Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | . 1 − dec Open Access 2 Discussions 895 300 km − , or 1 The Cryosphere − 2 14.0 % dec − 2180 2179 , and A. Schweiger 1 , M. Serreze 1 , A. Barrett 1 This discussion paper is/has beenPlease under refer review to for the the corresponding journal final The paper Cryosphere in (TC). TC if available. National Snow and Ice Data Center, Cooperative Institute for ResearchPolar in Science Environmental Center, Applied Physics Laboratory, University of Washington, cline in between SeptemberAssimilation System 1979 (PIOMAS). While and September 2012 ice extent according rebounded in to 2013, the partly Pan-Arctic Ice Ocean The last four decadessea have ice seen at a the remarkableover end the decline of 1979 in the through the meltThe 2013 spatial period, season. downward extent stands The trend of at linear hasand Arctic trend been warming for that linked September, as is tohouse calculated gases a a (e.g. Notz response combination and to Marotzke, ofSeptember 2012; 2012 increasing natural Stroeve et (the concentrations al., record 2012a). of low variability in Extent in atmospheric recorded the the for green- late satellite 1970s era) to was only early 50 1980s. % Volume of losses values were recorded even greater showing 80 % de- the decline of Arctic seamay ice be and realized. project the timing of when a seasonally ice-free Arctic 1 Introduction observations, the spatial patternsmodels. of The sea poor ice spatial thickness representationure are of of poorly thickness models represented to patterns represent in is detailserns most associated of with the the a mean transport atmospheric fail- and circulationalso pattern spatial that tend distribution gov- to ofthe sea underestimate multi-model ice. the ensemble The mean rate climatePan-Arctic trend models of Ice remains Ocean as ice Modeling within a and volume the Assimilation whole regarding uncertainty System. loss the These of from results ability that raise of 1979 concerns from CMIP5 the to models 2013, to though realistically represent the processes driving Arctic thicknessResearch distributions Programme Coupled from Model models Intercomparisonagainst Project participating observations Phase in 5 from the are submarines,that evaluated World aircraft the Climate mean and thickness satellites. distributions While from it’s the encouraging models are in general agreement with Abstract Using records from submarine, aircraft and satellite to evaluate climatesimulations model of Arctic sea iceJ. thickness Stroeve 1 Sciences, University of Colorado,2 Boulder, CO, USA Seattle, WA, USA Received: 14 March 2014 – Accepted: 7Correspondence April to: 2014 J. – Stroeve Published: ([email protected]) 28 AprilPublished 2014 by Copernicus Publications on behalf of the European Geosciences Union. The Cryosphere Discuss., 8, 2179–2212,www.the-cryosphere-discuss.net/8/2179/2014/ 2014 doi:10.5194/tcd-8-2179-2014 © Author(s) 2014. CC Attribution 3.0 License. 5 20 25 15 10 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | N, but ◦ cient coverage for the ffi erent models remains large. ff 2182 2181 ciently long time period. It was not until 2003 that ffi culty in evaluating thickness distributions in GCMs is the lack of consis- ffi ort, with 67 % of the models (or 16 out of 24) having a 1953–1995 mean Septem- This paper examines biases in contemporary Arctic sea ice thickness and ice volume A major di Realistically simulating the past and future evolution of the Arctic’s floating sea ice Coupled global climate models (GCMs) consistently project that if greenhouse gas ff We evaluate models using three criteria:distribution (1) of how sea well ice they thickness replicate based the on observed aggregating mean all available data across the Arc- period is further evaluated againstOcean volume estimates Modeling simulated and from Assimilation thethe Pan-Arctic System months Ice of (PIOMAS; March Zhang and and September. Rothrock, 2003) for 2 Methodology 2.1 Evaluation framework from the CMIP5 modelsevaluated making for the use whole of of all thecoverage. Since Arctic of Ocean radar these and measurements on data are(e.g. a influenced sets. March) regional by estimates Model basis of snowmelt, depending thicknesses ice we on thickness. are focus data Modeled on ice spring volume spanning the 1979 to 2013 2009, additional seaborne ice flights thickness as measurementshas part have since of become resumed, NASA’s available startingSpace Operation from in Agency’s IceBridge 2011 air- CryoSat-2. program. from Together,validation Arctic-wide the these of radar coverage data spatial altimeter provide patterns on-boardservations su of the have sea been European used ice to thickness.PIOMAS, provide In validation which of addition, in sea satellite turn icecomparison and reanalysis may with in-situ systems provide climate such ob- model a as long-term consistent trends record (Schweiger et of al., thickness 2011). and volume for and US submarines collected duringnear the the 1980s pole and as well 1990s,say, as mainly 2010). several covering moorings The the providing first region time European seriestimeter Remote in that fixed Sensing locations provided satellite (Lind- fields (ERS-1)only of included for estimated a the radar sea 1993 al- ice to thickness 2001 up period to (Laxon latitude et 81.5 al., 2003). Since the failure of ICESat in Sat, information was largely limited to data from upward looking sonars on board British Elevation Satellite (ICESat) Geoscience Laser Altimeter System (GLAS). Prior to ICE- with an overly thick initial (earlylater 21st century) than ice models cover will with tend initiallyet to thinner lose al., their ice summer 2010), given ice theimpacting the same on ice both climate the forcing thickness ice (e.g.of mass distribution Holland budget Arctic strongly and amplification ice determines – lossthe rate, the surface Arctic which outsized Ocean heat is compared rise in fluxes, to turn in lower a lower-tropospheric latitudes major air (Serreze driver et temperatures al.,tent over 2009). observations spanning aArctic-wide estimates su of thickness became available from NASA’s Ice, Cloud, and land However, historical trends fromobserved, most and model the ensemble spread members in simulated remain extent smaller between than di cover is one of thethickness most distribution challenging has facets emerged of as climate a modeling. key Simulating issue. the While sea it ice follows that climate models ties to Arctic residents,useful government projections agencies of and when industry.fidence While a in GCMs seasonally these can ice-free projectionsthe provide Arctic depends present-day Ocean in climate. may Stroeve part beWorld et on realized, Climate al. their con- Research (2012b) ability Programme found5 to Coupled that (CMIP5) reproduce models Model are features more participating Intercomparison consistent of e in Project with the Phase observations than thoseber from the ice previous extent CMIP3 falling within the minimum and maximum bounds of observed values. record. concentrations continue to rise,multiyear the ice eventual cover, outcome thatOcean will is, (e.g. be sea Stroeve a ice et complete will al., loss become 2007, of 2012b), but the presenting a both seasonal challenges feature and of opportuni- the Arctic a result of anomalously cool summer conditions, it was still the 6th lowest in the satellite 5 5 25 15 20 10 25 15 20 10 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | at 2 ers an op- ff 2184 2183 ). The archive consists of both atmosphere–ocean by 2100. It is perhaps a conservative scenario given 2 − ciently homogeneous to evaluate thickness or volume trends, ffi culty in our model evaluation, amplified by the piecemeal nature of the ffi To compare aggregate mean thickness (evaluation criterion 1), frequency distribu- Monthly mean thickness fields for the 1981 to 2010 period were calculated for ev- Monthly mean fields of sea ice thickness for 79 ensemble members of 27 climate A further di 650 ppm at the end of the century (e.g. Thompson et al., 2011), corresponding to http://cmip-pcmdi.llnl.gov/cmip5/ lations between individual ensembles within0.98) a (not model shown). This are suggests above that 0.9 the (and fragmented observational mostly record above o ble members can be used forthe validation natural of spatial variability patterns, of itmodels is the with important sea more to first ice than assess fivepatterns thickness ensemble and spatial members, Arctic-wide patterns we mean within evaluated thicknessrule the (Fig. the over variability 1). models. in the As For expected, spatial NorthEC-EARTH higher Atlantic variability and is near HadCM3) the stand theBeaufort out sea Sea because sector ice in of margin. CCSM4.is high Three However, less for local of than most variability, the of 8 such % the models an of models (CCSM4, the in evaluated, variability mean the over the . In addition, spatial pattern corre- could be extracted correspondingdata to sets. the For coverage example, ofmodel only each were grid used of cells when the evaluating withness observed how thicknesses distribution well thickness the from during models both the representused IceBridge IceBridge the to and aggregate time-period. evaluate thick- spatial the Regridded thickness model patterns fields (criterion were 2). To also ensure that model ensem- ensemble average. Spatial resolutionsmodelling grids vary to considerably coarse grids from withcomparisons a high-resolution roughly between ocean 1 models degree-by-1-degree spacing. andgridded To enable to the the observations, 100 km mean2002) Equal thickness using Area a Scaleable fields drop-in-the-bucket Earth approach. were (EASE)olution The re- grid of 100 km (Brodzik the resolution and coarser corresponds Knowles, model to grids. res- tions were derived forbutions each were model produced using for the each regridded observed mean thickness fields. field Separate so distri- that model thicknesses a radiative forcing ofcurrent 4.5 emission Wm rates. A listing of the models used canery be ensemble found member. in For Table 2. thickness models fields from having each more ensemble than for a one given ensemble model member, were mean averaged to form a single models from thethe CMIP5 Program archive for were Climate( Model downloaded Diagnosis from and the Intercomparison Earth data portal System (PCMDI) Grid of North Atlantic Oscillation) will likelysimulations. be Thus, out discrepancies of in phase modeled withbiases natural ice or variability thickness natural in can climate the either variability. model need Ideally, be climatologies to due of be to modeled sea compared model enough ice with (e.g., thickness observed 30 years) climatologies to that average out are most of of similar the natural length variability. and long centration, sea surface temperaturesensitive and to the ice atmospheric velocity. reanalysis Whilein-situ used, estimates observations, PIOMAS of submarines, is thickness airborne compare aand well measurements, model with Rothrock, and 2003; and from Schweiger satellites et (Zhang al., 2011; Lindsay et al.,ice 2012; thickness Laxon record, et is al.,with 2013). that the individual same years in yearsvariability CMIP5 in in model the the time observational observational do record. record not (such Imprints correspond as of that intrinsic associated natural with climate the phase of the (3) how well they replicateevaluations the make best use estimate of ofsubmarine, the trends aircraft- in and combined satellite-borne sea thickness instruments iceThis records introduced volume. record in from The the first is in-situ previous two not moorings, section. which and su is why we make use of the PIOMAS record. PIOMAS assimilates sea ice con- tic; (2) how well they replicate the observed spatial pattern of sea ice thickness; and sion scenarios were processed andscenarios the were same used. number RCP4.5 of∼ ensembles is for a both medium-mitigation emission scenario that stabilizes CO global climate models (AOGCMs)incorporate and interactive Earth biogeochemical cycles System into Models AOGCMs.2005) Both (ESMs), and the the future historical latter Representative (1850– Concentration which Pathway (RCP) 4.5 (2006–2100) emis- 5 5 25 15 20 10 15 20 10 25 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | ). ULS on moorings also measure 2186 2185 erent sources must be combined. We provide gridded fields at ff http://www.whoi.edu/beaufortgyre Upward Looking Sonar (ULS) instruments on bottom anchored moorings in the East- Unlike submarine sonar, satellite and aircraft radar and laser altimeters measure the Unclassified sonar data from US Navy and UK Royal Navy submarine missions pro- To evaluate criterion 3 (trends in ice volume using PIOMAS records), March ice vol- the Institute of Ocean Sciencesand (Melling end and in Riedel, 2005. 2008). Mooringsavailable Data in by records the the start Beaufort Beaufort in Gyre Gyre 1990 graphic region Exploration Institution are ( Project maintained based and at data made the Woods Hole Oceano- prior to comparison withwe the only CMIP5 use model data output.is from Following uncertain. US Schweiger cruises Submarine et because cruises al.cruises, the are (2011), defined processing designated as history as occurring for springthe between UK central March or cruise Arctic and summer. Ocean, data June. We away Most use from cruises the spring provide shallow data continental shelves. for ern ,thickness. Beaufort Moorings Gyre in and the Eastern Chukchi Beaufort Sea Sea provide and Chukchi further Sea estimates are of maintained by ice derived from draft estimateswater using densities, Archimedes and principle the withknown depth assumed and ice, of the snow snow Warren and onWenshahan snow (2007) the climatology estimate ice. (Warren an et In averageto thickness al., most bias direct 1998) cases, from observations is snow the of used. sonar depth 0.29 data Rothrock m. is compared and We un- subtracted this bias from the submarine data set (2007) fully describe the conversion of ice draft into thickness. Briefly, ice thickness is annually, is also hostedsurements by of NSIDC ice (Lindsay, 2013). draft Submarine (the depth sonars provide of mea- ice below sea level). Rothrock and Wenshahan (100 km resolution). vide the earliest estimates, starting infrom 1975 submarines and and ending in other 1993. platformstent Ice have thickness been format estimates collated by and processedproduce R. into the Lindsay a Unified consis- at SeaThe the Ice most University Thickness recent Climate of version DataWashington, Washington of Polar Record Polar Science the (CDR) Science Center. (Lindsay, submarine An 2010). Center data archive to was version obtained of the from CDR, the which University is of updated from 1975 throughporally the over present, the no wholeseries one of from this data observations period, source alone makingof is impossible. ice the To thickness continuous construction provide from spatially of di atwo a or long-term resolutions homogenous on picture, tem- the time estimates EASEPIOMAS grid (distributed (25 at and 25 100 km km) spatial that facilitate resolution) comparisons and with the both CMIP5 mean thickness fields ice-free portions of thegrid-cell thickness grid and cell. grid-cell area. Grid-celldomain, Grid to ice cell give volumes volume a were time is summed series simply for of the the monthly PIOMAS mean2.2 product ice of volume. mean Data: observations As previously introduced, the observedbination record of of in-situ, sea submarine, ice aircraft thickness and is satellite based data. on Although a records are com- available by natural variability. ume was calculated forof each the model PIOMAS ensemble estimates. membermodel Unlike corresponding grid. thickness, Ice to ice thickness the in volume domain the was CMIP5 calculated archive on is given the as native the grid cell mean including portunity to compare characteristics of the thickness patterns, which are less impacted models. Second, this regionaldetermined sea from height radar is andlocal adjusted laser sea using returns surface local from is sea open then surface water subtracted heights surfaces from within the leads. ice The or snow surface heights to give the height of bare-ice, snow-covered ice anddepending snow surfaces relative on to a instrument reference ellipsoid, these characteristics surfaces and above ice-surface thesurface. conditions. geoid This is The must done be heights in converted of two into steps. heights First, sea above surface the height local is sea estimated using gravity ice draft. The mosttained recent from versions the Polar of Scienceusing these Center. the in-situ Thickness same was ice method calculated as draft from applied estimates in-situ to were ice the also drafts submarine data. ob- 5 5 15 10 25 20 25 15 20 10 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | ). N. Kwok et al. ◦ erent techniques have ff N. The winter sea ice area ◦ erence between the two mea- ff www.meereisportal.de/cryosat N. We use the preliminary thick- ◦ and includes the Beaufort, Chukchi, East 2 2188 2187 km 6 10 × erently. For comparison with CMIP5, all observed thickness ff longitude grid. erent wavelengths and footprints, and di ◦ ff erences in sensor types and retrieval approaches. Radar and laser ff latitude by 0.5 ◦ Along with temporal sampling problems, the various thickness records have a range The in-situ mooring data, Airborne EM, IceBridge and submarine sonar track data All satellite-derived ice thickness fields were regridded as needed from their original Ice thickness is also measured using a combination of airborne electromagnetic (EM) ICESat, with its laser altimeter, provided the first thickness data set to cover almost Laxon et al. (2003) demonstrated the ability to retrieve ice thickness from the technologies use di spatial resolution were additionallyple produced flight for lines individual andvary, years cruise composites by tracks of combining in multi- icea a thickness range single from of year. IceBridge Since timesmonthly the and during averages. This time-periods the submarine will of observational data introduceisons a coverage intervals are temporal between and sampling based the do error on observations when notand making from exactly PIOMAS compar- these output. correspond data to sets and the monthlyof CMIP5 biases model due to di needed to be handledestimates di within 70 km ofcell a mean 100 km thickness. EASE To provide grid thedistributions, box center best all were coverage thickness to averaged to estimates compareerage give with for a field modeled all grid thickness for years the were period used of to record. calculate Grids a of single IceBridge av- and submarine data at 25 km gridded format to 25 kmThis and provides 100 km a EASE mean grids 1993–2001the using thickness five a field ICESat drop-in-the-bucket from years averaging. ERS-1, (spring a 2004to yearly to 2013). field 2009) Period-of-record and for mean each each fieldsculated, of of from the by ICESat three and first CryoSat CryoSat averaging years wereresolution. (2011 on additionally their cal- native grids and then regridding to 25 and 100 km surements provides the combinedusing snow-ice information thickness. about Ice snow thicknessavailable can thickness for the be and central and obtained density. western EMare Arctic derived Ocean also between ice included 2002 thicknesses and in 2012.Polar are the These Science data Unified Center. Sea Ice Thickness CDR and were obtained from the provides the height of the snow or ice surface. The di of the program but subsequently improves.timates Each IceBridge at track 40 gives m ice thickness spacing.and es- Kurtz Thickness et retrievals al. are (2013). Finally, detailed CryoSat-2lite thickness by radar estimates Kurtz are altimeter derived and with using Farrell a coverageness satel- (2011) extending product produced up by to the 88 Alfred Wegner Institute ( (2009) estimate an uncertainty ofan 0.5 m ongoing for airborne each laser 25 altimeter kmand mission grid the aimed cell. follow-on at Operation ICESat-2 bridging IceBridge scheduled the is tracks gap to of between launch ICESat ice in 2016. thickness, IceBridgeand generally April provides confined from individual 2009 to to present the (Kurtz western et al., Arctic 2012). Coverage Ocean is sparse during in March the early years covered by ERS-1 isSiberian, about Kara, 3.08 Laptev, Barentsprovided and as Greenland a single seas. mean ERS-1-deriveda field ice 0.1 averaged from thickness 1993 is to 2001 for the month ofthe March entire on Arctic Ocean. Thicknesses areby derived Kwok based on et the al.ded methodology described (2009). fields at The 25 ICESat km archive resolution. Estimates provides of five thickness years extend (2004–2009) up of to grid- 86 converted to ice thicknessconversion of using submarine Archimedes ice principle draft to in ice a thickness. similar13.8 same GHz way radar as altimetertic onboard the sea the ice ERS-1 thickness satellite from 1993 and to assessed 2001 changes up in to Arc- latitude 81.5 ice or total freeboard (the height of the snow or ice surface a.s.l.). Ice freeboard is induction instruments and laseris altimeter (Haas flown et above al, thetect 2009). sea the The distance ice instrument between package surface the by instrument helicopter. and The ice–water EM interface. The instrument laser is altimeter used to de- Data are available for 2011 through 2013 on the EASE-2 25 km grid. 5 5 25 15 20 10 25 15 20 10 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | erent ff . Given 3 km 3 10 × in March, PIOMAS is a suitable 1 − shore ice motion and ice divergence, dec ff 3 km 3 2190 2189 10 × Based on data comparisons and sensitivity studies, Schweiger et al. (2011) estimate The observed thickness patterns and magnitudes generally compare well with those Schweiger et al. (2011) found that PIOMAS ice thickness estimates agree well with In this paper, our focus is on representation of March ice thickness and volume. It the large observed volume trend oftool 2.8 for assessing long-term trends CMIP5spatial models. resolution Daily from ice PIOMAS volume estimates were atover averaged 25 the to km 1979 create monthly to means 2013 of record ice to volume compare with the CMIP5 output. PIOMAS was previously noted inare Schweiger smallest et with al. respectIceBridge, (2011). CryoSat to In and general the ERS-1 the submarine data. mean and errors ICESat dataan and upper are bound largest for for the the uncertainty of decadal PIOMAS trends of 1 data in the top scatterand plot Airborne includes EM. thicknesses Statistics from are in-situ summarized moorings, in US Table submarines 1. simulated by PIOMAS, providingsess further the confidence CMIP5 that volume PIOMAS trendseral can during negative be winter. (too However, used the thin) tonear scatter thickness as- the plots bias Canadian reveal Archipelago a in and gen- for PIOMAS north areas for of of higher Greenland). thin The ice. thickness reverse The values tends underestimation (found to of be thick true ice and overestimation of thin ice by estimates corresponding to the five observationalThe thickness data right sets hand used in column thisand of study. the Fig. observations 2 for showscolors each corresponding for scatter individual each year plots year of between of data,and except PIOMAS the for ERS-1, observations the which in-situ (plotted CDR, was as which provided includes di as 29 years mean of field data, over the entire time-period). The CDR To this end, the middle column of Fig. 2 (center column) shows the PIOMAS thickness is, therefore, useful tofrom assess ERS-1 and PIOMAS IceBridge, for which have this not period been used in in particular. previous comparison We studies. include data those from ICESattions (Kwok from et the al., seaOMAS ice 2009) ice thickness volume and CDR. and with They trends,of established and in-situ changes uncertainty concluded in and that estimates ice PIOMAS Airborneet for volume. provides PI- Comparisons al. EM useful estimates were (2013) observa- madefound compared for that derived concatenated all trends months time agree infurther within series the the arguing established of that year. uncertainty PIOMAS Laxon ICESat limits is from useful and PIOMAS, for CryoSat climate model data evaluation. and We assess CMIP5 ice thickness andPIOMAS volume (Zhang from and 1979 Rothrock, to 2003). 2013trations, PIOMAS against ice assimilates estimates motion observed from sea and iceice sea concen- volume surface on temperatures a into continuousNational a basis. Centers numerical The for model model Environmental to is Prediction forced estimate (NCEP) at atmospheric the reanalysis. surface by data from the Canadian Arctic Archipelago and Greenlandice motion where resulting there in is strong anof ridging. onshore the Mean thicknesses component Arctic are Ocean of lower where onleading there the to is Eurasian new a side ice persistent growth o inrecords open show water a areas. decline When through viewed as time a in whole, ice the thickness. 2.3 combined PIOMAS ice thickness patterns and volume trievals. This creates additional challenges intime-series. generating It a is consistent sea nevertheless icepatterns encouraging thickness of that all ice of thicknessdata, the (Fig. also records demonstrates 2: show persistence left similar offrom column), spatial the 1979 which general to pattern while present. of Mean lending Arctic thicknesses confidence sea are to ice greater thickness the along the northern coasts of the been used to estimate snow depth and density, which in turn impacts ice thickness re- 5 5 15 20 10 25 15 20 10 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | erent remotely-sensed data sets (Fig. 3). ff 2192 2191 erences in the spatial patterns of thickness (Fig. 4). Few models capture the ff erent observational data sets, such that if the spread of the observations for a given This good agreement with PIOMAS must be tempered by recognition of the pro- Given the limited temporal coverage of each observational data set, the above com- The CMIP5 models show the best agreement with the ICESat and CryoSat observa- In general, the thickness distributions from the models overlap those from each Ice thicknesses from the 27 individual CMIP5 models are presented as box and ff MIROC-ESM-CHEM). Instead, many models show a ridge of thick ice north of Green- CM5A-LR, MIROC5, and NORESM1-M) havebetween mean March the ice 10th thickness values andmean falling thicknesses 90th within the percentiles PIOMAS of interquartile the range (i.e. PIOMAS gray shading). nounced values, inter-model and spread 56 in % icelarge (15) thickness di have aggregated across thepattern Arctic of Ocean thin and ice closethickest to ice the along Eurasian the coast(i.e. Canadian and BCC-CSM1, several Arctic CanCM4, additionally Archipelago CanESM2, fail and CSIRO-Mk3, to place GISS-E2-R, northern the HadCM3, coast INMCM4, of Greenland parisons should be regarded aslong a qualitative PIOMAS assessment. On record theFig. (30 other years) hand, 2 facilitates the compares fairly moresame CMIP5 robust 1981 modeled comparisons. to ice The 2010 thicknesses time-period. bottom All with of but PIOMAS five estimates models over (EC-EARTH, FGOALS-g2, the IPSL- overall smaller mean thickness valuesis compared also to from a the time other period(e.g. data of Kwok sets. significant and ice The thinning Rothrock, coverage throughout 2009;with most Kwok ICESat, of all et the but Arctic al., one Ocean 2009; model90th Laxon (FGOALS-g2) percentiles has et a of al., mean the 2013). thicknessare within In observed slightly the comparison value. smaller 10th Mean than and for thicknessesMK3-6-0, ICESat, during EC-EARTH, resulting the FGOALS-g2, in CryoSat seven IPSL-CM5A-MR,mean models period MIROC5, thicknesses (CESM-CAM5, larger CSIRO- NorESM1-M) than having the 90th percentile of that from CryoSat. the models areCanadian underestimating Archipelago in sampled by regions the IceBridge of flights. thick icetions. north The ICESat of and Greenlandthe CryoSat and statistics thick integrate the ice more regions regions north of thin of ice Greenland along and with the Canadian Archipelago, resulting in CNRM-CM5, GFDL-ESM2M, MIROC ESM and MIROC-ESM-CHEM), suggesting that that show a negativeative bias bias compared with to respect the to in the situ IceBridge and ERS-1 data data (e.g. also BCC-CSM1, show CanCM4, a CanESM2, neg- additionally expect that the trendmodel in thickness thickness if should one be exists captured in in those the models. distributions of remotely-sensed data set. Thereels are have negative exceptions biases however. In inwith comparison particular, means to several the below mod- inspect the situ, to 10th ERS-1 and the percentile IceBridge infrom of data a situ sets, the thicker and observations. ice ERS-1 A regime data than negative is the bias not more with recent surprising re- two as decades. these However, some observations models sample period for CMIP5 isthe somewhat models arbitrary to as be we inonly cannot reflects phase how expect with well observed the the long-term natural naturaldi mean variability. variability This thickness fields in comparison in therefore theplatform/instrument models compare fall to the within thecaptures spread the for thickness. If a the given spread model, does we not conclude overlap, the then model there is a bias. We may The median springred thickness line, from together each with the observationalrange 10th (grey data shading). and 90th set percentiles is (green shown lines) and as the awhisker interquartile plots solid based on datathe for model interquartile years range 1981 to inand 2010, 90th thickness where percentiles, (25th the and boxes to themedian represent horizontal and 75th mean, bars percentiles), respectively. As and mentioned the asterisks earlier, the within whiskers 1981 each the to box 2010 10th define averaging time- the 3.1 Ice thickness We first compare observedthe and coverage corresponding CMIP5 to mean each sea of the ice di thickness fields averaged over 3 Results 5 5 25 15 20 10 25 15 20 10 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | (GISS-E2-R) 3 ) throughout the , respectively. 1 σ ects ice mobility. − ff 27.9 % dec − ects the surface , sea erent models: average March ice ff ff 9.9 % and − 2194 2193 (MIROC4h) to above 87 000 km 3 shore of Siberia suggest the presence of a strong anticyclonic ff ect the Beaufort Gyre ice drift and hence the thickness pattern. Models erent atmospheric reanalysis. Note that correlations between the reanaly- ff ff The CMIP5 multi-model ensemble mean March ice volume averaged over this period We directly evaluated the annual mean annual sea level pressure fields and the as- Kwok (2011) previously attributed deficiencies in ice thickness fields in the CMIP3 An analysis of spatial pattern correlations and root-mean-square error (RMSE) of the close match of thezling. ensemble average We with speculate the PIOMAS that average modeling is somewhat groups puz- participating in the CMIP5 collection may 1979–2013 time-period (Fig. 7). Thus, whenels viewed realistically as capture a the group, last thising three suggests that the decades PIOMAS mod- of provides changes a infind good Arctic good representation agreement ice of between volume, these PIOMAS assum- changes. iceble However, volume mean, while and ice we the volume CMIP5 varies multi-model substantiallyvolume ensem- between ranges di from around 14(Fig. 470 km 7 – dashed lines).observed Additionally, spatial as noted pattern earlier, of few thickness. models Given correctly the capture the wide range of CMIP5 model results, the decline in ice extenttant (e.g. Schweiger climate et indicator al., than 2011). extentbudget. Ice through The volume its current is rates direct also of connection athe ice more with 1979 volume impor- the to loss sea 2013 for ice period March energy from and PIOMAS September are calculated over agrees well with PIOMAS, and remains within 1 standard deviation (1 the failure of modelsin to terms capture of the biases observed inadditional issues the thickness such surface distribution as wind can near fields,ice surface be this stability rheology, explained is ocean that heat not a fluxes always and the the case. ice This thickness points3.2 itself to as this a Ice volume Recent studies suggest that because of thinning, sea ice volume is declining faster than also have poor spatialexception thickness is CCSM4. pattern While correlations CCSM4 showsness and good and large spatial the pattern RMSEs correlation lowest (Fig.mean in RMSE ice 4). sea of thick- The level all pressure thefails pattern models to capture (computed does the with not Beaufort respect feature Gyre to a and ICESat), the closed the Transpolar BSH Drift and Stream. Thus, the while mean part flow of 1, CCSM4, FGOALS, IPSL-CM5A-MR, MIROC-ESM-CHEM and NorESM1) generally Arctic (e.g. IPSL-CM5A-LR). Models that do not feature a closed BSH (e.g. bcc-csm1- drift that extends closeever, to the the presence coast, of allowing thickfrequency ice of ice to occurrence on of pile the a up Siberian specific on side atmospheric the circulation could upwind anomaly alsosociated pattern. side. be surface How- a geostrophic result wind offrom fields a four di higher in thesis CMIP5 themselves range models between (Fig. 0.91 anda 6) 0.99 closed (Table against 3). BSH, In fields general, though mostwards in models the feature some pole it (e.g. is CanCM4, CSIRO-Mk3-6-0, not MIROC-ESM), well-defined or (e.g. towards MPI-ESM-LR), the is eastern shifted to- models to thepressure inability and of hence the surfacestructured models winds. Beaufort to For Sea simulate example, Highadversely if the a (BSH) a observed in model pattern thewith fails of correct overly to thick location sea produce ice north level a o of well- Alaska, this will ice thickness between CMIP5 modelsous and model ICESat shortcomings. observations documents Spatial some patternmodels seri- correlations (CCSM4, are MIROC5 and less MRI-GCGM3) thanexceed (Fig. 0.4 0.7 5, for m top) all (Fig. and but 5, RMSE three cantly bottom). values smaller generally Given than that those these betweenpoor spatial ensembles correlations pattern from cannot correlations the be are same explainedmodels. signifi- model by suggests natural that variability the but rather a bias within the ice in the Beaufort/Chukchito and overestimate the ice Kara/Barents seas. thicknesscoast As over and a the underestimate whole, central ice theGreenland thickness Arctic and models along the Ocean tend Canadian the and Archipelago. North along American the coast Eurasian and north of land and across the Lomonosov Ridge towards the East Siberian shelf, with thinner 5 5 25 15 20 10 25 15 20 10 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | 1 − ) for 3 of the dec 3 σ km 3 km 3 10 × 10 × uncertainty of the 1.01 σ − 2.79 cult. relative to the 1979– − ffi 1 − erent from zero (i.e. 2 2.34 and ff − 7.3 % dec − between its 10 ensemble members . While all trends are negative, 10 en- 3 3 (or ects of temporal autocorrelation. This (MIROC5) as assessed over the period ff 1 km 1 )km − 3 3 − ort or success by these groups is randomly uncertainty of that value. ff yr 10 dec 2196 2195 3 erence in trends between PIOMAS and CMIP5 σ 3 × ff km 3 0.67 − 10 427.8 km × − 2.05 3.19 to − − ) but remains within 2 1 − (INMCM3) to 1 − yr 3 uncertainty with that from PIOMAS and can therefore be considered compat- σ 10.3 % dec − It is important to recognize that the di Finally, several ensembles show pronounced interannual variability in ice volume, To evaluate CMIP5 ice volume further, volume trends were computed using linear 49.3 km ume simulated by PIOMAS.and We find reduction that in the ice CMIP5 volume models over show a the general period thinning of observations. The CMIP5 ensemble Evaluating model skill is important givenframing the the large role debate that on the how modelmodels to projections more play address accurately in global hindcast environmental seaet change. ice While al., extent the 2012a), than CMIP5 trends thecern CMIP3 from that models most a (e.g. seasonally models Stroeve ice-freeclimate remain models. Arctic smaller Here state we than may have be observed, evaluatedmodels sea realized lending through ice sooner comparisons con- thickness than and with suggested volume observed by from records 27 CMIP5 of sea ice thickness and ice vol- while HadCM3 features a substantiallyits smaller 10 range ensemble members. ( Thisof makes the models identification based of on model how biases well or they the represent filtering observed trends di 4 Conclusions or from the factbutions that from the natural climate trend variability.3.1 in % For example, the of Day PIOMAS the (2012) 1979 timeMeridional attribute to Overturning series about 2010 Circulation 0.5 includes (AMOC). September to significant Theshown sea contri- in range ice Table of 2 extent provides trends a trend fortributor good to individual to indication models changes ice that natural volume in variability trends theto maybe over strongly a Atlantic the vary strong last in con- 35 the years. amount6-0 However of models trends natural themselves variability range seem in from their integration. The CSIR0-MK3- in March ice volume2013 of mean). This is smaller(or than the PIOMAS rate of decline of ensemble members can arise from systematic model errors, observational uncertainty together the individual ensemble members yields a multi-model ensemble mean trend variability in the ensembles likely reflects variability in atmospheric forcing. Averaging trend overlaps with zero). Neglectingfrom ensemble members zero, with 21 trends of indistinguishable slower, the and remaining two ensemble faster membersPIOMAS (IPSL-CM5A-LR have trend. and mean Nevertheless, MIROC5) March the than volume majoritytheir the 2 trends of the 2 ensembleible member with trends PIOMAS. overlap Only in 18PIOMAS ensembles model have trends. trends than cannot be reconciled with the with periods of increasing volume not captured by PIOMAS (not shown). Interannual 1979 to 2013 (Table 2shading and Fig. for 8). one The (darkshading corresponding PIOMAS does gray) trend not and is represent shown twoSchweiger the in et standard al. gray uncertainty (2011) deviations estimate in (light tosemble the be members gray). 1 PIOMAS have (10 Note volume trends estimates, the that are which gray insignificantly di least squares with aand test statistic the that observation combinesapproach, and the which standard accounts follows error for Santer(2012a) of the et to both examine the al. e ice model (2008),those extent was trends trends in previously compared both used tohypothesis the by is the CMIP3 Stroeve that and observed et the CMIP5 trend.Ice al. CMIP5 models As volume volume and in trends how trends Stroeve are during− et consistent March with al. from those (2012a), from individual the PIOMAS. ensemble null members range between torical ice extent and thicknesses.distributed, If then the a e closewhich match assimilates of observed sea the ice ensembletions, concentrations would mean and be volume is expected. and tuned to PIOMAS thickness volume, observa- each individually be working to construct and tune their models to match observed his- 5 5 25 20 10 15 25 20 10 15 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | , erent ff , 2009. , 2011. erent trends 10.1088/1748- ff , 2011. the coast of northern the spatial pattern of respectively) does not ff 10.1029/2011JC007004 1 − both dec 3 10.1029/2009GL039035 , 2009. ing, A.: Helicopter-borne measure- 10.1029/2011GL049303 km ffl 3 10.1029/2011GL049216 10 × 1.15 − 2198 2197 10.1038/NGEO467 and 1 − , 2010. dec 3 , 2009. km 3 10 × This work was funded under NASA Grant NNX12AB75G. , 2012. 3.6 − 10.1029/2009JC005312 10.1007/s00382-008-0493-4 Several techniques have been employed to sub-select models based on di Furthermore, while mean thickness and volume for the Arctic Ocean as a whole ap- records: 1958–2008, Geophys. Res. Lett., 36, L15501, doi: ning and volumedoi: loss of Arctic seation ice: icebridge, Geophys. 2003–2008, Res. Lett., J. 38, L20505, Geophys. doi: Res., 114, C07005, CMIP3 climate simulations,2011. J. Geophys. Res., 116, C00D05, doi: its future changedoi: as simulated by coupled climate models, Clim. Dynam., 34, 185–200, by 2100, Nat. Geosci., 2, 341–343, doi: 9326/7/3/034011 ments of sea icephys., thickness, 67, using 234–241, a 2009. small and lightweight, digital EMing system, Arctic climate, J. Geophys. Appl. Res. Geo- Lett., 38, L18501, doi: variability in Arctic sea ice extent, Environ. Res. Lett., 7, 034011, doi: in ice volume ( bode well for constraining climate models based on sea ice thickness patterns. 2009, 2012; Boe eteven al., if a 2009; model captures Massonnet the etthe seasonal model cycle al., in may 2012). extent, still or poorly It trendsthickness represent in is distributions. the extent clear and/or prevalent Indeed, volume, atmospheric we from circulation show ouror patterns that study and ice a that extent model reasonably mayand get correct, the the yet trend Canadian fail in to Archipelago.sea ice locate Only volume ice the two thickness thickest modelsMRI-CGCM3), and ice capture further north the reducing of general confidence Greenland on pattern in CMIP5 the of climate veracity atmospheric of models. circulation future The (MIROC5 projections fact based and that both models display rather di a seasonal ice cover.thickness Thus distribution, as maybe ice become more thins, relevant. the ability of modelsmetrics of to model represent performance the duringuncertainty spatial the as historical time-period to in when an an attempt to ice-free reduce Arctic may be realized (e.g. Wang and Overland, poorly represented. Many models failGreenland to and locate the the Canadian thickest ice ArcticShelf. o Archipelago Part and of thinner the ice explanation lies overprevailing in the atmospheric model East circulation deficiencies Siberian over in the representing Arctic. theof details This ice of is the a extent critical areHolland failure as et strongly projections al., related 2010; to(2011) e.g. suggest the Holland that and initial the Stroeve, ice variance 2011).the of winter-spring thickness Moreover, Holland September ice pattern thickness sea and increases distribution ice Stroeve as extent (e.g. the anomalies ice-cover explained thins by and transitions towards the PIOMAS values. Althoughis the strikingly Arctic-wide similar ensemble tomodels mean the exist. ice PIOMAS volume sea and ice trend volume and trend,pears huge well variations represented among by many of the models, spatial patterns of sea ice thickness are mean ice volume trend over the 1979–2013 is smaller but within the uncertainties of Kwok, R. and Rothrock, D. A.: Decline in ArcticKwok, sea R., ice thickness Cunningham, from submarine G. and F., ICESat Wensnahan, M., Rigor,Kurtz, N. I., T. and Zwally, Farrell, H. S. L.: J., Large-scale surveys and of snow Yi, depth on D.: arctic Thin- sea ice from opera- Kwok, R.: Observational assessments of Arctic Ocean sea ice motion, export, and thickness in Boe, J., Hall, A., and Qu, X.: September sea-ice cover in the Arctic ocean projected to vanish Holland, M. M. and Stroeve, J. C.: Changing seasonal sea ice predictorHolland, relationships in M. a M., chang- Serreze, M. C., and Stroeve, J.:. The sea ice mass budget of the Arctic and References Haas, C., Lobach, J., Hendricks, S., Rabenstein, L., Pfa Acknowledgements. Day, J. J., Hargreaves, J. C., Annan, J. D., and Abe-Ouchi, A.: Sources of multi-decadal 5 5 25 20 15 10 25 15 20 10 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | , , 2013. , 2007. , 2007. , 2012. 10.1007/s10584- 10.5194/tc-7-1035- , 2011. 10.5194/tc-3-11-2009 10.7265/N5D50JXV , 2003. , 2012. 10.1029/2007GL029703 10.1175/JTECH2097.1 10.1029/2012GL052868 , 2009. 10.1029/2011JC007110 2200 2199 , 2013. , 2012a. 10.1038/nature02050 10.1029/2012GL051094 , 2012. , 2012b. , 2011. , 1998, updated 2006. 10.1002/grl.50193 10.1029/2009GL037820 , 2008. 10.1029/2012GL052676 , 2011. 10.5194/tc-6-1383-2012 , 2013. 10.1002/joc.1756 10.1029/2011JC007084 10.1007/s10584-011-0101-1 10.7265/N54Q7RWK tribution model in generalized2003. curvilinear coordinates, Mon. Weather Rev., 131, 845–861, 1814–1829, 1998. of surface-based Arctic amplification,2009. The Cryosphere, 3, 11–19, doi: faster than forecast, Geophys. Res. Lett., 34, L09501, doi: Lett., 36, L07502, doi: observed temperature trends indoi: the tropical troposphere, Int. J. Climatol.,tainty 28, 1703–1722, indoi: modeled Arctic sea ice volume, J. Geophys. Res.-Oceans, 116, C00D06, edited by: Dyer, I. and Chryssostomidis,179–195, C., Hemisphere 1984. Publishing Corp., Washington, D. C., CMIP5 models, Geophys. Res. Lett., 39, L18501, doi: Trends in Arctic sea ice39, extent L16502, from doi: CMIP5, CMIP3 and observations, Geophys. Res.Arctic’s rapidly Lett., shrinking sea ice cover:doi: a research synthesis, Clim. Change, 110, 1005–1027, Bond-Lamberty, B., Wise, M.stabilization A., of Clark, radiative L. forcing011-0151-4 E., by and 2100, Edmonds, Clim. J. A.: Change, RCP4.5: 109, a 77–94, pathway doi: for Barriat, Constraining P.-Y.: projections of1394, summer doi: Arctic sea ice, The Cryosphere, 6, 1383– National Snow and Ice Data Center, Boulder, Colorado, USA, DigitalGeophys. media, Res. 2008. Lett., 39, L08502, doi: Arctic sea ice driftdecline, acceleration: J. consequences Geophys. Res., in 116, terms C00D07, doi: of projected sea ice thinning and Panzer, B., and Sonntag, J.Operation G.: Sea IceBridge ice airborne thickness, data, freeboard, The and snow Cryosphere, depth 7, products 1035–1056, from doi: Arctic region, Nature, 425, 947–950, doi: Schweiger, A., Zhang, J.,Davidson, Haas, M.: C., CryoSat-2 Hendricks, estimates S.,Lett., of Krishfield, 40, Arctic 732–737, R., sea doi: Kurtz, ice N., thickness Farrell, and S., volume, and Geophys. Res. cate subsetdoi: used], National Snow and Ice Data Center, Boulder, Colorado, USA, Solomon, S., Free, M.,chka, Gleckler, D., P. J., Schmidt, Jones, G. P. D., A., Karl, Sherwood, T. S. R., C., Klein, and S. Wentz, A., F. Mears, J.: C., Consistency Ny- of modeled and Snow Depth, and Thickness.Center, Boulder, Version Colorado, 1, USA, 2012. NASA DAAC at the National Snow and Ice Data 2013 National Snow and Ice Data Center, Boulder, Colorado, USA, doi: submarines, J. Atmos. Ocean. Tech., 24, 1936–1949, doi: Zhang, J. L. and Rothrock, D. A.: Modeling global sea ice with a thickness and enthalpy dis- Warren, S., Rigor, I. G., and Untersteiner, N.: Snow depth on Arctic sea ice, J. Climate, 12, Serreze, M. C., Barrett, A. P., Stroeve, J. C., Kindig, D. N., and Holland, M.Stroeve, M.: J., The Holland, emergence M. M., Meier, W., Scambos, T., and Serreze, M.: Arctic sea ice decline: Schweiger, A., Lindsay, R., Zhang, J., Steele, M., Stern, H., and Kwok, R.: Uncer- Wang, M. and Overland, J. E.: A seaWang, ice M. free and summer Overland, Arctic J. E.,: within A 30 years?, sea Geophys. ice Res. free summer Arctic within 30 years – an update from Thompson, A. M., Calvin, K. V., Smith, S. J., Kyle, G. P., Volke, A., Patel, P., Delgado-Arias, S., Wadhams, P.:Arctic sea ice morphology and its measurement, in: Arctic Technology and Policy, Stroeve, J. C., Kattsov, V., Barrett, A., Serreze, M., Pavlova, T., Holland, M., andStroeve, Meier, J. W. C., N.: Serreze, M. C., Kay, J. E., Holland, M. M., Meier, W. N., and Barrett, A. P.: The Massonnet, F., Fichefet, T., Goosse, H., Bitz, C. M., Philippon-Berthier, G., Holland, M. M., and Notz, D. and Marotzke, J., J.: Observations reveal external driver for Arctic sea-ice retreat, Melling, H. and Riedel, D. A.: Ice Draft and Ice Velocity Data in the Beaufort Sea, 1990–2003, Rampal, P., Weiss, J., Dubois, C., and Campin, J.-M.: IPCC climate models do not capture Laxon, S. W., Giles, K. A., Ridout, A. L., Wingham, D. J., Willatt, R., Cullen, R., Kwok, R., NSIDC: Submarine Upward Looking Sonar Ice Draft Profile Data and Statistics [indi- Santer, B. D., Thorne, P. W., Haimberger, L., Taylor, K. E., Wigley, T. M. L., Lanzante, J. R., Kurtz, N. T., Farrell, S. L., Studinger, M., Galin, N., Harbeck, J. P., Lindsay, R., Onana, V. D., Kurtz, N. T., Studinger, M., Harbeck, J., Onana, V., and Farrell, S.: IceBridge Sea Ice Freeboard, Laxon, S., Peacock, N., and Smith, D.: High interannual variability of sea ice thickness in the Lindsay, R.: Unified Sea Ice Thickness Climate Data Record Collection Spanning 1947–2012, Rothrock, D. A. and Wensnahan, M.: The accuracy of sea-ice drafts measured from US Navy 5 5 30 20 25 15 10 20 10 25 15 30 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | 1 1 1 1 1 1 1 1 1 1 1 27 1 Ensembles per year. 3 ) 0.85) 3 2.98) 2 1.27) 3 2.31) 4 1.01) 10 1.77) 3 0.67) 10 3.08) 2 1.49) 6 0.74) 5 0.68) 9 r − − − − − − − − − − − 1.66 to 4.28 to 2.34 to 3.85 to 2.34 to 3.20 to 3.19 to 3.18 to 2.79 to 1.15 to 1.23 to − − − − − − − − − − − ( ( ( ( ( ( ( ( ( ( ( Range of Trends Number of ) 1 − , respectively. dec 3 ++ + ++ ++ ++ ++ ++ ++ ++ ++ ++ ++ ++ ++ ++ ++ ++ ++ ++ ++ ++ ++ ++ ++ ++ ++ km 3 2.41 1.15 1 1.37 3.63 1.95 1.76 0.96 2.48 0.492.90 1 2.26 2.92 2.32 1.72 2.54 0.75 1 1.63 1.68 3.39 2.21 1 2.09 2.34 3.13 2.37 1.03 0.94 2.05 2.79 1.83 − − − − − − − − − − − − − − − − − − − − − − − − − − − − (10 − and + erent remotely-sensed data sets. ff MIROC5 MIROC4h MIROC-ESM-CHEM MIROC-ESM FGOALS-g2 CSIRO-Mk3-6-0 CNRM-CM5 Multi-model Mean PIOMAS 0.150.360.47 0.780.37 0.55 0.56 0.81 0.70 0.70 0.47 0.38 − − − − 2202 2201 ObservationsIn Situ and Submarine ERS-1 ICESat MeanIceBridge Error (m)CryoSat-2 RMSE (m) Correlation ( 0.20 0.50 0.68 Mean ice thickness bias, root-mean-square error estimate and correlation between Linear trends in Arctic sea ice volume for Mar based on the period 1979 to 2013 from ce Hadley Centre HadGEM2-ES ce Hadley Centre HadGEM2-CC ce Hadley Centre HadGEM2-AO ce Hadley Centre HadCM3 ffi ffi ffi ffi Norwegian Climate Centre NorEMS1-M Meteorological Research Institute MRI-CGCM3 Max-Planck-Institut fur Meteorologie MPI-ESM-LR Japan Agency for Marine-EarthResearch Institute Science (The and University of Technology,Studies Tokyo) Atmosphere and and National Institute Ocean for Environmental Japan Agency for Marine-EarthResearch Institute Science (The and University of Technology,Studies Tokyo) Atmosphere and and National Institute Ocean for Environmental Japan Agency for Marine-EarthResearch Institute Science (The and University of Technology,Studies Tokyo) Atmosphere and and National Institute Ocean for Environmental Japan Agency for Marine-EarthResearch Institute Science (The and University of Technology,Studies Tokyo) Atmosphere and and National Institute Ocean for Environmental Institut Pierre-Simon Laplace IPSL-CM5A-MR Institut Pierre-Simon Laplace IPSL-CM5A-LR Institute for Numerical Mathematics INMCM4 Met O Met O Met O Met O NASA Goddard Institute for Space Studies GISS-E2-R NOAA Geophysical Fluid Dynamics Laboratory GFDL-ESM2M NOAA Geophysical Fluid Dynamics Laboratory GFDL-ESM2G NOAA Geophysical Fluid Dynamics Laboratory GFDL-CM3 LASG, Institute of AtmosphericTsinghua Physics, University Chinese Academy of Sciences and CESS, EC-EARTH consortium EC-EARTH Commonwealth Scientific and Industrial ResearchQueensland Organization Climate in Change collaboration Centre with of Excellence Centre National de Recherches Meteorologiques/CentreFormation Europeen Avancee de en Recherche Calcul et Scientifique National Center for Atmospheric Research CESM1-CAM5 National Center for Atmospheric Research CCSM4 Canadian Centre for Climate Modelling and Analysis CanESM2 Canadian Centre for Climate Modelling and Analysis CanCM4 Modeling Center (or Group)Beijing Climate Center, China Meterological Administration BCC-CSM1-1 Model Name Trend Trends statistically significant at 95 and 99 % are denoted by 27 CMIP5 models and PIOMAS.trend For is models given with along more with than the one range ensemble in member, trend the (in mean parenthesis). Trends are listed as km Table 2. Table 1. PIOMAS modeled ice thickness and thicknesses from di Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | cient of vari- ffi

culties in reducing surface pressures

ffi 17

2204 2203 stddev/average). This is a normalized measure of variability so that variability can becan variability that so variability measure of normalized a is This stddev/average). Variability of thicknesses in six models is attached. The values are coefficient of Figure 1. Figure ( variability betweenandmodels. comparedspatially

581 582 583 584 585 586 587 580 Model1. bcc-csm1-12. CanCM43. CanESM24. CCSM45. CESM1-CAM5 0.766. (10) CNRM-CM57. ERA-Interim CSIRO-Mk3-6-0 0.698. MERRA (4) EC-EARTH 0.74 0.72 (8) (5) 0.939. (25) FGOALS-g2 CFSR10. 0.73 GFDL-CM3 0.58 (11) 0.62 (1) (2) 0.73 0.74 (6)11. (9) GFDL-ESM2G 0.89 0.71 0.77 (22) (12) NCEP (10)12. GFDL-ESM2M 0.93 N 0.65 0.67 (26) (3) (6)13. 0.92 0.67 GISS-E2-R (24) 0.51 (2) (0) Closed 0.79 0.43 0.91 BSH? (12) (0) (27)14. HadCM3 0.61 0.63 0.75 (3) (5) 0.87 (8) (21) 0.67 0.52 Y 0.6615. (5) (1) (4) HadGEM2-AO 0.76 0.94 (11) (25) Y 16. Y HadGEM2-CC 0.63 0.52 0.47 0.70 0.89 (4) (1) (1) (10) (24) 0.8817. (20) HadGEM2-ES 0.82 (14) 0.81 0.82 (15) N 0.8618. (15) (22) 0.85 inmcm4 Y 0.36 Y 0.94 (18) (0) 0.70 (26) (8)19. 0.71 Y IPSL-CM5A-LR (9) 0.89 0.82 0.63 (22) (17) (3) 0.3120. 0.84 (0) 0.65 IPSL-CM5A-MR (17) (6) 0.90 Y 0.66 (23) 0.97 (8)21. (29) MIROC4h 0.78 (14) N 0.94 Y 22. 0.83 0.92 (24) MIROC5 (17) (25) Y 0.72 0.74 (5) (14) 0.81 0.9523. (14) 0.86 (27) MIROC-ESM 0.88 (19) (23) Y 0.8624. (18) 0.58 0.87 MIROC-ESM-CHEM 0.81 (2) (22) 0.78 Y (16) (11)25. 0.75 MPI-ESM-LR 0.73 0.83 (9) (6) Y (19) 0.84 0.53 (17)26. (2) 0.78 MRI-CGCM3 (12) 0.84 (18) Y 0.73 0.83 0.8327. (7) (18) (15) NorESM1-M Y 0.80 0.86 (13) (20) 0.71 Y 0.84 (3) (21)ERA-Interim 0.83 (16) 0.83 0.86 (20) (20) N MERRA 0.74 0.73 0.73 0.86 0.86 (12) Y (10) (7) (19) (19)CFSR 0.70 0.71 0.82 0.76 (9) (11) 0.89 (16)NCEP (13) 0.69 (21) (7) N 0.79 0.71 0.83 (13) Y (13) (16) 0.66 (7) 0.71 0.87 Y 0.81 1.00 (4) (23) (15) (30) 0.89 Y Y (24) 0.86 (21) Y 0.96 0.96 (28) 0.89 (27) (25) 0.99 (29) N 0.99 (29) 0.97 1.00 (28) (30) 0.97 (28) 0.94 (27) 0.94 (26) 0.91 0.91 (26) (23) 1.00 (30) 0.99 (28) 0.99 (29) 1.00 (30) Spatial correlations between observed mean annual sea level pressure from four dif- Variability of thicknesses in six models is attached. The values are coe Fig. 1. ability (stddev/average). This iscompared a spatially normalized and between measure models. of variability so that variability can be Table 3. ferent reanalysis data setsparentheses, running and lowest from to the highest. CMIP5 Because models. of Ranks di of correlations are given in to sea level, pressures overferent Greenland reanalysis have been are screened alsoBeaufort out. included Sea Correlations High as between (BSH). the well dif- as whether or not the models represent a closed Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | 1, 1, - 2 sea ice ice sea 2 -

period. -

18 1, ICESat, IceBridge and CryoSat and IceBridge ICESat, 1, - erent colors, except for ERS-1, which was ff campaigns period of record, with ice thickness fields fields thickness ice with record, of campaignsperiod 2206 2205 plots, individual years are shown in different colors, except for ERS . Comparison of submarine, ERS

Figure 2 Figure thickness fields (left column), for each PIOMAS column). (right plots scatter PIOMAS bycorresponding and column) simulated (middle fields are the average March thicknesses for the same periods as corresponding observed records.In the scatter time entire the for field mean a as provided was which

588 589 590 591 592 593 594 Comparison of submarine, ERS-1, ICESat, IceBridge and CryoSat-2 sea ice thickness provided as a mean field for the entire time-period. Fig. 2. fields (left column), for eachPIOMAS campaigns (middle period column) of record, andthe with corresponding average ice scatter March thickness plots fields thicknessesthe simulated (right for by scatter column). the plots, PIOMAS same individual fields periods years are as are corresponding shown observed in records. di In Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | e Figure Figure

19 2208 2207 models. Model results are presented as box and whisker plots from 1981 to

Comparison of thickness distributions between five observational data sets, PIOMAS Spatial patterns of sea ice thickness from 1981 to 2010 from 27 CMIP5 models. . Comparison of thickness distributions between five observational data sets, PIOMAS and 27

3 individual CMIP5 the and percentiles) 75th to (25th range interquartile the represent boxes the where 2010, Th respectively. mean, and median the define box each within asterisks and bars horizontal solid a as shown are PIOMAS and set data observational each from thicknesses spring median redline, together with the 10th and 90th percentiles (green lines) and the interquartile range (greyshading).

Fig. 4. Fig. 3. and 27 individual CMIP51981 models. to Model 2010, results where the arethe boxes presented horizontal represent as the bars box interquartile andThe and range asterisks median (25th whisker within spring to plots 75th each thicknesses from a percentiles) from box and solid each define red observational the line, data medianrange together set (grey and with shading). and mean, the PIOMAS 10th respectively. are and shown 90th as percentiles (green lines) and the interquartile 601 602 603 604 597 598 599 600 595 596 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |

square error (RMSE) (bottom) of ice ice of (RMSE) (bottom) error square - mean -

23

Spatial pattern correlations (top) and root

Figure 5. Figure thickness in 27 CMIP5 models and ICESat.

2210 2209 610 611 612 613 Mean annual sea level pressure and geostrophic wind from 27 CMIP5 models and from Spatial pattern correlations (left) and root-mean-square error (RMSE) (right) of ice thick- Fig. 5. ness in 27 CMIP5 models and ICESat. Fig. 6. ERA-Interim spanning 1981–2010. Contour intervalused is as 1 a hPa. proxy Near-surface for geostrophicwind sea wind speed. ice is Vectors motion are and curved is tangent shown to by red the vectors. instantaneous Vector flow. length is proportional to Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |

2010 PIOMAS trendsPIOMAS are

). 1 standard deviation of CMIP5 σ ± modelensemble mean ice - 2000

26 27 March confidence intervals of 2212 2211 σ

) and) light gray shading (2 modelensemble mean (shown in black) with confidence and 2 - σ 1990 σ hemodel ensemble. Multi ). σ PIOMAS CMIP5 confidence intervals of PIOMAS trends are shown in dark gray shading σ Changeinseaice Arctic volume asshown from theCMIP5 ensemble and 1980

March ice volume trends from 1979 2013to allfor individual CMIP5 model and 2

0

σ

20 10 50 40 30

Ice Volume [10 Volume Ice ] km

3 3 March ice volume trends from 1979 to 2013 for all individual CMIP5 model ensembles Change in Arctic sea ice volume as shown from the CMIP5 ensemble and from PIOMAS

fromtheperiodfor PIOMAS 1979 2012,to March.for Grey shading shows the±1 standarddeviation CMIP5of ensemble. Upper and lower pecked lines show maximum andminimum ice volume t of volumeisshown astheblack line. Figure 7.

ntervals(vertical lines). The1

Figure 8. ensemblesaswell asthemulti i shownin dark gray shading (1 ) and light gray shading (2

σ (1 Fig. 8. as well as the multi-modellines). ensemble The mean 1 (shown in black) with confidence intervals (vertical for the period 1979 to 2012, for March. Grey shading shows the Fig. 7. ensemble. Upper and lower pecked linesensemble. show Multi-model maximum and ensemble minimum mean ice ice volume volume of is the model shown as the black line. 627 621 622 623 624 625 626 620 627 628 629 630 631 632 633