New Perspectives on Climate, Earth Surface Processes and Thermal–Hydrological Conditions in High–Latitude Systems Juha Aalto

Home , Nivation

New perspectives on climate, Earth surface processes and thermal–hydrological conditions in high–latitude systems juha aalto

ACADEMIC DISSERTATION To be presented, with the permission of the Faculty of Science of the University of Helsinki, for public examination in lecture room 6, University main building, on 23 January 2015, at 12 o’clock noon.

Department of Geosciences and Geography A29 / Helsinki 2015 © Juha Aalto (synopsis) © Springer (Paper I) © Regents of the University of Colorado (Paper II) © John Wiley and Sons (Papers III and IV) Cover photo: Juha Aalto

Author´s address: Juha Aalto Department of Geosciences and Geography P.O.Box 64 00014 University of Helsinki Finland [email protected]

Supervised by: Professor Miska Luoto Department of Geosciences and Geography University of Helsinki, Finland

Co–supervised by: Dr. Ari Venäläinen Finnish Meteorological Institute, Finland

Reviewed by: Professor Jan Hjort Department of Geography University of Oulu, Finland

Professor Alfred Colpaert Department of Geographical and Historical Studies University of Eastern Finland, Finland

Discussed with: Assistant Professor Stephan Harrison College of Life and Environmental Sciences University of Exeter, United Kingdom

ISSN 1798-7911 ISBN 978-952-10-9469-9 (paperback) ISBN 978-952-10-9470-5 (PDF) http://ethesis.helsinki.fi

Unigrafia Helsinki 2015 Aalto J., 2015. New perspectives on climate, Earth surface processes and thermal–hydrological conditions in high–latitude systems. Unigrafia. Helsinki. 38 pages and 8 figures.

Abstract

Climate, Earth surface processes and soil ther- sis was based on unique datasets from north- mal–hydrological conditions drive landscape de- ern Fennoscandia: climate station records from velopment, ecosystem functioning and human Finland, Sweden and Norway; state–of–the–art activities in high–latitude regions. Such areas are climate model simulations; fine–scale field mea- characterized by large annual variations in air surements collected in arctic–alpine tundra; and temperatures, frost–related geomorphic activity remotely-sensed geospatial data. The study area and extreme spatial heterogeneity of the ground covers the main environmental gradients thus surface conditions due to complex topographical providing suitable study settings for theoretical and edaphic settings. These systems are at the and applied research in arctic–alpine environ- focal point of concurrent global change studies ments. as the ongoing shifts in climate regimes has al- Overall, the models successfully related the ready changed the dynamics of fragile and high- geographical variation in investigated high–lati- ly specialized environments across pan–Arctic. tude phenomena to main environmental gradi- This thesis aims to 1) analyze and model ex- ents. In paper I, accurate extreme air tempera- treme air temperatures, soil thermal and hydro- ture maps were produced, which were notably logical conditions, and the main Earth surface improved after incorporating the influence of lo- processes, ESP, (cryoturbation, solifluction, ni- cal factors such as topography and water bodies vation and palsa mires) controlling the function- into the spatial models. In paper II, the results ing of high–latitude systems in current and future show extreme variation in soil temperature and climate conditions; 2) identify the key environ- moisture over very short distances, while reveal- mental factors driving the spatial variation of the ing the factors controlling the heterogeneity of phenomena studied; and 3) develop methodolo- surface thermal and hydrological conditions. Fi- gies for producing novel high–quality datasets, nally, the modelling outputs in papers III and which can be used in other applications and dis- IV provided new insights into the determination ciplines, such as climatology, ecology and geo- of geomorphic activity patterns across arctic– science. To accomplish these objectives, spatial alpine landscapes, while stressing the need for analyses were conducted throughout a range of accurate climate data for predictive geomorpho- geographical scales by utilizing multiple statis- logical distribution mapping. Importantly, Earth tical modelling approaches, such as regression surface processes were found to be extremely and machine learning techniques. The robust- climatic sensitive, and drastic changes in geo- ness of these models was further increased by morphic systems towards the end of 21st cen- adopting an ensemble approach, where the out- tury can be expected. The increase over current puts of different statistical algorithms were com- temperatures by 2 ˚C was projected to cause a bined to give single agreement outputs. This the- near–complete loss of active ESPs in the high–

3 Department of Geosciences and Geography A29 latitude study area. This thesis demonstrates the applicability of spatial modelling techniques as a useful framework for multiple key challenges in contemporary physical geography. Moreover, with the model ensemble approach utilized, the modelling uncertainty can be reduced while represent- ing the local trends in response variables more robustly. Such a methodology is required, since complex topography, soil conditions and vegetation produce substantial spatial heterogeneity in arctic–alpine landscapes. This local variation in environmental conditions is integral, since it potentially buffers against climate change and aids in protecting the diversity in both Earth surface processes and in biota. In conclusion, this thesis provides important perspectives for the determination of multiple key phenomena typical for high–latitude regions, and the established statistical relationships based on extensive sampling are applicable over pan–Arctic regions. In future Earth system studies, it is essential to further assess the dynamics of arctic–alpine landscapes under changing climatic conditions and identify potential tipping–points of these sensitive systems. Forthcoming studies will require novel collaboration across disciplines, spatially comprehensive datasets and robust methodological approaches.

4 Acknowledgements

I’ve been lucky to be surrounded by amazing mother Pirjo, and my twin sister Tuikku, for re- people without whom this work would have been minding me what is important in life. Thanks greatly diminished. First of all, I’m indebted to for listening tirelessly to the latest updates on my supervisor, Prof. Miska Luoto for his crucial my work. Special thanks to Maija for love, help support and advice throughout this project. Dur- in the field and at the office, and all the support. ing the past three years, he has always provided This work was funded by the Geography me time for constructive and critical discussions Graduate School of the Academy of Finland. and not once I’ve needed to question his enthusi- Additionally, I would like to acknowledge the astic vision. My co-supervisor Dr. Ari Venäläinen funding by Oskar Öflunds Stiflelse, Societas Pro was also an important part of this project and I Fauna et Flora Fennica, an University of Helsinki thank him for his support. Grant, and Nordenskiold Sambundet. I would like to acknowledge the co-authors of the papers; I have had the privilege to work with Three other publications are also related to the exceptional talented scientists. Special thanks go PhD project, but have not been included in the to Dr. Peter C. le Roux; the multiple discussions thesis: with him at the slopes of Saana as well as at the 1) Aalto, J., Pirinen, P., Heikkinen, J., Venäläinen, office have had a big influence on my work. Ad- A. 2012. Spatial interpolation of monthly climate ditionally, I want to thank my reviewers, Prof. data for Finland: comparing the performance of Jan Hjort and Prof. Alfred Colpaert for encour- kriging and generalized additive models. The- aging and helpful comments. oretical and Applied Climatology 112, 99-111. I wish to thank Reija Ruuhela and Hilppa 2) le Roux, PC., Aalto, J., Luoto, M. 2013. Soil Gregow from FMI for support and providing moisture’s underestimated role in climate change me the opportunity to work on both sides of the impact modelling in low-energy systems. Global street. I’m particularly thankful to my FMI-wolf- Change Biology 19, 2965-2975. pack-leader Pentti Pirinen for helping me count- 3) Aalto, J., Luoto, M. 2014. Field-measured soil less of times with all sort of data related prob- temperature and moisture data improve fine-scale lems. I also acknowledge Dr. Ewan O’Connor models of periglacial activity patterns. Perma- for polishing the language of the final version frost and Periglacial Processes. Under review. of the manuscript. Kilpisjärvi biological research station has been my second home for the past three years. Therefore I would like to thank the staff (especially Pirjo!) for providing the facilities that a tired fieldworker needs. I also thank the research- ers and staff at the Department of Geoscienc- es and Geography for a friendly and supporting work atmosphere. On a personal note, I thank my family: my two beautiful daughters Emma and Venla, my

5 Department of Geosciences and Geography A29

6 Contents Abstract...... 3 Acknowledgements...... 5 List of original publications...... 8 Authors’ contribution to the publications...... 9 Abbreviations...... 10 List of figures...... 11

1 Introduction...... 12 1.1 Background and motivation...... 12 1.2 Climate change context...... 13 1.3 Methodological development...... 14 1.4 Objectives of the thesis...... 15 2 Material and methods...... 16 2.1 Study areas...... 16 2.2 Climate data...... 16 2.2.1 Climate station data...... 16 2.2.2 Derived climate indices...... 18 2.2.3 Global climate model simulation data...... 18 2.3 Field data...... 19 2.4 Geospatial data...... 19 2.5 Statistical analysis...... 21 3 Summary of original publications...... 22 Paper I...... 22 Paper II...... 22 Paper III...... 23 Paper IV...... 23 4 Discussion...... 24 4.1 The drivers of the investigated high–latitude phenomena...... 24 4.2 Methodological issues...... 28 4.3 Future perspectives...... 30 5 Conclusions...... 31 References...... 32 Errata...... 38

Publications I–IV

7 Department of Geosciences and Geography A29

List of original publications

This thesis is based on the following publications:

I Aalto, J., le Roux, P.C., Luoto, M. 2014. The meso–scale drivers of temperature extremes in high–latitude Fennoscandia. Climate Dynamics 42, 237–252.

II Aalto, J., le Roux, P.C., Luoto, M. 2013. Vegetation mediates soil temperature and moisture in arctic–alpine Environments. Arctic, Antarctic, and Alpine Research 45, 429–439.

III Aalto, J., Luoto, M. 2014. Integrating climate and local factors for geomorphological distribution models. Earth Surface Processes and Landforms 39, 1729–1740.

IV Aalto, J., Venäläinen, A., Heikkinen, R.K., Luoto, M. 2014. Potential for extreme loss in high–latitude Earth surface processes due to climate change. Geophysical Research Letters 41, 3914–3924.

The publications are referred to in the text by their roman numerals.

8 Authors’ contribution to the publications

І The study was planned by M. Luoto. J. Aalto conducted the data collection (data synthesis, GIS analyses) and statistical analyses (multivariate regression). J. Aalto and P. C. le Roux were responsible for preparing the manuscript, while all authors commented and contributed.

ІІ The study was planned by J. Aalto and M. Luoto. All authors contributed to field work. J. Aalto and P. C. le Roux conducted statistical analyses (multivariate regression). J. Aalto and P. C. le Roux were responsible for preparing the manuscript, while all authors commented and contributed.

ІІІ The study was planned by M. Luoto and J. Aalto. M. Luoto conducted the related field investigations while J. Aalto was responsible for the aerial photography in- terpretation, GIS analyses and statistical modelling. J. Aalto was responsible for preparing the manuscript, while M. Luoto commented and contributed.

ІV The study was planned by J. Aalto and M. Luoto. M. Luoto conducted the related field investigations while J. Aalto was responsible for the data synthesis and GIS analyses. J. Aalto conducted data analysis (down–scaling of the climate model data, statistical modelling). J. Aalto was responsible for preparing the manuscript, while A. Venäläinen, R. K. Heikkinen and M. Luoto commented and contributed.

9 Department of Geosciences and Geography A29

Abbreviations

ANN artificial neural network AUC area under the receiver operating characteristics curve CMIP3 coupled model intercomparison project phase 3 CTA classification tree analysis DEM digital elevation model ESP Earth surface process FDA flexible discriminant analysis FDD freezing degree days Ga giga–annum GAM generalized additive model GBM generalized boosting method GCM general circulation model GDM geomorphological distribution model GIS geographical information system GLM generalized linear model MARS multiple adaptive regression splines m a.s.l. meters above sea level MAXENT maximum entropy NW northwest PET potential evapotranspiration RF random forest SRE surface range envelope SRES special report on emission scenarios SW southwest TDD thawing degree days TMEAN mean annual temperature TSS true skill statistics TWI topographical wetness index VWC volumetric water content (%) WAB water balance

10 List of figures

Fig. 1 Forecasted changes in mean temperatures and precipitation by the end of 21st century in northern Europe, page 14 Fig. 2 The location of the study areas in high–latitude Fennoscandia, page 17 Fig. 3 The four key Earth surface processes (ESPs) occurring in the study area, page 20 Fig. 4 An example of the modelled temperature and geomorphic activity, page 25 Fig. 5 Schematic of the generalized patterns of high–latitude phenomena, page 26 Fig. 6 The key drivers determining air temperature, ground thermal–hydrological con ditions, and ESP activity patterns, page 27 Fig. 7 Ensemble forecasting of cryoturbation distribution under two different temperature conditions, page 29 Fig. 8 The forecasted occurrence of nivation along increasing temperatures in relation to local topographic conditions, page 31

11 Department of Geosciences and Geography A29

1 Introduction and Taylor, 1994; French, 2007; Bertoldi et al., 2010). These ground surface characteristics are the fundamental drivers of soil physical–chemi- 1.1 Background and motivation cal processes, such as microbial activity, carbon cycling, nutrient availability and the activity of High–latitude environments are strongly con- frost–related processes (Broll et al., 1999; Saito strained by their latitudinal and altitudinal loca- et al., 2009). Recent literature has recognized ex- tion which is generally characterized by a cold treme temperature and moisture variations with- climate and pronounced topographical complex- in distances less than one meter (Scherrer and ity (Bowman and Seastedt, 2001). In these sys- Körner, 2010; le Roux et al., 2013a), thus creat- tems (located roughly at 60–90˚N/S), climatic ing a mosaic of processes operating at fine spatial conditions, soil thermal and hydrological re- scales. Presumably, this spatial heterogeneity in gimes, and geomorphic activity drive the land- ground thermal and hydrological patterns is ulti- scape development, ecosystem functioning and mately driven by complex topographical settings, human activities (Johnson and Billings, 1962; (Isard, 1986; Takahashi, 2005) and, within a few Washburn, 1979; Parmesan et al., 2000; Post meters, may exceed long latitudinal or altitudinal et al., 2009). To cover these different aspects, gradients (Billings, 1974; le Roux et al., 2013a). this thesis is divided into three main themes: 1) A manifestation of the broad scale geologi- climate, 2) soil thermal–hydrological conditions cal and climatic factors and soil temperature and and 3) Earth surface processes. In the climate moisture, ESPs are characteristic features of arc- part of the thesis (paper I) the focus is on ex- tic–alpine landscapes (papers III and IV) (Wash- treme annual air temperature variations, which burn, 1979; French, 2007). ESPs in these regions potentially possess a major stress factor for biotic are mainly driven by the formation of ground ice processes (Chapin, 1983; Marchand et al., 2006; in the topmost soil layer and the spatial distri- Zimmermann et al., 2009), and partly control the bution of permafrost (French, 2007; Etzelmül- activity of Earth surface processes (ESP) in these ler, 2013). Geomorphic systems are important systems (French, 2007). Additionally, lower at- factors effecting landscape and vegetation dy- mospheric conditions are strongly coupled with namics in arctic–alpine systems (Malanson et al., soil temperatures and moisture conditions which 2012; Frost et al., 2013; le Roux et al., 2013b). are further important drivers of vegetation assem- In the third part of the thesis (papers III and blages and frost–driven Earth surface processes IV), the focus is on four key ESPs occurring in in high–latitude environments (Swanson et al., high–latitude regions: solifluction, cryoturbation, 1988; Scherrer and Körner, 2010). Strong varia- nivation and palsa mires. Geomorphic features tions in the annual temperature cycles (both air reflect the landscape evolution during the Ho- and in soil) are therefore defining elements of locene while processes still remain active today arctic–alpine regions, shaping both the abiotic (Allard, 1996). Solifluction is gradual mass wast- and biotic environment (Greenland and Losle- ing driven by freeze–thaw cycles of the upper- ben, 2001). most soil layers combined with gravity (Harris Soil thermal and hydrological conditions (the et al., 2001a; Matsuoka, 2005). These slow mass second part of the thesis, paper II) are the key movements create various features such as lobes, determinants of ecosystem dynamics and geo- steps and stripes (Matsuoka, 2001; Harris et al., morphic activity in high–latitude regions (Lloyd 2008). Cryoturbation (frost churning) refers to

12 the mixing of materials from various horizons of (Daly, 2006). This can further hinder the investi- the soil down to the bedrock due to freezing and gation of individual effects among variables and thawing, generating features such as patterned potentially causal links (Graham, 2003). ground and earth hummocks (Washburn, 1979; Matthews et al., 1998; French, 2007). Nivation 1.2 Climate change context represents local snow accumulation sites close- ly related to other hillslope processes including Over the last decades, high–latitude regions es- mass wasting, weathering and fluvial processes pecially have experienced a rapid increase in (Thorn, 1979; Wasburn, 1979; French, 2007). mean temperatures (Serreze et al., 2000; IPCC, Palsa mires are mire complexes with a perma- 2013). Simultaneously, the dynamics of these nently frozen core (Seppälä, 1986), located at landscapes has changed; for example the veg- the outer margins of the discontinuous perma- etation cover has increased notably in response frost zone in high–latitude peatlands (Luoto et to changing climatic conditions (Sturm et al., al., 2004). 2001; Tape et al., 2006) and fragile permafrost At regional scales (spatial resolution 10 km2– formations have started to degrade (Luoto and 1000 km2), climatic conditions, such as average Seppälä, 2003; Payette et al., 2004; Bosiö et al., temperature and precipitation, has been found to 2012). High–latitude regions may be very sen- control the geomorphic activity as well as soil sitive to climate warming due to their marginal thermal and hydrological patterns (Isard, 1986; location, specialized biota at their distributional Fronzek et al., 2006). These conditions, however, limits and the fact that the projected relative rise establish very general distributional patterns of in temperatures increases pole wards (Fountain response variables due to the spatial extent of the et al., 2012; IPCC, 2013) (Fig. 1). In addition, meteorological variables (Luoto et al., 2004; Pot- various land surface conditions and permafrost ter et al., 2013). Moreover, recent literature im- in these environments are found to be strongly plies that towards the landscape and local scale linked to the prevailing climate (Fronzek et al., (spatial resolution 1 km2 – 0.01 km2) other factors 2006; Etzelmüller et al., 2013; Farbrot et al., such as local topography, soil characteristics and 2013). Importantly, multiple environmental gra- vegetation, control the soil thermal–hydrologi- dients are potentially responding simultaneously cal patterns, with geomorphic activity filtering to the changing climate, with an as yet uncertain the coarse–grained effects of climate and cre- rate and amplitude (Chapin et al., 2005; Starr et ating distinct microclimatic spaces (Hjort and al., 2008; Virtanen et al., 2010). Luoto, 2009; Wundram et al., 2010; Scherrer Changes in arctic–alpine systems can trig- and Körner, 2011; Graham et al., 2012; Malan- ger multiple opposing feedbacks with poten- son et al., 2012). Further, the increasing level of tially global implications (Knight and Harrison, spatial heterogeneity causes strong coupling and 2013). For example, warmer and wetter climate feedbacks among environmental gradients (both conditions in the future can cause permafrost abiotic and biotic) (Isard, 1986; Ehrenfeld et al., thaw to accelerate further, amplifying the deg- 2005; le Roux et al., 2013a). From a method- radation of palsa mire complexes (Payette et al., ological point of view, these connections can be 2004; Fronzek et al., 2006). Similarly, dimin- challenging as the level of collinearity, i.e., the ishing frost–activity enables vegetation to re– statistical association between explanatory vari- establish, which in turn stabilizes the topmost ables, tends to increase towards fine spatial scales soil and further modifies heat fluxes and nutri-

13 Department of Geosciences and Geography A29 ent cycles (Kade and Walker, 2008). The under- 1.3 Methodological development standing of such feedbacks is essential for current global change impact studies. For example, The spatial modelling of response variables and changes in land surface processes across pan– the identification of the most influential predic- Arctic might effect ecosystem dynamics (Vir- tors is an essential theme in contemporary en- tanen, et al., 2010; Macias–Fauria and Johnson, vironmental and climate change impact studies 2013) and lower atmospheric conditions through (Guisan and Thuiller, 2005; Hjort and Luoto, various feedbacks related to changes in ground 2009; Boeckli et al., 2012). Modern statistical reflectance, heat fluxes and biochemical cycles approaches (e.g., Breiman, 2001; Venables and (Callaghan et al., 2011; Koven et al., 2011; Pear- Ripley, 2002; Luoto and Hjort, 2005; Elith et son et al., 2013). Moreover, permanently frozen al., 2008) provide flexible methods for capturing peat soils are major storages of organic carbon, multivariate relationships between response and and the thawing of these releases greenhouse environmental predictors across large geographical gradients (Walsh et al., 1998; Hjort and Luo- gases (CO2, CH4 and N2O) with potentially major effects on the climate system (Christensen et to, 2013), connections that are often accompa- al., 2004; Bosiö et al., 2012). In this thesis, the nied by nonlinearities and thresholds (Schumm, impacts of climate change were examined in a 1979; Phillips, 2003; Hjort and Luoto, 2011). predictive geomorphological context (paper IV). Moreover, different predictor variables often possess a distinct effective scale (Daly, 2006; Potter et al., 2013), indicating that the spatial extent of the predictors’ effect might vary, e.g.,

Figure 1. Forecasted changes in A) mean temperatures (∆ T) and B) precipitation (∆ RR) by the end of 21st century in northern Europe based on the ensemble of 19 general circulation models (GCM) (special report on emission scenarios [SRES] scenario A1B assuming CO2 emission roughly equal to 700 parts per million, baseline period 1971–2000). The black boxes indicate the approximate location of the study domain.

14 from regional to local (i.e., 1000 km2–0.01 km2). 1.4 Objectives of the thesis Therefore, a useful approach for examining the influence of various environmental factors on re- This thesis has three main objectives: firstly to an- sponse variables is through a hierarchical model- alyze and model the spatial variation in extreme ling perspective (Walsh et al., 1998; Albrecht and temperatures (paper I), ground thermal–hydro- Car, 1999; Pearson and Dawson, 2003). More logical conditions (paper II), and geomorphic precisely, this encompasses the integration of activity patterns (paper III), which are impor- different data sources e.g., climate, topography tant controllers of high–latitude environments. and soil into spatial models in stepwise man- Moreover, paper IV examines the climatic sen- ner (e.g., Pearson et al., 2004; Sormunen et al., sitivity of the four key ESPs, after modifying 2011). Consequently, such a modelling approach the prevailing temperature and precipitation re- helps to conceptualize and structure the effects gimes. Secondly, the aim is to identify the most of environmental drivers on response variables influential factors driving the spatial variation (papers I and III). in described response variables under realistic However, uncertainty in spatial modelling is multivariate settings. Thirdly, this thesis aims introduced to the study from a variety of sources, to provide new perspectives on the links and for example sampling errors, inaccuracies in geo- feedbacks among various environmental gradi- spatial datasets and modelling algorithms (Walsh ents operating in the arctic–alpine regions, while et al., 1998; Guisan and Zimmermann, 2000). developing methodologies for producing spatial The choice of the most suitable modelling tech- datasets to be used by other applications and dis- nique can be challenging since different statis- ciplines, such as ecology and geoscience. To ac- tical algorithms have their own strengths and complish the described objectives, spatial anal- weaknesses (Luoto and Hjort, 2005; Marmion yses were conducted across large geographical et al., 2008). Recently, the compilation of en- gradients utilizing modern statistical modelling semble prediction, i.e., the spatial forecast based approaches, comprehensive field–quantified ob- on the outputs of multiple different modelling servations and remotely sensed geospatial da- methods, has gained momentum in the environ- ta sources. By assessing the climatic sensitivity mental sciences (Marmion et al., 2009; Gallien of various land surface processes, this thesis at- et al., 2012). By utilizing such an approach, it is tempts to deepen public discussion about impacts possible to examine the majority trends in data of climate change in high–latitude regions, and while considering the methodology related un- further strengthen the scientific understanding of certainty (Araújo and New, 2007; Marmion et al., these environments. 2009). Moreover, this is especially useful when extrapolating the modelled present day patterns to future (or past) environmental conditions, as the predicted patterns of a single technique inside an ensemble might differ notably depending on the algorithm (Thuiller, 2004; Araújo et al., 2005; Fronzek et al., 2011). In this thesis, the ensemble modelling approach was utilized for predicting geomorphic activity patterns in current and future climate conditions (papers III and IV).

15 Department of Geosciences and Geography A29

2 Material and methods Fennoscandia is mainly covered by glacigenic till deposits, peat soils and bare rock, but sandy es- ker formations from Weichselian glaciations are 2.1 Study areas also widespread. Vegetation shifts from spruce (Picea abies) and Scots pine (Pinus sylvestris) In this thesis, temperature extremes, soil ther- dominated forests in the south, to mountain birch mal–hydrological conditions and ESPs were (Betula pubescens ssp. czerepanovii) in the north modelled, focusing on multiple study areas in of the study area. Alpine vegetation, above the high–latitude Fennoscandia (Fig. 2). This region tree line, is characterized by shrubs (e.g., Betu- covers extensive environmental gradients with la nana, Juniperus communis ssp. alpina) and low human disturbance, thus providing an ideal dwarf–shrubs (e.g., Empetrum hermaphroditum) location for spatial modelling studies and global (Sormunen et al., 2011; le Roux et al., 2014). change investigations. In general, the cold cli- In this thesis, each case study (papers I–IV) mate of this region is affected by its northern lo- represents a subset of the described geographi- cation, the strong continental–oceanic gradient cal domain (Fig. 2). The study area in paper I and the Scandes mountains (Tikkanen, 2005). is located between 68˚N and 70˚N (Fig. 2b in The study area represents the marginal zone of paper I). The two study sites in paper II are lo- discontinuous permafrost (Fig. 2a) (Christiansen cated approximately 100–200 m above the tree et al., 2010). Based on the climate station data line on the Saana massif, both at an elevation used in paper I (n=61; Fig. 2b), the mean annual of ca. 700 m a.s.l. (Fig. 2d). The study areas in temperature over the period of 1971–2000 was papers III and IV cover ca. 20 000 km2 and 26 –0.3 °C. However, due to a strong land–ocean 000 km2, respectively, mainly in north–western gradient and pronounced topographical varia- Finland (including minor parts from Sweden and tions, the mean annual temperature drops from Norway) (Fig. 2c). 4.3 °C (Borkenes, Norway; N 68°46’ S 16°15’, 36 meters above sea level [m a.s.l.]) to –3.6 °C 2.2 Climate data (Kilpisjärvi Saana, Finland; N 69°2’ S 20°49’, 1007 m a.s.l.) over a distance of approximate- 2.2.1 Climate station data ly 190 kilometers. Similarly, the mean annual precipitation sum is 549 mm, ranging from 323 The temperature dataset used in paper I cov- mm (Abisko Scientific Research Station, Swe- ers the period 1971–2000 and comprises 61 sta- den; N 68°21’ S 18°49’, 394 m a.s.l.) to 1049 tions covering the northern parts of Fennoscan- mm (Tromsø, Norway; N 69°39’ S 18°56’, 100 dia from the national observation networks of m a.s.l). Finland, Sweden and Norway. The observations Due to the long–term development of the bed- were collected from the climate databases of the rock (0.4–3.0 Ga), topography varies throughout Finnish Meteorological Institute and other na- the study domain with the highest fell tops lo- tional meteorological offices and research sta- cated in the geologically younger regions of the tions (Abisko Scientific Research Station 2012; study area of Caledonian rocks (i.e., Scandes at Norwegian Meteorological Institute, 2012; the west). The middle and southern parts con- Swedish Meteorological and Hydrological In- sist of eroded Precambrian bedrock with a gently stitute, 2012). Daily minimum, maximum and sloping landscape (Laitakari, 1998). This part of mean temperatures were extracted from station

16 andIII (n=61). The activity ofThe activity (n=61). The location of the study the of location The igure 2. igure F Fennoscandia. high–latitude in areas shows the study area in A P anel relation to the circumpolar extent of permafrost (based on the data in Brown following:as indicated 1998), al ., et continuous=90–100 % of the area covered by permafrost, discontinuous= 50–90 %, sporadic=10–50 isolated= <10 %, respectively. Panel B presents network station climatological the utilized in paper I geomorphic processes for papers I V was mapped at locations shown at The two study sitespanel C (n=1200). (NW = northwest; S W southwest) of the fine–scale dataset used in paper II is shown in panel D with 100–m– interval contour lines indicating elevation. meanthe represents map the C, I n annual absolute minimum temperature 1971–2000. of period the over

17 Department of Geosciences and Geography A29 records and were used to determine annual absolute extremes and mean annual temperatures. The average of the yearly values across the pe- PET = 58.93 Tabove 0 ˚C / 12, (3) riod 1971–2000 was subsequently used in paper × I to analyze the spatial variations of the temper- where T denotes the monthly mean temperatures. ature parameters. To predict the activity patterns of ESPs under current and future climates, these four indices 2.2.2 Derived climate indices were subsequently averaged to a spatial resolution of 200 m × 200 m (ArcGis 10.1 Zonal Sta- In papers III and IV, the distributional patterns of tistics –function). ESPs were investigated in relation to four climate indices: mean annual temperature (TMEAN), 2.2.3 Global climate model simulation data freezing degree days (FDD), thawing degree days (TDD) and water balance (WAB). Such Climate projections for the 21st century are based predictors have been shown to correlate with the on an ensemble of 19 global climate model sim- occurrence of permafrost features in high–lati- ulations obtained from the coupled model in- tude regions (Luoto et al., 2004; Fronzek et al., tercomparison project phase 3 (CMIP3) archive 2006). To obtain the climatic indices, monthly (Meehl et al., 2007). In this thesis, the future cli- mean temperatures and annual precipitation sum mate over two periods, from 2040 to 2069 and were modelled based on the statistical specifi- from 2070 to 2099, was calculated by adding cations (generalized additive model; GAM) in the mean change as predicted by the 19 gen- paper I, accounting for topography, water cover eral circulation models (GCM) to the observed and geographical location. The FDD and TDD 1971–2000 climate (spatial resolution 10 km × are based on effective temperature sum below 10 km) (Jylhä et al., 2009). The data represents and above base temperature (0˚C), respectively the average changes in temperatures and precipi- (Carter et al., 1991): tation under the B1, A1B and A2 emission scenarios (Nakićenović et al., 2000); B1 represent-

FDD = , if (Ti – Tb) < 0, (1) ing low, A1B medium and A2 high greenhouse 𝑛𝑛 𝑖𝑖=1 𝑖𝑖 𝑏𝑏 TDD = ∑ (𝑇𝑇 − 𝑇𝑇 ), if (Ti – Tb) > 0, (2) gas emissions, leading to CO2 concentrations at 𝑛𝑛 the end of the 21st century of roughly 540, 700 ∑𝑖𝑖=1(𝑇𝑇𝑖𝑖 − 𝑇𝑇𝑏𝑏 ) and above 800 parts per million, respectively. where, Ti denotes the mean temperature at day i, In order to match the resolution of the modelled

Tb the base temperature, and n the length of the baseline climate, the GCM data was bi–linear- summation period. However, as daily tempera- ly downscaled to 200 m × 200 m. For the sen- ture data was not available for this thesis, we es- sitivity analysis in paper IV, constant changes timated FDD and TDD using monthly data (fol- were applied to modelled monthly climate data; lowing e.g., Araújo and Luoto, 2007). WAB was changes in temperature from –2 °C to +6 °C at calculated as the difference between the mean 0.5 °C intervals, and of precipitation from –50 % annual precipitation sum and potential evapora- to 50 % at 10 % intervals were tested. In paper tion (PET) following Skov and Svenning (2004): IV, the climate predictors used (i.e., TMEAN, TDD, FDD and WAB) were re–calculated for each sensitivity and emission scenario analysis.

18 2.3 Field data High–resolution aerial photography (spatial resolution of 0.25m2; Land Survey of Finland, 2013) The spatial variation in soil temperature and and targeted field investigations were used for moisture (paper II) were investigated in a fine– the compilation of the geomorphological dataset scale study setting; on the Saana massif six sam- (see e.g., Luoto and Hjort, 2005). The activity pling grids were established at two sites, with of the ESPs (1=presence, 0=absence) was visu- each grid comprising 160 1 m2 plots in a regular 8 ally estimated based on the evidence in topsoil x 20 arrangement. Both response variables were material e.g., mass wasting, frost–heaving and measured on two consecutive days (northwest- cracking as well as soil displacement (le Roux ern site; 16 July 2012; southwestern site: 17 July and Luoto, 2014). In paper III, the geomorpho- 2012) from a depth of 10 cm using a handheld logical dataset comprised 1150 observations (531 digital temperature probe VWR–TD11 (VWR sites visited). Moreover, this dataset was com- international, Radnor, Penn., USA; accuracy of plemented in paper IV by increasing the total 0.8 °C). Volumetric soil moisture was measured number of observations to 1200. using a hand–held time–domain reflectometry sensor (FieldScout TDR 300; Spectrum Tech- 2.4 Geospatial data nologies, Plainfield, IL, USA) up to a depth of A wide range of geospatial information sourc- 10 cm, taking the mean of ca. 3 measurements es and geographical information system (GIS) per quadrat. techniques (ArcGis 10.1 Spatial analyst –func- In addition to the two response variables, tions) was utilized throughout this thesis. In pa- three groups of predictors (each comprising of per I, the extreme temperature variations were four variables) were measured and/or calculat- modelled in relation to topography, water cover ed: topography, soil characteristics and vegeta- and geographical location. The digital elevation tion (le Roux et al., 2013a; Mod et al., 2014). model (DEM) used is a global Gtopo with a spa- The four predictor variables related to topogra- tial resolution of 30 arc seconds (900 m; USGS, phy were: mesotopography (a measure of local 2004). The land cover data was obtained from topography: 1=depressions, 10=ridge tops; see the Corine land cover 2006 dataset with a spatial Billings, 1973; Bruun et al., 2006), slope angle, resolution of 100 m × 100 m (European Environ- potential annual direct radiation, and elevation. ment Agency, 2012). The water cover variables The four soil predictors were: soil temperature were spatially filtered with varying kernel sizes (when modelling soil moisture), soil moisture to account the gradually diminishing effects of (when modelling soil temperature), peat depth, the Arctic Ocean and lake cover (ArcGis 10.1 and the cover of rock. Finally, the four predictors Focal statistics –function). in the vegetation group were: vegetation volume, In addition to the climatic predictors de- biomass, cover of moss, and cover of lichen. The scribed in section 2.2, papers III and IV focus detailed description of the measuring protocol on relating the activity patterns of ESPs with and predictors is provided in paper II. topographical, soil and vegetation variables (pa- In the geomorphology part of this thesis (paper III). Two DEMs were utilized with differ- pers III and IV), the focus was on the activity ent spatial resolutions: 1) 25 m × 25 m (Land patterns of four ESPs occurring in arctic–alpine Survey of Finland, 2013; paper III) and 2) 30 regions: solifluction (Fig. 3a), cryoturbation (Fig. m × 30 m (NASA Land Processes Distributed 3b), nivation (Fig. 3c), and palsa mires (Fig. 3d). Active Archive Center LP DAAC, 2013; paper

19 Department of Geosciences and Geography A29

Figure 3. The four key Earth surface processes (ESPs) occurring in the study area: A) solifluction (20°59’E 68°59’N, ca. 810 m a.s.l); B) cryoturbation (21°4’E 68°59’N, ca 880 m a.s.l); C) nivation (local snow accumulation site, 20°48’E 69°3’N, ca. 800 m a.s.l); and D) palsa mire (21°25’E 68°43’N, ca. 408 m a.s.l). Photos: A–C, Author, and D, M Luoto.

IV). From these DEMs, four terrain parameters paper III: coniferous forest cover (%) and de- were derived: slope angle, topographical wetness ciduous forest cover (%). The vegetation data index (TWI) (Beven and Kirkby, 1979), poten- was compiled from the Corine 2006 land cov- tial annual direct solar radiation (MJ/cm2/a), and er –dataset with a spatial resolution of 25 m × total curvature (positive value indicating ridge 25 m (Finnish Environmental Institute, 2006). tops and negative values valley bottoms). TWI To obtain spatial predictions of the ESPs across was calculated using a Python script written by the two study areas, all the predictors described Prasad Pathak (Esri, 2013), whereas potential an- in this section were resampled to 200 m × 200 nual direct solar radiation (McCune and Keon, m resolution by spatial averaging (ArcGis 10.1 2002) was calculated using ArcView 3.2 Solar Zonal statistics –function). analyst –extension accounting for latitude, slope angle and slope aspect. The three soil predictors used in papers III and IV were peat cover, bare rock and sand cover. The soil classes were reclassified from the digital soil database (Geological Survey of Finland, 2010; spatial resolution of 20 m × 20 m) and the binary masks created were spatially filtered to a continuous scale. Additionally, two vegetation variables were included in the analysis of

20 2.5 Statistical analysis multiple times (e.g., 1000 runs in paper II) to account for sampling variability. This produces The response variables in relation to multiple a distribution of the evaluation metrics of inter- explanatory variables were examined within a est, rather than a single value. For continuous spatial modelling framework (see e.g., Guisan response variables (paper I and II), the model and Zimmermann, 2000; Marmion et al., 2008; evaluation was based on the amount of deviance Ridefelt et al., 2010), in which the geograph- explained by the models (i.e., the goodness of the ical distribution of a response variable is sta- fit) and the predictive performance i.e., how well tistically associated with present environmental the predicted values explained the observed ones. conditions. In this thesis, ten different statisti- In papers III and IV, the predicted occurrences cal modelling techniques were used (Thuiller of ESPs were evaluated using the area under the et al., 2013), ranging from parametric regres- receiver operating characteristics curve (AUC) sion to complex machine learning and classifi- and true skill statistics (TSS). AUC is a thresh- cation methods (e.g., Breiman, 2001; Venables old–independent measure of predictive accuracy and Ripley, 2002; Luoto and Hjort, 2005; Elith assessing the agreement between the observed et al., 2008). Such techniques included: general- presence/absence values and model predictions ized linear model (GLM), generalized additive (Fielding and Bell, 1997). The AUC values range model (GAM), artificial neural network (ANN), from zero to one; a model providing excellent classification tree analysis (CTA), generalized predictive performance has an AUC value higher boosting method (GBM), random forest (RF), than 0.9 and a fair model has AUC values ranging multiple adaptive regression splines (MARS), from 0.7 to 0.9 (see Swets, 1988). TSS is an ac- surface range envelope (SRE), flexible discrim- curacy measure that takes into account sensitiv- inant analysis (FDA) and maximum entropy ity (true positive rate) and specificity (true nega- (MAXENT). These modelling methods are de- tive rate) and is not sensitive to prevalence (i.e., scribed in more detail in papers I–IV. The en- the frequency of occurrence). TSS ranges from semble modelling approach adopted in papers –1 to 1, where 1 indicates perfect agreement, 0 III and IV combines the outputs of different al- random performance and –1, perfect disagree- gorithms (with varying performance) to a single ment (Allouche et al., 2006). agreement prediction. This technique allows for Additionally, two statistical techniques were accounting for the uncertainty related to differ- used to calculate the relative importance of envi- ent modeling techniques and their underlying as- ronmental factors in multivariate study arrange- sumptions (Walsh et al., 1998; Guisan and Zim- ments: 1) variation partitioning, based on GLMs, merman, 2000), further improving the predic- parcels out the independent or joint contribution tive performance of geomorphical distribution of variable groups (Borcard et al., 1992); while models (GDM) (Marmion et al., 2009; Gallien 2) variable importance in BIOMOD2 identifies et al., 2012). the relative importance of individual predictors Throughout this thesis, a cross–validation (Thuiller et al,. 2013). More precisely, the vari- approach was used to evaluate the spatial mod- able importance compares correlations between els. However, instead of splitting the data once the fitted values and predictions (thus indepen- for model calibration and evaluation (a common dent from the techniques used) where the pre- split–sample approach; Van Houwelingen and dictor of interest has been randomly permutated. Le Cessie, 1990), this procedure was repeated High correlation (i.e., the two predictions show

21 Department of Geosciences and Geography A29 little difference) indicates that the predictor per- the strongest effects due to the vertical lapse mutated is not considered important for the mod- rate. Additionally, temperature maxima were re- el. Subsequently, each of the predictors is ranked lated to the sea proximity, which tends to buf- based on the correlation coefficients and the pro- fer temperature variations and further decreas- portion of the relative influence is scaled from ing maximum temperatures. Thus, the highest 0 to 1. Hence, the higher the variable impor- temperature maxima in the study area are likely tance, the more influential the predictor is in the to occur at low elevation sites at considerable model. The two methods described are useful to distance from large water bodies. The models overcome statistical pitfalls such as collinearity mainly associated the spatial variation of mean among the predictors (Graham, 2003). All sta- temperatures with geographical location and to tistical analysis in this thesis was conducted in the proximity of the Arctic Ocean. These results the R statistical programming environment (R thus underline the governing role of oceanic and Development Core Team, 2011). topographical gradients as the key meso–scale drivers of temperature variations in this high–latitude region. Moreover, the valuable outputs of 3 Summary of original this paper were the accurate temperature maps publications describing the meso–scale variation in extreme and mean temperature conditions. Subsequently, Paper I these forecasts were used in the later part of this thesis as GDM input data (papers III and IV). Paper I focuses on investigating the meso–scale air temperature variations in topographically Paper II complex high–latitude environments. More precisely, mean annual absolute temperature max- In paper II, the fine–scale variation in soil tem- ima and minima, and mean annual temperature perature and moisture was quantified at an arc- in Northern Fennoscandia were modelled by tic-alpine site in northern Europe. Additionally, combining digital elevation model and remote- by utilizing GAM and GLM modelling and a ly sensed land cover data with 61 climate series robust cross–validation scheme, the effects of from northern Finland, Norway and Sweden. vegetation in controlling thermal and hydrologi- GAM and GLM were used to relate the varia- cal patterns were examined. tion in air temperature extremes to the predictors Soil temperature was found to vary by ≥ 5 and to partition the response to the most influ- ˚C and moisture by ≥ 50 % volumetric water ential environmental variables. content (VWC) over very short distances (≥ 1 The results indicate that minimum tempera- m). These results thus reflect the extreme spa- tures at the meso–scale are mainly driven by wa- tial heterogeneity of thermal and hydrological ter cover variables. The effects were positive with conditions in these arctic–alpine systems. The proximity to Arctic Ocean generally increasing inclusion of vegetation variables significantly minima, while the lowest temperatures are most improved both the model fit and the predictive likely to occur at topographical depressions such performance of the spatial models. While veg- as large mires and frozen lakes. In turn, maxi- etation showed marked effects on the studied mum temperatures were most strongly controlled parameters, the abiotic variables such as local by topography. Elevation, particularly, showed topography and soil characteristics were the most

22 influential predictors controlling the soil temper- ditions with seasonal frost and permafrost related ature and moisture patterns. Furthermore, the re- ESPs. However, three out of four models benefit- sults demonstrate how vegetation can mediate ted from the inclusion of local predictors. The edaphic conditions in arctic–alpine environments compiled ensemble predictions clearly show that at a fine–spatial scale. Pan–Arctic vegetation is the activity patterns of ESPs becomes increas- potentially sensitive to soil thermal and hydro- ingly detailed and patchy as additional predictors logical alterations due to e.g., complex physical– are included in the models. Therefore, the results chemical connections and feedbacks. Therefore, indicate that, while the climate of the study area the understanding of these fine–scale thermal and is the main component driving the coarse–scale hydrological patterns is crucial for future global activity patterns of ESPs, the local patchwork– change impact studies. like variability of ESPs is strongly constrained by the variation in topographical and soil condi- Paper III tions. Importantly, disregarding such local factors in GDMs will introduce additional bias into the In paper III, the aim was firstly to integrate ac- spatial analysis of ESPs. curate climate data and multiple local factors to develop realistic models of the four key ESPs Paper IV occurring in high–latitude regions: solifluction, cryoturbation, nivation, and palsa mires. Second- In paper IV, the climatic sensitivity of the four ly, we tested whether the spatial models of ESPs key ESPs (paper III) were examined, by mod- are improved after incorporating topographical, elling their occurrence in relation to changing soil and vegetation predictors to the climate–on- temperature and precipitation regimes, local to- ly models. Finally, the relative importance of pography and soil in northernmost Europe. Ad- these predictors was examined in a multivariate ditionally, utilizing ensemble modelling based arrangement. This study was based on a com- on ten statistical techniques, the distribution of prehensive geomorphic data set (n=1150) from ESPs across the 21st century under three green- northernmost Europe and modelled climate pre- house gas emission scenarios was forecasted. dictors. In addition, to reduce the model–relat- This study was based on empirical geomorpho- ed uncertainty and to correctly predict the oc- logical observations (n=1200), complementing currence of the ESPs, model ensembles based the dataset presented in paper III. The climate on ten statistical techniques were used. The cli- model data was based on an ensemble of 19 glob- matic predictors, such as TMEAN and WAB, al climate model simulations from the CMIP3 were derived from the modelled climate dataset archive. The future climate was calculated over based on the modelling specifications and out- two periods (from 2040 to 2069 and from 2070– puts of paper I. 2099) and for three (SRES) emission scenarios The results suggest that the occurrence of (B1, A1B and A2). ESPs can be modelled with good accuracy by uti- The results indicate that high–latitude Earth lizing only climate predictors. Furthermore, the surface processes are extremely sensitive to analysis highlighted the pronounced role of the changes in climatic conditions. Based on robust climatic predictors as the most influential vari- modelling assessments, the forecasts implied a ables for all four ESPs studied. This reflects the nearly complete disappearance of ESPs by the strong coupling of prevailing climatological con- end of the 21st century. Moreover, the increase

23 Department of Geosciences and Geography A29 in baseline climate conditions by 2 ˚C was found as a methodological advancement over tradition- to result in a drastic decrease in geomorphic ac- al descriptive research, by combining fine–reso- tivity in the study area. Similarly, studied ESPs lution databases with modern multivariate meth- strongly responded to manipulated precipitation odology to explain the observed landscape pat- conditions. The sites where geomorphic activ- terns of multiple high–latitude phenomena. ity was maintained were characterized by high Throughout this thesis, the importance of lo- altitude, compound topography and low radia- cal environmental heterogeneity, especially to- tion input generally associated with north–facing pography related, is underlined. Furthermore, slopes. These results thus stress the sensitivity of the spatial analysis implies a strong scale de- ESP activity to altering climatic conditions in arc- pendency of environmental drivers, where dif- tic–alpine systems. Moreover, this study shows ferent predictors are effective at their distinctive the potential buffering effects of topographical geographical distances (Pearson and Dawson, and soil conditions against climate change, fur- 2003). Effective scale is evident (Fig. 4) for ex- ther promoting the local persistence of ESPs. ample when modelling temperature extremes in The forecasted changes in geomorphic systems paper I; latitudinal position defines the amount of could have major impacts on both vegetation solar radiation received and the influence of large and regional climate system through changes in scale atmospheric circulation (here, the moving albedo, heat fluxes and biogeochemical cycles. low–pressure systems along the Polar Front). In This is the first study to examine the climatic turn, the location with respect to the ocean–land sensitivity of multiple ESPs at landscape–scale gradient determines the susceptibility to conti- while also accounting for local factors. nental air masses from the east (large effective scale; Barry and Chorley, 2009) (Fig. 5a). How- ever, established meso–scale temperature maps 4 Discussion differ considerably from the broad–scale patterns and are strongly constrained by topography (Fig. 4.1 The drivers of the investigated 4a). The complex topographical conditions, for high–latitude phenomena example slope inclination, aspect as well as ter- By modelling multiple key high–latitude phe- rain ruggedness produces substantial local varia- nomena, this thesis has provided new perspec- tion in e.g., radiation and wind conditions, (Rol- tives on climate, Earth surface processes and land, 2003; Scherrer and Körner, 2011; Yang et thermal–hydrological conditions across arctic– al., 2011; Pike et al., 2013) subsequently modi- alpine landscapes. The results, based on exten- fying the finer–scale variation in temperatures sive datasets and modern multivariate statistics, close to ground level. Importantly, this is where suggest diverse process–environment relation- it possesses the most relevance for ESP activ- ships which are organized in a hierarchical man- ity and vegetation (Daanen et al., 2008; Malan- ner. Additionally, the analysis has provided new son et al., 2012; le Roux et al., 2013a) (Fig. 4b; alarming evidence for the climatic sensitivity of 5b). Despite their local effects (i.e., small effec- high–latitude Earth surface processes. The spa- tive scale), disregarding additional factors such tial modelling framework and the ensemble–ap- as water cover and peat lands in meso–scale cli- proach utilized throughout the thesis have prov- mate models in these systems might severely re- en to be important analytical tools in the field of duce their accuracy (paper I). physical geography. Moreover, this work serves Paper II provided new insights into fine–

24 scale thermal and hydrological conditions of geneity in ground surface conditions is ecologi- the soil in topographically compound mountain cally significant, since the observed large differ- areas. Again, the complex terrain acts as initial ences in thermal and hydrological patterns in- filter for broad–scale temperature and moisture side very short horizontal distances (exceeding patterns, while soil characteristics and vegeta- coarse scale latitudinal and elevation gradients) tion partly define the patchwork–like variation challenges the prevailing global change estimates of soil thermal–hydrological regimes. This modi- (Scherrer and Körner, 2011; Scherrer et al., 2011; fication is potentially due to multiple inter–con- Lenoir et al., 2013; le Roux et al., 2013a). nections; for example different soil properties, The majority of the fine–scale heterogeneity such as pore size and the amount of organic mat- in high–latitude environments is a consequence ter, partly control thermal and hydrological prop- of the topographical control on the formation erties of soils (Wundram et al., 2010; Legates and persistence of perennial snow packs (Wash- et al., 2011). In turn, vegetation cover alters soil burn, 1979; Bruun et al., 2006; Kivinen et al., temperature and moisture patterns by modify- 2012). Depending on the position along the lo- ing transpiration rates, ground reflectance and cal topographical gradient (i.e., mesotopography, wetness in topmost soil layers (Cahoon et al., see Billings, 1973; Bruun et al., 2006) (Fig. 5b), 2012; Graham et al., 2012). This spatial hetero- these nivation sites strongly control the thermal–

Figure 4. An example of the modelled temperature and geomorphic activity patterns in the study area. The maps present the effect of sequential inclusion of predictors possessing different effective scales: A) geographical location, G, elevation, E, and water cover to minimum temperature forecasts (paper I); and B) climate, C, local topography, T, and soil characteristics to slope process distribution model (paper III). The black boxes in panel A) indicate the spatial domain of the slope process forecasts presented in panel B).

25 Department of Geosciences and Geography A29

Figure 5. Schematic of the generalized patterns of high–latitude phenomena along A) relief–continentality, and B) mesotopographical gradient. Mesotopography is a measure of local topography recorded on a 10–point scale (1=depressions, 10=ridge tops; following Bruun et al., 2006). Tmean=annual mean temperature (°C), Tmax=mean annual absolute maxima (°C), Tmin=mean annual absolute minima (°C), Rad=potential annual radiation (MJ/cm2/a). The relative air temperature variations in panel A are based on the modelling outputs in paper I. The response curves are based on bivariate GAM modelling with data utilized in paper III, while in panel B, the relative variation in soil temperature and moisture is based on the field measurements conducted at July 2012 (see paper II for details).

26 Figure 6. The drivers determining A) air temperature (Abs Tmin=mean annual absolute minimum temperature; Abs Tmax=mean annual absolute maximum temperature; Tmean=mean annual temperature), B) soil thermal and hydrological conditions, and C) ESP activity patterns in high–latitude regions. The width of the arrow indicates the strength of the effects, while the dashed arrows represent potential links and feedbacks among environmental variables not quantified in this thesis. In subfigure C, the edaphic group contains soil quality (peat depth and the cover of rock) and soil temperature and moisture marked with a star in subfigure B. hydrological regimes of uppermost soil, vegeta- environmental variables is evident; climate, be- tion assemblages (e.g., Litaor et al., 2008) and ing the most influential factor for the investigated the activity of Earth surface processes (Thorn and ESPs, producing coarse–scale geomorphic activ- Hall, 2002; French, 2007). Noteworthy, while ity patterns (Fig. 4b) (Luoto et al., 2004; Fronzek not quantified in this thesis, the various gradi- et al., 2006). This is due to the fact that high– ents investigated in relation to soil temperature latitude ESPs are strongly related to the forma- and moisture are strongly coupled (Fig. 6b). For tion of seasonal frost or permafrost (Washburn, example, complicated soil–topography–vegeta- 1979; Ballantyne and Matthews, 1982). There- tion interactions exist, where the direction of ef- fore, as identified in paper III, these processes fects can be highly ambiguous (see e.g., Legates in general require sub–zero mean air tempera- et al., 2010). tures with adequate soil moisture input for the Finally, as summarized in Fig. 6c, the envi- ground ice to form (Vliet-Lanoë, 1991; Luoto et ronmental factors controlling geomorphic activ- al., 2004; French, 2007). Moreover, this climate– ity at high–latitudes are diverse and the estab- ESP coupling is highlighted in paper IV, where lished links can be non–linear (Fig. 5a; papers III the model assessments suggest marked varia- and IV) (e.g., Hjort et al., 2007; Hjort and Luoto, tion in geomorphic activity patterns in relation 2011). Yet again, a hierarchical organization of to minor changes in temperature and precipita-

27 Department of Geosciences and Geography A29 tion regimes (Fronzek et al., 2006). integral part of modern environmental research Despite the role of climatic factors, papers (e.g., Heikkinen et al., 2006; Luoto et al., 2010; III and IV emphasize the importance of local Fronzek et al., 2011). factors, such as topography and soil character- Uncertainty is introduced into the spatial istics, controlling ESP activity (Fig. 4b). For ex- models from a variety of sources, such as field ample, cryoturbation and palsa mires are more measurements, variable selection, statistical al- likely to occur at low inclinations with adequate gorithms, model extrapolation and biased inter- moisture supply (Luoto et al., 2004; Hjort et al., pretation (Walsh et al., 1998; Guisan and Zim- 2007; Hjort, 2014). In turn, slope processes and mermann, 2000). GIS databases and tools of- nivation are active at topographically complex fer the widest possibilities and functionality for high–elevation sites with increased mass–move- sampling variables for spatial models (papers I, ment potential (solifluction) and low radiation III–IV) (Walsh et al., 1998). For example, ma- conditions (nivation) (Matsuoka, 2005; Kivin- ny GDMs are based on variables derived from a en et al., 2012) (Fig. 5a). For ESP activity, the DEM and a digital land cover classification (e.g., strong relationship between soil characteristics Hjort et al., 2007; Ridefelt et al., 2010). Despite and ground freezing is widely acknowledged their importance for concurrent environmental (Washburn, 1979; French, 2007). The potential research, the use of GIS–derived variables can be effects are derived from thermal and hydrologi- challenging due to systematic problems, which cal properties of the topmost soil layer mainly are often related to geo–referencing, interpola- related to the soil texture (i.e., grain distribution). tion, and calculation algorithms (Oksanen and Thus soils with small pore sizes (e.g., till) and Sarjakoski, 2005; Van Niel and Austin, 2007; increased water retention potential are suscep- le Roux et al., 2013a). tible to frost–action (Daanen et al., 2008). Veg- A common challenge in multivariate mod- etation presumably stabilizes the uppermost soil elling studies is collinearity among explana- layers and modifies the hydrology thus generally tory variables (Graham, 2003) and the use of limiting cryogenic activity (Stallins, 2006; Hjort indirect predictors with confounded causality and Luoto, 2009). (Guisan and Zimmermann, 2000). Additional- ly, geographical datasets are often spatially au- 4.2 Methodological issues tocorrelated thus violating the independent–as- sumptions of the statistical tests (Legendre et al., The spatial modelling framework adopted 2002). Statistical techniques exists which are able throughout this thesis has provided new in- to account for such pitfalls (Mac Nally, 2002; sights on process–environment relationships Dormann et al., 2007; 2013). In this thesis, spa- in arctic–alpine landscapes. Overall, the con- tial autocorrelation was routinely tested (papers I structed models showed consistently good per- and II) while the results imply that the methods formance, i.e., the predicted patterns matched used were sufficient. Moreover, the multivariate well with the observations, even though, when partitioning methods (papers I–III) proved to be coupled with strong correlation structures and useful tools for identifying the most important extensively sampled gradients, several factors predictors (or predictor groups) in multivariate might hinder the reliability of the modelling re- settings, where at least moderate level of collin- sults. Consequently, the recognition, quantifi- earity is expected. cation and presentation of uncertainties are an In papers III and IV, the model ensemble

28 Figure 7. Ensemble forecasting of cryoturbation distribution under two different temperature conditions (paper IV). The monthly temperatures were modified by –1 °C and +1 °C with respect to the 1971–2000 average. The maps in A show the number of models predicting for occurrence at each grid cell. Consequently, the maps in B demonstrate the majoritys’ vote approach (papers III and IV), where the predicted occurrence of cryoturbation is set to locations with 6 out of 10 modelling techniques voting for presence. In contrast, the number of votes less than six equals absence.

29 Department of Geosciences and Geography A29 approach was adopted to account for the meth- et al., 2006; Myers–Smith et al., 2011; Pauli et odological differences, as the results are not sen- al., 2012), vegetation–ESP interactions (Frost et sitive to the choice of a single modelling tech- al., 2013; Macias–Fauria and Johnson, 2013) and nique (Fig. 7) (Araújo and New, 2007; Marmion geomorphic systems (Christensen et al., 2004; et al., 2009; Luoto et al., 2010). Furthermore, this Kade and Walker, 2008). To obtain realistic fore- methodology enhanced the reliability of the pre- casts of the future land surface conditions in arc- dicted patterns as it combines the model outputs tic–alpine regions, a novel collaboration among of various algorithms (e.g., regression, classifica- scientists from across the disciplines is required. tion trees and machine learning) into a single con- In the field of geosciences, there is an emerg- sensus map, thus presenting the majority trend ing necessity for spatially comprehensive and in the response variables (Marmion et al., 2009; high–quality data sets as well as robust model- Gallien et al., 2012). Additionally, it allowed the ling exercises over different spatial scales (e.g., comparisons between modelling algorithms to Pope and Baeseman, 2014). be made. This thesis recognizes good model- Modern remote sensing and modelling tech- ling performance of e.g., GAM, GBM, RF and niques are already well established tools in pan- MAXENT, thus mostly agreeing with the previ- Arctic, permafrost and vegetation monitoring ous studies by e.g., Marmion et al., (2008; 2009), (e.g., Harris et al., 2001b; Stow et al., 2004; Ep- Heikkinen et al., (2012) and Hjort et al., (2014) stein et al., 2012). At the regional scale, one of the conducted in different study settings. Notewor- key contemporary research questions is the effect thy, the ensemble approach was especially useful, of changing land cover on the climate system when models were applied with predictor val- through alterations in ground reflectance (Chapin ues outside the range of the calibration sample et al., 2005; Callaghan et al., 2011, Loranty et (paper IV), i.e., space–time extrapolation (Fig. al., 2012). Locally, for example, the potentially 7) (Heikkinen et al., 2006). This increased the governing role of soil moisture on temperature robustness of the forecasts, as the predicted oc- variation as a key factor controlling high–latitude currences of response variables were set only in ecosystems has recently gained interest (Crim- locations with the highest agreement between mins et al., 2011; le Roux et al., 2013a). There- the modelling algorithms (i.e., majoritys’ vote, fore, in situ observation networks need to be es- papers III and IV; Fig. 7b) consequently ex- tablished to continuously survey the focal param- cluding sites with most uncertainty (Araújo and eters (e.g., soil temperature and moisture, snow New, 2007; Gallien et al., 2012). cover, permafrost, vegetation dynamics) driving the functioning of topographically complex ar- 4.3 Future perspectives eas. Topics such as local soil–vegetation–atmo-

sphere interactions, carbon cycling (e.g., CO2

High–latitude areas are a hot spot of current en- and CH4) and thawing of permafrost features vironmental research due to their observed and are under increasing interest due to their poten- forecasted responses to altering climate condi- tial implications on climate, landscape, ecosys- tions (Hinzman et al., 2005; Post et al., 2009; tems and human activities (Koven et al., 2011; Knight and Harrison, 2013; Pearson et al., Hipp et al., 2012; Huggel et al., 2012; Knight 2013). Moreover, a growing body of literature and Harrison, 2013). Additionally, recent litera- provides evidence for changes in both terrestrial ture has recognized the potential sheltering ef- and aquatic biota (Walther et al., 2002; Wrona fect of extreme spatial heterogeneity in land sur-

30 Figure 8. The forecasted occurrence of nivation for increasing temperatures in relation to local topographical conditions (paper IV); A) elevation (m a.s.l.) and B) slope angle (in degrees). Dotted black line indicates the mean of predictions based on ten modelling techniques, while inner and outer shaded areas represent the 50th and 95th percentiles, respectively. The diagrams present how the distributional limits of nivation are first restricted to elevated and steep sites before the suitable climate conditions for the activity completely disappears from the study area. face conditions against climate change (Scherrer a function of e.g., land cover. However, the use and Körner, 2011; Lenoir et al., 2013). As paper of such approach will require co–operation be- IV implies, these pockets of suitable microcli- tween physical geographers, ecologists, clima- mates created by local conditions (Fig. 8) might tologists, hydrologists, geophysicists and soil sci- be important for the preservation of both bio– and entists. Importantly, the spatial and holistic per- geodiversity under altering climatic conditions. spective combined with strong methodological Even though statistical modelling is a valu- capabilities opens new possibilities for geogra- able framework for spatial analysis, mechanistic phers to be an integral part of future environ- models describing, for example, heat and mass mental research. transfer processes might reduce uncertainties in predictions for future climate scenarios (Cramer et al., 2001; Fronzek et al., 2006). These pro- 5 Conclusions cess–based models, however, rely on heavy pa- rameterization and their spatial extent is limited By modelling multiple key high–latitude phe- (Lischke et al., 1998). Therefore, the next step nomena with modern statistical techniques and in spatial modelling studies would be the incor- an ensemble approach, this thesis has provided poration of statistical and mechanistic models new perspectives on process–environment rela- (e.g., Shipley et al., 2006 and Bocedi et al., 2014 tionships across arctic–alpine landscapes. Sub- in an ecological context) for an improved land sequently, the analyses highlighted the influence surface model. For example, the local develop- of local conditions, for example topography and ment of cryoturbated soils could be described soil, thus reflecting the spatial heterogeneity of with a mechanistic model based on thermody- these environments. Furthermore, this thesis pro- namics and hydrology, and further statistically vided strong evidence for the argument that Earth transferred over broad high–latitude regions as surface processes pan–Arctic are potentially ex-

31 Department of Geosciences and Geography A29 tremely climatically sensitive. rent temperature conditions was projected to The main results and implications of this the- cause a near–complete loss of active ESPs sis can be summarized as follows: in the arctic–alpine study area. As climate • The spatial modelling approach proved to be change proceeds, the suitable climatic spac- an effective tool for investigating the struc- es for the occurrence of ESPs may global- ture and functioning of high–latitude sys- ly shrink to a few topographically complex tems. The spatial variation in temperature ex- mountain areas. tremes, ground thermal–hydrological condi- • The methodology developed in this thesis can tions and ESPs were robustly modelled and be used to produce novel datasets for the use predicted. Importantly, the analytical mod- of other disciplines, for example climatolo- elling tools help to conceptualize and struc- gist, ecologist and geoscientist. ture the complex and hierarchical process– environment relationships in arctic–alpine regions. References • The results underline the governing role of oceanic and topographical gradients as key Abisko Scientific Research Station, 2012. http://www. polar.se/. (20.08.2012). meso–scale drivers of temperature variations Albrecht, J., Car, A., 1999. GIS analysis for scale–sen- in high–latitude region. Moreover, accurate sitive environment modelling based on hierarchy theory. In: Dikau, R., Saurer, H. (Eds.), GIS for temperature maps describing the meso–scale Earth Surface Systems. Gebrüder Borntraeger, variation in extreme and mean temperature Berlin, pp. 1–23. conditions were produced. Allard, M. 1996. Geomorphological changes and permafrost dynamics: Key factors in changing arctic • Extreme variation in soil temperature and ecosystems. An example from Bylot Island, Nuna- moisture was observed over short distanc- vut, Canada. Geoscience Canada 23: 205–212. Allouche, O., Tsoar, A., Kadmon, R. 2006. Assess- es, reflecting the strong spatial heterogene- ing the accuracy of species distribution models: ity of thermal and hydrological conditions in Prevalence, kappa and the true skill statistic (TSS). these systems. While vegetation has an im- Journal of Applied Ecology 43: 1223–1232. Araújo, M.B., Whittaker, R.J., Ladle, R.J., Erhard, M. portant role in mediating soil temperatures 2005. Reducing uncertainty in projections of ex- and moisture patterns at fine spatial scale in tinction risk from climate change. Global Ecology and Biogeography 14: 529–538. the arctic–alpine system, topography and soil Araújo, M.B., Luoto, M. 2007. The importance of characteristics revealed to be the most influ- biotic interactions for modelling species distribu- ential factors. However, when modeling soil tions under climate change. Global Ecology and Biogeography 16: 743–753. temperature and moisture, vegetation proper- Araujo, M.B., New, M. 2007. Ensemble forecasting of ties need to be explicitly considered. species distributions. Trends in Ecology & Evolu- tion 22: 42–47. • The modelling of ESPs will benefit from Ballantyne, C.K., Matthews, J.A. 1982. The devel- the inclusion of local predictors, such as to- opment of sorted circles on recently deglaciated pography and soil characteristics, with in- terrain, Jotunheimen, Norway. Arctic, Antarctic and Alpine Research 14: 341–354. creased transferability compared to climate– Barry, R.G., Chorley, R.J. 2009. Atmosphere, weather only models. This highlights the role of local and climate. Routledge. Bertoldi, G., Notarnicola, C., Leitinger, G., Endrizzi, factors determining the geomorphic activity S., Zebisch, M., Della Chiesa, S., Tappeiner, U. patterns across arctic–alpine landscapes. 2010. Topographical and ecohydrological controls • ESPs are potentially extremely climatic sen- on land surface temperature in an alpine catch- ment. Ecohydrology 3: 189–204. sitive. The increase of 2 °C relative to cur- Beven, K., Kirkby, M. 1979. A physically based vari-

32 able contributing area model of basin hydrology. Chapin, F.S. 1983. Direct and Indirect Effects of Tem- Hydrological Sciences Bulletin. 24: 43–69. perature on Arctic Plants. Polar Biology 2: 47–52. Billings, W. 1973. Arctic and alpine vegetations: simi- Chapin, F.S., Sturm, M., Serreze, M.C., McFadden, larities, differences, and susceptibility to distur- J.P., Key, J.R., Lloyd, A.H., McGuire, A.D., bance. Bioscience 23: 697–704. Rupp, T.S., Lynch, A.H., Schimel, J.P., Beringer, Billings, W. 1974. Adaptations and origins of alpine J., Chapman, W.L., Epstein, H.E., Euskirchen, plants. Arctic, Antarctic and Alpine Research 6: E.S., Hinzman, L.D., Jia, G., Ping, C.L., Tape, 129–142. K.D., Thompson, C.D.C., Walker, D.A., Welker, Bocedi, G., Palmer, S.C., Pe’er, G., Heikkinen, R.K., J.M. 2005. Role of land–surface changes in Arctic Matsinos, Y.G., Watts, K., Travis, J.M. 2014. summer warming. Science 310: 657–660. RangeShifter: a platform for modelling spatial Christensen, T.R., Johansson, T.R., Akerman, H.J., eco-evolutionary dynamics and species’ responses Mastepanov, M., Malmer, N., Friborg, T., Crill, to environmental changes. Methods in Ecology P., Svensson, B.H. 2004. Thawing sub–arctic per- and Evolution 5: 388–396. mafrost: Effects on vegetation and methane emis- Boeckli, L., Brenning, A., Gruber, S., Noetzli, J. 2012. sions. Geophysical Research Letters 31: L04501. A statistical approach to modelling permafrost dis- Christiansen, H.H., Etzelmüller, B., Isaksen, K., Ju- tribution in the European Alps or similar mountain liussen, H., Farbrot, H., Humlum, O., Johansson, ranges. Cryosphere 6: 125–140. M., Ingeman-Nielsen, T., Kristensen, L., Hjort, Borcard, D., Legendre, P., Drapeau, P. 1992. Partialling J. 2010. The thermal state of permafrost in the Out the Spatial Component of Ecological Varia- nordic area during the international polar year tion. Ecology 73: 1045–1055. 2007–2009. Permafrost and Periglacial Pro- Bosiö, J., Johansson, M., Callaghan, T.V., Johansen, cesses 21:156-181. B., Christensen, T.R. 2012. Future vegetation Cramer, W., Bondeau, A., Woodward, F.I., Prentice, changes in thawing subarctic mires and impli- I.C., Betts, R.A., Brovkin, V., Cox, P.M., Fisher, cations for greenhouse gas exchange–a regional V., Foley, J.A., Friend, A.D. 2001. Global response assessment. Climate Change 115: 379–398. of terrestrial ecosystem structure and function to Bowman, W.D., Seastedt, T.R. 2001. Structure and CO2 and climate change: results from six dynamic function of an alpine ecosystem: Niwot Ridge, global vegetation models. Global Change Biol- Colorado. Oxford University Press. ogy 7: 357–373. Breiman, L. 2001. Random forests. Machine learn- Crimmins, S.M., Dobrowski, S.Z., Greenberg, J.A., ing 45: 5–32. Abatzoglou, J.T., Mynsberge, A.R. 2011. Changes Broll, G., Tarnocai, C., Mueller, G. 1999. Interac- in climatic water balance drive downhill shifts in tions between vegetation, nutrients and moisture plant species’ optimum elevations. Science 331: in soils in the Pangnirtung Pass area, Baffin Is- 324–327. land, Canada. Permafrost Periglacial Processes Daanen, R.P., Misra, D., Epstein, H., Walker, D., Ro- 10: 265–277. manovsky, V. 2008. Simulating nonsorted circle Brown, J., Ferrians, O.J.J., Heginbottom, J.A., Mel- development in arctic tundra ecosystems. Journal nikov, E.S. 1998. Circum–arctic map of perma- of Geophysical Research: Biogeosciences (2005– frost and ground ice conditions. Boulder, Colo- 2012) 113: G3. rado USA: National Snow and Ice Data Center. Daly, C. 2006. Guidelines for assessing the suitability Bruun, H.H., Moen, J., Virtanen, R., Grytnes, J., Ok- of spatial climate data sets. International Journal sanen, L., Angerbjörn, A. 2006. Effects of altitude of Climatology 26: 707–721. and topography on species richness of vascular Dormann, C.F., McPherson, J.M., Araujo, M.B., plants, bryophytes and lichens in alpine commu- Bivand, R., Bolliger, J., Carl, G., Davies, R.G., nities. Journal of Vegetation Science 17: 37–46. Hirzel, A., Jetz, W., Kissling, W.D., Kuehn, I., Cahoon, S.M., Sullivan, P.F., Shaver, G.R., Welker, Ohlemueller, R., Peres–Neto, P.R., Reineking, J.M., Post, E. 2012. Interactions among shrub B., Schroeder, B., Schurr, F.M., Wilson, R. 2007. cover and the soil microclimate may determine Methods to account for spatial autocorrelation in future Arctic carbon budgets. Ecology Letters 15: the analysis of species distributional data: a review. 1415–1422. Ecography 30: 609–628. Callaghan, T. V., Johansson, M., Key, J., Prowse, T., Dormann, C.F., Elith, J., Bacher, S., Buchmann, C., Ananicheva, M., Klepikov, A. 2011. Feedbacks Carl, G., Carre, G., Garcia Marquez, J.R., Gruber, and interactions: From the Arctic cryosphere to B., Lafourcade, B., Leitao, P.J., Muenkemueller, the climate system, Ambio 40 (S1): 75–86. T., McClean, C., Osborne, P.E., Reineking, B., Carter, T.R., Porter, J.H., Parry, M.L. 1991. Climatic Schroeder, B., Skidmore, A.K., Zurell, D., Laut- Warming and Crop Potential in Europe – Pros- enbach, S. 2013. Collinearity: a review of methods pects and Uncertainties. Global Environmen- to deal with it and a simulation study evaluating tal Change–Human and Policy Dimensions 1: their performance. Ecography 36: 27–46. 291–312. Ehrenfeld, J., Ravit, B., Elgersma, K. 2005. Feedback

33 Department of Geosciences and Geography A29

in the plant–soil system. AnnualReview of Envi- alpine fellfield habitat, White Mountains, Cali- ronment and Resources 30: 75–115. fornia. Arctic, Antarctic, and Alpine Research 44: Elith, J., Leathwick, J.R., Hastie, T. 2008. A work- 288–295. ing guide to boosted regression trees. Journal of Graham, M.H. 2003. Confronting multicollinearity Animal Ecology 77: 802–813. in ecological multiple regression. Ecology 84: Epstein, H.E., Raynolds, M.K., Walker, D.A., Bhatt, 2809–2815. U.S., Tucker, C.J., Pinzon, J.E. 2012. Dynamics of Greenland, D., Losleben, M. 2001. Climate. In: Bow- aboveground phytomass of the circumpolar Arctic man, W.D., Seastedt, T.R. (Eds.), Structure and tundra during the past three decades. Environmen- function of Alpine ecosystem. Oxfrod University tal Research Letters 7: 015506. Press, Oxford. Esri, 2013. ArcScripts. www.arcsripts.esri.com. Guisan, A., Zimmermann, N.E. 2000. Predictive (23.9.2013). habitat distribution models in ecology. Ecologi- Etzelmüller, B. 2013. Recent Advances in Mountain cal Modelling 135: 147–186. Permafrost Research. Permafrost and Periglacial Guisan, A., Thuiller, W. 2005. Predicting species distri- Processes 24: 99–107. bution: offering more than simple habitat models. European Environment Agency, 2012. Corine Land Ecology Letters 8: 993–1009. Cover 2006 raster data. Harris, C., Rea, B., Davies, M. 2001a. Scaled physi- Farbrot, H., Isaksen, K., Etzelmüller, B., Gisnas, K. cal modelling of mass movement processes on 2013. Ground Thermal Regime and Permafrost thawing slopes. Permafrost Periglacial Processes Distribution under a Changing Climate in North- 12: 125–135. ern Norway. Permafrost Periglacial Processes 24: Harris, C., Haeberli, W., Vonder Mühll, D., King, L. 20–38. 2001b. Permafrost monitoring in the high moun- Fielding, A.H., Bell, J.F. 1997. A review of methods tains of Europe: the PACE project in its global con- for the assessment of prediction errors in conser- text. Permafrost Periglacial Processes 12: 3–11. vation presence/absence models. Environmental Harris, C., Smith, J.S., Davies, M.C.R., Rea, B. 2008. Conservation 24: 38–49. An investigation of periglacial slope stability in Finnish Environmental Institute, 2006. Corine Land relation to soil properties based on physical model- Cover 2006. (28.1.2014). ling in the geotechnical centrifuge. Geomorphol- Fountain, A.G., Campbell, J.L., Schuur, E.A.G., Stam- ogy 93: 437–459. merjohn, S.E., Williams, M.W., Ducklow, H.W. Heikkinen, R.K., Luoto, M., Araujo, M.B., Virkkala, 2012. The Disappearing Cryosphere: Impacts and R., Thuiller, W., Sykes, M.T. 2006. Methods and Ecosystem Responses to Rapid Cryosphere Loss. uncertainties in bioclimatic envelope modelling Bioscience 62: 405–415. under climate change. Progress in Physical Ge- French, H.M. 2007. The periglacial environment. John ography 30: 751–777. Wiley & Sons, Chichester. Heikkinen, R.K., Marmion, M., Luoto, M. 2012. Does Fronzek, S., Luoto, M., Carter, T.R. 2006. Potential the interpolation accuracy of species distribution effect of climate change on the distribution of models come at the expense of transferability? palsa mires in subarctic Fennoscandia. Climate Ecography 35: 276–288. Research 32: 1–12. Hinzman, L.D., Bettez, N.D., Bolton, W.R., Chap- Fronzek, S., Carter, T.R., Luoto, M. 2011. Evaluating in, F.S., Dyurgerov, M.B., Fastie, C.L., Griffith, sources of uncertainty in modelling the impact of B., Hollister, R.D., Hope, A., Huntington, H.P. probabilistic climate change on sub–arctic palsa 2005. Evidence and implications of recent climate mires. Natural Hazards and Earth System Sci- change in northern Alaska and other arctic regions. ences 11: 2981–2995. Climate Change 72: 251–298. Frost, G.V., Epstein, H.E., Walker, D.A., Matyshak, G., Hipp, T., Etzelmuller, B., Farbrot, H., Schuler, T.V., Ermokhina, K. 2013. Patterned–ground facilitates Westermann, S. 2012. Modelling borehole tem- shrub expansion in Low Arctic tundra. Environ- peratures in Southern Norway – insights into mental Research Letters 8: 015035. permafrost dynamics during the 20th and 21st Gallien, L., Douzet, R., Pratte, S., Zimmermann, N.E., century. Cryosphere 6: 553–571. Thuiller, W. 2012. Invasive species distribution Hjort, J., Luoto, M., Seppalä, M. 2007. Landscape models – how violating the equilibrium assump- scale determinants of periglacial features in sub- tion can create new insights. Global Ecology and arctic Finland: A grid–based modelling approach. Biogeography 21: 1126–1136. Permafrost Periglacial Processes 18: 115–127. Geological Survey of Finland, 2010. Digital soil data- Hjort, J., Luoto, M. 2009. Interaction of geomorphic base. Geological Survey of Finland: Espoo. and ecologic features across altitudinal zones Graham, E.A., Rundel, P.W., Kaiser, W., Lam, Y., in a subarctic landscape. Geomorphology 112: Stealey, M., Yuen, E.M. 2012. Fine–scale pat- 324–333. terns of soil and plant surface temperatures in an Hjort, J., Luoto, M. 2011. Novel theoretical insights

34 into geomorphic process–environment relation- nocai, C. 2011. Permafrost carbon–climate feed- ships using simulated response curves. Earth backs accelerate global warming. Proceedings of Surface Processes and Landforms 36: 363–371. the National Academy of Sciences of the United Hjort, J., Luoto, M. 2013. Statistical methods for geo- States of America 108: 14769–14774. morphic distribution modeling. Treatise on Geo- Laitakari, I. 1998. Peruskallion myöhäiset kehitysvai- morphology, Shroder J, Jr (ed). Academic Press: heet – miljardi rauhallista vuotta. In: Lehtinen, M., San Diego, 59–73. Nurmi, P., Rämö, T. (Eds.), Suomen kallioperä – Hjort, J., Ujanen, J., Parviainen, M., Tolgensbakk, J., 3000 vuosimiljoonaa. The Geological Society of Etzelmüller, B. 2014. Transferability of geomor- Finland, pp. 309–325. (In Finnish). phological distribution models: Evaluation using Land Survey of Finland, 2013. MapSite. www.kans- solifluction features in subarctic and Arctic re- alaisen.karttapaikka.fi. (19.9.2013). gions. Geomorphology 204: 165–176. Legates, D.R., Mahmood, R., Levia, D.F., DeLiberty, Hjort, J. 2014. Which environmental factors determine T.L., Quiring, S.M., Houser, C., Nelson, F.E. 2011. recent cryoturbation and solifluction activity in a Soil moisture: A central and unifying theme in subarctic landscape? A comparison between ac- physical geography. Progress in Physical Geog- tive and inactive features. Permafrost and Peri- raphy 35: 65–86. glacial Processes 25: 136–143. Legendre, P., Dale, M.R.T., Fortin, M.J., Gurevitch, Huggel, C., Clague, J.J., Korup, O. 2012. Is climate J., Hohn, M., Myers, D. 2002. The consequences change responsible for changing landslide activity of spatial structure for the design and analysis of in high mountains? Earth Surface Processes and ecological field surveys. Ecography 25: 601–615. Landforms 37: 77–91. Lenoir, J., Graae, B.J., Aarrestad, P.A., Alsos, I.G., IPCC, 2013. Summary for Policymakers. In: Stocker, Armbruster, W.S., Austrheim, G., Bergendorff, C., T.F., Qin, D., Plattner, G.K., Tignor, M., Allen, Birks, H.J.B., Bråthen, K.A., Brunet, J. 2013. Lo- S.K., Boschung, J., Nauels, A., Xia, Y., Bex, V., cal temperatures inferred from plant communities Midgley, P.M. (Eds.), Climate Change 2013: The suggest strong spatial buffering of climate warm- Physical Science Basis. Contribution of Working ing across Northern Europe. Global Change Biol- Group I to the Fifth Assessment Report of the ogy 19: 1470–1481. Intergovernmental Panel on Climate Change. Lischke, H., Guisan, A., Fischlin, A., Williams, J., Cambridge University Press, Cambridge, United Bugmann, H. 1998. Vegetation responses to cli- Kingdom and New York, NY, USA. mate change in the Alps: modeling studies. MIT Isard, S.A. 1986. Factors Influencing Soil–Moisture Press, Boston, Massachusetts. and Plant Community Distribution on Niwot Litaor, M., Williams, M., Seastedt, T. 2008. Topo- Ridge, Front Range, Colorado, Usa. Arctic, Ant- graphic controls on snow distribution, soil mois- arctic, and Alpine Research 18: 83–96. ture, and species diversity of herbaceous alpine Johnson, P.L., Billings, W.D. 1962. The Alpine Veg- vegetation, Niwot Ridge, Colorado. Journal of etation of the Beartooth Plateau in Relation to Geophysical Research: Biogeosciences 113: G2. Cryopedogenic Processes and Patterns. Ecologi- Lloyd, J., Taylor, J.A. 1994. On the Temperature De- cal Monographs 32: 105–135. pendence of Soil Respiration. Functional Ecology Jylhä, K., Ruosteenoja, K., Räisänen, J., Venäläinen, 8: 315–323. A., Tuomenvirta, H., Ruokolainen, L., Saku, S., Loranty, M.M., Goetz, S.J. 2012. Shrub expansion and Seitola, T. 2009. Arvioita Suomen muuttuvasta climate feedbacks in Arctic tundra. Environmental ilmastosta sopeutumistutkimuksia varten. Finn- Research Letters 7: 011005. ish Meteorological Institute, Helsinki, Finland. Luoto, M., Seppalä, M. 2003. Thermokarst ponds as (In Finnish). indicators of the former distribution of palsas in Kade, A., Walker, D.A. 2008. Experimental altera- Finnish lapland. Permafrost Periglacial Processes tion of vegetation on nonsorted circles: Effects 14: 19–27. on cryogenic activity and implications for climate Luoto, M., Fronzek, S., Zuidhoff, F.S. 2004. Spatial change in the arctic. Arctic, Antarctic, and Alpine modelling of palsa mires in relation to climate Research 40: 96–103. in northern Europe. Earth Surface Processes and Kivinen, S., Kaarlejärvi, E., Jylhä, K., Räisänen, J. Landforms 29: 1373–1387. 2012. Spatiotemporal distribution of threatened Luoto, M., Hjort, J., 2005. Evaluation of current statis- high–latitude snowbed and snow patch habitats tical approaches for predictive geomorphological in warming climate. Environmental Research Let- mapping. Geomorphology 67: 299–315. ters 7: 034024. Luoto, M., Marmion, M., Hjort, J. 2010. Assessing Knight, J., Harrison, S. 2013. The impacts of climate spatial uncertainty in predictive geomorphological change on terrestrial Earth surface systems. Nature mapping: A multi–modelling approach. Computer Climate Change 3: 24–29. & Geosciences 36: 355–361. Koven, C.D., Ringeval, B., Friedlingstein, P., Ciais, P., Mac Nally, R. 2002. Multiple regression and infer- Cadule, P., Khvorostyanov, D., Krinner, G., Tar- ence in ecology and conservation biology: fur-

35 Department of Geosciences and Geography A29

ther comments on identifying important predic- ronmental Research Letters 6: 045509. tor variables. Biodiversity and Conservation 11: Nakićenović, N., Joseph, A., Gerald, D., de Vries, B., 1397–1401. Fenhann, J., Gaffin, S., Gregory, K., Grubler, A., Macias–Fauria, M., Johnson, E.A. 2013. Warming– Jung, T.J., Kram, T., La Rovere, E.L., Michae- induced upslope advance of subalpine forest is lis, L., Mori, S., Morita, T., Pepper, W., Pitcher, severely limited by geomorphic processes. Pro- H.M., Price, L., Riahi, K., Roehrl, A., Rogner, ceedings of the National Academy of Sciences of H.H., Sankovski, A., Schlesinger, M., Shukla, P., the United States of America 110: 8117–8122. Smith, S.J., Swart, R., van Rooijen, S., Victor, N., Malanson, G.P., Bengtson, L.E., Fagre, D.B. 2012. Dadi, Z. 2000. Special Report on Emission Sce- Geomorphic Determinants of Species Compo- narios: a Special Report of Working Group III of sition of Alpine Tundra, Glacier National Park, the Intergovernmental Panel on Climate Change. U.S.A. Arctic, Antarctic, and Alpine Research NASA Land Processes Distributed Active Archive 44: 197–209. Center (LP DAAC), 2013. ASTER DEM. https:// Marchand, F.L., Kockelbergh, F., van de Vijver, B., lpdaac.usgs.gov/data_access. (13.03.2014). Beyens, L., Nijs, I. 2006. Are heat and cold re- Norwegian Meteorological Institute, 2012. eKlima. sistance of arctic species affected by successive https://eklima.met.no/. (20.08.2012). extreme temperature events? New Phytologist Oksanen, J., Sarjakoski, T. 2005. Error propagation 170: 291–300. analysis of DEM–based drainage basin delinea- Marmion, M., Hjort, J., Thuiller, W., Luoto, M. 2008. tion. International Journal of Remote Sensing 26: A comparison of predictive methods in modelling 3085–3102. the distribution of periglacial landforms in Finnish Parmesan, C., Root, T.L., Willig, M.R. 2000. Impacts Lapland. Earth Surface Processes and Landforms of extreme weather and climate on terrestrial biota. 33: 2241–2254. Bulletin of American Meteorological Society 81: Marmion, M., Hjort, J., Thuiller, W., Luoto, M. 2009. 443–450. Statistical consensus methods for improving pre- Pauli, H., Gottfried, M., Dullinger, S., Abdaladze, O., dictive geomorphology maps. Computers & Geo- Akhalkatsi, M., Benito Alonso, J.L., Coldea, G., sciences 35: 615–625. Dick, J., Erschbamer, B., Fernandez Calzado, R., Matsuoka, N. 2001. Solifluction rates, processes and Ghosn, D., Holten, J.I., Kanka, R., Kazakis, G., landforms: a global review. Earth–Science Re- Kollar, J., Larsson, P., Moiseev, P., Moiseev, D., views 55: 107–134. Molau, U., Molero Mesa, J., Nagy, L., Pelino, G., Matsuoka, N. 2005. Temporal and spatial variations Puscas, M., Rossi, G., Stanisci, A., Syverhuset, in periglacial soil movements on alpine crest A.O., Theurillat, J.P., Tomaselli, M., Unterlug- slopes. Earth Surface Processes and Landforms gauer, P., Villar, L., Vittoz, P., Grabherr, G. 2012. 30: 41–58. Recent plant diversity changes on Europe’s moun- Matthews, J.A., Shakesby, R.A., Berrisford, M.S., tain summits. Science 336: 353–355. McEwen, L.J. 1998. Periglacial patterned ground Payette, S., Delwaide, A., Caccianiga, M., Beauche- on the Styggedalsbreen glacier foreland, Jotun- min, M. 2004. Accelerated thawing of subarctic heimen, southern Norway: Micro–topographic, peatland permafrost over the last 50 years. Geo- paraglacial and geoecological controls. Perma- physical Research Letters 31: L18208. frost Periglacial Processes 9: 147–166. Pearson, R.G., Dawson, T.P. 2003. Predicting the McCune, B., Keon, D. 2002. Equations for potential impacts of climate change on the distribution of annual direct incident radiation and heat load. species: are bioclimate envelope models useful? Journal of Vegetation Science 13: 603–606. Global Ecology and Biogeography 12: 361–371. Meehl, G.A., Covey, C., Delworth, T., Latif, M., McA- Pearson, R.G., Dawson, T.P., Liu, C. 2004. Modelling vaney, B., Mitchell, J.F.B., Stouffer, R.J., Taylor, species distributions in Britain: a hierarchical inte- K.E. 2007. The WCRP CMIP3 multimodel data- gration of climate and land-cover data. Ecography set – A new era in climate change research. Bul- 27: 285–298. letin of the American Meteorological Society 88: Pearson, R.G., Phillips, S.J., Loranty, M.M., Beck, 1383–1394. P.S., Damoulas, T., Knight, S.J., Goetz, S.J. Mod, H.K., Roux, P.C., Luoto, M. 2014. Outcomes 2013. Shifts in Arctic vegetation and associated of biotic interactions are dependent on multiple feedbacks under climate change. Nature Climate environmental variables. Journal of Vegetation Change 3: 673–677. Science 25, 1024–1032. Phillips, J.D. 2003. Sources of nonlinearity and com- Myers–Smith, I.H., Forbes, B.C., Wilmking, M., Hal- plexity in geomorphic systems. Progress in Physi- linger, M., Lantz, T., Blok, D., Tape, K.D., Ma- cal Geography 27: 1–23. cias–Fauria, M., Sass–Klaassen, U., Lévesque, Pike, G., Pepin, N., Schaefer, M. 2013. High latitude E. 2011. Shrub expansion in tundra ecosystems: local scale temperature complexity: the example dynamics, impacts and research priorities. Envi- of Kevo Valley, Finnish Lapland. International

36 Journal of Climatology 33: 2050–2067. national Journal of Biometeorology 55: 645–654. Pope, A., Baeseman, J. 2014. The Cryosphere in a Schumm, S.A. 1979. Geomorphic Thresholds: The Changing Climate. Eos, Transactions American Concept and Its Applications. Transactions of Geophysical Union 95: 128–128. the Institute of British Geographers 4: 485–515. Post, E., Forchhammer, M.C., Bret–Harte, M.S., Cal- Seppälä, M. 1986. The origin of palsas. Geografiska laghan, T.V., Christensen, T.R., Elberling, B., Fox, Annaler 68A: 141–147. A.D., Gilg, O., Hik, D.S., Hoye, T.T., Ims, R.A., Serreze, M., Walsh, J., Chapin Iii, F., Osterkamp, T., Jeppesen, E., Klein, D.R., Madsen, J., McGuire, Dyurgerov, M., Romanovsky, V., Oechel, W., A.D., Rysgaard, S., Schindler, D.E., Stirling, I., Morison, J., Zhang, T., Barry, R. 2000. Observa- Tamstorf, M.P., Tyler, N.J.C., van der Wal, R., tional evidence of recent change in the northern Welker, J., Wookey, P.A., Schmidt, N.M., Aastrup, high–latitude environment. Climate Change 46: P. 2009. Ecological Dynamics Across the Arctic 159–207. Associated with Recent Climate Change. Science Shipley, B., Vile, D., Garnier, E. 2006. From plant traits 325: 1355–1358. to plant communities: A statistical mechanistic ap- Potter, K.A., Woods, H.A., Pincebourde, S. 2013. Mi- proach to biodiversity. Science 314: 812–814. croclimatic challenges in global change biology. Skov, F., Svenning, J.C. 2004. Potential impact of cli- Global Change Biology 19: 2932–2939. matic change on the distribution of forest herbs R Development Core Team, 2011. R: A language and in Europe. Ecography 27: 366–380. environment for statistical computing. R Founda- Sormunen, H., Virtanen, R., Luoto, M. 2011. Inclu- tion for Statistical Computing. sion of local environmental conditions alters high– Ridefelt, H., Etzelmuller, B., Boelhouwers, J. 2010. latitude vegetation change predictions based on Spatial Analysis of Solifluction Landforms and bioclimatic models. Polar Biology 34: 883–897. Process Rates in the Abisko Mountains, North- Stallins, J.A. 2006. Geomorphology and ecology: uni- ern Sweden. Permafrost Periglacial Processes fying themes for complex systems in biogeomor- 21: 241–255. phology. Geomorphology 77: 207–216. Rolland, C. 2003. Spatial and seasonal variations of air Starr, G., Oberbauer, S.F., Ahlquist, L.E. 2008. The temperature lapse rates in Alpine regions. Journal photosynthetic response of Alaskan tundra plants of Climate 16: 1032–1046. to increased season length and soil warming. Arc- le Roux, P.C., Aalto, J., Luoto, M. 2013a. Soil mois- tic, Antarctic, and Alpine Research 40: 181–191. ture’s underestimated role in climate change im- Stow, D.A., Hope, A., McGuire, D., Verbyla, D., pact modelling in low–energy systems. Global Gamon, J., Huemmrich, F., Houston, S., Racine, Change Biology 19: 2965–2975. C., Sturm, M., Tape, K. 2004. Remote sensing of le Roux, P.C., Virtanen, R., Luoto, M. 2013b. Geomor- vegetation and land–cover change in Arctic Tun- phological disturbance is necessary for predicting dra Ecosystems. Remote Sensing of Environment fine–scale species distributions. Ecography 36: 89: 281–308. 800–808. Sturm, M., Racine, C., Tape, K. 2001. Increasing shrub le Roux, P.C., Luoto, M. 2014. Earth surface processes abundance in the Arctic. Nature 411: 546–547. drive the richness, composition and occurrence Swanson, F., Kratz, T., Caine, N., Woodmansee, R. of plant species in an arctic–alpine environment. 1988. Landform effects on ecosystem patterns and Journal of Vegetation Science 25: 45–54. processes. BioScience 38: 92–98. le Roux, P.C., Pellissier, L., Wisz, M.S., Luoto, M. Swedish Meteorological and Hydrological Institute, 2014. Incorporating dominant species as proxies 2012. Smhi. http://www.smhi.se/. (20.08.2012). for biotic interactions strengthens plant commu- Swets, J.A. 1988. Measuring the accuracy of diagnos- nity models. Journal of Ecology 102: 767–775. tic systems. Science 240: 1285–1293. Saito, M., Kato, T., Tang, Y. 2009. Temperature con- Takahashi, K. 2005. Seasonal changes in soil tempera- trols ecosystem CO2 exchange of an alpine mead- ture on an upper windy ridge and lower leeward ow on the northeastern Tibetan Plateau. Global slope in Pinus Pumila scrub on Mt. Shogigashira, Change Biology 15: 221–228. central Japan. Polar Bioscince 18: 82–89. Scherrer, D., Körner, C. 2010. Infra–red thermom- Tape, K., Sturm, M., Racine, C. 2006. The evidence etry of alpine landscapes challenges climatic for shrub expansion in Northern Alaska and the warming projections. Global Change Biology Pan–Arctic. Global Change Biology 12: 686–702. 16: 2602–2613. Thorn, C.E. 1979. Ground Temperatures and Surficial Scherrer, D., Körner, C. 2011. Topographically con- Transport in Colluvium during Snowpatch Melt- trolled thermal–habitat differentiation buffers out – Colorado Front Range. Arctic, Antarctic, alpine plant diversity against climate warming. and Alpine Research 11: 41–52. Journal of Biogeography 38: 406–416. Thorn, C.E., Hall, K. 2002. Nivation and cryoplana- Scherrer, D., Schmid, S., Körner, C. 2011. Elevational tion: the case for scrutiny and integration. Progress species shifts in a warmer climate are overesti- in Physical Geography 26: 533–550. mated when based on weather station data. Inter- Thuiller, W. 2004. Patterns and uncertainties of spe-

37 Department of Geosciences and Geography A29

cies’ range shifts under climate change. Global Change Biology 10: 2020–2027. Errata Thuiller, W., Georges, D., Engler, R. 2013. biomod2: Ensemble platform for species distribution modeling. R package version 2.1.15. Paper II: Tikkanen, M. 2005. Climate. In: Seppälä, M. (Ed.), The physical geography of Fennoscandia. Oxford University Press, Oxford. Page 4, in the caption of Figure 2 is written “… USGS, 2004. Shuttle radar topography mission. U.S. (D) peat depth … ”. It should be changed to “… Department of the Interior. Van Houwelingen, J., Le Cessie, S. 1990. Predic- (D) soil depth…”. tive value of statistical models. Stat.Med. 9(11), 1303–1325. Van Niel, K.P., Austin, M.P., 2007. Predictive vegetation modeling for conservation: Paper III: Impact of error propagation from digital elevation data. Ecological Applications 17: 266–280. The citation “Hjort and Luoto, 2010” should be Venables, W.N., Ripley, B.D. 2002. Modern Applied Statistics with S. Springer, Berlin. changed to: Virtanen, R., Luoto, M., Rama, T., Mikkola, K., Hjort, J., Grytnes, J., Birks, H.J.B. 2010. Recent vegetation changes at the high–latitude tree line ecotone “Hjort and Luoto, 2011”. are controlled by geomorphological disturbance, productivity and diversity. Global Ecology and Similarly in the reference list, “Hjort J, Luoto M. Biogeography 19: 810–821. Vliet-Lanoë, V. 1991. Differential frost heave, load 2010. Novel theoretical insights into geomorphic casting and convection: converging mechanisms; process–environment relationship using simulat- a discussion of the origin of cryoturbations. Per- mafrost Periglacial Processes 2: 123–139. ed response curves.Earth Surface Processes and Walsh, S.J., Butler, D.R., Malanson, G.P. 1998. An Landforms 35: 363–371.” should be changed to: overview of scale, pattern, process relationships in geomorphology: a remote sensing and GIS perspective. Geomorphology 21: 183–205. “Hjort J, Luoto M. 2011. Novel theoretical in- Walther, G., Post, E., Convey, P., Menzel, A., Parme- sights into geomorphic process–environment re- san, C., Beebee, T.J., Fromentin, J., Hoegh–Guld- berg, O., Bairlein, F. 2002. Ecological responses lationship using simulated response curves. Earth to recent climate change. Nature 416: 389–395. Surface Processes and Landforms 35: 363–371.” Washburn, A.L. 1979. Geocryology – a survey of periglacial processes and environments. Arnold, London. Wrona, F.J., Prowse, T.D., Reist, J.D., Hobbie, J.E., Lévesque, L.M., Vincent, W.F. 2006. Climate change effects on aquatic biota, ecosystem structure and function. Ambio 35: 359–369. Wundram, D., Pape, R., Löffler, J. 2010. Alpine soil temperature variability at multiple scales. Arctic, Antarctic, and Alpine Research 42: 117–128. Yang, Z., Hanna, E., Callaghan, T.V. 2011. Modelling surface–air–temperature variation over complex terrain around Abisko, Swedish Lapland: uncertainties of measurements and models at different scales. Geografiska Annaler 93A: 89–112. Zimmermann, N.E., Yoccoz, N.G., Edwards, T.C., Meier, E.S., Thuiller, W., Guisan, A., Schmatz, D.R., Pearman, P.B. 2009. Climatic extremes improve predictions of spatial patterns of tree species. Proceedings of the National Academy of Sciences 106 (S2): 19723–19728.