Scaling-Up Climate Change Effects in Greenland Stine Højlund Pedersen

Scaling-up Climate Change Effects in Greenland Stine Højlund Pedersen

PhD Thesis 2017

Scaling-up Climate Change Effects in Greenland

Stine Højlund Pedersen

PhD Thesis

2017

Graduate School of Science and Technology, Aarhus University, Denmark.

Department of Bioscience, Greenland Ecosystem Monitoring, Arctic Research Centre

Data sheet Title: Scaling-up Climate Change Effects in Greenland Subtitle: PhD Thesis Author: Stine Højlund Pedersen Institute: Department of Bioscience, Arctic Research Centre, Aarhus University, Denmark and Greenland Ecosystem Monitoring. Publisher: Aarhus University, Denmark URL: www.bios.au.dk/en/ Year of publication: 2017 PhD supervisors: Senior Researcher Niels Martin Schmidt and Senior Researcher Mikkel P. Tamstorf, Aarhus University, Denmark, and Dr. Glen E. Liston, Cooperative Institute for Research in the Atmosphere, Colorado State University, U.S.A. Assessment Associate Professor Dr. Marc Macias-Fauria, School of Geography and the committee: Environment, University of Oxford, UK., Professor Dr. Jason E. Box, Department of Glaciology and Climate, The Geologic Survey of Denmark and Greenland (GEUS), Denmark, and Senior Researcher Dr. Morten Frederiksen, Department of Bioscience, University of Aarhus, Denmark. Please cite as: Højlund Pedersen, S. 2017. Scaling-up Climate Change Effects in Greenland. PhD Thesis. Aarhus University. Department of Bioscience, Denmark. 178 pp Abstract: In ice-free Greenland, extensive knowledge and a mechanistic understanding of interactions between abiotic and biotic ecosystem components have been gained from observations collected in the Greenland Ecosystem Monitoring (GEM) sites during 10-20 years. This PhD project was initiated to facilitate an understanding of these interactions also in between observational sites using up-scaling. Since, the seasonal snow cover is a key driver of changes in Arctic ecosystems, snow is the main focus of this interdisciplinary study. The project aim was twofold; (i) to quantify the spatial and temporal changes and variability in snow characteristics across multiple spatial scales and time periods in ice-free Greenland, and (ii) to investigate the effects of these snow changes and variability on biotic components of the ecosystems. This was accomplished by combining ground-based observations, spatial-temporal snow and atmospheric modeling tools, and remotely sensed vegetation greenness in a stepwise up-scaling to gradually larger domain extents. Keywords: Snow, vegetation greenness, gradients, observations, modeling, scales, remote sensing, Greenland. Layout and all photos: Stine Højlund Pedersen Front cover: Freshly fallen snow in the morning near Tasiilaq, East Greenland. Back cover: Heavily eroded snow surface at sunset near Qassiarsuk, West Greenland. ISBN: 978-87-93129-39-9 Printed by: LaserTryk.dk A/S Circulation: 50 Time and place of 24 February 2017 at 13:00, Niels Bohr Auditorium, Aarhus University, Department defence: of Bioscience, Frederiksborgvej 399, 4000 Roskilde, Denmark Animations:

Table of Contents Data sheet ...... 2 Preface ...... 5 Acknowledgements ...... 6 List of included papers ...... 8 List of additional contributions...... 9 Summary ...... 10 Resumé (Summary in Danish) ...... 12 Eqikkaaneq (Summary in Greenlandic) ...... 14 Seasonal snow cover in the Arctic ecosystems...... 16 Snow monitoring in Greenland...... 17 Challenges in research of changing Arctic snow cover: a scale/resolution mismatch between observations, modeling, and impacts...... 17 Scaling-up from point to region to entire ice-free Greenland ...... 18 Scale-dependent modeling representing processes at relevant resolution ...... 18 Inclusion of ground observations ...... 20 Stepwise knowledge-gain in time and space across multiple scales ...... 21 Local scale ...... 21 Regional scale ...... 22 Entire ice-free Greenland ...... 22 Looking into the future ...... 23 Perspectives and Concluding remarks ...... 23 References ...... 25 Paper I ...... Paper II ...... Paper III ...... Paper IV ...... Paper V ...... Paper VI ...... Paper SI ...... Paper SII ...... Paper SIII ...... Paper SIV ......

Preface This thesis represents a partial fulfilment of the requirements for the degree of Doctor of Philosophy (PhD) at the Faculty of Science and Technology, Aarhus University, Denmark. The research presented in this thesis is the results of a four year PhD project conducted at Department of Bioscience and Arctic Research Centre, Aarhus University under the supervision of Niels Martin Schmidt, Mikkel P. Tamstorf, and Glen E. Liston. A part of the results are produced in collaboration with international as well as Danish colleagues, all of whom are listed as co-authors of the respective papers. The thesis includes an introduction describing the aims, data sources, challenges, and methodology of the PhD project. Hereafter, I summarize and synthesize the findings of the four published papers, a manuscript, and an extended abstract (Paper I-VI) comprising the thesis.

When I started my university education almost ten years ago, my wish was to understand and figure out how the nature around me was functioning. One of the requests to the part of nature, I was going to study, was that it had to vary over relatively short time spans, for example between seasons or from year to year, unlike processes evolving over millions of years. Therefore, I pursued the field of physical geography and learned about passing weather systems, river discharge, plant and soil gas-flux exchange, and temperature fluctuations. But I fell in love with the science of snow and not least in Greenland. And I promise you that very early in my PhD studies my desire for the volatile nature of snow was met. An early morning in March 2013, I woke up in Nuuk after a windy night and opened the window while the curtains were still closed. An unusual, earthy, spring-like odor met me as the fresh air streamed into my room. This confused me since the day before had been, literally, freezing cold. So I ripped the curtains to the sides and quickly realized the reason for the unexpected smell. Outside my window, entire snow drifts had disappeared over night and the snow had started melting away. Mind you, the snow that I had traveled 3000 km to measure and study! My plans changed. Plans change and nothing is permanent, neither in snow science nor in life. I have come to realize that, and many other important things, during the past four years. Most of these more important things are presented and summarized on the following pages of this thesis.

St. Nord, N-Greenland (2015) 6

Acknowledgements

Mange tak, qujanaq, and sincere thank you to…

Greenland Ecosystem Monitoring, the Danish Environmental Protection Agency and the Danish Energy Agency, and the Arctic Research Centre (ARC) at Aarhus University for making this project possible and financing my four incredible Greenland journeys.

All co-authors for your valuable ideas, edits, contributions to the papers included in this PhD thesis.

Niels Martin Schmidt, my supervisor, for introducing me to your world of biology, and for your contagious enthusiasm and encouragements. I truly appreciated your straightforward, sincere, and down-to-earth way of being through these four challenging years. Thank you for keeping your office door and heart open for both minor and major questions and issues.

Mikkel P. Tamstorf, my co-supervisor, for your mentoring, motivation, remote sensing support, strategic tips, and ambitious initiatives. Thank you for always greeting me enthusiastically ‘Jamen, det er jo Stiiiine, min ph.d. studerende!’ every single time I bothered you in your office and stole away your time. It has been a pleasure to share the science with you.

Glen E. Liston, my mentor and field-companion. Thank you for believing in me from the moment I first stepped into your office in Colorado. You are a great inspiration to me both as a scientist and as a human. Thank you for introducing me to this wonderful snow world both inside and outside the computer. You have pushed me forward and while it sometimes has been tough, I have learned more than I had ever imagined in the past four years.

Without doubt, I have been granted with an inter-disciplinary committee simply including the best people. At times it has been challenging to bridge the fields of biology, physical geography, and snow science, but what we have accomplished from it definitely made it worthwhile. The four of us set out on a mission four years ago, I believe we reached the goal together. It has been a lot of fun. Thank you.

Means of transport in Qaanaaq, NW-Greenland (2014) 7

Collaborators and colleagues at the Department of Bioscience, Aarhus University. Ole Lund for your IT support and being my Linux-goto-person. Members of the ‘GEM Family’ from Asiaq – Greenland Survey, Lund University, GEUS, CENPERM, and University of Copenhagen.

Kelly Elder for introducing me to snow field measurements, sampling equipment, and snow photography, and foremost your great company out in the snow. I also appreciate the help and down-suit tips from the CIRA staff and technical support from Steve Finley during my visits at Colorado State University.

All field assistants and logistic supporters in Greenland including Jørgen Skafte & Kenny Madsen at Zackenberg Research Station, Glen in Zackenberg and the surrounding region, Stian and Markus from the Sirius Dog Sled Patrol in Daneborg, Pele in Tasiilaq, Kuluk & Ellen Frederiksen in Qassiarsuk, Ivali, Carl, Katrine, and Jakob in Nuuk, Jørgen Larsen in Kangerlussuaq, Mads Ole Kristensen & the sledge dogs in Qaanaaq, Birte & Hans Jensen at Hotel Qaanaaq, and Bjarne Jensen & Kasper Hancke at Station Nord. Thank you for driving me by boat, truck, snow mobile, or dog sledge to all my field sites and thank you for noting hundreds of snow observations in my little curly notebook and assisting in digging snowpits. Thanks to your warm housing, I was never cold despite the frosty weather outdoors. Katrine & Jose, thank you for welcoming, accommodating, and serving me delicious Greenlandic meals while staying in Nuuk.

Katherine & Glen for including me in your life and for welcoming me in your home in all its peace and quiet on SnowMan Road in Colorado. Katherine, I thank you for your friendship and for extending my snow skills into snow ploughing using your tractor.

My cherished friends Ane, Cecilie, Sine, Malene, Line, Ane, Kirstine, Lærke, Jesper, Rozemien, Thea, Anne, Nanna, Louise, Katrine, Magda, Lonnie, Gitte, Karin, Herbert, JP, Rune, Martin, Bo, & Preben for your smile, hugs, ideas, fun times, opinion, high-fives, shoulders, winter and summer dips, hiking trips, and lunch-walks in the fresh air.

Brormand Søren, Christina, Mikkel & Anna, I gør mig glad og har givet mig de bedste pauser fra arbejdet, som en søster/faster kan ønske sig.

Min kære mor & far, tak for I trofast støtter mig i det, der gør mig glad her i livet. Jeg værdsætter jeres ærlighed. Tak for de mange friweekender på Fyn med hygge, eftermiddagskaffe, syltning & bagning, frisk luft og en hel køkkenhave at hakke igennem, når det var dét, jeg havde allermest brug for.

Af hele mit hjerte tak,

List of included papers

Paper I Bokhorst, S., Pedersen, S. H., Brucker, L., Anisimov, O., Bjerke, J. W., Brown, R. D., Ehrich, D., Essery, R. L. H., Heilig, A., Ingvander, S., Johansson, C., Johansson, M., Jónsdóttir, I. S., Inga, N., Luojus, K., Macelloni, G., Mariash, H., McLennan, D., Rosqvist, G. N., Sato, A., Savela, H., Schneebeli, M., Sokolov, A., Sokratov, S. A., Terzago, S., Vikhamar-Schuler, D., Williamson, S., Qiu, Y., and Callaghan, T. V. Changing Arctic snow cover: A review of recent developments and assessment of future needs for observations, modelling, and impacts. Published: Ambio. Vol. 45, No. 5, pp. 516-537 (2016).

Paper II Pedersen, S. H., Tamstorf, M. P., Abermann, J., Westergaard-Nielsen, A., Lund, M., Skov, K. Sigsgaard, C., Mylius, M. R., Hansen, B. U., Liston, G. E., and Schmidt, N. M. Spatiotemporal characteristics of seasonal snow cover in Northeast Greenland from in situ observations. Published: Arctic, Antarctic, and Alpine Research. Vol. 48, No. 4, 2016, pp. 653-671 (2016).

Paper III Pedersen, S. H., Liston, G. E., Tamstorf, M. P., Westergaard-Nielsen, A., and Schmidt, N. M. Quantifying episodic snowmelt events in Arctic ecosystems. Published: Ecosystems. Vol. 18, No. 5, pp. 839-856 (2015).

Paper IV Pedersen, S. H., Liston, G. E., Tamstorf, M. P., and Schmidt, N. M. Snow-free date drives maximum NDVI timing: A mesocosm study across compressed Arctic environmental gradients. Manuscript: aimed at Journal of Geophysical Research: Biogeosciences.

Paper V Pedersen, S. H., Liston, G. E., Tamstorf, M. P., and Schmidt, N. M. Timing of maximum vegetation greenness across ice-free Greenland. Extended abstract.

Paper VI Westermann, S., Elberling, B., Pedersen, S. H., Stendel, M., Hansen, B.U., and Liston, G.E. Future permafrost conditions along environmental gradients in Zackenberg, Greenland. Published: The Cryosphere. Vol. 9, No.2, pp. 719-735 (2015).

List of additional contributions

Paper S I Pirk, N., Lund, M., Mastepanov, M., Parmentier, F.-J. W., Pedersen, S. H., Mylius, M. R., Tamstorf, M. P., Christiansen, H. H., and Christensen, T. R. Snowpack fluxes of methane and carbon dioxide from high Arctic tundra. Published: Journal of Geophysical Research: Biogeosciences, Vol. 121, No. 10, doi: 10.1002/2016JG003486 (2016).

Paper S II Westergaard-Nielsen, A., Lund, M., Pedersen, S. H., Schmidt, N. M., Klosterman, S., Abermann, J., and Hansen, B. U. Transitions in high-Arctic vegetation growth patterns and ecosystem productivity tracked with automated cameras from 2000 to 2013. Published: Ambio. DOI: 10.1007/s13280-016-0864-8.

Paper S III Schmidt, N. M., Pedersen, S. H., Mosbacher, J. B., and Hansen, L. H. Long-term patterns of muskox (Ovibos moschatus) demographics in High Arctic Greenland. Published: Polar Biology. Vol. 38, pp. 1667–1675 (2015).

Paper S IV Lund, M., Hansen, B. U., Pedersen, S. H., Stiegler, C., and Tamstorf, M. P. Characteristics of summer-time energy exchange in a high Arctic tundra heath. Published: Tellus. Series B: Chemical and Physical Meteorology. Vol. 66, [21631] (2014).

Kobbefjord,W-Greenland (2012) 10

Summary In ice-free Greenland, extensive knowledge and a mechanistic understanding of interactions between abiotic and biotic ecosystem components have been gained from observations collected in the Greenland Ecosystem Monitoring (GEM) sites during 10-20 years. This PhD project was initiated to facilitate an understanding of these interactions also in between observational sites using up-scaling. Since, the seasonal snow cover is a key driver of changes in Arctic ecosystems, snow is the main focus of this interdisciplinary study. The project aim was twofold; (i) to quantify the spatial and temporal changes and variability in snow characteristics across multiple spatial scales and time periods in ice-free Greenland, and (ii) to investigate the effects of these snow changes and variability on biotic components of the ecosystems. This was accomplished by combining ground- based observations, spatial-temporal snow and atmospheric modeling tools, and remotely sensed vegetation greenness in a stepwise up-scaling to gradually larger domain extents.

In Paper I, we review the role of the changing Arctic snow cover as being the prominent driver of changes in both societies and ecosystems in snow-covered environments. We emphasize the challenges in the research of a changing Arctic snow cover, mainly originating from a mismatch in spatial and temporal scale and resolution between observations, modeling efforts, and data requests from researchers studying the impacts of snow- cover changes. Paper I illustrates snow’s importance in the Arctic ecosystem, which reasons the strong focus on seasonal snow in this thesis. From Paper II through Paper VI, the thesis gradually advance in scale and resolution using the adaptable modeling tools, MicroMet and SnowModel, to counteract the scale issues addressed in Paper I.

In Paper II, we quantified the spatial variability, the temporal trends, and interannual as well as seasonal variation in the observed snow characteristics during the past 18 years at local scale in Zackenberg Valley in NE-Greenland. We found pronounced interannual variability, particularly in the timing and magnitude of snow, which may have masked significant temporal trends.

In Paper III, we quantified the snow-related effects of a 2-days episodic snowmelt event (ESE) in March 2013 across a valley domain in W-Greenland. The knowledge gained about ESEs on local scale was used in quantifying the spatial distributions and trends of past ESEs across the entire ice-free Greenland. We find that ESEs are common in Greenland snow-cover dynamics, and because of their impact on ecological relevant snow properties, should be accounted for in snow-related ecosystem studies.

In Paper IV, regional snow and vegetation greenness distributions are investigated across decadal time series to quantify inland-coast gradients in NE-Greenland. The inland-coast temperature gradient is eight times stronger than the south-to-north temperature gradient observed along the Greenland east coast. Hence, the region may comprise a scientific mesocosm, which offers opportunities to study biological responses to environmental drivers within relatively short distances. Additionally, we find that the spring snowpack water content and timing of snowmelt govern the spatially and temporally distribution of the seasonal maximum vegetation greenness and its timing.

In the extended abstract for Paper V, the project pioneers into representing and mapping climate variables across the entire ice-free Greenland in the highest spatially and temporally possible resolution through 36 years to investigate how climate controls vegetation greenness and its timing. The latter were estimated from daily MODIS reflectance data. The spatial patterns of the distributed drivers accentuate the complexity found within the Greenland study area. Yet it stresses the need to disentangle the different driver’s importance for vegetation greenness within climatically contrasting regions. 11

In Paper VI we couple SnowModel with a ground thermal model and regional climate model outputs to predict changes in permafrost conditions towards year 2100. The predicted average ground temperatures are increasing, but remain generally below freezing in the Zackenberg region.

The Greenland-wide representations of climate variables in Paper V show that the variability, which was observed on local scale in Paper II, 3, and 4, is identifiable in other regions at varying altitude and latitude. Hence, the local-scale variability accentuates a part of the variability found on larger regional scales. As also argued for in Paper I, this confirms the importance of local-scale observations collected at the GEM sites, since they have extensive explanatory power exceeding their domain and/or time period, in which they are observed, exemplified by the permafrost predict research in Paper VI.

Trapperhut at Falskebugt, NE-Greenland (2014)

Resumé (Summary in Danish) Indenfor den kystnære, isfri del af Grønland har man de sidste 10-20 år fået en detaljeret viden om de geofysiske og biologiske dele af økosystemet og hvordan disse enkelte dele interagere. Forståelsen er frembragt via data fra overvågningsprogrammet, Greenland Ecosystem Monitoring (GEM). GEM har to undersøgelsesområder beliggende på Grønlands vest - og østkyst. Dette ph.d.-projekt blev begyndt i år 2012 med henblik på at øge forståelsen for økologiske processer og interaktioner udenfor for disse to undersøgelsesområder.

Sne er det primære fokus i dette tværfaglige ph.d.-projekt, da snedækket i Arktis er drivkraft for ændringer i økosystemet. Projektet har to mål; (i) at kvantificere de rumlige og tidsmæssige variationer i sneens egenskaber, på tværs af rumlige skalaer og indenfor forskellige tidsperioder af det isfri Grønland, og (ii) at undersøge effekterne på de arktiske økosystemer, som følge af sneens variabilitet. Begge mål blev nået ved at anvende en kombination af landbaserede snemålinger, snemodelleringsværktøjer og satellit-baserede observationer. Via en trinvis opskalering igennem ph.d.-projektet blev viden om snedynamikker og sneprocesser anvendt på stadig større geografiske områder af den kystnære, isfri del af Grønland.

I artikel I præsenterer vi hvilke effekter, ændringer i det arktiske snedække kan have på samfund og økosystemer, hvor den primære vinternedbør falder som sne. Ligeledes fremhæver vi nuværende udfordringer indenfor forskning af det arktiske snedække. Disse består primært i uoverensstemmelser i rumlige og tidsmæssige skala og opløsning mellem indsamlede observationer, model-repræsentation og datakrav fra forskere, som studerer økologiske eller samfundsmæssige effekter af ændringer i sneen. For at imødekomme denne udfordring anbefaler vi en forbedret dialogen mellem de individuelle fagfelter omkring kravene til den rumlige og tidsmæssige opløsning af data, og styrker og begrænsninger ved modellering. Artikel I påviser sneens betydning for det arktiske økosystem, hvilket begrunder det stærke fokus på netop det arktiske snedække i denne ph.d.-afhandling.

Artikel II er baseret på sneobservationer indsamlet igennem 18 år i Zackenberg-dalen i Nordøstgrønland. Vi fandt en enorm variation imellem årene i bl.a. snesmeltningstidspunktet om foråret og tidspunktet for den første snebyge i efteråret. Generelt fandt vi ingen tegn på betydningsfulde ændringer i de målte sneparametre på grund af den markante år-til-års variation.

I artikel III undersøgte and kvantificerede vi, hvordan et 2-dages akut snesmeltningsevent (SSE) i marts 2013 påvirkede sneen i et dalområde nær Nuuk i Vestgrønland. Med viden om SSEs karakteristika på lokal skala kunne vi kortlægge SSE’ernes rumlige og tidsmæssige udbredelse, indenfor det isfri Grønland i årene 1979- 2013. Af det lærte vi, at SSE’er er en almindelig del af de grønlandske snedynamikker. Deres indvirkning på snedækket og potentielt på økosystemet skal dog medregnes i sne-relaterede studier foretaget i Grønland.

I artikel IV er årtiers regionale rumlige udbredelse, herunder fordeling af sne og vegetationsdækkets grønhed, kortlagt, for at kvantificere kyst-indland gradienter i Nordøstgrønland. Vi finder i denne artikel, at regionen indeholder komprimerede og kraftige klimagradienter. Dette gør regionen til et ideelt studieområde at forstå effekten af klimatiske ændringer og variationer på biologiske processer indenfor korte afstande. Vi finder ydermere at den rumlige fordeling af vegetationsdækket maksimale grønhed, samt hvornår på vækstsæsonen dette indtræffer, er reguleret af vandindholdet i forårets snedække og tidspunktet for snesmeltning.

I artikel V kortlægger vi, som de første, klimafaktorer indenfor det isfri Grønland, i den højest mulige rumlige og tidsmæssige opløsning gennem 36 år. Formålet er at undersøge, hvordan klimaet styrer vegetationsdækkets grønhed og vegetationens fænologi, som bliver estimeret via daglige satellitdata, som dækker hele Grønland. De rumlige mønstre af klimafaktorenes fordeling fremhæver den kompleksitet i terræn og gradienter, der findes 13

indenfor det grønlandske studieområde. Kompleksiteten understreger desuden behovet for at kende de forskellige klimafaktorers betydning for vegetationsdækket i klimatisk kontrasterende regioner.

I artikel VI vender vi blikket mod fremtiden. Her kobler vi snemodellen, SnowModel, sammen med en geotermiske model og klimadata fra en regional klimamodel, for at forudsige ændringer i præmisserne for permafrost frem mod år 2100. Vores forudsigelser påviser, at den gennemsnitlige temperatur i jordlagene i Zackenberg-området i Nordøstgrønland vil stige betydeligt de næste hundrede år, men generelt vil forblive under frysepunktet, således at permafrosten ikke påvirkes. Studiet viser dog at jordtemperaturene er letpåvirkelige over for ændringer i snedækkets tykkelse.

Kortlægningen af de primære klimafaktorer inklusiv sne i artikel V fremhæver, hvordan variabiliteten, observeret på lokal skala i artikel II, III og IV, kan identificeres i andre regioner, i forskellige højdeniveauer og på forskellige breddegrader indenfor det isfrie Grønland. Dette, som der også argumenteres for i artikel I, bekræfter vigtigheden af observationer indsamlet på GEM lokaliteter, da disse rummer en forklaringskraft, som rækker ud over det geografiske område og/eller tidsrum, hvori de er observeret.

Falskebugt, NE-Greenland (2014)

Eqikkaaneq (Summary in Greenlandic) Kalaallit Nunaanni sineriammut qanittumi ukiut kingulliit 10-12t ingerlaneranni nunap akui uumassusillillu uumassusillit sunneqatigiittarnerannut ataatsimut immikkoorlutillu sunniuteqartarneri sukumiisumik ilisimasaqarfigineqalernikuupput. Sunneqatigiittartut paasisassarsiorneqarput alapernaarsuinissamut aaqqissuussaq Greenland Ecosystem Monitoring (GEM) atorlugu. Sumiiffiit GEMimik misissugarineqartut Kitaani Tunumilu inissisimapput. Sumiiffiit taakkua marluk avataanni uumassusillit sunneqatigiittarnerat paasisassarsiorfiginiarlugu ph.d.-nngorniut una 2012imi aallartinneqarpoq.

Issittumi uumassusillit sunneqatigiittarnerisa allanngorarnerannut aput sunniuteqarnerpaasarmat, ph.d.- nngorniummi uani assigiinngitsunik sammisaqarfiusumi aput immikkut misissugarineqarpoq. Suliariniagaq marlunnik siunertaqarpoq; (i) Kalaallit Nunaata piffissani assigiinngitsuni sikoqannginnerata nalaani aputip qanoq angitigisup piffissallu qanoq sivisutigisup qanoq sunniisinnaaneri annertusulersorneqarput, kiisalu (ii) Issittumi aputip assigiinngiiaartup allanngorarnerata uumassusilinnut sunniutaanera misissorneqarpoq. Uuttortarneqartut nunami apummmik uuttuutitigut, apummik ilusilersukkat sakkutut atornerisigut kiisalu qaammataasaniit alapernaarsukkat atorlugit uuttortaasoqarpoq. Ph.d.-nngorniutip ingerlannerani aputip assigiinngitsup imminut qanoq sunnertarnera aputillu allanngorarnera, Kalaallit Nunaanni nuna sikoqanngitsoq annerunerujartuinnartoq sinerissamut qanittumi misissugarineqarpoq.

Artikel 1imi saqqummersippagut Issittumi ukiuunerani aputip allanngorarnera inuiaqatigiinnut uumassusillillu sunneqatigiittarnerannut qanoq sunniuteqartarneranut tunngasut. Issittumi aputip ilisimasassarsiorfiginerani ullutsinni unammilligassat aamma erseqqarissarpagut. Tamakkua tassaanerupput angissusilersuinerit piffissamillu nalunaarsuinerit kiisalu misissukkat nalunaarsornerisa, assersuutit saqqummersinneranni kiisalu ilisimatuunit uumassusillit sunneqatigiittarnerisa inuiaqatigiillu apummit sunnerneqartarnerat pillugu paasissutissat ilanngunneqartussatut piumasaqaatigineqartut, imminnut assigiinnginnerat. Unammilligassat tamakkua akiorniarlugit assigiinngitsunik ataasiakkaanik sammisallit tamarmik akornanni angissusilersuinissamut piffissalersuinermullu paasissutissat piumasaqaatigineqartut pillugit pitsaanerusumik isumasioqatigiittarnissaq kaammattuutigineqarpoq, tamatumuuna assersuutissat tutsuiginarnerullutillu killilersuutaassammata. Artikel 1imi Issittumi uumassusillit sunneqatigiittarnerannut aput sunniuteqartartoq paasinarsivoq, tassalu uani ph.d.-nngorniummi Issittumi qanoq aputeqartiginerata salliutillugu misissugarinera tunngavissillugu.

Artikel 2mi paasissutissat tassaapput Tunup avnnaata kangiani Zackenbergip qooruani ukiuni 18ini apummik misissuinermit paasissutissat. Assersuutigalugu upernaakkut aputip aalersarneri ukiakkullu apeqqaarneri ukiuni taakkunani annertoqisumik assigiinngiiaartoq paasivarput. Ukiumiilli ukiumut aputeqarnerata allanngorarnera ataatsimut eqqarsaatigalugu pinartumik sunniuteqarneq ajortoq.

Artikel 3mi sammineqartoq tassaavoq, Nunatta Kitaani Nuup eqqaani qooroqarfimmi 2013imi martsip qaammataani, ulluni marlunni aputip aakkiartortup qanoq annertutigineranik (SSE) misissuinerit paasissutissartaat. SSE pillugu paasisat naapertorlugit, 1979imiit 2013ip tungaanut aputip qanoq annertutigisup qanoq sivisutigisumik aakkiartornikuunera uuttortarsinnaavarput. Taamatut paasivarput Kalaallit Nunaanni aputip pissusaanut SSE-t nalinginnaasumik sunniuteqartartut. Taakkuali qanoq aputeqartigineranut sunniutaat kiisalu immaqa aamma uumassusillit imminnut sunneqatigiissinnaanerannut paasissutissat, Kalaallit Nunaanni aput pillugu paasisassarsiornernut ilanngunneqassapput.

Artikel 4mi sammineqartut tassaapput ukiuni qulikkaani nunap immikkoortortaani aputip qanoq atsigisup aattarneranik kiisalu aputip aputerqannginnerasanilu naasut qanoq qorsooqqitsiginerannik misissuinerit sumiiffii ilanngullugit nalunaasorneqarput, Tunup avannaata kitaata timaani paasissutissat tutsuiginartuuneri 15

qulakkeerniarlugit. Immikkoortumi tassani atuarneqarsinnaapput, nunap immikkoortortaani tamaani silap pissusaa sukkasuumik sakkortuumillu allanngoriarsinnaasartoq. Taamaammat tamanna misissuiffissaqqippoq, taama sivikitsigisumi isorartunngitsumi silap allanngorarnerata kiisalu uumassusillit allanngorarnerisa sunniutaat paasiniarlugit. Naasut qaqugukkut qorsooqqissinerusarnersut kiisalu naaffigisinnaasaasa qanoq ilinerani tamanna pisarnersoq, tamannalu upernaakkut qanoq aputeqartiginera kiisalu aputip aalernerata nalaani tamakkua sunniutaasarnerat aamma paasivarput.

Artikel 5imi Kalaallit Nunaata sikoqannginnersaani silap pissusaanit sunnerneqartartut, ukiuni 36ni angissusilersuinerit piffissalersuinerillu aallaavigalugit siullersaalluta sumiissusaannik nalunaarsuivugut. Tamatumani siunertarineqarpoq, naasoqarfiit qanoq qorsooqqitsiginerat kiisalu naasut silap pissusaanit sunnerneqartarnerat paasiniarlugu, ullut tamaasa qaammataasaniit Kalaallit Nunaat tamaat pillugu paasissutissat atorlugit missingersuisoqartarluni. Silap pissusaanik sunnerneqartut sumiinnerannik nalunaarsuinikkut, Kalaallit Nunaanni misissuiffigisaq maniillunilu assigiinngitsunik sunniuteqarfissartaqartoq. Nunap immikkoortortaani imminnut assigiinngeqisuni sunniutaasut amerlaqisut apeqqutaasut paasiniarneranni taamaammat silap pissusaata naasunut sunniutigisartagaasa ilisimaarinissaat pingaaruteqarpoq.

Artikel 6imi siunissaq sammineqarpoq. Tassani apummik assersuusiaq ”SnowModel” atortoraarput, nunami silamut assersuummik nunamilu immikkoortortami pineqartumi silap pissusaata sunniutaatut nalunaarsukkat katillugit, ukioq 2100mi nunap ikerata qeriinnarnissaanut apeqqutaasut allannguuteqassanersut siulittuutiginiarneqarluni. Siulittuutitta takutippaat, Tunup avannaata kangiani Zackenbergip eqqaani nunap ikera ukiuni tulliuttuni 100ni annertuumik kissatsissasoq kisiannili qerinartumiit kissarnerulissanngimmat, nunap ikera qeriuaannartoq qeriuaannassaaq. Misissuinernili paasinarpoq, nunap kissassusaanut qanoq aputeqartiginera sunniutiasoq.

Silap pissusaata kinguneranit sunnerneqartarput, artikel 5imi aput eqqartorneqartoq ilanngullugu, taakkua ersersippaat sumiiffinni artikel 2, 3 aamma 4mi misissukkanit paasisat, qaffasissutsini assigiinngitsuni Kalaallit Nunaannilu sikoqanngitsuni avannarpasissutsini assigiinngitsuni missingersuusersuisoqarsinnaalluni. Artikel 1imi sumiiffinni GEMimik katersat pingaarutaaneri tunngavilersorneqarpoq, taakkua aqqutigalugit misissuiffinni sumiikkaluartuniluunniit kiisalu qaqugukkut katersugaaneri apeqqutaatinnagit nassuiaataasinnaasunik imaqarput.

Seasonal snow cover in the Arctic ecosystems In ice-free Greenland, interactions between abiotic and biotic ecosystem components have been monitored and studied in detail during previous decades mainly in two sites by Greenland Ecosystem Monitoring. However, this PhD project was initiated to gain the knowledge about these interactions taking place also in between the two observational sites, i.e., within approximately 19 % of the total Greenland area. Essential for this up- scaling process is the understanding of the Arctic ecosystems, gained from long-term monitoring in these ‘observation points’, relatively speaking. Therefore, the first aim of the PhD project is to quantify the spatial and temporal changes and variability in climate, mainly snow, characteristics across multiple spatial scales and different time periods in ice-free Greenland. The second aim is to investigate the effects of these snow changes and variability on biotic components of the Arctic ecosystems.

The main assumption of the PhD project is that abrupt to gradual changes as well as inter-annual to interdecadal variability in seasonal snow cover may explain changes observed in biotic and abiotic ecosystem components and their dynamics. Therefore, my research has particularly been focused on seasonal snow, since it is a key driver of climate change effects and controlling a range of Arctic ecosystem processes (Jones 1999, Post et al. 2009, Brooks et al. 2011, Callaghan et al. 2011, Bokhorst et al. 2016). Hence my first paper, Paper I, highlights how changes in snow conditions impact both the society and ecosystems in multiple ways. More specifically, snow properties, such as snow depth, density, snow-water-equivalent (SWE), thermal conductivity, and timing of snowmelt, influence and drives the biotic and abiotic ecosystem processes. For instance, the presence and absence of snow cover influence the surface energy balance both locally (Marks and Dozier 1992, Stiegler et al. 2016) and globally (Groisman et al. 1994), which in turn affects the belowground surface thermal regime. The snow cover acts as an efficient insulator during winter with its low thermal conductivity (Goodrich 1982, Sturm et al. 1997) and high thermal resistance (Liston et al. 2002). This insulating effect keeps soil thermal conditions relatively stable during snow-covered periods (Zhang 2005), regulate the decomposition and sequestration rates in the soil (Schimel et al. 2004, Johansson et al. 2013), controls the active-layer depth (Paper VI), and protects the vegetation cover from frost damages (Bokhorst et al. 2011). In turn, soil thermal conditions drive the microbial activity, the respiration rate, and the amount of soil organic carbon produced during winter (Elberling 2007). Hence, the snow-depth evolution through autumn and winter governs the amount and timing of plant-available nutrients at the end of winter and the following spring in tundra ecosystems (Schimel et al. 2004, Buckeridge and Grogan 2008). In addition, the snowpack characteristics affect the winter soil-atmosphere gas emissions of methane and carbon dioxide. In Paper SI, we investigated the effects of snow depth, snowpack stratigraphy, and snow density on gas emissions measured during late winter at two high-Arctic, permafrost-underlain wetland sites in Zackenberg Valley and Adventdalen in Svalbard. We found that the spatial pattern of gas emissions of methane and carbon dioxide from the frozen soil mirrors the patterns observed during the growing season. The winter observations also showed that snowpack ice layers block the gas emission and result in increased gas concentrations below the ice layers. Hence, the snowpack can be thought of as a porous buffer zone that delays or even blocks the gas emissions to the atmosphere.

Throughout the Arctic, solid precipitation accumulates during autumn, winter, and spring in snowpacks, which act as water reservoirs (Jones, 1999). In spring, the water is released during snowmelt and provides moisture for plant growth, not only at the growing season initiation, but also into the summer. Particularly the timing of snowmelt is identified as a strong driver for the maximum level of the vegetation greenness and its timing during the growing season both in Greenland (Paper IV) and other Arctic regions (Pudas et al. 2008, Zeng and Jia 2013). The snow-water equivalent (SWE) of the snowpack at the end of winter even affects the timing of a range of vegetation phenological events. This was found in Paper SII, where we examined vegetation 17

phenological events derived from digital camera photos on a landscape scale. Finally, the winter snow cover plays an important role in ecology. For instance, the snow hardness and snowpack ice layers have a limiting effect on caribou forage availability (Collins and Smith 1991, Tyler 2010). Also, the spring snow cover is often used as a measure of the snow conditions of the preceding winter. In Paper SIII, we find that the spring snow cover, registered on 10 June, is the main driver of musk ox (Ovibos moschatus) population dynamics during the 18-year observational period in Zackenberg in NE-Greenland. The snow conditions mainly impacted the calf percentage, i.e., a negative proportional relationship, which in turn was closely linked to the recruitment to the population.

Snow monitoring in Greenland The main study area of the PhD project is the land parts, which are only seasonally snow-covered during 7-11 months per year (Paper V) and occupy ~19 % of the total Greenland area, thus excluding the Greenland Ice Sheet (GrIS) and mountain glaciers. Despite the snow-cover extent and its importance, snow observations are lacking in the coastal and ice-free areas of Greenland. Only few snow-cover studies on landscape scale have been conducted (Hinkler et al. 2002, Mernild et al. 2007) and snow field observations are limited to research projects and point measurements outside GrIS and glaciers (Hansen et al. 2006, Rogers et al. 2011). An exception to this data deficiency is two Greenland Ecosystem Monitoring (GEM) sites: Kobbefjord in SW- Greenland and Zackenberg in NE-Greenland. Nuuk Ecological Research Operations (NERO) and Zackenberg Ecological Research Operations (ZERO) have since 2007 and 1996, respectively, run two ecosystem baseline monitoring programs responsible for collecting extensive snow-observation datasets using manual, automated, and remote-sensing methods. The available snow datasets provide a unique opportunity to describe and examine the spatial and temporal distributions of snow-cover features and interactions in a low-Arctic and high-Arctic setting, where snow-dependent ecosystem components and processes are also being monitored. Hence, the snow observations are key in getting a mechanistic understanding of ecosystem processes and dynamics and constitute a valuable validation potential for modeling and remote-sensing datasets. However, while these observations cover a very small fraction of the ice-free area of Greenland, the knowledge derived from them is often applicable across a larger geographical area. Hence, to meet the request of quantifying the variations in snow in between these two well-studied areas, Kobbefjord and Zackenberg, should the valuable observations from here be included in the up-scaling process.

Challenges in research of changing Arctic snow cover: a scale/resolution mismatch between observations, modeling, and impacts. The biota and fauna in the Arctic may be affected by climate warming through snow-cover changes. Hence, studying and quantifying snow is important in order to gain an understanding of the impact of present and past changes and to equip us with knowledge to predict their potential consequences in a future Arctic. Paper I reviews recent advances in snow observing and modeling techniques, and we present an overview of recent impacts of the changing Arctic snow cover on societies and ecosystems. The snow research has escalated in the recent decades. Observational methods are being developed and monitoring and modeling efforts are initiated. This review argues that investigating snow and its variability may help explain some of the changes seen in the Arctic, since many of them are caused directly or indirectly by snow-related changes. These reviewed developments meet the need for methods and procedures to study these changes. We found that within the snow monitoring field, the development advances towards increasingly automated and subjective techniques to quantify snow metrics, which often results in a larger amount of available data than derived exclusively from manually collected observations. However, with an increasing availability of both observed and modeled snow data comes a need to maximize the usage of these data and to improve their potential explanatory power. This requires a stronger link between the fields of monitoring, modeling and studying the

impacts of snow. The changes in snow properties are not uniformly distributed across the Arctic and the affected processes in, e.g., ecosystems, are responding at different temporal and spatial scales and resolutions. Hence, one of the pitfalls in this matter, which we identified in Paper I, is that different fields are working with snow metrics and evaluate their related effects on different temporal and spatial scales and resolution. This results in a potential mismatch in the availability and requirements of snow data between snow scientists, modelers, ecologists, and sociologists and limits the comparability and corporation between fields. Instead, to meet this challenge and to promote increased data exchange, we recommend establishing a dialog between science fields about data resolution requirements, model constrains, and sampling strategies. Furthermore, we encourage scientists to improve the communication with society and management bodies about up-to-date knowledge on snow changes and risks and in return receive socially relevant concerns and questions that can help guide or/and improve the snow monitoring, modeling, and impact research (Figure 5 in Paper I).

Scaling-up from point to region to entire ice-free Greenland Observations, be it snow or biotic measures, are essential in gaining a better understanding of changes in snow, and their associated effects in the ecosystem. Also, observations are regarded as the ‘truth’ and are of high value in especially Arctic sciences because of their sparse availability due to the remoteness of many high- latitude observation sites. Since this project is initiated from the monitoring tradition of GEM, in which high- quality observations are collected, the research included in this project is founded on these observations, which are used to develop quantitative methods, to adjust model inputs and parameterizations, and in the validation of model outputs. To counteract this challenge of scale issues between observations and model outputs, I have applied dynamical and flexible modeling tools in the following subprojects (Paper II - Paper VI). Herein, the temporal and spatial resolution and extent are adjusted from project to project to match the scale defined by biological processes of interest, topographic features, and/or the temporal resolution required to resolve the processes or dynamics of interest. Hence, the subprojects are presented according to progression in increasing spatial and temporal scale.

Initially, the research is based on point observations through multiple-year time series (Paper II) and evolves through a stepwise increase in complexity, to gridded snow distributions in a valley domain, but within relative short extent in time, examining 2-days melt events (Paper III). Further, regional snow and vegetation greenness distributions are investigated across decadal time series to quantify inland-coast gradients in NE- Greenland (Paper IV). Then, the project pioneers into representing and mapping distributions of climate across the entire ice-free Greenland in the highest possible spatial and temporal resolution through 36 years (Paper V). Finally, the last subproject extents into the future examining climate change effects on permafrost conditions (Paper VI). Hence, the two overall project aims of examining processes at multiple scale and resolution requested a dynamic modeling scheme, which was adaptable to the temporal and spatial resolutions and extents that fitted the research purpose the best.

Scale-dependent modeling representing processes at relevant resolution Models are useful tools for describing biotic and abiotic ecosystem components and processes. Most ground- based, airborne, and space-borne observations of ecosystem components and processes are tied to a specific scale; they are confined to a specific grain (grid-cell size and/or time step) and extent (spatial and/or temporal domain) (Wiens 1989). Thus models used for simulating ecosystem features and/or processes should be scale- specific and match the inherent spatial and temporal scales of the features or processes of interest (Wu and Li 2009). The end result of a scale-dependent modeling approach is a stepwise up-scaling (in space and time) of ecosystem-relevant features and processes, where each component of the system has been accounted for by a modeling tool operating at a resolution appropriate for the processes it is simulating. 19

The meteorological and snow modeling systems, MicroMet and SnowModel (Liston and Elder 2006a, b) are ideal for this stepwise up-scaling approach. Their performance in distributing meteorological variables and evolve snow distributions in complex and mountainous terrain have been validated in Arctic and seasonally snow-covered areas around the world (Liston and Sturm 1998, Greene et al. 1999, Liston et al. 2000, Liston and Sturm 2002, Hasholt et al. 2003, Hiemstra et al. 2006, Liston et al. 2007, Liston and Hiemstra 2008, Randin et al. 2009, Liston and Hiemstra 2011, Stuefer et al. 2013, Mernild et al. 2016). Furthermore, MicroMet and SnowModel are applicable in the any computationally possible resolution and extent to fit the ecologically relevant scale of interest to my PhD subprojects. MicroMet and SnowModel (Figure 1) can be thought of as detailed process models that use our understanding of snow physics and dynamics, and convert basic meteorology such as air temperature, humidity, precipitation, wind, and radiation, into the evolution of complex snow variables such as snow depth, density and water content. When coupled with the SnowAssim submodel (Liston and Hiemstra 2008), the resulting SnowModel distributions include all of that physics, plus the added feature of observed snow-property distributions, where and when these are available, and the simulations will produce distributions that maintain that physical realism in the spatial and temporal space between the observations. The result is spatially- and temporally-continuous snow-property distributions that match both our physical understanding of snow processes (e.g., snowfall/precipitation distributions, albedo evolution, conservation of mass and energy, blowing snow redistribution, melt rates and timing, snow-covered- area evolution, etc.) and the available field snow observations.

Figure 1 Schematic description of MicroMet and SnowModel (From Paper III).

The winter precipitation measurements are challenging and comes with large uncertainties (Rasmussen et al. 2012). Networks of manual and automated snow observations and measurements are extensive and standardized in North America, e.g., by SNOwpack TELemetry (SNOTEL) sites (Barton and Burke 1977), but remains limited, less extensive, or data are collected through short-term research projects in large parts of the pan-Arctic area. Consequently, most Arctic spatially distributed snow data currently available rely on remotely sensed acquired data or model outputs. Remote sensing products, mainly originating from visible imagery, are limited to detect two-dimensional snow metrics, e.g. snow cover and snow albedo, which enable identification of arrival, duration, and depletion of the snow cover (Ramsay 1998, Hall et al. 2002). However, validation of these data sets for northern latitudes and mountain areas remains limited (Déry and Brown 2007) and the snow classifications scheme rely on persistence, i.e., if a grid cell is snow-covered today, it likely that it is snow-covered tomorrow. Hence, day-to-day changes in the snow cover may not be captured in these weekly synthesis products. Attempts to estimate snow depth from airborne radar measurements show accurate results for sea-ice surfaces. However, the application of this method is still limited to flat surfaces and the data

sets represents a snap shot in time (e.g., one day) and in space (e.g., 15m by 11m) due to costly operations acquiring the data (Kurtz et al. 2013). Also SWE estimates from both passive microwave sensors and optical satellite imagery remain available in a coarse spatial resolution (> 25 km) (Hancock et al. 2013). Snow outputs from global circulation models are similarly available in a coarse spatial resolution and attached with large uncertainties due to lack of simulated snowpack evolution included in the modeling scheme (Brown and Mote 2009). Common for these data resources is that spatial and temporal scaling of the snow data is required in order to apply the snow metrics on a landscape scale, i.e., within a catchment or valley. In contrast, MicroMet and SnowModel outputs are originating from a user-specified and ecologically relevant extent and resolution. An additional strength of MicroMet and SnowModel is that the modeling schemes can be constrained to resemble the real world by assimilating ground-based observations of snow and meteorological variables.

Figure 2 Locations of sites and stations providing data for Paper II- Paper VI. Left: Automated weather stations of Greenland Ecosystem Monitoring (GEM), Danish Meteorological Institute (DMI), and Arctic Station, University of Copenhagen. Right: Field sites, where snow observations were collected as part of my PhD project in March 2013 in Nuuk Fjord; February-April 2014 in the areas near Qaanaaq, Ilulissat, Kangerlussuaq, Nuuk, Qassiarsuk, Tasiilaq, and Zackenberg; and during May 2015 near Station Nord/Villum Station.

Inclusion of ground observations In the PhD project, I have aimed at including as many of the available meteorology and snow observations in ice-free Greenland as possible in order to validate and tie, e.g., the modeling results, to the observed value range and levels witnessed in the field. The meteorological data from GEM sites, coastal automated weather 21

stations by Danish Meteorological Institute, and Arctic Station located on Disko Island (Figure 2, left) have been included in the analyses. In addition to the GEM snow monitoring data, I have collected snow observations during three field campaigns at multiple sites across Greenland in spring 2013, 2014, and 2015 (Figure 2, right). In each site, snow-depth observations were collected along transects and snow density and stratigraphy were observed in snowpits. These observations served partly as assimilation data, for adjusting the reanalysis precipitation rates used as input to MicroMet, and for validation of SnowModel outputs.

Stepwise knowledge-gain in time and space across multiple scales

Local scale Paper II presents the time series of snow observations collected in Zackenberg (Figure 2) on local-scale defined as within a valley. These in-situ point observations, transect surveys, and spatial representations of snow were used to characterize the seasonal snow in a high-Arctic setting. The aim was to quantify the spatial variability, the temporal trends, and interannual as well as seasonal variation in the observed snow variables during the period 1997 through 2014. Furthermore, we want to investigate whether the interannual variation in snow variables had changed during the study period and whether this observed variability was extraordinary compared to variations seen in the past three decades. We found pronounced interannual variability, particularly in the timing of snow-cover onset and melt, and the annual maximum accumulation varied up to an order of magnitude between years. We found little evidence of significant trends in the observed snow- cover characteristics, apart from the snow-cover fraction registered annually on 10 June, which exhibited a significant decrease over the 18-year period. Moreover, SnowModel results for the Zackenberg region confirmed that the pronounced interannual variability in snow precipitations has persisted in this high-Arctic region since 1979 and may have masked potential temporal trends. From this paper, we gained an insight on the magnitude of the variability in snow conditions in NE-Greenland. A variability that the ecosystems are exposed to and have been able to cope with for more than three decades. Such long-term snow monitoring time series, as also emphasized in Paper I, are important since they provide an overview of the snow conditions through time, and can be used to quantify year-to-year variability, and to place single-year events or observations in a context of being (extra-)ordinary for the region (Paper IV). Despite most of the data are point observations (one-dimensional), they are valuable as assimilation and validation data in snow modeling and remote-sensing analyses (Paper V).

The second local-scale subproject presented in Paper III was based on the west coast of Greenland, in Kobbefjord, near the capital Nuuk. Here, we investigated an abrupt snowmelt event during which the majority of the snow cover depleted and the snow depth was reduced with approximate 50% within 2 days in March 2013. The aim of the paper was first to characterize the episodic snowmelt event that I observed in March 2013, in order to identify previous ESEs in recent years (2008-2013). Second, we quantified how biologically relevant snow properties, such as snowpack water content, the insulating effect of the snowpack (snow thermal resistance), and the timing of spring snowmelt (snow-free date), had changed during the identified ESEs using MicroMet and SnowModel outputs. Last, we used the acquired knowledge about the ESE characteristics to quantify the spatial distributions and trends in the occurrence of ESEs within the entire ice-free Greenland during the past 34 years (1979-2013). As a consequence of this single March 2013 ESE, we found a 50%–80% meltwater loss of the pre-melt snowpack water content, a 40%–100% loss of snow thermal resistance, and a 4-day earlier spring snowmelt snow-free date. Furthermore, the meltwater amount lost during the ESE equaled a substantial part of the annual precipitation and may potentially have advanced spring snowmelt by more than a week. The Greenland-wide frequency of ESEs showed large interannual variation, and a maximum number of ESEs was found in SW-Greenland. However, the analysis showed that ESEs are a common part of snow- cover dynamics in Greenland and should, because of their impact, be accounted for in snow-related ecosystem

and climate-change studies. This study is an example of how major snow changes promptly occur, and it was only by chance I had the unique opportunity to quantify the snow-related changes. This single-event observation is valuable since it provides an understanding of a so-far unknown mechanism. This knowledge enabled us to investigate whether these events are common in Greenland by applying modeling tools to quantify their spatial distribution across a larger extent and longer time period than the ESE was originally observed in. This was the first up-scaling effort of the PhD project.

Regional scale In Paper II, we discovered that the snow conditions in Zackenberg have been dominated by pronounced variability through time. Once we ‘zoomed out’ to the larger surrounding region, extending from the east coast to the GrIS margin, we realized that we had been studying one point in a region, which comprises strong gradients and pronounced spatial variability (Paper IV). In this study, we first mapped and quantified the inland-coast continentality gradients of temperature and snow characteristics, which are drivers of Arctic vegetation growth and phenology, using MicroMet and SnowModel within the NE-Greenland region. Herein, we also included in-situ observations of snow and meteorology to validate and verify the modeled variables. Secondly, we investigated whether the across-region gradients in the SnowModel outputs of pre-melt SWE depth and derived snow-free date were driving the spatially and temporally distributed vegetation greenness (NDVI) covering the entire domain. Finally, we evaluated the potential of the spatio-temporal highly-resolved regional gradients for being a scientific mesocosm. To do this, we compared the regional environmental gradients with large-scale gradients, as they are observed in the south-north directions along the Greenland east coast and in other Arctic regions. Our regional air temperature gradient was eight times stronger than the one observed along the east coast of Greenland. A significant inland-coast gradient was missing for the estimated maximum level of NDVI (MaxNDVI), which was relatively stable across years and strongly correlated with pre-melt SWE distributions within the region. In contrast, the timing of MaxNDVI showed a significant regional gradient and noticeable interannual variation, which was mainly driven by the timing of snowmelt (snow-free date). Due to the magnitude of the quantified gradients, this region constitutes a geographically compressed system with unique opportunities for studying biological responses to environmental drivers within relatively short distances. This means that both the pronounced temporal variability (Paper II) and spatial variation (Paper IV) in snow and vegetation greenness observed locally and regionally is also distinguished in other Greenland regions examined in Paper V. Hence, the snow and vegetation dynamics found on local- and regional scales (Paper II, Paper III, and Paper IV) are representative of large parts of ice-free Greenland.

Entire ice-free Greenland The entire ice-free area of Greenland is the multi-latitudinal study area in Paper V. For this vast area, we aim at answering the question of how climate controls the seasonal maximum vegetation greenness and timing of maximum greenness. To our knowledge, climate variables including snow amounts and derived snow characteristics have not previously been mapped in such great detail (300-m spatial increments) within this entire area. Assumed vegetation-greenness drivers including the sum of air temperature, sum of incoming shortwave radiation (Qsi), liquid precipitation (rain) during the snow-free period, and the derived snow characteristics counting pre-melt SWE depth, snow-free day of year (DOY), and the fraction of the year, where the ground is snow-free, i.e., the potential length of the growing season were quantified and mapped using MicroMet and SnowModel. The ESEs, identified within the same ice-free Greenland domain, have a huge effect on the timing of snowmelt (Paper III), and the March 2013 ESE potentially advanced the snow-free date by four days. This effect is taking into account in the Greenland-wide modeling results since SnowModel handle these satisfactory, as shown in Paper III. However, the ESEs constitute an uncertainty in the estimated 23

pre-melt SWE, since this measure of accumulated water equivalents excludes the potential ESE-related meltwater losses. Hence, the meltwater, which may be lost during ESE, is not included in pre-melt SWE, thus this measure does not comprise the entire amount of accumulated winter precipitation.

The average distributions across Greenland showed both regional contrasts from north to south, between the east and west coast, and more local differences and gradients from outer coastal areas to inland areas bordering the GrIS. These spatial patterns accentuate the complexity found within the study area and stresses the need to disentangle the question of the diversity of the different vegetation driver’s importance in various regions, e.g., in the high-Arctic NW-Greenland versus the sub-Arctic SW-Greenland. This will be investigated using multiple regression analysis across the climatically contrasting regions.

In the detailed representation of the climate and snow variables within ice-free Greenland (Paper V), inland- coast gradients, similar to the ones quantified in the NE-Greenland region as part of Paper IV, can also be identified in other Greenland regions. These are particularly visible in W-Greenland in areas with the longest distance across ice-free land from the coast to the GrIS margin. In addition, snow distributions and derived snow variables vary spatially with elevation increase from valley bottoms to the mountain tops on local scale, to the same degrees as they vary on regional scale, e.g., in a west-to-east or north-to-south direction in Greenland. This is similar to the mesocosm concept described in Paper IV. Indeed the variability found on local scale accentuates a part of the variability found on larger regional scale. This confirms the importance of local-scale observations, e.g., collected at the GEM sites, since they have extensive explanatory power exceeding their domain, in which they are observed.

Looking into the future The previous four papers (Paper II - Paper V) have demonstrated that MicroMet and SnowModel are powerful modeling tools for temporally and spatially distributing snow at multiple scales and resolutions mainly looking back in time. However, as part of the last subproject, we demonstrate how these tools can be coupled with additional models to predict changes. The last subproject advance in complexity by extending into the future with its climate-change scenario simulations forward to year 2100 in 10-m spatial resolution. We present simulations for the ZERO-line transect, also included in Paper II, which extents 4 km across the lower parts of Zackenberg Valley. Using SnowModel snow-depth distributions as input to the ground thermal model CryoGrid 2 (Westermann et al. 2013) together with down-scaled RCM HIRHAM5 input (Christensen et al. 1996), we estimated the range of ground thermal conditions predicted until the end of the century. We find that the model outputs match the observed range in maximum active layer thickness between vegetation classes and that the modeled ground temperatures agree with observed borehole temperatures. For the predictions the modeled average ground temperatures increase towards 2100, but remain in general below freezing. However in a few years of the future simulation, the CryoGrid model sensitivity to snow depth result in above-zero average soil temperatures in locations with snow drifts, which in theory would result in sporadic permafrost degradation in the NE-Greenland region.

Perspectives and Concluding remarks In Greenland, the research and knowledge about the ice-free ecosystem’s dynamics and processes have increased significantly within the past four years of this PhD project. A great deal of the research is executed in one single or in a few sites, which are easily accessible. However, knowledge gaps remain between these sites. The results of my PhD project are exactly able to fill the gaps both spatially and temporally using the

temporal and spatial distribution and quantifications of snow variables, which are unprecedented in ice-free Greenland. Especially, the high resolution data, both in time and space that are inputted to our analyses, are unique since these are observed, modeled, analyzed, and presented at the ecologically relevant scales.

One of the major forces of this PhD project is that the model simulations are tied to ground-based observations - both in the validation and the development of methods, e.g., in the NDVI processing (Paper IV and Paper V) or, e.g., by assimilating SWE observations for adjusting reanalysis precipitation rates and to include automated weather station data as MicroMet inputs (Paper III-V). The inclusion of observations has been of highest priority in order to model realistic representations of snow and climate variables.

It is furthermore rare for a project to be able to connect both biotic and abiotic data sets in order to map interactions. This PhD project is an example of an interdisciplinary approach, which has been proven to be successful in explaining more than single-discipline research. Moreover, the scale-dependent modeling framework of combining observations and modeling tools is requested by both the snow and biology research community and is easily applicable in other seasonally snow-covered regions.

I have demonstrated that interactions between abiotic and biotic ecosystem components are possible by using all available observations in ice-free Greenland in combination with the dynamical and flexible MicroMet and SnowModel modeling tools. They are able to handle the complex and unstable nature of the seasonal snow cover, from which ecologically relevant variables can be derived at scales and resolutions appropriated for the process or dynamics being studied. Finally, the findings presented in this PhD thesis, and the model data as well as the observational data obtained during PhD project, provide new opportunities for projects previously not possible. It is my hope that these data, distributed in time and space, may lay the foundation for future studies of more dynamical character in Greenland, such as ungulate movement, bird migration patterns, or gas flux dynamics, all investigable within a physically realistic framework.

Ilulissat, W-Greenland (2014) 25

References

Barton, M. and M. Burke. 1977. SNOTEL: An Operational Data Aquisition System Using Meteor Burst Technolog. Pages pp. 82-84. in Proceedings of the 45th Annual Western Snow Conference. Colorado State University, Albuquerque, New Mexico. Bokhorst, S., J. W. Bjerke, L. E. Street, T. V. Callaghan, and G. K. Phoenix. 2011. Impacts of multiple extreme winter warming events on sub-Arctic heathland: phenology, reproduction, growth, and CO2 flux responses. Global Change Biology 17:2817-2830. Bokhorst, S., S. H. Pedersen, L. Brucker, O. Anisimov, J. W. Bjerke, R. D. Brown, D. Ehrich, R. L. H. Essery, A. Heilig, S. Ingvander, C. Johansson, M. Johansson, I. S. Jónsdóttir, N. Inga, K. Luojus, G. Macelloni, H. Mariash, D. McLennan, G. N. Rosqvist, A. Sato, H. Savela, M. Schneebeli, A. Sokolov, S. A. Sokratov, S. Terzago, D. Vikhamar-Schuler, S. Williamson, Y. Qiu, and T. V. Callaghan. 2016. Changing Arctic snow cover: A review of recent developments and assessment of future needs for observations, modelling, and impacts. Ambio 45:516- 537. Brooks, P. D., P. Grogan, P. H. Templer, P. Groffman, M. G. Öquist, and J. Schimel. 2011. Carbon and Nitrogen Cycling in Snow‐Covered Environments. Geography Compass 5:682-699. Brown, R. D. and P. W. Mote. 2009. The Response of Northern Hemisphere Snow Cover to a Changing Climate. Journal of Climate 22:2124-2145. Buckeridge, K. M. and P. Grogan. 2008. Deepened snow alters soil microbial nutrient limitations in arctic birch hummock tundra. Applied Soil Ecology 39:210-222. Callaghan, T. V., M. Johansson, R. D. Brown, P. Y. Groisman, N. Labba, V. Radionov, R. S. Bradley, S. Blangy, O. N. Bulygina, T. R. Christensen, J. E. Colman, R. L. H. Essery, B. C. Forbes, M. C. Forchhammer, V. N. Golubev, R. E. Honrath, G. P. Juday, A. V. Meshcherskaya, G. K. Phoenix, J. Pomeroy, A. Rautio, D. A. Robinson, N. M. Schmidt, M. C. Serreze, V. P. Shevchenko, A. I. Shiklomanov, A. B. Shmakin, P. Skold, M. Sturm, M. K. Woo, and E. F. Wood. 2011. Multiple Effects of Changes in Arctic Snow Cover. Ambio 40:32-45. Christensen, J. H., O. B. Christensen, P. Lopez, E. van Meijgaard, and M. Botzet. 1996. The HIRHAM 4 regional atmospheric climate model. Danish Meteorological Institute, Denmark. Collins, W. and T. Smith. 1991. Effects of wind-hardened snow on foraging by reindeer (Rangifer tarandus). Arctic:217- 222. Déry, S. J. and R. D. Brown. 2007. Recent Northern Hemisphere snow cover extent trends and implications for the snow- albedo feedback. Geophysical Research Letters 34. Elberling, B. 2007. Annual soil CO 2 effluxes in the High Arctic: the role of snow thickness and vegetation type. Soil Biology and Biochemistry 39:646-654. Goodrich, L. 1982. The influence of snow cover on the ground thermal regime. Canadian geotechnical journal 19:421- 432. Greene, E. M., G. E. Liston, and R. A. Pielke. 1999. Simulation of above treeline snowdrift formation using a numerical snow-transport model. Cold Regions Science and Technology 30:135-144. Groisman, P. Y., T. R. Karl, and R. W. Knight. 1994. Observed Impact of Snow Cover on the Heat Balance and the Rise of Continental Spring Temperatures. Science 263:198-200. Hall, D. K., G. A. Riggs, V. V. Salomonson, N. E. DiGirolamo, and K. J. Bayr. 2002. MODIS snow-cover products. Remote Sensing of Environment 83:181-194. Hancock, S., R. Baxter, J. Evans, and B. Huntley. 2013. Evaluating global snow water equivalent products for testing land surface models. Remote Sensing of Environment 128:107-117. Hansen, B. U., B. Elberling, O. Humlum, and N. Nielsen. 2006. Meteorological trends (1991–2004) at Arctic Station, Central West Greenland (69°15'N) in a 130 years perspective. Geografisk Tidsskrift-Danish Journal of Geography 106:45-55. Hasholt, B., G. E. Liston, and N. T. Knudsen. 2003. Snow-distribution modelling in the Ammassalik region, south east Greenland. Nordic Hydrology 34:1-16. Hiemstra, C. A., G. E. Liston, and W. A. Reiners. 2006. Observing, modelling, and validating snow redistribution by wind in a Wyoming upper treeline landscape. Ecological Modelling 197:35-51. Hinkler, J., S. B. Pedersen, M. Rasch, and B. U. Hansen. 2002. Automatic snow cover monitoring at high temporal and spatial resolution, using images taken by a standard digital camera. International Journal of Remote Sensing 23:4669-4682. Johansson, M., T. V. Callaghan, J. Bosiö, H. J. Åkerman, M. Jackowicz-Korczynski, and T. R. Christensen. 2013. Rapid responses of permafrost and vegetation to experimentally increased snow cover in sub-arctic Sweden. Environmental Research Letters 8:035025.

Jones, H. G. 1999. The ecology of snow-covered systems: a brief overview of nutrient cycling and life in the cold. Hydrological Processes 13:2135-2147. Kurtz, N. T., S. L. Farrell, M. Studinger, N. Galin, J. P. Harbeck, R. Lindsay, V. D. Onana, B. Panzer, and J. G. Sonntag. 2013. Sea ice thickness, freeboard, and snow depth products from Operation IceBridge airborne data. The Cryosphere 7:1035-1056. Liston, G. E. and K. Elder. 2006a. A distributed snow-evolution modeling system (SnowModel). Journal of Hydrometeorology 7:1259-1276. Liston, G. E. and K. Elder. 2006b. A meteorological distribution system for high-resolution terrestrial modeling (MicroMet). Journal of Hydrometeorology 7:217-234. Liston, G. E., R. B. Haehnel, M. Sturm, C. A. Hiemstra, S. Berezovskaya, and R. D. Tabler. 2007. Instruments and methods simulating complex snow distributions in windy environments using SnowTran-3D. Journal of Glaciology 53:241-256. Liston, G. E. and C. A. Hiemstra. 2008. A simple data assimilation system for complex snow distributions (SnowAssim). Journal of Hydrometeorology 9:989-1004. Liston, G. E. and C. A. Hiemstra. 2011. The Changing Cryosphere: Pan-Arctic Snow Trends (1979-2009). Journal of Climate 24:5691-5712. Liston, G. E., J. P. McFadden, M. Sturm, and R. A. Pielke. 2002. Modelled changes in arctic tundra snow, energy and moisture fluxes due to increased shrubs. Global Change Biology 8:17-32. Liston, G. E. and M. Sturm. 1998. A snow-transport model for complex terrain. Journal of Glaciology 44:498-516. Liston, G. E. and M. Sturm. 2002. Winter precipitation patterns in arctic Alaska determined from a blowing-snow model and snow-depth observations. Journal of Hydrometeorology 3:646-659. Liston, G. E., J. G. Winther, O. Bruland, H. Elvehoy, K. Sand, and L. Karlof. 2000. Snow and blue-ice distribution patterns on the coastal Antarctic Ice Sheet. Antarctic Science 12:69-79. Marks, D. and J. Dozier. 1992. Climate and energy exchange at the snow surface in the alpine region of the Sierra Nevada: 2. Snow cover energy balance. Water Resources Research 28:3043-3054. Mernild, S. H., G. E. Liston, and B. Hasholt. 2007. Snow-distribution and melt modelling for glaciers in Zackenberg river drainage basin, north-eastern Greenland. Hydrological Processes 21:3249-3263. Mernild, S. H., G. E. Liston, C. A. Hiemstra, J. K. Malmros, J. C. Yde, and J. McPhee. 2016. The Andes Cordillera. Part I: snow distribution, properties, and trends (1979–2014). International Journal of Climatology. Post, E., M. C. Forchhammer, M. S. Bret-Harte, T. V. Callaghan, T. R. Christensen, B. Elberling, A. D. Fox, O. Gilg, D. S. Hik, T. T. Hoye, R. A. Ims, E. Jeppesen, D. R. Klein, J. Madsen, A. D. McGuire, S. Rysgaard, D. E. Schindler, I. Stirling, M. P. Tamstorf, N. J. Tyler, R. van der Wal, J. Welker, P. A. Wookey, N. M. Schmidt, and P. Aastrup. 2009. Ecological dynamics across the Arctic associated with recent climate change. Science 325:1355-1358. Pudas, E., A. Tolvanen, J. Poikolainen, T. Sukuvaara, and E. Kubin. 2008. Timing of plant phenophases in Finnish Lapland in 1997-2006. Boreal environment research 13:31-43. Ramsay, B. H. 1998. The interactive multisensor snow and ice mapping system. Hydrological Processes 12:1537-1546. Randin, C. F., G. Vuissoz, G. E. Liston, P. Vittoz, and A. Guisan. 2009. Introduction of Snow and Geomorphic Disturbance Variables into Predictive Models of Alpine Plant Distribution in the Western Swiss Alps. Arctic Antarctic and Alpine Research 41:347-361. Rasmussen, R., B. Baker, J. Kochendorfer, T. Meyers, S. Landolt, A. P. Fischer, J. Black, J. M. Thériault, P. Kucera, and D. Gochis. 2012. How well are we measuring snow: The NOAA/FAA/NCAR winter precipitation test bed. Bulletin of the American Meteorological Society 93:811-829. Rogers, M. C., P. F. Sullivan, and J. M. Welker. 2011. Evidence of Nonlinearity in the Response of Net Ecosystem CO2 Exchange to Increasing Levels of Winter Snow Depth in the High Arctic of Northwest Greenland. Arctic, Antarctic, and Alpine Research 43:95-106. Schimel, J. P., C. Bilbrough, and J. M. Welker. 2004. Increased snow depth affects microbial activity and nitrogen mineralization in two Arctic tundra communities. Soil Biology and Biochemistry 36:217-227. Stiegler, C., M. Lund, T. R. Christensen, M. Mastepanov, and A. Lindroth. 2016. Two years with extreme and little snowfall: effects on energy partitioning and surface energy exchange in a high-Arctic tundra ecosystem. The Cryosphere 10:1395-1413. Stuefer, S., D. L. Kane, and G. E. Liston. 2013. In situsnow water equivalent observations in the US Arctic. Hydrology Research 44:21. Sturm, M., J. Holmgren, M. Konig, and K. Morris. 1997. The thermal conductivity of seasonal snow. Journal of Glaciology 43:26-41. Tyler, N. J. C. 2010. Climate, snow, ice, crashes, and declines in populations of reindeer and caribou (Rangifer tarandus L.). Ecological Monographs 80:197-219. Westermann, S., T. V. Schuler, K. Gisnas, and B. Etzelmuller. 2013. Transient thermal modeling of permafrost conditions in Southern Norway. Cryosphere 7:719-739. 27

Wiens, J. A. 1989. Spatial scaling in ecology. Functional ecology 3:385-397. Wu, H. and Z.-L. Li. 2009. Scale issues in remote sensing: A review on analysis, processing and modeling. Sensors 9:1768-1793. Zeng, H. and G. Jia. 2013. Impacts of snow cover on vegetation phenology in the arctic from satellite data. Advances in Atmospheric Sciences 30:1421-1432. Zhang, T. 2005. Influence of the seasonal snow cover on the ground thermal regime: An overview. Reviews of Geophysics 43.

South of Upernavik, NW-Greenland (2014)

Paper I

Changing Arctic snow cover: A review of recent developments and assessment of future needs for observations, modelling, and impacts.

Bokhorst, S., S. H. Pedersen, L. Brucker, O. Anisimov, J. W. Bjerke, R. D. Brown, D. Ehrich, R. L. H. Essery, A. Heilig, S. Ingvander, C. Johansson, M. Johansson, I. S. Jónsdóttir, N. Inga, K. Luojus, G. Macelloni, H. Mariash, D. McLennan, G. N. Rosqvist, A. Sato, H. Savela, M. Schneebeli, A. Sokolov, S. A. Sokratov, S. Terzago, D. Vikhamar-Schuler, S. Williamson, Y. Qiu, and T. V. Callaghan. 2016. Changing Arctic snow cover: A review of recent developments and assessment of future needs for observations, modelling, and impacts. Ambio. Vol. 45, pp. 516-537. DOI:10.1007/s13280-016-0770-0.

North of Qaanaaq, NW-Greenland (2014)

Ambio 2016, 45:516–537 DOI 10.1007/s13280-016-0770-0

REVIEW

Changing Arctic snow cover: A review of recent developments and assessment of future needs for observations, modelling, and impacts

Stef Bokhorst , Stine Højlund Pedersen, Ludovic Brucker, Oleg Anisimov, Jarle W. Bjerke, Ross D. Brown, Dorothee Ehrich, Richard L. H. Essery, Achim Heilig, Susanne Ingvander, Cecilia Johansson, Margareta Johansson, Ingibjo¨rg Svala Jo´nsdo´ttir, Niila Inga, Kari Luojus, Giovanni Macelloni, Heather Mariash, Donald McLennan, Gunhild Ninis Rosqvist, Atsushi Sato, Hannele Savela, Martin Schneebeli, Aleksandr Sokolov, Sergey A. Sokratov, Silvia Terzago, Dagrun Vikhamar-Schuler, Scott Williamson, Yubao Qiu, Terry V. Callaghan

Received: 29 October 2015 / Revised: 3 November 2015 / Accepted: 5 February 2016 / Published online: 17 March 2016

Abstract Snow is a critically important and rapidly seasons but are likely to be affected by climate warming changing feature of the Arctic. However, snow-cover and with unexpected impacts for ecosystems and society. For snowpack conditions change through time pose challenges example, Arctic snow-cover duration is decreasing for measuring and prediction of snow. Plausible scenarios rapidly (*3–5 days/decade), particularly due to earlier of how Arctic snow cover will respond to changing Arctic spring melt (20 %/decade) and later onset of snow cover climate are important for impact assessments and (Derksen et al. 2015). However, the Eurasian Arctic adaptation strategies. Although much progress has been region has experienced larger declines in the duration of made in understanding and predicting snow-cover changes the snow-covered period (12.6 days), i.e. prolonged veg- and their multiple consequences, many uncertainties etation growing season, compared to the North American remain. In this paper, we review advances in snow Arctic region (6.2 days) between 1982 and 2011 (Bar- monitoring and modelling, and the impact of snow ichivich et al. 2013). In addition, climate warming changes on ecosystems and society in Arctic regions. increases the potential for unseasonal thaws, early Interdisciplinary activities are required to resolve the snowmelt, and rain-on-snow events (ROS) (Liston and current limitations on measuring and modelling snow Hiemstra 2011). These changes impact snow properties characteristics through the cold season and at different and runoff (Semmens et al. 2013), which in turn affect spatial scales to assure human well-being, economic Arctic ecosystems and societies (Meltofte 2013; Cooper stability, and improve the ability to predict manage and 2014; Hansen et al. 2014). However, changes in snow adapt to natural hazards in the Arctic region. properties are not uniform across the Arctic and affected processes operate/respond at different temporal and spa- Keywords Climate change Á Ecosystem services Á tial scales. Moreover, the various disciplines working on Human health Á Societal costs Á Indigenous Á Snow snow measure and evaluate its properties at different temporal and spatial scales. Therefore, there are potential mismatches on the availability and requirements of snow INTRODUCTION data between snow scientists, modellers, ecologists, and sociologists. Snow is a critically important element of the Arctic and To address these issues, an interdisciplinary workshop is rapidly changing due to climate warming (Callaghan was held to develop a road map to improve measurement, et al. 2011). Snow cover, stratigraphy, and physical modelling, and prediction of changing snow characteristics characteristics are naturally changing throughout the and to collate developments in the ﬁeld since the ‘‘Snow Water Ice and Permafrost in the Arctic’’ assessment of 2011(Callaghan et al. 2011). This paper builds on the results presented at the workshop and presents an overview Electronic supplementary material The online version of this of recent developments in studies of changing Arctic snow article (doi:10.1007/s13280-016-0770-0) contains supplementary material, which is available to authorized users. cover and its consequences.

Ó The Author(s) 2016. This article is published with open access at Springerlink.com 123 www.kva.se/en Ambio 2016, 45:516–537 517

UNDERSTANDING THE IMPACTS OF CHANGING affect the water supply for aquatic ecosystems, forestry, SNOW CONDITIONS ON SOCIETIES and agriculture (Jeelani et al. 2012; Clarke et al. 2015). AND ECOSYSTEMS The increasingly wetter and milder Arctic climate can lead to increased frequency of avalanches threatening Economy, human health, and well-being growing populations and infrastructure (Eckerstorfer and Christiansen 2012; Qiu 2014). When comparing snow The direct impact of snow temporal and spatial variability avalanche risk assessments between regions, losses are on economic development of the Arctic has to our often associated with an increase in land use, population knowledge not been comprehensively evaluated and density, and economic activities (Shnyparkov et al. 2012). quantified. Such a study would need to take into account Healthcare costs can rise due to increasing occurrence of among others: Snow clearing costs of transportation routes bone fractures resulting from unusual snow and ice con- (Hanbali 1994; Riehm and Nordin 2012) (Fig. 1), which ditions (Bjerke et al. 2015). Snow can also become a health varies annually and is complicated by extreme snowfalls issue when supporting biological pathogens (Biedunkie- (Borzenkova and Shmakin 2012). The prevention of wicz and Ejdys 2011; Shen and Yao 2013; Simon et al. freezing damage to water pipes and drainage systems 2013; Ejdys et al. 2014). The impacts of changing snow- (Bjerke et al. 2015). Associated risks to winter-crops and melt dynamics on snow-pathogens for humans, livestock, forestry production due to changes in snow-season duration and agriculture are unclear (Parham et al. 2015). (Hanewinkel et al. 2011; Krenke et al. 2012), increased frequency of desiccation, exposure to snow moulds (Mat- Ecosystems sumoto and Hoshino 2009), and encasement in ground ice (Bjerke et al. 2014, 2015). Furthermore, ice-based con- Snow cover is an important determinant of community and struction procedures relying on firn-ice (e.g. winter roads) ecosystem structure in polar regions (AMAP 2011) and can be affected (Sosnovsky et al. 2014). Seasonal snow winter temperatures are increasing in the Arctic more than conditions are crucial for the way of life of indigenous those during summer (Walsh 2014). However, impacts of people and local residents for reindeer herding practices changing winter climate and snow regimes have received and access to hunting grounds (Riseth et al. 2011), harvest much less attention compared to the effects of climate yields of cultivated and wild berries (Bokhorst et al. 2011; change during summer. Different aspects of the snowpack Niemi and Ahlstedt 2012), and game animals (Stien et al. play crucial roles in ecosystem processes and the life of 2012; Hansen et al. 2013). Snow-season duration and Arctic organisms (e.g. Cooper 2014). Relevant snowpack snow-cover depth also affect the economy through changes characteristics include thermal insulation, snow depth, in the magnitude and timing of spring runoff and floods. In microstructure, temporal changes of these aspects, as well Siberia, the frequency of dangerous river ice jams and as snow-cover duration, all of which have been shown to be spring river flooding events are increasing (Popova 2011; affected by climate change, with important consequences Semenov 2013), while decreased snow precipitation will for Arctic ecosystems (AMAP 2011).

Fig. 1 Increases in heavy snowfall affect the function of cities above the Arctic Circle. Snow clearance (left) has economic costs, whereas lack of snow clearance (right) can perhaps have even greater costs (left Kirovsk and right Norilsk: photos M.N. Ivanov)

Ó The Author(s) 2016. This article is published with open access at Springerlink.com www.kva.se/en 123 518 Ambio 2016, 45:516–537

Terrestrial ecosystems snowpack can therefore affect the freezing regime, having consequences for the freshwater ecosystem with feedbacks Snow acts as an insulating blanket against freezing Arctic to habitat structure, food availability, and survival of spe- temperatures for many organisms. Snow is also a major cies (Prowse and Brown 2010; Prowse et al. 2011; Surdu determinant of the mosaic of ecological communities et al. 2014). For shallow waters (\3 m) and wetlands, the through its uneven landscape distribution and the influence timing and duration of ice defines the open water, produc- of snowmelt-driven spring flooding on wetland communi- tive period and limits the active state of aquatic organisms ties. Changes in snow quantity, quality, and seasonality by freezing to the bottom. Winter-dormancy allows species can, therefore, result in changes in the distribution and to survive such frozen conditions but the breaking of winter- composition of Arctic communities with resulting effects dormancy depends on the photoperiod and temperature on their many inherent ecological processes, functions, and (Dupuis and Hann 2009) which is affected by the snow feedbacks. Extreme weather events (unseasonal warm cover. Particularly the formation of ‘white ice’, formed temperatures and ROS see Fig. 2) can cause complete loss when the snowpack exceeds the buoyance of the ice, affects of snow cover, changes in the snow stratigraphy, snow the light transfer to the water column below (Dibike et al. hardness, and formation of ice layers with great impacts on 2012). Changing snow conditions affecting freshwater plants (Bokhorst et al. 2011; Preece et al. 2012), herbivores freezing and melting conditions may cause mismatches for (Bartsch et al. 2010; Ims et al. 2011; Stien et al. 2012; organisms in terms of when winter-dormancy ends com- Bilodeau et al. 2013), soil organisms and CO2 fluxes pared to peak food availability. Ecosystem phenology (Bokhorst et al. 2012, 2013), and agriculture (Bjerke et al. associated with ice and snow cover in freshwater systems is 2014, 2015). However, species responses to extreme an area that needs more research. weather events and snowmelt are dependent on the timing Spring snowmelt is also an important conduit for trans- of events (Bokhorst et al. 2010, 2011), while the mecha- porting organic matter from the land into rivers and lakes. nisms behind species responses are unclear (Rumpf et al. This pulse of organic matter into freshwater affects the 2014; Bowden et al. 2015) and processes are often inferred clarity (light attenuation), nutrient and carbon cycling, pri- based on indirect correlative information (e.g. Ims et al. mary productivity, and overall food web dynamics of aquatic 2011). Furthermore, changing snow conditions can have ecosystems (Ask et al. 2009; Rautio et al. 2011). Further- wide-ranging indirect effects mediated by ecological more, dissolved and suspended concentrations of metals are interactions. For instance, shrub growth affects snow highest in rivers and lakes during the spring freshet (Hole- accumulation which in turn influences soil temperatures mann et al. 2005) indicating that the snowpack acts as a and ecosystem process rates (Myers-Smith and Hik 2013) reservoir for contaminants that are released as a pulse highlighting the importance of interactions between vege- (Douglas et al. 2012). The timing of mercury (Hg) runoff, for tation structure and snow properties. Snow-induced chan- example, is greatly affected by the spatial variability in hill- ges in mortality and dynamics of reindeer and lemming slope flow paths and the magnitude of snowmelt inputs (Hansen et al. 2013) affect predator populations (Schmidt (Haynes and Mitchell 2012) indicating that predictions of et al. 2012) which in turn may shift to alternative prey mercury runoff in water streams need to be developed at (McKinnon et al. 2013; Nolet et al. 2013). These examples small scales and that up-scaling will be challenging. highlight the need to identify critical periods when species and ecosystems are vulnerable to winter climate change, Sea ice and snow especially with regard to periods of snowpack build-up, ROS and ground icing, and spring snowmelt. Variations in snow-covered sea ice affect the Earth’s cli- Aside from the species-specific and ecosystem responses mate by affecting ocean–atmosphere interactions. Snow to changing snow conditions, there is a major research cover on top of sea ice has a high albedo that dominates the challenge in linking the predictions of snow changes to the surface solar energy exchange, and a changing thermal scales that are relevant for the organisms or ecosystem that conductivity that regulates ice/atmosphere heat transfer is being studied (Table 1). Specifically, there is a need for that greatly modifies the sea ice thermodynamic processes. accurate predictions of the build-up and change in the snow The snow cover also modifies surface roughness with stratigraphy across scales of a few square metres to land- implications for the ice/air drag coefficient and sensible scapes covering several km2. and latent heat fluxes. Snow depth and snow properties (e.g. thermal conductivity and density) on sea ice are thus Freshwater systems of crucial importance, and must be accurately retrieved on a large scale. Snow on lake and river ice affects the temperature and light Snow across sea ice influences algal communities with transmission to the underlying ice and water. Changes in the thin snow cover promoting productivity in the ocean

Ó The Author(s) 2016. This article is published with open access at Springerlink.com 123 www.kva.se/en Ambio 2016, 45:516–537 519

Fig. 2 Examples of changing snow conditions in terrestrial ecosystems: a Vegetation captured in ice layer following rain-on-snow event leading to b mortality among reindeer (Yamal Russia) and c delayed breeding of Black-bellied Plover (Pluvialis squatarola) (Southampton Island, Nunavut, Canada); d Muskoxen (Ovibos moschatus) grazing at high elevation to ﬁnd snow-free patches during spring 2012, Zackenberg in Northeast Greenland; e Experimental simulation of extreme winter warming near Tromsø (Norway). Photos a and b Aleksandr Sokolov, c K. Young, d S. Højlund Pedersen, and e S. Bokhorst

(Alou-Font et al. 2013). This suggests that reduced snow Teleconnections and snow cover in Arctic precipitation or quicker melt out may promote higher pri- ampliﬁcation mary production underneath sea ice with potential positive impacts higher up the food chain. Conversely, snow-cover Research has been dedicated to investigate the linkages removal from the sea ice surface can inhibit spring growth between the changing Arctic snow cover and tropospheric of Arctic ice algae through physiological and behavioural processes (Cohen et al. 2014) and the impacts of Arctic effects (Lund-Hansen et al. 2014). ampliﬁcation to temperature variability at low and high

Ó The Author(s) 2016. This article is published with open access at Springerlink.com www.kva.se/en 123 520 Ambio 2016, 45:516–537

Table 1 Overview of the various expected changes in snow conditions, affected groups of organisms, processes, or activities and the modelling requirements that are required to predict their occurrence in the near future. The different affected groups, processes, and/or activities have different spatial and temporal extent and resolution; hence models are required to resolve these speciﬁc spatial and temporal dimensions Changes in climate Affected groups/processes Modelling requirements to predict these changes Scale and snow

Temperature Soil organisms, dwarf shrubs, cryptogams Snow depth, snow density, snow type, stratigraphy, 0–1 m2 variability under and temporal evolution of these through the cold 2 Ecosystem CO2 fluxes 0–1 m the snow (snow season Shrubs and trees 1–10 m2 insulation) Ice-layer formation Humans, sub-Arctic agroecosystems, vegetation, Timing, duration/longevity, compactness, and spread 1–10 m2 small rodents, reindeer, and species depending on of (ground) ice formation across the landscape, in and them through direct or indirect trophic interactions urban areas, and on transportation infrastructure [km2 (roads, airports, culverts) Avalanche risk Society, infrastructure, large grazers, and Snow stratigraphy/stability through the cold season 100 m2 mountainside vegetation, especially trees Snow accumulation Infrastructure/society, water supply, large grazers and Snow depth, snow water equivalent, timing of heavy \100 m2 flooding risk snowfall events, and snow (re-)distribution by wind Snow-cover duration Agriculture, freshwater ecosystems, terrestrial Snow depth, timing of snow deposition and \100 m2 and timing ecosystems, energy use, northern food security, snowmelt, and resultant sea ice melt out transportation, and recreation latitudes (Francis and Vavrus 2012; Screen 2014). accurate snow data at different spatial and temporal reso- Declining terrestrial spring snow cover in the Arctic is lutions to address the challenges of changing snow condi- contributing to Arctic amplification (Serreze and Barry tions. We present an overview of recent advances in 2011; Matsumura et al. 2014). Changing snow on fresh- methods for quantifying and monitoring snow variables, water systems affect local climate conditions (Rouse et al. and a summary of widely used ground-based snow obser- 2008; Brown and Duguay 2010). Observations of Arctic vational methods is presented in Table 2. In addition, we sea ice reduction in autumn are shown to be causing cold indicate data/knowledge gaps where progress is required in extremes (e.g. additional snowfall) in mid-altitude and terms of spatial and temporal resolution of snow variables. northern continents/sub-Arctic areas (Cohen et al. 2013; Tang et al. 2013). Arctic amplification depends on heat- Overview of recent advances in methods transport from lower latitudes but local factors on surface and findings in Arctic snow monitoring warming is still a matter of debate because it is difficult to isolate local forcings from simultaneously occurring Ground-based snow-depth monitoring external forcings and feedbacks (Screen and Simmonds 2012). Furthermore, high-latitude responses in the multiple Several well-known methods for measuring snow depth types of forcing between models were broad, making it exist (Table 2). Recent developments in snow-depth mea- difficult to define the particular causes of Arctic tempera- surements include remote sensing methods that enable an ture amplification (Crook et al. 2011). Improved process objective monitoring of spatial distributions of snow depth. understanding, additional Arctic observations, and further These methods include polarimetric phase differences modelling efforts in collaboration with observation data are (Leinss et al. 2014), ground-based laser scans (Deems et al. required to elucidate the teleconnections with the Arctic 2013), and electromagnetic wave technology (e.g. Koch (Cohen et al. 2014). et al. 2014; McCreight et al. 2014).

Spaceborne snow-cover monitoring OBSERVATIONS OF CHANGING SNOW CONDITIONS Snow-cover has high spatial and temporal variability and satellites provide observations at the hemispherical scale. Quantifying snow-cover extent, thickness, and speciﬁc Both passive and active remote sensing methods are used snow characteristics in the Arctic is challenging mainly due with sensors operating in the visible and microwave to the inclement weather conditions, polar night, and domains. Visible sensors observe snow-surface properties redistribution of snow by wind. In addition, the limited (with solar illumination, in cloud-free conditions), and are Arctic snow-observation stations challenge the up-scaling used for mapping snow-cover extent (e.g. Hall et al. 2002, process to larger regions. However, there is a great need for 2006). Microwave sensors are sensitive to snow properties,

Ó The Author(s) 2016. This article is published with open access at Springerlink.com 123 www.kva.se/en Ambio 2016, 45:516–537 521

Table 2 Overview of observation methods in quantifying various snow parameters Target parameter(s) Method(s) Reference(s)

Destructive ground-based snow observations Snow depth Simple (avalanche) or semi-automated probes (e.g. e.g. Sturm et al. (2006) MagnaProbe) Specific surface area (SSA) (i.e. Near-infrared photography and infrared reflectance e.g. Matzl and Schneebeli (2006), Gallet et al. the surface area of ice per unit methods (2009) Arnaud et al. (2011), and Montpetit mass) et al. (2012) Penetration resistance and SnowMicroPen (Highly resolved measurements (250 Schneebeli and Johnson (1998) and Proksch deviation of snow density, measurements/mm) et al. (2015) grain parameters, and SSA. Snowfall/new snow Snow board (i.e. new-snow observations are being e.g. Fierz et al. (2009) conducted by placing a board (snow board) on the snow surface and revisiting it every 24 h to read the additional snow height Liquid water content in snow ‘Denoth capacity probe’ or ‘Finnish Snow Fork’ (e.g. Denoth (1994) and Sihvola and Tiuri (1986) used to deriving dielectric/conduction properties of the snow) Non-destructive ground-based snow observations Snow depth Acoustic snow-depth sensors, ultrasonic methods, lasers, manual readings at stakes, and automatic readings utilizing time-lapse cameras Snow density and snow bulk Upward-looking ground penetrating radar (upGPR) e.g. Mitterer et al. (2011), Avanzi et al. (2014), liquid water content Combination of upGPR with buried GPS sensors (allows Heilig et al. (2015), Schmid et al. (2014, for direct conversion for density, SWE and liquid water 2015), and Stacheder (2005) content) Time domain reflectometer (TDR) Snow water equivalent (SWE) Snow pillows or snow scales weigh the mass of the snowpack above the sensors and convert this to SWE Snow albedo Net radiometer e.g. Michel et al. (2008) Snow-cover fraction Derived from hourly-daily digital photos acquired from e.g. Bernard et al. (2013) automatic time-lapse digital cameras installed in terrestrial areas, e.g. near glaciers and ice fields Avalanche hazard and activity Seismic sensor Reiweger et al. (2015) Infrasound arrays e.g. Van Herwijnen and Schweizer (2011), Havens et al. (2014)

and operate independently from solar illumination with a above the snow cover have been used for deriving snow weak sensitivity to the atmosphere. The main limitation of depth, density, bulk liquid water content, and for deriving using microwave radiometers is the coarse resolution (i.e. SWE (Heilig et al. 2009; Schmid et al. 2014) and allow tens of kilometres), whereas radars lack the appropriate monitoring of the temporal evolution of the overlying frequencies. Existing radar sensors, which can provide snow. In addition, recent advances in SWE quantification information on snow-cover with fine resolution, are able to have shown the benefit of combining passive microwave work only in the presence of wet snow. radiometer and ground-based synoptic weather station observations to provide robust information on hemispher- Snow water equivalent (SWE) ical scale (Takala et al. 2011). Mobile measurements allow for monitoring spatial differences in SWE or liquid water Satellite algorithms have been developed to monitor SWE content but only provide snapshots in time. Hence, there at the hemispherical scale since the 1980s (e.g. Kelly are major challenges to compare satellite-derived infor- 2009). In the early 2000s, surface-based Frequency-Mo- mation with ground-based in situ data. In addition, further dulated Continuous-Wave (FMCW) radar measurements development on sensors for satellites and aircrafts is nec- were used to estimate SWE to within 5 % (Marshall et al. essary including new technologies for data interpretation 2005). Furthermore, fixed radars installed underneath or together with up-scaling methods for temporal continuous

Ó The Author(s) 2016. This article is published with open access at Springerlink.com www.kva.se/en 123 522 Ambio 2016, 45:516–537 point measurements. Further investigations are required to responses to deposits of black carbon (BC) on the snow convert satellite observations into accurate SWE retrievals surface are shown to cause accelerating snowmelt rates in and remote sensing of SWE is currently restricted to flat Alaska, Norway, and Greenland (Doherty et al. 2013). areas thereby excluding mountains. Particle size of snow impurities can be used to identify their source and have been linked to peripheral snow-free Snow microstructure (grain size, snow-specific surface areas or locations with early snowmelt and fires (Aoki et al. area) and liquid water content (LWC) 2014; Dumont et al. 2014). A decreasing snow-cover extent may play a major role in the surface mass balance of Arctic Snow microstructure is complex, but can be characterized ice bodies. by snow-specific surface area (SSA). SSA controls the snow albedo and is a more objective measure of snow’s Snow on sea, lake, and river ice complexity than grain size. SSA typically decreases with time with a rate depending on temperature and the shape of Snow cover on sea ice influences the Earth’s climate and the initial snow grain (Hachikubo et al. 2014). SSA mea- biology in the ocean. The only current snow-depth-on-sea- surements have been successfully conducted in the field ice algorithm that uses satellite data is based on passive using near IR methods (Gallet et al. 2009; Arnaud et al. microwave observations (Cavalieri et al. 2012; Brucker and 2011; Montpetit et al. 2012). The SnowMicroPen, which Markus 2013). Since 2009, NASA has supported the air- uses highly resolved penetration resistance (250 measure- borne Operation IceBridge mission, which operates mul- ments/mm), can be used to quantify snow density, grain tiple radars to retrieve snow depth on sea ice (Kurtz et al. size, and SSA (Proksch et al. 2015). Time-lapse X-ray 2013; Panzer et al. 2013). Recent work on IceBridge data micro-tomography methods provide a 3D reconstruction of and from drifting ice station indicates a substantial thinning the snow structure (Pinzer et al. 2012) and enable visual- of the snowpack in the western Arctic and in the Beaufort ization of the recrystallization distribution on depth hoar and Chukchi seas (Webster et al. 2014). This thinning is crystals through time (Fig. 3). Recent development of SSA negatively correlated with the delayed onset of sea-ice measurements led to implementation of SSA parametriza- freeze-up during autumn. Thin snowpack and sea ice tions in snow evolution modelling (Carmagnola et al. increase the heat flux between the ocean and atmosphere 2014). Advances in thermal and short IR remote sensing with potential feedbacks for the Earths’ climate but are not allow for determining surface snow types and surface thoroughly investigated. Although snow on lake ice has temperature (Hori et al. 2014). major implications for lake ecology, ice thickness, and the In snow hydrology, the onset and the total amount of local climate (Brown and Duguay 2010), studies on these runoff are essential for flood and reservoir management, systems appear to be under-represented in the literature and impact on terrestrial ecosystems. The change in (Cheng et al. 2014; Duguay et al. 2015). Furthermore, there dielectric permittivity of snow during melt highly influ- is currently little focus on quantifying changes in lake-ice ences remote sensing data from microwave to infrared, snow cover. The most recent progress in remote sensing is allowing us to monitor the extent of surficial melt (e.g. summarized in Duguay et al. (2015). Steffen et al. 2004). Modelling of LWC and snowpack runoff is still very challenging and water transport schemes Avalanche detection like a multi-layer bucket model or Richards equation underestimate observed maximum LWC in the course of a Recent advances in avalanche detection include the use of season (Heilig et al. 2015). LWC retention in the snow is seismic sensors and infrasound arrays (Table 2). Further- important to improve modelled runoff performance (Essery more, Synthetic Aperture Radar (SAR), e.g. Radarsat-2, et al. 2013; Heilig et al. 2015). TerraSAR-X, and Cosmo-Skymed, have been shown useful in detecting avalanche activity. Especially, the SAR data Snow-surface albedo and light-absorbing impurities properties as the spatial resolution (2–3 m), high temporal resolution (2–5 days), and their application during cloudy Impurities in the snowpack can affect the snowmelt rates conditions make them ideal for this purpose (Caduff et al. through decreased surface albedo. Such light-absorbing 2015). snow impurities include organic carbon, mineral dust, and micro-organisms (Langford et al. 2010), and can be Indigenous knowledge: Sa´mi snow observational quantified in manually collected snow samples and by methods and terminology reflectance measurements. Algal communities have been associated with glacial melt and reducing snow-surface Snow plays a central role in the cultures of indigenous albedo (e.g. Tedesco et al. 2013; Lutz et al. 2014). Similar Arctic people, notably for the reindeer herders of Eurasia.

Ó The Author(s) 2016. This article is published with open access at Springerlink.com 123 www.kva.se/en Ambio 2016, 45:516–537 523

Fig. 3 Age distribution of ice in a depth hoar sample from a laboratory experiment. The depth hoar sample has been exposed to typical temperature gradients of an Arctic snowpack (5°K snow temperature increase per 10 cm depth). Depth hoar recrystallizes completely and the oldest parts of the sample are just 5-days old ice (dark red), although the snow was made 28 days before (M. Schneebeli, WSL-SLF, unpublished)

They have developed a holistic snow terminology inte- periods leading to unseasonal melt periods and isolated grating the effects on the ecology, grazing opportunities, freeze–thaw cycles (Bokhorst et al. 2011; Semenchuk et al. and management of the herd (Fig. 4) which differs from 2013; Semmens et al. 2013; Wilson et al. 2013). These scientiﬁc standard terms (Eira et al. 2013). However, the events are caused by different factors such as heavy rainfall combination of traditional ecological knowledge (TEK) of (Rennert et al. 2009; Hansen et al. 2014) and movement of reindeer herders with natural science measurements and warm air masses through katabatic winds, e.g. Chinook snow classiﬁcation may guide future strategies for a sus- (Fuller et al. 2009) and foehn winds (Pedersen et al. 2015). tainable future of reindeer herding in a changing climate These extreme and anomalous events may be caused by (Riseth et al. 2011; Eira et al. 2013). TEK in general has different weather phenomena, but they all have the fol- been formally recognized by the Arctic Council as lowing in common: (1) they have an abrupt and sporadic important to understanding the Arctic (Arctic-Council nature, (2) they are unusual for the season in the geo- 1996) and the Ottawa traditional knowledge principles can graphical locations where they occur, (3) they cause be found here: http://www.arcticpeoples.org/images/2015/ changes in snowpack properties, and (4) they have imme- ottradknowlprinc.pdf. diate impacts on humans and ecosystems. Their temporal extent varies from a few hours to many days, and their Extreme events spatial extent is controlled by the spatial scale of the driving weather phenomenon (e.g. synoptic). Snow properties are increasingly impacted by extreme and The sparse distribution of meteorological stations and anomalous events such as ROS (Rennert et al. 2009), icing remoteness of areas across the Arctic region limit ground- (Bartsch et al. 2010; Hansen et al. 2013), and warming based observation of extreme events, their effect on the

Ó The Author(s) 2016. This article is published with open access at Springerlink.com www.kva.se/en 123 524 Ambio 2016, 45:516–537

Fig. 4 Schematic overview of Sa´mi snow concepts used during the cold season in reindeer herding in Guovdageaidnu, sub-Arctic Norway. The concepts are shown as they occur in and above the snowpack (blue frost on trees, green snow formation related to the surface and snowpack top layer, white mid snowpack layer, pink illustrates bottom snow layer). The arrows illustrate the duration of different concepts used by reindeer herders. This figure is modified from Fig. 4 by Eira et al. (2013). Further descriptions of the snow characteristics, rather than position and timing, can be found in Riseth et al. (2011) snowpack, and modelling efforts (e.g. Bulygina et al. 2010; Quantification and prediction of these extreme events Johansson et al. 2011; Hansen et al. 2014; Pedersen et al. requires increased research focus. 2015). However, Pedersen et al. (2015) quantified the spatially distributed snow property (SWE, snow depth, snow thermal resistance, and timing of snow-free date) MODELLING CHANGING SNOW CONDITIONS changes associated with episodic snowmelt events through in situ snow observations, meteorological data, and snow Types and applications of snow models modelling. Extreme events are also detectable through remote sensing using differencing 3-day averages of Terrestrial snow-cover models are used to simulate the backscatter (Bartsch et al. 2010; Semmens et al. 2013; snow temporal evolution in multiple hydrological, meteo- Wilson et al. 2013). Additionally, extreme events are rological, climatological, glaciological, and ecological detectable through modelling, e.g. by Liston and Hiemstra applications. Depending on the snow-model sophistication (2011) who showed an increased trend in ROS events over (i.e. the complexity of parameterisations used to describe maritime regions of the Arctic since 1979. Observed snow properties and the processes taking place within the (Hansen et al. 2014) and predicted (Bjerke et al. 2014) snow and at the interfaces with the atmosphere and the abrupt changes in snow properties and snow conditions soil), some models can also simulate snow stratigraphy (i.e. associated with extreme events add complexity to the the vertical evolution of snow properties in the various impacts of current warming in the Arctic (Walsh 2014). layers forming the snowpack).

Ó The Author(s) 2016. This article is published with open access at Springerlink.com 123 www.kva.se/en Ambio 2016, 45:516–537 525

Simple (empirical) snow models have been widely used reanalyses or climate model fields downscaled to a fine in impacts studies (e.g. Van Den Broeke et al. 2010; scale grid in order to account for the strong horizontal Saloranta 2012). These models have fewer data require- variability caused, for example, by complex orography ments (e.g. just temperature and precipitation) than physi- (Fiddes and Gruber 2014). The choice of input data cally based models, but require calibration. For example, depends on the application, and NWP data are used for Kumar et al. (2013) compared the impact of using a tem- snow prediction on large scales. perature index and a physically based snow model on Recent developments within the NWP community have streamflow simulations. They found that un-calibrated resulted in increased cooperation and interests among temperature-index models predict streamflow poorly. various disciplines (e.g. hydrology and ecology). The Therefore, simple empirical models need to be carefully increased spatial resolution of NWP models increases their calibrated in both time and space, whereas physically based potential utility for user groups who depend on modelling snow and hydrological models provide better accuracy. In regional- and local-scale processes. This is also supported fact, even calibrated models may be unreliable outside their by the development of off-line land-surface models which regions and periods of calibration (Bougamont et al. 2007). can be run stand-alone (e.g. Crocus snow physics model). Moreover, models based on energy balance principles are essential when snow models are required to provide Progress and key achievements in Arctic snow boundary conditions for atmospheric models in weather modelling and climate prediction applications and physically based snow models therefore remain essential. Modelling snow cover accurately is important, particularly Three main categories of physically based snow models because of the crucial role it plays in energy transfer exist: between the land and the atmosphere. Recent model intercomparison projects have improved our understanding of • Zero-layer (combined with soil) or single-layer snow how snow models perform and have prompted develop- models ments in individual models and parameterisations of snow • Intermediate complexity snow models accounting for processes. In this section, we highlight some achievements some physical processes within the snowpack, typically in snow modelling and look forward to upcoming inter- with 2–5 model layers comparison experiments. • Detailed snowpack models Snow models can be driven with measured or simulated Snow simulation achievements and limitations meteorological data. Usually, the higher the snow model sophistication, the simpler the framework within which Phase 5 of the Coupled Model Inter-comparison Project they are used. There are three main configurations in which (CMIP5; http://cmip-pcmdi.llnl.gov/cmip5/) provided an snow models are run: opportunity for assessing the simulation of snow in the current generation of climate models. Progress and limi- • Stand-alone models tations of CMIP5 models representing SWE, snow cover, • Coupled models with atmosphere, soil, and vegetation and snowfall compared to observations and reanalyses have components been identified (Brutel-Vuilmet et al. 2013; Kapnick and • Modules within Earth System Models (ESMs) Delworth 2013; Terzago et al. 2014). A key result was that ESMs typically use zero- and single-layer snow models the decreasing trend in Northern Hemisphere spring snow- because they have few parameterisations leading to fast cover extent over the 1979–2005 period (Derksen et al. computations, but they have limitations. Successful 2015) was underestimated by CMIP5 models (Brutel- attempts to couple intermediate complex snow models with Vuilmet et al. 2013). Snow-albedo feedbacks were mod- atmospheric and soil models have been made (e.g. within elled well but the spread in modelled snow-albedo feed- numerical weather prediction (NWP) systems and ESMs back has not narrowed since CMIP3, probably due to the such as HTESSEL (Dutra et al. 2010), RACMO (Kuipers widely varying treatment of the masking of snow-covered Munneke et al. 2011), and CLM4 (Oleson et al. 2010). surfaces by vegetation in the models (Qu and Hall 2014). Detailed snowpack models are typically used in simple Most CMIP5 models overestimate the contrast in albedo stand-alone configurations. Simulation results from these between snow-covered and snow-free land, but fewer models provide the temporal evolution of snow properties models had large cold temperature or high snow-cover with depth (Vionnet et al. 2012). It is possible to drive biases in CMIP5 than in CMIP3 (Fletcher et al. 2015). these sophisticated models either with weather station Because snow cover forms an interface between the measurements or with atmospheric reanalyses (e.g. Brun atmosphere and the land surface, differences in simulations et al. 2013). A similar approach is to use coarse-grid of the insulating effect of snow leads to disagreements in

Ó The Author(s) 2016. This article is published with open access at Springerlink.com www.kva.se/en 123 526 Ambio 2016, 45:516–537 modelled soil temperatures (Koven et al. 2013). Repre- Picard et al. 2014) and it is now parameterized in some sentation of snow properties may also affect the accuracy models (Carmagnola et al. 2014). SSA can now be mea- of air temperature calculated by climate models. Analysis sured in the field using observer-independent near-infrared of data from 48 CMIP5 models indicates that the calculated sensors (Gallet et al. 2009; Arnaud et al. 2011; Montpetit monthly-mean surface temperature for Northern Eurasia et al. 2012). Process studies have identified weaknesses of has the largest inter-model spread during the snowmelt snow models in simulating water percolation and ice-layer period indicating that accurate representation of the formation (e.g. Brucker et al. 2011; Wever et al. 2014). snowmelt is needed to improve the overall performance of However, physically based snow models may help in models and narrow the range of associated uncertainties in identifying ice layers in the snow (Vikhamar-Schuler et al. climate projections. 2013; Bjerke et al. 2014). Snow water mass still varies Large sets of simulations will soon be available from widely (50 %) among models and datasets relying solely on climate models and ESMs in CMIP6 (http://www.wcrp- satellite-derived information show approximately 40 % climate.org/wgcm-cmip/wgcm-cmip6) and from stand- less total snow for the peak accumulation seasons, com- alone land-surface models in GSWP3 (http://hydro.iis.u- pared with retrievals combining satellite- and ground-based tokyo.ac.jp/GSWP3/intro.html). The CliC ESM-SnowMIP data (Mudryk et al. 2015). project (http://www.climate-cryosphere.org/activities/ targeted/esm-snowmip) has been initiated to assess the Modelling soil–snow–vegetation interactions strengths and weaknesses of snow simulations in these experiments and to provide guidelines for the improvement Forests affect snow dynamics, and models have been of models. developed to incorporate this (Essery 2013). However, there are still issues with simulated snow-albedo feedbacks Snow model forcing data and the transition from snow-covered to snow-free canopies when temperatures rise above freezing (Thackeray Improved simulations can result from improvements in the et al. 2014). Shrubs trap windblown snow thereby affecting forcing data used to run snow models as well as from snow distribution (Myers-Smith et al. 2011) and this effect improvements in snow parameterizations. Snow-cover may be accentuated by the expansion of shrubs in some builds up due to solid precipitation and its properties are Arctic regions (e.g. Pearson et al. 2013; Urban et al. 2014). dramatically sensitive to liquid and mixed-phase precipi- The impact of snow-trapping by shrubs on soil tempera- tation. Though recent progress has been made (Marks et al. tures and gas fluxes have been modelled (e.g. Lawrence 2013; Mizukami et al. 2013), accurately partitioning pre- and Swenson 2011; Menard et al. 2014), but these pro- cipitation into rain and snow remains a challenge. Multi- cesses have not yet been included in dynamic vegetation ple-year snow model forcing datasets with multiple models. Progress on modelling freeze–thaw processes has evaluation data have recently been collated for several been made by increasing the numbers of layers and depth well-instrumented research sites in mid-latitude alpine of soil models, but modelling of permafrost conditions is locations (Brun et al. 2013), but there is a comparative lack degraded by biases in snow-depth simulations (Slater and of suitable data for the Arctic. For large-scale studies, Lawrence 2013). global gridded forcing datasets available from reanalyses have been used successfully (e.g. Brun et al. 2013). ESM- Modelling contaminants in snow SnowMIP includes comparisons between snow simulations at reference sites with in situ forcing data and large-scale Models now parameterize the impacts of contaminants simulations using reanalyses or coupled atmospheric with different spectral properties on the snow-surface models. albedo (Qian et al. 2015), but it remains challenging to couple these parameterisations with the atmospheric Snow parameterizations transport and deposition of contaminants such as BC. Current aerosol models can simulate mean BC concentra- Physical parameterizations of snow metamorphism are tions in snow reasonably well, but modelled distributions important because snow microstructure determines snow are poorly correlated with measurements; models generally properties, including those controlling energy exchanges at underestimate BC concentrations in snow in northern the snow/soil and snow/air interfaces. Specific surface area Russia and Norway but overestimate BC elsewhere in the (SSA) has attracted attention as a microstructural property Arctic (Jiao et al. 2014). Algae and bacteria living in snow that determines the physical, optical, and chemical prop- and ice are also considered contaminants, and the spectral erties of snow (Domine et al. 2008). It affects microwave properties of snow are affected by the species composition remote sensing (e.g. Brucker et al. 2011; Roy et al. 2013; (Lutz et al. 2014).

Ó The Author(s) 2016. This article is published with open access at Springerlink.com 123 www.kva.se/en Ambio 2016, 45:516–537 527

Table 3 Identiﬁcation of knowledge gaps related to changing Arctic snow cover and its consequences: gaps, recommendations, and implementation strategy Gaps Recommendations Implementation strategy

A. Observations There are large spatial scaling issues that (a) Increase the number of stations for manual INTERACT can provide additional measuring need to be resolved, from snow grain and automatic recording stations but needs information on methods characteristics to the circumpolar Arctic (b) Develop remote sensing tools that can and on making the data accessible region to the full Earth system. detect snow-depth differences across small GEO Cold Regions Initiative, which scale landscape topography coordinates existing in situ and remote sensing observations of snow can facilitate, through the Global Earth System of Systems (GEOSS), data sharing and method standardization The temporal evolution of the Arctic (a) Initiate year-round ground observations are INTERACT can provide year-round measuring snowpack throughout an entire cold season needed at intervals of hours or day stations but the number and location is poorly investigated, specifically, the (b) Improve methods to derive reliable depends on whether or not the methods are evolution of ice crusts and soil properties information at a proper spatial and temporal manual or remotely controlled (temperature and soil frost depth) resolution from remote sensing techniques from both optical and active (SAR) and passive (radiometer) microwave spaceborne sensors (c) Resolve technological difficulties in microwave and SAR (Synthetic Aperture Radar) remote sensing techniques The Arctic is vast but is sparsely populated (a) Extend the number of human-based snow and observing power is limited measurements to obtain a more detailed grid of snow parameters across the Arctic Region (b) Include citizen observations to extend the distribution of observations Ground-based observations of impacts of Develop detection methods (manual and extreme events on the snowpack are limited remote) to quantify and record impacts on the snowpack by extreme events The effects of physical properties of the (a) Improvement in the application and snowpack on sea ice have been measured development of new and coordinated but by out-dated methods and understanding methodologies are required of the snow-on-sea ice feedback is poor (b) Develop remote sensing techniques to quantify snowpack on sea ice The accuracy of remote sensing of SWE is Develop and improve remote sensing INTERACT can provide Arctic-wide ground- limited by topography and forest cover techniques for quantification of SWE validation of RS techniques over multiple topographies GEO Cold Regions Initiative can facilitate availability of remote sensing data through its Participant Organizations for intercomparison and validation efforts For modelling of snow precipitation, reliable (a) Increase the number of precipitation INTERACT can provide additional measuring measurements of total precipitation and measuring stations to meet the needs of the stations but needs information on methods solid precipitation fractions are crucial for modelling community and on making the data accessible properly driving snow models (b) Equip automated weather stations with SPICE is evaluating current instrumentation instrumentation to estimate precipitation (http://www.wmo.int/pages/prog/www/ phase—such as optical disdrometers (SPICE) IMOP/intercomparisons/SPICE/SPICE.html) There is great variety in methods used Share and compare techniques between INTERACT is already compiling a list of between different long-term measuring monitoring teams to increase the support for methods used at research stations and will stations long-term complete validation sites with help implement new observations and sensors probing the atmosphere, snow, and methods soil

Ó The Author(s) 2016. This article is published with open access at Springerlink.com www.kva.se/en 123 528 Ambio 2016, 45:516–537

Table 3 continued Gaps Recommendations Implementation strategy

B. Modelling The spread of model output needs to be More accurate representation of the snowmelt WCRP CliC ESM-SnowMIP experiments reduced in relation to snow-albedo feedback, is needed to improve the overall under CMIP6 will be investigating sources most models overestimate the contrast in performance of the models and narrow the of model spread in snow simulations and albedo between snow-covered and snow- range of associated uncertainties in climate their influence on climate free land. Differences in simulations of the projections insulating effect of snow leads to disagreements in modelled soil temperatures Aerosol models can simulate mean Black Inclusion of particle transport from snow-free Carbon (BC) concentrations in snow areas in GCM/regional snow models are reasonably well, but modelled distributions needed and the simulation of surface albedo are poorly correlated with measurements change due to dust deposition and microorganism growth Potential feedbacks between snow and sea ice The snow science community urgently needs are of critical importance, but not to quantify these feedbacks and include experimentally investigated them in models if relevant Potential feedbacks between snow and The snow science community needs to INTERACT can provide facilities around the freshwater ice are likely to be important quantify these feedbacks and include them Arctic for observations and experiments on because of the spatial coverage of tundra in models if relevant. Also, processes should feedbacks and for validation of models lakes and ponds. However, this has not been be identified and quantified using investigated in the field or in the laboratory experimental manipulations of snow while snow manipulation experiments on analogues to those deployed on land lake ice are absent Progress on modelling soil freeze and thaw Snow-depth simulations need to be improved WCRP CliC ESM-SnowMIP experiments processes has been made by increasing the and coupling of snow and soil models is under CMIP6 will be investigating sources numbers of layers and depth of soil models, needed of model spread in snow simulations and but modelling of permafrost conditions is their influence on climate degraded by biases in snow-depth simulations Process studies have identified weaknesses of Physically based snow models may help in snow models in simulating water identifying ice layers in the snow percolation and ice-layer formation Impacts of changing snow conditions on Increase the modelling effort on how changing teleconnections within the Arctic and with snow conditions impact on Arctic other regions of Earth require more research teleconnections attention C. Impacts studies Effects of earlier or late snowmelt impacts on (a) Initiate base-line studies to assess the INTERACT can help monitor spread of human well-being, such as physical injuries current threats and where in the Arctic pathogens and vectors throughout the Arctic and degree of exposure of people to region large changes may be expected and is developing a coordinated system to do pathogens from various sources transported (b) Promote research and monitoring this in snow and melt water coordination across the Arctic for inter- GEO Cold Regions Initiative can provide the comparability of methodologies societal benefits assessment and awareness crossing the GEO societal benefits areas via the GEO new work programme for 2016–2025 Recent studies on avalanche risk assessments Risk assessments need to be re-considered in indicate that these may be inaccurate light of changing snow conditions The direct impact of the temporal and spatial Initiate an economic assessment on the cost of variability of snow on the economic management and the costs associated with development of the Arctic, especially lack of appropriate management expressed in monetary value, is hard to evaluate. Determining these impacts is difficult as snow conditions are changing at the same time as economic growth

Ó The Author(s) 2016. This article is published with open access at Springerlink.com 123 www.kva.se/en Ambio 2016, 45:516–537 529

Table 3 continued Gaps Recommendations Implementation strategy

The detailed timing of changes in snow cover From an ecosystem perspective there is a INTERACT can facilitate to increase the during the cold season is uncertain. These pressing need to identify when the largest number of appropriate observations include periods of snowpack build-up, mid- changes in snow conditions will occur, e.g., National funding agencies need to be made winter rain events, spring snowmelt, and start, middle, or late winter aware of the requirement of seasonal timing as well as increased soil moisture monitoring and experiments deficits later in the growing season Impacts of changing snow conditions are We need to identify which species are most INTERACT can facilitate to start appropriate species-specific both for plants and animals. responsive to snow changes and why, and observations and host relevant experiments However, species vary in the magnitude of how they will impact ecosystem processes Protocols for monitoring snow conditions and their contribution to key ecosystem and surface feedback to climate impacts in the same places and at the same processes scales need to be further developed in the frame of CPMP The influences of snow and ground ice on Facilitate greater representation of snow-cover GEO Cold Regions Initiative can initiate a vegetation have been investigated in some in all its complexity including ice layers dedicated aim that may bridge the ecosystem models but these processes have not yet needs to be developed in vegetation/ mapping and snow-cover interaction been included in large scale dynamic ecosystem models vegetation models D. Linking and communicating Information exchange between science and (a) Facilitate information exchange between INTERACT offers a system for society is generally poor with inadequate society and the science community communication between field researchers communication. Sometimes there is low (b) Inform communities of ongoing and and local communities and has outreach relevance of the science for community projected changes relevant at the local scale activities needs. On the other hand, there are (c) Design observation strategies for traditional GEO Cold regions aims to establish a proactive sometimes excessive expectations of science to work together with citizens framework for the development of governments on researchers and lack of information and related services over Cold understanding of science by policy makers Region: the Global Cold Regions Community Portal The Arctic science community is well (a) Improve the integration between GEO Cold Regions can help by bridging the integrated and coordinated by various activities—monitoring, modelling, and different activities, domains, and organizations but their agendas for research evaluating impacts—and between Earth communities (remote sensing and in situ) in and monitoring, for example of snow cover, system domains—terrestrial, marine, the field of cold regions’ earth observations are often implemented independently, even atmospheric, and freshwater. GEO Cold Regions is promoting free access to though there are numerous interactions (b) We need to establish archives (metadata the earth observations data over the Cold within the Arctic and Earth systems portals) and/or a hub of in situ snow products Regions, including the Global Observation that are relevant for the snow science System of Systems (GEOSS) products and disciplines and communicate awareness of GEOSS-DataCORE the existence of these archives to other end- users (Policy makers and society)

CURRENT GAPS AND RECOMMENDATIONS determining changes in Arctic snow cover and their con- FOR FUTURE RESEARCH sequences is a lack of integration among domains (land, AND IMPLEMENTATION PLANS sea, lakes, and atmosphere) and between approaches. Monitoring of snow identiﬁes change but needs to be Without duplicating recommendations suggested by other linked to manipulations of climate, environment, and programmes (AMAP 2011), our intention was to review ecosystems to understand the impacts. This understanding and up-date the perceived gaps in current research activi- needs to be linked to modelling at relevant scales that ties on Arctic snow changes as a contribution to the ICARP project into the future (or past). With this predictive III process towards a roadmap for future research. To focus capability, knowledge-based management may be devel- these developments, we identiﬁed key gaps, formulate oped and implemented (Johansson et al. 2012). One pos- recommendations, and seek commitments by stakeholders sibility to improve integration of activities across domains and major Arctic and Global organisations to implement and approaches is to develop coordinated activities, hosted these recommendations (Table 3). In addition, many by a regional or global organization. detailed requirements exist which are listed in Supple- Therefore, in order to develop ESM that can be used in mentary material S1. A key limitation to progress on the documentation and/or prediction of snow-cover

Ó The Author(s) 2016. This article is published with open access at Springerlink.com www.kva.se/en 123 530 Ambio 2016, 45:516–537

Fig. 5 Conceptual model of required interactions between society and management and science including the snow monitoring, snow modelling, and snow-impact communities changes and their impacts, there is a need for improved communication to end-users could be achieved through the communication and cooperation between discipline-speci- ICARP process and associated organizations IASC, fic communities (ecologist/biologist, social scientists, and INTERACT, CliC, GEO (GEOSS), and WMO (GCW). snow scientist) and between the approaches (monitoring/ With this paper, we have attempted to provide a basis, and observers in the field/remote sensing and modellers) stimulus, for the implementation of key priorities (Table 3) (Fig. 5). For instance, ecologists need to identify at which to address the limitations in our understanding of Arctic spatial and temporal resolutions snow-cover changes are snow conditions and how they may change in the near relevant and make this known to the modelling community. future. This will assure that the outputs of modelled snow variables match the given resolution of ecosystem processes Acknowledgments The writing of this paper was initiated by an and dynamics. Conversely, modellers require validation IASC ICARP III Activity grant to TVC enabling a workshop hosted by the European Environment Agency. The authors acknowledge data of snow variables on relevant scales (Table 1). funding from their respective national and international funding Therefore, the timing, frequency, and spatial resolution of bodies, which has enabled the contribution of all authors to this work. snow surveys and snow monitoring should match the snowmodel resolution in order to generate useful snow outputs Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http:// for the ecosystem scientists/snow-impact community creativecommons.org/licenses/by/4.0/), which permits unrestricted (Fig. 5). For this interaction to be successful, detailed use, distribution, and reproduction in any medium, provided you give cross-disciplinary coordination of field campaigns, moni- appropriate credit to the original author(s) and the source, provide a toring, research projects, and model development is link to the Creative Commons license, and indicate if changes were made. required. Since society and its infrastructure have to cope with the challenges of changing snow conditions (Fig. 1), it requires REFERENCES easy access to snow predictions. Therefore, an open dia- logue needs to be established or expanded to facilitate Alou-Font, E., C.J. Mundy, S. Roy, M. Gosselin, and S. Agusti. 2013. information exchange between society and the science Snow cover affects ice algal pigment composition in the coastal community. Implementation of these recommendations Arctic Ocean during spring. Marine Ecology Progress Series should ideally be considered by organizations, such as the 474: 89–104. AMAP. 2011. Snow, water, Ice and Permafrost in the Arctic Arctic Council, that span science and human dimensions. (SWIPA): Climate change and the cryosphere, xii–538. Oslo: Integration between the different snow disciplines and Arctic Monitoring and Assessment Programme (AMAP).

Ó The Author(s) 2016. This article is published with open access at Springerlink.com 123 www.kva.se/en Ambio 2016, 45:516–537 531

Aoki, T., S. Matoba, S. Yamaguchi, et al. 2014. Light-absorbing snow Bougamont, M., J.L. Bamber, J.K. Ridley, et al. 2007. Impact of impurity concentrations measured on Northwest Greenland ice model physics on estimating the surface mass balance of the sheet in 2011 and 2012. Bulletin of Glaciological Research 32: Greenland ice sheet. Geophysical Research Letters. doi:10.1029/ 21–31. 2007GL030700. Arctic-Council. 1996. Declaration on the establishment of the Arctic Bowden, J.J., A. Eskildsen, R.R. Hansen, K. Olsen, C.M. Kurle, and Council, 1–5. Ottawa: Arctic-Council. T.T. Høye. 2015. High-Arctic butterflies become smaller with Arnaud, L., G. Picard, N. Champollion, et al. 2011. Measurement of rising temperatures. Biology Letters 11: 20150574. vertical profiles of snow specific surface area with a 1 cm Brown, L.C., and C.R. Duguay. 2010. The response and role of ice resolution using infrared reflectance: Instrument description and cover in lake–climate interactions. Progress in Physical Geog- validation. Journal of Glaciology 57: 17–29. raphy 34: 671–704. Ask, J., J. Karlsson, L. Persson, P. Ask, P. Bystro¨m, and M. Jansson. Brucker, L., and T. Markus. 2013. Arctic-scale assessment of satellite 2009. Terrestrial organic matter and light penetration: Effects on passive microwave-derived snow depth on sea ice using bacterial and primary production in lakes. Limnology and Operation IceBridge airborne data. Journal of Geophysical Oceanography 54: 2034–2040. Research-Oceans 118: 2892–2905. Avanzi, F., M. Caruso, C. Jommi, C. De Michele, and A. Ghezzi. Brucker, L., A. Royer, G. Picard, A. Langlois, and M. Fily. 2011. 2014. Continuous-time monitoring of liquid water content in Hourly simulations of the microwave brightness temperature of snowpacks using capacitance probes: A preliminary feasibility seasonal snow in Quebec, Canada, using a coupled snow study. Advances in Water Resources 68: 32–41. evolution–emission model. Remote Sensing of Environment Barichivich, J., K.R. Briffa, R.B. Myneni, et al. 2013. Large-scale 115: 1966–1977. variations in the vegetation growing season and annual cycle of Brun, E., V. Vionnet, A. Boone, et al. 2013. Simulation of northern atmospheric CO2 at high northern latitudes from 1950 to 2011. Eurasian local snow depth, mass, and density using a detailed Global Change Biology 19: 3167–3183. snowpack model and meteorological reanalyses. Journal of Bartsch, A., T. Kumpula, B.C. Forbes, and F. Stammler. 2010. Hydrometeorology 14: 203–219. Detection of snow surface thawing and refreezing in the Brutel-Vuilmet, C., M. Meńe´goz, and G. Krinner. 2013. An analysis Eurasian Arctic with QuikSCAT: Implications for reindeer of present and future seasonal Northern Hemisphere land snow herding. Ecological Applications 20: 2346–2358. cover simulated by CMIP5 coupled climate models. The Bernard, E., J.M. Friedt, F. Tolle, M. Griselin, G. Martin, D. Laffly, Cryosphere 7: 67–80. and C. Marlin. 2013. Monitoring seasonal snow dynamics using Bulygina, O.N., P.Y. Groisman, V.N. Razuvaev, and V.F. Radionov. ground based high resolution photography (Austre Lovenbreen, 2010. Snow cover basal ice layer changes over Northern Eurasia Svalbard, 79°N). ISPRS Journal of Photogrammetry and Remote since 1966. Environmental Research Letters 5: 015004. Sensing 75: 92–100. Caduff, R., A. Wiesmann, Y. Bu¨hler, and C. Pielmeier. 2015. Biedunkiewicz, A., and E. Ejdys. 2011. Icicles as carriers of yeast- Continuous monitoring of snowpack displacement at high spatial like fungi potentially pathogenic to human. Aerobiologia 27: and temporal resolution with terrestrial radar interferometry. 333–337. Geophysical Research Letters 42: 813–820. Bilodeau, F., G. Gauthier, and D. Berteaux. 2013. Effect of snow Callaghan, T., M. Johansson, R. Brown, et al. 2011. The changing cover on the vulnerability of lemmings to mammalian predators face of Arctic snow cover: A synthesis of observed and projected in the Canadian Arctic. Journal of Mammalogy 94: 813–819. changes. Ambio 40: 17–31. Bjerke, J.W., S.R. Karlsen, K.A. Høgda, et al. 2014. Record-low Carmagnola, C.M., S. Morin, M. Lafaysse, et al. 2014. Implemen- primary productivity and high plant damage in the Nordic Arctic tation and evaluation of prognostic representations of the optical Region in 2012 caused by multiple weather events and pest diameter of snow in the SURFEX/ISBA-Crocus detailed snow- outbreaks. Environmental Research Letters 9: 084006. pack model. The Cryosphere 8: 417–437. Bjerke, J.W., H. Tømmervik, M. Zielke, and M. Jørgensen. 2015. Cavalieri, D.J., T. Markus, A. Ivanoff, et al. 2012. A comparison of Impacts of snow season on ground-ice accumulation, soil frost snow depth on sea ice retrievals using airborne altimeters and an and primary productivity in a grassland of sub-Arctic Norway. AMSR-E simulator. IEEE Transactions on Geoscience and Environmental Research Letters 10: 095007. Remote Sensing 50: 3027–3040. Bokhorst, S., J.W. Bjerke, M. Davey, et al. 2010. Impacts of extreme Cheng, B., T. Vihma, L. Rontu, A. Kontu, H.K. Pour, C. Duguay, and winter warming events on plant physiology in a sub-Arctic heath J. Pulliainen. 2014. Evolution of snow and ice temperature, community. Physiologia Plantarum 140: 128–140. thickness and energy balance in Lake Oraja¨rvi, northern Finland. Bokhorst, S., J.W. Bjerke, L. Street, T.V. Callaghan, and G.K. Tellus A. doi:10.3402/tellusa.v3466.21564. Phoenix. 2011. Impacts of multiple extreme winter warming Clarke, G.K.C., A.H. Jarosch, F.S. Anslow, V. Radic, and B. events on sub-Arctic heathland: phenology, reproduction, Menounos. 2015. Projected deglaciation of western Canada in growth, and CO2 flux responses. Global Change Biology 17: the twenty-first century. Nature Geoscience 8: 372–377. 2817–2830. Cohen, J., J. Jones, J.C. Furtado, and E. Tziperman. 2013. Warm Bokhorst, S., G.K. Phoenix, J.W. Bjerke, T.V. Callaghan, F. Huyer- Arctic, cold continents a common pattern related to Arctic sea Brugman, and M.P. Berg. 2012. Extreme winter warming events ice melt, snow advance, and extreme winter weather. Oceanog- more negatively impact small rather than large soil fauna: Shift raphy 26: 152–160. in community composition explained by traits not taxa. Global Cohen, J., J.C. Furtado, J. Jones, M. Barlow, D. Whittleston, and D. Change Biology 18: 1152–1162. Entekhabi. 2014. Linking Siberian snow cover to precursors of Bokhorst, S., D.B. Metcalfe, and D.A. Wardle. 2013. Reduction in stratospheric variability. Journal of Climate 27: 5422–5432. snow depth negatively affects decomposers but impact on Cooper, E.J. 2014. Warmer shorter winters disrupt Arctic terrestrial decomposition rates is substrate dependent. Soil Biology & ecosystems. Annual Review of Ecology Evolution and System- Biochemistry 62: 157–164. atics 45: 271–295. Borzenkova, A.B., and A.B. Shmakin. 2012. Changes in the snow Crook, J.A., P.M. Forster, and N. Stuber. 2011. Spatial patterns of cover thickness and of daily snowfall intensity affecting the modeled climate feedback and contributions to temperature highways cleaning expenses in Russian cities. Ice and Snow 2: response and Polar amplification. Journal of Climate 24: 59–70. 3575–3592.

Ó The Author(s) 2016. This article is published with open access at Springerlink.com www.kva.se/en 123 532 Ambio 2016, 45:516–537

Deems, J.S., T.H. Painter, and D.C. Finnegan. 2013. Lidar measure- Francis, J.A., and S.J. Vavrus. 2012. Evidence linking Arctic ment of snow depth: A review. Journal of Glaciology 59: amplification to extreme weather in mid-latitudes. Geophysical 467–479. Research Letters 39: L06801. Denoth, A. 1994. An electronic devise for long-term snow wetness Fuller, M.C., T. Geldsetzer, and J.J. Yackel. 2009. Surface-based recording. Annals of Glaciology 19: 104–106. polarimetric C-band microwave scatterometer measurements of Derksen, C., R. Brown, L. Mudryk, and K. Luojus. 2015. Arctic: snow during a Chinook event. IEEE Transactions on Geoscience Terrestrial Snow. State of the Climate in 2014. J. Blunden and D. and Remote Sensing 47: 1766–1776. S. Arndt. Bulletin of the American Meteorological Society 96: Gallet, J.C., F. Domine, C.S. Zender, and G. Picard. 2009. Measure- 133–135. ment of the specific surface area of snow using infrared Dibike, Y., T. Prowse, B. Bonsal, L. de Rham, and T. Saloranta. 2012. reflectance in an integrating sphere at 1310 and 1550 nm. Simulation of North American lake-ice cover characteristics Cryosphere 3: 167–182. under contemporary and future climate conditions. International Hachikubo, A., S. Yamaguchi, H. Arakawa, et al. 2014. Effects of Journal of Climatology 32: 695–709. temperature and grain type on time variation of snow specific Doherty, S.J., T.C. Grenfell, S. Forsstrom, D.L. Hegg, R.E. Brandt, surface area. Bulletin of Glaciological Research 32(1): 33–45. and S.G. Warren. 2013. Observed vertical redistribution of black Hall, D.K., G.A. Riggs, and V.V. Salomonson. 2006. MODIS snow carbon and other insoluble light-absorbing particles in melting and sea ice products. In Earth science satellite remote sensing, snow. Journal of Geophysical Research—Atmospheres 118: vol. I: Science and Instruments, ed. J.J. Qu, W. Gao, M. Kafatos, 5553–5569. R.E. Murphy, and V.V. Salomonson, 154–181. New York: Domine, F., M. Albert, T. Huthwelker, et al. 2008. Snow physics as Springer. relevant to snow photochemistry. Atmospheric Chemistry and Hall, D.K., G.A. Riggs, V.V. Salomonson, N.E. DiGirolamo, and K.J. Physics 8: 171–208. Bayr. 2002. MODIS snow-cover products. Remote Sensing of Douglas, T.A., L.L. Loseto, R.W. Macdonald, et al. 2012. The fate of Environment 83: 181–194. mercury in Arctic terrestrial and aquatic ecosystems: A review. Hanbali, R.M. 1994. Economic impact of winter road maintenance on Environmental Chemistry 9: 321–355. road users. Transportation Research Record 1442: 151–161. Duguay, C.R., M. Bernier, Y. Gauthier, and A. Kouraev. 2015. Hanewinkel, M., S. Hummel, and A. Albrecht. 2011. Assessing Remote sensing of lake and river ice. In Remote sensing of the natural hazards in forestry for risk management: A review. cryosphere, ed. M. Tedesco, 273–306. Oxford: Wiley-Blackwell. European Journal of Forest Research 130: 329–351. Dumont, M., E. Brun, G. Picard, et al. 2014. Contribution of light- Hansen, B.B., V. Grøtan, R. Aanes, et al. 2013. Climate events absorbing impurities in snow to Greenland’s darkening since synchronize the dynamics of a resident vertebrate community in 2009. Nature Geoscience 7: 509–512. the high Arctic. Science 339: 313–315. Dupuis, A.P., and B.J. Hann. 2009. Climate change, diapause Hansen, B.B., K. Isaksen, R.E. Benestad, et al. 2014. Warmer and termination and zooplankton population dynamics: An experi- wetter winters: Characteristics and implications of an extreme mental and modelling approach. Freshwater Biology 54: weather event in the High Arctic. Environmental Research 221–235. Letters 9: 114021. Dutra, E., G. Balsamo, P. Viterbo, P.M.A. Miranda, A. Beljaars, C. Havens, S., H.-P. Marshall, J.B. Johnson, and B. Nicholson. 2014. Scha¨r, and K. Elder. 2010. An improved snow scheme for the Calculating the velocity of a fast-moving snow avalanche using ECMWF land surface model: Description and offline validation. an infrasound array. Geophysical Research Letters 41: Journal of Hydrometeorology 11: 899–916. 6191–6198. Eckerstorfer, M., and H.H. Christiansen. 2012. Meteorology, topog- Haynes, K.M., and C.P.J. Mitchell. 2012. Inter-annual and spatial raphy and snowpack conditions causing two extreme mid-winter variability in hillslope runoff and mercury flux during spring slush and wet slab avalanche periods in high Arctic maritime snowmelt. Journal of Environmental Monitoring 14: 2083–2091. Svalbard. Permafrost and Periglacial Processes 23: 15–25. Heilig, A., M. Schneebeli, and O. Eisen. 2009. Upward-looking Eira, I.M.G., C. Jaedicke, O.H. Magga, N.G. Maynard, D. Vikhamar- ground-penetrating radar for monitoring snowpack stratigraphy. Schuler, and S.D. Mathiesen. 2013. Traditional Sami snow Cold Regions Science and Technology 59: 152–162. terminology and physical snow classification—Two ways of Heilig, A., C. Mitterer, L. Schmid, N. Wever, J. Schweizer, H.P. knowing. Cold Regions Science and Technology 85: 117–130. Marshall, and O. Eisen. 2015. Seasonal and diurnal cycles of liquid Ejdys, E., A. Biedunkiewicz, M. Dynowska, and E. Sucharzewska. water in snow—Measurements and modeling. Journal of Geo- 2014. Snow in the city as a spore bank of potentially pathogenic physical Research: Earth Surface. doi:10.1002/2015JF003593. fungi. Science of the Total Environment 470: 646–650. Holemann, J.A., M. Schirmacher, and A. Prange. 2005. Seasonal Essery, R. 2013. Large-scale simulations of snow albedo masking by variability of trace metals in the Lena River and the southeastern forests. Geophysical Research Letters 40: 5521–5525. Laptev Sea: Impact of the spring freshet. Global and Planetary Essery, R., S. Morin, Y. Lejeune, and C.B. Meńard. 2013. A Change 48: 112–125. comparison of 1701 snow models using observations from an Hori, M., T. Aoki, T. Tanikawa, K. Kuchiki, M. Niwano, S. alpine site. Advances in Water Resources 55: 131–148. Yamaguchi, and S. Matoba. 2014. Dependence of thermal Fiddes, J., and S. Gruber. 2014. TopoSCALE v. 1.0: Downscaling infrared emissive behaviors of snow cover on the surface snow gridded climate data in complex terrain. Geoscientific Model type. Bulletin of Glaciological Research 32: 33–45. Development 7: 387–405. Ims, R.A., N.G. Yoccoz, and S.T. Killengreen. 2011. Determinants of Fierz, C.R., R.L. Armstrong, Y. Durand, P. Etchevers, E. Greene, lemming outbreaks. Proceedings of the National Academy of D.M. McClung, K. Nishimura, P.K. Satyawali, and S.A. Sciences of the United States of America 108: 1970–1974. Sokratov. 2009. The international classification for seasonal Jeelani, G., J.J. Feddema, C.J. van der Veen, and L. Stearns. 2012. snow on the ground. Paper presented at the UNESCO-IHP, Paris, Role of snow and glacier melt in controlling river hydrology in France. Liddar watershed (western Himalaya) under current and future Fletcher, C.G., C.W. Thackeray, and T.M. Burgers. 2015. Evaluating climate. Water Resources Research 48: W12508. biases in simulated snow albedo feedback in two generations of Jiao, C., M.G. Flanner, Y. Balkanski, et al. 2014. An AeroCom climate models. Journal of Geophysical Research—Atmospheres assessment of black carbon in Arctic snow and sea ice. 120: 12–26. Atmospheric Chemistry and Physics 14: 2399–2417.

Ó The Author(s) 2016. This article is published with open access at Springerlink.com 123 www.kva.se/en Ambio 2016, 45:516–537 533

Johansson, C., V.A. Pohjola, C. Jonasson, and T.V. Callaghan. 2011. and its possible linkage to accelerated sea ice decrease. Journal Multi-decadal changes in snow characteristics in sub-Arctic of Climate 27: 6551–6558. Sweden. Ambio 40: 566–574. Matzl, M., and M. Schneebeli. 2006. Measuring specific surface area Johansson, M., C. Jonasson, M. Sonesson, and T.R. Christensen. of snow by near-infrared photography. Journal of Glaciology 52: 2012. The man, the myth, the legend: Professor Terry V. 558–564. Callaghan and his 3M concept. Ambio 41: 175–177. McCreight, J.L., E.E. Small, and K.M. Larson. 2014. Snow depth, Kapnick, S.B., and T.L. Delworth. 2013. Controls of global snow density, and SWE estimates derived from GPS reflection data: under a changed climate. Journal of Climate 26: 5537–5562. Validation in the western U.S. Water Resources Research 50: Kelly, R. 2009. The AMSR-E snow depth algorithm: Description and 6892–6909. initial results. Journal of The Remote Sensing Society of Japan McKinnon, L., D. Berteaux, G. Gauthier, and J. Beˆty. 2013. Predator- 29: 307–317. mediated interactions between preferred, alternative and inci- Koch, F., M. Prasch, L. Schmid, J. Schweizer, and W. Mauser. 2014. dental prey in the arctic tundra. Oikos 122: 1042–1048. Measuring snow liquid water content with low-cost GPS Meltofte, H. 2013. Arctic biodiversity assessment. Status and trends receivers. Sensors 14: 20975–20999. in Arctic biodiversity. Akureyri: Conservation of Arctic Flora Koven, C.D., W.J. Riley, and A. Stern. 2013. Analysis of permafrost and Fauna. thermal dynamics and response to climate change in the CMIP5 Menard, C.B., R. Essery, and J. Pomeroy. 2014. Modelled sensitivity earth system models. Journal of Climate 26: 1877–1900. of the snow regime to topography, shrub fraction and shrub Krenke, A.N., E.A. Cherenkova, and M.M. Chernavskaya. 2012. height. Hydrology and Earth System Sciences 18: 2375–2392. Stability of snow cover on the territory of Russia in relation to Michel, D., R. Philipona, C. Ruckstuhl, R. Vogt, and L. Vuilleumier. climate change. Ice and Snow 1: 29–37. 2008. Performance and uncertainty of CNR1 net radiometers Kuipers Munneke, P., M.R. van den Broeke, J.T.M. Lenaerts, M.G. during a one-year field comparison. Journal of Atmospheric and Flanner, A.S. Gardner, and W.J. van de Berg. 2011. A new Oceanic Technology 25: 442–451. albedo parameterization for use in climate models over the Mitterer, C., A. Heilig, J. Schweizer, and O. Eisen. 2011. Upward- Antarctic ice sheet. Journal of Geophysical Research: Atmo- looking ground-penetrating radar for measuring wet-snow prop- spheres. doi:10.1029/2010JD015113. erties. Cold Regions Science and Technology 69: 129–138. Kumar, M., D. Marks, J. Dozier, M. Reba, and A. Winstral. 2013. Mizukami, N., V. Koren, M. Smith, D. Kingsmill, Z. Zhang, B. Evaluation of distributed hydrologic impacts of temperature- Cosgrove, and Z. Cui. 2013. The impact of precipitation type index and energy-based snow models. Advances in Water discrimination on hydrologic simulation: Rain–snow partitioning Resources 56: 77–89. derived from HMT-West radar-detected brightband height Kurtz, N., J. Richter-Menge, S. Farrell, M. Studinger, J. Paden, J. versus surface temperature data. Journal of Hydrometeorology Sonntag, and J. Yungel. 2013. IceBridge airborne survey data 14: 1139–1158. support Arctic sea ice predictions. Eos, Transactions American Montpetit, B., A. Royer, A. Langlois, et al. 2012. New shortwave Geophysical Union 94: 41–41. infrared albedo measurements for snow specific surface area Langford, H., A. Hodson, S. Banwart, and C. Boggild. 2010. The retrieval. Journal of Glaciology 58: 941–952. microstructure and biogeochemistry of Arctic cryoconite gran- Mudryk, L.R., C. Derksen, P.J. Kushner, and R. Brown. 2015. ules. Annals of Glaciology 51: 87–94. Characterization of northern hemisphere snow water equivalent Lawrence, D.M., and S.C. Swenson. 2011. Permafrost response to datasets, 1981–2010. Journal of Climate 28: 8037–8051. increasing Arctic shrub abundance depends on the relative Myers-Smith, I.H., and D.S. Hik. 2013. Shrub canopies influence soil influence of shrubs on local soil cooling versus large-scale temperatures but not nutrient dynamics: An experimental test of climate warming. Environmental Research Letters 6: 045504. tundra snow–shrub interactions. Ecology and Evolution 3: Leinss, S., G. Parrella, and I. Hajnsek. 2014. Snow height determi- 3683–3700. nation by polarimetric phase differences in X-Band SAR data. Myers-Smith, I.H., B.C. Forbes, M. Wilmking, et al. 2011. Shrub IEEE Journal of Selected Topics in Applied Earth Observations expansion in tundra ecosystems: Dynamics, impacts and and Remote Sensing 7: 3794–3810. research priorities. Environmental Research Letters 6: 045509. Liston, G.E., and C.A. Hiemstra. 2011. The changing cryosphere: Niemi, J., and J. Ahlstedt. 2012. Finnish agriculture and rural Pan-Arctic snow trends (1979–2009). Journal of Climate 24: industries, 112a. Helsinki: Agrifood Research. 5691–5712. Nolet, B.A., S. Bauer, N. Feige, Y.I. Kokorev, I.Y. Popov, and B.S. Lund-Hansen, L., I. Hawes, B. Sorrell, and M. Nielsen. 2014. Ebbinge. 2013. Faltering lemming cycles reduce productivity Removal of snow cover inhibits spring growth of Arctic ice and population size of a migratory Arctic goose species. Journal algae through physiological and behavioral effects. Polar of Animal Ecology 82: 804–813. Biology 37: 471–481. Oleson, K.W., D.M. Lawrence, B. Gordon, et al. 2010. Technical Lutz, S., A.M. Anesio, S.E.J. Villar, and L.G. Benning. 2014. description of version 4.0 of the community land model (CLM). Variations of algal communities cause darkening of a Greenland NCAR Technical Notes. glacier. FEMS Microbiology Ecology 89: 402–414. Panzer, B., D. Gomez-Garcia, C. Leuschen, et al. 2013. An ultra- Marks, D., A. Winstral, M. Reba, J. Pomeroy, and M. Kumar. 2013. wideband, microwave radar for measuring snow thickness on sea An evaluation of methods for determining during-storm precip- ice and mapping near-surface internal layers in polar firn. itation phase and the rain/snow transition elevation at the surface Journal of Glaciology 59: 244–254. in a mountain basin. Advances in Water Resources 55: 98–110. Parham, P.E., J. Waldock, G.K. Christophides, et al. 2015. Climate, Marshall, H-P., G. Koh, and R.R. Forster. 2005. Estimating alpine environmental and socio-economic change: weighing up the snowpack properties using FMCW radar. Annals of Glaciology balance in vector-borne disease transmission. Philosophical 40: 157–162. Transactions of the Royal Society of London B: Biological Matsumoto, N., and T. Hoshino. 2009. Fungi in snow environments: Sciences 370: 20130557. psychrophilic molds—A group of pathogens affecting plants Pearson, R.G., S.J. Phillips, M.M. Loranty, P.S.A. Beck, T. under snow. Enfield: Science Publishers Inc. Damoulas, S.J. Knight, and S.J. Goetz. 2013. Shifts in Arctic Matsumura, S., X. Zhang, and K. Yamazaki. 2014. Summer Arctic vegetation and associated feedbacks under climate change. atmospheric circulation response to spring Eurasian snow cover Nature Climate Change 3: 673–677.

Ó The Author(s) 2016. This article is published with open access at Springerlink.com www.kva.se/en 123 534 Ambio 2016, 45:516–537

Pedersen, S.H., G.E. Liston, M.P. Tamstorf, A. Westergaard-Nielsen, Schmid, L., A. Heilig, C. Mitterer, J. Schweizer, H. Maurer, R. and N.M. Schmidt. 2015. Quantifying episodic snowmelt events Okorn, and O. Eisen. 2014. Continuous snowpack monitoring in Arctic ecosystems. Ecosystems 18: 839–856. using upward-looking ground-penetrating radar technology. Picard, G., A. Royer, L. Arnaud, and M. Fily. 2014. Influence of Journal of Glaciology 60: 509–525. meter-scale wind-formed features on the variability of the Schmidt, N.M., R.A. Ims, T.T. Høye, et al. 2012. Response of an microwave brightness temperature around Dome C in Antarc- arctic predator guild to collapsing lemming cycles. Proceedings tica. The Cryosphere 8: 1105–1119. of the Royal Society B—Biological Sciences 279: 4417–4422. Pinzer, B.R., M. Schneebeli, and T.U. Kaempfer. 2012. Vapor flux Schneebeli, M., and J.B. Johnson. 1998. A constant-speed penetrom- and recrystallization during dry snow metamorphism under a eter for high-resolution snow stratigraphy. Annals of Glaciology steady temperature gradient as observed by time-lapse micro- 26: 107–111. tomography. Cryosphere 6: 1141–1155. Screen, J.A. 2014. Arctic amplification decreases temperature vari- Popova, V.V. 2011. The snow storage contribution in major changes ance in northern mid- to high-latitudes. Nature Climate Change of river runoff in the Arctic Ocean drainage basin during the 4: 577–582. current warming period. Ice and Snow 51: 69–78. Screen, J.A., and I. Simmonds. 2012. Declining summer snowfall in Preece, C., T.V. Callaghan, and G.K. Phoenix. 2012. Impacts of the Arctic: Causes, impacts and feedbacks. Climate Dynamics winter icing events on the growth, phenology and physiology of 38: 2243–2256. sub-arctic dwarf shrubs. Physiologia Plantarum. doi:10.1111/j. Semenchuk, P.R., B. Elberling, and E.J. Cooper. 2013. Snow cover 1399-3054.2012.01640.x. and extreme winter warming events control flower abundance of Proksch, M., H. Lo¨we, and M. Schneebeli. 2015. Density, specific some, but not all species in high arctic Svalbard. Ecology and surface area, and correlation length of snow measured by high- Evolution 3: 2586–2599. resolution penetrometry. Journal of Geophysical Research: Semenov, V.A. 2013. Climate-related change of snow contribution in Earth Surface 120: 346–362. the development of dangerous hydrological phenomena on Prowse, T.D., and K. Brown. 2010. Hydro-ecological effects of rivers. Ice and Snow 53: 107–112. changing Arctic river and lake ice covers: A review. Hydrolog- Semmens, K.A., J. Ramage, A. Bartsch, and G.E. Liston. 2013. Early ical Research 41: 454–461. snowmelt events: Detection, distribution, and significance in a Prowse, T., K. Alfredsen, S. Beltaos, et al. 2011. Past and future major sub-arctic watershed. Environmental Research Letters 8: changes in Arctic lake and river ice. Ambio 40: 53–62. 11. Qian, Y., T.J. Yasunari, S.J. Doherty, et al. 2015. Light-absorbing Serreze, M.C., and R.G. Barry. 2011. Processes and impacts of Arctic particles in snow and ice: Measurement and modeling of climatic amplification: A research synthesis. Global and Planetary and hydrological impact. Advances in Atmospheric Sciences 32: Change 77: 85–96. 64–91. Shen, F.X., and M.S. Yao. 2013. Are we biologically safe with snow Qiu, J. 2014. Avalanche hotspot revealed. Nature 509: 142–143. precipitation? A case study in Beijing. PLoS ONE 8: 11. Qu, X., and A. Hall. 2014. On the persistent spread in snow-albedo Shnyparkov, A., S. Fuchs, S. Sokratov, K. Koltermann, Y.G. feedback. Climate Dynamics 42: 69–81. Seliverstov, and M. Vikulina. 2012. Theory and practice of Rautio, M., H. Mariash, and L. Forsstro¨m. 2011. Seasonal shifts individual snow avalanche risk assessment in the Russian Arctic. between autochthonous and allochthonous carbon contributions Geography, Environment, Sustainability 5: 64–81. to zooplankton diets in a subarctic lake. Limnology and Sihvola, A., and M. Tiuri. 1986. Snow fork for field determination of Oceanography 56: 1513–1524. the density and wetness profiles of a snow pack. IEEE Reiweger, I., K. Mayer, K. Steiner, J. Dual, and J. Schweizer. 2015. Transactions on Geoscience and Remote Sensing GE-24: Measuring and localizing acoustic emission events in snow prior 717–721. to fracture. Cold Regions Science and Technology 110: 160–169. Simon, A., M.B. Poulin, A.N. Rousseau, and N.H. Ogden. 2013. Rennert, K.J., G. Roe, J. Putkonen, and C.M. Bitz. 2009. Soil thermal Fate and transport of Toxoplasma gondii Oocysts in seasonally and ecological impacts of rain on snow events in the circumpolar snow covered watersheds: A conceptual framework from a Arctic. Journal of Climate 22: 2302–2315. melting snowpack to the Canadian Arctic coasts. International Riehm, M., and L. Nordin. 2012. Optimization of winter road Journal of Environmental Research and Public Health 10: maintenance energy costs in Sweden: A critique of site specific 994–1005. frost warning techniques. Meteorological Applications 19: Slater, A.G., and D.M. Lawrence. 2013. Diagnosing present and 443–453. future permafrost from climate models. Journal of Climate 26: Riseth, J.A., H. Tømmervik, E. Helander-Renvall, et al. 2011. Sa´mi 5608–5623. traditional ecological knowledge as a guide to science: Snow, ice Sosnovsky, A.V., P.R. Nakalov, and S.V. Nenashev. 2014. Physical- and reindeer pasture facing climate change. Polar Record 47: geographical aspects of formation of artificial firn-ice massives. 202–217. Ice and Snow 54: 113–119. Rouse, W.R., P.D. Blanken, C.R. Duguay, C.J. Oswald, and W.M. Stacheder, M. 2005. TDR and low-frequency measurements for Schertzer. 2008. Climate lake interactions. In Cold region continuous monitoring of moisture and density in a snow pack. atmospheric and hydrologic studies: The Mackenzie GEWEX International Agrophysics 19: 75–78. experience, vol. 2, ed. M.K. Woo, 139–160. Berlin: Springer. Steffen, K., S.V. Nghiem, R. Huff, and G. Neumann. 2004. The melt Roy, A., G. Picard, A. Royer, et al. 2013. Brightness temperature anomaly of 2002 on the Greenland Ice Sheet from active and simulations of the Canadian seasonal snowpack driven by passive microwave satellite observations. Geophysical Research measurements of the snow specific surface area. IEEE Transac- Letters. doi:10.1029/2004GL020444. tions on Geoscience and Remote Sensing 51: 4692–4704. Stien, A., R.A. Ims, S.D. Albon, et al. 2012. Congruent responses to Rumpf, S.B., P.R. Semenchuk, S. Dullinger, and E.J. Cooper. 2014. weather variability in high arctic herbivores. Biology Letters 8: Idiosyncratic responses of high Arctic plants to changing snow 1002–1005. regimes. PLoS ONE 9: 10. Sturm, M., J.A. Maslanik, D.K. Perovich, J.C. Stroeve, J. Richter- Saloranta, T.M. 2012. Simulating snow maps for Norway: Descrip- Menge, T. Markus, J. Holmgren, J.F. Heinrichs, and K. Tape. tion and statistical evaluation of the seNorge snow model. The 2006. Snow depth and ice thickness measurements from the Cryosphere 6: 1323–1337. Beaufort and Chukchi seas collected during the AMSR-Ice03

Ó The Author(s) 2016. This article is published with open access at Springerlink.com 123 www.kva.se/en Ambio 2016, 45:516–537 535

campaign. IEEE Transactions on Geoscience and Remote changes in winter climate and the impacts of extreme weather events Sensing 44: 3009–3020. for dwarf shrubs, soil arthropods, and decomposition. Surdu, C.M., C.R. Duguay, L.C. Brown, and D. Fernańdez Prieto. Address: FRAM – High North Research Centre on Climate and the 2014. Response of ice cover on shallow lakes of the North Slope Environment, Norwegian Institute for Nature Research (NINA), PO of Alaska to contemporary climate conditions (1950–2011): Box 6606, Langnes 9296, Tromsø, Norway. Radar remote-sensing and numerical modeling data analysis. The Address: Department of Ecological Sciences, Vrije Universiteit Cryosphere 8: 167–180. Amsterdam, De Boelelaan 1085, 1081 HV Amsterdam, The Nether- Takala, M., K. Luojus, J. Pulliainen, et al. 2011. Estimating northern lands. hemisphere snow water equivalent for climate research through e-mail: [email protected]; [email protected] assimilation of space-borne radiometer data and ground-based measurements. Remote Sensing of Environment 115: 3517–3529. Stine Højlund Pedersen is a PhD candidate at Aarhus University as Tang, Q., X. Zhang, X. Yang, and J.A. Francis. 2013. Cold winter part of the Greenland Ecosystem Monitoring program. Her research extremes in northern continents linked to Arctic sea ice loss. interest lies in snow–ecosystem interactions across multiple temporal Environmental Research Letters 8: 014036. and spatial scales using a mix of observations and models. Tedesco, M., X. Fettweis, T. Mote, J. Wahr, P. Alexander, J.E. Box, Address: Department of Bioscience, Arctic Research Centre, Aarhus and B. Wouters. 2013. Evidence and analysis of 2012 Greenland University, Frederiksborgvej 399, 4000 Roskilde, Denmark. records from spaceborne observations, a regional climate model e-mail: [email protected] and reanalysis data. Cryosphere 7: 615–630. Terzago, S., J. von Hardenberg, E. Palazzi, and A. Provenzale. 2014. Ludovic Brucker received his Ph.D. degree in Earth System Sciences Snowpack changes in the Hindu Kush-Karakoram-Himalaya from the University of Grenoble/Centre National de la Recherche from CMIP5 global climate models. Journal of Hydrometeorol- Scientifique, Grenoble, France, in 2009. Since then, he has been with ogy 15: 2293–2313. the NASA Goddard Space Flight Center, Greenbelt, MD, USA. He Thackeray, C.W., C.G. Fletcher, and C. Derksen. 2014. The influence has participated in polar deployments in Antarctica, Greenland, and of canopy snow parameterizations on snow albedo feedback in northern Canada. His research interests include the investigation of boreal forest regions. Journal of Geophysical Research: Atmo- climate evolution in polar regions by interpreting spaceborne micro- spheres 119: 9810–9821. wave observations of snow-covered surfaces (terrestrial snowpacks, Urban, M., M. Forkel, J. Eberle, C. Huettich, C. Schmullius, and M. sea ice, and ice sheet), the comprehension of the relationships Herold. 2014. Pan-Arctic climate and land cover trends derived between passive microwave airborne and spaceborne observations from multi-variate and multi-scale analyses (1981–2012). Re- and snow/ice physical properties using modelling approaches to mote Sensing 6: 2296–2316. provide climate-related variables to the community for the satellite Van Den Broeke, M., C. Bus, J. Ettema, and P. Smeets. 2010. era. Temperature thresholds for degree-day modelling of Greenland Address: NASA GSFC Cryospheric Sciences Laboratory, Code 615, ice sheet melt rates. Geophysical Research Letters 37: L18501. Greenbelt, MD 20771, USA. Van Herwijnen, A., and J. Schweizer. 2011. Seismic sensor array for Address: Goddard Earth Sciences Technology and Research Studies monitoring an avalanche start zone: Design, deployment and and Investigations, Universities Space Research Association, preliminary results. Journal of Glaciology 57: 267–276. Columbia, MD 21044, USA. Vikhamar-Schuler, D., I. Hanssen-Bauer, T.V. Schuler, S.D. Mathie- e-mail: [email protected] sen, and M. Lehning. 2013. Use of a multilayer snow model to assess grazing conditions for reindeer. Annals of Glaciology 54: Oleg Anisimov is a Professor of Physical Geography at the Hydro- 214–226. logical Institute in St. Petersburg, Russia, the coordinating lead author Vionnet, V., E. Brun, S. Morin, et al. 2012. The detailed snowpack of the Polar regions chapters in the Third, Fourth and Fifth IPCC scheme Crocus and its implementation in SURFEX v7.2. reports and the lead author of the international scientific assessments Geoscientific Model Development 5: 773–791. focused on the Arctic (ACIA, SWIPA). Walsh, J.E. 2014. Intensified warming of the Arctic: Causes and Address: State Hydrological Institute of Roshydromet, 23 Second impacts on middle latitudes. Global and Planetary Change 117: Line V.O., St.Petersburg, Russia 199053. 52–63. Address: International Centre for Science and Education ‘‘Best’’, Webster, M.A., I.G. Rigor, S.V. Nghiem, N.T. Kurtz, S.L. Farrell, North-East Federal University, Yakutsk, Russia. D.K. Perovich, and M. Sturm. 2014. Interdecadal changes in e-mail: [email protected] snow depth on Arctic sea ice. Journal of Geophysical Research- Oceans 119: 5395–5406. Jarle W. Bjerke is a Senior Researcher at the Norwegian Institute for Wever, N., C. Fierz, C. Mitterer, H. Hirashima, and M. Lehning. Nature Research. His research interests include the impacts of climate 2014. Solving Richards Equation for snow improves snowpack change and land use on vegetation and terrestrial ecosystems of cold meltwater runoff estimations in detailed multi-layer snowpack biomes. model. The Cryosphere 8: 257–274. Address: FRAM – High North Research Centre on Climate and the Wilson, R.R., A. Bartsch, K. Joly, J.H. Reynolds, A. Orlando, and Environment, Norwegian Institute for Nature Research (NINA), PO W.M. Loya. 2013. Frequency, timing, extent, and size of winter Box 6606, Langnes 9296, Tromsø, Norway. thaw-refreeze events in Alaska 2001–2008 detected by remotely e-mail: [email protected] sensed microwave backscatter data. Polar Biology 36: 419–426.

AUTHOR BIOGRAPHIES Ross D. Brown is a Research Scientist with the Climate Research Division of Environment Canada located at the Ouranos Climate Stef Bokhorst (&) is a Post-Doctoral Researcher at the VU Consortium in Montreal, Canada. His research interests include docu- University Amsterdam, Netherlands and at NINA in Tromsø. His menting and understanding snow-cover variability and change, the research interests include the above and belowground response of representation of snow processes in climate and hydrological models Polar ecosystems to climate change, having great interest in the and the validation of snow cover in regional and global climate models.

Ó The Author(s) 2016. This article is published with open access at Springerlink.com www.kva.se/en 123 536 Ambio 2016, 45:516–537

Address: Climate Research Division, Environment Canada Ouranos, ecosystems and plant–herbivore interactions. 550 Sherbrooke St. West, 19th Floor, Montreal, QC H3A 1B9, Canada. Address: University Centre in Svalbard, PO Box 156, 9171 e-mail: [email protected] Longyearbyen, Norway. Address: Faculty of Life- and Environmental Sciences, University of Dorothee Ehrich is an Arctic Ecologist and her research focuses on Iceland, Sturlugata 7, 101 Reykjavı´k, Iceland. food web interactions in the Arctic tundra, ecosystem monitoring in e-mail: [email protected] the context of climate change, and natural resource use in Arctic terrestrial ecosystems. Her experience includes field work mostly in Niila Inga is Reindeer Herder and Chair of Laevas Sami community, Russia, but also in Canada, northern Norway, and Svalbard, as well as northern Sweden. He is engaged in several research projects dealing the use of molecular ecological methods such as genetics and with the effect of climate and land-use change on reindeer husbandry. stable isotopes. She works also on developing a collaborative moni- Address: Leavas Sa´mi Community, Box 53, 981 21 Kiruna, Sweden. toring programme for the tundra ecosystem between northern Norway e-mail: [email protected] and several sites in Russia. Address: Department of Arctic and Marine Biology, University of Kari Luojus is a Senior Research Scientist at the Arctic Research Tromsø, 9037 Tromsø, Norway. Division of Finnish Meteorological Institute in Helsinki, Finland. His e-mail: [email protected] research interests include the development of active and passive microwave remote sensing techniques for cryosphere and in particular Richard L. H. Essery is a Reader in the School of GeoSciences at the snow-cover monitoring. He has worked on several international University of Edinburgh with research interests in modelling and projects on satellite snow remote sensing and also worked as the observation of land–atmosphere interactions in cold regions. project manager for the ESA GlobSnow-1/2 projects between 2008 Address: School of GeoSciences, University of Edinburgh, Edin- and 2014, which focused on constructing long-term essential climate burgh, UK. variables concerning terrestrial snow cover. e-mail: [email protected] Address: Arctic Research, Finnish Meteorological Institute, P.O. Box 503, 00101 Helsinki, Finland. Achim Heilig received a Diploma in Physical Geography from LMU e-mail: kari.luojus@fmi.fi Munich in 2005 and the Dr. rer. nat. in Environmental Physics in 2009 from the University of Heidelberg, Germany. He is currently pursuing Giovanni Macelloni is a Senior Scientist at IFAC-CNR, Florence. Postdoc at WSL in Davos, Switzerland, and LMU Munich. He has also His main research interest includes microwave active and passive used electromagnetic wave technologies to non-destructively record remote sensing for the study of the cryosphere using data from temporal changes in snow, firn and ice in alpine and polar regions. ground, airborne and satellite. He has been the leader of several Address: Institute of Environmental Physics, University of Heidelberg, national and international programmes granted by the National and Im Neuenheimer Feld 229, 69120 Heidelberg, Germany. International and Entities (PNRA, ESA, ASI, NASA and NASDA) e-mail: [email protected] and has also participated in the development and assessment of future spaceborne missions for studying the cryosphere. Susanne Ingvander received her PhD in 2011 from Stockholm Address: IFAC-CNR - Institute of Applied Physics ‘‘Nello Carrara’’, University. Her PhD focused on ground truth validation of surface National Research Council, Via Madonna del Piano 10, 50019 Sesto snow conditions at the Antarctic plateau. She is currently leading a Fiorentino, FI, Italy. research project funded by the Swedish National Space Board e-mail: [email protected] studying spatial and temporal variability of the snow pack in the Swedish mountain range using remote sensing. Address: Department of Physical Geography, Stockholm University, Heather Mariash is an Aquatic Biologist with a focus on Arctic 106 91 Stockholm, Sweden. inland waters, specializing in water chemistry, food web dynamics e-mail: [email protected] and using biochemical-tracing techniques. She holds a Garfield Weston Post-Doctoral Fellowship for Northern Studies, and is active Cecilia Johansson is a Senior Lecturer in Meteorology at the in the Association for Polar Early Career Scientists (APECS). Department of Earth Sciences, Uppsala University, Sweden. Her Address: National Wildlife Research Centre, Environment Canada, research interest includes boundary layer meteorology, climatology 1125 Colonel By Drive, Ottawa K1A 0H3, Canada. and snow physics. e-mail: [email protected] Address: Department of Earth Sciences, Uppsala University, Villa- va¨gen 16, 75236 Uppsala, Sweden. Donald McLennan is the Head of Monitoring Science at the Cana- e-mail: [email protected] dian High Arctic Research Station-Polar Knowledge Canada in Cambridge Bay, Nunavut. He has been directly involved in arctic Margareta Johansson is a Researcher at the Department of Earth ecosystem monitoring and mapping, both in his recent position at and Ecosystem Sciences, Lund University and at the Royal Swedish CHARS, and for 12 years as Head of Ecological Integrity Monitoring Academy of Sciences, Stockholm, Sweden. She has specialized in for Canada’s National Park system. His background as an applied permafrost dynamics in relation to climate change and its impact on scientist is in the mapping, monitoring and interpretation of landscape ecosystems. scale pattern and process for scientific and land management appli- Address: Department of Physical Geography and Ecosystem Science, cations. Lund University, So¨lvegatan 12, 223 62 Lund, Sweden. Address: Canadian High Arctic Research Station (CHARS), 360 Address: Royal Swedish Academy of Sciences, PO Box 50005, 104 Albert Street, Suite 1710, Ottawa, ON K1R 7X7, Canada. 05 Stockholm, Sweden. e-mail: [email protected] e-mail: [email protected] Gunhild Ninis Rosqvist is a Professor in Geography at Stockholm Ingibjo¨rg Svala Jońsdo´ttir is a Professor in Ecology at the University, Sweden, Professor II at University of Bergen, Norway, and University Centre in Svalbard and the University of Iceland. Her the Director of Tarfala Research Station, northern Sweden. She has research interests focus on impact of climate change on plants and established links between cross-disciplinary science and Sami

Ó The Author(s) 2016. This article is published with open access at Springerlink.com 123 www.kva.se/en Ambio 2016, 45:516–537 537 knowledge to develop research focusing on the effect of multiple and balance of the snow cover, physical processes in snow and the cumulative impacts of climate and land-use change on mountain quantification of snow-related natural hazards. ecosystems. Address: Arctic Environment Laboratory, Faculty of Geography, Address: Department of Physical Geography, Stockholm University, 106 M.V. Lomonosov Moscow State University, Leninskie gory 1, 91 Stockholm, Sweden. Moscow, Russia 119991. Address: Department of Earth Sciences, University of Bergen, 5020 e-mail: [email protected] Bergen, Norway. e-mail: [email protected] Silvia Terzago is a Post-Doctoral researcher at the Institute of Atmospheric Sciences and Climate of the Italian National Research Atsushi Sato is a senior expert researcher at the Snow and ice Council in Torino. Her expertise is on climate variability and change Research Center of the National Research Institute for Earch Science in high elevation areas, with a focus on snow-related processes. She and Disaster Prevention. His interests include physical snow charac- has worked on the representation of snowpack dynamics in land- teristics and their impact on avalanche risk. surface models and on the assessment of the models uncertainties Address: Snow and Ice Research Center, National Research Institute when they are used in ‘‘stand-alone’’ configuration or within global for Earth Science and Disaster Prevention, 187-16 Suyoshi, Nagaoka, climate models. Niigata 940-0821, Japan. Address: Institute of Atmospheric Sciences and Climate, National e-mail: [email protected] Research Council (ISAC-CNR), Corso Fiume 4, 10133 Turin, Italy. e-mail: [email protected] Hannele Savela completed her PhD in Physiological Zoology from the University of Oulu, Finland, in 2005. Her research focused on experi- Dagrun Vikhamar-Schuler is a Senior Scientist at MET Norway mental and field studies in applied animal physiology of one of the most with a PhD in Geosciences (remote sensing and snow modelling). Her iconic Arctic animals—the reindeer. She is currently working at Thule research focuses mainly on analysis of winter climate and snow Institute in the University of Oulu as the Transnational Access Coor- conditions for various impact studies, namely snow avalanche dinator of INTERACT (International Network for Terrestrial Research warning, hydrological applications, geohazards, permafrost, and and Monitoring in the Arctic) and as the Research Area Coordinator for ecology (impact of snow cover on plants, reindeer, crops). She has the University of the Arctic. She is one of the co-leads in the Group on been involved in numerous national and international projects dealing Earth Observations (GEO) Cold Regions Initiative. with analysis of past, present and future climate. Her specialities are Address: Thule Insitute, University of Oulu, PO Box 7300, 90014 Oulu, land-surface modelling with emphasis on snow and soil schemes, and Finland. Arctic winter climate. e-mail: hannele.savela@oulu.fi Address: Division for Model and Climate Analysis, R&D Depart- ment, The Norwegian Meteorological Institute, Postboks 43, Blin- dern, 0313 Oslo, Norway. Martin Schneebeli received his Dipl.Ing. and Dr.Sc. degrees in e-mail: [email protected] Environmental Engineering from ETH Zurich, Switzerland, in 1984 and 1991, respectively. He is currently a Group Leader of snow physics at WSL and lecturer at ETH Zurich. He developed several Scott Williamson is a Post-Doctoral Fellow at the University of new methods to measure the microstructure and stratigraphy of snow Alberta. His research interests include trying to understand how cli- quantitatively in the field and in the laboratory. mate and cryosphere influence each other. Address: WSL Institute for Snow and Avalanche Research SLF, Address: Department of Biological Sciences, University of Alberta, Fluëlastrasse 11, 7260 Davos Dorf, Switzerland. CW 405, Biological Sciences Bldg., Edmonton, AB T6G 2E9, e-mail: [email protected] Canada. e-mail: [email protected] Aleksandr Sokolov defend his PhD thesis in 2003 on small rodents and birds of prey interactions. Since 1999, he has been leading a field group at Yubao Qiu is an associate researcher at the Institute of Remote Tundra monitoring site ‘‘Erkuta’’ in southern Yamal, where ecosystem- Sensing and Digital Earth where he uses remote sensing to quantify based monitoring of different groups of organisms was conducted on changes in snow characteristics. year-round basis. He has also led the expeditions to remote areas of Address: Institute of Remote Sensing and Digital Earth, Chinese Yamal, Taymir, Lena Delta and Kolyma Delta. He has been the leader of Academic of Science, Beijing 100094, China. Yamal group of researchers at IPY projects ‘‘Arctic predators’’ and Address: Group on Earth Observations, Cold Regions Initiative, ‘‘Arctic WOLVES’’and also an active participant of International World Geneva, Switzerland. Working Group on Snowy Owl. He has also been the station manager of Labytnangi Research Station in the project ‘‘INTERACT’’. Terry V. Callaghan is a Distinguished Research Professor at the Address: Arctic Research Station of Institute of Plant and Animal Royal Swedish Academy of Sciences and Professor of Arctic Ecology Ecology, Ural Branch, Russian Academy of Sciences, Labytnangi, at Universities of Sheffield, UK and Tomsk, Russia. He has special- Russia 629400. ized in arctic ecology, and climate and UV-B radiation impacts on Address: Science Center for Arctic Studies, State Organization of Yamal- arctic ecosystems. Nenets Autonomous District, Salekhard, Russia. Address: Department of Physical Geography and Ecosystem Science, e-mail: [email protected] Lund University, So¨lvegatan 12, 223 62 Lund, Sweden. Address: Department of Animal and Plant Sciences, University of Sergey A. Sokratov is senior research scientist in Natural Risks Sheffield, Sheffield S10 2TN, UK. Assessment Laboratory and in the Research Laboratory of Snow Address: National Research Tomsk Stated University, 36, Lenin Ave., Avalanches and Debris Flows, Faculty of Geography, Lomonosov Tomsk, Russia 634050. Moscow State University. His interests include mass and energy e-mail: [email protected]

Ó The Author(s) 2016. This article is published with open access at Springerlink.com www.kva.se/en 123 Paper II

Spatiotemporal characteristics of seasonal snow cover in Northeast Greenland from in situ observations.

Pedersen, S. H., Tamstorf, M. P., Abermann, J., Westergaard-Nielsen, A., Lund, M., Skov, K. Sigsgaard, C., Mylius, M. R., Hansen, B. U., Liston, G. E., Schmidt, N. M. (2016). Arctic, Antarctic, and Alpine Research. Vol. 48, No. 4, 2016, pp. 653–671. DOI:http://dx.doi.org/10.1657/AAAR0016-028.

Miniscule snow drifts behind pebble stones, NE-Greenland (2012)

Arctic, Antarctic, and Alpine Research, Vol. 48, No. 4, 2016, pp. 653–671 DOI: http://dx.doi.org/10.1657/AAAR0016-028

Spatiotemporal characteristics of seasonal snow cover in Northeast Greenland from in situ observations Stine Højlund Pedersen1,*, Mikkel P. Tamstorf1, Jakob Abermann2, Andreas Westergaard-Nielsen3, Magnus Lund1, Kirstine Skov3, Charlotte Sigsgaard3, Maria Rask Mylius3, Birger Ulf Hansen3, Glen E. Liston4, and Niels Martin Schmidt1 1Arctic Research Centre, Department of Bioscience, Aarhus University, Frederiksborgvej 399, DK-4000 Roskilde, Denmark 2Asiaq, Greenland Survey, Qatserisut 8, GL-3900 Nuuk, Greenland 3Center for Permafrost (CENPERM), University of Copenhagen, Department of Geosciences and Natural Resource Management, Oester Voldgade 10, DK-1350 Copenhagen K, Denmark 4Cooperative Institute for Research in the Atmosphere (CIRA), Colorado State University, Fort Collins, Colorado 80523, U.S.A. *Corresponding author’s email: [email protected]

ABSTRACT In this study, we quantified the spatiotemporal variability and trends in observations of multiple snow characteristics in High Arctic Zackenberg in Northeast Greenland through 18 years. Annual premelt snow-depth observations collected in 2005–2014 along an elevation gradient showed significant differences in snow depth between vegetation types. The seasonal snow cover was characterized by strong interannual variability in the Zackenberg region. Particularly the timing of snow-cover onset and melt, and the annual maximum accumulation, varied up to an order of magnitude between years. Hence, apart from the snow-cover fraction registered annually on 10 June, which exhib- its a significant trend of –2.3% per year over the 18-year period, we found little evidence of significant trends in the observed snow-cover characteristics. Moreover, SnowModel results for the Zackenberg region confirmed that the pronounced interannual variability in snow precipitations has persisted in this High Arctic setting since 1979 and may have masked potential temporal trends. In exception, a significant difference in interannual variability of snow-cover onset timing was observed through the period 1997–2014, which in the recent period since 2006 was 7.3 times more variable.

Introduction erage in the pan-Arctic area since the start of the satellite era in 1979 (Tedesco et al., 2009). Snow Within the last decades, the Arctic terrestrial model results for the terrestrial pan-Arctic region snow cover has undergone changes, both in thick- support in general an earlier snow-free date in ness and duration. The snow-cover extent (SCE) in spring and show that the maximum winter snow- the Arctic region has decreased by approximately water equivalent (SWE) decreased, the snow-cover 20% per decade during 1979–2013 (Blunden and onset occurred later in the autumn/early winter, Arndt, 2014). Moreover, the observed SCE reduc- and the length of the snow-covered period was re- tion rate exceeds the simulated and projected rates duced through the period 1979–2009 (Liston and of decreasing SCE, which are based on global cli- Hiemstra, 2011). However, trends vary from posi- mate model ensembles (Derksen and Brown, 2012; tive to negative in regions across the Arctic area for Blunden and Arndt, 2014). In addition, the timing the period 1979–2009 (Liston and Hiemstra, 2011) of snowmelt onset has advanced 2 weeks on av- and 2001–2014 (Chen et al., 2015). Within North

© 2016 Regents of the University of Colorado – 1523-0430/04 $7.00 653 America, Northern Eurasia, and the ice-free area of at the end of the winter and the following spring in Greenland, there are regions with opposing trends tundra ecosystems (Schimel et al., 2004; Buckeridge in, for example, the annual snow precipitation, the and Grogan, 2008). The nutrient availability along timing of onset, and the end of core snow-covered with the meltwater released from the snowpack in period (Liston and Hiemstra, 2011). spring (Jones, 1999) regulate, in turn, the vegetation In this study, we present a ground-based long- growth far into the growing season (Blankinship et term time series of snow observations from Zack- al., 2014) and thus the main food source for herbi- enberg in the High Arctic region of Greenland. vores—for example, the muskoxen (Ovibos moschatus) Only a few ground-based studies using long-term (Kristensen et al., 2011; Schmidt et al., 2015; Mos- snow observations have been conducted in the bacher et al., 2016). Hence, the snow conditions in High Arctic (e.g., Zhang et al., 2000; Bulygina et the preceding winter may have legacy effects on the al., 2011; Dyrrdal et al., 2013; Stuefer et al., 2013). following growing season(s), which makes snow ob- However, these studies are valuable for mapping the servations essential in the understanding of ecosys- diversity in changes and trends in snow variables tem functions and feedbacks (Hollesen et al., 2015). within the pan-Arctic area. Furthermore, ground- Ultimately, due to the governing role of the snow, based observations are requested as calibration and any observed changes, trends, and/or altered variabil- validation data sets for model simulations and re- ity in snow conditions, may explain observed (inter-) mote sensing products and the development of seasonal effects, changes, and variability in the biotic such (Derksen et al., 2014; Bokhorst et al., 2016). components of the ecosystem. To identify significant Finally, since the terrestrial snow cover is a key trends and variability in ground-based snow obser- variable controlling Arctic ecosystem processes, these vations, we examine an 18-year record comprising time series of snow observations are valuable in ex- continuous observations of multiple, both point and plaining changes seen in both biotic and abiotic spatially distributed, seasonal snow metrics collected components of the Arctic ecosystems (e.g., Jones, as part of the Greenland Ecosystem Monitoring pro- 1999; Post et al., 2009; Brooks et al., 2011; Callaghan gram (http://www.g-e-m.dk). We quantify the spa- et al., 2011; Cooper, 2014). The spatial distribution of tial variability, the temporal trends, and interannual snow depth is primarily driven by a combination of as well as seasonal variation in a suite of ecologi- the dominant wind direction during and after snow- cally relevant snow variables, which are all observed fall (Liston and Sturm, 1998; Winstral et al., 2002), the during the period 1997 through 2014. Additionally, topographic relief and slope orientation (Schirmer we investigate whether the snow variables showed a et al., 2011), and the vegetation cover, which traps change in interannual variability through the study wind-transported snow (Sturm et al., 2001). In ad- period. Finally, modeled winter snow amounts for dition, during the melt season, the snow-depth dis- the period 1979–2014 allow us to explore whether tribution is controlled by the availability of melt en- the observed temporal variability and trends in the ergy, which is governed by terrain elevation, slope, past 18 years of snow observations are unique in and aspect (Clark et al., 2011). Especially during the comparison to previous decades. snow-covered season, when insulating properties of the snow cover (Goodrich, 1982; Sturm et al., 1997; Study Area Liston et al., 2002) provide stable thermal conditions in the below-snow environment, including the veg- The Zackenberg study area is located in the ice- etation cover and soil (Schimel et al., 2004; Zhang, free coastal part of Northeast Greenland (74°27′N, 2005; Bokhorst et al., 2011; Johansson et al., 2013). 20°34′W) (Fig. 1). The area has been covered by The soil thermal conditions moreover drive the mi- the Greenland Ice Sheet several times through geo- crobial activity, the respiration rate, and the amount logical history (Bennike et al., 2008). About 10,000 of soil organic carbon produced during winter (El- years ago, the lowland surrounding Zackenberg berling, 2007), and it controls the active-layer depth became ice-free, and today the ice sheet margin is (Westermann et al., 2015). Hence, the snow-depth located 70 km west of Zackenberg. The landscape evolution through the autumn and winter governs in and surrounding Zackenberg is highly hetero- the amount and timing of plant-available nutrients geneous in terms of bedrock and sediment type,

654 / stine Højlund Pedersen et al. / arctic, Antarctic, and Alpine Research FIGURE 1. Zackenberg study area (QuickBird image from 31 May 2012) showing locations of climate stations C1, M2, and M3 (orange squares); ZERO-line transect (black line); the two regions Valley floor and Hill slope (blue line); the region for annual snow-cover estimations on 10 June (red line); and Zackenberg Research Station, ZAC (yellow triangle). and the topography is characterized by deep fjords produces this asymmetry). The predominant wind and valleys with elevations varying from sea level to direction is north during winter and southeast dur- mountain peaks reaching 1000–1400 meters above ing summer, the latter mainly being triggered by sea level (m a.s.l.). Zackenberg is located within the land-sea interaction. The annual average pre- the continuous permafrost zone, and the landscape cipitation (snow and rain) for the hydrological years development is thus dominated by periglacial pro- in Zackenberg (defined here to be 1 September to cesses (Westermann et al., 2015). 31 August and denoted, e.g., “2012/2013”) was The Zackenberg study area is situated in the High 367 mm (1996–2014) varying between 222 mm Arctic zone (Bliss and Matveyeva, 1992) with a tun- (2003/2004) and 547 mm (2013/2014). dra climate (Kottek et al., 2006), which is mainly governed by the proximity to the Greenland Ice Sheet to the west and the Greenland Sea to the east. Methods For the period 1997 through 2014, the monthly mean temperatures were between –19.8 °C (Febru- Snow Quantities and Snow-Cover ary) and 6.3 °C (July), and the yearly average tem- Timing perature was –9.0 °C. There is polar night during The snow variables observed in Zackenberg in- 89 days and polar day during 106 days (refraction clude snow cover (i.e., presence or absence), snow

Arctic, Antarctic, and Alpine Research / stine Højlund Pedersen et al. / 655 TABLE 1 Snow observations collected in Zackenberg by GeoBasis and ClimateBasis. All station locations are indicated in Figure 1. Data collection procedures are described by Sigsgaard et al. (2014). All data are available from http:// www.zackenberg.dk, and data descriptions are given in ZERO Annual Reports 1–18 (http://www.zackenberg. dk/publications/annual-reports/). Snow variable Unit Accuracy Time series Sampling method Snow depth m ±0.01 1997–2014 (C1) 3-hourly measurements by automated sonic ranging snow- (sensor) 2003–2014 (M2) depth sensor (Campbell SR50a) located at the climate stations: C1 (74°28′20″N, 20°33′08″W, 38 m a.s.l.), M2 (74°27′56″N, 2003–2014 (M3) 20°33′47″W, 19 m a.s.l.), and M3 (74°30′11″N, 20°27′35″W, 410 m a.s.l.).

Snow density kg m–3 * 2004–2014 Estimated in snow pits through layers using a RIP cutter, 250 cm3 (http://www.snowmetrics.com), density cutter, 100 cm3 (http:// www.snowhydro.com) or ring samples (100–300 cm3), and for full depth snow packs using Standard Federal Snow Sampler (Clyde, 1932). The density samples were collected only in March– November, when Zackenberg Research Station was manned.

Snow-water m w.e. * 2004–2014 From snow pit total depth and bulk snow density or from snow equivalent core length and snow density collected with Standard Federal Snow Sampler.

Snowfall # — 1997–2014 Snow-depth increase from snow-depth sensor at C1. events

Snow depth m ±0.02 2005–2014 Premelt snow depth point measurements along the ZERO-line (ZERO-line) transect (Fig. 1) with GPS/MagnaProbe (http://www.snowhydro. com) every 10 m, or ground penetrating radar with a 500 MHz shielded antenna (http://www.malags.com).

Snow-cover % — 1997–2014 Based on georeferenced, orthorectified, and snow-classified photos fraction from automatic cameras (Buus-Hinkler et al., 2006). 10 June snow cover was estimated for a 47 km2 region (Fig. 1).

*Accuracy varies with, e.g., snowpack stratigraphy and density cutter (Proksch et al., 2016). depth and snow density, hence snow-water equiva- quantification of the temporal trends and interan- lent (SWE), and snowfall events. The metadata and nual variability, are described in the section Statisti- sampling methods for these variables are described cal Analysis. in Table 1, and the data are given in Tables 2 and 3. To quantify the changes, trends, and temporal All observations are made by the climate and geo- variability in the observed snow variables, we have physical monitoring programs, ClimateBasis and calculated several snow metrics. Snowfall events GeoBasis, respectively. The spatial analysis of snow (large snowfall events) were defined as times with a cover and topographic features was performed for snow-depth increase (at climate station C1, Fig. 1) two separate regions, Valley floor and Hill slope, of more than 0.05 m (0.20 m) from one day to the where the snow-cover monitoring was conducted next. In this way, snowfall events also include the by GeoBasis (Fig. 1). The sampling methods were large snowfall events. The thresholds (0.05 m and kept constant throughout the time series to allow 0.20 m) were set according to the local snow con- for interannual comparison (Sigsgaard et al., 2014). ditions, where the snow depth varied up to ±0.05 The statistical methods, which are used for the m from day to day because blowing snow was de-

656 / stine Højlund Pedersen et al. / arctic, Antarctic, and Alpine Research TABLE 2 Snow variables observed in Zackenberg from 1997 through 2014. Large Maximum Hydrological 10 June Snow-cover Snow- Snow- Snowfall snowfall Mean snow snow depth year snow cover duration cover end cover onset events events depth (C1) (C1) (1 Sep–31 Aug) (%) (days) (DOY) (DOY) (#) (#) (m) (m) 1997/1998 80 219 176 322 10 2 0.36 0.89 1998/1999 92 250 184 299 11 4 0.48 1.30 1999/2000 54 166 166 365 6 1 0.17 0.49 2000/2001 82 221 175 319 8 0 0.27 0.69 2001/2002 77 214 171 322 15 2 0.50 1.33 2002/2003 83 191 165 339 7 0 0.07 0.60 2003/2004 49 203 165 327 5 0 0.24 0.69 2004/2005 37 163 158 360 7 0 0.19 0.74 2005/2006 77 195 182 352 16 1 0.40 1.09 2006/2007 43 148 159 11 8 1 0.15 0.55 2007/2008 65 243 176 298 15 2 0.48 1.30 2008/2009 28 85 136 51 4 0 0.04 0.18 2009/2010 44 265 167 267 11 1 0.30 0.73 2010/2011 46 136 161 25 9 0 0.11 0.45 2011/2012 78 257 178 286 11 2 0.47 1.30 2012/2013 3 74 138 64 0 0 0.05 0.13 2013/2014 77 187 175 353 7 1 0.42 0.90

DOY = day of year. C1 = climate station. posited and eroded from the snow surface and not in the observational data set. These point measure- because of snowfall. The snow-depth threshold for ments are considered representative for the Valley determining whether the ground was continuously floor region (Fig. 1), because it is a flat and homo- snow-covered or snow-free was defined to be 0.10 geneous area, and also in terms of vegetation type m. This height equaled the maximum vegetation and cover. height and microtopographic relief in the Valley floor region surrounding the snow-depth sensor at C1 (Fig. 1). Further, the snow-cover fraction for the Spatial Distribution of Snow Valley floor region (described in the section Spatial To describe the spatial distribution of snow in the Distribution of Snow) of 90%–100% coincided in landscape in Zackenberg, we investigated the snow- time with snow depths above 0.10 m being meas- depth temporal and spatial variability along an eleva- ured at the automated snow-depth sensor at C1. tion gradient on a mountain slope (ZERO-line) and Snow-cover onset is, herein, defined as the first day the spatial snow-cover depletion in the two confined of a period, where the snow depth at C1 is con- regions, Valley floor and Hill slope. The monitoring tinuously above 0.10 m. Snow-cover end is the last transect, ZERO-line (Fredskild and Mogensen, 1997; day in this period—that is, the day when the snow Meltofte et al., 2008a), was established in 1992 on a depth decreases below 0.10 m. The core snow- southwest-facing slope covering an elevation gradi- covered period corresponds to the period between ent from sea level to 420 m a.s.l. (Fig. 1). In order snow-cover onset and end and is the longest con- to get an estimate of the snow accumulation dur- tinuous period in a winter when the snow depth is ing the preceding winter, snow-depth observations larger than 0.10 m. All metrics are calculated for the have been annually collected since 2005 along ZE- longest snow-depth time series from C1 (Table 1) RO-line at the end of winter or early spring before

Arctic, Antarctic, and Alpine Research / stine Højlund Pedersen et al. / 657 TABLE 3 Timing and duration of the snowmelt periods in the Valley floor region and Hill slope region. Valley floor Valley floor Valley floor Hill slope Hill slope Hill slope Valley timing of timing of timing of timing of timing of timing of floor Hill slope Hydrological 20% snow 50% snow 80% snow 20% snow 50% snow 80% snow snowmelt snowmelt year cover cover cover cover cover cover duration duration (1 Sep–31 Aug) (DOY) (DOY) (DOY) (DOY) (DOY) (DOY) (days) (days) 1997/1998 — 173 — 212 186 173 — 39 1998/1999 187 184 180 228 188 181 7 47 1999/2000 168 162 160 178 168 160 8 18 2000/2001 177 175 174 196 186 181 3 15 2001/2002 177 170 164 197 178 160 13 37 2002/2003 165 164 158 165 161 147 7 18 2003/2004 164 158 157 174 158 156 7 18 2004/2005 160 155 154 171 163 153 6 18 2005/2006 186 170 175 210 196 177 11 33 2006/2007 165 162 157 174 165 156 8 18 2007/2008 178 174 167 185 171 163 11 22 2008/2009 151 149 143 174 164 159 8 15 2009/2010 168 167 162 175 167 156 6 19 2010/2011 166 162 158 179 168 160 8 19 2011/2012 180 177 170 195 179 166 10 29 2012/2013 150 148 139 150 148 139 11 11 2013/2014 178 169 165 203 180 167 13 36

DOY = day of year. substantial melting had occurred in the snowpack. 2005–2014, we extracted the corresponding veg- One premelt snow-depth transect per year was in- etation type from a vegetation map by Elberling et cluded in the analysis for the period 2005–2014. To al. (2008), including Dryas heath, fell-field, fen,Cas - verify that no or only limited melting had occurred siope heath, grassland, and Salix snowbed. Likewise, previous to the snow-depth measurements, we de- we extracted the elevation from a digital elevation fined two requirements: (1) the positive degree-day model (DEM) with a 10-m horizontal resolution (PDD) sum during spring in Zackenberg should be (originally rescaled from a DEM with 0.10-m hor- less than 5 PDDs, and (2) the cumulated snow-depth izontal and vertical resolution based on digital pho- decrease should be less than 0.07 m from the day of tos taken from drone flights over Zackenberg Valley maximum snow depth to the day when the ZERO- in summer 2012). This 10-m resolution adequately line observations were made. The 0.07 m threshold described the Zackenberg area landscape features. corresponded to our observed maximum premelt The temporal evolution of the snow-cover snowpack settling. For these estimates, we used air fractions for the Valley floor and Hill slope regions temperature and snow-depth observations recorded was estimated during the snowmelt season, to de- at station C1 (Table 1, Fig. 1). All years in the period rive the timing of the spatially distributed spring 2005–2014 met these requirements. snowmelt and to investigate its change and vari- To investigate variations in the snow distribution ability through the study period. For this, we used amongst vegetation types and along the altitudinal digital photos, which were daily acquired from gradient, we used the ZERO-line end-of-winter April through August by an automated camera. snow depths (Table 1). For each snow-depth meas- These photos were geo-referenced and orthorec- urement point along the ZERO-line in the period tified by applying the method of Buus-Hinkler et

658 / stine Højlund Pedersen et al. / arctic, Antarctic, and Alpine Research al. (2006) and also used by Mernild et al. (2007), data from ERA-Interim (Dee et al., 2011). This and subsequently we performed a snow-classifi- model setup, including a precipitation correction cation in the image-processing program ENVI by assimilating observed SWE using SnowAs- (http://www.exelisvis.com). Similarly, the snow- sim (Liston and Hiemstra, 2008) for a terrestrial cover fraction was estimated on 10 June within a Greenland domain, is described in detail by Ped- region covering 47 km2 of the Zackenberg Valley ersen et al. (2015). For the period 2004–2014, and the eastern mountain slopes from sea level to modeled SWE was extracted from the grid cell 880 m a.s.l. (see region marked in red in Fig. 1). corresponding to the location of the ground ob- The annual 10 June snow-cover fraction was re- servations of SWE from snow pits dug in the end ferred to as “spring snow cover.” Previous studies of winter (within the Valley floor region in all have demonstrated that this date can be linked to years) and at the same date as the ground ob- a range of ecological processes in the tundra eco- servations were made. We compared the variabil- system in Zackenberg (e.g., Forchhammer et al., ity of the observed and modeled end-of-winter 2005; Schmidt et al., 2006; Meltofte et al., 2008b; SWE to validate the ability of SnowModel to Pellissier et al., 2013). The timing of 80%, 50%, reproduce the interannual variability. To quan- and 20% snow cover marked the onset, middle, tify if the variability in annual snow amounts and end of the core snowmelt season, respectively, had changed through the period 1979–2014, we for the regions Valley floor and Hill slope (Fig. 1). compared the temporal variability in modeled Hence, the duration of the core snowmelt period end-of-winter SWE in both 5-year and 11-year was defined as the difference in days between 80% periods within the 35-year modeled time series. and 20% snow cover. In order to obtain a measure of the topograph- Statistical Analysis ic variability that may influence the spatial snow distribution within each region, the mean terrain For the trend analysis we used linear regression slope was calculated using the ArcMap GIS 10.1 (ordinary least squares) between a given variable “Spatial Analyst Surface Slope” tool and is defined and time in order to quantify the direction of the as the average of slopes for all grid cells within a trend (positive or negative) as well as the statisti- region. The terrain ruggedness (TR) was estimated cal significance of the trend (different from 0.0, using a method by Riley et al. (1999) and the GIS i.e., p < 0.05). Furthermore, linear regression was terrain analysis tool “Spatial Analyst Neighborhood also used for finding relationships between snow Focal Statistics.” TR is the summed change in ele- characteristics. Linear relationships between snow- vation between a grid cell and the eight-cell neigh- melt duration in the two regions were quantified borhood (Riley et al., 1999). The mean TR is the with correlation coefficients (r) as well as the co- average TR of all grid cells within a defined region. variance between annual maximum snow depths Both the terrain slope and TR analysis was con- observed at the three Zackenberg meteorologi- ducted on the 10-m horizontal resolution DEM, cal stations (Table 1). F-tests were used to evalu- and the terrain slope was calculated over areas of ate whether the variances of a variable (e.g., tim- 100 m by 100 m. ing of onset or end of snow cover) in two periods were significantly different. We used an ANOVA to examine if snow depth varied among vegetation SnowModel 1979–2014 types and with elevation. Snow-depth data were In order to investigate whether the observed logarithmic transformed—that is, log(snow depth temporal variability and trends in the past 18 + 0.01m)—to stabilize the variances to allow the years of snow observations have changed with application of the ANOVA. To account for the in- respect to the previous decades, we used mod- terannual variation in snow depth, we included the eled SWE to extend the observational time se- “year” as an explanatory variable in the model. As ries. The spatially distributed snow modeling post hoc test we used Tukey’s Honest Significant tools, MicroMet (Liston and Elder, 2006a) and Differences for multiple comparisons of means SnowModel (Liston and Elder, 2006b), were ap- (p < 0.01) to identify for which vegetation-types plied in the Zackenberg Valley using reanalysis the mean snow depth were significantly different

Arctic, Antarctic, and Alpine Research / stine Højlund Pedersen et al. / 659 at a 95% confidence level. We used the software R However, the timing of 50% snow cover in Valley (http://www.r-project.org/) for the statistical analysis. floor was tightly linked with snow-cover duration (R2 = 0.84, F = 60.87, p < 0.001) (Fig. 3, Tables 2 and 3). Results Snow Quantities and Temporal Snow Depth Variation Across years, the 10% highest snow depth (above Snow-Covered Period 90th percentile for each winter) occurred between The 17 winters of snow depth from C1 in February and late May. No temporal trends were Zackenberg showed variable duration of the core observed in either the mean snow depth or the an- snow-covered period (Fig. 2). The maximum nual maximum snow depth or the timing of the snow-cover duration of 265 days was observed in latter through the study period (Table 4). The an- winter 2009/2010 and the minimum duration was nual maximum snow depth had a pronounced in- observed in winter 2012/2013, where the snow terannual variation ranging from 0.13 m in winter depth was above 0.10 m for only 74 consecutive 2012/2013 to 1.33 m in 2001/2002 (Fig. 2). Meas- days. In 2008/2009, the second shortest snow- ured maximum snow depth showed an interannual cover duration of 85 days was registered (Figs. 2 covariance in the period from 2003 through 2014 and 3). Snow-poor winters (e.g., 2008/2009 and between C1 and M3 (covariance (C1, M3) = 0.12, t = 2012/2013) had several shorter periods (2–17 days) 2.533, p = 0.035, r = 0.667), but no correlation with with snow-covered ground before or after the core M2 (data not shown). The highest annual maximum snow-covered period. The onset of the snow-cov- snow depth (2.53 m in 2013/2014) was found at ered period was also highly variable, ranging be- M2, which is installed in a snowdrift, and the lowest tween 24 September and 5 March, through the (0.09 m in 2012/2013) at M3, installed on a south- 17-year period and thus contributed to the high west-facing slope at 410 m a.s.l. (Fig. 1, Table 1). interannual variability in the snow-cover duration (Fig. 3, Table 2). However, the end of the snow- Snow Density and Snow-Water Equivalent covered period (snow-cover end) varied less—that The snow densities observed during the period is, between 16 May (2009) and 3 July (1999). The 2004–2014 ranged between 173 kg m–3 (September variance in timing of snow-cover onset in the early 2009) and 685 kg m–3 (June 2005) and the mean den- period (1997/1998–2005/2006) was significantly sity was 398 kg m–3 ± 104 kg m–3, which compared different (F = 0.124, p < 0.001) from the variance with the bulk density of 380 kg m–3 for the tundra in the most recent period (2006/2007–2013/2014), snow-cover class defined by Sturm et al. (1995). The and the variance was 7.3 times larger in the lat- mean monthly snow density showed an increase ter. This difference was not found for snow-cover through the snow-covered season from September end (F = 0.274, p = 0.090) between the same two through June. The densities in June were in some years periods. Likewise, the variances in the timing of collected from melting snowpacks that may have con- 50% snow cover during spring in the Valley floor tained liquid and/or refrozen meltwater causing the and Hill slope areas were not significantly differ- density increase in June (Fig. 4). SWE, estimated from ent in the two periods 1997/1998–2005/2006 and all available bulk density observations and correspond- 2006/2007–2013/2014 (F = 0.730, p = 0.664 and ing snow depths, had a median of 0.22 m of water F = 1.914, p = 0.408, respectively). Additionally, the equivalent (w.e.) in the period 2004–2014, and the variance in snow-cover onset was approximately an maximum SWE observed in the period was 2.13 m order of magnitude (ratio = 8.7) higher than the w.e. in a 5-m-deep snowdrift (11 April 2008). variance in snow-cover end during the 17 winters. A trend analysis of the snow-cover timing through the period 1998–2014 showed no significant trends Snowfall Events in snow-cover onset, snow-cover end, snow-cover Snowfall events occurred from September duration, or timing of 50% snow cover (Table 4). through May and totaled 150 snowfall events in

660 / stine Højlund Pedersen et al. / arctic, Antarctic, and Alpine Research FIGURE 2. Daily snow depth (m) measured at C1 in the hydrological year (1 September–31 August) for the period 1 September 1997 through 31 August 2014. The gray shaded area marks the core snow-covered period from snow-cover onset (snow depth at C1 is above 0.10 m) to snow-cover end/snowmelt (snow depth at C1 is below 0.10 m), that is, the timing of the longest continuous period with snow depth above 0.10 m for each hydrological year.

Arctic, Antarctic, and Alpine Research / stine Højlund Pedersen et al. / 661 FIGURE 3. Snow-cover fraction estimated from georeferenced, orthorectified, and snow-classified photos taken at 10 June in years 1997–2014 (black points). Linear fit (dashed black regression line) statistics: slope = –2.3, R2 = 0.27, F = 5.803, p = 0.028. Timing of 50% snow cover in Valley floor region, estimated from georeferenced, orthorectified, and snow-classified photos in years 1997–2014 (gray points). Snow-cover duration (days) for each hydrological year defined as the length of the period between snow-cover onset and end, where snow depth was consistently above 0.10 m at C1 (red points).

1997–2014 with the highest frequency (36) in Jan- uary (Fig. 5, total given without parentheses). The highest number of annual snowfall events (16) occurred during the winter 2005/2006, whereas the winters 2001/2002 and 2007/2008 both included 15 snowfall events. Included in the 150 snowfall events were 17 large snowfall events (Fig. 5, totals given in parentheses). The highest frequency of large snowfall events (4) was found in winter 1998/1999, and over the years the month of Feb- ruary had the highest frequency of large snowfall events. There were no significant temporal trends in the number of all snowfall events or large snowfall events through the study period (Table 4). FIGURE 4. Snow bulk density (kg m–3) collected in 2004– 2014 within 5–10 m distance of the three meteorological Spring Snow Cover stations: C1 on the Valley floor, M2 in a snowdrift, and M3 on a southwest-facing mountain slope. The total number The 10 June snow cover showed a pronounced of observations was 260, and most densities (91) were interannual variation through the 18-year period collected in May. No density observations were collected ranging between 3% (7 June 2013) and 92% (10 in November–February because the Zackenberg Research June 1999) (Fig. 3). Furthermore, the 10 June snow- Station is unmanned during these months. The whiskers mark the data ranging between 1st quartile – 1.5 × IQR cover fraction showed a statistically significant de- (the interquartile range) and 3rd quartile + 1.5 × IQR. creasing trend and decreased on average 50% from

662 / stine Højlund Pedersen et al. / arctic, Antarctic, and Alpine Research TABLE 4 Statistical details for temporal trends (significant trend at the 95% confidence level in bold) for snow variables observed in Zackenberg, Northeast Greenland.

Linear Regression Statistics Station/ Slope Snow variable region Period Unit Mean SD R2 F p (year–1) 10 June snow cover SCR 1997–2014 % 61 24 0.27 5.803 0.03 –2.3 Mean annual snow depth C1 1998–2014 m 0.28 0.16 0.05 0.734 0.41 <0.1 Maximum snow depth C1 1998–2014 m 0.79 0.38 0.06 0.984 0.34 <0.1 Max. snow-depth timing C1 1998–2014 date/days 20 Apr 16 0.13 2.272 0.15 –1.1 Snow-cover onset C1 1998–2014 date/days 8 Dec 45 0.08 1.217 0.29 2.3 Snow-cover end C1 1998–2014 date/days 16 Jun 14 0.14 2.427 0.14 –1.0 Snow-cover duration C1 1998–2014 days 189 54 0.11 1.769 0.20 –3.6 Mean annual snow depth M2 2003–2014 m 0.49 0.25 0.23 2.654 0.14 <0.1 Mean annual snow depth M3 2003–2014 m 0.13 0.13 0.28 3.181 0.11 <0.1 Snowfall events, all C1 1998–2014 # 9 4 0.04 0.676 0.42 –0.2 Snowfall events, large C1 1998–2014 # 1 1 0.10 1.724 0.21 –0.1 Snowmelt duration VF 1998–2014 days 9 3 0.19 3.328 0.09 0.3 Snowmelt duration HS 1998–2014 days 24 10 0.10 1.622 0.22 –0.6 Timing of : 80% snow-cover fraction VF 1998–2014 date/days 10 Jun 11 0.16 2.623 0.13 –0.9 50% snow-cover fraction VF 1998–2014 date/days 15 Jun 10 0.14 2.427 0.14 –1.0 20% snow-cover fraction VF 1998–2014 date/days 19 Jun 11 0.07 1.208 0.29 –0.7 80% snow-cover fraction HS 1998–2014 date/days 10 Jun 11 0.17 2.971 0.11 –0.9 50% snow-cover fraction HS 1998–2014 date/days 21 Jun 13 0.13 2.152 0.16 –0.9 20% snow-cover fraction HS 1998–2014 date/days 5 Jul 20 0.16 2.847 0.11 –1.6

SCR = 10 June snow-cover region; C1, M2, and M3 = climate stations; VF = Valley floor region; HS = Hill slope region (all seen in Fig. 1). SD = Standard deviation, R2 = coefficient of determination,p = P-value, F = F-statistics.

1997 to 2014 (R2 = 0.27, F = 5.803, p = 0.028, climate station (Fig. 1). The maximum snow depths Table 4, Fig. 3). There was no significant difference of the ZERO-line time series were observed in the in the variance between the periods prior to and end of the snow-rich winter 2011/2012—that is, after 2006. However, the 10 June snow-cover time in spring 2012 (red points in Fig. 6, part a)—while series varied up to 24 percentage points in standard the minimum snow depths were observed in spring deviation throughout the period. 2009 along the transect (blue points in Fig. 6, part a). The snow-depth observations along ZERO-line Spatial Distribution of Snow (Fig. 1) showed the highest snow depth in the lowland (0–150 m a.s.l., 0–4 km in distance from the Snow-Depth Variability with Elevation and Vegeta- transect starting point) with varying slopes of smaller tion Type hills and depressions, where snow accumulated (Fig. The mean snow depth in the 250-m distance in- 6, parts a and c). The lower snow depths were found tervals along the ZERO-line ranged between 0.3 m at higher elevation (150–300 m a.s.l. and 4–5 km and 0.9 m for the years 2005–2014 (black points in distance), where steeper slopes (~15%) and outcrops Fig. 6, part a), and the maximum snow-depth point are blown free of snow. At elevation 300–350 m a.s.l., measurement was 3.5 m in 2012 at the lowest el- we found locally higher snow depths potentially due evations (0–50 m a.s.l.) in a snowdrift near the M2 to snowdrifts building up on the lee-side of the edge

Arctic, Antarctic, and Alpine Research / stine Højlund Pedersen et al. / 663 tation. We identified the vegetation types that had significantly different mean snow depth from other vegetation types (p < 0.01) (Fig. 6, part d).

Snowmelt and Snow-cover Depletion The Valley floor and Hill slope, located in elevation levels at 0–50 m and 150–300 m a.s.l., respectively, were characterized by different topographic relief with a mean terrain ruggedness of 8 ± 5 m and 42 ± 8 m, and a mean terrain slope of 3.3% ± 3.7% and 14.3% ± 1.6%, respectively. Snow-cover depletion curves derived from the automatic cameras showed a difference in snowmelt timing and depletion rate between the two regions (Figs. 7 and 8). The timing of 80% and 20% snow cover was derived for each year-specific depletion curve for each region (Fig. 8, Table 3). The duration of the snowmelt period (80%–20% snow cover) lasted 3–13 days in the Valley floor and 11–47 days in the Hill slope. In all years, the Valley floor had the shortest snowmelt period of the two regions (Fig. 8). The snowmelt start (80% snow cover) occurred between 19 May and 30 June in the two regions (Figs. 7 and 8). The timing of snowmelt end (20% snow cover) occurred between 30 May and 6 July in the Valley floor and 30 May and 16 August in the Hill slope—that is, the snowmelt end occurred lat- FIGURE 5. The monthly and annual sums of identified snowfall events (>0.05 m snow-depth er in the Hill slope than in the Valley floor. Hence, increase per day, black points) and their distribution the timing of snowmelt end varied less between across the period September 1997 through August years in the Valley floor than in the Hill slope re- 2014. Large snowfall events (>0.20 m snow-depth gion. We found no significant trends in the dura- increase per day) are marked with red points (large tion of snowmelt period for the two regions from red point = 2 and small red point = 1 large snowfall 1998 through 2014 (Table 4). Episodic snowmelt event). In the margin, annual and monthly sums of snowfall events are given with the number of large events, when rapid snowmelt occurs during foehn snowfall events in parentheses. wind events, are identified in other regions of ice- free Greenland (Pedersen et al., 2015). However, no such events have been identified in Zackenberg using either local meteorological station data or rea- of a relatively flat, wind-blown plateau at approxi- nalysis data (Pedersen et al., 2015). mately 350 m a.s.l., where the slope decreased. There was no significant temporal trend in the annual variance in snow depth observations through the period Modeled SWE 1979–2014 2005–2014 (p > 0.05). The modeled and observed end-of-winter SWE Across the years, snow depth exhibited a consist- during the period 2004–2014 showed strong cor- ent pattern of statistically significant different snow respondence (Fig. 9, R2 = 0.82, F = 37.51, p < 0.001, depths among vegetation types (p < 0.01) (Fig. 6, intercept = –0.02, slope = 1.03), and the interannual part b) and altitude (p < 0.01). The effect of “year” variances for the two data sets were not significantly was also significant p( < 0.01), reflecting the inter- different in either 5-year or 11-year intervals F( = annual variation in the amounts of snow precipi- 0.85–1.80, p > 0.05). Hence, the SnowModel repro-

664 / stine Højlund Pedersen et al. / arctic, Antarctic, and Alpine Research FIGURE 6. (a) Mean snow depth (m) (error bars are error on the mean) aggregated per 250 m along the ZERO-line transect in distance (km) from the starting point for all observations in the years 2005– 2014 (black points), a snow-rich year 2011/2012 (red points), and a snow-poor year 2008/2009 (blue points). Gray shading: all snow- depth observations for all years excluding the 10% lowest and 10% highest values. (b) Mean snow depth (m) (error bars are error on the mean) for the period 2005– 2014 for the six vegetation types observed along ZERO-line. (c) ZERO-line elevation (m above sea level) and slope (%) with distance (m) from the starting point of the transect (Fig. 1). (d) Matrix presenting results from Tukey’s Honest Significant Differences for multiple comparisons of mean snow depths in six vegetation types. Red circles mark significantly different (p < 0.01) vegetation types in the pairwise comparison of mean snow depths. duced the temporal variability in the snow condi- dicating an increasingly patchy snow cover in the tions in Zackenberg and was a valid tool to model beginning of June. In addition, the spring snow snow variables back in time. The variances in mod- cover on 10 June showed large interannual varia- eled end-of-winter SWE in 5-year periods (1980– tion from 3% to 92% in the Valley floor region. The 1984, 1985–1989, 1990–1994, 1995–1999, 2000– amount of snow—that is, the annual maximum 2004, 2005–2009, and 2010–2014) and 11-year snow depth—showed no significant temporal trend periods (1980–1990, 1991–2001, and 2002–2014) but a pronounced interannual variation up to an did not differ significantly (F = 0.10–5.69, p > 0.05). order of magnitude between years in Zackenberg. The modeled end-of-winter SWE showed no linear This interannual variation was particularly related trend through the 35-year time series (R2 < 0.01, p to the variable number of snowfall events per win- > 0.05, slope = 0.001), but the interannual variability ter, ranging between 0 and 16 events through the was as variable as in the recent 11 years (2004–2014). period 1997–2014, and occurring throughout the winter season from September through May. More- over, the highly variable number of snowfall events Discussion and variable timing of snow-cover onset, which In the present study, we found few temporal ranged over a period of approximately 162 days (~5 trends but high interannual variability in a suite months) caused the durations of the snow-covered of snow variables in the High Arctic ecosystem period to range from 74 to 265 days. Such variable in Zackenberg. The spring snow cover decreased duration of the snow-covered period effectively led significantly through the period 1997–2014, in- to a relatively prolonged and shortened snow-free

Arctic, Antarctic, and Alpine Research / stine Højlund Pedersen et al. / 665 FIGURE 7. Snow-cover fraction per day during the snowmelt season (gray circles) for the two regions, Valley floor (top) and Hill slope (bottom) in the years 1998–2014. Snow-cover fractions in the snow-poor spring 2013 (red circles) and the snow-rich spring 1999 (blue circles). Arrows indicate the range in timing of 50% snow cover. period from 1997 through 2014. In Svalbard, the ter and 2008/2009 being a snow-poor winter, onset of snowmelt shows large interannual varia- the relative spatial distribution of snow depth was tion (Rotschky et al., 2011), which can cause simi- similar from year to year (Fig. 6, part a), and we lar fluctuations in the length of snow-free period. found no temporal trend in the variability of the In Zackenberg, where the lemming, a key species in transect snow depths through the observational the Arctic food web, has been monitored since 1996 period. The spatial pattern of relatively deeper (Meltofte and Berg, 2006), the predation pressure snow and shallow snow packs along ZERO-line (by Arctic fox, stoat, snowy owl, and long-tailed persisted from year to year given that the snow skua) is high (Schmidt et al., 2008; Barraquand et (re-)distribution during accumulation and erosion al., 2014). Since the snow cover acts as protection by wind is mainly controlled by the topography and limits the foraging on lemmings for the lo- (Schirmer et al., 2011) and the spatial variability cal predators during winter, the lemming popula- in slope. More specifically, there is a clear corre- tion size is sensitive to the length of the snow-free spondence between areas with slope increase or period and the variation between prolonged and decrease (i.e., a change in elevation; Fig. 6, part shortened snow-free periods. Furthermore, the in- c) and mean snow depth (Fig. 6, part a). Hence creased variability in the snow-cover onset in the in these areas, where wind-transported snow will recent years, may have caused the disappearance accumulate on the lee-side slope of hills, the snow of the lemming population cycles in Zackenberg depth is found to be controlled by topography. (Gilg et al., 2009; Schmidt et al., 2012) and hence The predominant wind direction during winter affected the food web as a whole in this Arctic eco- was from north in Zackenberg (observed at M3 system. and C1) and showed no interannual variation However, despite years with contrasting snow during the observation period. A similar persistent conditions within the 11-year ZERO-line time pattern in snow-depth distributions is found in series, with 2011/2012 being a snow-rich win- annually repeated snow-depth transects in, for ex-

666 / stine Højlund Pedersen et al. / arctic, Antarctic, and Alpine Research ample, Alaska, where it is primarily driven by the topographic relief but also the vegetation height (0–100 cm) and stand structure (Sturm and Wag- ner, 2010). However, for Zackenberg, located in the High Arctic climate zone where the vegetation cover is sparse and the vegetation height does not exceed 20 cm (Bay, 1998), it is most likely that the persistent pattern of the snow-depth distribution along the ZERO-line mainly controls the vegetation growth conditions and thus the spatial distribution of the vegetation types and not vice versa. This linkage was supported by the fact that the observed snow depths in the six vegetation types along the ZERO-line were significantly different and that vegetation types characteristic of wet and dry (i.e., snow-rich and snow-poor) environments could be distinguished from the snow depth data. The deepest snowpack was found in snowbeds dominated by Salix arctica (Fig. 6, part b), and the mean snow depth herein was consistently different from the mean snow depth in all other vegetation types across the 10 years (Fig. 6, part d). This may be expected since snowbed vegetation develop in moist environments, preferably down-slope from perennial snowdrifts (Björk and Molau, 2007). However, the mean snow depths in Cassiope heath, Fen, and Grassland did not differ significantly from each other, whereas bothDryas FIGURE 8. Timing and duration of the snowmelt period between 80% (left end of lines) and 20% (right heath and Fell-field vegetation, which occur in end of lines) snow cover for two regions, Valley floor more snow-poor areas (Fig. 6, part b), had a signif- (black) and Hill slope (gray) in Zackenberg in 1998– icantly different snow depth than observed for the 2014. No data was available for Valley floor in 1998. remaining four vegetation types (Fig. 6, part d).

FIGURE 9. Modeled end-of-winter snow-water equivalent (SWE, m water equivalent [m w.e.]) from SnowModel in the period 1979 through 2014 compared with observed end-of-winter SWE (m w.e.) from snow pits dug in the period 26 March–6 June in 2004–2014 in Valley floor region, Zackenberg. Subplot: linear regression between observed and modeled end-of-winter SWE in 2004– 2014 (no available SWE observation in 2009). Linear fit statistics:R 2 = 0.82, F = 37.51, p < 0.001, intercept = –0.02, slope = 1.03, n = 10.

Arctic, Antarctic, and Alpine Research / stine Højlund Pedersen et al. / 667 Despite the larger snow depth in the lowland depth, spring snow cover, and indirectly timing of than at higher altitude on the slope, seen on the snowmelt). Hence, the magnitude of interannual ZERO-line (Fig. 6, part a), the duration of the variation in these snow variables has likely remained snowmelt period was more than 1.5 times long- unchanged at least since 1979. Additionally, the in- er on the Hill slope than in the Valley floor. The terannual variation seen the last decade, including snowmelt started on average at the same time in the particularly dry winter 2012/2013, did not ap- the two regions (Figs. 7 and 8), which is controlled pear unusual in the 35-year time series. The finding by the time when ambient temperature rises above of no change in temporal variability through the freezing (Liston, 1995; Liston and Hall, 1995; Clow, 35-year period was supported by the fact that we 2009). However, the Hill slope snowmelt almost found no significant difference between snow vari- consistently ended later than in the Valley floor ables in recent years (2006/2007–2013/2014) and (Figs. 7 and 8). This extended snowmelt season on the early years (1997/1998–2005/2006), except for the Hill slope may be explained by the snowdrifts the timing of snow-cover onset, which in the re- that built up in the more rough terrain on the slope, cent period was 7.3 times more variable. This in- which eventually created a larger spatial variability creased temporal variability may point to a change in snow depth, resulting in a less steep snow-cover in the autumn weather conditions, when the snow depletion curve than for the Valley floor (Clark et cover is established. al., 2011). These snowdrifts take a longer time to melt away, hence postponing the snowmelt end Conclusions until July/August. In Zackenberg snowdrifts persisted throughout the summer in some snow-rich This study presented and discussed interannual years—for example, 1999 and 2014 (unpublished variability and trends in a suite of snow variables observations from automatic digital camera photos in the High Arctic environment of Zackenberg in in summer 1999 and 2014). This may explain the Northeast Greenland between 1979 and 2014. In higher interannual standard deviation in the length the observation period (1997–2014), apart from of the snowmelt period at the Hill slope than at 10 June snow-cover fraction, no presented snow the Valley floor (Table 4), since it is dependent on variables showed significant temporal trends across the presence and absence of snowdrifts and their the 18-year time series. The observed pronounced extent and thickness from year to year; that is, the interannual variability mainly masked any poten- timing of snowmelt (20% snow cover) is primar- tial trends. The distribution of end-of-winter snow ily dependent on the maximum snow depth (R2 = depth observed in the ZERO-line transect was 0.76, F = 42.62, p < 0.001). Furthermore, we iden- significantly affected by elevation and showed a tified that the 80% snow cover occurred on aver- significant difference in snow depth between the age on 10 June in both regions. This may highlight present vegetation types. SnowModel reproduced that currently 10 June is a relevant indicator for the temporal variability in the snow conditions in spring snow cover since it marks the beginning of Zackenberg. Extending the time series of end-of- the snowmelt season. However, in a warming Arc- winter SWE to 1979 using SnowModel showed tic, when snow is predicted to melt earlier in the that the pronounced interannual variability in snow spring than currently observed, we could expect a amounts observed in the recent decade, also was change in this date (e.g., Barnett et al., 2005). present over the last several decades back to 1979. Modeled end-of-winter SWE lacked significant We have left the detailed explanation of the in- temporal trends, and the pronounced interannu- teractions between the abiotic and biotic compo- al variation in end-of-winter SWE in the recent nents of this ecosystem to future studies. Likewise, years (2004–2014) did not differ from the interan- we acknowledge that the temporal variability in the nual variation in end-of-winter SWE in the peri- time series of snow observations presented herein ods 1980–1990 and 1991–2001. SWE was directly possibly originate from changes in large-scale pat- linked to winter precipitation amounts, and there- terns and trends in, for example, Arctic synoptic- fore a main driver for the interannual variation in scale circulation patterns, the Greenland Ice Sheet– the snow variables presented in this study (snow related weather patterns, and/or the near-shore

668 / stine Højlund Pedersen et al. / arctic, Antarctic, and Alpine Research sea-ice extent and concentration in the Greenland Björk, R. G., and Molau, U., 2007: Ecology of alpine snowbeds Sea. However, it is beyond the scope of this paper to and the impact of global change. Arctic, Antarctic, and Alpine explain the effect of these large-scale patterns and Research, 39(1): 34–43. Blankinship, J. C., Meadows, M. W., Lucas, R. G., and Hart, trends on our results. Hence, further research into S. C., 2014: Snowmelt timing alters shallow but not deep large-scale circulation patterns’ effect on local-scale soil moisture in the Sierra Nevada. Water Resources Research, snowfall events and timing of snowmelt is required 50(2): 1448–1456. in order to explain the interannual variation ob- Bliss, L., and Matveyeva, N., 1992: Circumpolar Arctic served in Zackenberg. vegetation. In Chapin, F. S., III, Jefferies, R. L., Reynolds, J. F., Shaver, G. R., Svoboda, J., and Chu, E. W. (eds.), Arctic Ecosystems in a Changing Climate: An Ecophysiological Acknowledgments Perspective. San Diego, California: Academic Press, 59–89. Blunden, J., and Arndt, D. S., 2014: State of the climate in We owe enormous gratitude to all GeoBasis field 2013. Bulletin of the American Meteorological Society, 95(7): assistants, who have collected the snow observa- S1–S279. Bokhorst, S., Bjerke, J. W., Street, L. E., Callaghan, T. V., and tions in the Zackenberg Valley during the period Phoenix, G. K., 2011: Impacts of multiple extreme winter 1997–2014. We wish to thank the logistics team at warming events on sub-Arctic heathland: phenology,

Zackenberg Research Station, Aarhus University, reproduction, growth, and CO2 flux responses. Global for their support and flexibility in the “shoulder- Change Biology, 17(9): 2817–2830. seasons.” Thank you to two anonymous reviewers Bokhorst, S., Pedersen, S. H., Brucker, L., Anisimov, O., Bjerke, and R. De Troch for valuable comments. Data from J. W., Brown, R. D., Ehrich, D., Essery, R. L. H., Heilig, A., Ingvander, S., Johansson, C., Johansson, M., Jónsdóttir, I. S., the Greenland Ecosystem Monitoring Program Inga, N., Luojus, K., Macelloni, G., Mariash, H., McLennan, ClimateBasis were provided by Asiaq–Greenland D., Rosqvist, G. N., Sato, A., Savela, H., Schneebeli, M., Survey, Nuuk, Greenland. Data from the Green- Sokolov, A., Sokratov, S. A., Terzago, S., Vikhamar-Schuler, D., land Ecosystem Monitoring Program GeoBasis Williamson, S., Qiu, Y., and Callaghan, T. V., 2016: Changing were provided by the Department of Bioscience, Arctic snow cover: a review of recent developments and Aarhus University, Denmark, in collaboration with assessment of future needs for observations, modelling, and impacts. Ambio, 45(5): 516–537. Department of Geosciences and Natural Resource Brooks, P. D., Grogan, P., Templer, P. H., Groffman, P., Öquist, Management, Copenhagen University, Denmark. M. G., and Schimel, J., 2011: Carbon and nitrogen cycling Data collection was financed through the DAN- in snow‐covered environments. Geography Compass, 5(9): CEA funds from the Danish Ministry of Environ- 682–699. ment and Danish Ministry of Climate and Energy. Buckeridge, K. M., and Grogan, P., 2008: Deepened snow alters soil microbial nutrient limitations in arctic birch hummock tundra. Applied Soil Ecology, 39(2): 210–222. References Cited Bulygina, O. N., Groisman, P. Y., Razuvaev, V. N., and Korshunova, N. N., 2011: Changes in snow cover Barnett, T. P., Adam, J. C., and Lettenmaier, D. P., 2005: characteristics over Northern Eurasia since 1966. Potential impacts of a warming climate on water availability Environmental Research Letters, 6(4): doi http://dx.doi. in snow-dominated regions. Nature, 438(7066): 303–309. org/10.1088/1748-9326/6/4/045204. Barraquand, F., Høye, T. T., Henden, J.-A., Yoccoz, N. G., Buus-Hinkler, J., Hansen, B. U., Tamstorf, M. P., and Pedersen, Gilg, O., Schmidt, N. M., Sittler, B., and Ims, R. A., 2014: S. B., 2006: Snow-vegetation relations in a High Arctic Demographic responses of a site-faithful and territorial ecosystem: inter-annual variability inferred from new predator to its fluctuating prey: long-tailed skuas and Arctic monitoring and modeling concepts. Remote Sensing of lemmings. Journal of Animal Ecology, 83(2): 375–387. Environment, 105(3): 237–247. Bay, C., 1998: Vegetation mapping of Zackenberg Valley, Callaghan, T. V., Johansson, M., Brown, R. D., Groisman, Northeast Greenland. Danish Polar Center & Botanical P. Y., Labba, N., Radionov, V., Bradley, R. S., Blangy, S., Museum, University of Copenhagen, Denmark. Bulygina, O. N., Christensen, T. R., Colman, J. E., Essery, R. Bennike, O., Sørensen, M., Fredskild, B., Jacobsen, B. L. H., Forbes, B. C., Forchhammer, M. C., Golubev, V. N., H., Böcher, J., Amsinck, S. L., Jeppesen, E., Andreasen, Honrath, R. E., Juday, G. P., Meshcherskaya, A. V., Phoenix, C., Christiansen, H. H., and Humlum, O., 2008: Late G. K., Pomeroy, J., Rautio, A., Robinson, D. A., Schmidt, N. Quaternary environmental and cultural changes in the M., Serreze, M. C., Shevchenko, V. P., Shiklomanov, A. I., Wollaston Forland Region, Northeast Greenland. In High- Shmakin, A. B., Skold, P., Sturm, M., Woo, M. K., and Wood, Arctic Ecosystem Dynamics in a Changing Climate. Advances in E. F., 2011: Multiple effects of changes in Arctic snow cover. Ecological Research, 40: 45–79. Ambio, 40(Supplement 1): 32–45.

Arctic, Antarctic, and Alpine Research / stine Højlund Pedersen et al. / 669 Chen, X., Liang, S., Cao, Y., He, T., and Wang, D., 2015: Observed Common Plant Communities in the Zackenberg Valley, Northeast contrast changes in snow cover phenology in northern Greenland, Greenland. Copenhagen, Denmark: Botanical middle and high latitudes from 2001–2014. Scientific Reports, Survey & Botanical Museum,University of Copenhagen. 5: doi http://dx.doi.org/10.1038/srep16820. Gilg, O., Sittler, B., and Hanski, I., 2009: Climate change and Clark, M. P., Hendrikx, J., Slater, A. G., Kavetski, D., Anderson, cyclic predator-prey population dynamics in the High B., Cullen, N. J., Kerr, T., Örn Hreinsson, E., and Woods, Arctic. Global Change Biology, 15(11): 2634–2652. R. A., 2011: Representing spatial variability of snow water Goodrich, L., 1982: The influence of snow cover on the equivalent in hydrologic and land-surface models: a review. ground thermal regime. Canadian Geotechnical Journal, Water Resources Research, 47(7). 19(4): 421–432. Clow, D. W., 2009: Changes in the timing of snowmelt and Hollesen, J., Buchwal, A., Rachlewicz, G., Hansen, B. U., streamflow in Colorado: a response to recent warming. Hansen, M. O., Stecher, O., and Elberling, B., 2015: Winter Journal of Climate, 23(9): 2293–2306. warming as an important co-driver for Betula nana growth Clyde, G. D., 1932: Circular no. 99—Utah snow sampler and in western Greenland during the past century. Global scales for measuring water content of snow. UAES Circulars, Change Biology, 21(6): 2410–2423. Paper 90: http://digitalcommons.usu.edu/uaes_circulars/90. Johansson, M., Callaghan, T. V., Bosiö, J., Åkerman, H. Cooper, E. J., 2014: Warmer shorter winters disrupt Arctic J., Jackowicz-Korczynski, M., and Christensen, T. R., terrestrial ecosystems. Annual Review of Ecology, Evolution, 2013: Rapid responses of permafrost and vegetation to and Systematics, 45: 271–295. experimentally increased snow cover in sub-Arctic Sweden. Dee, D. P., Uppala, S. M., Simmons, A. J., Berrisford, P., Poli, Environmental Research Letters, 8(3): doi http://dx.doi. P., Kobayashi, S., Andrae, U., Balmaseda, M. A., Balsamo, G., org/10.1088/1748-9326/8/3/035025. Bauer, P., Bechtold, P., Beljaars, A. C. M., van de Berg, L., Jones, H. G., 1999: The ecology of snow-covered systems: Bidlot, J., Bormann, N., Delsol, C., Dragani, R., Fuentes, M., a brief overview of nutrient cycling and life in the cold. Geer, A. J., Haimberger, L., Healy, S. B., Hersbach, H., Hólm, Hydrological Processes, 13(14–15): 2135–2147. E. V., Isaksen, L., Kållberg, P., Köhler, M., Matricardi, M., Kottek, M., Grieser, J., Beck, C., Rudolf, B., and Rubel, F., 2006: McNally, A. P., Monge-Sanz, B. M., Morcrette, J. J., Park, B. World map of the Köppen-Geiger climate classification K., Peubey, C., de Rosnay, P., Tavolato, C., Thépaut, J. N., and updated. Meteorologische Zeitschrift, 15(3): 259–263. Vitart, F., 2011: The ERA-Interim reanalysis: configuration Kristensen, D. K., Kristensen, E., Forchhammer, M. C., and performance of the data assimilation system. Quarterly Michelsen, A., and Schmidt, N. M., 2011: Arctic herbivore Journal of the Royal Meteorological Society, 137(656): 553–597. diet can be inferred from stable carbon and nitrogen Derksen, C., and Brown, R., 2012: Spring snow cover extent isotopes in C3 plants, faeces, and wool. Canadian Journal of reductions in the 2008–2012 period exceeding climate Zoology, 89(10): 892–899. model projections. Geophysical Research Letters, 39(19). Liston, G. E., 1995: Local advection of momentum, heat, and Derksen, C., Lemmetyinen, J., Toose, P., Silis, A., Pulliainen, moisture during the melt of patchy snow covers. Journal of J., and Sturm, M., 2014: Physical properties of Arctic Applied Meteorology, 34(7): 1705–1715. versus subarctic snow: implications for high latitude passive Liston, G. E., and Elder, K., 2006a: A meteorological microwave snow water equivalent retrievals. Journal of distribution system for high-resolution terrestrial modeling Geophysical Research: Atmospheres, 119(12): 7254–7270. (MicroMet). Journal of Hydrometeorology, 7(2): 217–234. Dyrrdal, A. V., Saloranta, T., Skaugen, T., and Stranden, H. B., Liston, G. E., and Elder, K., 2006b: A distributed snow- 2013: Changes in snow depth in Norway during the period evolution modeling system (SnowModel). Journal of 1961–2010. Hydrology Research, 44(1): 169–179. Hydrometeorology, 7(6): 1259–1276.

Elberling, B., 2007: Annual soil CO2 effluxes in the High Liston, G. E., and Hall, D. K., 1995: An energy-balance model Arctic: the role of snow thickness and vegetation type. Soil of lake-ice evolution. Journal of Glaciology, 41(138): 373–382. Biology and Biochemistry, 39(2): 646–654. Liston, G. E., and Hiemstra, C. A., 2008: A simple data Elberling, B., Tamstorf, M. P., Michelsen, A., Arndal, M. F., assimilation system for complex snow distributions Sigsgaard, C., Illeris, L., Bay, C., Hansen, B. U., Christensen, (snowassim). Journal of Hydrometeorology, 9(5): 989–1004. T. R., and Hansen, E. S., 2008: Soil and plant community- Liston, G. E., and Hiemstra, C. A., 2011: The changing characteristics and dynamics at Zackenberg. In Meltofte, cryosphere: Pan-Arctic snow trends (1979–2009). Journal of H., Christensen, T. R., Elberling, B., Forchhammer, M. C., Climate, 24(21): 5691–5712. and Rasch, M. (eds.), High-Arctic Ecosystem Dynamics in a Liston, G. E., and Sturm, M., 1998: A snow-transport model Changing Climate. Amsterdam: Elsevier, Advances in Ecological for complex terrain. Journal of Glaciology, 44(148): 498–516. Research, 40: 223–248. Liston, G. E., McFadden, J. P., Sturm, M., and Pielke, R. A., Forchhammer, M. C., Post, E., Berg, T. B., Høye, T. T., and 2002: Modelled changes in Arctic tundra snow, energy and Schmidt, N. M., 2005: Local-scale and short-term moisture fluxes due to increased shrubs. Global Change herbivore-plant spatial dynamics reflect influences of large- Biology, 8(1): 17–32. scale climate. Ecology, 86(10): 2644–2651. Meltofte, H., and Berg, T. B., 2006: BioBasis—Conceptual Fredskild, B., and Mogensen, G. S., 1997: Zero Line: Final Design and Sampling Procedures of the Biological Programme Report 1997: a Description of the Plant Communities Along the of Zackenberg Basic. Roskilde, Denmark: National ZERO Line from Young Sund to the Top of Aucellabjerg and the Environmental Research Institute.

670 / stine Højlund Pedersen et al. / arctic, Antarctic, and Alpine Research Meltofte, H., Christensen, T. R., Elberling, B., Forchhammer, M. C., and Morten, R. (eds.), Advances in Ecological Research. M. C., and Rasch, M., 2008a: High-Arctic Ecosystem Dynamics London and New York: Academic Press, 345–370. in a Changing Climate. London: Elsevier Academic Press. Schmidt, N. M., Ims, R. A., Høye, T. T., Gilg, O., Hansen, Meltofte, H., Høye, T. T., and Schmidt, N. M., 2008b: Effects L. H., Hansen, J., Lund, M., Fuglei, E., Forchhammer, M. of food availability, snow and predation on breeding C., and Sittler, B., 2012: Response of an Arctic predator performance of waders at Zackenberg. Advances in Ecological guild to collapsing lemming cycles. Proceedings of the Royal Research, 40: 325–343. Society of London B: Biological Sciences, 279(1746): 4417– Mernild, S. H., Liston, G. E., and Hasholt, B., 2007: Snow- 4422. distribution and melt modelling for glaciers in Zackenberg Schmidt, N. M., Pedersen, S. H., Mosbacher, J. B., and Hansen, river drainage basin, north-eastern Greenland. Hydrological L. H., 2015: Long-term patterns of muskox (Ovibos Processes, 21(24): 3249–3263. moschatus) demographics in High Arctic Greenland. Polar Mosbacher, J. B., Kristensen, D. K., Michelsen, A., Stelvig, Biology, 38: 1667–1675. M., and Schmidt, N. M., 2016: Quantifying muskox plant Sigsgaard, C., Mylius, M. R., and Skov, K., 2014: Zackenberg biomass removal and spatial relocation of nitrogen in a Ecological Research Operations. GeoBasis—Guidelines and High Arctic Tundra ecosystem. Arctic, Antarctic, and Alpine Sampling Procedures for the Geographical Monitoring Programme of Research, 48(2): 229–240. Zackenberg Basic. Department of Bioscience, Aarhus University Pedersen, S. H., Liston, G. E., Tamstorf, M. P., Westergaard-Nielsen, and Department of Geosciences and Natural Resource A., and Schmidt, N. M., 2015: Quantifying episodic snowmelt Management, University of Copenhagen, Denmark. events in Arctic ecosystems. Ecosystems, 18(5): 839–856. Stuefer, S., Kane, D. L., and Liston, G. E., 2013: In situ snow Pellissier, L., Meltofte, H., Hansen, J., Schmidt, N. M., water equivalent observations in the US Arctic. Hydrology Tamstorf, M. P., Maiorano, L., Aastrup, P., Olsen, J., Guisan, Research, 44(1): 21–34. A., and Wisz, M. S., 2013: Suitability, success and sinks: Sturm, M., and Wagner, A. M., 2010: Using repeated How do predictions of nesting distributions relate to fitness patterns in snow distribution modeling: an Arctic parameters in High Arctic waders? Diversity and Distributions, example. Water Resources Research, 46: doi http://dx.doi. 19(12): 1496–1505. org/10.1029/2010WR009434. Post, E., Forchhammer, M. C., Bret-Harte, M. S., Callaghan, Sturm, M., Holmgren, J., and Liston, G. E., 1995: A seasonal T. V., Christensen, T. R., Elberling, B., Fox, A. D., Gilg, O., snow cover classification-system for local to global Hik, D. S., and Høye, T. T., 2009: Ecological dynamics across applications. Journal of Climate, 8(5): 1261–1283. the Arctic associated with recent climate change. Science, Sturm, M., Holmgren, J., Konig, M., and Morris, K., 1997: The 325(5946): 1355–1358. thermal conductivity of seasonal snow. Journal of Glaciology, Proksch, M., Rutter, N., Fierz, C., and Schneebeli, M., 2016: 43(143): 26–41. Intercomparison of snow density measurements: bias, precision, Sturm, M., McFadden, J. P., Liston, G. E., Chapin, F. S., Racine, and vertical resolution. The Cryosphere, 10(1): 371–384. C. H., and Holmgren, J., 2001: Snow-shrub interactions Riley, S. J., DeGloria, S. D., and Elliot, R., 1999: A terrain in Arctic tundra: a hypothesis with climatic implications. ruggedness index that quantifies topographic heterogeneity. Journal of Climate, 14(3): 336–344. Intermountain Journal of Sciences, 5(1–4): 23–27. Tedesco, M., Brodzik, M., Armstrong, R., Savoie, M., and Rotschky, G., Schuler, T. V., Haarpaintner, J., Kohler, J., and Ramage, J., 2009: Pan Arctic terrestrial snowmelt trends Isaksson, E., 2011: Spatio-temporal variability of snowmelt (1979–2008) from spaceborne passive microwave data across Svalbard during the period 2000–08 derived from and correlation with the Arctic Oscillation. Geophysical QuikSCAT/SeaWinds scatterometry. Polar Research, 30: doi Research Letters, 36(21): L21402, doi http://dx.doi. http://dx.doi.org/10.3402/polar.v30i0.5963. org/10.1029/2009GL039672. Schimel, J. P., Bilbrough, C., and Welker, J. M., 2004: Increased Westermann, S., Elberling, B., Højlund Pedersen, S., Stendel, snow depth affects microbial activity and nitrogen M., Hansen, B., and Liston, G., 2015: Future permafrost mineralization in two Arctic tundra communities. Soil conditions along environmental gradients in Zackenberg, Biology and Biochemistry, 36(2): 217–227. Greenland. The Cryosphere, 9(2): 719–735. Schirmer, M.,Wirz, V., Clifton,A., and Lehning, M., 2011: Persistence Winstral, A., Elder, K., and Davis, R. E., 2002: Spatial snow in intra-annual snow depth distribution: 1. Measurements and modeling of wind-redistributed snow using terrain-based topographic control. Water Resources Research, 47(9): doi http:// parameters. Journal of Hydrometeorology, 3(5): 524–538. dx.doi.org/10.1029/2010WR009426. Zhang, T., 2005: Influence of the seasonal snow cover on the Schmidt, N. M., Baittinger, C., and Forchhammer, M. C., ground thermal regime: an overview. Reviews of Geophysics, 2006: Reconstructing century-long snow regimes using 43(4): doi http://dx.doi.org/10.1029/2004RG000157. estimates of High Arctic Salix arctica radial growth. Arctic, Zhang, X., Vincent, L. A., Hogg, W. D., and Niitsoo, A., 2000: Antarctic, and Alpine Research, 38(2): 257–262. Temperature and precipitation trends in Canada during the Schmidt, N. M., Berg, T. B., Forchhammer, M. C., Hendrichsen, 20th century. Atmosphere-Ocean, 38(3): 395–429. D. K., Kyhn, L. A., Meltofte, H., and Høye, T. T., 2008:Vertebrate predator—prey interactions in a seasonal environment. In MS submitted 29 March 2016 Meltofte, H., Christensen, T. R., Elberling, B., Forchhammer, MS accepted 23 August 2016

Arctic, Antarctic, and Alpine Research / stine Højlund Pedersen et al. / 671

Paper III

Quantifying episodic snowmelt events in Arctic ecosystems.

Pedersen, S. H., Liston, G. E., Tamstorf, M. P., Westergaard-Nielsen, A., and Schmidt, N. M. (2015). Ecosystems. Vol. 18, No. 5, pp. 839-856. DOI: 10.1007/s10021-015-9867-8.

Kobbefjord, W-Greenland (2013)

Ecosystems (2015) 18: 839–856 DOI: 10.1007/s10021-015-9867-8 Ó 2015 Springer Science+Business Media New York

Quantifying Episodic Snowmelt Events in Arctic Ecosystems

Stine Højlund Pedersen,1* Glen E. Liston,2 Mikkel P. Tamstorf,1 Andreas Westergaard-Nielsen,3 and Niels Martin Schmidt1

1Arctic Research Centre, Department of Bioscience, Aarhus University, Frederiksborgvej 399, 4000 Roskilde, Denmark; 2Cooperative Institute for Research in the Atmosphere (CIRA), Colorado State University, Fort Collins, Colorado 80523, USA; 3Center for Permafrost (CENPERM), Department of Geosciences and Natural Resource Management, University of Copenhagen, Oester Voldgade 10, 1350 Copenhagen K, Denmark

ABSTRACT Rapid and extensive snowmelt occurred during ESE, we investigated the origin, past occurrences, 2 days in March 2013 at a low-Arctic study site in frequency, and abundance of ESEs at spatial scales the ice-free part of southwest Greenland. Me- ranging from local (using 2008–2013 meteoro- teorology, snowmelt, and snow-property observa- logical station data) to all of Greenland (using tions were used to identify the meteorological 1979–2013 atmospheric reanalysis data). The fre- conditions associated with this episodic snowmelt quency of ESEs showed large interannual varia- event (ESE) occurring prior to the spring snowmelt tion, and a maximum number of ESEs was found in season. In addition, outputs from the SnowModel southwest Greenland. The investigations suggested snowpack-evolution tool were used to quantify the that ESEs are driven by foehn winds that are typical snow-related consequences of ESEs on ecosystem- of coastal regions near the Greenland Ice Sheet relevant snow properties. We estimated a 50–80% margin. Therefore, ESEs are a common part of meltwater loss of the pre-melt snowpack water snow-cover dynamics in Greenland and, because of content, a 40–100% loss of snow thermal resis- their substantial impact on ecosystem processes, tance, and a 4-day earlier spring snowmelt snow- they should be accounted for in snow-related free date due to this March 2013 ESE. Further- ecosystem and climate-change studies. more, the accumulated meltwater loss from all ESEs in a hydrological year represented 25–52% of the annual precipitation and may potentially have Key words: snow; meltwater; modeling; growing advanced spring snowmelt by 6–12 days. Guided season; snow thermal properties; foehn; upscaling; by the knowledge gained from the March 2013 Greenland.

INTRODUCTION Terrestrial snow cover is a key variable controlling Arctic ecosystem processes (Jones 1999; Post and

Received 4 July 2014; accepted 15 February 2015; others 2009; Brooks and others 2011; Callaghan published online 26 March 2015 and others 2011). Snow properties, such as depth, Author contributions Stine Højlund Pedersen: Designed the study, density, snow-water-equivalent (SWE), thermal performed research, analyzed data, contributed with new methods, and wrote the paper. Glen E. Liston: contributed with new methods and conductivity, and timing of snowmelt, affect biotic models and wrote the paper. Mikkel P. Tamstorf: contributed with new and abiotic components of the Arctic ecosystem. methods and wrote the paper. Andreas Westergaard-Nielsen: wrote the The presence and the absence of snow cover in- paper. Niels Martin Schmidt: wrote the paper. *Corresponding author; e-mail: [email protected] ﬂuence the surface energy balance both locally

839 840 S. H. Pedersen and others

(Marks and Dozier 1992) and globally (Groisman lowing in common: (1) an abrupt and sporadic and others 1994), which in turn affects below- nature, (2) they are unusual for the season and ground surface thermal regime that controls a winter climate in the geographic locations where range of ecosystem processes (Schimel and others they occur, and (3) they cause changes in snow- 2004; Johansson and others 2013). The snow cov- pack properties that affect ecosystems. Their tem- er, with its low thermal conductivity (Goodrich poral extent varies from a few hours to many days, 1982; Sturm and others 1997) and high thermal and their spatial extent is controlled by the weather resistance (Liston and others 2002), acts as an ef- phenomenon that drives them. The area of influ- fective insulator. This keeps soil thermal conditions ence can range from large regions of the Siberian relatively stable during snow-covered periods tundra exposed to ROSs (Bartsch and others 2010), (Zhang 2005) and protects vegetation from frost to leeside areas of the Canadian Rockies experi- damages (Bokhorst and others 2011). Throughout encing warm and dry chinook winds (Nkemdirim the Arctic, solid precipitation accumulates during 1997; Fuller and others 2009). autumn, winter, and spring in a snowpack, which In recent decades, the ability of extreme weather acts as a water reservoir (Jones 1999). In spring, the events to change snowpack properties and impact water is released during snowmelt and provides local Arctic ecosystems has gained increased at- moisture for plant growth not only at the growing tention. This interest is largely due to the recogni- season initiation, but also into the summer period tion that these snow-related impacts can be (Blankinship and others 2014). Since the presence significant, influencing not only single ecosystem and the absence of snow cover play a major role in components, but also entire communities (Hansen ecosystem functions and dynamics, snow-cover and others 2013). An example of this is the creation timing and duration must be accounted for in of ice layers on the snow surface, within the Arctic process and climate-change ecosystem stud- snowpack, and at the ground surface, as result of ies (Høye and others 2007; Callaghan and others refreezing meltwater or rainwater. These ice layers 2011; Bokhorst and others 2012). can severely limit the foraging area for Arctic un- Snow properties, and the spatial and temporal gulates (Forchhammer and Boertmann 1993; development in these, are directly influenced by Bartsch and others 2010; Hansen and others 2011, the weather and climate processes that occur dur- 2013, 2014). Another consequence of extreme ing the snow season. Often these snow properties weather events can be snow-cover depletion and can change dramatically and abruptly in response thereby loss of insulating effect for the below-snow to extreme weather events. These abrupt changes vegetation. This increases the risk of frost damage in snow properties can, in turn, have important to the vegetation when it is exposed to subsequent impacts on ecosystem components and processes. freezing temperatures (Inouye 2000; Bokhorst and Extreme weather events are occurring throughout others 2010; Semenchuk and others 2013). the Arctic and are predicted to become more fre- An extreme warming event and the associated quent in a warming climate (AMAP 2012). Cur- abrupt changes in terrestrial snow-cover properties rently, these events are infrequent and record- were observed in Kobbefjord in southwest Green- breaking, but with increased frequency, they may land on March 15 and 16, 2013. Air temperature become a common part of the future Arctic climate increased from -7.9°C to above freezing within system. 24 h and caused extensive snowmelt across the Ecologically relevant extreme events include landscape (Figure 1). Observations of this relatively rain-on-snow (ROS) events (Rennert and others short-term melt event (occurring prior to the onset 2009), early snowmelt events (Semmens and oth- of spring snowmelt) formed the basis for this study ers 2013), extreme winter warming events (Bo- and it is, herein, referred to as an episodic snow- khorst and others 2010; Semenchuk and others melt event (ESE). The focus of our study is to in- 2013), icing events (Hansen and others 2013), re- vestigate such ESEs and their potential importance freeze events (Bartsch and others 2010), and win- in Arctic ecosystem processes and components. The ter thaw–freeze events (Wilson and others 2013). study consists of three interconnected parts (I, II, These events are caused by different factors such as and III). In Part I, on local-scale in the Kobbefjord winter warming or warm spells, high wind speeds study area (71 km2), meteorological station data from storms or katabatic winds (for example, Fuller and field observations of snow properties were used and others 2009), or heavy winter rainfall (for to characterize the March 2013 ESE. From those example, Rennert and others 2009; Hansen and results, we developed an automated algorithm that others 2014). They may be caused by different processed meteorological data to identify past ESEs weather phenomena, but they all have the fol- and their temporal distributions on a local scale for Episodic Snowmelt Events in Arctic Ecosystems 841

Figure 1. Digital photos from March 14, 2013 (left) and March 16, 2013 (right) showing the center part of the Kobbefjord study area. The photos were taken by an automated camera installed at 550 meters above sea level (m a.s.l.). Photo: GeoBasis, Nuuk Ecological Research Operations. the period 2008–2013. This led to Part II, where we surrounded by steep mountains up to 1375 m a.s.l. quantified how biologically relevant snow proper- The valley floor topography varies on the order of ties, such as snowpack water content, the insulat- 100 m with hill tops and depressions with fresh- ing effect of the snowpack (snow thermal water lakes. The mountain slopes above 200 m resistance), and the timing of spring snowmelt a.s.l. are characterized by alluvial cones, recent (snow-free date), had changed during the local rock slides, small hanging glaciers, and snow ESEs identified in Part I. This was accomplished by patches. The hill slopes below 200 m a.s.l. are running and validating the SnowModel (Liston and vegetated with heath and dwarf shrubs and Elder 2006a, b) spatially distributed snow-evolu- dominated by patterns of inactive solifluction tion modeling system over the Kobbefjord study sheets and 3–4-m-deep depressions eroded by area. Finally, the local-scale results from Part I and streams, which favor growth of 0.7–1.0 m tall Salix II were used, elaborated on, and scaled-up to all of glauca shrubs. Greenland in Part III. Herein, we applied the local- The observed mean annual air temperature scale results from Kobbefjord (Part I and II) to during the study period of 2008–2013 was 0.19°C, identify the ESE spatial distributions and trends on with the lowest temperatures measured in Febru- a regional-scale, using a 34-year time series of ary and the highest in July. The dominant wind historical climate data covering all ice-free parts of direction during May–August was west–northwest Greenland (410,500 km2). Lastly, we discussed the (from the fjord) and during September–April east– driving mechanisms for the ESEs. northeast (from inland). Easterly winds originated from the inner parts of the fjord system and were channeled by the main valley (Jensen and Rasch PART I: CHARACTERIZING EPISODIC 2008, 2009, 2010, 2011, 2013; Jensen 2012). SNOWMELT EVENTS (ESES) The first-hand observations of an ESE in March 2013 initiated the investigation of the meteoro- Method and Data logical characteristics and snowpack changes asso- Local Meteorological Data ciated with this ESE, aiming to develop an To support the analyses, meteorological variables algorithm to identify ESEs in Kobbefjord. including air temperature, wind speed, and snow Study Area depth were provided by the main meteorological station, KOB, located in the study area (Table 1; The low-Arctic, 7.6 km by 9.3 km, Kobbefjord Figure 2). The time series covered the period study area is located at the head of the fjord, September 2008–August 2013. All meteorological Kangerluarsunnguaq, in the Godtha˚bsfjord system data were provided by Nuuk Ecological Research in southwest Greenland (64°7¢59¢¢N, 51°20¢35¢¢W) Operations (NERO); technical information, for ex- (Figure 2). The Kobbefjord study area includes a ample, sensor types and measurement frequency 3-km-wide main valley with two adjacent elevated are available in Jensen and Rasch (2008, 2009, valleys at 200–250 meters above sea level (m a.s.l.) 2010, 2011, 2013), and Jensen (2012). 842 S. H. Pedersen and others

Figure 2. Kobbefjord study area in west Greenland. Brown color scale indicates elevation (m a.s.l.). Red triangles mark the location of four climate stations (SoilFen, KOB, M1000, and M500). The blue triangle is a soil temperature station. Black boxes are annually repeated snow- observation sites with cross transects of snow- depth measurements and snow pits; the west-most site was only repeated on and March 14 and 19, 2013.

Table 1. Kobbefjord Meteorological Stations and Available Variables (Red Triangles in Figure 2)

Station Established Latitude Longitude Elevation (m. a.s.l.) Climate variables

KOB 2007 64°7¢59¢¢ 51°20¢35¢¢ 30 tair, rh, wspd, wdir, Qsi, Qli, snod SoilFen 2007 64°7¢50¢¢ 51°23¢60¢¢ 30 tair, rh M500 2007 64°7¢20¢¢ 51°22¢19¢¢ 550 tair, rh, Qsi M1000 2008 64°9¢13¢¢ 51°21¢10¢¢ 1000 tair, rh, wspd tair = 2-m air temperature (°C), rh = relative humidity (%), wspd = wind speed (m s-1), wdir = wind direction (°), Qsi = incoming shortwave radiation (W m-2), Qli = incoming longwave radiation (W m-2), and snod = snow depth (m).

Table 2. Episodic Snowmelt Event (ESE) Char- and turbulent fluxes at high wind speeds are often acteristics associated with high melt rates (Dadic and others 2013). The duration of the local past ESEs (2008– Snow depth > 0.0 m 2013) was determined by the number of con- Daily mean air temperature > 0.0°C secutive days, where these three criteria were met. Daily mean wind speed > 5.5 m s-1 Part I Findings Identification of Past ESEs in Kobbefjord Over the 5-year period, September 2008–August To identify past ESEs, we used local meteorological 2013, we identified 31 ESEs in Kobbefjord using data and snow observations collected before, dur- available meteorological data (Table 1) and the ing, and after the March 15–16, 2013 ESE. The automated algorithm (Table 2). The ESEs occurred three-point characterization of ESEs (Table 2), from mid-October until mid-May, and ranged be- based on the March 2013 ESE, included conditions tween 3 and 13 ESEs per year. The identified ESEs required for snowmelt, such as snow presence and varied in duration from 1 to 3 days; however, 81% above-freezing air temperatures. We also included of all identified ESEs were 1-day events. Further- a relatively high daily wind speed (corresponding more, decreases in snow depth during the ESEs to the 90th percentile of daily mean wind speed were observed. In Part II, we investigated how data for 2012–2013) for an ESE to occur. The wind these changes in the snowpack and snow cover speed threshold was introduced because high wind were affecting ecologically relevant snow proper- speeds were present during the March 2013 ESE, ties in Arctic ecosystems during and after the ESEs. Episodic Snowmelt Events in Arctic Ecosystems 843

PART II: ECOLOGICAL RELEVANCE OF ESES validated against observations from Kobbefjord. IN KOBBEFJORD Second, the model outputs were used to estimate the snowpack changes during the 31 identified Quantifying the impact of ESEs on snow properties ESEs in Part I. Finally, we quantified the spatial in low-Arctic ecosystems required application of a distribution of the snow meltwater loss and re- spatially distributed snow-evolution modeling sys- duction in the snow thermal resistance, and esti- tem to convert the simple environmental variables, mated the number of days that the spring snow- given in the ESE identification algorithm (Table 2), free date would occur earlier in a point in the valley into more complex, ecologically relevant variables. center (KOB, Figure 2) due to the meltwater loss Because the snowpack changes potentially affect a associated with the March 2013 ESE. wide range of ecosystem components and processes (see ‘‘Introduction’’ section), we chose to focus our analyses on quantifying the ESE-induced changes in Methods and Data (1) snowpack water content, (2) snow thermal re- Model Description sistance (that is, the snow insulating effect), and (3) snow-free date. Each of these three snow properties To obtain spatial and temporal snow distributions can affect a range of low-Arctic ecosystem compo- through the 5-year period, 2008–2013, for the nents, even after the snow season is over. Most Kobbefjord study area, we implemented Snow- notably, these snow-related features can impact the Model (Liston and Elder 2006b). SnowModel con- early and middle portion of the growing season. For sists of three interconnected submodels (Figure 3): example, the meltwater lost during an ESE will be EnBal, SnowPack, and SnowTran-3D. EnBal cal- unavailable during the onset of plant growth in the culated the surface energy exchanges and snow- spring (Blankinship and others 2014). In addition, melt (Liston 1995, 1999); SnowPack modeled the changes in snow thermal resistance may change the evolution of the snowpack in time and space by soil thermal conditions, limiting the decomposition accounting for snowfall, snow density evolution, processes and nutrient availability in the soil during and snowmelt (Liston and Hall 1995; Liston and the growing season (Pattison and Welker 2014). As Mernild 2012); and the blowing-snow transport further examples, changes in spring snowmelt tim- was generated by SnowTran-3D (Liston and Sturm ing may potentially alter the timing of net CO2 up- 1998; Liston and others 2007). The three sub- take (Lund and others 2012), insect emergence models were coupled with a high-resolution simple (Høye and others 2007), plant flowering (Cooper meteorological model, MicroMet (Liston and Elder and others 2011), and vegetation green-up (Elleb- 2006a), which spatially distributed the meteoro- jerg and others 2008). logical input variables over the simulation domain To quantify these potential affects, first a spatially and provided input to SnowModel. distributed snow-evolution model was run over the MicroMet and SnowModel require meteorological Kobbefjord area and model outputs of SWE, snow station and/or gridded atmospheric forcing inputs of depth, and timing of the snow-covered period were air temperature, relative humidity, precipitation,

Figure 3. Input variables, SnowModel submodels, and output variables. SWE = Snow-water-equivalent, AWS = Auto- matic weather station, and RCM = Regional climate model. 844 S. H. Pedersen and others wind speed, and wind direction. All meteorological shrub heath’ and ‘Tall shrub copse,’ were based on variables, except precipitation, were provided by the 48 manual vegetation height measurements dis- four meteorological stations installed in the study tributed in the main valley in July 2012 (Wester- area (Table 1; Figure 2) for the period September gaard-Nielsen and others 2013). 2008–August 2013, which defined the temporal boundaries of the Kobbefjord SnowModel simula- Assimilation Data tions. Because of uncertainties associated with in si- Observed SWEs were used as part of the SnowModel tu winter precipitation measurements (Goodison integrations to overcome uncertainties in the MER- and others 1998), SnowModel precipitation inputs RA reanalysis precipitation rates (Reichle and others were provided by NASA’s Modern-Era Retrospective 2011). NERO conducted annual snow surveys every Analysis for Research and Applications (MERRA) 2/ spring (March–May) during 2008–2013, where 3 longitude by 1/2 latitude gridded atmospheric ° ° snow pits were dug in the same locations (Figure 2) reanalysis precipitation data (Rienecker and others each year to provide detailed measurements of the 2011); daily precipitation rates from the six nearest snowpack grain size, grain type, hardness, stratigra- MERRA data grid points were spatially extrapolated phy, density, and temperature. Observed SWE val- by MicroMet to cover the Kobbefjord simulation ues were calculated from the bulk density domain. Furthermore, measured incoming short- observations and the total depth of these snow pits wave and longwave radiation were included in the (Figure 2) and used to adjust MERRA precipitation SnowModel energy balance calculations as an im- fluxes using SnowAssim (Liston and Hiemstra 2008) proved alternative to the MicroMet calculation of running within SnowModel. This was done under the radiation components. In addition to the three the constraint that modeled SWE matched the ob- submodels, a data assimilation scheme, SnowAssim served SWE both in time and location. (Liston and Hiemstra 2008), was included in the model runs. SnowAssim was used to correct the Validation Data input of precipitation rates by constraining the modeled field of SWE by the observed pre-melt Model outputs were validated against observations SWE. This technique, pioneered by Liston and of SWE, snow depth, and albedo. Observed SWE Sturm (2002), has proven to be an effective way to was obtained from 3 to 6 snow pit locations per year produce realistic precipitation fluxes when available in the main valley; SWE was calculated using mean precipitation datasets may be inadequate. snow depths measured along 100 m by 100 m cross SnowModel was run over the spatial domain transects and from snow bulk density from snow pits shown in Figure 2 using a 10 m by 10 m grid in- (Figure 2). We used the observed density values crement and daily time steps. Required digital obtained from the snow pits in the assimilation elevation model (DEM) data were based on a di- (letting SnowModel simulate the snow depth) and gitized version of a 25-m topographic contour in- validation because snow density is conservative and terval map and provided on the same grid as its spatial variation typically falls within well-de- SnowModel. Also required by SnowModel are fined limits (Sturm and others 2010); the available snow-holding depths (SHDs), that is, the depth bulk density observations made within 1–2 days in below which the vegetation captures the snow and the valley varied less than 8%. prevents snow-transport by wind. The assigned Automated snow depth (sonic ranging sensor) SHDs (Table 3) for the two vegetation types, ‘Low measurements available during 2008–2013 from the

Table 3. Estimated Snow-Holding Depths (SHDs) for Land-Cover Types Present in Kobbefjord

Land-cover type1 SHD Land-cover (m) fraction (%)2

Permanent snow/glaciers 0.01 8.3 Fjord and lakes (possible frozen) 0.01 3.0 Exposed bedrock and fell ﬁeld/wind-blown area 0.01 69.0 Fen (Carex rariﬂora, Scirpus caespitosus, Eriophorum angustifolium) 0.15 <0.1 Low shrub heath (Empetrum nigrum, Vaccinium uliginosum, Betula nana) 0.15 8.2 Tall shrub copse (Salix glauca) 0.60 2.3

1From Bay and others (2008). 2Based on the area of the simulation domain of approximately 71 km2. Episodic Snowmelt Events in Arctic Ecosystems 845

KOB climate station (Figure 2) were used to validate where SD is the snow depth (m), A is a unit area 2 -1 -1 modeled snow depth. The timings of the start and (1 m ), and keff_snow_type (W m K ) is the snow end of the modeled snow-covered periods were thermal conductivity specific for the two observed compared with the observed snow-covered periods. snow types in the snowpack (Liston and others To define the start and end of the observed snow- 2002). Snowmelt during ESEs can produce snow- coveredperiods,weusedobservedalbedoduring pack SWE reductions that lead to an earlier snow- 2008–2013 from KOB. The observed start of the free date during spring snowmelt. This means that snow-covered period was defined as the day the the amount of meltwater lost during the ESE rep- observed albedo exceeded 0.8, and the end of the resents an amount of snow, which is lacking in the period as the day the albedo went below 0.2. For the end-of-winter snowpack, regardless of additional modeled snow cover, at the SnowModel grid cell precipitation between the time of the ESE occur- coincident with the KOB station location, the start of rence and the end of the snow-covered period. the snow-covered period was defined to be the first How much earlier the snow-free date is depends on day with snow depth greater than 0.0 m, and the end the amount of snowpack SWE lost during the ESE was the day before the snow depth equaled 0.0 m. and the spring snowmelt rate. The spring snowmelt rate is given for each hydrological year, because it Estimating ESE-Related Snow-Property Changes depends on year-specific cloud cover and snow albedo evolution. The change in snow-free date (in We estimated the spatial distribution of meltwater days) was assumed to equal to the ratio between loss during the March 2013 ESE by calculating the the total ESE SWE loss in a given year (Table 4) difference in modeled SWE before and after these and the year-specific spring snowmelt rate (0.004– two dates for all grid cells in the simulation domain. 0.030 m water equivalent per day), both derived The meltwater loss associated with the 31 identified from the SnowModel outputs for one point in the ESEs was estimated by calculating the difference in valley. SWE depth between the start and end day of each ESE and accumulated for each hydrological year. Similarly, the change in snowpack thermal resis- Part II Findings tance, R (K W-1), resulting from the March 2013 The modeled snow depth showed pronounced in- ESE, was calculated as the difference between the terannual variation from 2008 to 2013 (Figure 4). spatially distributed R prior to and after the March Snow depth was expected to decrease during ESEs, 2013 ESE. R was based on the sum of R for two and during the March 2013 ESE, we saw a 55% different snow type layers, new/recent snow and decrease in the modeled snow depth (Figure 4,black depth hoar, that we observed in the snowpack dur- line). The hydrological year 2009–2010 was snow- ing field work in March 2013. R was estimated from poor with a mean snow depth of 0.15 m and several SD Â snow type 1 fraction snow-free periods during the winter season (Fig- R ¼ ure 4). The earliest end of the snow-covered period k Â A eff snow type1 was in late April 2010. In all other years, the snow- ð1Þ free date occurred between mid-May and mid-June. SD Â snow type 2 fraction þ ; The longest continuous snow-covered period of keff snow type2 Â A 236 days (7.7 months), the maximum annual mean

Table 4. ESEs Identiﬁed for the Hydrological Years (September 1–August 31) from 2008 to 2013 and Their Modeled Accumulated Effect on the Snowpack in the Grid Cell Corresponding to the Location of the KOB Climate Station (Figure 2)

Hydrological year Number of ESE Total water Earlier Total water ESEs (#) duration loss from ESEs snow-free loss as fraction (days) (m water equivalent) date (days) of annual precipitation (%)

2008–2009 13 15 0.31 10 35 2009–2010 2 4 0.31 10 52 2010–2011 4 4 0.19 6 25 2011–2012 3 4 0.22 12 23 2012–2013 (March 2013 ESE) 9 12 0.19 11 23 (1) (2) (0.12) (4) (14) 846 S. H. Pedersen and others

Figure 4. Modeled snow depth (m) from one grid point with low shrub vegetation in ﬂat terrain, located in the center of the study area (Figure 2, near KOB) for the hydrological years (September 1–August 31) from 2008 to 2013.

snow depth (0.71 m), and the maximum snow depth observed snow-covered periods showed a high (1.24 m) were all found during 2011–2012. correlation with the timing of modeled snow cover (Linear regression statistics: intercept = 0.081, Model Validation slope = 1.000, n = 15, P < 0.001, R2 = 0.999). The modeled snow depth was validated against observed snow depth measured by an automated ESE-Related Snow-Property Changes snow-depth sensor at KOB (Figure 5A). The linear regression between all available snow depth mea- Snow depth and snow density observations col- surements (September 2008–August 2013, n = lected before and after the March 2013 ESE en- 1560) and modeled snow depths showed that the abled evaluation of the SnowModel performance relationship between the modeled and the ob- during an ESE. Modeled and observed SWE dif- served snow depths was statistically significant fered by 0.03 m on March 14, 2013 and 0.04 m on (P < 0.001), and the SnowModel output explained March 19, 2013 in the point of the west-most snow 85.3% of the variation in the observed snow depth. pit (Figure 2). The simulated snowpack water loss The largest difference between modeled and ob- between the two dates was partitioned into 98.9% served snow depth of 0.41 m was found during meltwater runoff and a minor moisture loss (1.1%) 2011–2012. The validation of the modeled SWE from the snowpack surface by sublimation. against the observed snowpack water content On March 14, we observed a 0.90 m deep snow- showed that the modeled SWE explained 74.3% of pack in the west-most snow pit (Figure 2) with a top the interannual variation in observed SWE, and the 0.40 m thick fine-grained recent/new snow layer -3 linear fit between the modeled and observed SWEs with a density of 185 kg m (44% of the snow was statistically significant ( < 0.001) (Fig- depth), and a bottom 0.50 m thick layer of coarse- P -3 ure 5B).This interannual SWE variation is strongly grained depth hoar with a density of 300 kg m related to the interannual water equivalent snow (56% of the snow depth). The thermal conductivity -1 -1 (keff) for the new snow layer is 0.062 W m K precipitation used in the simulations (not shown). -1 -1 Modeled snow-covered period timing and dura- and keff for depth hoar is 0.126 W m K based on tion were compared with observed start and end of the quadratic equation in Sturm and others (1997). the snow-covered period(s) defined by the albedo The vertically integrated R for that snow pit was evolution (Figure 5C). On average, the timing of estimated using equation 1 modeled start and end matched the observed with 1–2 days difference. However, during 2011–2012, 0:90 m Â 0:44 RMarch 14; 2013 ¼ 1 À1 2 the start and end were offset by 7 and 5 days, re- 0:062 W mÀ K Â 1m spectively (Figure 5C). Modeled snow cover also 0:90 m Â 0:56 þ captured the timing of shorter snow-covered and 0:126 W mÀ1 KÀ1 Â 1m2 snow-free periods during October–February. All À1 dates of start and end from both shorter and longer ¼ 10:4KW : Episodic Snowmelt Events in Arctic Ecosystems 847

Figure 5. A Regression between modeled and observed snow depths in the location of the climate station, KOB, at time increments (daily), where observed snow depth were available, that is, September 2008–August 2013. Linear ﬁt statistics 2 (solid line): Intercept = -0.016, slope = 0.826; R = 0.853, F1,1560 = 9070, P < 0.001. Dotted line is 1:1 line. B Regression between the observed and the modeled snow-water-equivalents (SWEs). Linear ﬁt statistics (solid line): Intercept = 0.089, 2 slope = 0.530; R = 0.743, F1,46 = 133.5, P < 0.001. Dotted line is 1:1 line. C Snow-covered periods, where modeled snow depth (black lines) is above 0.0 m in comparison with the observed timing (red crosses) of snow-cover onset (albedo is above 0.8) and snow-cover end (albedo is less than 0.2) from 2008 to 2013. No observed albedo data were available prior to August 2008.

Due to small spatial variability in snow depth and and 16, 2013 (end-of-day modeled outputs) be- vegetation height in the area, on March 19, 2013 cause the observed snow depth (0.38 m) in the we dug a new snow pit 2 m away from the March snow pit on March 19 was representative for the 14, 2013 snow pit. The top 0.40 m of new snow range of modeled snow depth for the valley on that had melted away and only the lowest 0.38 m of the date (0–50 cm). We assumed that the snowpack did depth hoar layer remained; the bulk density of the not change much from March 16, until the obser- bottom layer had increased 2%, which corre- vations were made on March 19, 2013, because -1 -1 sponded to an increase in keff of 0.006 W m K . both automated snow-depth sensor measurements The resulting R on March 19, 2013 was calculated at KOB and modeled snow depth showed that the for the remaining bottom depth hoar layer as, majority of the snowmelt took place during the 2 days (March 15 and 16). 0:38m Â 1:00 During the March 2013 ESE, the mean water loss R ¼ ¼ 2:9KWÀ1: March19;2013 0:131WmÀ1 KÀ1 Â 1m2 over the 2 days per grid cell (10 m by 10 m) ranged between 0.06 and 0.12 m water equivalent in the These estimates resulted in 72% snow thermal re- valley below 200 m a.s.l. (Figure 6A), which rep- sistance reduction from the observed snowpack resents 50–80% of the pre-melt snowpack water from March 14 to March 19, in the west-most snow content. These differences are primarily related to pit. We applied this partitioning of 44% new snow slope, aspect, and elevation variations associated and 56% depth hoar to the spatially distributed with the simulation domain. The resulting modeled modeled snow depth on the days March 14, 15, snow thermal resistance loss during the March 848 S. H. Pedersen and others

Figure 6. A Summed water loss (m water equivalent) from snowpack during ESE on March 15 and 16, 2013. B Loss of snow thermal resistance (%) during March 2013 ESE. Lower left corner coordinates are 64°7¢25.4¢¢N, 51°25¢24.8¢¢W.

2013 ESE varied spatially as well, varying from 40 than our local Kobbefjord study domain. Hence, we to 100% in the valley area below 200 m a.s.l. then looked beyond the Kobbefjord valley and fo- (Figure 6B). Furthermore, observations of snow- cused in Part III on the greater coastal ice-free area free areas in the digital photo from March 16, 2013 of Greenland using a meteorological dataset (Figure 1) was consistent with the modeled 100% (MERRA; Rienecker and others 2011) that mat- snow depletion being confined primarily to small, ched the synoptic scale (1000 km) over which the local hill tops in the valley area. low-pressure weather patterns occur. Therefore, in By applying SnowModel to the Kobbefjord study Part III we investigated whether past ESEs could be area, we were able to quantify the recent local ESE- identified further north, south, and east of Kobbe- associated changes in ecologically relevant snow fjord and estimated their temporal frequency and properties in the studied low-Arctic ecosystem. spatial extent. Among the 31 identified ESEs (Table 4) we found maximum meltwater losses of 0.12 m water PART III: PAST ESESINGREENLAND equivalent during October 27–28, 2008 and 0.12 m during March 15–16, 2013. The meltwater lost Methods and Data during these two ESEs alone represented more To identify trends in the occurrence of past ESEs than a tenth of the annual precipitation sum and over the ice-free part of Greenland, we used the were estimated to cause up to 3.8 and 3.8 days automated ESE identification algorithm (Table 2) earlier snow-free date, respectively. The majority of on this regional scale. As input, we used a Green- the identified 1-day ESEs led to thinning of the land subset of MERRA reanalysis data (see Part II; snowpack, and thereby a loss of snow thermal re- Rienecker and others 2011) that covered the 34- sistance. The meltwater loss associated with single year period from September 1, 1979 to August 31, ESEs (1- to 3-day ESEs) ranged between 0.0 and 2013 and included the variables: air temperature, 0.05 m water equivalent. However, the accumu- SWE, and wind speed. lated meltwater loss through the winter for all events occurring within the same hydrological year Part III Findings equaled 23–52% of the annual precipitation, which potentially represented an advancement of 6– We found that the ESE identification algorithm 12 days in spring snowmelt per year (Table 4). For applied on the MERRA data reproduced all the seven of the ESEs identified in the meteorological Kobbefjord ESEs. This allowed us to conclude that data (Part I), the SnowModel outputs showed no it was appropriate to run the ESE identification snowmelt. The meltwater loss along with a 40– algorithm over Greenland using the MERRA en- 100% decrease in the snow insulation effect can vironmental data. The results also showed that potentially influence the dynamics and functioning ESEs have indeed been occurring outside the of the ecosystem on this local scale. Kobbefjord area during the past 34 years. However, In light of the ecological importance of ESEs on the majority of the identified ESEs were mainly snow properties in the low-Arctic terrestrial confined to west–southwest Greenland (Figure 7). ecosystem, and that the meteorological variables Furthermore, the spatial representation of the controlling the occurrence of ESEs are assumed to identified ESEs in west Greenland (Figure 7) be strongly tied to middle- and high-latitude low showed a relatively larger number of ESEs along pressure systems (Steffen and Box 2001; Mernild the Greenland Ice Sheet margin and near the coast. and others 2014), we expected the scope of influ- The timings of the identified ESEs differed between ence associated with ESEs to cover a larger area west–southwest Greenland (59°–75°N) and north- Episodic Snowmelt Events in Arctic Ecosystems 849

2011–2012 could be explained by a precipitation event resulting in either snowfall or rainfall in SnowModel, depending on the simulated air temperature. Because we were operating in a valley surrounded by mountains (up to 1375 m a.s.l.), temperature lapse rates heavily affect the snowfall variation with elevation. To account for this, we incorporated daily temperature lapse rate estimations in MicroMet, based on air temperature observations from climate stations in the valley (KOB, Table 1) and on the mountain tops (M1000 and M500, Table 1), when data were available. During time steps where data were unavailable from M1000 and M500, we used monthly mean lapse rates based on data from the whole time series (2008–2013). This was the case for the period January 1–August 31, 2012, where M1000 data Figure 7. Mean annual number of ESEs identified dur- were lacking. Hence, the applied monthly mean ing the period 1979–2013. lapse rate during this end-of-winter/spring season may have led to a later snowmelt onset than the west Greenland (75°–84°N). In southwest Green- observed and, therefore, created a mismatch be- land, the ESEs occurred mostly from October to tween the modeled and the observed snow-free November, while the lower number of ESEs iden- date (Figure 5C). In addition, on days when the tified in north Greenland occurred from June to monthly mean lapse rates were applied, it could September. Conducting a linear temporal trend occur that the model output showed snowfall in analysis for each individual grid cell on the basis of both the valley and on mountain tops, but in re- the 34-year time series of the annual number of ality the snowfall was confined to the mountain top identified ESEs showed a statistically significant because of a relatively higher temperature lapse positive trend (P < 0.05) in a few areas of west– rate present that day. During snow-cover onset in southwest Greenland and southeast Greenland. In October 2011, air temperature observations were addition, a statistically significant negative trend available for all three meteorological stations: KOB, was found in the outer coastal areas of northwest M500, and M1000 (Table 1). This enabled inclu- Greenland (>81.0°N), where relatively few ESEs sion of observed lapse rates in the model run and were identified (Figure 7). on October 2, 2011, the daily mean air temperatures were 1.93; -2.62; and -5.14°C at the three stations, respectively. However, it resulted in DISCUSSION modeled snowfall both at the mountain tops and in The validation of model outputs demonstrated that the valley, because the snow-rain phase threshold SnowModel reproduced the snow distribution in in MicroMet was set to 2.0°C according to Auer Kobbefjord well, both in terms of timing and du- (1974). Furthermore, automated camera photos ration of the snow-covered periods and in match- showed snowfall only on the mountain tops that ing the observed snow depth and SWE during the same day. Valley snowfall events similar to this years 2008–2013. Only during 2011–2012 did the occurred until October 23, and resulted in a mod- modeled snow-covered period and deep snowpack eled snow depth of 0.39 m on that date (October show less-than-ideal correspondence with the ob- 23, 2011) when the observed snow depth went servations. This was partly caused by uncertainties above 0.0 m in the valley for the first time that in the measured snow depth at the snow-cover winter. This caused a modeled snow depth that was onset in October 2011. Although daily automated 0.39 m greater than the observed valley snow camera photos (as in Figure 1) and observed albedo depth. This depth increment persisted throughout from KOB (Figure 2) showed the first snowfall on the winter and produced the differences found October 9, the snow-depth sensor at KOB showed between the modeled and the observed snow and 0.0-m snow depth until October 23. The difference SWE depths during the 2011–2012 winter. between modeled and observed snow depth is Based on the model validation, we used Snow- likely to be explained by these snow-depth mea- Model outputs for Kobbefjord to estimate snow- surement errors. In addition, the difference during property changes for ESEs identified with the 850 S. H. Pedersen and others automated identification algorithm during 2008– frost sensitivity varies between plant species. Two 2013. The snow thermal resistance (R) calculations dwarf shrub types, Empetrum hermaphroditum and were not validated because no such measurements Vaccinium vitis-idaea found in south and west Green- have been conducted in Kobbefjord. However, the land, including Kobbefjord (Bo¨cher and others 1978; RMarch 14, 2013 and RMarch 19, 2013 have been found Bay and others 2008), respond differently to expo- to correspond to R of snow covers found on Arctic sure to freezing temperatures. Extreme winter dry and scrub tundra (Liston and others 2002). warming experiments in a sub-Arctic ecosystem have Nevertheless, some observations of snow stratigra- shown that E. hermaphroditum is prone to frost phy were available in the valley prior of the March damage, whereas V. vitis-idaea showsnooronly 2013 ESE to support the extrapolation of 44% new limited sign of frost damage (Bokhorst and others snow and 56% depth hoar from one observation 2011). Hence, because ESEs appear to be a common point to the rest of the valley. This means that the part of the snow-cover dynamics in west Greenland, decrease in snow thermal resistance (Figure 6B) the species living there are likely to be resilient or may be underestimated in areas with less depth resistant to frequent frost exposure (see Bokhorst and hoar fraction than 56%, while it could be overes- others 2010; Semenchuk and others 2013). However, timated in areas with higher depth hoar fraction, the adaptive capacity of some species in the Arctic depending on the depth hoar development in the ecosystem may be challenged due to a future in- snowpack on different surfaces (Sturm and John- creasing frequency of ESEs in west Greenland. son 1992; Benson and Sturm 1993). Still, depth In Arctic ecosystems dependent on winter pre- hoar crystals were observed in all (6) snow pits cipitation and snow accumulation as a moisture after March 2013 and are common in a tundra source for plant growth during the growing season snowpack (Benson and Sturm 1993; Sturm and (Elberling and others 2008; Brooks and others others 1995). Hence, the depth hoar fraction in- 2011), the consequences of ESE-related meltwater troduces uncertainty in these results. loss of up to 52% of the annual precipitation from the snowpack may reach into the plant growing Ecological Effects of ESEs season and impact growing conditions. The meltwater loss may have caused a moisture shortage at To assess the potential effects on ecosystem com- the beginning of and during the growing season ponents and processes caused by ESEs, knowledge (Ellebjerg and others 2008; Schmidt and others of ESE-related changes in snow properties, such as 2012) and potentially limited the vegetation growth those presented in Part II, are required (see Bo- for some species (Ellebjerg and others 2008; Schmidt khorst and others 2010). For instance, the March and others 2012; Rumpf and others 2014). Fur- 2013 ESE caused changes in the soil thermal re- thermore, the lack of soil moisture through the gime due to loss of the snow-cover insulating ef- growing season may have increased decomposition fect. Observations showed a 3.0°C increase in 3-day rates of organic matter in the soil and limited the average soil temperature between before and after gross primary production (Lund and others 2012), the March 2013 ESE in a valley area (Soil station, thereby affecting the ecosystem CO2 balance Figure 2), which experienced a 60–70% reduction (Brooks and others 2011). Meltwater lost from the in thermal resistance (Figure 6B). Also, an ESE snowpack during winter and spring during an ESE is may cause frost damage to vegetation, if the pre- likely to be lost through surface runoff, because the melt snow depth is shallow enough for the snow ground is frozen, and the water is routed under the cover to deplete completely during an ESE (Bo- snowpack into streams and rivers (Bayard and oth- khorst and others 2011). The ESE observed in ers 2005; this study). The meltwater loss during the Kobbefjord in March 2013 (Table 4) may have re- March 2013 ESE represented up to 80% of pre-melt sulted in such vegetation frost damage, because the snowpack water content and up to 14.2% of the snow cover depleted completely on the hill tops annual precipitation. The model outputs showed and in other thin-snow areas. This resulted in a that this substantial amount of water was lost pri- patchy insulating snow cover (see Figure 1) and marily through runoff. This is supported by auto- exposed vegetation to freezing temperatures (daily mated soil temperature measurements in and below mean air temperature of -3.3°C). 10-cm depth in heath- and fen-covered soils in The majority of the identified ESEs in Kobbefjord Kobbefjord. The data showed freezing temperatures resulted in a thinning of the snowpack and a reduced from November to March, which thereby limited insulating effect, which may not prevent frost dam- the meltwater percolation into the soil and fa- age and reduced flowering the following growing cilitated lateral runoff into streams and lakes (Ba- season(s) (Semenchuk and others 2013). However, yard and others 2005). Episodic Snowmelt Events in Arctic Ecosystems 851

The March 2013 ESE was the single ESE having round, because changes and interannual variations the largest impact on the snowpack in terms of observed during the heavily studied summer field thinning of the snow cover (0.20 m), meltwater season may be explained by events or changes oc- loss (0.12 m water equivalent = 14% of the 2012/ curring during winter and/or the shoulder seasons. 2013 annual precipitation), and potential advancement of the spring snowmelt (4 days). This ESE Scale-Issues, Driving Factors, and emphasizes that the March 2013 ESE was stronger Timing than the regularly occurring ESEs. However, because the cumulated meltwater loss from all annual The comparison between the number of identified ESEs potentially equal 23–52% of the annual pre- ESEs locally in Kobbefjord through station obser- cipitation, the additive effect from all ESEs plays a vations (Part I) and a MERRA grid cell matching major role for the hydrological cycle of the the same location is challenged by the lack of ecosystem. Furthermore, the potential 6–12 days subgrid variability in topography and physical advancement of the start of the snow-free season processes within a MERRA grid cell. The 2/3° by 1/ can potentially cause a shift in a range of biotic and 2° MERRA grid cell covering the Kobbefjord valley abiotic terrestrial processes, such as plant-flowering and surroundings spans an elevation range from phenology (Høye and others 2007; Cooper and sea level to 1400 m a.s.l. For comparison, the others 2011; Iler and others 2013), gas–flux ex- photos in Figure 1 and the study area map in Fig- change (Brooks and others 2011; Lund and others ure 2 correspond to approximately 1/4 of the 2012), arthropod emergence (Høye and others MERRA grid cell. The topographic variation is the 2007), and avian-breeding phenology (Meltofte main driver for the spatial variations in air tem- and others 2007), which are dependent on the perature, wind speed, and snow cover with eleva- timing of the spring snowmelt and thereby ground tion and across the landscape (Liston and Elder exposure. 2006a). Such variation is inevitably not captured All these examples and observations illustrate the with one single data value per MERRA grid cell for potential impact that ESEs have on the Arctic any of the three climate variables used in the ecosystem through changes in snow properties. identification algorithm (Part 1). A direct compar- Our analysis documented that ESEs have been a ison between the MERRA grid cell, covering the common and natural part of the snow-cover dy- location of the KOB station, and the KOB station namics in west Greenland at least since 1979, and data for the years 2008–2013 resulted in 12% most species found in Greenland are thus likely higher mean annual air temperature and 22% adapted to this phenomenon. However, an in- higher mean annual wind speed in MERRA than in creasing frequency of ESEs may put the ecosystems KOB station observations. Hence, due to this scal- under severe pressure, ultimately resulting in al- ing-issue, more ESEs were identified with MERRA terations in local species composition (see for ex- data as input to the identification algorithm than ample, Elmendorf and others 2012a, b) to cope when using Kobbefjord station data (Part I). We are with reoccurring abrupt meltwater losses and de- therefore aware that the mean annual number of creases in snow-cover insulating effect. We focused identified ESEs in Kobbefjord, and likely in other on three snow properties in this study; snow ther- areas in Greenland, may be overestimated when mal resistance (controlling the thermal conditions using the MERRA data. We are, however, confi- in the soil and thereby the soil heterotrophic ac- dent that the MERRA dataset captures the general tivity during the snow-covered period (Nobrega patterns of the Greenland atmospheric environ- and Grogan 2007; Gouttevin and others 2012)), ment and most importantly, the weather features meltwater content available for plant growth, and causing the ESEs, for example, latitudinal and the snow-free date defining the growing-season coast–inland gradients in air temperature. The onset. As discussed above, these three snow prop- MERRA data thus represent the relative spatial and erties are, individually and combined, driving and temporal distributions and differences in ESEs be- controlling ecosystem processes occurring outside tween regions in north, south, east, and west the snow-covered period, that is, within the Greenland. growing season. Hence, changes occurring during The driver for the ESEs is likely related to large- winter, such as an ESE, may cause ecosystem scale weather patterns occurring on the west coast changes during the following growing season(s) of Greenland and particularly in southwest (Bokhorst and others 2011; Semenchuk and others Greenland, where we found the highest ESE fre- 2013). This highlights and emphasizes the need for quency (Figure 7). The ESE-associated high wind ecosystem-monitoring programs that run year- speeds, rapid increase in air temperatures, and the 852 S. H. Pedersen and others location of the highest ESE frequency suggested a September, when air temperatures reached above possible relation to foehn winds. The assumption 0°C (DMI 2014). The combination of small solar that foehn winds are driving the ESEs is further- angle, that is, relatively limited energy amounts more supported by the finding that the highest reaching these northern latitudes, and air tem- frequency values of ESEs are in southwest Green- peratures ranging between -30 and -15°C (Cap- land (Figure 7), where foehn winds are the stron- pelen 2011), limits the chance for air temperatures gest and most frequent (Fristrup 1953; Gorter and in northwest Greenland rising above freezing for others 2014). The first scientific description of most of the year. Therefore, ESEs are less likely to foehn winds in west Greenland was by Hoffmeyer occur in the north, and may have less ecosystem (1877). This early publication year suggests that impact than at lower latitudes because they mainly foehn winds have been a feature of west Greenland occur during spring or summer, when the snow climate for a long time. During foehn events, warm cover is depleting. Furthermore, the low ESE fre- dry wind is able to create high snowmelt rates. In quency north of 81.0°N has significantly decreased Greenland, foehn winds are driven by the Green- over the last 34 years. In middle-west Greenland, land katabatic wind system (Figure 5 in Cappelen the ESEs predominantly occurred during the onset and others 2001; Gorter and others 2014), which is of the snow-covered season during October and driven by the density difference between the heavy November. The region below the Arctic Circle cold air close to the ice surface of the Greenland Ice (approximately 66°34¢N) is less affected by reduced Sheet interior and the lower density and warmer solar radiation during winter. This region has gen- air above. When the winds approach the ice mar- erally higher air temperatures than further north, gin, the wind speed is accelerated due to topo- so that a temperature rise during winter would graphic channeling and the steeper slopes of the more likely result in above-freezing temperatures, coastal mountains and outlet glaciers. The out- and therefore ESEs were also identified through the flowing air mass is compressed as it descends from winter months in this area. The early-winter timing the ice sheet because of the altitude change, and of ESEs may cause frost penetrating into the soil if the air pressure increase results in heating of the air the ground is exposed to subfreezing temperatures mass. Figure 7 includes not only the ice-free areas, after an ESE. These frozen soils can be preserved but also the Greenland Ice Sheet margin to identify during the winter, when the snow cover is whether the frequency of the identified ESEs reestablished (Zhang 2005). Also, meltwater from would potentially be higher on the ice and there- early-winter ESEs may create ice layers within or fore support the argument that foehn winds are a below the remaining snow cover, which will po- driver for ESEs, because foehn winds originate tentially block food access for herbivores through from the Greenland Ice Sheet and reach high wind the winter and have fatal consequences for a speeds in the ice marginal area. Indeed, Figure 7 population (for example, Hansen and others 2011). shows a pattern of more ESEs being closest to the In south Greenland, the identified ESEs occurred ice sheet margin and at the outer coast in west through the autumn, winter, and spring, that is, Greenland (65°–72°N). The climatic conditions for from October to mid-May. In this region, local ESEs to occur are likely to be favorable in both sheep herding is dependent on soil moisture places. During a foehn event, the area near the ice originating from spring snowmelt for vegetation sheet margin is relatively warm and relatively more growth, especially in meadows and snow-beds that windy than further out by the coast (Cappelen and provide the majority of the plant forage for sheep others 2001), whereas near the outer coast, the air grazing in the mountains (Austrheim and others temperature is generally higher than inland in 2008). If meltwater amounts are reduced due to winter because of the relatively warm ice-free Da- ESEs during winter, it may result in reduced vid Strait. However, the outer coast is also windier vegetation growth, and thereby limit food avail- because the high wind speeds associated with ability for the sheep during summer grazing in ar- foehn winds may be sustained through channeling eas already at risk of overgrazing (Aastrup and in fjord systems and valleys on their way to the others 2014). The north–south difference in ESE outer coast and fjord heads. This pattern is consis- frequency is most likely to be tied to the locations, tent with our March 2013 ESE observations, be- where foehn winds are most frequent and cause Kobbefjord is located approximately 100 km physically possible to occur in terms of air tem- from the ice sheet margin. perature and wind regime, which in turn are con- The timing of ESE occurrences differed between trolled by synoptic-scale weather circulation northwest and southwest Greenland and the few patterns and topography and slope gradients, re- northern-identified ESEs occurred from June to spectively (Gorter and others 2014). Episodic Snowmelt Events in Arctic Ecosystems 853

ESEs in a Changing Climate a weakening of the Jetstream due to a reduced temperature gradient between the Northern For future perspectives, the predicted reduction in Hemisphere middle-latitudes and high-latitudes the temperature deficit over the center of the caused by increased warming in the Arctic. A Greenland Ice Sheet will result in wind-speed re- weakened Jetstream potentially generates slower ductions in the center of the ice sheet, where movement of pressure systems across the North katabatic forcing is predominant. However, as a Atlantic, creating more persistent weather and al- subsequent consequence, the wind speeds are tered large-scale circulation patterns in the North- predicted to increase in the coastal areas with steep ern Hemisphere. This change in circulation pattern topography (Gorter and others 2014), thereby in- is likely to influence the location and strength of creasing the foehn wind frequency. This may cause low-pressure systems in relation to the high-pres- a continued increase in the number of extreme sure over the Greenland Ice Sheet, which may events in, for example, air temperature (Mernild potentially affect future ESE frequency, distribu- and others 2014) and snowmelt (ESEs). Further- tion, and strength, which in turn may challenge more, because the annual mean surface air tem- the Arctic ecosystems in Greenland. perature in coastal Greenland continues to increase (Hartmann and others 2013; Cappelen and Vinther SUMMARY AND CONCLUSIONS 2014), especially during winter (Hanna and others 2012), future increased magnitude and frequency First-hand observations of snow-property changes of air temperatures above freezing may result in and weather conditions during an ESE in March higher snowmelt rates than we have seen in the 2013 in the Kobbefjord coastal low-Arctic study site recent, identified ESEs. in southwest Greenland enabled us to define ESEs During the March 2013 ESE, the foehn event that occur during times when snow depth is greater appeared to be tied to and driven by the location, than 0.0 m, daily air temperature is above 0.0°C, -1 strength, and the associated wind patterns of syn- and wind speed is greater than 5.5 m s . These optic-scale pressure systems. A high-pressure ridge ESE characteristics were the basis for developing an was located over the Greenland Ice Sheet and a automated algorithm that was used to identify 31 low-pressure trough over Baffin Island, which are recent ESEs (2008–2013) occurring on a local scale typical locations for anticyclones and cyclones in Kobbefjord. Next, we applied SnowModel as a during winter (Serreze and others 1993). The tool to model snow distributions in Kobbefjord. The pressure gradient generated south-easterly winds, model outputs were used to quantify ecologically passing in a westward direction over the ice sheet, relevant snow-property changes resulting from the presumably resulting in foehn winds and ESEs in March 2013 ESE. We estimated a water loss west Greenland. If the foehn events are tied to the equaling 50–80% of the pre-melt snowpack water locations and strength of these high- and low- content, and a 40–100% decrease in snow thermal pressure systems, it might explain the pronounced resistance caused by changes in snow depths, snow difference in the timings of ESEs between north- type composition of the snowpack, and thereby west and southwest Greenland, as the location and changes in snow thermal resistance during the ESE. strength of the high- and low-pressure systems Potential impacts from these changes in snow over and surrounding the Greenland Ice Sheet are properties on ecosystem processes and components tied to the seasons and dark/light periods (Serreze include frost damage to vegetation and changes in and others 1993). soil moisture and soil thermal conditions. In addi- In this analysis, we found a positive temporal tion, ESEs occur frequently during the winter and trend in the annual frequency of ESEs in few areas are a part of the snow-cover dynamics in Kobbe- of west and east Greenland (59.9°–81.0°N) and fjord and constitute a significant part of the annual identified large interannual variations in the precipitation (23–52%) lost primarily through number of ESEs through the 34 years. If we assume surface runoff, which potentially could result in 6– that ESEs are driven by foehn winds and that the 12 days earlier spring snowmelt. timing of the foehn winds to some extent is con- On a regional scale of the ice-free parts of trolled by locations and strengths of low- and high- Greenland, we found that ESEs are common in pressure systems in and around Greenland, an west and southwest Greenland. The annual num- amplification of a future continued increase in ESE ber of ESEs in a few areas in west and east frequency could originate from changes in synop- Greenland from 1979 to 2013 showed a statistically tic-scale weather systems. Such changes are de- significant positive trend, but also large interannual scribed by Francis and Vavrus (2012), who suggest variation in the occurrence of ESEs. Based on the 854 S. H. Pedersen and others geographic location of the highest ESE frequencies region—A long term perspective on management, resource and the characteristics identified for ESEs, we economy and ecology. Rapport zoologisk serie 2008-3. Trondheim: Norges teknisk-naturvitenskapelige universitet suggest that foehn winds are likely to be a key Vitenskapsmuseet. p 86. driver of ESEs in Greenland. The ESEs seem to be a Bartsch A, Kumpula T, Forbes BC, Stammler F. 2010. Detection natural and common part of the snow-cover dy- of snow surface thawing and refreezing in the Eurasian Arctic namics in low-Arctic west and east Greenland, and with QuikSCAT: implications for reindeer herding. Ecol Appl the local species assemblages have likely adapted to 20:2346–58. these. However, in a warmer Greenland climate, Bay C, Aastrup P, Nymand J. 2008. The NERO line. A vegetation the frequency of ESEs may continue to increase transect in Kobbefjord. West Greenland: National Environ- mental Research Institute, Aarhus University. p 40. and the associated snowmelt rates might also in- Bayard D, Sta¨hli M, Parriaux A, Flu¨ hler H. 2005. The influence crease due to increasing air temperatures. This will, of seasonally frozen soil on the snowmelt runoff at two Alpine in turn, gradually increase the pressure on the sites in southern Switzerland. J Hydrol 309:66–84. ecosystem, including its species, and the processes Benson CS, Sturm M. 1993. Structure and wind transport of associated with moisture availability and timing, seasonal snow on the Arctic slope of Alaska. Ann Glaciol thermal protection, and growing season (snow- 18:261–7. free) duration. Blankinship JC, Meadows MW, Lucas RG, Hart SC. 2014. Snowmelt timing alters shallow but not deep soil moisture in ACKNOWLEDGMENTS the Sierra Nevada. Water Resour Res 50:1448–56. Bo¨ cher TW, Fredskild B, Holmen K, Jakobsen K. 1978. Grøn- We wish to thank Nuuk Ecological Research Op- lands flora. Copenhagen: P. Haase & Søn. erations and Asiaq, Greenland Survey for providing Bokhorst S, Bjerke JW, Davey MP, Taulavuori K, Taulavuori E, data and helping us with data collection in March Laine K, Callaghan TV, Phoenix GK. 2010. Impacts of extreme winter warming events on plant physiology in a sub-Arctic 2013 in Kobbefjord; and NASA for permission to use heath community. Physiol Plant 140:128–40. Modern-Era Retrospective Analysis for Research Bokhorst S, Bjerke JW, Street LE, Callaghan TV, Phoenix GK. and Applications (MERRA) reanalysis datasets. We 2011. Impacts of multiple extreme winter warming events on offer our special thanks to K. Elder for thorough sub-Arctic heathland: phenology, reproduction, growth, and guidance and recommendations on snow-sampling CO2 flux responses. Glob Change Biol 17:2817–30. methods and strategies used during the field cam- Bokhorst S, Bjerke JW, Tommervik H, Preece C, Phoenix GK. paign. We also thank two anonymous reviewers 2012. Ecosystem response to climatic change: the importance of the cold season. Ambio 41:246–55. whose comments greatly improved this manuscript. Brooks PD, Grogan P, Templer PH, Groffman P, Oquist MG. We gratefully acknowledge the logistic support of 2011. Carbon and nitrogen cycling in snow-covered envi- Arctic Research Centre (ARC), Aarhus University. ronments. Geogr Compass 5:682–99. Support was also provided by the Canada Excellence Callaghan TV, Johansson M, Brown RD, Groisman PY, Labba N, Research Chair (CERC). This study was funded by Radionov V, Bradley RS, Blangy S, Bulygina ON, Christensen the Environmental Protection Agency and the TR, Colman JE, Essery RLH, Forbes BC, Forchhammer MC, Danish Energy Agency, and it is a contribution to Golubev VN, Honrath RE, Juday GP, Meshcherskaya AV, Phoenix GK, Pomeroy J, Rautio A, Robinson DA, Schmidt the Arctic Science Partnership (ASP) asp-net.org. NM, Serreze MC, Shevchenko VP, Shiklomanov AI, Shmakin AB, Skold P, Sturm M, Woo MK, Wood EF. 2011. Multiple effects of changes in arctic snow cover. Ambio 40:32–45. REFERENCES Cappelen J. 2011. DMI monthly climate data collection 1768- Aastrup P, Raundrup K, Feilberg J, Krogh PK, Schmidt NM, 2010, Denmark, The Faroe Islands and Greenland technical Nabe-Nielsen J. 2014. Effects of large herbivores on biodi- report. Copenhagen: Danish Meteorological Institute. p 54. versity of vegetation and soil microarthropods in low Arctic Cappelen J, Vinther BM. 2014. SW Greenland temperature data Greenland—Akia, West Greenland and Southern Greenland. 1784-2013. Technical report. Copenhagen: Danish Meteoro- Scientific report from DCE—Danish Centre for Environment logical Institute. and Energy. Roskilde: Aarhus University, DCE—Danish Cappelen J, Vraae Jørgensen B, Vaarby Laursen E, Sligting Centre for Environment and Energy, p. 40. Stannius L, Sjølin Thomsen R. 2001. The observed climate of AMAP. 2012. Arctic Climate Issues 2011: Changes in Arctic Greenland, 1958-99—with climatological standard normals, Snow, Water, Ice and Permafrost. SWIPA 2011 Overview re- 1961-90. Technical report. Copenhagen: Danish Meteoro- port. Oslo: Arctic Monitoring and Assessment Programme logical Institute. p 151. (AMAP), p. xi + 97. Cooper EJ, Dullinger S, Semenchuk P. 2011. Late snowmelt Auer AH. 1974. The rain versus snow threshold temperature. delays plant development and results in lower reproductive Weatherwise 27:67. success in the High Arctic. Plant Sci 180:157–67. Austrheim G, Asheim L-J, Bjarnason G, Feilberg J, Fosaa AM, Dadic R, Mott R, Lehning M, Carenzo M, Anderson B, Mack- Holand Ø, Høegh K, Jo´ nsdo´ ttir IS, Magnu´ sson B, Mortensen intosh A. 2013. Sensitivity of turbulent fluxes to wind speed LE, Mysterud A, Olsen E, Skonhoft A, Steinheim G, over snow surfaces in different climatic settings. Adv Water Tho´ rhallsdo´ ttir AG. 2008. Sheep grazing in the North-Atlantic Resour 55:178–89. Episodic Snowmelt Events in Arctic Ecosystems 855

DMI. 2014. Danish Meteorological Institute Weather archive: snow affect soil carbon distribution in the continental pan- climate archive of climate normal period 1961–1990. Copen- Arctic area. J Geophys Res 117:G02020. hagen: DMI. Groisman PY, Karl TR, Knight RW, Stenchikov GL. 1994. Elberling B, Tamstorf MP, Michelsen A, Arndal MF, Sigsgaard C, Changes of snow cover, temperature, and radiative heat-bal- Illeris L, Bay C, Hansen BU, Christensen TR, Hansen ES, Jakobsen ance over the Northern-Hemisphere. J Clim 7:1633–56. BH, Beyens L. 2008. Soil and plant community-characteristics Hanna E, Mernild SH, Cappelen J, Steffen K. 2012. Recent and dynamics at Zackenberg. In: Meltofte H, Christensen TR, warming in Greenland in a long-term instrumental (1881– Elberling B, Forchhammer MC, Rasch M, Eds. Advances in 2012) climatic context: I. Evaluation of surface air tem- ecological research. London: Academic Press. p 223–48. perature records. Environ Res Lett 7:045404. Ellebjerg SM, Tamstorf MP, Illeris L, Michelsen A, Hansen BU. Hansen BB, Aanes R, Herfindal I, Kohler J, Saether BE. 2011. 2008. Inter-annual variability and controls of plant phenology Climate, icing, and wild arctic reindeer: past relationships and and productivity at Zackenberg. In: Meltofte H, Christensen future prospects. Ecology 92:1917–23. TR, Elberling B, Forchhammer MC, Rasch M, Eds. Advances Hansen BB, Grotan V, Aanes R, Saether BE, Stien A, Fuglei E, in ecological research. London: Academic Press. p 249–73. Ims RA, Yoccoz NG, Pedersen AO. 2013. Climate events Elmendorf SC, Henry GHR, Hollister RD, Bjo¨ rk RG, Bjorkman synchronize the dynamics of a resident vertebrate community AD, Callaghan TV, Collier LS, Cooper EJ, Cornelissen JHC, in the high Arctic. Science 339:313–15. ´ ´ Day TA, Fosaa AM, Gould WA, Gretarsdottir J, Harte J, Her- Hansen BB, Isaksen K, Benestad RE, Kohler J, Pedersen AO, Loe ´ ´ manutz L, Hik DS, Hofgaard A, Jarrad F, Jonsdottir IS, Keuper LE, Coulson SJ, Larsen JO, Varpe O. 2014. Warmer and wetter F, Klanderud K, Klein JA, Koh S, Kudo G, Lang SI, Loewen V, winters: characteristics and implications of an extreme May JL, Mercado J, Michelsen A, Molau U, Myers-Smith IH, weather event in the High Arctic. Environ Res Lett 9:114021. Oberbauer SF, Pieper S, Post E, Rixen C, Robinson CH, Sch- Hartmann DL, Klein Tank AMG, Rusticucci M, Alexander LV, midt NM, Shaver GR, Stenstro¨ m A, Tolvanen A, Totland O, Troxler T, Wahren CH, Webber PJ, Welker JM, Wookey PA. Bro¨ nnimann S, Charabi Y, Dentener FJ, Dlugokencky EJ, 2012a. Global assessment of experimental climate warming on Easterling DR, Kaplan A, Soden BJ, Thorne PW, Wild M, Zhai PM. 2013. Observations: atmosphere and surface. In: Stocker tundra vegetation: heterogeneity over space and time. Ecol Lett 15:164–75. TF, Qin D, Plattner G-K, Tignor M, Allen SK, Boschung J, Nauels A, Xia Y, Bex V, Midgley PM, Eds. Climate change Elmendorf SC, Henry GHR, Hollister RD, Bjork RG, Boulanger- ¨ 2013: The Physical Science Basis. Contribution of Working Lapointe N, Cooper EJ, Cornelissen JHC, Day TA, Dorrepaal E, Group I to the Fifth Assessment Report of the Intergovern- Elumeeva TG, Gill M, Gould WA, Harte J, Hik DS, Hofgaard A, mental Panel on Climate Change. Cambridge: Cambridge ´ ´ Johnson DR, Johnstone JF, Jonsdottir IS, Jorgenson JC, University Press. Klanderud K, Klein JA, Koh S, Kudo G, Lara M, Levesque E, Magnusson B, May JL, Mercado-Diaz JA, Michelsen A, Molau Hoffmeyer N. 1877. The Greenland Foehn. Nature 16:296–7. U, Myers-Smith IH, Oberbauer SF, Onipchenko VG, Rixen C, Høye TT, Post E, Meltofte H, Schmidt NM, Forchhammer MC. Schmidt NM, Shaver GR, Spasojevic MJ, Porhallsdottir PE, 2007. Rapid advancement of spring in the high Arctic. Curr Tolvanen A, Troxler T, Tweedie CE, Villareal S, Wahren CH, Biol 17:R449–51. Walker X, Webber PJ, Welker JM, Wipf S. 2012b. Plot-scale Iler AM, Høye TT, Inouye DW, Schmidt NM. 2013. Nonlinear evidence of tundra vegetation change and links to recent flowering responses to climate: are species approaching their summer warming. Nat Clim Change 2:453–7. limits of phenological change? Philos Trans R Soc B Forchhammer M, Boertmann D. 1993. The Muskoxen Ovibos 368:20120489. moschatus in north and northeast Greenland: population Inouye DW. 2000. The ecological and evolutionary significance trends and the influence of abiotic parameters on population of frost in the context of climate change. Ecol Lett 3:457–63. dynamics. Ecography 16:299–308. Jensen LM. 2012. Nuuk Ecological Research Operations, 5th Francis JA, Vavrus SJ. 2012. Evidence linking Arctic amplifica- annual report 2011. Roskilde: Aarhus University, tion to extreme weather in mid-latitudes. Geophys Res Lett DCE—Danish Centre for Environment and Energy. p 84. 39:L06801. Jensen LM, Rasch M. 2008. Nuuk Ecological Research Op- Fristrup B. 1953. De grønlandske foehnvinde. Tidsskriftet erations, 1st annual report 2007. Roskilde: Danish Polar Grønland 3:4. Centre, Danish Agency for Science, Technology and Inno- Fuller MC, Geldsetzer T, Yackel JJ. 2009. Surface-based polari- vation, Ministry of Science, Technology and Innovation. p metric C-band microwave scatterometer measurements of 112. snow during a chinook event. IEEE Trans Geosci Remote Sens Jensen LM, Rasch M. 2009. Nuuk Ecological Research Opera- 47:1766–76. tions, 2nd annual report 2008. Roskilde: National Environ- Goodison BE, Louie PYT, Yang D. 1998. WMO solid precipitation mental Research Institute, Aarhus University. p 80. intercomparison. Final report WMO/TD-872 Instruments and Jensen LM, Rasch M. 2010. Nuuk Ecological Research Opera- observing methods report. Geneva: World Meteorological tions, 3rd annual report 2009. Roskilde: National Environ- Organization. p 212. mental Research Institute, Aarhus University. p 80. Goodrich LE. 1982. The influence of snow cover on the ground Jensen LM, Rasch M. 2011. Nuuk Ecological Research Opera- thermal regime. Can Geotech J 19:421–32. tions, 4th annual report 2010. Roskilde: Aarhus University, Gorter W, van Angelen JH, Lenaerts JTM, van den Broeke MR. DCE—Danish Centre for Environment and Energy. p 84. 2014. Present and future near-surface wind climate of Jensen LM, Rasch M. 2013. Nuuk Ecological Research Opera- Greenland from high resolution regional climate modelling. tions, 6th annual report 2012. Roskilde: Aarhus University, Clim Dyn 42:1595–611. DCE—Danish Centre for Environment and Energy. p 92. Gouttevin I, Menegoz M, Domine F, Krinner G, Koven C, Ciais Johansson M, Callaghan TV, Bosio J, Akerman HJ, Jackowicz- P, Tarnocai C, Boike J. 2012. How the insulating properties of Korczynski M, Christensen TR. 2013. Rapid responses of 856 S. H. Pedersen and others

permafrost and vegetation to experimentally increased snow gaard S, Schindler DE, Stirling I, Tamstorf MP, Tyler NJC, van cover in sub-arctic Sweden. Environ Res Lett 8:035025. der Wal R, Welker J, Wookey PA, Schmidt NM, Aastrup P. Jones HG. 1999. The ecology of snow-covered systems: a brief 2009. Ecological dynamics across the Arctic associated with overview of nutrient cycling and life in the cold. Hydrol recent climate change. Science 325:1355–8. Process 13:2135–47. Reichle RH, Koster RD, De Lannoy GJM, Forman BA, Liu Q, Liston GE. 1995. Local advection of momentum, heat, and Mahanama SPP, Toure´ A. 2011. Assessment and enhance- moisture during the melt of patchy snow covers. J Appl Me- ment of MERRA land surface hydrology estimates. J Clim teorol 34:1705–15. 24:6322–38. Liston GE. 1999. Interrelationships among snow distribution, Rennert KJ, Roe G, Putkonen J, Bitz CM. 2009. Soil thermal and snowmelt, and snow cover depletion: implications for atmo- ecological impacts of rain on snow events in the circumpolar spheric, hydrologic, and ecologic modeling. J Appl Meteorol Arctic. J Clim 22:2302–15. 38:1474–87. Rienecker MM, Suarez MJ, Gelaro R, Todling R, Bacmeister J, Liu Liston GE, Elder K. 2006a. A meteorological distribution system E, Bosilovich MG, Schubert SD, Takacs L, Kim GK, Bloom S, for high-resolution terrestrial modeling (MicroMet). J Hy- Chen JY, Collins D, Conaty A, Da Silva A, Gu W, Joiner J, Koster drometeorol 7:217–34. RD, Lucchesi R, Molod A, Owens T, Pawson S, Pegion P, Redder Liston GE, Elder K. 2006b. A distributed snow-evolution mod- CR, Reichle R, Robertson FR, Ruddick AG, Sienkiewicz M, eling system (SnowModel). J Hydrometeorol 7:1259–76. Woollen J. 2011. MERRA: NASA’s modern-era retrospective analysis for research and applications. J Clim 24:3624–48. Liston GE, Hall DK. 1995. An energy-balance model of lake-ice evolution. J Glaciol 41:373–82. Rumpf SB, Semenchuk PR, Dullinger S, Cooper EJ. 2014. Idiosyncratic responses of high arctic plants to changing snow Liston GE, Hiemstra CA. 2008. A simple data assimilation system regimes. PLoS ONE 9:E86281. for complex snow distributions (SnowAssim). J Hy- drometeorol 9:989–1004. Schimel JP, Bilbrough C, Welker JA. 2004. Increased snow depth affects microbial activity and nitrogen mineralization in Liston GE, Mernild SH. 2012. Greenland freshwater runoff. Part two Arctic tundra communities. Soil Biol Biochem 36:217–27. I: A runoff routing model for glaciated and Nonglaciated landscapes (HydroFlow). J Clim 25:5997–6014. Schmidt NM, Kristensen DK, Michelsen A, Bay C. 2012. High Arctic plant community responses to a decade of ambient Liston GE, Sturm M. 1998. A snow-transport model for complex warming. Biodiversity 13:191–9. terrain. J Glaciol 44:498–516. Semenchuk PR, Elberling B, Cooper EJ. 2013. Snow cover and Liston GE, Sturm M. 2002. Winter precipitation patterns in arctic extreme winter warming events control flower abundance of Alaska determined from a blowing-snow model and snow- some, but not all species in high arctic Svalbard. Ecol Evol depth observations. J Hydrometeorol 3:646–59. 3:2586–99. Liston GE, McFadden JP, Sturm M, Pielke RA. 2002. Modelled Semmens KA, Ramage J, Bartsch A, Liston GE. 2013. Early changes in arctic tundra snow, energy and moisture fluxes snowmelt events: detection, distribution, and significance in a due to increased shrubs. Glob Change Biol 8:17–32. major sub-arctic watershed. Environ Res Lett 8:014020. Liston GE, Haehnel RB, Sturm M, Hiemstra CA, Berezovskaya S, Serreze MC, Box JE, Barry RG, Walsh JE. 1993. Characteristics Tabler RD. 2007. Instruments and methods simulating com- of Arctic synoptic activity, 1952–1989. Meteorol Atmos Phys plex snow distributions in windy environments using Snow- 51:147–64. Tran-3D. J Glaciol 53:241–56. Steffen K, Box J. 2001. Surface climatology of the Greenland Ice Lund M, Falk JM, Friborg T, Mbufong HN, Sigsgaard C, Soegaard Sheet: Greenland climate network 1995–1999. J Geophys Res H, Tamstorf MP. 2012. Trends in CO exchange in a high 2 106:33951–64. Arctic tundra heath, 2000-2010. J Geophys Res 117:G02001. Sturm M, Johnson JB. 1992. Thermal-conductivity measure- Marks D, Dozier J. 1992. Climate and energy exchange at the ments of depth hoar. J Geophys Res Solid Earth 97:2129–39. snow surface in the alpine region of the Sierra-Nevada. 2. Sturm M, Holmgren J, Liston GE. 1995. A seasonal snow cover Snow cover energy-balance. Water Resour Res 28:3043–54. classification-system for local to global applications. J Clim Meltofte H, Høye TT, Schmidt NM, Forchhammer MC. 2007. Dif- 8:1261–83. ferences in food abundance cause inter-annual variation in the breeding phenology of high Arctic waders. Polar Biol 30:601–6. Sturm M, Holmgren J, Konig M, Morris K. 1997. The thermal conductivity of seasonal snow. J Glaciol 43:26–41. Mernild SH, Hanna E, Yde JC, Cappelen J, Malmros JK. 2014. Coastal Greenland air temperature extremes and trends 1890– Sturm M, Taras B, Liston GE, Derksen C, Jonas T, Lea J. 2010. 2010: annual and monthly analysis. Int J Climatol 34:1472–87. Estimating snow water equivalent using snow depth data and climate classes. J Hydrometeorol 11:1380–94. Nkemdirim LC. 1997. On the frequency and sequencing of Westergaard-Nielsen A, Lund M, Hansen BU, Tamstorf MP. chinook events. Phys Geogr 18:101–13. 2013. Camera derived vegetation greenness index as proxy for Nobrega S, Grogan P. 2007. Deeper snow enhances winter res- gross primary production in a low Arctic wetland area. ISPRS J piration from both plant-associated and bulk soil carbon pools Photogram Remote Sens 86:89–99. in birch hummock tundra. Ecosystems 10:419–31. Wilson RR, Bartsch A, Joly K, Reynolds JH, Orlando A, Loya WM. Pattison RR, Welker JM. 2014. Differential ecophysiological re- 2013. Frequency, timing, extent, and size of winter thaw-response of deciduous shrubs and a graminoid to long-term freeze events in Alaska 2001-2008 detected by remotely sensed experimental snow reductions and additions in moist acidic microwave backscatter data. Polar Biol 36:419–26. tundra, Northern Alaska. Oecologia 174:339–50. Zhang T. 2005. Influence of the seasonal snow cover on the Post E, Forchhammer MC, Bret-Harte MS, Callaghan TV, ground thermal regime: an overview. Rev Geophys Christensen TR, Elberling B, Fox AD, Gilg O, Hik DS, Høye TT, 43:RG4002. Ims RA, Jeppesen E, Klein DR, Madsen J, McGuire AD, Rys- Paper IV

Snow-free date drives maximum NDVI timing: A mesocosm study across compressed Arctic environmental gradients.

Pedersen, S. H., Liston, G. E., Tamstorf, M. P., Abermann, J., Lund, M., and Schmidt, N. M. [Manuscript]

Zackenberg Valley,NE-Greenland in late winter (S. H. Pedersen) and summer (K. Skov) (2014)

Paper IV

Snow-free date drives maximum NDVI timing: A mesocosm study across compressed Arctic environmental gradients

Stine Højlund Pedersen1, Glen E. Liston2, Mikkel P. Tamstorf1, Jakob Abermann3, Magnus Lund1, Niels Martin Schmidt1

1 Arctic Research Centre, Department of Bioscience, Aarhus University, Frederiksborgvej 399, DK-4000 Roskilde, Denmark. 2 Cooperative Institute for Research in the Atmosphere (CIRA), Colorado State University, Fort Collins, Colorado 80523, USA. 3Asiaq, Greenland Survey, Qatserisut 8, GL-3900 Nuuk, Greenland.

1 Abstract A well-known approach to understanding ecosystem variations and changes in space and time is to study the associated processes across natural or designed environmental gradients. The ice-free regions along Greenland’s east coast are located between the Greenland Ice Sheet (GrIS) and the periodically sea-ice- covered North Atlantic. This creates inland-coast continentality differences that exert strong control over ecosystems in these areas. While regional environmental gradients have been qualitatively described in numerous Greenland climate-system studies, defining the details of these gradients has been elusive. Here we quantified the inland-coast gradients of air temperature, winter precipitation (using pre-melt snow-water equivalent (SWE) depth), and timing of snowmelt (using snow-free day of year (DOY)) in a topographically heterogeneous high-Arctic region in northeast Greenland. From the GrIS margin towards the coast, we found gradients in mean annual air temperature of -0.05 °C km-1, mean pre-melt SWE depth of 0.003 m km-1, and mean snow-free DOY of 0.4 days km-1. Across the 97.5 km-wide domain, this meant that, at the coast, the annual mean air temperature was 4.5 °C lower, the mean pre-melt SWE was 0.3 m greater, and the mean snow-free date was 37 days later than near the GrIS margin. This region thus comprises a continentality gradient which is eight times stronger than the mean annual air temperature gradient found from south to north along the east coast of Greenland. To examine whether these strong gradients were reflected in the vegetation growth phenology, we quantified gradients in the annual maximum level of Normalized Difference Vegetation Index (MaxNDVI) and the timing of MaxNDVI over the same domain. We found no significant gradient of MaxNDVI across the domain, but MaxNDVI was strongly correlated with pre-melt SWE depth. However, the inland-coast gradient of average timing of MaxNDVI was 0.1 days km-1 and varied up to 7 days in total from west to east in the domain. The pre-melt SWE and timing of snowmelt gradient was identified in both snow-rich and snow-poor years. The MaxNDVI level was relatively stable over the years, whereas the timing of MaxNDVI varied from year to year depending on the pre-melt SWE depth and snow-free DOY. This region thus constitutes a geographically compressed system with unique opportunities for studying biological responses to environmental drivers within relatively short distances.

Keywords Continentality, environmental gradient, winter precipitation, snow, air temperature, Normalized Difference Vegetation Index (NDVI), mesocosm.

Paper IV

2 Introduction Frequently, research questions are formulated with the aim that their answers will have an impact across a larger spatial and/or temporal extent than that of the original research domain. This is particularly true for research conducted in the Arctic, where ground-observations are often relatively sparsely distributed in space and/or time. In this case, up-scaling efforts are key to understanding the extent, magnitude, and importance of the observed processes and effects (Azaele et al. 2015). However, prior to any up-scaling, the appropriate ecosystem inter-relations and mechanisms must be identified, and the spatial and temporal extent that they are valid for must be defined (Jarvis 1995, Williams et al. 2008, Blok et al. 2011).

To increase the extension of an experiment to other places or time periods, observations may be collected across environmental gradients. Here an environmental gradient constitutes a change in, e.g., temperature or winter precipitation during a defined time period, reflecting expected future conditions (Henry and Molau 1997), or across space, e.g., several degrees of latitude (e.g., Kottek et al. 2006, Legagneux et al. 2014), or along an increase in elevation (Körner 2007). For example, biological mechanisms or responses, which are observed on plot-scale across a designed environmental gradient, can be representative and have predictive value for a future time period and/or an extended area (e.g., spanning multiple latitude degrees) within the Arctic region. For instance, since winter precipitation is a key driver of Arctic vegetation phenology (e.g., Cooper 2014), the investigations of vegetation phenology and biomass along precipitation gradients are conducted at plot-scale and across experimental winter precipitation gradients, e.g., using snow fences (e.g., Leffler and Welker 2013, Semenchuk et al. 2013). Additionally, the response of vegetation growth to future climate is observed across temperature gradients of accelerated warming, within standardized plot warming experiments (Elmendorf et al. 2012a). Furthermore, similar warming observations are collected in multiple sites located across the Arctic, where the differing latitude and climate zones between sites act as a surrogate for a temperature gradient (space-for-time substitution) (Elmendorf et al. 2015) or a gradient in snow conditions (Bienau et al. 2014). Last, observations of vegetation responses are also made across elevation levels, e.g., on a mountain, representing temperature and precipitation gradients. However, the use of this type of gradients for up-scaling purposes is challenging because of the lacking comparability between a latitudinal temperature gradient and an elevational temperature gradient. For instance, the way plants disperse across the two types of gradients may be different (Halbritter et al. 2013) and the phenological response to environmental conditions may differ between alpine and Arctic vegetation (Billings and Mooney 1968). Moreover, the comparability between altitudinal and latitudinal gradients is challenged by the fact that the mountain environment, e.g., temperature and precipitation lapse rates, may differ from global altitudinal gradients, i.e., decreasing air temperature with elevation, hence prior quantification of the physical environment is essential (Körner 2007).

Knowledge gained from plot-based studies have shown that higher air temperatures result in enhanced vegetation growth both across plots exposed to naturally increasing summer temperature and in experimentally warmed plots across pan-Arctic sites (Elmendorf et al. 2012b, Myers-Smith et al. 2015). However, the vegetation response to experimental warming and ambient temperature differs. Similarly, the response to naturally varying climate using repeat sampling in the same location, differ from the vegetation responses observed along spatial environmental gradients, such as across latitudes (Elmendorf et al. 2015). Furthermore, the vegetation response to experimental warming may diminish over time (Kremers et al. 2015). Other driving factors of vegetation phenology in the Arctic are snowmelt timing and post-snowmelt temperature sums (Pudas et al. 2008, Hollesen et al. 2015). A delay in timing of snowmelt can even counteract the expected phenological response of increased temperatures (Bjorkman et al. 2015). Again, this emphasizes the importance of a quantification of the physical environment prior to experimental setup

Paper IV

design, and to be aware of declining responses and artifacts/feedback mechanisms across an experimental gradients. Furthermore, these uncertainties encourage one to conduct studies along more naturally varying environmental gradients if possible. This has been done using spatially distributed data, e.g., remote sensing imagery. The vegetation response to temperature and winter precipitation has been studied spatially across domains spanning multiple latitude degrees that represent temperature, irradiation, or/and moisture gradients (Zeng et al. 2011, Zeng and Jia 2013). The seasonal peak plant greenness as inferred from Normalized Difference Vegetation Index (NDVI) (Riedel et al. 2005, Raynolds et al. 2012), shows a positive relationship with growing season temperatures across the Arctic. However, the strength of this relationship has declined between 1982 and 2011 (Piao et al. 2014). Furthermore, there are regional differences in the direction of maximum vegetation greenness trends in the past decades (Bhatt et al. 2013), and in recent years the Arctic tundra has experienced a general ‘browning’ (Epstein et al. 2015). The spatially distributed vegetation phenology and timing of peak vegetation greenness show a significant positive correlation with the timing of snowmelt in parts of the Arctic (Zeng and Jia 2013), while the direction of this relationship varies across latitudes (Grippa et al. 2005).

These different types of gradient studies, both plot and remote sensing-based, are challenged in different ways. For example, when examining multiple sites/plots, a high level of standardization (sampling methods) and coordination is required in order to obtain a satisfactory inter-site comparability of the observations (Walker et al. 2016). Furthermore, in gradient studies using remote sensing data, or other spatially continuous data, the validation of the satellite-acquired measurements is challenging since it requires expensive ground-truthing collected during the overflight time of the satellite. However, despite observation data collection/validation being both challenging (timing of observation, coordination) and expensive (logistics) in the Arctic, we, as scientists, strive to understand the vegetation-environment relations across large scales. Therefore, in this study we set out to find local-scale and naturally varying environmental gradients that represented condensed large-scale gradients, hence to understand snow-vegetation relations, which were valid over an extensive area, i.e., larger than the scale it was monitored on.

First, we mapped and quantified the inland-coast continentality gradients of air temperature and snow amounts within an ice-free NE-Greenland region. These two key drivers for Arctic vegetation growth and phenology were modeled using temporally and spatially distributed modeling tools (MicroMet/SnowModel). In-situ observations of snow and meteorological variables enabled validation and verification of the modeled variables. Secondly, we investigated whether the across-region gradients in the SnowModel outputs of pre- melt snow-water equivalent (SWE) depth and derived snow-free date were reflected in a spatially and temporally distributed and independent data set of vegetation greenness (i.e. NDVI) covering the entire domain. Furthermore, we quantified snow-vegetation relations across the region. Last, we evaluated the potential of the spatio-temporal highly-resolved regional gradients being a scientific mesocosm by comparing the regional environmental gradients with large-scale gradients, as observed along the Greenland east coast and in other Arctic regions.

3 Materials and methods

3.1 Study area The study region is located in the high-Arctic NE-Greenland, centered at 74°27' N, 20°34' W, and extends from the Greenland Ice Sheet (GrIS) margin to the Greenland east coast near the North Atlantic Ocean, and covers 97.5 km west-east by 75.0 km south-north (3°15’ longitude by 0°40’ latitude). Across the region, the

Paper IV

topographic relief varies between sea level and 1570 m a.s.l. (Figure 1). The mean annual air temperature (MAAT) equaled -9.0 °C in the center area near Zackenberg through the period 1996-2015. The region is located in the high-Arctic zone (Kottek et al. 2006) with a tundra climate (Bliss and Matveyeva 1992) and more specifically within the bioclimate subzone C, which is characterized by a 5%-50% cover of vascular plants and open patchy vegetation (Walker et al. 2005). Long-term observations, spanning the components of the Arctic ecosystem including ground-based measurements of NDVI and snow, were available for this region at Zackenberg Research Station (www.zackenberg.dk), which also served as logistical platform during the regional snow field work conducted in April 2014.

Figure 1 Topographic map of study area in NE-Greenland with elevation in meters above sea level (m a.s.l.), coordinates are in UTM kilometers, the north direction is along the y-axis. Gray points mark the center of 12 ECMWF Interim Re-Analysis (ERA-Interim) grid cells each 0.75 degrees longitude by 0.75 degrees latitude, red open circles marks location of automated weather stations (AWS).

3.2 Spatially distributed snow characteristics and air temperature In addition to air temperature, we focused our study on two ecologically relevant snow metrics, the pre-melt SWE depth, i.e., the maximum amount of water available in the snowpack prior to snowmelt, and the timing of snowmelt referred to as snow-free DOY, i.e., the DOY when SWE depth equaled zero. Both snow metrics are proved important for vegetation growth and phenology in Greenland (Groendahl et al. 2006, Lund et al. 2012, Westergaard-Nielsen et al. 2013, Hollesen et al. 2015). To obtain regional distributions and to capture the landscape-scale (5-10 km) heterogeneity of daily mean air temperature, pre-melt SWE depth, and snow- free DOY, we used the high-resolution meteorological model, MicroMet (Liston and Elder 2006b) coupled

Paper IV

with the snow modeling tools, SnowModel (Liston and Elder 2006a). Nine automated weather stations (AWS) (Figure 1, Table 1), distributed within the region, provided inputs of air temperature, relative humidity, wind speed, and wind direction. Due to large uncertainties attached to gauge measurements of winter precipitation (Goodison et al. 1998), we used ECMWF Interim Re-Analysis (ERA-Interim) reanalysis meteorological inputs including precipitation rates from 12 grid cells (Dee et al. 2011) distributed across the domain (gray points in Figure 1). The model was run with a 300-m spatial resolution and covered the period from 1 August 1979 through 31 July 2015. Since synoptic weather systems producing precipitation occur on a daily time scale or shorter across Greenland (Schuenemann et al. 2009), we used 3-hourly time increments to resolve the diurnal cycle. Furthermore, we analyzed the output data in a daily temporal resolution in order to resolve sub-weekly/daily variations and the timing of events that may change from one day to the next. To adjust the reanalysis precipitation rates, the SnowModel setup included the SnowAssim submodel (Liston and Hiemstra 2008), for assimilating observed SWE depth. The SnowModel configurations for this simulation also included incorporating distributed large-scale curvature weights that affected the snow distribution by reducing the threshold surface shear velocity in regions of large-scale positive curvature. The net effect of this was to increase the wind-related snow erosion on high mountain tops and, in this study area, to force the snow drifts to form lower on the mountain slopes.

Table 1 Automated weather stations providing meteorological input for MicroMet/SnowModel simulations in the period 1979-2015. Data were provided by GeoBasis, ClimateBasis, and GlacioBasis Zackenberg, Greenland Ecosystem Monitoring, the Climate Research Section, Central Institute for Meteorology and Geodynamics (ZAMG) in Austria, and the Danish Meteorological Institute (DMI).

UTM UTM Elevation Easting Northing Name Time series [m a.s.l.] [km] [km] Location Data provider C1 1995-2015 38 513.382 8264.743 Zackenberg Valley ClimateBasis

M2 2003-2015 17 513.058 8264.019 Zackenberg Valley GeoBasis

M3 2003-2014 410 516.126 8268.250 Zackenberg Valley GeoBasis

M6 2003-2010 1282 507.453 8273.009 Zackenberg Valley GeoBasis

ZAC_M 2008-2014 660 488.877 8281.802 A.P. Olsen Glacier GlacioBasis

ZAC_S 2008-2014 880 486.106 8283.867 A.P. Olsen Glacier GlacioBasis

ZAC_T 2009-2014 1475 480.742 8284.476 A.P. Olsen Glacier GlacioBasis Freya Glacier, FRE 2011-2014 860 505.540 8254.393 ZAMG Clavering Island 04330 1979-2015 44 523.660 8245.664 Daneborg DMI

3.3 Snow observations Snowpack properties including SWE depth observations were collected across the domain in 55 sites during a snow field campaign in April 2014 (Table S1, Appendix I). SWE depths from five randomly-selected sites (a) were included in the SWE assimilation. The remaining 50 SWE depth observations (v) were used in the validation of the spatially distributed SWE depth across the region for the winter 2013/2014. Additionally,

Paper IV

we used observed pre-melt SWE depth in the SWE assimilation, which were collected in snowpits in the years 2004-2015 in the Zackenberg Valley by GeoBasis Zackenberg (Pedersen et al. 2016). Snow depth observations from a snow-depth sensor installed in the Zackenberg Valley (C1, Table 1) were used to estimate the snow-free DOY for validation of the modeled snow-free DOY.

3.4 Quantifying and adjusting for the SWE depth inland-coast gradient We quantified the observed pre-melt SWE depth gradient across the region based on observations made during April 2014. The inland-coast gradients were quantified as the slope of the linear regression between observed/modeled SWE depth and the west-east distance across the region. The observed SWE depth gradient was 0.00590 m km-1 distance from inland in the west towards the coast in the east. The unadjusted modeled SWE depth gradient was 0.0011 m km-1 distance (Figure 2). Five of the SWE depth observations were included in the SWE assimilation to generate a smoothed distribution of precipitation correction factors across the domain using the SnowAssim submodel. The correction factors varied between 0.83 and 1.8 from inland towards the coast. The input precipitation rates were adjusted using this correction-factor surface prior to being included in the SnowModel simulation. Initially, this correction was applied to a SnowModel simulation for the year 2013/2014. The resulting modeled SWE depth gradient, with adjusted precipitation rates, equaled 0.00515 m km-1, whereas the observed SWE depth gradient equaled 0.00590 m km-1 distance. Both gradients were significantly different from zero (observed and modeled p-values < 0.001, R2 = 0.43 and 0.64, respectively) (Figure 2) and there was no significant difference between the gradients and no offset in average SWE depth level. This was found from one single regression between all SWE depth data and distance from the GrIS margin with the data type (observed/modeled) as interaction term, which showed that the interaction was not significant, p-value = 0.519. Hence, the precipitation-correction surface was applied to all other years in the SnowModel simulation.

Figure 2 Inland-coast gradients in observed snow-water-equivalent (SWE) depth (Black); modeled, unadjusted SWE depth (Gray); and modeled, adjusted SWE depth (Red) in April 2014.

Paper IV

3.5 Quantifying inland-coast temperature, snow, and NDVI gradients The aim was to quantify the inland-coast gradients in the modeled air temperature, pre-melt SWE depth, and snow-free DOY for the low-altitudinal and vegetated areas across the region. Hence, to minimize the orographic effect on the inland-coast gradient estimates, we extracted each variable (MAAT, pre-melt SWE depth, snow-free DOY, MaxNDVI, and timing of MaxNDVI) in all land-covered grid cells at elevations below 100 m a.s.l. Based on these grid cells, we computed an average for every 300-m longitudinal increment along the west-east dimension of the domain. Eventually, we ended up with 325 averages along the distance from the inland to the coast for each variable.

3.6 Spatially distributed vegetation greenness To assess the impact of the quantified environmental gradients on vegetation growth and phenology across the region, we used an independent data set of NDVI, which is an integrative measure of the vegetation growth and above-ground phytomass (Tucker 1979, Myneni et al. 1997). NDVI was calculated from combining two daily Moderate Resolution Imaging Spectroradiometer (MODIS) products, MOD09GA and MOD09GA. For detailed product and preprocessing descriptions, see Appendix II. The generated daily NDVI data set extending from 1 January 2000 through 31 December 2015 was used to estimate the seasonal maximum level of vegetation greenness (MaxNDVI) and to find the DOY, when the MaxNDVI occurred.

3.7 Statistical analysis To find the overall relations between the variables, pre-melt SWE depth and snow-free DOY, and MaxNDVI and timing of MaxNDVI, we used linear regression (with year as random factor) for all available data across the years 2000-2015. Linear regression analyses were also used to compare MaxNDVI and the timing of MaxNDVI for the three Zackenberg stations with MODIS-derived NDVI. We used the statistical software R (www.r-project.org) for these analyses. To determine the association between snow variables and MaxNDVI and the MaxNDVI timing, we quantified the relationship between spatially distributed pre-melt SWE depth and snow-free DOY, and MaxNDVI and timing of MaxNDVI by calculating the spatial distribution of pixel- wise Pearson’s product-moment correlation coefficient, r, and coefficient of determination, r2, based on annual averages for the period 2000-2015. Also, to map the consistency of the relation between the two variables snow-free DOY and timing of MaxNDVI, we estimated the pixel-wise standard error of the mean (i.e., the standard deviation of the average) of the duration of the green-up period, i.e., the difference in days between snow-free DOY and timing of MaxNDVI.

4 Results

4.1 Validation of snow modeling There was a satisfactory correspondence between modeled and observed SWE depth in April 2014, which was linearly correlated (n = 50, intercept = 0.047, slope = 1.00, R2 = 0.61, p-value < 0.001) (Figure 3a). The modeled SWE depth captured the observed spatial distribution of SWE depth within the heterogeneous landscape of the region, where SWE depth increased with distance from the GrIS margin (Figure 4). Furthermore, the snow-free DOY estimated from SnowModel SWE depth, and the snow-free DOY, derived from a snow depth sensor measurements from Zackenberg Valley (C1, Table 1), showed a significant correspondence (n = 17, intercept = -25.9, slope = 1.18, R2 = 0.72, p-value < 0.001) (Figure 3b).

Paper IV

4.2 Validation of NDVI method and estimates Using linear regression, we validated how far the estimated timing of 40% seasonally cumulated NDVI (our method, see Appendix II) was occurring from the timing of the actual maximum NDVI per season for each automatic weather station (Table 2). The regression analysis showed that the two metrics were significantly correlated (R2 = 0.57, p-value < 0.001, slope = 1.20, intercept = -38.91, Figure 5).

Table 2 Stations providing Normalized Differenced Vegetation Index (NDVI) sensor measurements for validation.

Elevation Station Time series Vegetation type Location Data provider [m a.s.l.] M4 2010-2015 17 Cassiope heath Zackenberg Valley GeoBasis, Zackenberg MM2 2004-2013 40 fen wetland Zackenberg Valley GeoBasis, Zackenberg M3 2012-2015 420 Dryas heath Zackenberg Valley GeoBasis, Zackenberg

Figure 3 Modeled SWE depth (m) at 15 April 2014 and observed SWE depth (m) collected in 50 sites across the region during the period 4-28 April 2014 (a). Snow-free day of year (DOY) derived from SnowModel SWE depth and snow-free DOY estimated from snow-depth sensor observations in Zackenberg Valley (C1) during the period 1998-2015 (b). Black lines are linear regression lines, the statistics are described in the text.

Paper IV

Figure 4 Modeled distributed SWE depth (m) on 15 April 2014 (map) and observed SWE depth (m) (points) collected in sites across the region during the period 4-28 April 2014. The colored points have similar color scale as the map.

Paper IV

Figure 5 Linear regression (black line) between timing of 40% seasonally cumulated NDVI (in our method referred to as timing of the annual maximum level of Normalized Differenced Vegetation Index (MaxNDVI)) and the timing of the actual maximum NDVI per season for each station. The dashed line is the unity line.

Figure 6 Comparison between observed MaxNDVI and Moderate Resolution Imaging Spectroradiometer (MODIS)-derived MaxNDVI, no significant linear fit (a), and linear regression between observed timing of MaxNDVI and MODIS-derived timing of MaxNDVI for all stations. The dashed line is the unity line. The solid black line is the linear fit; the statistics are for (a): intercept = 0.159, slope = 0.726, R2= 0.181, p-value = 0.069, n = 20, and (b): intercept = 11.196, slope = 0.923, R2= 0.688, p- value = < 0.001, n = 20.

The MaxNDVI levels derived from the three sensors and MODIS data were compared for the years, where data was available from the NDVI sensors in Zackenberg (intercept = 0.159, slope = 0.726, R2= 0.181, p- value = 0.069, Figure 6a). The MaxNDVI from the three sensor locations had similar slope (not significantly

Paper IV

different, p-value < 0.001), but had different levels of MaxNDVI, which was to a lesser degree captured in the MODIS data (Figure 6a). This may originate from a more diverse surface in terms of NDVI level within the MODIS grid cell covering 300 m by 300 m than within the 5 m by 5 m area below each NDVI sensor. Hence, the MODIS MaxNDVI was averaged over a larger area and therefore had a smaller range in NDVI values than the range in sensor values. The range in MODIS-derived MaxNDVI was 0.43-0.68, whereas the range in sensor-derived MaxNDVI was 0.39-0.76 (Figure 6a). The observed timing of MaxNDVI and MODIS-derived timing of MaxNDVI for all station locations was significantly correlated (intercept = 11.196, slope = 0.923, R2= 0.688, p-value = < 0.001, Figure 6b). The robust validation for both snow and NDVI variables allowed us to derive spatial representations of the heterogeneous patterns of average pre- melt SWE depth, snow-free DOY, MaxNDVI, and timing of MaxNDVI to be used for quantifying the inland-coast gradients.

Figure 7 Time series of daily Normalized Differenced Vegetation Index (NDVI) data calculated from MODIS (green circles), estimated MaxNDVI and its timing (black triangles), and modeled SWE depth (m) (blue line) in the period 1 August 1999 through 30 September 2015. All data are extracted from the grid cell with the location of station C1 in Zackenberg Valley (Figure 1, Table 1).

4.3 Temporal accordance between independent snow and NDVI data sets The accordance between the two distributed and independent datasets, SnowModel outputs and MODIS NDVI, was visible when looking at the data in a non-aggregated form in an extracted daily time series for one point in the domain in the vegetated Zackenberg Valley (Figure 7). The timing of the daily SnowModel SWE depth matched up with the daily NDVI data. For instance in spring months, the NDVI values started increasing (vegetation greening up) after the snow-free DOY and the modeled snowpack had melted away. Furthermore in the autumn, when the first modeled snow storms had occurred and a persistent snow cover was established, the NDVI values were again below 0.1. Hence, the annual cycle of snow and NDVI

Paper IV

dynamics matched up despite the different data sources and the applied tools, SnowModel and NDVI preprocessing. This emphasized the validity of the method for describing these dynamics, and showed that the correspondence between the snow and NDVI dynamics were visible and originated from high temporal resolution.

4.4 Gradients Across the distance from the GrIS margin in west towards the east coast, we found gradients in air temperature of -0.05 °C km-1, pre-melt SWE depth of 0.003 m km-1, snow-free DOY of 0.4 days km-1, and the timing of MaxNDVI of 0.1 days km-1 (Table 3 and Figure 8). We found no significant gradient of MaxNDVI across the domain (p > 0.05). These gradients equaled a total difference across the 97.5 km-wide domain of -4.5 °C in average air temperature and 0.3 m average pre-melt SWE depth. While the average snow-free DOY and timing of MaxNDVI varied up to 37 and 7 days, respectively, across the domain.

Table 3 Statistics linear fit summary for quantified gradients across longitudinal averages of air temperature (1979-2015), SWE depth (2000-2015), snow-free day of year (DOY) (2000-2015), timing of MaxNDVI (DOY) (2000-2015), and MaxNDVI (2000-2015). All grid cells below 100m elevation, where data was available across the time series, were included in the analyses.

Average Minimum Maximum Intercept Slope/gradient R2 p-value Air temperature -9.7 -12.6 -7.9 -7.5 °C -0.05 °C km-1 0.91 < 0.001 Pre-melt SWE depth 0.23 0.11 0.39 0.08 m 0.003 m km-1 0.88 < 0.001 Snow-free DOY 166 152 184 DOY 148 0.4 days km-1 0.93 < 0.001 MaxNDVI timing 216 204 231 DOY 212 0.1 days km-1 0.24 < 0.001 MaxNDVI 0.44 0.30 0.58 0.44 0.00 0.00 0.807

The spatial distribution of pre-melt SWE depth also showed a clear gradient across the region with less winter precipitation in the inland area near Payer’s Land and western Clavering Island compared to the more snow-rich Wollaston Forland near the east coast (Figure 9a, see place names in Figure 1). The snowmelt timing pattern followed the distribution of pre-melt SWE depth, hence in areas with less SWE depth we found earlier snow-free DOY than in areas with higher pre-melt SWE depth (Figure 9b). Even during winters with contrasting SWE depth conditions, e.g., the snow-poor winter 2008/2009 and the snow-rich winter 2011/2012, the gradients in pre-melt SWE depth and snow-free DOY were identified across the region (Figure 10 a, b).

4.5 Timing and level of MaxNDVI While we found a significant inland-coast gradient of the timing of MaxNDVI, a similar gradient was not identified for MaxNDVI across the region (Table 3 and Figure 11a), although the annual average of MaxNDVI was pixel-wise significantly correlated with annual average pre-melt SWE depth and snow-free DOY (p < 0.001), and we found similar positive correlations spatially distributed within most vegetated areas of the region (Figure 12 a, b). In the Zackenberg Valley, the level of MaxNDVI was also persistent from year to year, e.g., compared to the inter-annually varying SWE depth (Figure 7) and the persistent pattern of MaxNDVI was also identified in years with contrasting SWE depth conditions (Figure 10d). In contrast, the timing of MaxNDVI co-varied between years with minimum, average, and maximum levels of pre-melt SWE depth and the associated dynamics in snow-free DOY (Figure 10c).

Paper IV

Figure 8 Inland-coast gradients (longitudinal annual averages; ± standard error of the mean) in air temperature (red regression line), pre-melt SWE depth (blue regression line), snow-free DOY (light blue regression line), timing of MaxNDVI (DOY) (green regression line), and MaxNDVI in the NE-Greenland region. The x-axis matches the upper x-axis in Figure 1.

Paper IV

Figure 9 Average pre-melt SWE depth (m) (a) and average snow-free DOY (b) for the period 2000-2015 based on daily SnowModel outputs of SWE depth. The topographic contours are in 100-m elevation increments and map specifications are as in Figure 1.

Paper IV

Figure 10 Pre-melt SWE depth (m) (a), snow-free DOY (b), timing of MaxNDVI in (DOY) (c), and MaxNDVI (d) in snow- rich, average, and snow-poor winters (defined by pre-melt SWE depth) in 2009, 2003, and 2012, respectively. The topographic contours are in 100-m elevation increments and map specifications are as in Figure 1.

Paper IV

4.6 Timing of MaxNDVI dependent on snow-free DOY Generally, the timing of MaxNDVI advanced in winters with early snowmelt, e.g., 2008/2009, and was likewise delayed in winters with above-average pre-melt SWE depth and delayed snow-free DOY, e.g., 2011/2012 (Figure 10 a, b ,d). Also, there was a general pattern of a 45-55 days green-up period (Figure 13a). Across years, the duration of the green-up period was most consistent, with a standard error of the mean of 1 to 2 days (Figure 14), in areas with 45-55 days green-up period. However, in some areas the green-up period differed from this general pattern. A shorter green-up period was found at higher elevation due to either late snowmelt or early timing of MaxNDVI. A longer green-up period was found in areas with relatively early snowmelt, primarily in the inland area (Figure 13a). However, these areas of exception were located outside the areas, where there was a significant positive correlation between the snow-free DOY and timing of MaxNDVI (p < 0.05) (Figure 13b).

4.7 Non-linear heterogeneity In addition to mapping the regional longitudinal gradient, the 300-m spatial resolution model output also described the heterogeneity, which dominated at a landscape scale, e.g., within valleys. Locally, the overall linearity of the inland-coast gradients was overruled by landscape-scale topography-related patterns. Here, the pre-melt SWE depth increased with elevation and the snow-free DOY was delayed with increasing elevation, due to the precipitation lapse rate (Figure 9a,b), e.g., in west Clavering Island. Likewise, the landscape-scale distribution of MaxNDVI showed a decrease in vegetation greenness with increasing elevation (Figure 11a), whereas the spatial distribution of the timing of MaxNDVI showed a more pronounced variation across the regional than with increasing elevation (Figure 11b).

Paper IV

Figure 11 Average MaxNDVI (a) and average timing of MaxNDVI (DOY) (b) for the period 2000-2015 based on daily MODIS reflectance data (MOD09GQ).The topographic contours are in 100-m elevation increments. Map specifications as in Figure 1.

Paper IV

Figure 12 Significant correlation coefficient (p < 0.05) between MaxNDVI and snow-free DOY (a). Significant correlation coefficient (p < 0.05) between MaxNDVI and pre-melt SWE depth (b). The topographic contours are in 100-m elevation increments and map specifications are as in Figure 1.

Paper IV

Figure 13 Duration of greening-up period, i.e., the difference in days between timing of MaxNDVI and Snow-free DOY (a). Significant correlation coefficient between timing of MaxNDVI and Snow-free DOY (b). The topographic contours are in 100-m elevation increments and map specifications are as in Figure 1.

Paper IV

Figure 14 Standard error of the mean estimated for the average greening-up period (days), for the years 2000-2015.

5 Discussion

5.1 Application of gradients By applying the high-resolution simulations of SnowModel in combination with daily MODIS NDVI spatial representations, we were able to quantify gradients in ecologically relevant variables across complex mountainous terrain within the ice-free NE-Greenland region. Furthermore, these data sets proved useful to study landscape-scale relations since they resolved the NDVI patterns and variation in SWE depth on landscape scale, e.g., variations between valleys and hill/mountain slopes, and limited the temporal uncertainty in estimated timing of snowmelt or MaxNDVI to 1-2 days because of the daily temporal resolution.

Inland-coast gradients in temperature and precipitation have previously been identified in a range of data sets (e.g., Ohmura and Reeh 1991, Schuenemann et al. 2009, Lucas-Picher et al. 2012), but so far not properly quantified by eliminating the orographic effects, nor mapped in detail as in this current study. The inland- coast gradient of annual precipitation found in other regions of ice-free Greenland, e.g., W- and SW- Greenland (Mernild et al. 2014) equal 0.0029 m km-1 and 0.0071 m km-1, respectively. These are comparable with the quantified pre-melt SWE depth gradient of 0.003 m km-1 for the NE-Greenland region (Table 3). Furthermore, the air temperature gradients were investigated by Taurisano et al. (2004), who found a continentality index (K km-1), i.e., inland-coast gradient, for Nuuk Fjord in W-Greenland varying in summer (March-August) up to -0.04 K km-1 and 0.01- 0.03 K km-1 in winter (October-March). Instead, we found the

Paper IV

air temperature gradient of -0.065 °C km-1 in summer and -0.034 °C km-1 in winter; hence the inland-coast gradients remained negative across seasons. This differs from an inverse, yet weak summer signal found locally between Zackenberg and Daneborg (Figure 1) (Hansen et al. 2008) .

Within our domain, the inland-coast gradients hold a potential to describe a variability in air temperature, winter precipitation, timing of snowmelt, and timing of MaxNDVI, which would be found across a more extensive area than the NE-Greenland study region contained. Within the coastal, ice-free Greenland area, which spans 23 latitudinal degrees, there are pronounced south-north gradients in air temperature along the west and east coast, which are strongest in winter and spring months (November-April) (Abermann et al. 2016). However, the regional gradient in our study area of -0.05 °C km-1 distance, when moving from the inland towards to the coast, was approximately eight times stronger than the south-north gradient of MAAT in E-Greenland. This gradient of MAAT for the period 1979-2015 was quantified from nine coastal weather stations along the east coast of Greenland owned by the Danish Meteorological Institute (DMI) (Cappelen 2014, 2016). This gradient equaled -0.0058 °C km-1 distance from south to north (Figure 15). Hence, for each 166 km traveled in a northern direction along the south-north E-Greenland gradient the MAAT will decrease by 1 °C. Similarly, in general one need only to hike 20 km towards the coast within our study area, to experience a decrease in MAAT of 1 °C, but will encounter landscape-scale-related variation along the way.

Figure 15 Mean annual air temperature for the period 1979-2015 for nine coastal DMI meteorological stations along the east coast of Greenland (Cappelen 2014, 2016). Linear fit (black line) statistics: slope = -0.0058, R2 = 0.96, p-value < 0.001, n = 9.

Furthermore, comparing the regional inland-coast gradients to other Arctic regions indicated that a gradient of this magnitude is usually found across much larger geographical distances. Kobayashi et al. (2016) estimated the latitudinal gradients in timing of start of the growing season (SOS), i.e., an indicator of the

Paper IV

growing season timing, and the snowmelt timing across Alaska, and found that gradients of SOS varied between 0.031 days km-1 and 0.044 days km-1 distance from south to north. This gradient, covering more than ten degrees latitude, is less than half of the gradient that we found in timing of MaxNDVI within our 97.5 km-wide region. Furthermore, the timing of snowmelt across Alaska was delayed between 0.018 days km-1 and 0.053 days km-1 from south to north (Kobayashi et al. 2016), whereas the gradient of snow-free DOY was almost an order of magnitude stronger (0.4 days km-1) across the NE-Greenland region.

5.2 Regional variation in Snow and NDVI This study mapped regional gradients, where MAAT decreased with distance from the GrIS margin. The average pre-melt SWE depth, snow-free DOY, and the timing of MaxNDVI increased significantly across the same distance. However, there was a high degree of landscape-scale variability superimposed on the gradients for all variables including MaxNDVI, which was captured in both the SnowModel outputs and the MODIS data set (Figure 11 and Figure 13). This landscape-scale variability originated both from the quantified regional gradients and the landscape-scale processes/dynamics of precipitation and air temperature lapse rates, topographic aspect, and exposure. This emphasizes that the landscape-scale gradients are nonlinear and more complex and differ from the linear inland-coast regional gradients. The combination of spatial landscape-scale heterogeneity and the regional gradient constitutes a mesocosm of large-scale environmental variation within a condensed geographical area. Such mesocosm may serve as excellent natural experimental sites, allowing researchers to study patterns and ecosystem processes, such as carbon fluxes (McGuire et al. 2009), phenology (Oberbauer et al. 2013), community structure and diversity (Legagneux et al. 2014), whilst minimizing logistic costs.

5.3 Perspectives SnowModel being run with reanalysis data and AWS data as inputs was able to capture both the overall gradients and the landscape-scale variability in snow variables and air temperature, which was reflected in the vegetation greenness. Hence, the distributed snow representations were modeled at an ecologically relevant spatial and temporal resolution. This dynamical model setup allowed explaining the temporally and spatially varying vegetation greenness and phenology, which were driven both by the amount of snow and snowmelt timing.

There are multiple advantages of choosing a relatively small region that comprises strong environmental gradients (usually found across a geographically larger area) because it enables us to acquire ground truthing and observations both spatially and temporally synchronized with MODIS satellite observations. We collected SWE depth observations across the region in less than a month, where the weather was relatively stable during the pre-melt spring period. Furthermore, we had a NDVI sensor installed during parts of the study period for ground truthing of the MODIS NDVI estimates. A similar setup can be performed elsewhere within a relatively small region with ‘compressed gradients’ representative of the Arctic climate.

Because the region in NE-Greenland comprises a strong air temperature and snow gradient it was the ideal area for studying the biological variables’ dependence on these environmental gradients since, e.g., the vegetation coverage is a product of the long-term gradients in snow distribution and temperature, which in turn are driven by topography. Furthermore, such regions in Greenland or Arctic, comprising a mesocosm, may serve as an alternative to suggested expansion of standardized plot-experiments in the Arctic (Walker et al. 2016), where plots can be established within relatively short distances but across strong environmental gradients to optimize the vegetation-change studies.

Paper IV

6 Conclusions • We quantified environmental gradients including annual average of air temperature, pre-melt SWE depth, snow-free DOY, MaxNDVI, and the timing of MaxNDVI across complex mountainous terrain within an ice-free NE-Greenland region. Both MODIS NDVI estimates and modeled snow variables were validated against ground observations and showed satisfactory results. Pre-melt SWE-depth and snow-free DOY gradients were identified in all years 2000-2015, also during contrasting snow-rich and snow-poor years. • While the MaxNDVI level was persistent through the study period 2000-2015, the timing of MaxNDVI varied from year to year. This interannual variation was mainly driven by variability in timing of snowmelt, since we found a significant strong correlation between the snow-free DOY and the timing of MaxNDVI. • The quantified continentality gradients, i.e., air temperature decreasing with distance from the GrIS margin, was eight times stronger than the mean annual air temperature gradients found from south to north along the east coast of Greenland. • Regions, similar to this NE-Greenland study area, comprising ‘compressed’ environmental gradients are ideal ‘laboratories’ for studying the biological variables’ dependence of these environmental gradients. Additionally, it logistically enables the collection of in situ ground truthing/observations under naturally varying conditions.

7 Acknowledgements We wish to thank the logistics team at Zackenberg Research Station, Aarhus University for their support during the snow fieldwork in April 2014, furthermore we gratefully acknowledge the logistic support of Arctic Research Centre (ARC), Aarhus University. Support was also provided by the Canada Excellence Research Chair (CERC). Data from the Greenland Ecosystem Monitoring (GEM) Program ClimateBasis were provided by Asiaq – Greenland Survey, Greenland. Data from the GEM Programme GlacioBasis were provided by the Geological Survey of Denmark and Greenland (GEUS), Denmark. Data from the GEM Program GeoBasis were provided by the Department of Bioscience, Aarhus University, Denmark in collaboration with Department of Geosciences and Natural Resource Management, Copenhagen University, Denmark. We thank the Climate Research Section, Central Institute for Meteorology and Geodynamics (ZAMG) in Austria, for providing meteorology and snow data from Freya Glacier. We are thankful to the Danish Meteorological Institute for providing the station data. Thank you also to NASA’s Earth Science Program for the freely available MODIS data. This study was funded by the Environmental Protection Agency and the Danish Energy Agency, and it is a contribution to the Arctic Science Partnership (ASP) asp- net.org.

Paper IV

8 Appendix I Table S1 Observations of mean snow depth across transect, bulk density, and snow-water-equivalent (SWE) depth from 55 sites across the region within subregions, w= Wollaston Forland, z = Zackenberg Valley, i = Inland/ Clavering Island (Figure 1, main text). Site types a = SWE depth use in SWE-assimilation in SnowAssim (Liston and Hiemstra 2008). Site types v = SWE depth used for validating modeled SWE-depth.

Site id UTM Easting UTM Northing SWE depth Bulk density Snow depth Elevation Site type km km m kg m-3 m m a.s.l. w101 524.264 8249.850 0.303 317.3 0.956 36 v w103 546.981 8245.171 0.083 358.0 0.233 263 v w104 544.295 8246.991 0.249 404.8 0.615 268 v w105 544.974 8240.552 0.021 307.1 0.067 1 v w107 524.483 8244.863 0.194 365.5 0.531 21 v w108 523.927 8251.294 0.023 356.5 0.063 75 v w109 547.543 8275.219 0.450 325.2 1.384 9 a w110 546.108 8272.648 0.694 325.2 2.135 9 v w111 542.400 8274.565 0.724 325.2 2.226 9 v w112 537.086 8276.818 0.609 322.6 1.886 30 v w113 529.279 8260.258 0.594 257.6 2.307 30 v w114 531.158 8261.086 0.580 257.6 2.251 30 a w115 531.160 8267.048 0.469 257.6 1.823 30 v w116 528.805 8274.331 0.469 322.6 1.453 30 v w117 532.285 8278.317 0.480 322.6 1.489 30 v w118 532.599 8264.695 0.519 257.6 2.016 30 v w119 531.288 8274.673 0.541 322.6 1.677 30 a w120 553.957 8268.726 0.654 325.2 2.012 9 v w121 557.352 8262.735 0.387 325.2 1.189 9 v w122 555.534 8264.908 0.777 325.2 2.388 9 v z201 511.413 8270.075 0.364 360.0 1.011 130 v z202 513.473 8270.150 0.164 337.1 0.488 293 v z203 513.800 8268.591 0.482 345.0 1.396 210 v z204 522.714 8254.945 0.382 398.0 0.960 4 v z205 519.479 8261.692 0.328 362.5 0.904 50 v z206 512.795 8266.821 0.333 355.3 0.937 59 v z207 514.063 8266.727 0.391 365.7 1.070 82 v z208 517.340 8265.244 0.417 378.5 1.102 160 v z209 506.216 8269.815 0.268 349.1 0.767 164 v z210 508.927 8269.410 0.356 385.2 0.923 137 v z211 520.066 8263.693 0.520 392.2 1.326 221 v z212 518.121 8263.213 0.247 378.5 0.653 75 v z213 512.967 8264.022 0.202 369.2 0.546 26 v z214 513.392 8264.713 0.260 337.2 0.771 44 a z215 513.332 8265.821 0.301 329.8 0.911 46 v z216 511.321 8267.708 0.269 369.7 0.728 49 v z217 516.080 8268.224 0.316 393.7 0.802 421 v z218 514.442 8265.704 0.328 329.8 0.993 46 v z219 516.105 8263.288 0.309 331.7 0.932 18 v z220 511.598 8265.635 0.193 347.6 0.556 64 v z221 510.708 8264.145 0.409 380.2 1.076 13 v z222 514.423 8263.606 0.220 287.2 0.765 18 v i301 487.252 8270.166 0.148 309.3 0.477 251 v i302 480.235 8263.868 0.062 321.4 0.191 151 v i303 481.982 8267.066 0.083 329.0 0.252 330 v i304 483.419 8268.644 0.182 319.2 0.571 347 a i305 492.409 8268.262 0.222 330.7 0.672 259 v i306 493.920 8270.125 0.229 346.9 0.661 251 v i307 474.534 8253.679 0.029 305.3 0.096 43 v i308 476.708 8249.316 0.067 317.4 0.211 80 v i309 479.856 8255.870 0.079 319.2 0.247 34 v i310 475.736 8259.119 0.003 325.1 0.009 43 v i311 486.635 8266.962 0.091 333.9 0.273 237 v i312 483.459 8264.076 0.096 331.1 0.289 295 v i313 499.808 8270.569 0.110 333.6 0.328 160 v

Paper IV

9 Appendix II

9.1 Moderate Resolution Imaging Spectroradiometer (MODIS) data To calculate Normalized Differenced Vegetation Index (NDVI) through time (2000-2015) and across the region we used the reflectance band 1 (620-670 nm) and band 2 (841-876 nm) and the quality check flag from MODIS Daily Surface Reflectance MOD09GQ (Collection 6) in 250-m resolution and the State Quality Assessment (QA) flag from MOD09GA (Collection 6) 1-km resolution. The 1 km State QA data held within the MOD09GA product is considered the primary overall quality data source for MODIS surface reflectance products and is recommended for quality screening for the 250-m reflectance data (Vermote et al. 2016). MOD09GA contains information about the state of a pixel, i.e., characteristics of each pixel that are not dependent upon band or resolution (Vermote et al. 2016). The MODIS tile h17v01, covering the Greenland east coast including our study area, was downloaded for both products from NASA Earthdata Search (https://search.earthdata.nasa.gov).

The 250-m Surface Reflectance band 1 and band 2 were in 16-bit signed integer, and we applied the scaling factor of 0.0001 to these values (Vermote et al. 2016). The 250-m Level 2/Level 2G Surface Reflectance Quality Band was a bit field in 16-bit unsigned integer that needed interpretation in order to set the quality flags as given in Table S1. The 1 km State QA (regridded to 300-m resolution to match the reflectance bands resolution) is a 16-bit layer containing per pixel quality flags relating to cloud state, cloud shadow, aerosol quantity, and basic land cover characteristics such as the land/water, snow, and fire flags. We aimed at including only cloud-free data in the analysis; hence, we focused the quality assessment on removing cloudy grid cell using the quality flag settings as in Table S2.

Table S1 Description of the applied quality flags in MOD09GQ (16-bit) (Vermote et al. 2016, Table 8).

Bit no. Parameter Bit combination Flag meaning corrected product produced at 0-1 MODLAND QA bits 00 ideal quality all bands 4-7 band 1 data quality four bit range 0000 highest quality 8-11 band 2 data quality four bit range 0000 highest quality

Table S2 Description of the applied State Quality Assessment (QA) flags in MOD09QA (16-bit) (Vermote et al. 2016, Table 13).

Bit no. Parameter Bit combination Flag meaning 0-1 cloud state 00 clear 2 cloud shadow 0 no 8-9 cirrus detected 00 none 10 internal cloud algorithm flag 1 0 no cloud

9.2 MODIS imagery preprocessing For the preprocessing of the reflectance and QA data, we used the GDAL programs, gdal_translate, for converting the downloaded .hdf files to .tif files and gdal_warp for subsetting the region of interest within the h17v01 MODIS tile. Furthermore, we used the program gdal_warp to reproject the data from the MODIS sinusoidal projection to Universal Transverse Mercator (UTM) and regridding the initial 250-m spatial

Paper IV

resolution to 300-m to make it comparable with the 300-m spatial resolution of the SnowModel outputs. Additionally, the MODIS preprocessing included:

• A requirement for the input NDVI data is that each grid cell should have at least 15 values (non- interpolated) above 0.2 in NDVI level per year to be included in the analysis in order to include only vegetation-covered grid cells (Raynolds et al. 2006, Ding et al. 2016). The minimum number of valid data points per growing season (above 0.2 NDVI) was set to 15 based on counts of valid grid cells in known vegetated areas, e.g., the Zackenberg Valley, to ensure that these were included in the analysis. • The original NDVI data set included several missing data, due to missing data files or removed as part of the quality check (MOD09GA). These gaps were filled using temporal linear interpolation between the available valid original NDVI values, under the assumption that the NDVI values from one day to the next were temporally auto-correlated. • To find the annual maximum level of NDVI (MaxNDVI), the 10% highest NDVI values per growing season (occurring between 1 April and 30 October) were averaged. • To find the annual timing of MaxNDVI given as DOY, we summed all NDVI values above or equal to 0.2 during the period 1 April through 30 October and cumulated the increasing and decreasing NDVI through the same period. We defined the timing of MaxNDVI as the DOY, where the cumulated NDVI equaled 40% of the summed NDVI. • The 40%-threshold was found by testing different percentage thresholds from 30% to 60% on ground-based NDVI observations. We used daily NDVI ground observations from three automatic stations located in the Zackenberg Valley (M4, M3, and MM2). We found the actual timing of MaxNDVI, i.e., the DOY, when the maximum NDVI value of the growing season occur, which is easily distinguished in these type of daily dataset (do not include much noise). Furthermore, we estimated the timing of MaxNDVI for each year for all three stations using the same method as for the MODIS data, with different percentage thresholds (30%-60%). The ‘40 %’ was the threshold resulting in the lowest mean difference between the actual timing of MaxNDVI and estimated timing of MaxNDVI. • To validate the method for finding MaxNDVI and the timing of MaxNDVI we also used the daily NDVI ground observations from the three Zackenberg stations (M4, M3, and MM2). We found the MaxNDVI and the timing of MaxNDVI for each year for all three stations using the same method as for the MODIS data. Hereafter, we compared these with MaxNDVI and the timing of MaxNDVI estimated for the MODIS grid cells matching the locations of the respective stations in a regression analysis.

Paper IV

10 References Abermann, J., B. U. Hansen, M. Lund, S. Wacker, M. Karami, and J. Cappelen. 2016. Hotspots and key periods of Greenland climate change during the past 6 decades. Ambio In review. Azaele, S., A. Maritan, S. J. Cornell, S. Suweis, J. R. Banavar, D. Gabriel, and W. E. Kunin. 2015. Towards a unified descriptive theory for spatial ecology: predicting biodiversity patterns across spatial scales. Methods in Ecology and Evolution 6:324-332. Bhatt, U., D. Walker, M. Raynolds, P. Bieniek, H. Epstein, J. Comiso, J. Pinzon, C. Tucker, and I. Polyakov. 2013. Recent Declines in Warming and Vegetation Greening Trends over Pan-Arctic Tundra. Remote Sensing 5:4229. Bienau, M. J., D. Hattermann, M. Kröncke, L. Kretz, A. Otte, W. L. Eiserhardt, A. Milbau, B. J. Graae, W. Durka, and R. L. Eckstein. 2014. Snow cover consistently affects growth and reproduction of Empetrum hermaphroditum across latitudinal and local climatic gradients. Alpine Botany 124:115-129. Billings, W. D. and H. A. Mooney. 1968. Ecology of Arctic and Alpine Plants. Biological Reviews of the Cambridge Philosophical Society 43:481-&. Bjorkman, A. D., S. C. Elmendorf, A. L. Beamish, M. Vellend, and G. H. R. Henry. 2015. Contrasting effects of warming and increased snowfall on Arctic tundra plant phenology over the past two decades. Global Change Biology 21:4651-4661. Bliss, L. and N. Matveyeva. 1992. Circumpolar Arctic Vegetation. Pages 59-89 in F. S. C. III, R. L. Jefferies, J. F. Reynolds, G. R. Shaver, J. Svoboda, and E. W. Chu, editors. Arctic ecosystems in a changing climate: an ecophysiological perspective. Academic Press Inc., San Diego, California, USA. Blok, D., G. Schaepman-Strub, H. Bartholomeus, M. M. P. D. Heijmans, T. C. Maximov, and F. Berendse. 2011. The response of Arctic vegetation to the summer climate: relation between shrub cover, NDVI, surface albedo and temperature. Environmental Research Letters 6:035502. Cappelen, J. 2014. Weather observations from Greenland 1958-2014 - Observation data with description. Danish Meteorological Institute. Cappelen, J. 2016. Guide to Climate Data and Information from the Danish Meteorological Institute. Updated July 2016. Danish Meteorological Institute. Cooper, E. J. 2014. Warmer Shorter Winters Disrupt Arctic Terrestrial Ecosystems. Annual Review of Ecology, Evolution, and Systematics 45:271-295. Dee, D. P., S. M. Uppala, A. J. Simmons, P. Berrisford, P. Poli, S. Kobayashi, U. Andrae, M. A. Balmaseda, G. Balsamo, P. Bauer, P. Bechtold, A. C. M. Beljaars, L. van de Berg, J. Bidlot, N. Bormann, C. Delsol, R. Dragani, M. Fuentes, A. J. Geer, L. Haimberger, S. B. Healy, H. Hersbach, E. V. Hólm, L. Isaksen, P. Kållberg, M. Köhler, M. Matricardi, A. P. McNally, B. M. Monge-Sanz, J. J. Morcrette, B. K. Park, C. Peubey, P. de Rosnay, C. Tavolato, J. N. Thépaut, and F. Vitart. 2011. The ERA-Interim reanalysis: configuration and performance of the data assimilation system. Quarterly Journal of the Royal Meteorological Society 137:553- 597. Ding, Y. L., X. M. Zheng, K. Zhao, X. P. Xin, and H. J. Liu. 2016. Quantifying the Impact of NDVIsoil Determination Methods and NDVIsoil Variability on the Estimation of Fractional Vegetation Cover in Northeast China. Remote Sensing 8. Elmendorf, S. C., G. H. Henry, R. D. Hollister, R. G. Bjork, A. D. Bjorkman, T. V. Callaghan, L. S. Collier, E. J. Cooper, J. H. Cornelissen, T. A. Day, A. M. Fosaa, W. A. Gould, J. Gretarsdottir, J. Harte, L. Hermanutz, D. S. Hik, A. Hofgaard, F. Jarrad, I. S. Jonsdottir, F. Keuper, K. Klanderud, J. A. Klein, S. Koh, G. Kudo, S. I. Lang, V. Loewen, J. L. May, J. Mercado, A. Michelsen, U. Molau, I. H. Myers-Smith, S. F. Oberbauer, S. Pieper, E. Post, C. Rixen, C. H. Robinson, N. M. Schmidt, G. R. Shaver, A. Stenstrom, A. Tolvanen, O. Totland, T. Troxler, C. H. Wahren, P. J. Webber, J. M. Welker, and P. A. Wookey. 2012a. Global assessment of experimental climate warming on tundra vegetation: heterogeneity over space and time. Ecol Lett 15:164-175. Elmendorf, S. C., G. H. R. Henry, R. D. Hollister, R. G. Bjork, N. Boulanger-Lapointe, E. J. Cooper, J. H. C. Cornelissen, T. A. Day, E. Dorrepaal, T. G. Elumeeva, M. Gill, W. A. Gould, J. Harte, D. S. Hik, A. Hofgaard, D. R. Johnson, J. F. Johnstone, I. S. Jonsdottir, J. C. Jorgenson, K. Klanderud, J. A. Klein, S. Koh, G. Kudo, M. Lara, E. Levesque, B. Magnusson, J. L. May, J. A. Mercado-Diaz, A. Michelsen, U. Molau, I. H. Myers- Smith, S. F. Oberbauer, V. G. Onipchenko, C. Rixen, N. M. Schmidt, G. R. Shaver, M. J. Spasojevic, P. E. Porhallsdottir, A. Tolvanen, T. Troxler, C. E. Tweedie, S. Villareal, C. H. Wahren, X. Walker, P. J. Webber, J. M. Welker, and S. Wipf. 2012b. Plot-scale evidence of tundra vegetation change and links to recent summer warming. Nature Climate Change 2:453-457. Elmendorf, S. C., G. H. R. Henry, R. D. Hollister, A. M. Fosaa, W. A. Gould, L. Hermanutz, A. Hofgaard, I. S. Jónsdóttir, J. C. Jorgenson, E. Lévesque, B. Magnusson, U. Molau, I. H. Myers-Smith, S. F. Oberbauer, C. Rixen, C. E. Tweedie, and M. D. Walker. 2015. Experiment, monitoring, and gradient methods used to infer

Paper IV

climate change effects on plant communities yield consistent patterns. Proceedings of the National Academy of Sciences 112:448-452. Epstein, H., U. Bhatt, M. Raynolds, D. Walker, P. Bieniek, C. Tucker, J. Pinzon, H. Zeng, G. Jia, and K. Guay. 2015. Tundra Greenness. NOAA Arctic Report Card 2015. Goodison, B. E., P. Y. T. Louie, and D. Yang. 1998. WMO solid precipitation intercomparison. Final report WMO/TD- 872 Instruments and observing methods report., Geneva: World Meteorological Organization. Grippa, M., L. Kergoat, T. Le Toan, N. Mognard, N. Delbart, J. L'hermitte, and S. Vicente‐Serrano. 2005. The impact of snow depth and snowmelt on the vegetation variability over central Siberia. Geophysical Research Letters 32. Groendahl, L., T. Friborg, and H. Soegaard. 2006. Temperature and snow-melt controls on interannual variability in carbon exchange in the high Arctic. Theoretical and Applied Climatology 88:111-125. Halbritter, A. H., J. M. Alexander, P. J. Edwards, and R. Billeter. 2013. How comparable are species distributions along elevational and latitudinal climate gradients? Global Ecology and Biogeography 22:1228-1237. Hansen, B. U., C. Sigsgaard, L. Rasmussen, J. Cappelen, J. Hinkler, S. H. Mernild, D. Petersen, M. P. Tamstorf, M. Rasch, and B. Hasholt. 2008. Present-day climate at Zackenberg. Advances in ecological research 40:111-150. Henry, G. and U. Molau. 1997. Tundra plants and climate change: the International Tundra Experiment (ITEX). Global Change Biology 3:1-9. Hollesen, J., A. Buchwal, G. Rachlewicz, B. U. Hansen, M. O. Hansen, O. Stecher, and B. Elberling. 2015. Winter warming as an important co-driver for Betula nana growth in western Greenland during the past century. Global Change Biology 21:2410-2423. Jarvis, P. G. 1995. Scaling Processes and Problems. Plant Cell and Environment 18:1079-1089. Kobayashi, H., A. P. Yunus, S. Nagai, K. Sugiura, Y. Kim, B. Van Dam, H. Nagano, D. Zona, Y. Harazono, M. S. Bret- Harte, K. Ichii, H. Ikawa, H. Iwata, W. C. Oechel, M. Ueyama, and R. Suzuki. 2016. Latitudinal gradient of spruce forest understory and tundra phenology in Alaska as observed from satellite and ground-based data. Remote Sensing of Environment 177:160-170. Körner, C. 2007. The use of ‘altitude’ in ecological research. Trends in Ecology & Evolution 22:569-574. Kottek, M., J. Grieser, C. Beck, B. Rudolf, and F. Rubel. 2006. World map of the Köppen-Geiger climate classification updated. Meteorologische Zeitschrift 15:259-263. Kremers, K. S., R. D. Hollister, and S. F. Oberbauer. 2015. Diminished Response of Arctic Plants to Warming over Time. PLoS One 10:e0116586. Leffler, A. J. and J. M. Welker. 2013. Long-term increases in snow pack elevate leaf N and photosynthesis in Salix arctica: responses to a snow fence experiment in the High Arctic of NW Greenland. Environmental Research Letters 8:025023. Legagneux, P., G. Gauthier, N. Lecomte, N. M. Schmidt, D. Reid, M. Cadieux, D. Berteaux, J. Bêty, C. Krebs, and R. Ims. 2014. Arctic ecosystem structure and functioning shaped by climate and herbivore body size. Nature Climate Change 4:379-383. Liston, G. E. and K. Elder. 2006a. A distributed snow-evolution modeling system (SnowModel). Journal of Hydrometeorology 7:1259-1276. Liston, G. E. and K. Elder. 2006b. A meteorological distribution system for high-resolution terrestrial modeling (MicroMet). Journal of Hydrometeorology 7:217-234. Liston, G. E. and C. A. Hiemstra. 2008. A simple data assimilation system for complex snow distributions (SnowAssim). Journal of Hydrometeorology 9:989-1004. Lucas-Picher, P., M. Wulff-Nielsen, J. H. Christensen, G. Adalgeirsdottir, R. Mottram, and S. B. Simonsen. 2012. Very high resolution regional climate model simulations over Greenland: Identifying added value. Journal of Geophysical Research-Atmospheres 117. Lund, M., J. M. Falk, T. Friborg, H. N. Mbufong, C. Sigsgaard, H. Soegaard, and M. P. Tamstorf. 2012. Trends in CO2exchange in a high Arctic tundra heath, 2000-2010. Journal of Geophysical Research: Biogeosciences 117:n/a-n/a. McGuire, A. D., L. G. Anderson, T. R. Christensen, S. Dallimore, L. Guo, D. J. Hayes, M. Heimann, T. D. Lorenson, R. W. Macdonald, and N. Roulet. 2009. Sensitivity of the Carbon Cycle in the Arctic to Climate Change. Ecological Monographs 79:523-555. Mernild, S. H., E. Hanna, J. C. Yde, J. Cappelen, and J. K. Malmros. 2014. Coastal Greenland air temperature extremes and trends 1890-2010: annual and monthly analysis. International Journal of Climatology 34:1472-1487. Myers-Smith, I. H., S. C. Elmendorf, P. S. A. Beck, M. Wilmking, M. Hallinger, D. Blok, K. D. Tape, S. A. Rayback, M. Macias-Fauria, B. C. Forbes, J. D. M. Speed, N. Boulanger-Lapointe, C. Rixen, E. Levesque, N. M. Schmidt, C. Baittinger, A. J. Trant, L. Hermanutz, L. S. Collier, M. A. Dawes, T. C. Lantz, S. Weijers, R. H. Jorgensen, A. Buchwal, A. Buras, A. T. Naito, V. Ravolainen, G. Schaepman-Strub, J. A. Wheeler, S. Wipf, K.

Paper IV

C. Guay, D. S. Hik, and M. Vellend. 2015. Climate sensitivity of shrub growth across the tundra biome. Nature Climate Change 5:887-+. Myneni, R. B., C. D. Keeling, C. J. Tucker, G. Asrar, and R. R. Nemani. 1997. Increased plant growth in the northern high latitudes from 1981 to 1991. Nature 386:698-702. Oberbauer, S., S. Elmendorf, T. Troxler, R. Hollister, A. Rocha, M. Bret-Harte, M. Dawes, A. Fosaa, G. Henry, and T. T. Høye. 2013. Phenological response of tundra plants to background climate variation tested using the International Tundra Experiment. Philosophical Transactions of the Royal Society of London B: Biological Sciences 368:20120481. Ohmura, A. and N. Reeh. 1991. New precipitation and accumulation maps for Greenland. Journal of Glaciology 37:140-148. Pedersen, S. H., M. P. Tamstorf, J. Abermann, A. Westergaard-Nielsen, M. Lund, K. Skov, C. Sigsgaard, M. R. Mylius, B. U. Hansen, G. E. Liston, and N. M. Schmidt. 2016. Spatiotemporal characteristics of seasonal snow cover in Northeast Greenland from in situ observations. Arctic, Alpine, and Antarctic Research 48:653–671. Piao, S., H. Nan, C. Huntingford, P. Ciais, P. Friedlingstein, S. Sitch, S. Peng, A. Ahlström, J. G. Canadell, and N. Cong. 2014. Evidence for a weakening relationship between interannual temperature variability and northern vegetation activity. Nature communications 5. Pudas, E., A. Tolvanen, J. Poikolainen, T. Sukuvaara, and E. Kubin. 2008. Timing of plant phenophases in Finnish Lapland in 1997-2006. Boreal environment research 13:31-43. Raynolds, M., D. Walker, and H. Maier. 2006. NDVI patterns and phytomass distribution in the circumpolar Arctic. Remote Sensing of Environment 102:271-281. Raynolds, M. K., D. A. Walker, H. E. Epstein, J. E. Pinzon, and C. J. Tucker. 2012. A new estimate of tundra-biome phytomass from trans-Arctic field data and AVHRR NDVI. Remote Sensing Letters 3:403-411. Riedel, S. M., H. E. Epstein, and D. A. Walker. 2005. Biotic controls over spectral reflectance of arctic tundra vegetation. International Journal of Remote Sensing 26:2391-2405. Schuenemann, K. C., J. J. Cassano, and J. Finnis. 2009. Synoptic forcing of precipitation over Greenland: Climatology for 1961-99. Journal of Hydrometeorology 10:60-78. Semenchuk, P. R., B. Elberling, and E. J. Cooper. 2013. Snow cover and extreme winter warming events control flower abundance of some, but not all species in high arctic Svalbard. Ecol Evol 3:2586-2599. Taurisano, A., C. E. Bøggild, and H. G. Karlsen. 2004. A Century of Climate Variability and Climate Gradients from Coast to Ice Sheet in West Greenland. Geografiska Annaler. Series A, Physical Geography 86:217-224. Tucker, C. J. 1979. Red and Photographic Infrared Linear Combinations for Monitoring Vegetation. Remote Sensing of Environment 8:127-150. Vermote, E. F., J. C. Roger, and J. P. Ray. 2016. MODIS Surface Reflectance User’s Guide Collection 6 Version 1.4. MODIS Land Surface Reflectance Science Computing Facility Greenbelt, Maryland, USA. Walker, D. A., F. J. A. Daniels, I. Alsos, U. S. Bhatt, A. L. Breen, M. Buchhorn, H. Bultmann, L. A. Druckenmiller, M. E. Edwards, D. Ehrich, H. E. Epstein, W. A. Gould, R. A. Ims, H. Meltofte, M. K. Raynolds, J. Sibik, S. S. Talbot, and P. J. Webber. 2016. Circumpolar Arctic vegetation: a hierarchic review and roadmap toward an internationally consistent approach to survey, archive and classify tundra plot data. Environmental Research Letters 11. Walker, D. A., M. K. Raynolds, F. J. Daniëls, E. Einarsson, A. Elvebakk, W. A. Gould, A. E. Katenin, S. S. Kholod, C. J. Markon, and E. S. Melnikov. 2005. The circumpolar Arctic vegetation map. Journal of Vegetation Science 16:267-282. Westergaard-Nielsen, A., M. Lund, B. U. Hansen, and M. P. Tamstorf. 2013. Camera derived vegetation greenness index as proxy for gross primary production in a low Arctic wetland area. ISPRS Journal of Photogrammetry and Remote Sensing 86:89-99. Williams, M., R. Bell, L. Spadavecchia, L. E. Street, and M. T. Van Wijk. 2008. Upscaling leaf area index in an Arctic landscape through multiscale observations. Global Change Biology 14:1517-1530. Zeng, H. and G. Jia. 2013. Impacts of snow cover on vegetation phenology in the arctic from satellite data. Advances in Atmospheric Sciences 30:1421-1432. Zeng, H., G. Jia, and H. Epstein. 2011. Recent changes in phenology over the northern high latitudes detected from multi-satellite data. Environmental Research Letters 6:045508.

Paper V

Timing of maximum vegetation greenness across ice-free Greenland.

[Extended abstract]

Pedersen, S. H., Liston, G. E., Tamstorf, M. P. and Schmidt, N. M.

South of Nuuk, W-Greenland (2012)

Paper V

Timing of maximum vegetation greenness across ice-free Greenland

Stine Højlund Pedersen1, Glen E. Liston2, Mikkel P. Tamstorf1, Niels Martin Schmidt1.

1 Arctic Research Centre, Department of Bioscience, Aarhus University, Frederiksborgvej 399, DK-4000 Roskilde, Denmark.2 Cooperative Institute for Research in the Atmosphere (CIRA), Colorado State University, Fort Collins, Colorado 80523, USA.

Motivation The growth and distribution of current Arctic above-ground biomass can be regarded as the end-result of fluctuating climate through decades and the adaptation of vegetation to the varying growing conditions. Climate variables such as air temperature, solar irradiance, and precipitation govern vegetation greenness and phenology (e.g., Billings 1952). Particularly in the Arctic, where the majority of the annual precipitation falls as snow, winter precipitation is a main driver of vegetation growth (Cooper 2014). While the amount of water accumulated and stored in the snowpack during winter, and released during spring snowmelt, limits or provides moisture during the growing season (Jones 1999), vegetation growth is also affected by derived snow characteristics such as the timing of snowmelt in spring (Pudas et al. 2008), the insulating effect of the snowpack during winter, and the length of the snow-covered period (Goodrich 1982, Sturm et al. 1997, Johansson et al. 2013). In order to investigate snow’s governing role on vegetation growth, a quantification of the spatially and temporally varying climate conditions, including snow and meteorological characteristics, is required. Furthermore, the importance of the individual drivers of vegetation growth may vary across the Arctic climate zones and across pronounced topographic relief and continentality gradients (e.g., Havström et al. 1993, Wookey et al. 1993). In Greenland, this can be investigated across multiple latitudes and complex terrain to strengthen the applicability of such an analysis. Therefore, we set out to answer the question: How does climate control vegetation greenness and timing of peak greenness within ice-free Greenland?

We answered this question by quantifying the importance of likely drivers of seasonal maximum vegetation greenness (MaxNDVI) and timing of MaxNDVI within climatically contrasting regions of ice-free Greenland. We focused on environmental drivers associated with winter, spring, and summer processes. The atmospheric-related drivers included: 1) annual sum of daily positive air temperature; 2) annual sum of daily incoming shortwave radiation (Qsi); and 3) annual sum of liquid precipitation (rain), all summed during the snow-free (potential growing) season. The analysis also included the following snow drivers: 4) pre-melt snow-water equivalent (SWE) depth, defined to be the maximum SWE achieved before spring snowmelt; 5) timing of snowmelt, defined to be the day of year (DOY) when the ground became snow-free (i.e., the day the SWE equaled 0.0 m); and 6) the fraction of the year when the ground was snow-free (i.e. the potential growing season length).

Two tasks were required to answer the above question. First, the analysis to identify regions with contrasting climate, i.e. in temperature, radiation, and moisture regimes, by mapping the spatial patterns of the climate variables in the complex mountainous terrain, and across continentality differences and latitudinal gradients in ice-free Greenland. These were computed using the spatially and temporally distributed modeling tools MicroMet and SnowModel (Liston and Elder 2006a, b) and derived from annual averages of the climate

Paper V

variables from the model simulations run at 3-hour and 300-m resolutions across the entire ice-free Greenland.

Second, we will quantify which of the ecologically relevant climate variables, mapped in the first step, are the most important in terms of explanatory power for the variations in observed MaxNDVI and in timing of MaxNDVI in each of the contrasting regions. This quantification is based on multiple regression analyses of the climate variables versus MaxNDVI and timing of MaxNDVI extracted in each year during the period 2000-2015. The analyses only include grid cells within each region that are located below 100 m a.s.l., and we use only grid cells that provide a sufficient number of valid NDVI values to estimate the annual MaxNDVI and timing of MaxNDVI for all 16 years.

The first task of the study was fulfilled and is presented within this extended abstract (Paper V). The second task, including the multiple regression analyses of potential drivers of vegetation greenness, is described in the paragraph below entitled ‘Region-specific multiple regression’. Additional work is required to complete the second task.

Combining observational and modeled data To illustrate the contrasts and complexity of the inter-regional patterns of climate conditions within ice-free Greenland, we estimated the atmospheric and snow variables using MicroMet and SnowModel (Liston and Elder 2006a, b) outputs. We performed a MicroMet and SnowModel simulation using 3-hour time increments and saved daily outputs covering the period 1 September 1979 - 31 August 2015. We used a 300- m spatial resolution to capture landscape-scale features and melt patterns, and to resolve latitudinal gradients and regional difference across ice-free Greenland; a total area of 410.449 km2. We used the ECMWF Interim Re-Analysis (ERA-Interim) reanalysis data as precipitation input (Dee et al. 2011); these inputs were adjusted using SWE observations collected during the period February through April 2014 in locations distributed across the entire ice-free Greenland in Qaanaaq, Illulissat, Kangerlussuaq, Nuuk, Qassiarsuk, Tasiilaq, and Zackenberg regions (Figure 1 Left panel). The SnowAssim submodel (Liston and Hiemstra 2008) was included in the SnowModel simulations to adjust the precipitation, using a method similar to Pedersen et al. (2015) (Paper III). In addition to the reanalysis input data, we included meterological input data from the automatic climate stations from the Greenland Ecosystem Monitorings sites Kobbefjord and Zackenberg (http://www.data.g-e-m.dk), from Arctic Station on Disko Island, and from a selection of 45 coastal automated weather stations by the Danish Meteorological Institute (DMI) (Cappelen 2014, 2016). As topographic elevation input for MicroMet and SnowModel we used the MEaSURES Greenland Ice Mapping Project (GIMP) Digital Elevation Model (Version 1) (Figure 1 Right panel). Since the vegetation distribution pattern is the product of long-term climate and persistant snow distribution patterns (Evans et al. 1989, Walker et al. 1993, Liston et al. 2002), we calculated the regional averages for each climate and snow variable, which were based on 36 annual averages during the period 1979-2015. In order to constrain the analyses and comparisons to potentially vegetated areas, we extracted annual averages for grid cells at or below 100 m a.s.l. within each region. The standard error of the means were estimated for the regional averages. To calculate vegetation greenness (NDVI) and estimate the annual maximum NDVI and its timing, we used daily MODIS reflectance data (MODGQ09, originally at 250 m resolution) that were filtered using MODGA09 quality flag information (originally at 1 km resolution) (Vermote et al. 2016). Six tiles (h15v2, h16v1, h16v2, h17v0, h17v1, h17v2) covered the entire ice-free Greenland during the period 2000-2015. The resulting NDVI datasets were resampled and reprojected to match the 300 m spatial resolution and geo- location of the SnowModel outputs. The preprocessing of the MODIS tiles was described in Paper IV.

Paper V

Figure 1 Left: Locations within ice-free Greenland (light gray area) of automated weather stations (AWS) (red, black and purple circles), ERA-Interim gridcells (black points) included in the SnowModel simulation and the locations of snow observations made in spring 2014 (blue triangles). This map and the all following maps are in a polar stereographic projection (WGS84). Right: MEaSURES Greenland Ice Mapping Project (GIMP) Digital Elevation Model (Version 1), [subset extent: N: 83, S: 60, E: -14, W: -75]. Boulder, Colorado USA. NASA National Snow and Ice Data Center Distributed Active Archive Center. doi: http://dx.doi.org/10.5067/NV34YUIXLP9W. [1 February 2016]. The elevation data (in meters above sea level [m a.s.l.]) were downloaded at 30-m resolution (horizontal uncertainty is ±18.3 m) (Howat et al. 2014) and resampled to 300-m spatial resolution. Precipitation correction and SnowModel validation In Greenland, ERA-Interim reanalysis precipitation rates included in SnowModel simulations require adjustment using ground SWE observations in order to reproduce realistic snow distributions (Paper III and Paper IV). A comparison between modeled SWE depth outputs (Figure 2, black lines) and ground-based SWE observations collected during the spring 2014 showed that adjustments ranging between a 50% reduction and 10% increase in precipitation rates were required for the Greenland domain (data not shown). 25% of SWE-depth observations, included in the data assimilation using SnowAssim to generate the spatially distributed correction-factor surface, were distributed across the ice-free area (Figure 1 Left panel). The generated correction-factor surface covering the entire Greenland was mainly reducing the precipitation rates but also slightly increasing the precipitation in NE-Greenland. The largest reductions, corresponding to correction factors between 0.50 and 0.65, were found in NW-Greenland (e.g., Qaanaaq in Figure 2) and between 0.70 and 0.90 in SE-Greenland (e.g., Tasiilaq in Figure 2). In contrast, a minor reduction, with correction factors between 0.85 and 0.90, was found in SW-Greenland (e.g., Qassiarsuk). In NE-Greenland, between Zackenberg and Station Nord, the precipitation rates were slightly increased with correction factors between 1.00 and 1.12. The resulting outputs of SWE depth using the corrected ERA-Interim precipitation rates showed high correspondence with the remaining independent set of SWE-depth observations collected in eight locations across the ice-free Greenland region (Figure 2). The correction factor of 1.07-1.10

Paper V

found for the NE-Greenland area near Zackenberg (74 °N) in 2014 match well the precipitation correction previously found for the Zackenberg region, where the average correction factors vary 1.1 ±0.4. These were found for the period 2004-2015 between observed SWE depth and modeled SWE depth based also on ERA- Interim precipitation rates (Paper II). Hence, these spatially distributed precipitation correction factors were applied to all precipitation rates included in the SnowModel simulation during the period 1979-2015.

Figure 2 The average SnowModel snow-water equivalent (SWE) depth in eight observational sites with (red line) and without (black line) corrected ERA-Interim precipitation input. Independent SWE depth observations collected in spring 2014 (blue triangle) were used as validation of the modeled SWE depth including the corrected precipitation input. Spatial climate and snow distributions across ice-free Greenland In general, the spatial representations of the ecologically relevant climate variables showed pronounced variation across latitudes, between Greenland’s west and east coasts, between coastal and inland areas, and with local elevation differences (Figure 3a-d). The elevational gradients, primarily controlled by air temperature and precipitation lapse rates, dominated all regions of Greenland and created a distinct difference between valley areas and mountain tops in all mapped variables; this was particularly visible in

Paper V

the snow variables (Figure 3b,d). The snow-free fraction of the year (Figure 3a) was controlled by the timing of snowmelt in spring/summer (Figure 3b), the timing of the first snowfall in autumn, and the occurrence of episodic snowmelt events, which have previously been quantified in the ice-free Greenland areas by Pedersen et al. (2015) (Paper III). In land areas below 100 m a.s.l., the snow-free fraction of the year varied from the northern-most regions with 5%-10 % (approximately 0.5-1 month) to the southern part of Greenland with 35%-45% (approximately 4-5 months) (Figure 3a). Within large parts of N- and SE- Greenland and areas at elevations above 900 m a.s.l., the ground was either snow-covered year round and the snow never melted away completely, or the ground was snow-free during less than a month of the year (Figure 3a). The latitudinal gradients differed between E- and W-Greenland, e.g., for summer precipitation sums (Figure 3c). Due to the relatively short snow-free period in the north (approximately 0.5-1 month; Figure 3a), and the limited sum of positive air temperatures during this snow-free period, the amounts of rain were limited in the northern part of Greenland and along the majority of Greenland’s east coast. In SW- and SE-Greenland, where low pressure systems pass most frequently and thus receive the maximum amount of precipitation (Mernild et al. 2015), we found the maximum rain totals. The average winter precipitation amount, given as the average pre-melt SWE depth, was also found in SE-Greenland with an average of above 0.7 m (Figure 3d). In comparison, everywhere else in Greenland the pre-melt SWE-depth ranged between 0.0 m and 0.7 m and only at higher elevations (Figure 1 Right panel) in W-Greenland above 1000 m a.s.l. reached pre-melt SWE-depth values above 0.7 m (Figure 3d). A clear gradient in pre-melt SWE depth was found from inland near the Greenland Ice Sheet (GrIS) margin towards the coast line in N-Greenland (81 °N-82 °N), W-Greenland (66 °N-69 °N), and NE-Greenland (72 °N-75 °N), which is also observed in the NE-Greenland area near Zackenberg (74 °N) (Paper III). In SW- and SE-Greenland the snow-free DOY occurred in mid-May through June (DOY 120-180) in the areas nearest to the coast and below 200 m a.s.l. (Figure 3b). On average, the snowmelt occurred later in NW-, N-, and NE-Greenland, namely within the first half of August (DOY 210-230).

The average air temperature sum during snow-free days showed a marked difference between NW-, N-, and NE-Greenland (ranging 100-500 °C) and W-, SW-, and SE-Greenland (ranging 500-1000 °C). Furthermore, there was a difference between coastal and inland areas, which was both visible in W-Greenland (66 °N-69 °N) and in E-Greenland (70 °N-76 °N), where the ice-free region extents across the largest distance from the GrIS margin to the most eastward or westward coastlines. Within this E-Greenland region, strong air temperature gradients were identified and quantified by Pedersen et al. (Paper IV). When Qsi was summed across the snow-free period, its distribution pattern became complex and dependent on the altitude and the timing of snowmelt varying across Greenland. However, despite the heterogeneous spatial distribution in summed Qsi, the expected north-south gradient, i.e. decreasing summed Qsi with increasing latitude, remained visible in the lowland areas (below 100 m a.s.l.).

Paper V

Figure 3 a: Average percentage of the year, where the ground was snow-free (SWE depth = 0.0 m) during the period 1979- 2015. b: Average timing of snowmelt, i.e., the snow-free day of year (DOY) where SWE depth equaled 0.0 m during the period 1979-2015. c: Average total liquid precipitation (rain) (mm) during snow-free days (SWE depth = 0.0 m) derived from MicroMet outputs during the period 1979-2015. d: Average of the annual pre-melt SWE depth during the period 1979-2015.

Paper V

Regions of contrasting climate Regional differences in air temperature and precipitation within ice-free Greenland have previously been documented using the long-term DMI weather station data (Box 2002, Mernild et al. 2014, Mernild et al. 2015). However, the mapped climate variables presented in Figure 3 argued that the distribution of potential vegetation drivers was far more complex than simply differences between N- and S-Greenland or between the east and west coasts. These complex maps were used to depict five Greenland regions, which together spanned the range in contrasting climates found within ice-free Greenland. These five regions were named the NW-, W-, SW-, E- and NE-Greenland regions and were located across the study domain (Figure 4, Right panel). From the regional averages, the SW (NE) region was identified as the warmest (coldest) (Figure 4b), and the region receiving the largest (smallest) amount of irradiance (Figure 4c), rain precipitation (Figure 4d), but also had the earliest (latest) timing of snowmelt (Figure 4f) and the longest (shortest) snow-free period (Figure 4a). Whereas the SW and E regions were the most snow-rich regions, the NW and NE regions were snow-poorest (Figure 4e). Also the W and E region showed pronounced contrasts in precipitation. On average, the amount of precipitation falling as rain was largest in the W region compared to the E region, whereas the proportion of precipitation falling as snow was largest in the E region compared to the W region. The NW region was also a relatively dry region compared to the other five regions but, on average, it was as warm as the E region, which is located six degrees latitude further south, and despite the snow-free period in the NW region was shorter than in the E region.

Region-specific multiple regression analysis [to be completed] The five regions had different seasonal snow characteristics and meteorology during the period 1979-2015 (Figure 4). We will extract the average climate variables from distributed points across each region and use these data to identify and quantify the region’s most important drivers of the MaxNDVI and timing of MaxNDVI patterns. Examples of estimated MaxNDVI and timing of MaxNDVI for the NW - and SW regions (Figure 5 and Figure 6) illustrate the assumed differences in vegetation greenness as a result of the differing climate that the two regions have been exposed to through time (Figure 4). In general, we found higher MaxNDVI levels and earlier MaxNDVI timing in the lowlands (below 100m a.s.l.) in the relatively warm and precipitation-rich SW region than in the relatively colder and dryer NW region. We will quantify the individual climate variables’ explanatory power in relation to MaxNDVI and timing of MaxNDVI using multiple regression analyses.

Perspectives The mapped climate variables emphasize the contrasting and diverse growing conditions for the sub-, low-, and high-Arctic vegetation types found across the complex terrain within ice-free Greenland (Figure 3). We expect to derive the most important drivers of maximum vegetation greenness and its timing from these spatially distributed data. Hereafter, an improvement and refinement of the driving variables is expected to follow, e.g., by combining the currently chosen variables to more specialized, ecologically relevant variables. These could be, e.g., measures of the below-snowpack thermal conditions during winter, which may control MaxNDVI and timing of MaxNDVI through winter soil processes of decomposition and nutrient cycling. Another example of a more specialized variable includes converting the incoming shortwave radiation, i.e., global solar radiation, to photosynthetically active radiation (PAR), which is the part of the incoming solar radiation used by the plants for photosynthesis. Hereafter will follow a discussion of the differences in the identified and quantified most-important climate variables driving MaxNDVI and timing of MaxNDVI between regions across ice-free Greenland in order to place these in an Arctic-wide context.

Paper V

Figure 4 Left panel: Regional averages inferred from SnowModel outputs of (a) snow-free fraction of the year, i.e., the percentage of the year where the ground was snow-free (%), (b) snow-free Tair sum, i.e., annually summed positive air temperatures during snow-free days (SWE depth = 0.0) (°C), (c) snow-free Qsi sum, i.e., the annually summed incoming shortwave radiation during snow-free days (kW m-2), (d) precipitation (rain) sum, i.e., the annually summed liquid precipitation (mm), i.e. rain, during snow-free days), and (e) pre-melt SWE depth, i.e., the maximum SWE depth before snowmelt occurred (m), and (f) timing of snowmelt, i.e., average day of year where SWE depth equaled 0.0 m (DOY), for the five regions (right panel) in NW-, W-, SW-, E-, and NE-Greenland during the period 1979-2015.

Paper V

Figure 5 Maximum Normalized Differenced Vegetation Index (MaxNDVI) and the timing of MaxNDVI (DOY) estimated from daily MODIS imagery (MOD09GQ) at 300-m spatial resolution for a NW-Greenland region.

Paper V

Figure 6 MaxNDVI and the timing of MaxNDVI (DOY) estimated from daily MODIS imagery (MOD09GQ) at 300-m spatial resolution for a SW-Greenland region.

Paper V

References Billings, W. D. 1952. The environmental complex in relation to plant growth and distribution. The Quarterly Review of Biology 27:251-265. Box, J. E. 2002. Survey of Greenland instrumental temperature records: 1873–2001. International Journal of Climatology 22:1829-1847. Cappelen, J. 2014. Weather observations from Greenland 1958-2014 - Observation data with description. Danish Meteorological Institute. Cappelen, J. 2016. Guide to Climate Data and Information from the Danish Meteorological Institute. Updated July 2016. Danish Meteorological Institute. Cooper, E. J. 2014. Warmer Shorter Winters Disrupt Arctic Terrestrial Ecosystems. Annual Review of Ecology, Evolution, and Systematics 45:271-295. Dee, D. P., S. M. Uppala, A. J. Simmons, P. Berrisford, P. Poli, S. Kobayashi, U. Andrae, M. A. Balmaseda, G. Balsamo, P. Bauer, P. Bechtold, A. C. M. Beljaars, L. van de Berg, J. Bidlot, N. Bormann, C. Delsol, R. Dragani, M. Fuentes, A. J. Geer, L. Haimberger, S. B. Healy, H. Hersbach, E. V. Hólm, L. Isaksen, P. Kållberg, M. Köhler, M. Matricardi, A. P. McNally, B. M. Monge-Sanz, J. J. Morcrette, B. K. Park, C. Peubey, P. de Rosnay, C. Tavolato, J. N. Thépaut, and F. Vitart. 2011. The ERA-Interim reanalysis: configuration and performance of the data assimilation system. Quarterly Journal of the Royal Meteorological Society 137:553- 597. Evans, B. M., D. A. Walker, C. S. Benson, E. A. Nordstrand, and G. W. Petersen. 1989. Spatial Interrelationships between Terrain, Snow Distribution and Vegetation Patterns at an Arctic Foothills Site in Alaska. Holarctic Ecology 12:270-278. Goodrich, L. 1982. The influence of snow cover on the ground thermal regime. Canadian geotechnical journal 19:421- 432. Havström, M., T. V. Callaghan, and S. Jonasson. 1993. Differential Growth Responses of Cassiope tetragona, an Arctic Dwarf-Shrub, to Environmental Perturbations among Three Contrasting High- and Subarctic Sites. Oikos 66:389-402. Howat, I. M., A. Negrete, and B. E. Smith. 2014. The Greenland Ice Mapping Project (GIMP) land classification and surface elevation data sets. The Cryosphere 8:1509-1518. Johansson, M., T. V. Callaghan, J. Bosiö, H. J. Åkerman, M. Jackowicz-Korczynski, and T. R. Christensen. 2013. Rapid responses of permafrost and vegetation to experimentally increased snow cover in sub-arctic Sweden. Environmental Research Letters 8:035025. Jones, H. G. 1999. The ecology of snow-covered systems: a brief overview of nutrient cycling and life in the cold. Hydrological Processes 13:2135-2147. Liston, G. E. and K. Elder. 2006a. A distributed snow-evolution modeling system (SnowModel). Journal of Hydrometeorology 7:1259-1276. Liston, G. E. and K. Elder. 2006b. A meteorological distribution system for high-resolution terrestrial modeling (MicroMet). Journal of Hydrometeorology 7:217-234. Liston, G. E. and C. A. Hiemstra. 2008. A Simple Data Assimilation System for Complex Snow Distributions (SnowAssim). Journal of Hydrometeorology 9:989-1004. Liston, G. E., J. P. McFadden, M. Sturm, and R. A. Pielke. 2002. Modelled changes in arctic tundra snow, energy and moisture fluxes due to increased shrubs. Global Change Biology 8:17-32. Mernild, S. H., E. Hanna, J. R. McConnell, M. Sigl, A. P. Beckerman, J. C. Yde, J. Cappelen, J. K. Malmros, and K. Steffen. 2015. Greenland precipitation trends in a long-term instrumental climate context (1890–2012): evaluation of coastal and ice core records. International Journal of Climatology 35:303-320. Mernild, S. H., E. Hanna, J. C. Yde, J. Cappelen, and J. K. Malmros. 2014. Coastal Greenland air temperature extremes and trends 1890-2010: annual and monthly analysis. International Journal of Climatology 34:1472-1487. Pedersen, S. H., G. E. Liston, M. P. Tamstorf, A. Westergaard-Nielsen, and N. M. Schmidt. 2015. Quantifying episodic snowmelt events in Arctic ecosystems. Ecosystems 18:839-856. Pudas, E., A. Tolvanen, J. Poikolainen, T. Sukuvaara, and E. Kubin. 2008. Timing of plant phenophases in Finnish Lapland in 1997-2006. Boreal environment research 13:31-43. Sturm, M., J. Holmgren, M. Konig, and K. Morris. 1997. The thermal conductivity of seasonal snow. Journal of Glaciology 43:26-41. Vermote, E. F., J. C. Roger, and J. P. Ray. 2016. MODIS Surface Reflectance User’s Guide Collection 6 Version 1.4. MODIS Land Surface Reflectance Science Computing Facility Greenbelt, Maryland, USA. Walker, D. A., J. C. Halfpenny, M. D. Walker, and C. A. Wessman. 1993. Long-Term Studies of Snow-Vegetation Interactions. BioScience 43:287-301.

Paper V

Wookey, P. A., A. N. Parsons, J. M. Welker, J. A. Potter, T. V. Callaghan, J. A. Lee, and M. C. Press. 1993. Comparative Responses of Phenology and Reproductive Development to Simulated Environmental Change in Sub-Arctic and High Arctic Plants. Oikos 67:490-502.

Paper VI

Future permafrost conditions along environmental gradients in Zackenberg, Greenland.

Westermann, S., Elberling, B., Pedersen, S. H., Stendel, M., Hansen, B. U., and Liston, G. E. (2015). The Cryosphere. Vol. 9, No.2, pp. 719-735. DOI: 10.5194/tc-9-719-2015.

Zackenberg, NE-Greenland (2012)

The Cryosphere, 9, 719–735, 2015 www.the-cryosphere.net/9/719/2015/ doi:10.5194/tc-9-719-2015 © Author(s) 2015. CC Attribution 3.0 License.

Future permafrost conditions along environmental gradients in Zackenberg, Greenland

S. Westermann1,2, B. Elberling1, S. Højlund Pedersen3, M. Stendel4, B. U. Hansen1, and G. E. Liston5 1Center for Permafrost (CENPERM), Department of Geosciences and Natural Resource Management, University of Copenhagen, Øster Voldgade 10, 1350 Copenhagen K., Denmark 2Department of Geosciences, University of Oslo, P.O. Box 1047, Blindern, 0316 Oslo, Norway 3Department of Bioscience – Arctic Research Centre, Aarhus University, Frederiksborgvej 399, 4000 Roskilde, Denmark 4Danish Climate Centre – Danish Meteorological Institute, Lyngbyvej 100, 2100 Copenhagen, Denmark 5Cooperative Institute for Research in the Atmosphere, Colorado State University, 1375 Campus Delivery, Fort Collins, CO 80523-1375, USA

Correspondence to: B. Elberling ([email protected])

Received: 14 May 2014 – Published in The Cryosphere Discuss.: 16 July 2014 Revised: 25 March 2015 – Accepted: 26 March 2015 – Published: 17 April 2015

Abstract. The future development of ground temperatures in 1 Introduction permafrost areas is determined by a number of factors varying on different spatial and temporal scales. For sound projections of impacts of permafrost thaw, scaling procedures The stability and degradation of permafrost areas are exten- are of paramount importance. We present numerical simula- sively discussed regarding future climate changes as poten- tions of present and future ground temperatures at 10 m res- tially important source of greenhouse gases (Schuur et al., olution for a 4 km long transect across the lower Zackenberg 2008, 2009; Elberling et al., 2010, 2013), infrastructure sta- valley in northeast Greenland. The results are based on step- bility (Wang et al., 2003, 2006) and farming potential (Mick wise downscaling of future projections derived from general and Johnson, 1954; Merzlaya et al., 2008). Depending on the circulation model using observational data, snow redistribu- emission scenario, future projections based on coarse-scale tion modeling, remote sensing data and a ground thermal general circulation models (GCMs) suggest a loss of 30 to model. A comparison to in situ measurements of thaw depths 70 % of the current permafrost extent by 2100, in conjunction at two CALM sites and near-surface ground temperatures with a significant deepening of the active layer in the remain- at 17 sites suggests agreement within 0.10 m for the maxi- ing areas (Lawrence et al., 2012). However, such projections mum thaw depth and 1 ◦C for annual average ground tem- are based on the modeled evolution of coarse-scale grid cells perature. Until 2100, modeled ground temperatures at 10 m which may not represent significantly smaller variability of depth warm by about 5 ◦C and the active layer thickness in- environmental factors governing the thermal regime typical creases by about 30 %, in conjunction with a warming of av- for many permafrost landscapes. Hence, a detailed impact as- erage near-surface summer soil temperatures by 2 ◦C. While sessment of the thermal regime remains problematic, which ground temperatures at 10 m depth remain below 0 ◦C until precludes sound projections of future greenhouse gas emis- 2100 in all model grid cells, positive annual average temper- sions from permafrost areas. atures are modeled at 1 m depth for a few years and grid cells Regional climate models (RCMs) facilitate downscaling at the end of this century. The ensemble of all 10m model of GCM output to scales of several kilometers so that, grid cells highlights the significant spatial variability of the for example, regional precipitation patterns and topography- ground thermal regime which is not accessible in traditional induced temperature gradients are much better reproduced. coarse-scale modeling approaches. Based on RCM output, projections of the future ground thermal regime have been performed for a number of permafrost regions, e.g., northeast Siberia (50 km resolution,

Published by Copernicus Publications on behalf of the European Geosciences Union. 720 S. Westermann et al.: Permafrost in northeast Greenland

Stendel et al., 2007), Greenland (25 km resolution, Daanen tions of the ground thermal regime at the Zackenberg per- et al., 2011) and Alaska (2 km resolution, Jafarov et al., mafrost observatory in northeast Greenland (Meltofte et al., 2012). While this constitutes a major improvement, many 2008) until 2100. MicroMet/SnowModel is employed as part processes governing the ground thermal regime vary strongly of a sequential downscaling procedure, including the RCM at even smaller spatial scales so that the connection between HIRHAM5 (Christensen et al., 1996) and the ground thermal model results and ground observations is questionable. In model CryoGrid 2 (Westermann et al., 2013). With a spatial high-Arctic and mountain permafrost areas exposed to strong resolution of 10 m, the effect of snow distribution patterns winds, redistribution of blowing snow can create a pattern of and different subsurface and surface properties on ground strongly different snow depths on distances of a few meters. temperatures can be accounted for. The study aims to fill the Since snow is an effective insulator between ground and at- gap between the coarse- and the point-scale modeling studies mosphere (Goodrich, 1982), a distribution of ground temper- on the future ground thermal regime which are available for atures with a range between average maximum and minimum the Zackenberg valley so far. The 25 km scale, Greenland- temperatures of 5 ◦C and more is created (e.g., Gisnås et al., wide assessment of Daanen et al.(2011) puts Zackenberg 2014), which is of a similar order of magnitude to the pro- in the zone of “high thaw potential” until the end century, jected increase of near-surface air temperatures in many po- with modeled ground temperatures of −5 to −2.5 ◦C and an lar areas. Consequently, the susceptibility to climate change active layer thickness of 0.5 to 0.75 m for the period 2065– can display a dramatic variability on local scales and per- 2075. However, the detailed point-scale study by Hollesen mafrost degradation can occur significantly earlier in parts of et al.(2011) suggests a future active layer thickness of 0.8 to a landscape than suggested by coarse-scale modeling. Fur- 1.05 m for a site with average soil moisture conditions which thermore, the thermal properties and cryostratigraphy of the are not representative of many other sites found in the Za- ground can be highly variable as a result of geomorphol- ckenberg valley, such as the wetlands. Extending this ear- ogy, vegetation and hydrological pathways, with profound lier work, we present simulations for a 4 km transect cutting implications for the thermal inertia and thus the dynamics across typical vegetation zones in the lower parts of Zack- of permafrost degradation. In a modeling study for south- enberg valley which allow estimating the range of ground ern Norway, Westermann et al.(2013) highlight that near- thermal conditions that could be encountered until the end of surface permafrost in bedrock areas disappears within a few the century. years after the climatic forcing crosses the thawing threshold, while near-surface permafrost is conserved for more than 2 decades in areas with high organic and ground ice contents 2 The Zackenberg site and/or a dry, insulating surface layer. In addition, the soil carbon content in Arctic landscapes is unevenly distributed Zackenberg is located in northeast Greenland at 74◦300 N, (Hugelius et al., 2013), and greenhouse gas emissions from 20◦300 W (Fig.1). Zackenberg valley is a wide lowland val- localized carbon-rich hotspots can contribute a significant ley dominated by Quaternary non-calcareous sediments with part to the landscape signal (e.g., Walter et al., 2006; Mas- significant periglacial activity and continuous permafrost tepanov et al., 2008). Therefore, both the carbon stocks and (Elberling et al., 2004, 2008), with a mean annual air temper- the physical processes governing permafrost evolution must ature of −9.5 ◦C (1996–2007) according to Elberling et al. be understood at the appropriate spatial scales to facilitate (2010). Maximum active layer thickness varies from 40 cm improved predictions of the permafrost–carbon feedback. to more than 2 m and has increased significantly by 0.8 cm In recent years, modeling schemes capable of comput- to 1.5 cm per year between 1996 and 2012 (Elberling et al., ing the ground thermal regime at significantly higher spa- 2013), which has been determined at two sites (denoted Zero- tial resolutions of 10 to 30 m have been developed and ap- Calm 1 and 2, Fig.1) of the Circumpolar Active Layer Mon- plied in complex permafrost landscapes (e.g., Zhang, 2013; itoring (CALM) program (Brown et al., 2000). Zhang et al., 2012, 2013; Fiddes and Gruber, 2012, 2014; From the hilltops towards the depressions, an increase in Fiddes et al., 2015). These approaches can capture small- soil water content is seen from dry to wet conditions at the scale differences in altitude, aspect and exposition, as well foot of the slopes due to snowmelt water being released dur- as in surface and subsurface properties, but the redistribution ing large parts of the summer. Roughly one-third of the low- of snow through wind drift is only included in a simplified land area in Zackenberg is poorly drained. Given the low way through precipitation correction factors (Fiddes et al., summer precipitation, water availability during the growing 2015; Zhang et al., 2012). On the other hand, dedicated snow season is mainly controlled by the location of large snow redistribution models of various levels of complexity exist patches melting during the growing season, resulting in the (e.g., Winstral et al., 2002; Lehning et al., 2006) with which distinct vegetation zonation around these. the pattern and evolution of snow depths can be simulated. The topography, landscape forms and wind direction are In this study we make use of such an approach, the deter- the main factors controlling both water drainage and snow ministic snow modeling system MicroMet/SnowModel (Lis- distribution. These patterns are found on both a landscape ton and Elder, 2006a,b), to achieve high-resolution simula- scale and a small scale (100–200 m) and can therefore be il-

The Cryosphere, 9, 719–735, 2015 www.the-cryosphere.net/9/719/2015/ S. Westermann et al.: Permafrost in northeast Greenland 721

Figure 1. Left: location of the Zackenberg site and ZERO-line in Greenland. Right: NDVI image (derived from a multi-spectral Quickbird 2 image from 7 July 2011) of the modeled part of ZERO-line, with the CALM sites ZeroCalm (ZC) 1 and 2 and the locations of in situ measurements of ground temperatures Tground at different depths, as employed in Sect. 4.1. Two additional in situ measurements of ground temperatures at shallow depths are located approx. 0.5 km NE and SW of the displayed scene. Coordinates are in UTM zone 27; note that ZERO-line continues further NE to the top of Aucellabjerg. lustrated conceptually as a transect across typical landscape For monitoring purposes, an 8km transect cutting across forms in the valley from hilltops to depressions. The top of the main ecological zones of the Zackenberg valley from the hills are windblown and exposed throughout the year with sea level to 1040 m a.s.l. at the summit of Aucellabjerg has little or no accumulation of snow. From the hilltops towards been established, which is considered representative of the the depressions there is an increase in soil water content from Zackenberg valley (Fredskild and Mogensen, 1997; Meltofte dry conditions (even arid conditions and salt accumulation at et al., 2008). Along this so-called ZERO (“Zackenberg eco- the soil surface) at the hilltops to wet conditions in the bottom logical research operations”) line (Fig.1), changes in species of the depression. The dominant wind pattern during winter composition and distribution of plant communities are in- leaves large snow patches on the south-facing slopes ensur- vestigated regularly. In this study, we focus on lower 4 km ing high surface and soil water contents during a large part of ZERO-line from the coast to an elevation of 200 m a.s.l., of the growing season. which is characterized by a strong variability as exemplified Bay(1998) described and classified the plant communi- by the normalized difference vegetation index (NDVI) values ties in the central part of the Zackenberg valley and mapped (Fig.1). their distribution. The vegetation zones range from fens in the depressions to fell-fields and boulder areas towards the hill tops. East of the river Zackenbergelven the lowland is domi- 3 Modeling tools nated by Cassiope tetragona heaths mixed with Salix arctica snow beds, grasslands and fens; the latter occurring in the In order to determine the spatial variability of ground tem- wet, low-lying depressions, often surrounded by grassland. peratures in the Zackenberg valley, simulations from 1960 to On the transition from the lowland to the slopes of Aucellab- 2100 are performed for grid cells of 10 m resolution for the jerg (50–100 m a.s.l.), the vegetation is dominated by grass- lower 4 km of ZERO-line (in total 437 grid cells). In addition, land. Between 150 and 300 m a.s.l., open heaths of mountain the 100m × 100m large CALM sites ZeroCalm 1 and 2 are avens, Dryas sp., dominate and gradually the vegetation be- simulated (Fig.1, in total 200 grid cells). To compile forcing comes more open with increasing altitude towards the fell- data sets at such high resolution, a multi-step downscaling fields with a sparse plant cover of Salix arctica and Dryas sp. procedure is employed which is schematically depicted in Grassland, rich in vascular plant species and mosses, occurs Fig.2. It is designed to account for the spatial variability of along the wet stripes from the snow patches in the highland snow depths, differences in summer surface temperature (due (250–600 m a.s.l.). to, e.g., different evapotranspiration rates caused by surface soil moisture and land cover) and spatially variable ground www.the-cryosphere.net/9/719/2015/ The Cryosphere, 9, 719–735, 2015 722 S. Westermann et al.: Permafrost in northeast Greenland

0.25 Wm−1 K−1 (e.g., Côté and Konrad, 2005) for peat is employed. The latent heat from freezing soil water or melting ice is accounted for in terms of an effective heat capacity ceff [Jm−3 K−1], which increases strongly in the temperature range in which latent heat effects occur. This curve is determined by the soil freezing characteristic, i.e., the function linking the soil water content to temperature, which is related to the hydraulic properties of the soil in CryoGrid 2 (Dall’Amico et al., 2011) for three soil classes: sand, silt and clay. To account for the buildup and disappearance of the snow cover, the position of the upper boundary is allowed to change dynamically by adding or removing grid cells. Move- ment of soil water is not accounted for so that the sum of the soil water and ice contents are constant in CryoGrid 2. For spatially distributed modeling, the target domain is decom- Figure 2. Schematic workflow of the modeling scheme depicting posed in independent grid cells, each featuring a set of model field data (green), remote sensing data (red), models (blue) and the parameters. principal forcing data (yellow) for the thermal model CryoGrid 2, delivering spatially resolved fields of ground temperatures. See text. 3.1.1 Model initialization The initial temperature profile for each grid cell is obtained thermal properties and water/ice contents. Differences in in- by a multi-step initialization procedure which allows us to solation due to exposition and aspect are not accounted for, approximate steady-state conditions in equilibrium with the which is acceptable for the gentle topography (average slope climate forcing for the first model decade (September 1958– 2.8◦) in the modeled part of ZERO-line. The different parts August 1968) in a computationally efficient way. The method of the scheme and their interplay are described as follows. which is described in more detail in Westermann et al.(2013) accounts for the insulating effect of the seasonal snow cover 3.1 The permafrost model CryoGrid 2 as well as the thermal offset (Osterkamp and Romanovsky, 1999). CryoGrid 2 is a one-dimensional, physically based thermal 3.1.2 Driving data sets subsurface model driven by time series of near-surface air temperature and snow depth and has been recently employed As driving data sets for CryoGrid 2 we use gridded data sets to assess the evolution of permafrost extent and temperatures of daily average air temperature and snow depth obtained in southern Norway (Westermann et al., 2013). The physical from a downscaling scheme and a snow redistribution model basis and operational details of CryoGrid 2 are documented (Sects. 3.3, 3.4). To account for differences in surface soil in Westermann et al.(2013), and only a brief overview over moisture between grid cells, which give rise to spatially dif- the model properties is given here. CryoGrid 2 numerically ferent surface temperatures, we employ the empirical con- solves Fourier’s law of conductive heat transfer in the ground cept of n factors which relate average air temperature Tair to to determine the evolution of ground temperature T [K] over surface temperature Ts by Ts = nt Tair: time t, ◦ Tair for Tair ≤ 0 C Ts = ◦ (2) ∂T ∂ ∂T nt Tair for Tair > 0 C. ceff(z,T ) − k(z,T ) = 0, (1) ∂t ∂z ∂z This rough treatment of summer surface temperatures (which has been applied in previous modeling studies, e.g., Hipp with the thermal conductivity k [Wm−1 K−1] being a func- et al., 2012) is focused on seasonal averages and can not re- tion of the volumetric fractions and thermal conductivities of produce surface temperatures on shorter timescales, e.g., the the constituents water, ice, air, mineral and organic (West- daily cycle. As a result, a comparison of temperatures in ermann et al., 2013) following the formulation of Cosenza upper soil layers is less meaningful than for deeper layers, et al.(2003). For the thermal conductivity of the mineral frac- which are only influenced by seasonal or even multi-annual tion of the soil, we assume 3.0 Wm−1 K−1, which is a typ- average temperatures. However, the n factor-based approach ical value for sedimentary and metamorphic rock with low precludes the need to compute the surface energy balance quartz content (Clauser and Huenges, 1995), as dominant and allows employing measured historic time series of air in most parts of the Zackenberg valley (Koch and Haller, temperatures (such as the one from Daneborg, Sect. 3.4) for 1965). For the organic soil fraction, the standard value of ground thermal modeling.

The Cryosphere, 9, 719–735, 2015 www.the-cryosphere.net/9/719/2015/ S. Westermann et al.: Permafrost in northeast Greenland 723

3.1.3 Ground properties

Based on a NDVI-classification, six ecosystems were identified in Zackenberg valley (Bay, 1998; Tamstorf et al., 2007; Ellebjerg et al., 2008). Areas with NVDI < 0.2 are dominated by fell-field with a sparse vegetation. In the high mountains such areas are found on solifluction soils, patterned ground and rocky ravines. Dryas heath dominates areas with NDVI between 0.2 and 0.3. Fell-field and Dryas heath are both situated at exposed plateaus, where snow often blows off during the winter months causing thinner snow cover. Here, plant species experience an early snowmelt and hence an early start of the growing season. Cassiope heath (NDVI between 0.3 and 0.4) depends on a protective snow cover during winter and occurs mainly in the lowland on gentle slopes facing south and leeward from the northerly winds which Figure 3. Summer nt factor vs. NDVI based on in situ measure- dominate the winter period (Hansen et al., 2008). Salix snow ments from Zackenberg and Kobbefjord in Greenland and from beds feature NDVI values between 0.4 and 0.5. This ecosys- northern Alaska (Klene et al., 2001; Walker et al., 2003). The black tem, which is unique to eastern Greenland, occurs mostly on 2 line represents the fit following Eq. (3), R = 0.97. sloping terrain, often below the Cassiope heath belt on the slopes, where the snow cover is long lasting so that the soil moisture in the Salix snow-bed areas are higher. In the wet- The summertime n factor n is computed according to the t land areas with NDVI higher than 0.5, grassland and fen ar- NDVI of each grid cell (at the maximum of the growing sea- eas are distinguished. Grassland occurs mostly on slightly son) using sloping terrain with an adequate supply of water early in the 2 season, while the soil water regime can change from wet to nt = 2.42NDVI − 3.01NDVI + 1.54. (3) moist later in the season. The fen areas occur on flat terrain

The relationship is compiled with nt as the ratio of degree- in the lowland, where the soil is permanently water-saturated day sums at the soil surface to those in the air over the sum- throughout the growing season. In August 2013, a classifica- mer season at both Zackenberg (74.5◦ N) and Kobbefjord tion of ecosystem classes according to the dominating plant (65.6◦ N), close to Nuuk in western Greenland. Figure3 also species and qualitative surface moisture conditions was con- shows a strong correlation between nt values (Klene et al., ducted along the modeled part of ZERO-line at spatial reso- 2001) and NDVI values (Walker et al., 2003) from the Ku- lution of 10 m, which resulted in 5 % fell, 20 % Dryas, 35 % paruk River basin, Alaska, USA, with an R2 value of 0.97 for Cassiope, 15 % Salix snow bed and 25 % wetland (fen and the combined data set. Summer nt factors above 1 indicate grassland areas were not distinguished). that the soil-surface temperatures are warmer than air tem- Using satellite-derived NDVI values (see previous sec- peratures; this mostly occurs on nearly barren mineral soils. tion), these ecosystem fractions could be well reproduced The minimum nt values of approx. 0.65 are found in moist for fell (9 %), Dryas (22 %) and Cassiope (39 %), while a fen areas, indicating a strong cooling effect during the sum- strong discrepancy was encountered for the Salix and wet- mer on the mineral soils of these sites. land classes. Therefore, Salix snow bed was merged with For each 10 m model grid cell, an NDVI value was de- wetland, yielding a wetland fraction of 30 %. The “true” termined from a 2.5 m multi-spectral Quickbird 2 image of Salix class is hereby split between Cassiope and wetland, the Zackenberg area acquired around noon local time on 7 which is reflected in the strong concentration of grid cells July 2011 (Fig.1). Whereas the acquisition date is close to with NDVI values around 0.4. This suggests a significant the annual maximum NDVI values, it represents a single overlap of the NDVI values from the different classes in this point in the time, and there is strong seasonal and interan- region for the particular satellite acquisition date, so that the nual variability in plant growth and consequent evolution of classes can not be separated by their NDVI value. While the NDVI values (Tamstorf et al., 2007). While this error source NDVI-derived ecosystem classification constitutes a poten- is hard to quantify, the general agreement in the coverage of tially important source of uncertainty in the modeling chain, the different vegetation classes (see next section) with field it provides the possibility to use satellite images and thus ap- observations suggests that the satellite image is an adequate ply the classification procedure for larger regions, e.g., the basis to capture the pattern of surface soil moisture and sum- entire Zackenberg valley, at high spatial resolutions, which mer surface temperatures along ZERO-line. can hardly be achieved by manual mapping. For the remaining four classes fell, Dryas, Cassiope and wetland, typical soil stratigraphies were assigned based on www.the-cryosphere.net/9/719/2015/ The Cryosphere, 9, 719–735, 2015 724 S. Westermann et al.: Permafrost in northeast Greenland

Table 1. Sediment stratigraphies in CryoGrid 2 with volumetric situ measurements, a snow density of 300 kgm−3 is em- fractions of the soil constituents and soil type for each layer given. ployed, which results in a volumetric heat capacity of csnow = 0.65MJm−3 K−1. In the absence of in situ measurements of Depth (m) Water/ice Mineral Organic Air Type the thermal conductivity of the snow cover, we use the em- Fell pirical relationship between density and thermal conductivity from Yen(1981), which is also employed in the detailed 0–3 0.05 0.6 0.0 0.35 sand snowpack scheme CROCUS (Vionnet et al., 2012). The re- 3–10 0.4 0.6 0.0 0.0 sand −1 −1 sulting value is ksnow = 0.25Wm K , slightly lower than > 10 0.03 0.97 0.0 0.0 sand those employed in CryoGrid 2 simulations for the mountain Dryas environments of southern Norway where average winter tem- 0–1 0.15 0.55 0.0 0.3 sand peratures are higher than in Zackenberg, but predominantly 1–10 0.4 0.6 0.0 0.0 sand wind-packed snow is encountered as well. > 10 0.03 0.97 0.0 0.0 sand Cassiope heath 3.2 Future climate scenario with HIRHAM 0–0.8 0.25 0.55 0.0 0.2 sand 0.8–10 0.4 0.6 0.0 0.0 sand There are several types of uncertainties related to climate > 10 0.03 0.97 0.0 0.0 sand projections. Apart from “external” uncertainties such as the Wetland future evolution of greenhouse gas emissions, there are also 0–0.6 0.5 0.45 0.05 0.0 silt “internal” uncertainties related to different parameterizations 0.6–10 0.4 0.6 0.0 0.0 silt of subgrid-scale processes. Even though it is possible to > 10 0.03 0.97 0.0 0.0 sand model the distribution of permafrost on rather coarse scales (Stendel and Christensen, 2002), it is desirable to use a GCM with as high resolution as possible, which serves as the basis for downscaling to the target grid of a RCM driven with these and guided by in situ measurements in soil samples (Ta- fields. ble1). The stratigraphies are designed to represent the char- The climate model EC-EARTH (v2.3) is such a GCM. It acteristics of the different ecosystem classes at least in a consists of the Integrated Forecast System (IFS) developed at semi-quantitative way: from fell to wetland, the water con- the European Centre for Medium-Range Weather Forecasts tents in the active layer increase from dry to saturated con- (ECMWF) as the atmospheric component, the Nucleus for ditions, while the soil texture changes from coarse to more European Modelling of the Ocean (NEMO) version 2 as the fine-grained in conjunction with increasing porosity. The ab- ocean component and the Louvain-la-Neuve sea ice model solute values are derived from soil samples taken at depths (LIM2). These components are coupled using the OASIS3 between 0 and 0.5 m in the different classes mainly in July coupler (Hazeleger et al., 2010, 2012). The IFS in the cur- 2006 and 2007. For wetland and Cassiope, the average of all rent EC-EARTH model is based on ECMWF cycle 31r1 with values yielded volumetric water contents of 0.52 and 0.28, some improvements from later cycles implemented, includ- respectively. Furthermore, transient simulations of the one- ing a new convection scheme and a new land surface scheme dimensional water balance and ground thermal regime with (H-TESSEL) as well as a new snow scheme (Hazeleger et al., the COUP model suggest average soil water contents be- 2012). The atmospheric part of EC-EARTH is configured tween 0.2 and 0.3 for the active layer at a Cassiope site with a horizontal spectral truncation of T159, which is ap- (Hollesen et al., 2011). For the Dryas and fell classes, large proximately 125km × 125km in latitude and longitude. The changes in soil moisture were encountered after rain falls vertical resolution is 62 layers. The ocean and sea ice com- which made the values strongly dependent on the timing of ponents have 42 vertical layers and a roughly 1 ◦ horizontal the sampling. The volumetric organic material contents are resolution with refinement to 1/3◦ around the equator. EC- low in all classes (5 % or less) and have negligible influence Earth is one of the models of CMIP5 (Coupled Model Inter- on the thermal properties of the soil. Following measure- comparison Project) and has been used for the experiments ments of soil cores to 2 m depth (Elberling et al., 2010), sat- for the IPCC AR5 report. urated conditions are assumed below the current active layer To resolve the topography of Greenland adequately, a hor- for all classes (Table1), except for fell for which no in situ izontal resolution of 5 km or finer is required (Lucas-Picher data are available and saturated conditions are assumed be- et al., 2012). The output of EC-Earth is therefore downscaled low a depth of 3m. Furthermore, bedrock is assumed below to the RCM grid. The RCM used here is HIRHAM5 in its 10 m, which is a pure estimate but has limited influence on newest version, which includes calculation of the surface the outcome of the simulations. mass balance of the Greenland Ice Sheet. A surface snow Snow properties: in CryoGrid 2, constant thermal prop- scheme has been implemented over glaciers. The model erties in space and time are assumed for the snow cover setup is described in Rae et al.(2012) except that the res- (see Westermann et al., 2013, for details). Following in olution here is 0.05◦ (5.5 km) instead of 0.25◦ (27 km), as in

The Cryosphere, 9, 719–735, 2015 www.the-cryosphere.net/9/719/2015/ S. Westermann et al.: Permafrost in northeast Greenland 725

Langen et al.(2015). EC-Earth has a slight cold bias, prob- was based on an August 2000 aerial survey, and the land- ably caused by albedo values that are too high, so that the cover map was based on the Elberling et al.(2008) vegeta- estimates of surface mass balance under climate change con- tion classification (see Sect. 3.1 – Ground properties). From ditions are slightly higher than observed. the land-cover map, a snow-holding depth (shd) was assigned EC-Earth and HIRHAM have been run for three time to each class, i.e., the depth to which the vegetation is able to slices, namely 1991–2010, 2031–2050 and 2081–2100. The hold the snow and prevent snow transport by wind (snow ex- scenario used was RCP 4.5 (Thomson et al., 2011; Clarke ceeding this depth is available for wind redistribution). This et al., 2007; Smith and Wigley, 2006; Wise et al., 2009), snow-holding depth was set according to vegetation/canopy which gives an additional radiative forcing in 2100 with re- height but also included the micro-topographic relief within spect to preindustrial values of 4.5 Wm−2. In this rather con- a 10m × 10m grid cell. The classes “fell”, “Dryas”, “Cas- servative scenario, CO2 emissions peak around 2040 and de- siope heath” and “Wetland” were assigned a shd of 0.01, cline thereafter, resulting in a CO2 concentration of 550 ppm 0.05, 0.20 and 0.20m, respectively. The modeled mean snow in 2100, which is just below a doubling with respect to prein- depth along ZERO-line was on the order of tens of cm, while dustrial values. the modeled maximum snow depth was several meters in the winters 2003/2004–2009/2010. Both the annual mean and 3.3 Modeling snow distribution by maximum snow depth varied by a factor 1.5 from year to MicroMet/SnowModel year. The modeled mean snow depth exceeded the snow- holding depth in all vegetation classes, so that the parame- SnowModel is a spatially distributed snow-evolution mod- ter shd had minor influence on snow distributions and win- eling system (Liston and Elder, 2006a) which was ap- ter accumulation. The modeled snow depths were validated plied in the Zackenberg study area (14km × 12km) to de- against automated and manual measurements conducted at scribe the snow distribution through a 7-year period cover- the ZeroCalm sites close to the ZERO-line. Automated mea- ing August 2003 to September 2010. SnowModel consists surements of snow depth acquired at a point near ZeroCalm of three interconnected submodels: Enbal, SnowPack and 1 were compared to the model results at the closest grid SnowTran-3D. Enbal calculates surface energy exchanges cell. Linear regression analyses showed that the modeled and snowmelt (Liston, 1995, 1999), SnowPack models the snow depth represented 77–97 % of the variability in the ob- evolution of the snow depth and snow-water equivalent in served snow depth in 5 of the 7 hydrological years and ap- time and space (Liston and Hall, 1995; Liston and Mernild, proximately 47 % in 2 years (2004/2005 and 2008/2009). 2012) and SnowTran-3D generates the transport of blow- However, MicroMet/SnowModel results showed an earlier ing snow (Liston and Sturm, 1998; Liston et al., 2007). snowfall than in reality, most likely due to the monthly ap- SnowModel was coupled with a high-resolution atmospheric plied lapse rates which caused snowfall instead of rain in the model, MicroMet (Liston and Elder, 2006b), which spatially simulations. As a result, the modeled snow depths featured distributed the micrometeorological input parameters over a positive bias of on average of 0.16m (2005–2010) com- the simulation domain. MicroMet requires meteorological pared to the observed snow depths. The performance of Mi- station and/or atmospheric (re)analysis inputs of air temper- croMet/SnowModel in reproducing the spatial distribution of ature, relative humidity, precipitation, wind speed, and wind snow depths was investigated by comparing to snow depths direction. Furthermore, available observed incoming short- measured manually at one date between mid-May and mid- wave and long-wave radiation were included. All meteoro- June for the years 2005–2008 and 2010 at >150 sites within logical parameters except precipitation were measured by ZeroCalm 1 and 2. Figure4a displays the comparison of five automatic weather stations distributed in the valley and the cumulative distributions of all measurements to the mod- on mountains contained within the simulation domain (Ta- eled snow depths for the corresponding dates using all grid ble2). cells within ZeroCalm 1 and 2. The results suggest that Mi- Because of missing data and uncertainties associated with croMet/SnowModel can generally reproduce the range and in situ winter precipitation measurements, MicroMet pre- distribution of snow depths to a satisfactory extent, but some cipitation inputs were provided by the North American Re- deviations occur in particular for low and high snow depths. gional Reanalysis (NARR) (Mesinger et al., 2006). These Note that the measurements were conducted at the end of the NARR precipitation fluxes were adjusted using the SnowAs- snow season and in some years are heavily influenced by on- sim (Liston and Hiemstra, 2008) data assimilation scheme going snowmelt. under the constraint that modeled snow-water-equivalent In addition, the timing of the snowmelt was compared to depth matched observed pre-melt snow depth and snow den- in situ measurements similar as in Pedersen et al.(2015). At sity at locations where those observations were made. Addi- the automated station near ZeroCalm 1 (see above), Snow- tionally, a digital elevation model (DEM) and a land-cover Model/MicroMet represented the timing of snowmelt with map were required for the MicroMet/SnowModel simula- on average ±4 days, while the maximum deviation was 8 tions. These distributions were provided over the simula- days (Fig.4b). For ZERO-line, the modeled melt-out dates tion domain at a 10m × 10m spatial resolution. The DEM were validated by comparing them to orthorectified images www.the-cryosphere.net/9/719/2015/ The Cryosphere, 9, 719–735, 2015 8 S. Westermann et al.: Permafrost in NE Greenland 726 S. Westermann et al.: Permafrost in northeast Greenland Table 2. The five climate stations in Zackenberg used to provide MicroMet/SnowModel meteorological inputs. Table 2. The five climate stations in Zackenberg used to provide MicroMet/SnowModel meteorological inputs. station altitude Time series UTM UTM Station[ma Altitude.s.l.] Time series Easting UTM Northing UTM Main climate station(ma.s. 38l.) 1996–present 513 Easting 382 8 264Northing 743 M2Main climate station 17 38 2003–present 1996–present 513 513 058 382 8 8 264 264 019 743 M3M2 (Aucella) 410 17 2003–present 2003–present 516 513 126 058 8 8 268 264 250 019 M6M3 (Dome) (Aucella) 1283 410 2003–present 2006–2012 507 516 453 126 8 8 269 268 905 250 M7M6 (Stor (Dome) Sødal) 1283 145 2008–present 2006–2012 496 507 815 453 8 8 269 269 905 905 M7 (Stor Sødal) 145 2008–present 496 815 8 269 905

570 on(resolution a mountain 5m) slope taken at by 400 anm automatic a.s.l. overlooking camera system ZERO- lo- linecated (Hinkler on a mountain et al., 2002) slope for at400 the yearsm a.s. 2006l. overlooking to 2009. FromZERO- grayscaleline (Hinkler images, et al. the, 2002 presence) for or the absence years 2006of snow towas 2009. deter- From minedgrayscale using images, a simple the threshold presence filter, or absence which of was snow adapted was deter- for eachmined year. using In case a simple of missing threshold images filter, due which to clouds was in adapted front of for 575 theeach camera, year. In the case date of of missing the snowmelt images was due set to cloudsto the midpoint in front of betweenthe camera, thelast the date snow-covered of the snowmelt and the was first set snow-free to the midpoint date. Thebetween results the confirm last snow-covered the results from and the the comparison first snow-free to point date. observations:The results confirm in 2006, the the results deviation from theof the comparison melt-out to dates point between measurements and SnowModel/MicroMet results observations: in 2006, the deviation of the melt-out dates 580 was 0.0 8.6 days, -1.8 5.6 days in 2007, 0.7 8.2 days in between measurements and SnowModel/MicroMet results 2008 and± 5.4 6.0 days±in 2009. The melt-out date± is, there- was 0.0 ± 8.6days, −1.8 ± 5.6days in 2007, 0.7 ± 8.2days fore, represented± within one week for most grid cells, but in 2008 and 5.4 ± 6.0days in 2009. The melt-out date is, larger deviations can occur for a number of grid cells. Note therefore, represented within 1 week for most grid cells, but that cloudy periods with no images of up to four days lead larger deviations can occur for a number of grid cells. Note 585 to an uncertainty of several days in the determination of that cloudy periods with no images of up to 4 days lead to the snowmelt date for some years and pixels. Furthermore, Hinkleran uncertainty et al. (2002) of several suggest days an absolute in the referencingdetermination error ofof the aboutsnowmelt 10 m datefor foreach some pixel, years which and also pixels. contributes Furthermore, to a re-Hin- ducedkler et match al.(2002 between) suggest images an and absolute model referencing results. error of about 10m for each pixel, which also contributes to a reduced

590 3.4match Downscaling between images scheme and from model GCM results. to plot scale

To3.4 run Downscaling simulations of scheme permafrost from temperatures GCM to plot from scale 1958 to 2100, a continuous record of the driving data air tempera- tureTo run and simulations snow depth was of permafrost compiled from temperatures various sources. from 1958 The to method2100, a assumes continuous that trends record in of air the temperature driving data and air precipita tempera--

595 tionture measured and snow at depth one was point compiled or modeled from by various a medium-scale sources. The atmosphericmethod assumes scheme that are trends representative in air temperature for the trends and precipita- along ZERO-Line.tion measured at one point or modeled by a medium-scale atmospheric scheme are representative for the trends along ZERO-Line.– For the period from 2003 to 2010, a continuous record of forcing data is derived for all 10 m-grid cells from 600 – theFor output the period of MicroMet/SnowModel from 2003 to 2010, a (Sect. continuous 3.3). Thisrecord FigureFigure 4. 4. (a)a) Cumulative histogram histogram of of measured measured and and modeled modeled snow snow dataof forcing set constitutes data is derived the basis for upon all 10 which m grid statistical cells from depthsdepths at at ZeroCalm ZeroCalm 1 1 and and 2 2 for for 20 May May 20, 2005, 2005, 7 June 7,2006, 2006, 26 May May downscalingthe output of of MicroMet/SnowModel point measurements and (Sect. RCM 3.3 output). This 26, 2007, June 2, 2008, and May 16, 2010. The measurements were 2007, 2 June 2008 and 16 May 2010. The measurements were taken (Sect.data 3.2)set constitutes is performed the for basis the remaining upon which time periods. statistical taken along transects across ZeroCalm 1 and 2, and do not represent along transects across ZeroCalm 1 and 2 and do not represent the downscaling of point measurements and RCM output locationsthe locations of the of model the model grid grid cells. cells. The Thefive modeled five modeled grid grid cells cells with – To synthesize past air temperature, we employ the long- with snow depths >3.0 m feature snow depths of 3.2 m (2x), 4.0 m, (Sect. 3.2) is performed for the remaining time periods. snow depths >3.0m feature snow depths of 3.2m (2×), 4.0, 4.5 and605 term air temperature record from Daneborg (74◦18′ N, 4.5 m, and 5.4 m. b) Modeled vs. measured day of year (DOY) of 5.4m. (b) Modeled vs. measured day of year (DOY) of the termi- – 20To◦13 synthesize′ E), located past about air temperature, 25 km W of we Zackenberg, employ the long- for nationthe termination of snowmelt of snowmelt at the automated at the automated snow depth snow monitoring depth monitor- station which an hourly record is available for the periods◦ 0 ing station next to ZeroCalm 1 for the years 2004-2009. The dashed term air temperature record from Daneborg (74 18 N, next to ZeroCalm 1 for the years 2004–2009. The dashed line rep- 1958–1975◦ 0 and 1979–2011. For these periods, daily line represents the 1:1 line. 20 13 E), located about 25 km west of Zackenberg, resents the 1:1 line. meansfor which were ancalculated hourly for record each is year. available The gap for was the filled periods 1958–1975 and 1979–2011. For these periods, daily means were calculated for each year. The gap was filled

The Cryosphere, 9, 719–735, 2015 www.the-cryosphere.net/9/719/2015/ S. Westermann et al.: Permafrost in northeast Greenland 727

using random years selected from the 5 years before the temperatures colder than the MicroMet/SnowModel pe- gap for the first half and the first 5 years after the gap riod, this yields a later snowmelt, while the snow melts for its second half. In addition, a monthly trend was su- earlier for warmer conditions. perimposed on the randomly selected data, obtained by linear interpolation between the monthly averages from 5 years before and 5 years after the gap. With this pro- 4 Model results cedure, both a smooth transition between the time slices 4.1 Comparison to field data and a simulated natural variability was achieved. To build confidence that the modeling is a satisfactory rep- – For present-day and future air temperatures, the near- surface air temperature from the HIRHAM5 5 km grid resentation for the true ground thermal conditions, the model cell closest to the study area, which are available for results are compared to available in situ data sets. These com- three time slices, 1991–2010, 2031–2050 and 2081– prise in particular thaw depth measurements at ZeroCalm 1 2100. The gaps in between the time slices were filled and 2 since 1996, measurements of thaw depth along ZERO- similar to the gap in the Daneborg record. line in 2013 and measurements of ground temperatures conducted in the active layer and the permafrost between 1996 – To account for differences in the climate setting be- and 2014 at 17 sites. tween the study area and Daneborg/the HIRHAM grid cell, we calculate the offset of the average air tem- 4.1.1 Active layer thickness peratures between the Daneborg/HIRHAM records and The modeled and measured maximum thaw depths for 7 the MicroMet/SnowModel output for the period 2003– years for which MicroMet/SnowModel was run are shown in 2010, for which all time series are available simultane- Fig.5, with the areas selected for comparison equal to Elber- ously. A specific offset is calculated for each grid cell ling et al.(2013). Most importantly, CryoGrid 2 can capture and for each month of the year, thus accounting for both the significant differences between the three sediment classes the spatial gradients along Zero Line and the average Dryas, Cassiope and wetland caused by different ground and seasonal differences between the two sites. surface properties. With a few exceptions, CryoGrid 2 can – For both the past Daneborg and the future HIRHAM reproduce the measured thaw depth within the spatial vari- time series, the difference to the monthly average of ability in the validation areas (indicated by the error bars the 2003–2010 reference period (i.e., a monthly time in Fig.5), with the exception of the year 2006 which fea- series of offsets) was calculated. The final time se- tures stronger deviations from the measurements. The spa- ries was synthesized by selecting air temperatures from tial variability within the target areas is significantly smaller MicroMet/SnowModel for random years from 2003 to in the model runs than in nature, most likely since the sed- 2010 and subtracting the spatial and temporal offsets for iment classification assumes constant soil properties within each grid cell and each month, respectively. each class, while the soil composition can vary significantly within a class in reality. – Snow depths were obtained by a similar procedure. On 26 August 2013, thaw depths were measured manually Since a past record was not available and neither along the modeled part of ZERO-line at intervals of 30–40m. snow depth nor winter snowfall modeled by HIRHAM Although MicroMet/SnowModel data were not employed in showed a significant trend, the snow depth was taken the modeling of this year, a comparison to modeled data is from random years of the MicroMet/SnowModel pe- meaningful to assess the general range and distribution of riod (the same year as used for air temperatures) during thaw depths along ZERO-line. The measured and modeled the buildup period. To model past and future snowmelt distributions of thaw depths are displayed in Fig.6. Although in climate conditions different from the 2003–2010 Mi- thaw depths deeper than 1.0m could not be measured in the croMet/SnowModel period, a simple degree-day model field, the comparison shows that the modeling can generally linking melt rates to air temperature (e.g., Hock, 2003) reproduce the range of thaw depths. Furthermore, the mod- was applied. We assumed a constant melt factor of eled and measured fractions of thaw depths larger than 1.0m 2.5 mm snow water equivalent per degree day for tem- are approximately equal. All model grid cells with such large peratures exceeding −2 ◦C. The numbers were ob- thaw depths belong to the class fell, which is an indication tained by fitting the snowmelt dates delivered by Mi- that the modeling procedure is adequate also for fell. For croMet/SnowModel for the 437 10 m grid cells along thaw depths between 0.4 and 0.7m, differences in the mod- ZERO-line for the years 2003–2010. The average bias eled fractions occur (Fig.6). However, this can be explained in the snowmelt date of the degree-day melt model is by deviations between measured and modeled thaw depth on 1.2 day compared to MicroMet/SnowModel. The abla- the order of 0.1 to 0.2m, which is in agreement with the com- tion of the snow cover was subsequently calculated us- parison of Fig.5. ing the downscaled air temperatures for each day. For air www.the-cryosphere.net/9/719/2015/ The Cryosphere, 9, 719–735, 2015 728 S. Westermann et al.: Permafrost in northeast Greenland

Figure 7. Evolution of annual average ground temperatures at 1 m depth along the modeled part of ZERO-line for the period with in situ data from various depths for comparison. The white line is the average of all grid cells; red are the 25 and 75 % quartiles; yellow is minimum to maximum. In addition, minimum and maximum of the annual average ground temperatures at 0.3 m depth and the minimum of modeled temperatures with no snow cover (depth 1m) are Figure 5. Modeled (red) and measured (black) maximum thaw shown. The measurements are annual averages for the respective depths for the classes Dryas, Cassiope and wetland in ZeroCalm depths. The period for which MicroMet/SnowModel data are avail- (ZC) 1 and 2. The period for which MicroMet/SnowModel data are able is shaded gray. available is shaded gray. The error bars correspond to the standard deviation of the model grid cells and the in situ CALM measurements. of modeled ground temperatures at 1.0m depth, but small deviations of up to 0.5 ◦C are common, both in negative and positive directions. Two data points feature larger deviations, with annual average temperatures about 1 ◦C colder than the minimum of the modeled temperatures along ZERO- line in these years. As evident from the minimum and maximum modeled ground temperatures at 0.3m depth displayed in Fig.7, these deviations can in general not be explained by the fact that some of the measurements are from depths shallower than 1m. A possible explanation is the occurrence of spots with shallower snow depths than delivered by Mi- croMet/SnowModel, in particular at spatial scales of less than 10m. In addition, a too early onset of the snow cover, as found for the MicroMet/SnowModel grid cell at the automated snow depth station near ZeroCalm 1 (Sect. 3.3), Figure 6. Distributions of measured and modeled thaw depths along the modeled part of ZERO-line on 26 August 2013. Due to the lim- could cause a warm bias of modeled ground temperatures. ited length of the active layer probe, thaw depths exceeding 1.0m This is corroborated by model simulations assuming 0 snow could not be determined exactly and are plotted as a single bin at depth throughout the entire model period, which is still sig- 1.2m. niﬁcantly colder than the coldest measured annual average ground temperature (Fig.7). Note that snow depth measurements at the sites of the ground temperature measurements, 4.1.2 Ground temperatures which could prove this hypothesis, do not exist and other reasons, such as a systematic bias of employed model pa- To assess the model performance for ground tempera- rameters (e.g., the thermal conductivity of the snow) cannot tures, measurements conducted in the vicinity of ZERO-line be ruled out. Furthermore, it must be emphasized that the (Fig.1) between 1996 and 2014 are employed. The compar- sites with ground temperature measurements do not repre- ison focuses on annual average near-surface ground temper- sent a representative sample of the area, so that it is not pos- atures (depths between 0.15 and 1.0m), for which in total 47 sible to compare the distributions of ground temperatures (as data points from 17 different sites are available (Fig.7). The for thaw depth, Fig.6). Furthermore, most of the measure- majority of the data points are contained within the range ments are not directly located on ZERO-line, which is likely

The Cryosphere, 9, 719–735, 2015 www.the-cryosphere.net/9/719/2015/ S. Westermann et al.: Permafrost in NE Greenland 13

S. Westermann et al.: Permafrost in northeast Greenland 729

Figure 9. Evolution of annual maximum thaw depth until 2100 for the ecosystem classes Cassiope (ZeroCalm 1), Dryas and wetland (ZeroCalm 2). The yellow areas indicate the range of modeled maximum thaw depths.

4.2 Evolution of active layer thickness and ground temperatures

The modeled evolution of the temperature distribution at 1 m depth along ZERO-line is shown in Fig.8. The modeled temperatures extend over a range of 2 to 5 ◦C from minimum to Figure 8. Evolution of annual average ground temperatures at 1 m (top) and 10 m (bottom) depth along the modeled part of ZERO-line until 2100. White line: averageFigure of all grid 8. cells;Evolution Red: 25 and of 75 annual % quartiles; average yellow: ground minimum totemperatures maximum. at 1 m maximum which is evidence of the significant spatial vari- (top) and 10 m (bottom) depth along the modeled part of ZERO-line ability of the ground thermal regime in this landscape. In until 2100. The white line is the average of all grid cells; red are the order to investigate the sources for this spatial variability, a 25 and 75 % quartiles; yellow is minimum to maximum. sensitivity analysis was performed by running CryoGrid 2 for ZERO-line with a uniform ground stratigraphy and associated characteristic NDVI values (Sect. 3.1) for each of the to cause additional deviations between measurements and four stratigraphic classes: fell, Dryas, Cassiope and wetland. model results. Nevertheless, the comparison suggests that the This analysis suggests that snow depth has the largest ef- modeling approach is able to capture the spatial variability of fect on 1m ground temperatures, with a variability 3–5 times near-surface ground temperatures along and in the vicinity of larger than that caused by ground and surface properties. ZERO-line. However, modeled maximum thaw depths are much more In deeper layers, ground temperatures are influenced by influenced by ground and surface properties than by snow the temperature forcing of an extended period prior to the depths, which only lead to differences on the order of 0.1 to measurement. Therefore, measurements in deep boreholes 0.2m compared to differences of more than 0.5m for differ- are especially well suited to check the long-term performance ent stratigraphic classes/NDVI values. A statistically signifi- of a ground thermal model (in this case the model spin-up cant correlation between NDVI values (and thus stratigraphic produced by statistical downscaling). In 2012, the two deep classes) and snow depths modeled by SnowModel/MicroMet boreholes in the Zackenberg area featured temperatures at does not exist in the employed data set. − ◦ − ◦ 10 m depth of 5.2 C at a site with a snowdrift and 6.7 C According to the climate change scenario of the future at the meteorological station with more regular snow con- projections (Sect. 3.2), ground temperatures will warm by ditions. These point measurements are well in the range of about 4 ◦C until 2100, but permafrost will largely remain 10 m temperatures delivered by CryoGrid 2 along ZERO-line thermally sustainable along ZERO-line. However, the high- − ± ◦ in 2012, ( 6.0 0.6) C, and maximum and minimum tem- resolution simulations suggest a few sites where the yearly − − ◦ peratures of 5.1 and 8.0 C. The satisfactory agreement average 1m ground temperatures become positive in some suggests that the statistical downscaling procedure (Sect. 3.4) years at the end of this century (Fig.8). These sites are char- employed to produce the forcing data for the model spin-up acterized by above-average snow depths in the long-term av- is adequate for the area. erage, which suggests that talik formation may be initiated

www.the-cryosphere.net/9/719/2015/ The Cryosphere, 9, 719–735, 2015 S. Westermann et al.: Permafrost in NE Greenland 15

730 S. Westermann et al.: Permafrost in northeast Greenland at sites with topographically induced snowdrifts. The future warming of air temperatures predicted by HIRHAM is not constant over the year, with the most pronounced warming of 0.4–0.6 ◦C per decade occurring in fall, winter and spring, while summer (June to August) temperatures increase by less than 0.2 ◦C per decade. As a consequence, the annual maximum thaw depths increase only moderately until 2100, from 0.8–1.0 to 1.1–1.4 m for Dryas, from 0.65–0.85 to 0.8–1.1 m for Cassiope and from 0.5–0.65 to 0.6–0.8 m for the wetland class (Fig.9). The climate sensitivity of thaw depths is different between the classes, with a stronger increase for the classes with dryer soils than for the wetlands. It is remarkable that the projected increase is only 0.1–0.2 m in the wetlands, which can be related to the high ice content in the frozen active layer and to relatively smaller increase in summer surface temperatures due to the low summer nt factors assigned to the wetland class (Fig.3). The biological activity in this high-Arctic ecosystem is strongly related to summer conditions. The simulations suggest a signiﬁcant increase in average summer temperatures and thawing degree days (Fig. 10) within the effective root depth. The combination of deeper active layer (Fig.9) and warmer near-surface (Fig. 10) summer conditions is an important control for plant growth. Water and nutrients (mainly nitrogen) are being released from the thawing permafrost, and the longer growing season and warmer top soil conditions allow plants to beneﬁt from the additional nutrient and result in changes in the competition among plant species for light. This may lead to marked changes in vegetation over time, but this is beyond the scope of this study. Figure 10. The distribution of thawing degree days (top) and average summer (June–August) temperatures (bottom) at 0.1 m depth along ZERO-line until 2100. 5 Discussion Figure 10. The distribution of thawing degree days (top) and average summer (June–August) temperatures (bottom) at 0.1 m depth along ZERO-line until 2100. 5.1 Scaling strategies from GCM to plot scale ulate snow distributions over large areas (e.g., the ice-free parts of Greenland, several 100 000 km2) using, for example, The presented simulations of ground temperatures and per- subgrid snow distribution representations (e.g., Liston et al., mafrost state variables are derived from a multi-step down- 1999; Liston, 2004; Liston and Hiemstra, 2011). Gisnås et al. scaling approach (Sect. 3.4) which bridges the coarse spa- (2014) recently presented a statistical approach to account tial resolution of a GCM (hundreds of km) and the impact for the impact of the small-scale variability of snow depths scale on the ground (set to 10 m for this study). As such, on ground temperatures that is applicable on large spatial do- the scheme is technically capable of bridging 5 to 6 orders mains. of magnitude in space. The two main driving environmen- The surface temperature is derived from air temperature tal variables for the thermal model CryoGrid 2 are surface for which the regional gradients are based on the RCM at temperature and snow depth. a scale of 5 km. Within the target area along ZERO-line (a The snow depth is assumed to be controlled by wind distance of 4 km), variations in air temperature are generally drift of snow at the target scale of 10 m which is modeled small. Further downscaling to 10 m is accomplished by us- by the snow redistribution scheme MicroMet/SnowModel. ing a high-resolution NDVI satellite image and the NDVI MicroMet/SnowModel is a deterministic scheme capable of vs. n factor relationship (Sect. 3.1) which is used to con- predicting the snow depth for each model grid cell, thus re- vert air to surface temperatures during the snow-free season. producing the location of snow drifts and bare-blown spots. By this scheme, a high-resolution data set of surface tem- Such deterministic high-resolution modeling facilitates a bet- peratures is generated from comparatively low-resolution air ter comparison and validation with ground observations but temperature data. More physically based approaches make is restricted to small model domains for computational rea- use of the surface energy balance (SEB) to compute surface sons. However, SnowModel also includes the ability to sim- temperatures from air temperature, wind speed, humidity and

The Cryosphere, 9, 719–735, 2015 www.the-cryosphere.net/9/719/2015/ S. Westermann et al.: Permafrost in northeast Greenland 731 incoming radiation (e.g., Zhang et al., 2013; Fiddes et al., properties of the snow and the ground stratigraphy. As an ex- 2015). In addition to accounting for more complex topog- ample, the snow density and thermal conductivity are likely raphy through spatially distributed fields of incoming radi- to increase in a warmer climate, which would decrease the ation, the surface energy balance and thus the surface tem- insulation of the winter snow cover and thus lead to colder perature can directly be connected to surface soil moisture temperatures, as suggested by the model. A sensitivity study and land cover/vegetation type, which circumvents the use of for a transient thermal model similar to CryoGrid 2 in Siberia n factors. Nonetheless, SEB models require additional driv- showed that the thermal properties of the snow cover are ing data sets, in particular incoming radiation, which must be the critical source of uncertainty for successfully reproduc- compiled, e.g., from large-scale atmospheric modeling (Fid- ing ground temperatures (Langer et al., 2013). A similar re- des and Gruber, 2014) and/or from sparse in situ measure- sult was found in a sensitivity study with GEOtop (Endrizzi ments (Zhang et al., 2012). Due to the potential for serious et al., 2014) for a site in the European Alps for which the biases in such driving data sets in remote locations (such as assumed snow conditions crucially influenced the uncertain- Zackenberg), it remains unclear whether the capacity of SEB ties of modeled ground temperatures (Gubler et al., 2013). models in reproducing the true surface temperature is supe- Most likely, these findings are also applicable to this study rior to the simple empirical concept employed in this study. and the representation of the snow cover (including snow water equivalent, density and thermal conductivity) deserves 5.2 Model uncertainty increased attention in future modeling approaches. However, the ground thermal properties related to the ground stratigra- The presented model results must be considered a first-order phy proved to be the crucial source of uncertainty regarding approximation of the future thermal state of the permafrost, modeled thaw depths (Langer et al., 2013). In this study, con- which is subject to considerable uncertainty due to a variety stant soil water and ice contents are assumed in our model- of factors. Firstly, only one climate change scenario is con- ing, thus neglecting both seasonal and long-term changes in sidered, which does not account for the considerable spread the water cycle. However, at least for the Cassiope class, our in climate predictions. With permafrost approaching the thaw results for the future increase in maximum thaw depth are threshold at the end of this century for RCP 4.5 forcing, in good agreement with the study of Hollesen et al.(2011) wide-spread permafrost degradation is e.g., likely for more who used the coupled heat and water transfer model COUP aggressive climate change scenarios. (Jansson and Karlberg, 2004) to simulate the ground thermal Secondly, the downscaling procedure from large-scale and moisture regimes in this century. While a coupled energy model data to high-resolution fields of temperature and snow and water cycle is implemented in a number of modeling depth is susceptible to uncertainties, since it assumes con- schemes, such as GEOtop (Endrizzi et al., 2014) or Surfex stant offsets between the two data sets based on the climate (Masson et al., 2013), a major challenge is modeling lateral conditions of a 7-year reference period, which may not be water fluxes, which create spatially different soil moisture justified for a 100-year period. This is particularly critical conditions (as at the Zackenberg site) that subsequently can since the temperature regime in the study area is character- have a pronounced impact on the ground thermal regime. ized by strong inversion settings during a large part of the As pointed out by Gubler et al.(2011), spatially distributed year (Meltofte et al., 2008). A modification of such inver- in situ data sets are required to calibrate and validate spatially sions could lead to a different climate susceptibility of the distributed modeling schemes in heterogeneous permafrost study area compared to the large-scale trend, which cannot landscapes. These should capture the variability of the dif- be captured during the reference period. Furthermore, the fu- ferent environmental factors governing the ground thermal ture snow distribution patterns are based on random years regime, which in many permafrost landscapes will require a from the 7-year reference period, implying that the patterns significant effort with potentially dozens of measurement lo- are unchanged in a warmer future climate and that the ref- cations. However, such work is a crucial prerequisite to im- erence period allows a representative sample of the interan- prove the ability of modeling schemes to simulate the dis- nual variability within the rest of the century. It is not un- tribution of the ground thermal regime and its response to likely that both assumptions are violated at least to a cer- present and future changes. tain degree. In addition, new processes not accounted for by the modeling scheme might become relevant in the course 5.3 From model results to permafrost landscape of climate warming, e.g., the occurrence of wintertime rain development events, which can lead to significant additional ground warming (Westermann et al., 2011). Most real-world applications of permafrost models assume The CryoGrid 2 permafrost model assumes properties and non-interacting grid cells with spatially constant soil proper- relationships compiled and adjusted for the present state to ties. Consequently, permafrost degradation in model studies be valid in the future. This concerns in particular the NDVI- (e.g., Westermann et al., 2013) is generally described as talik based summer n-factor relationship employed to derive sur- formation manifested in the temperature profile of a one- face from air temperatures (Sect. 3.1), as well the thermal dimensional grid cell. While this is indeed observed in in- www.the-cryosphere.net/9/719/2015/ The Cryosphere, 9, 719–735, 2015 732 S. Westermann et al.: Permafrost in northeast Greenland strumented boreholes, it can be accompanied by significant and a regional climate model. The following conclusions can changes in the hydrological regime by thawing of hydrolog- be drawn from this study: ical barriers or the formation of new aquifers. Most operational permafrost models are not capable of predicting such – The model approach can capture the measured dif- developments, which is a significant limitation for sound pre- ferences in maximum thaw depth between different dictions on the permafrost–carbon feedback. For the study ecosystem classes encountered in the area. The simu- area, Elberling et al.(2013) demonstrated that the potential lated ground temperatures are in agreement with available borehole measurements. CO2 emissions from carbon-rich wetland soils strongly depend on the future hydrological regime of the wetland, with – The modeled average ground temperatures increase by a drying of the wetland leading to significantly faster car- about 4 ◦C until 2100 but generally remain below 0 ◦C. bon turnover. Furthermore, thawing of excess ground ice can However, a few model grid cells with topographically entirely modify the landscape, e.g., through thermokarst or induced snow drifts feature positive average 1 m tem- thaw slumps which can be hotspots of greenhouse gas emis- peratures in single years after 2060. sions and thus modify the carbon budget of an entire permafrost landscape. While excess ground ice has been in- – The modeled maximum thaw depths increase moder- cluded in land-surface models (Lee et al., 2014), the con- ately in all ecosystem classes, with the lowest value of siderable spatial variability and the interplay between excess 0.2 m for the wetland sites. ice thaw, microtopography and fluxes of energy and water represent major unresolved challenges. – The spatial variability of the average ground temperatures at 1 m depth within a distance of a few kilometers From the perspective of model development, a simple in- ◦ crease of the spatial resolution seems a prerequisite to resolve is between 3 and 5 C and thus on the order of the pro- such shortcomings in the next generation of permafrost mod- jected increase of ground temperatures until the end of els. At a 10 m resolution, this study captured two important this century. Therefore, both modeling and in situ mon- aspects which can be seen as part of the “thermal signature” itoring of the ground thermal regime may provide an of the permafrost landscape in Zackenberg: (a) the differ- incomplete assessment of present and future permafrost ences in maximum thaw depth between different ecosystem thaw if they are restricted to one or a few points within classes and (b) the spatial variability of ground temperatures an area. to a large extent caused by spatially variable snow depths. The study exemplifies that grid-based simulations at coarse Compared to large-scale (as in Daanen et al., 2011) or point- scales with only one or few model realizations cannot fully scale simulations (as in Hollesen et al., 2011), it provides account for the spatial variability of the ground thermal a far more detailed (though still incomplete) assessment of regime in many permafrost areas. They are hence not capable the possible development of the Zackenberg permafrost land- of correctly projecting the onset of permafrost thaw at the re- scape, which can be better linked to studies on the future quired scale to trigger transformative landscape changes due ecosystem carbon turnover (e.g., Elberling et al., 2013). For to erosion and hydrological processes. Despite the complex modeling of large spatial domains, a grid cell size of 10m model approach, the projections of the future ground thermal is generally not feasible due to computation power. Statisti- regime are associated with considerable uncertainties related cal representations of small-scale variability are a promising to a variety of environmental factors, which exemplifies the approach to overcome this problem, as recently explored by need for intensified process studies in permafrost environ- Fiddes et al.(2015) and Gisnås et al.(2014). ments.

Acknowledgements. We gratefully acknowledge the financial 6 Conclusions support for this study from the Danish National Research Foundation (CENPERM DNRF100), the European Union FP7- This study presents numerical simulations of present and fu- ENVIRONMENT project PAGE21 under contract no. GA282700 ture ground thermal conditions for a transect across the low- as well as the Faculty of Science at the University of Copenhagen. lying parts of the high-Arctic Zackenberg valley in northeast Also, we extend our gratitude to the Greenland Ecosystem Mon- Greenland. At the modeling scale of 10 m, the governing fac- itoring (GEM) programme, funded by the Danish Ministry for Climate, Energy and Building, for providing access to data. Finally, tors for the ground thermal regime are accounted for in a a warm thanks to staff from GeoBasis and BioBasis at Zackenberg deterministic way. This involves snow depth (derived from research station and the students involved in field measurements. a blowing snow model), ground properties (derived from an Two anonymous reviewers greatly contributed to improving the NDVI-based ecosystem classification) and surface properties manuscript. (derived from empirical correction factors for summer surface temperature based on NDVI). Past and future climate Edited by: S. Gruber trends for the general area are derived from in situ records

The Cryosphere, 9, 719–735, 2015 www.the-cryosphere.net/9/719/2015/ S. Westermann et al.: Permafrost in northeast Greenland 733

References Geosci. Model Dev., 5, 1245–1257, doi:10.5194/gmd-5-1245- 2012, 2012. Bay, C.: Vegetation mapping of Zackenberg valley, Northeast Fiddes, J. and Gruber, S.: TopoSCALE v.1.0: downscaling gridded Greenland, Danish Polar Center and Botanical Museum, Univer- climate data in complex terrain, Geosci. Model Dev., 7, 387–405, sity of Copenhagen, 75 pp., 1998. doi:10.5194/gmd-7-387-2014, 2014. Brown, J., Hinkel, K., and Nelson, F.: The circumpolar active layer Fiddes, J., Endrizzi, S., and Gruber, S.: Large-area land surface sim- monitoring (CALM) program: research designs and initial re- ulations in heterogeneous terrain driven by global data sets: ap- sults, Polar Geogr., 24, 166–258, 2000. plication to mountain permafrost, The Cryosphere, 9, 411–426, Christensen, J. H., Christensen, O. B., Lopez, P., van Meijgaard, E., doi:10.5194/tc-9-411-2015, 2015. and Botzet, M.: The HIRHAM 4 regional atmospheric climate Fredskild, B. and Mogensen, G. S.: Zero Line: Final Report 1997: model, Scientific Report 96-4, Danish Meteorological Institute, a Description of the Plant Communities Along the ZERO Line 51 pp., 1996. from Young Sund to the Top of Aucellabjerg and the Common Clarke, L., Edmonds, J., Jacoby, H., Pitcher, H., Reilly, J., and Plant Communities in the Zackenberg Valley, Northeast Green- Richels, R.: Scenarios of greenhouse gas emissions and atmo- land, Greenland Botanical Survey & Botanical Museum, Univer- spheric concentrations, Sub-report of Synthesis and Assessment sity of Copenhagen, 1997. Product – 2.1. Climate Change Science Program and the Sub- Gisnås, K., Westermann, S., Schuler, T. V., Litherland, T., Isaksen, committee on Global Change Research, Washington DC, 2007. K., Boike, J., and Etzelmüller, B.: A statistical approach to rep- Clauser, C. and Huenges, E.: Thermal conductivity of rocks and resent small-scale variability of permafrost temperatures due to minerals, AGU Reference Shelf, 3, 105–126, 1995. snow cover, The Cryosphere, 8, 2063–2074, doi:10.5194/tc-8- Cosenza, P., Guerin, R., and Tabbagh, A.: Relationship between 2063-2014, 2014. thermal conductivity and water content of soils using numerical Goodrich, L.: The influence of snow cover on the ground thermal modelling, European J. Soil Sci., 54, 581–588, 2003. regime, Canad. Geotech. J., 19, 421–432, 1982. Côté, J. and Konrad, J.-M.: A generalized thermal conductivity Gubler, S., Fiddes, J., Keller, M., and Gruber, S.: Scale- model for soils and construction materials, Canad. Geotechn. J., dependent measurement and analysis of ground surface temper- 42, 443–458, 2005. ature variability in alpine terrain, The Cryosphere, 5, 431–443, Daanen, R. P., Ingeman-Nielsen, T., Marchenko, S. S., Ro- doi:10.5194/tc-5-431-2011, 2011. manovsky, V. E., Foged, N., Stendel, M., Christensen, J. H., Gubler, S., Endrizzi, S., Gruber, S., and Purves, R. S.: Sensi- and Hornbech Svendsen, K.: Permafrost degradation risk zone tivities and uncertainties of modeled ground temperatures in assessment using simulation models, The Cryosphere, 5, 1043– mountain environments, Geosci. Model Dev., 6, 1319–1336, 1056, doi:10.5194/tc-5-1043-2011, 2011. doi:10.5194/gmd-6-1319-2013, 2013. Dall’Amico, M., Endrizzi, S., Gruber, S., and Rigon, R.: A robust Hansen, B. U., Sigsgaard, C., Rasmussen, L., Cappelen, J., Hinkler, and energy-conserving model of freezing variably-saturated soil, J., Mernild, S. H., Petersen, D., Tamstorf, M. P., Rasch, M., and The Cryosphere, 5, 469–484, doi:10.5194/tc-5-469-2011,2011. Hasholt, B.: Present-day climate at Zackenberg, Adv. Ecol. Res., Elberling, B., Jakobsen, B. H., Berg, P., Søndergaard, J., and Sigs- 40, 111–149, 2008. gaard, C.: Influence of vegetation, temperature, and water content Hazeleger, W., Severijns, C., Semmler, T., Stefanescu, S., Yang, on soil carbon distribution and mineralization in four High Arctic S., Wang, X., Wyser, K., Dutra, E., Baldasano, J. M., Bin- soils, Arct. Antarct. Alp. Res., 36, 528–538, 2004. tanja, R., Caballero, R., Dutra, E., Ekman, A., Christensen, J., Elberling, B., Tamstorf, M. P., Michelsen, A., Arndal, M. F., Sigs- van den Hurk, B., Jimenez, P., Jones, C., Kallberg, P., McGrath, gaard, C., Illeris, L., Bay, C., Hansen, B. U., Christensen, T. R., R., de Miranda, P., Molteni, F., van Noije, T., Palmer, T., Ro- Hansen, E. S., Jakobsen, B. H., and Beyens, L.: Soil and plant driguez Camino, E., Schmith, T., Selten, T., Sterl, A., Tapamo, community-characteristics and dynamics at Zackenberg, Adv. H., and Viterbo, P.: EC-earth: a seamless earth-system prediction Ecol. Res., 40, 223–248, 2008. approach in action, B. Am. Meteor. Soc., 91, 1357–1363, 2010. Elberling, B., Christiansen, H. H., and Hansen, B. U.: High nitrous Hazeleger, W., Wang, X., Severijns, C., ¸Stefanescu,˘ S., Bintanja, oxide production from thawing permafrost, Nature Geosci., 3, R., Sterl, A., Wyser, K., Semmler, T., Yang, S., Van den Hurk, 332–335, 2010. B., van Noije, T., van der Linden, E., and van der Wiel, K.: EC- Elberling, B., Michelsen, A., Schädel, C., Schuur, E. A., Chris- Earth V2. 2: description and validation of a new seamless earth tiansen, H. H., Berg, L., Tamstorf, M. P., and Sigsgaard, C.: system prediction model, Clim. Dynam., 39, 2611–2629, 2012. Long-term CO2 production following permafrost thaw, Nature Hinkler, J., Pedersen, S. B., Rasch, M., and Hansen, B. U.: Auto- Clim. Change, 3, 890–894, 2013. matic snow cover monitoring at high temporal and spatial reso- Ellebjerg, S. M., Tamstorf, M. P., Illeris, L., Michelsen, A., and lution, using images taken by a standard digital camera, Int. J. Hansen, B. U.: Inter-annual variability and controls of plant phe- Remote Sens., 23, 4669–4682, 2002. nology and productivity at Zackenberg, Adv. Ecol. Res., 40, 249– Hipp, T., Etzelmüller, B., Farbrot, H., Schuler, T., and Westermann, 273, 2008. S.: Modelling borehole temperatures in Southern Norway – in- Endrizzi, S., Gruber, S., Dall’Amico, M., and Rigon, R.: GEOtop sights into permafrost dynamics during the 20th and 21st century, 2.0: simulating the combined energy and water balance at The Cryosphere, 6, 553–571, doi:10.5194/tc-6-553-2012, 2012. and below the land surface accounting for soil freezing, snow Hock, R.: Temperature index melt modelling in mountain areas, J. cover and terrain effects, Geosci. Model Dev., 7, 2831–2857, Hydrol., 282, 104–115, 2003. doi:10.5194/gmd-7-2831-2014, 2014. Hollesen, J., Elberling, B., and Jansson, P. E.: Future active layer dy- Fiddes, J. and Gruber, S.: TopoSUB: a tool for efficient large area namics and carbon dioxide production from thawing permafrost numerical modelling in complex topography at sub-grid scales, www.the-cryosphere.net/9/719/2015/ The Cryosphere, 9, 719–735, 2015 734 S. Westermann et al.: Permafrost in northeast Greenland

layers in Northeast Greenland, Global Change Biol., 17, 911– Liston, G. E. and Elder, K.: A meteorological distribution system 926, 2011. for high-resolution terrestrial modeling (MicroMet), J. Hydrom- Hugelius, G., Bockheim, J. G., Camill, P., Elberling, B., Grosse, eteorol., 7, 217–234, 2006b. G., Harden, J. W., Johnson, K., Jorgenson, T., Koven, C. D., Liston, G. E. and Hall, D. K.: An energy-balance model of lake-ice Kuhry, P., Michaelson, G., Mishra, U., Palmtag, J., Ping, C.-L., evolution, J. Glaciol., 41, 373–382, 1995. O’Donnell, J., Schirrmeister, L., Schuur, E. A. G., Sheng, Y., Liston, G. E. and Hiemstra, C. A.: A simple data assimilation sys- Smith, L. C., Strauss, J., and Yu, Z.: A new data set for estimating tem for complex snow distributions (SnowAssim), J. Hydrome- organic carbon storage to 3 m depth in soils of the northern cir- teorol., 9, 989–1004, 2008. cumpolar permafrost region, Earth Syst. Sci. Data, 5, 393–402, Liston, G. E. and Hiemstra, C. A.: The Changing Cryosphere: Pan- doi:10.5194/essd-5-393-2013, 2013. Arctic Snow Trends (1979–2009), J. Climate, 24, 5691–5712, Jafarov, E. E., Marchenko, S. S., and Romanovsky, V. E.: Numerical 2011. modeling of permafrost dynamics in Alaska using a high spatial Liston, G. E. and Mernild, S. H.: Greenland freshwater runoff. Part resolution dataset, The Cryosphere, 6, 613–624, doi:10.5194/tc- I: A runoff routing model for glaciated and non-glaciated land- 6-613-2012, 2012. scapes (HydroFlow), J. Climate, 25, 5997–6014, 2012. Jansson, P. and Karlberg, L.: COUP manual: Coupled heat and Liston, G. E., Pielke, R. A., and Greene, E. M.: Improving first- mass transfer model for soil-plant-atmosphere systems, Techni- order snow-related deficiencies in a regional climate model, J. cal manual for the CoupModel, 1–453, 2004. Geophys. Res.-Atmos., 104, 19559–19567, 1999. Klene, A. E., Nelson, F. E., Shiklomanov, N. I., and Hinkel, K. M.: Liston, G. E., Haehnel, R. B., Sturm, M., Hiemstra, C. A., Bere- The N-factor in natural landscapes: Variability of air and soil- zovskaya, S., and Tabler, R. D.: Instruments and methods simu- surface temperatures, Kuparuk river basin, Alaska, USA, Arctic, lating complex snow distributions in windy environments using Antarctic, and Alpine Research, 140–148, 2001. SnowTran–3D, J. Glaciol., 53, 241–256, 2007. Koch, L. and Haller, J.: Geological map of Clavering Ø (74 Ø 2), Lucas-Picher, P., Wulff-Nielsen, M., Christensen, J. H., Aðalgeirs- Meddelelser om Grønland, Bd. 18.3, 1965. dóttir, G., Mottram, R., and Simonsen, S. B.: Very high resolu- Langen, P. L., Mottram, R. H., Christensen, J. H., Boberg, F., tion regional climate model simulations over Greenland: Iden- Rodehacke, C. B., Stendel, M., van As, D., Ahlstrøm, A. P., tifying added value, J. Geophys. Res.-Atmos., 117, D02108, Mortensen, J., Rysgaard, S., Petersen, D., Svendsen, K. H., Aðal- doi:10.1029/2011JD016267, 2012. geirsdóttir, G., Cappelen, J., Quantifying energy and mass fluxes Masson, V., Le Moigne, P., Martin, E., Faroux, S., Alias, A., controlling Godthåbsfjord freshwater input in a 5 km simula- Alkama, R., Belamari, S., Barbu, A., Boone, A., Bouyssel, F., tion (1991–2012), J. Climate, doi:10.1175/JCLI-D-14-00271.1, Brousseau, P., Brun, E., Calvet, J.-C., Carrer, D., Decharme, B., in press, 2015. Delire, C., Donier, S., Essaouini, K., Gibelin, A.-L., Giordani, H., Langer, M., Westermann, S., Heikenfeld, M., Dorn, W., and Boike, Habets, F., Jidane, M., Kerdraon, G., Kourzeneva, E., Lafaysse, J.: Satellite-based modeling of permafrost temperatures in a M., Lafont, S., Lebeaupin Brossier, C., Lemonsu, A., Mahfouf, tundra lowland landscape, Remote Sens. Environ., 135, 12–24, J.-F., Marguinaud, P., Mokhtari, M., Morin, S., Pigeon, G., Sal- 2013. gado, R., Seity, Y., Taillefer, F., Tanguy, G., Tulet, P., Vincendon, Lawrence, D. M., Slater, A. G., and Swenson, S. C.: Simulation of B., Vionnet, V., and Voldoire, A.: The SURFEXv7.2 land and present-day and future permafrost and seasonally frozen ground ocean surface platform for coupled or offline simulation of earth conditions in CCSM4, J. Climate, 25, 2207–2225, 2012. surface variables and fluxes, Geosci. Model Dev., 6, 929–960, Lee, H., Swenson, S. C., Slater, A. G., and Lawrence, D. M.: Effects doi:10.5194/gmd-6-929-2013, 2013. of excess ground ice on projections of permafrost in a warm- Mastepanov, M., Sigsgaard, C., Dlugokencky, E. J., Houweling, S., ing climate, Environ. Res. Lett., 9, 124 006, doi:10.1088/1748- Ström, L., Tamstorf, M. P., and Christensen, T. R.: Large tundra 9326/9/12/124006, 2014. methane burst during onset of freezing, Nature, 456, 628–630, Lehning, M., Völksch, I., Gustafsson, D., Nguyen, T. A., Stähli, 2008. M., and Zappa, M.: ALPINE3D: a detailed model of mountain Meltofte, H., Christensen, T. R., Elberling, B., Forchhammer, M. C., surface processes and its application to snow hydrology, Hydrol. and Rasch, M.: High-arctic ecosystem dynamics in a changing Proc., 20, 2111–2128, 2006. climate, 40, 1–563, Elsevier Academic Press Inc., San Diego, Liston, G. E.: Interrelationships among snow distribution, 2008. snowmelt, and snow cover depletion: Implications for atmo- Merzlaya, G., Stepanov, A., and Fedorov, A. Y.: Growing potatoes spheric, hydrologic, and ecologic modeling, J. Appl. Meteorol., above the arctic circle, Russ. Agric. Sci., 34, 373–376, 2008. 38, 1474–1487, 1999. Mesinger, F., DiMego, G., Kalnay, E., Mitchell, K., Shafran, P. C., Liston, G. E. and Sturm, M.: A snow-transport model for complex Ebisuzaki, W., Jovic, D., Woollen, J., Rogers, E., Berbery, E. H., terrain, J. Glaciol., 44, 498–516, 1998. Ek, B. M., Fan, Y., Grumbine, R., Higgins, W., Li, H., Lin, Y., Liston, G. E.: Local advection of momentum, heat, and moisture Manikin, G., Parrish, D., and Shi, W.: North American regional during the melt of patchy snow covers, J. Appl. Meteorol., 34, reanalysis, B. Am. Meteor. Soc., 87, 343–360, 2006. 1705–1715, 1995. Mick, A. H. and Johnson, H. A.: Soil resources and agricultural Liston, G. E.: Representing subgrid snow cover heterogeneities in development in Alaska, Arctic, 7, 236–248, 1954. regional and global models, J. Climate, 17, 1381–1397, 2004. Osterkamp, T. and Romanovsky, V.: Evidence for warming and Liston, G. E. and Elder, K.: A distributed snow-evolution modeling thawing of discontinuous permafrost in Alaska, Permafrost system (SnowModel), J. Hydrometeorol., 7, 1259–1276, 2006a. Periglac. Proc., 10, 17–37, 1999.

The Cryosphere, 9, 719–735, 2015 www.the-cryosphere.net/9/719/2015/ S. Westermann et al.: Permafrost in northeast Greenland 735

Pedersen, S. H., Liston, G. E., Tamstorf, M. P., Westergaard- Walker, D., Jia, G., Epstein, H., Raynolds, M., Chapin Iii, F., Nielsen, A., and Schmidt, N. M.: Quantifying episodic snowmelt Copass, C., Hinzman, L., Knudson, J., Maier, H., Michaelson, events in Arctic ecosystems, Ecosystems, doi:10.1007/s10021- G., et al.: Vegetation-soil-thaw-depth relationships along a low- 015-9867-8, 2015. arctic bioclimate gradient, Alaska: synthesis of information from Rae, J. G. L., Aðalgeirsdóttir, G., Edwards, T. L., Fettweis, X., Gre- the ATLAS studies, Permafrost Periglac. Proc., 14, 103–123, gory, J. M., Hewitt, H. T., Lowe, J. A., Lucas-Picher, P., Mot- 2003. tram, R. H., Payne, A. J., Ridley, J. K., Shannon, S. R., van de Walter, K., Zimov, S., Chanton, J., Verbyla, D., and Chapin III, F.: Berg, W. J., van de Wal, R. S. W., and van den Broeke, M. R.: Methane bubbling from Siberian thaw lakes as a positive feed- Greenland ice sheet surface mass balance: evaluating simula- back to climate warming, Nature, 443, 71–75, 2006. tions and making projections with regional climate models, The Wang, S., Li, Z., and Wu, J.: The state of the art and new subjects Cryosphere, 6, 1275–1294, doi:10.5194/tc-6-1275-2012, 2012. of highway research in permafrost regions of China, J. Glaciol. Schuur, E. A. G., Bockheim, J., Canadell, J. G., Euskirchen, E., Geocryol., 25, 471–476, 2003. Field, C. B., Goryachkin, S. V., Hagemann, S., Kuhry, P., Laﬂeur, Wang, S., Huang, X., and Hou, S.: Numerical analyses of pave- P. M., Lee, H., Mazhitova, G., Nelson, F. E., Rinke, A., Ro- ment deformation and stress in permafrost regions, J. Glaciol. manovsky, V. E., Shiklomanov, N., Tarnocai, C., Venevsky, S., Geocryol., 28, 217–222, 2006. Vogel, J. G., and Zimov, S. A.: Vulnerability of permafrost car- Westermann, S., Boike, J., Langer, M., Schuler, T. V., and Et- bon to climate change: Implications for the global carbon cycle, zelmüller, B.: Modeling the impact of wintertime rain events on Bioscience, 58, 701–714, 2008. the thermal regime of permafrost, The Cryosphere, 5, 945–959, Schuur, E. A., Vogel, J. G., Crummer, K. G., Lee, H., Sickman, doi:10.5194/tc-5-945-2011, 2011. J. O., and Osterkamp, T. E.: The effect of permafrost thaw on Westermann, S., Schuler, T., Gisnås, K., and Etzelmüller, B.: Tran- old carbon release and net carbon exchange from tundra, Nature, sient thermal modeling of permafrost conditions in Southern 459, 556–559, 2009. Norway, The Cryosphere, 7, 719–739, doi:10.5194/tc-7-719- Smith, S. J. and Wigley, T.: Multi-gas forcing stabilization with 2013, 2013. MiniCAM, The Energy Journal, 373–392, 2006. Winstral, A., Elder, K., and Davis, R. E.: Spatial snow modeling of Stendel, M. and Christensen, J.: Impact of global warming on per- wind-redistributed snow using terrain-based parameters, J. Hy- mafrost conditions in a coupled GCM, Geophys. Res. Lett., 29, drometeorol., 3, 524–538, 2002. 10-1–10-4, doi:10.1029/2001GL014345, 2002. Wise, M., Calvin, K., Thomson, A., Clarke, L., Bond-Lamberty, B., Stendel, M., Romanovsky, V. E., Christensen, J. H., and Sazonova, Sands, R., Smith, S. J., Janetos, A., and Edmonds, J.: Implica- T.: Using dynamical downscaling to close the gap between global tions of limiting CO2 concentrations for land use and energy, change scenarios and local permafrost dynamics, Global Planet. Science, 324, 1183–1186, 2009. Change, 56, 203–214, 2007. Yen, Y.-C.: Review of thermal properties of snow, ice and sea ice, Tamstorf, M., Illeris, L., Hansen, B., and Wisz, M.: Spectral mea- CRREL Rep., 81–10, 1981. sures and mixed models as valuable tools for investigating con- Zhang, Y.: Spatio-temporal features of permafrost thaw projected trols on land surface phenology in high arctic Greenland, BMC from long-term high-resolution modeling for a region in the Hud- Ecology, 7, p. 9, doi:10.1186/1472-6785-7-9, 2007. son Bay Lowlands in Canada, J. Geophys. Res.-Earth Surf., 118, Thomson, A. M., Calvin, K. V., Smith, S. J., Kyle, G. P., Volke, A., 542–552, 2013. Patel, P., Delgado-Arias, S., Bond-Lamberty, B., Wise, M. A., Zhang, Y., Li, J., Wang, X., Chen, W., Sladen, W., Dyke, L., Dredge, Clarke, L. E., et al.: RCP4.5: a pathway for stabilization of radia- L., Poitevin, J., McLennan, D., Stewart, H., Kowalchuk, S., Wu, tive forcing by 2100, Clim. Change, 109, 77–94, 2011. W., Kershaw, P., and Brook, R. K.: Modelling and mapping per- Vionnet, V., Brun, E., Morin, S., Boone, A., Faroux, S., Le Moigne, mafrost at high spatial resolution in Wapusk National Park, Hud- P., Martin, E., and Willemet, J.-M.: The detailed snowpack son Bay Lowlands, Canad. J. Earth Sci., 49, 925–937, 2012. scheme Crocus and its implementation in SURFEX v7.2, Geosci. Zhang, Y., Wang, X., Fraser, R., Olthof, I., Chen, W., Mclennan, D., Model Dev., 5, 773–791, doi:10.5194/gmd-5-773-2012, 2012. Ponomarenko, S., and Wu, W.: Modelling and mapping climate change impacts on permafrost at high spatial resolution for an Arctic region with complex terrain, The Cryosphere, 7, 1121– 1137, doi:10.5194/tc-7-1121-2013, 2013.

www.the-cryosphere.net/9/719/2015/ The Cryosphere, 9, 719–735, 2015

Paper SI

Snowpack fluxes of methane and carbon dioxide from high Arctic tundra.

Pirk, N., Lund, M., Mastepanov, M., Parmentier, F.-J. W., Pedersen, S. H., Mylius, M. R., Tamstorf, M. P., Christiansen, H. H., and Christensen, T. R. (2016). Journal of Geophysical Research: Biogeosciences, Vol. 121, No. 10, DOI: 10.1002/2016JG003486.

Journal of Geophysical Research: Biogeosciences

RESEARCH ARTICLE Snowpack ﬂuxes of methane and carbon dioxide 10.1002/2016JG003486 from high Arctic tundra

Key Points: 1,2 3 3 1,3 3 • Small gas fluxes from the frozen Norbert Pirk , Mikkel P. Tamstorf , Magnus Lund , Mikhail Mastepanov , Stine H. Pedersen , soil resemble the same spatial pattern Maria R. Mylius4, Frans-Jan W. Parmentier1,3, Hanne H. Christiansen2, and Torben R. Christensen1,3 as during the growing season • Snowpack ice layers block the 1Department of Physical Geography and Ecosystem Science, Lund University, Lund, Sweden, 2Geology Department, diffusive fluxes leading to high 3 snowpack gas concentrations University Centre in Svalbard, UNIS, Longyearbyen, Norway, Arctic Research Centre, Bioscience, Aarhus University, 4 • In polygonal tundra, Roskilde, Denmark, Department of Geosciences and Natural Resource Management, University of Copenhagen, geomorphological soil cracks Copenhagen, Denmark are the strongest methane source during wintertime

Abstract Measurements of the land-atmosphere exchange of the greenhouse gases methane (CH4) and Supporting Information: carbon dioxide (CO2) in high Arctic tundra ecosystems are particularly difficult in the cold season, resulting • Supporting Information S1 in large uncertainty on flux magnitudes and their controlling factors during this long, frozen period. We conducted snowpack measurements of these gases at permafrost-underlain wetland sites in Zackenberg ∘ ∘ Correspondence to: Valley (NE Greenland, 74 N) and Adventdalen Valley (Svalbard, 78 N), both of which also feature automatic N. Pirk, closed chamber flux measurements during the snow-free period. At Zackenberg, cold season emissions [email protected] were 1 to 2 orders of magnitude lower than growing season fluxes. Perennially, CH4 fluxes resembled the same spatial pattern, which was largely attributed to differences in soil wetness controlling substrate Citation: accumulation and microbial activity. We found no significant gas sinks or sources inside the snowpack but Pirk, N., M. P. Tamstorf, M. Lund, 𝛿13 detected a pulse in the C-CH4 stable isotopic signature of the soil’s CH4 source during snowmelt, which M. Mastepanov, S. H. Pedersen, M. R. Mylius, F.-J. W. Parmentier, suggests the release of a CH4 reservoir that was strongly affected by methanotrophic microorganisms. In the H. H. Christiansen, and polygonal tundra of Adventdalen, the snowpack featured several ice layers, which suppressed the expected T. R. Christensen (2016), Snowpack gas emissions to the atmosphere, and conversely lead to snowpack gas accumulations of up to 86 ppm CH4 fluxes of methane and carbon dioxide from high Arctic tundra, and 3800 ppm CO2 by late winter. CH4 to CO2 ratios indicated distinctly different source characteristics J. Geophys. Res. Biogeosci., in the rampart of ice-wedge polygons compared to elsewhere on the measured transect, possibly due to 121, 2886–2900, geomorphological soil cracks. Collectively, these findings suggest important ties between growing season doi:10.1002/2016JG003486. and cold season greenhouse gas emissions from high Arctic tundra.

Received 13 MAY 2016 Accepted 6 NOV 2016 Accepted article online 14 NOV 2016 1. Introduction Published online 25 NOV 2016 Fluxes of methane (CH4) and carbon dioxide (CO2) from Arctic tundra exhibit tremendous spatial variability due to complex microtopography [Sturtevant and Oechel, 2013; Olefeldt et al., 2013]. The large associated variations in soil wetness, temperature, and vegetation composition lead to varying rates of respiratory releases and microbial decomposition of organic material. Most measurements of the above are performed during the growing season, while much fewer studies exist that explain the magnitude and controls of wintertime emissions [McGuire et al., 2012]. Most biological activity in permafrost-underlain soils takes place in the uppermost, seasonally unfrozen part of the soil (also known as the active layer), whose thickness and wetness is a key control on gas exchange

processes [Christensen et al., 2003; Whalen, 2005]. Under waterlogged, anaerobic, conditions CH4 is produced by methanogenic microorganisms, while methanotrophic microorganisms in the aerobic part of the soil con-

sume CH4 as part of their metabolism [Lai, 2009]. The underlying microbial processes fractionate the carbon isotopes in distinctive ways, which can be studied by measuring the stable isotopic composition of CH4 [Hornibrook et al., 2000; Preuss et al., 2013; Vaughn et al., 2016]. Water table position also has an indirect effect ©2016. The Authors. This is an open access article under the on the gas exchange since it controls the abundance of vascular plants, which have been found to affect car- terms of the Creative Commons bon turnover and CH4 emissions through their root exudates and plant mediated gas transport [Schimel, 1995; Attribution-NonCommercial-NoDerivs Ström et al., 2005]. License, which permits use and distribution in any medium, provided The majority of studies on Arctic greenhouse gas dynamics have focused on the growing season, even though the original work is properly cited, the use is non-commercial and no this season covers a mere 2 to 3 months of the year [McGuire et al., 2012]. The present study, on the other modifications or adaptations are made. hand, focuses on the much longer Arctic winter, which can be classified into several periods based on the

PIRK ET AL. ARCTIC SNOWPACK FLUXES 2886

Paper SII

Transitions in high-Arctic vegetation growth patterns and ecosystem productivity tracked with automated cameras from 2000 to 2013.

Westergaard-Nielsen, A., Lund, M., Pedersen, S. H., Schmidt, N. M., Klosterman, S., Abermann, J., and Hansen, B.U. Ambio. DOI: 10.1007/s13280-016-0864-8.

Ambio 2017, DOI: 10.1007/s13280-016-0864-8.

Transitions in high-Arctic vegetation growth patterns and ecosystem productivity tracked with automated cameras from 2000 to 2013. Andreas Westergaard-Nielsen1,2*, Magnus Lund3, Stine Højlund Pedersen3, Niels Martin Schmidt3, Stephen Klosterman4, Birger Ulf Hansen1,2

1Department of Geosciences and Natural Resource Management, University of Copenhagen, Oestervoldgade 10, 1350 Copenhagen, Denmark 2Center for Permafrost (CENPERM), University of Copenhagen, Oestervoldgade 10, 1350 Copenhagen, Denmark 3Arctic Research Centre, Department for Bioscience, Aarhus University, Frederiksborgvej 399, 4000 Roskilde, Denmark 4Department of Organismic and Evolutionary Biology, Harvard University, Cambridge, Massachusetts 02138, USA *Corresponding Author: Andreas Westergaard-Nielsen, [email protected]

2010). Significant shifts in the timing of annual Abstract Climate induced changes in phenological events have been reported in vegetation phenology at northern latitudes are still monitoring studies based on satellite data (Jeong et poorly understood. Continued monitoring and al. 2011), as well as in in-situ observation data on research is therefore needed to improve the flowering and growing-season length (Kerby and understanding of abiotic drivers. Here we used 14 Post 2013). Such shifts in seasonality, and the years of time lapse imagery and climate data from duration of the individual seasons, can have high-Arctic Northeast Greenland to assess the important consequences for the functioning of seasonal response of a dwarf shrub heath, grassland, ecosystems and ultimately on the carbon cycle and fen, to inter-annual variation in snow-cover, soil (McGuire et al. 2009). Understanding the moisture, and air and soil temperatures. A late snow seasonality in relation to climate can thus be a key melt and start of growing season is counterbalanced to an improved understanding of ecosystem by a fast greenup and a tendency to higher peak response to a warmer climate (Richardson et al. greenness values. Snow water equivalents and soil 2013), including biologically driven fluxes of moisture explained up to 77% of growing season greenhouse gases (Menzel 2002). duration and senescence phase, highlighting that water availability is a prominent driver in the heath Several studies have found arctic site, rather than temperatures. We found a ecosystems to be particularly sensitive to shifts in significant advance in the start of spring by 10 days air temperature (Hinzman et al. 2005; Post et al. and in the end of fall by 11 days, resulting in an 2009), which again influence the vegetation unchanged growing season length. Vegetation functioning and phenology (Oberbauer et al. 2013; greenness, derived from the imagery, was correlated Høye et al. 2013). During the last two decades, to primary productivity, showing that the imagery vegetation phenology in the Arctic has been holds valuable information on vegetation monitored using both in-situ field measurements productivity. focusing on seasonal dynamics in growth (Ellebjerg et al. 2008; Michelsen et al. 2012) and its linkage to INTRODUCTION CO2 exchange (Kross et al. 2014). In parallel, arctic vegetation has been monitored from satellites (e.g. Vegetation growth and phenology are important Zeng et al. 2011), allowing for regional-scale indicators of climate change on both plant level studies. Regional studies report an increase in (Cleland et al. 2012) and global scale (Walther growing season length from both an advancement of

Paper SIII

Long-term patterns of muskox (Ovibos moschatus) demographics in High Arctic Greenland.

Schmidt, N. M., Pedersen, S. H., Mosbacher, J. B., and Hansen, L. H., (2015). Polar Biology, Vol. 38, pp. 1667–1675. DOI: 10.1007/s00300-015-1733-9.

Polar Biol (2015) 38:1667–1675 DOI 10.1007/s00300-015-1733-9

ORIGINAL PAPER

Long-term patterns of muskox (Ovibos moschatus) demographics in high arctic Greenland

Niels Martin Schmidt1 • Stine Højlund Pedersen1 • Jesper Bruun Mosbacher1 • Lars Holst Hansen1

Received: 15 September 2014 / Revised: 4 June 2015 / Accepted: 5 June 2015 / Published online: 17 June 2015 Ó Springer-Verlag Berlin Heidelberg 2015

Abstract Animal abundance is a key measure in con- Introduction servation and management and tightly linked to population demographics. Demographic data from remote regions, Measures of animal abundance are fundamental for con- however, are often scarce. Here, we present long-term servation strategies and management initiatives. Population (1996–2013) demographics on the muskox Ovibos demographic is one of the major determinants of popula- moschatus population at Zackenberg in northeast Green- tion dynamics, and changes in demographic parameters land. We examine both the inter- and intra-annual patterns may reflect ongoing or future changes in abundance (e.g., in demographic parameters and relate these to environ- Jenouvrier et al. 2009; Molna´r et al. 2010; Barraquand mental conditions. The sex and age composition of muskox et al. 2014). Knowledge of animal abundance is particu- groups changed during the study period, and changes were larly relevant in the Arctic, because the expected rapid particularly evident in the increasing versus the decreasing changes in environmental conditions may exert substantial phase of muskox abundance. The seasonal pattern of pressure on animal populations there. Current climate muskox density and group size was a parallel increase from change in the Arctic is dramatic and is affecting a host of late winter to autumn, which peaked at high densities species and their interactions (Post et al. 2009; Gilg et al. (approximately seven individuals per km2) in the autumn. 2012; Mortensen et al. 2014) and may in turn affect the The composition of muskox groups also changed between structure of the entire tundra food web (Legagneux et al. seasons. Across years, the muskox population dynamics 2014). Unfortunately, due to the remoteness and inacces- was mainly driven by spring snow cover (an indicator of sibility of much of the Arctic, detailed information on the winter conditions), which primarily impacted the calf and status and trends of many arctic species and the different yearling recruitments. This relationship, however, appeared populations is generally limited (CAFF 2013). to have a temporary decoupling, which may be attributable In tundra ecosystems, the muskox (Ovibos moschatus)isa to pathogens. Our study provides rare insight into the long- key species as one of few large herbivores. Like other term demographics of a remote ungulate population in ungulate herbivores, muskoxen may influence the ecosys- relation to drivers of change and thus aids the development tems through their selective foraging on preferred plant of adequate monitoring and management plans for musk- species or groups, which in turn affects the abundance of the oxen in a changing Arctic. various plant species, and which may alter the interspecific competition among these (Mulder 1999; Kristensen et al. Keywords Climate change Á Conservation Á Population 2011). Further, as the Arctic continues warming, grazing dynamics Á Snow Á Ungulates muskoxen may somewhat buffer climate-induced changes in vegetation composition and reduce the expansion of shrubs (Post and Pedersen 2008; Myers-Smith et al. 2011). & Niels Martin Schmidt Detailed information about the population development [email protected] of muskoxen in large parts of their distributional range is 1 Arctic Research Centre, Department of Bioscience, Aarhus lacking, and few coherent demographic data sets have been University, Frederiksborgvej 399, 4000 Roskilde, Denmark published. One of these is the long-term data series on 123

Paper SIV

Characteristics of summer-time energy exchange in a high Arctic tundra heath.

Lund, M., Hansen, B. U., Pedersen, S. H., Stiegler, C., and Tamstorf, M. P. (2014). Characteristics of summer-time energy exchange in a high Arctic tundra heath. Tellus. Series B: Chemical and Physical Meteorology, Vol. 66, [21631]. DOI: 10.3402/tellusb.v66.21631.

SERIES B CHEMICAL AND PHYSICAL METEOROLOGY PUBLISHED BY THE INTERNATIONAL METEOROLOGICAL INSTITUTE IN STOCKHOLM

Characteristics of summer-time energy exchange in a high Arctic tundra heath 2000Á2010

By MAGNUS LUND1,2*, BIRGER U. HANSEN3 , STINE H. PEDERSEN1, CHRISTIAN STIEGLER2 and MIKKEL P. TAMSTORF1, 1Department of Bioscience, Arctic Research Centre, Aarhus University, Frederiksborgvej 399, DK-4000 Roskilde, Denmark; 2Department of Physical Geography and Ecosystem Science, So¨lvegatan 12, SE-22362 Lund, Sweden; 3CENPERM Á Center for Permafrost, Department of Geosciences and Natural Resource Management, University of Copenhagen, Øster Voldgade 10, DK-1350 Copenhagen K, Denmark

(Manuscript received 6 June 2013; in final form 10 June 2014)

ABSTRACT Global warming will bring about changes in surface energy balance of Arctic ecosystems, which will have implications for ecosystem structure and functioning, as well as for climate system feedback mechanisms. In this study, we present a unique, long-term (2000Á2010) record of summer-time energy balance components

(net radiation, Rn; sensible heat flux, H; latent heat flux, LE; and soil heat flux, G) from a high Arctic tundra heath in Zackenberg, Northeast Greenland. This area has been subjected to strong summer-time warming with increasing active layer depths (ALD) during the last decades. We observe high energy partitioning into H, low partitioning into LE and high Bowen ratio (bH/LE) compared with other Arctic sites, associated with local climatic conditions dominated by onshore winds, slender vegetation with low transpiration activity

and relatively dry soils. Surface saturation vapour pressure deficit (Ds) was found to be an important variable

controlling within-year surface energy partitioning. Throughout the study period, we observe increasing H/Rn

and LE/Rn and decreasing G/Rn and b, related to increasing ALD and decreasing soil wetness. Thus, changes in summer-time surface energy balance partitioning in Arctic ecosystems may be of importance for the climate system. Keywords: energy budget, energy balance, Arctic, sensible heat, latent heat, ground heat, net radiation, climate change, global warming

1. Introduction Simmonds, 2010), will affect energy partitioning and hence the structure and functioning of Arctic terrestrial ecosystems The energy balance of northern high-latitude permafrost (Hinzman et al., 2005; Post et al., 2009). Changes in Arctic regions is crucial for most ecosystem processes in Arctic land energy balance partitioning may by itself induce further areas, including permafrost thermal conditions, plant feedback effects on the local and global climate system growth, microbial activity, carbon (C) and nutrient cycling, (Chapin et al., 2005). hydrology and geomorphology. Surface energy flux dy- Warming in the Arctic has accelerated during recent namics is regulated by a number of factors, including decades (Chapin et al., 2005; Overland et al., 2008). Obser- available radiation, meteorological conditions, surface char- vations from circumpolar Arctic permafrost monitoring acteristics and soil wetness (Boike et al., 2008; Westermann sites reveal increasing permafrost temperatures (Osterkamp, et al., 2009). Arctic climate warming, which has been 2005; A˚kerman and Johansson, 2008; Christiansen et al., estimated to be almost twice as large as the global average 2010; Romanovsky et al., 2010). Dependent upon site (Christensen et al., 2007; Graversen et al., 2008) due to a specific conditions in permafrost and hydrological regimes phenomenon known as Artic amplification (Screen and increasing active layer depths (ALD) and permafrost thawing may lead to wetter (Johansson et al., 2006) or dryer (Oechel et al., 1993) soil conditions. The observed increase *Corresponding author. email: [email protected] in shrub growth and associated increases in vegetation Responsible Editor: Annica Ekman, Stockholm University, Sweden. greenness and productivity across the circumpolar north

Tellus B 2014. # 2014 M. Lund et al. This is an Open Access article distributed under the terms of the Creative Commons CC-BY 4.0 License (http:// 1 creativecommons.org/licenses/by/4.0/), allowing third parties to copy and redistribute the material in any medium or format and to remix, transform, and build upon the material for any purpose, even commercially, provided the original work is properly cited and states its license. Citation: Tellus B 2014, 66, 21631, http://dx.doi.org/10.3402/tellusb.v66.21631

(page number not for citation purpose)

Scaling-up Climate Change Effects in Greenland

In ice-free Greenland, extensive knowledge and a mechanistic understanding of interactions between abiotic and biotic ecosystem components have been gained from observations collected in the Greenland Ecosystem Monitoring (GEM) sites during 10-20 years. This PhD project was initiated to facilitate an understanding of these interactions also in between observational sites using up-scaling. Since, the seasonal snow cover is a key driver of changes in Arctic ecosystems, snow is the main focus of this interdisciplinary study. The project aims; (i) to quantify the spatial and temporal changes and variability in snow characteristics across multiple spatial scales and time periods in ice-free Greenland, and (ii) to investigate the effects of these snow changes and variability on biotic components of the ecosystems. This was accomplished by combining ground-based observations, spatial-temporal snow and atmospheric modeling tools, and remotely sensed vegetation greenness in a stepwise up-scaling to gradually larger domain extents.

ISBN: 978-87-93129-39-9