Simulation and Projection of Arctic Freshwater Budget Components by the IPCC AR4 Global Climate Models

Transcription

1 JUNE 2007 K A T T S O V E T A L. 571 Simulation and Projection of Arctic Freshwater Budget Components by the IPCC AR4 Global Climate Models VLADIMIR M. KATTSOV Voeikov Main Geophysical Observatory, St. Petersburg, Russia JOHN E. WALSH International Arctic Research Center, University of Alaska Fairbanks, Fairbanks, Alaska WILLIAM L. CHAPMAN Department of Atmospheric Sciences, University of Illinois at Urbana Champaign, Urbana, Illinois VERONIKA A. GOVORKOVA AND TATYANA V. PAVLOVA Voeikov Main Geophysical Observatory, St. Petersburg, Russia XIANGDONG ZHANG International Arctic Research Center, University of Alaska Fairbanks, Fairbanks, Alaska (Manuscript received 10 June 2005, in final form 29 August 2006) ABSTRACT The state-of-the-art AOGCM simulations have recently (late 2004 early 2005) been completed for the Intergovernmental Panel on Climate Change (IPCC) in order to provide input to the IPCC s Fourth Assessment Report (AR4). The present paper synthesizes the new simulations of both the twentieth- and twenty-first-century arctic freshwater budget components for use in the IPCC AR4, and attempts to determine whether demonstrable progress has been achieved since the late 1990s. Precipitation and its difference with evapotranspiration are addressed over the Arctic Ocean and its terrestrial watersheds, including the basins of the four major rivers draining into the Arctic Ocean: the Ob, the Yenisey, the Lena, and the Mackenzie. Compared to the previous [IPCC Third Assessment Report (TAR)] generation of AOGCMs, there are some indications that the models as a class have improved in simulations of the Arctic precipitation. In spite of observational uncertainties, the models still appear to oversimulate area-averaged precipitation over the major river basins. The model-mean precipitation biases in the Arctic and sub-arctic have retained their major geographical patterns, which are at least partly attributable to the insufficiently resolved local orography, as well as to biases in large-scale atmospheric circulation and sea ice distribution. The river discharge into the Arctic Ocean is also slightly oversimulated. The simulated annual cycle of precipitation over the Arctic Ocean is in qualitative agreement between the models as well as with observational and reanalysis data. This is also generally the case for the seasonality of precipitation over the Arctic Ocean s terrestrial watersheds, with a few exceptions. Some agreement is demonstrated by the models in reproducing positive twentieth-century trends of precipitation in the Arctic, as well as positive area-averaged P E late-twentieth-century trends over the entire terrestrial watershed of the Arctic Ocean. For the twenty-first century, three scenarios are considered: A2, A1B, and B1. Precipitation over the Arctic Ocean and its watersheds increases through the twenty-first century, showing much faster percentage increases than the global mean precipitation. The arctic precipitation changes have a pronounced seasonality, with the strongest relative increase in winter and fall, and the weakest in summer. The river discharge into the Arctic Ocean increases for all scenarios from all major river basins considered, and is generally about twice as large as the increase of freshwater from precipitation over the Arctic Ocean (70 90 N) itself. The across-model scatter of the precipitation increase for each scenario is significant, but smaller than the scatter between the climates of the different models in the baseline period. Corresponding author address: Dr. John Walsh, International Arctic Research Center, University of Alaska Fairbanks, Fairbanks, AK jwalsh@iarc.uaf.edu DOI: /JHM American Meteorological Society

2 572 J O U R N A L O F H Y D R O M E T E O R O L O G Y VOLUME 8 1. Introduction The distribution of precipitation and evapotranspiration in the Arctic has been a subject of accelerating interest in recent years. This interest is driven largely by the realization that variations in hydrologic processes in the Arctic have major implications not only for arctic terrestrial and marine ecosystems, but also for the cryosphere and the global ocean. For example, the supply of freshwater to the Arctic Ocean is (over a period of multiple years) essentially equal to the difference between precipitation (P) and evaporation (evapotranspiration) (E) over the Arctic Ocean and the terrestrial watersheds draining into the Arctic Ocean. A change of this freshwater supply has potentially important implications for the Arctic Ocean s stratification, for its sea ice regime, and for its freshwater export to the North Atlantic. To the extent that deep oceanic convection in the North Atlantic is affected by the freshwater capping provided by Arctic Ocean export, the arctic freshwater budget has the potential to affect the global thermohaline circulation. An increase of P E implies generally wetter soils when soils are not frozen, increased surface flows above frozen soils, wetter active layers in the summer, and greater ice content of the upper soil layer during winter. If an increase of P is manifest as an increase of snowfall during the cold season, the Arctic Ocean and its terrestrial watersheds will experience increases of snow depth and snow water equivalent, although the seasonal duration may well be shorter if warming accompanies the increase of P. An increase of P E over the terrestrial watersheds will increase the moisture availability in the upper soil layers, favoring plant growth in regions that are otherwise moisture limited. An increase of P E and river discharge will likely result in enhanced fluxes of nutrients and sediments to the Arctic Ocean, with corresponding impacts on coastal marine ecosystems. In addition, a change of aquatic transport and associated heat fluxes across the coastal zone may accelerate the degradation of coastal permafrost in some areas. Variations of the arctic freshwater budget are not well documented and are poorly understood (Lewis et al. 2000). It should be noted that terrestrial components are better known than those of the ocean flux and storage. The network of river gauges, despite the absence of measurements of smaller streams and subsurface flows, is far denser than the network of moorings, buoys and ocean stations for ocean freshwater fluxes in programs such as Variability and Exchanges in the Northern Seas (VEINS; veins/) and Arctic Subarctic Ocean Fluxes (ASOF; For this reason, global climate models are important for diagnosing variations of the arctic freshwater budget, and they are essential for projections of changes over the coming decades to a century. Assessments of models capabilities in arctic freshwater simulations are hindered by the limitations of the observational data for validation and by the large scatter among models in their arctic simulations. This large scatter has been shown by Walsh et al. (1998) in a suite of atmospheric general circulation models (AGCMs) used in the Atmospheric Model Intercomparison Project (Gates 1992) and in a comparison of such simulations with coupled atmosphere ocean general circulation models (AOGCMs) by Walsh et al. (2002). The model simulations examined in these earlier studies were generally of late-1990s vintage used in the Intergovernmental Panel on Climate Change (IPCC) Third Assessment Report (TAR) of Houghton et al. (2001). A recent evaluation of AOGCM performance in the Arctic (Kattsov and Källén 2005; Walsh 2005) conducted as a part of the Arctic Climate Impact Assessment (ACIA), considered the same (TAR) generation of the AOGCMs. That evaluation included in particular an analysis of AOGCM systematic errors in simulation of the current arctic climate and uncertainties of arctic climate projections for the twenty-first century. Compared to the individual simulations, the multimodel ensemble means of precipitation showed reasonable agreement with available observations, at least in the area averages. The assessment of the ability of the AOGCM ensemble to simulate the observed current climate supported the suitability of the ensemble for use in constructing the twenty-first-century climate change scenarios for the Arctic region. Since the TAR, significant advances have been introduced into AOGCMs. The dynamical cores including both numerics and spatial resolution have improved. More processes have been incorporated, and parameterizations of physical processes have become more comprehensive. These improvements, while not always straightforwardly identifiable in the output, have resulted in improvements of AOGCM simulation of many aspects of the earth climate system. Among evidences of these improvements is the fact that most of the current AOGCMs no longer use flux adjustments to reduce the climate drift. Nevertheless, AOGCMs still show significant errors, as well as intermodel scatters both in simulating observed and projecting future climates, particularly at the regional scale. This is only partly a result of limitations of computing power (the highest-resolution models do not obviously outperform others), but also a result of insufficient scientific understanding.

3 JUNE 2007 K A T T S O V E T A L. 573 IPCC model name Country TABLE 1. IPCC AR4 model runs analyzed in this study. Atmosphere resolution Ocean resolution No. of 20C3M ensemble members SRES scenario missing Natural forcing in 20C3M 1 BCCR-BCM2.0 Norway T63 L L35 1 A1B, B1 Yes 2 CCSM3 United States T85 L L40 6 Yes 3 CGCM3.1(T47) Canada T47 L L29 1 No 4 CGCM3.1(T63) Canada T63 L L29 1 A2, A1B, B1 No 5 CNRM-CM3 a France T63 L L31 1 No 6 CSIRO-Mk3.0 b Australia T63 L L31 2 No 7 ECHAM5/MPI-OM Germany T63 L L40 3 No 8 ECHO-G c Germany/Korea T30 L L20 3 A1B, B1 Yes 9 GFDL-CM2.0 d United States L L50 3 Yes 10 GFDL-CM2.1 United States L L50 3 Yes 11 GISS-AOM e United States 3 4 L L16 2 A2 No 12 GISS-EH United States 4 5 L L16 5 A2, B1 Yes 13 GISS-ER United States 4 5 L L13 9 A1B Yes 14 INM-CM3.0 f Russia 4 5 L L33 1 Yes 15 IPSL-CM4 g France L L31 1 No 16 MIROC3.2(hires) Japan T106 L L47 1 A2 Yes 17 MIROC3.2(medres) Japan T42 L L43 3 Yes 18 MRI-CGCM2.3.2 Japan T42 L L23 5 Yes 19 PCM United States T42 L L40 2 Yes 20 UKMO-HadCM3 i United Kingdom L L20 2 No 21 UKMO-HadGEM1 United Kingdom L L40 2 B1 Yes a CNRM-CM3 Centre National de Recherches Météorologiques Coupled Global Climate Model version 3. b CSIRO-Mk3.0 Commonwealth Scientific and Industrial Research Organisation Mark version 3.0. c ECHO-G ECHAM and the global Hamburg Ocean Primitive Equation. d GFDL-CM2.0 Geophysical Fluid Dynamics Laboratory Climate Model version 2.0. e GISS-AOM Goddard Institute for Space Studies Atmosphere Ocean Model. f INM-CM3.0 Institute of Numerical Mathematics Coupled Model version 3.0. g IPSL CM4 L Institut Pierre-Simon Laplace Coupled Model version 4. i HadCM3 Third Hadley Centre Coupled Ocean Atmosphere General Circulation Model. The most current generation of AOGCM simulations has recently (late 2004 early 2005) been completed for the IPCC in order to provide input to the IPCC s Fourth Assessment Report (AR4). The AR4 is scheduled for publication in The present paper has as its objectives 1) a synthesis of the most recent simulations of the arctic freshwater budget components both for the twentieth and twenty-first centuries for use in the IPCC AR4, and 2) a determination of whether demonstrable progress has been achieved since the late 1990s in the simulation of the arctic freshwater budget. In section 2, we describe the models, the suite of simulations performed for the IPCC by the participating modeling centers, and the observational data used for the evaluation of the model simulations. Section 3 is an evaluation of the mean current climate simulations by these models, while section 4 addresses their ability to reproduce trends observed in the twentieth century. Section 5 is a synthesis of the twenty-first-century projections driven by different scenarios of greenhouse gases and aerosol forcing. Section 6 is a summary of the findings in the context of the two objectives stated above. 2. Model and observational data The AOGCMs whose outputs were analyzed in this study are listed in Table 1. This is a subset of the IPCC AR4 models for which precipitation and evapotranspiration fields were available by mid-may 2005 for analysis of the simulation of the twentieth-century climate (20C3M). Three IPCC Special Report on Emission Scenarios (SRES) scenarios for the twenty-first century were also considered in this study: A2, A1B, and B1. These are three of the six so-called marker SRES emission scenarios based on different assumptions on changes in population, economic growth, etc. through the twenty-first century. By the end of the twenty-first century, the A2 scenario has the greatest increase of greenhouse gas concentrations (to about 825 ppm CO 2 ) and suggests the strongest global warming, while the B1 is the least extreme of the three (about 550 ppm CO 2 by 2100; for details see Nakićenović and Swart 2000). The 20C3M simulations were forced with evolving concentrations of greenhouse gases and sulfate aerosol prescribed in accordance with observations. Some of the

4 574 J O U R N A L O F H Y D R O M E T E O R O L O G Y VOLUME 8 models additionally used time-varying natural (solar and volcanic) forcings. Altogether, the number of models included in this analysis is 21. Simulations with some of the models include several ensemble members started from different initial conditions. In this study, the entire ensembles were used in the analysis of the twentieth-century trends and variability in the 20C3M simulations, as well as in evaluation of signal-to-noise ratios for different time slices of the twenty-first century. Otherwise, whenever more than one simulation with an individual model was available, only single (first) members of the ensembles were included in the analysis. Some scenario simulations for the twenty-first century were not available for some IPCC AR4 models; thus different subsets of the models are used in obtaining the twenty-first-century estimates discussed in this study. Only monthly fields of precipitation and evaporation were used in this study. Original global fields were bilinearly interpolated into the latitude longitude grid in order to enable direct point-to-point intercomparison of the geographical distributions of the simulated variables, and also to permit comparisons of them against observational or reanalysis data. Observational estimates of arctic precipitation suffer from errors in gauge measurements of solid precipitation, especially when the solid precipitation occurs under windy conditions. Depending on the partitioning between falling snow (precipitation) and wind-blown snow from the surface, the error can range from a significant undercatch to a significant overcatch. Because different observational climatologies incorporate varying types and degrees of adjustments, there is some variance among the observational estimates (Walsh et al. 1998). The primary observational dataset used in this study is an outgrowth of an Arctic precipitation climatology originally compiled by Bryazgin (1976), whose monthly mean fields based on gauge-corrected in situ data were extended in Khrol (1996) and subsequently updated and enhanced by additional corrections. This compilation includes data from the Russian drifting ice stations and high-latitude land surface stations, and it is gridded over the N domain. Earlier analyses (Walsh et al. 2002) showed a close consistency (at least in area averages over the Arctic Ocean) between the Bryazgin data and another gauge-corrected precipitation climatology compiled by Legates and Willmott (1990, with subsequent updates). Progress in mapping the spatial and seasonal distributions of actual Arctic precipitation has resulted from the use of information on gauge bias adjustment procedures, for example, from the World Meteorological Organization s (WMO) Solid Precipitation Measurement Intercomparison (Goodison et al. 1998). Summaries of precipitation over the Arctic Ocean, where only coastal stations and drifting ice station measurements are available, have recently been completed by Colony et al. (1998), Yang (1999), and Bogdanova et al. (2002). In the latter study, which accounts for all the major systematic errors of precipitation, the annual mean bias-corrected precipitation for the central Arctic Ocean was found to be 16.9 cm, which is 32% higher than the value resulting from uncorrected measurements. The spatial pattern shows an increase from minimum values of less than 10 cm yr 1 over Greenland and cm yr 1 over much of the Arctic Ocean, to more than 50 cm yr 1 over parts of the North Atlantic subpolar seas north of 70 N. Estimates of evaporation over the Arctic Ocean are scarce. Probably the best estimates are from the 1-yr SHEBA project in the Beaufort Sea collected during 1997 and These observations show evaporation as nearly zero from October through April, then peaking in July at about 7 mm (Persson et al. 2002). Estimates of precipitation and evapotranspiration for the major terrestrial watersheds of the Arctic have recently been compiled by Serreze et al. (2003). In that study, basin-averaged values of the annual mean P were derived from objectively analyzed fields of gaugeadjusted station measurements; P E from the atmospheric moisture flux convergences in the National Centers for Environmental Prediction National Center for Atmospheric Research (NCEP NCAR) reanalysis; runoff from river gauges near the mouths of the major rivers. Evapotranspiration estimates were obtained from two types of calculations: the difference between the independently derived precipitation and P E; and the difference between basin-averaged precipitation and runoff. The two estimates of evapotranspiration differ by as much as 20%, providing a measure of the uncertainty in the basin-scale means of the hydrologic quantities. At least some, and probably most, of the uncertainty arises from biases in precipitation. Additional estimates, using data for earlier years, of the freshwater budget components of Arctic and worldwide rivers are provided by Oki et al. (1995). A more detailed analysis of the Mackenzie Basin water cycle has recently been provided by Rouse et al. (2003). An opportunity for model evaluation is provided by reanalyses employing numerical weather prediction models and variational assimilation techniques to convert irregularly spaced observational data into complete temporally regular global grids of precipitation (currently for periods of several decades). The quality of reanalyses-derived precipitation and evaporation cli-

5 JUNE 2007 K A T T S O V E T A L. 575 matologies in high latitudes is lower than that of assimilated variables, such as temperature or atmospheric pressure. However, in spite of this and some other limitations (see, e.g., Cullather et al. 2000), reanalyses are self-consistent and useful resources for model evaluation. In this study we use the 40-yr European Centre for Medium-Range Weather Forecasts (ECMWF) Re- Analysis (ERA-40), which begins in the International Geophysical Year of The documentation on ERA-40 data can be found online at research/era/products. Serreze et al. (2005) have evaluated the quality of ERA-40 estimates over the Arctic and have shown that in general ERA-40 performs well in comparison to bias-adjusted observations, at least over land, and is superior to the NCEP NCAR reanalysis. 3. Simulation of the recent climatology In this study, two time periods were selected for the analysis of the model performance in simulation of the present climatologies of precipitation and evaporation (evapotranspiration) over the Artic Ocean and its watersheds: the 20-yr period and the 30-yr period The former represents the most recent climate and coincides well with the climatological baseline period recommended by the IPCC AR4 in estimates of climate changes projected for the twentyfirst century, and was also used in earlier evaluations for the same purpose by ACIA (Kattsov and Källén 2005). The latter period ( ) was chosen because it enables straightforward comparison with the climatological component estimates of the Arctic Ocean freshwater budget by Serreze et al. (2003). Additionally, the 30 yr ( ) are the greater part of the period of operation of the Russian ice drifting stations under the North Pole program, which formed the basis of Bryazgin s (1976) arctic precipitation climatology. The ERA-40 data cover both time periods and can further provide a measure of differences between them in terms of multiyear averages. The 21-model multiyear ( ) annual, winter, and summer mean precipitation distributions over the Arctic Ocean and its terrestrial watersheds are presented in Fig. 1. The annual mean field shows an impressive similarity to that simulated with the eight en- FIG. 1. Twenty-one-model mean (first ensemble members) precipitation (mm day 1 ) geographical distribution for the current climate ( ): (a) annual mean; (b) winter [December February (DJF)]; and (c) summer [June August (JJA)].

6 576 J O U R N A L O F H Y D R O M E T E O R O L O G Y VOLUME 8 semble members of the IPCC TAR AOGCMs (Walsh et al. 2002, their Fig. 11b). Both ensembles depict the largest amounts in the North Atlantic, including southern Greenland and western Scandinavia, and in the northeastern Pacific. The smallest amounts are seen over the central Arctic Ocean and northern Greenland. In winter, the precipitation over the major part of the Arctic Ocean, extending from the Canadian Archipelago and the north of Greenland to the Laptev and East Siberian Seas, and further to Siberia, is below 0.5 mm day 1, while the precipitation is comparatively larger locally up to 5 mm day 1 over the subarctic oceans, including the Greenland Iceland Norwegian Seas (GIN), the Bering Sea, and the North Atlantic and North Pacific. By contrast, in summer, precipitation increases over the Arctic Ocean (up to 1 mm day 1 ) and the surrounding land areas, and decreases over the North Atlantic and North Pacific. This precipitation seasonality is consistent with, and can be accounted for by, the climatological annual cycle of Arctic storm activity, which shows more storms entering into the interior Arctic Ocean in summer than winter (Zhang et al. 2004). Figure 2 compares the simulated annual mean precipitation for the other period ( ) against the corresponding climatologies of (a) ERA-40 and (b) Bryazgin. The two difference fields show qualitative similarity indicating a general overestimate of precipitation over the western Arctic, particularly over Alaska and southeastern Greenland, and an underestimate over a broad area of the eastern Arctic, especially in the Norwegian Barents Sea region. Among likely causes of the precipitation biases (and possibly of some differences between ERA-40 and Bryazgin shown in Fig. 3c) are unresolved orographic effects due to insufficient spatial resolution of most models (too small an orographic enhancement of precipitation by the models smoothed topography, e.g., on the western Scandinavia slope). Other contributing factors include discrepancies in atmospheric circulations, for example, unrealistic positioning of the major storm track in the northeastern North Atlantic, and distorted thermodynamic forcing due to the poor representation of sea ice cover, for example, an overestimation of sea ice cover in the Barents Sea by most models (Arzel et al. 2006). The across-model scatter is illustrated by Fig. 3a, which shows annual mean precipitation averaged over FIG. 2. Geographical distribution of the annual mean precipitation biases (mm day 1 ) for the current climate: (a) 21-model mean (first ensemble member from each model) minus ERA-40; (b) 21-model mean minus Bryazgin; (c) ERA-40 minus Bryazgin.

7 JUNE 2007 K A T T S O V E T A L. 577 FIG. 3. Annual mean P and P E over the Arctic Ocean (70 90 N) ( ): (a) P in IPCC AR4 model simulations vs an estimate based on observations by Bryazgin (Khrol 1996) and ERA-40; (b) P in the subset of models allowing for a comparison between their IPCC AR4 and earlier TAR versions (the model means were calculated for the 8 TAR models and the 11 AR4 models); and (c) P E in IPCC AR4 model simulations vs an observationally based estimate (Korzun 1978). the Arctic Ocean (70 90 N) from each IPCC AR4 model simulation and from the ERA-40 for the period of The Bryazgin climatology is also shown in the figure. In addition, we show the annual mean precipitation from the ERA-40 for the period for comparison, since the Bryazgin dataset is based on the earlier years of observations. The largest annual mean precipitation (1.00 mm day 1 ) is simulated by the Model for Interdisciplinary Research on Climate 3.2, high-resolution version MIROC3.2(hires), which has the highest spatial resolution. The next two wettest (and next two highest resolution) models are the Community Climate System Model version 3 (CCSM3) and ECHAM5/Max Planck Institute Ocean Model (MPI- OM), showing a precipitation of 0.98 mm day 1. The atmospheric components of the driest model, the Canadian Centre for Climate Modelling and Analysis (CCCma) Coupled General Circulation Model version 3.1 [CGCM3.1(T63); 0.57 mm day 1 ], and the third driest Bjerknes Centre for Climate Research Bergen Climate Model version 2.0 (BCCR-BCM2.0) (0.63 mm day 1 ) have the same resolution as that of the ECHAM5/MPI-OM, and thus break the apparent relationship between the arctic precipitation and model resolution. The 21-model mean of precipitation of 0.79 mm day 1 is lower than that in the ERA-40 (0.90 mm day 1 ) for the same period. However, this mean value is apparently comparable to the averaged precipitation from the ERA-40 during a longer previous period from 1960 to 1989, which is 0.85 mm day 1, and is slightly higher than Bryazgin s estimate of 0.74 mm day 1. The model-derived estimates are generally within about 10% of the observational estimates. Given indications that arctic precipitation has generally increased in the later portion of the twentieth century (McBean 2005, his Table 2.2), and the fact that for the baseline period of the ERA-40 estimate is higher than Bryazgin s estimate, the multimodel estimate (lower than ERA-40) appears to realistically capture the actual precipitation. Figure 3b compares a subset of IPCC AR4 models with their earlier (IPCC TAR) versions considered by Walsh et al. (2002). Compared to the TAR versions, 9 of the 11 AR4 models show more or less pronounced decrease (especially strong in the two versions of the CGCM3.1 model) of the annual mean arctic precipitation. Arctic precipitation increases in the CCSM3 and ECHAM5/MPI-OM models, while there is no change in the Parallel Climate Model (PCM). It should be noted that the exclusion of the outlier versions of the CGCM from the Fig. 3b decreases the difference in the arctic annual mean precipitation between the TAR and AR4 subensembles by the factor of 2 (0.10 versus 0.05 mm day 1 ). Altogether, the multimodel ensemble mean of this subset shows a decrease of precipitation compared to the TAR simulations. The annual mean P E over the Arctic Ocean (70 90 N) simulated by the 21 models for the period is generally higher (multimodel mean of 0.45 mm day 1 ) than the existing observationally based climato-

8 578 J O U R N A L O F H Y D R O M E T E O R O L O G Y VOLUME 8 FIG. 4.P seasonal cycle over the Arctic Ocean ( ): (a) 21 models [first ensemble members compared against Bryazgin and ERA-40 ( )]; (b) model means for four subsets of models: 21 AR4 models, 11 AR4 models whose earlier versions were available from TAR, the corresponding 8 TAR versions, and 7 TAR versions (minus the TAR-generation CGCM outlier). logical estimate (0.38 mm day 1 ) of Korzun (1978; Fig. 3c). The larger model value could be partially attributed to the recent increases of precipitation discussed above (note that Korzun s estimate was based on earlier years observations). Over the same region, for the period , Cullather et al. (2000) give annual mean P E of mm yr 1 from ERA-15 and NCEP (see their Table 1). This corresponds to mm day 1. The estimate is higher than that given by Korzun for an earlier period and is closer to the model ensemble average. It is not presently possible to say whether the smaller values of Korzun are attributable to methodology (e.g., spatial interpolation, gauge correction) or to changes in P E from the earlier to the later period. While increases of Arctic precipitation may have occurred in earlier decades of the twentieth century (Kattsov and Walsh 2000), there are no indications of systematic changes of precipitation during the period on which many of our comparisons with reanalysis results are based. The 21-model mean seasonal cycle of P in the area N is in qualitative agreement both with the ERA-40 for the time period of , and with Bryazgin s climatology (Fig. 4a). While there are substantial differences among the individual simulations throughout the year (the seasonality of the intermodel scatter generally follows the seasonality of precipitation itself with the largest intermodel spread in early fall, and the least in late spring), it is noteworthy that the model mean varies seasonally almost entirely within the range between the observational data and the reanalysis. The amplitude of the model-mean seasonal cycle is smaller than that in the ERA-40 and Bryazgin. Compared to the IPCC TAR, the new models used for the IPCC AR4 appear to show an improvement, with a decrease of precipitation from fall to spring (Fig. 4b). For the evaluation of model performance in simulating precipitation and evapotranspiration over the Arctic Ocean terrestrial watersheds, four major river basins were chosen: the Ob, the Yenisey, the Lena, and the MacKenzie. Figure 5 compares the seasonality of precipitation in the IPCC AR4 simulations for each basin with the observationally based estimates of Serreze et al. (2003) and ERA-40 for the time period of The two datasets show good agreement with each other for all four basins. Although the simulated precipitation amounts clearly vary among the models, particularly in summer, most models qualitatively capture a realistic seasonal course of the precipitations in all four basins, which are characterized by summer maxima and winter minima. The only exceptions are the Goddard Institute for Space Studies Models E-R and E-H (GISS-ER and -EH), whose seasonal cycles of precipitation for the Ob and Yenisey show minima in the summer. Quantitatively, compared with the estimates by ERA-40 and by Serreze et al. (2003), the 21-model mean (as well as the 19-model mean with the two GISS models excluded) shows an underestimate in summer in the Ob Basin and an overestimate either in some seasons or throughout the year in the other three river basins. Among the problems responsible for the simulated precipitation biases and the intermodel scatters, the most suspicious are atmospheric large-scale circulation biases in winter (see Chapman and Walsh 2007) and the variety of both atmospheric convection and land hydrology schemes employed by the AR4 AOGCMs. It should be mentioned that summer precipitation over these watersheds is known to be strongly driven by surface evapotranspiration (Serreze and Etringer 2003), which is not easy to properly simulate. However, it is only through a series of controlled experiments with different models that the impact of different parameterizations on precipitation could be properly addressed. In the annual mean, the models show outstanding

9 JUNE 2007 K A T T S O V E T A L. 579 FIG. 5.P over AO terrestrial watersheds ( ): (a) the Ob; (b) the Yenisey; (c) the Lena; and (d) the MacKenzie. skill in producing precipitation in these river basins (Fig. 6). The differences of precipitation in the Ob and Yenisey River basins between observations and the 21- model mean are less than 0.1 mm day 1. The differences are larger in the Lena and MacKenzie River basins, exceeding by 0.2 and 0.4 mm day 1 the estimate by Serreze et al. (2003). Removing the outlier estimates by the GISS-ER and GISS-EH models only slightly increases the model-mean estimates of precipitation over the major river basins. Because of the high correlation between precipitation and evapotranspiration (Walsh et al. 1998), the annual mean P E shows even better agreement between the observational and model-mean estimates: correspondingly, 0.41 versus FIG. 6. Annual mean P and P E over four watersheds compared to observational estimates by Serreze et al. (2003) for the period : (a) the Ob; (b) the Yenisey; (c) the Lena; and (d) the MacKenzie.

10 580 J O U R N A L O F H Y D R O M E T E O R O L O G Y VOLUME mm day 1 for the Ob; 0.52 versus 0.55 mm day 1 for the Yenisey; 0.49 versus 0.51 mm day 1 for the Lena; and 0.39 versus 0.58 mm day 1 for the MacKenzie. 4. Simulation of the twentieth-century variability and trends Given the uncertainties in the climatologies of arctic precipitation and evaporation (evapotranspiration), it is not surprising that information on recent variations and trends of these quantities is quite limited and even more uncertain. Estimates by the reanalyses are subject to impacts of inhomogeneities resulting from changes in the input data (e.g., availability of satellite data after late 1970s), while estimates by the in situ observational data suffer from sparse observational network and instrument and measurement biases as noted earlier. The use of in situ measurements for trend determination is further complicated by the changes in the rain/snow ratio during periods of warming or cooling at high-latitude sites (Forland and Hanssen-Bauer 2000). Houghton et al. (1996, 2001) have consistently indicated twentieth-century increases of precipitation in northern high latitudes (55 85 N), as shown in Fig. 2.7 of Houghton et al. (1996). The increase is similar to that in Karl s (1998) Arctic region, which includes the area poleward of 65 N but excludes the waters surrounding southern Greenland. In both cases, the greatest increase appears to have occurred during the first half of the twentieth century. However, the time series are based on data from the synoptic station network, which is not only unevenly distributed but which has undergone changes over time. Nevertheless, the increase in the early twentieth century was reproduced by some pre-ar4 model simulations of twentieth-century climate (Paeth et al. 2002; Kattsov and Walsh 2002). Figure 7a shows the twentieth-century precipitation linear trends (mm day 1 per century) over the Arctic Ocean (70 90 N) in ensemble simulations with each of the 21 models. Wherever the number of ensemble simulations of each individual model was greater than one, maximum (MAX) and minimum (MIN) trend values and ensemble member means (Mean) were computed and are shown in this figure. (The three values obviously coincide for models with only one 20C3M simulation.) With the exception of three models {GISS- ER, GISS-EH, and Model for Interdisciplinary Research on Climate 3.2, medium-resolution version [MIROC3.2(medres)]}, all individual model ensemble members have positive twentieth-century trends in the arctic annual mean precipitation. All model ensemble means without exceptions are positive. This confirms that the models are robust in capturing the observed FIG. 7. Precipitation linear trends (mm day 1 per century) in all model ensemble members: maximum (MAX) and minimum (MIN) values for each model ensemble, and values for the ensemble means (Mean) over the Arctic Ocean (70 90 N) for : (a) annual mean; (b) DJF; and (c) JJA. general increase of arctic precipitation through the twentieth century (Kattsov and Walsh 2002). However, the twentieth-century trend of arctic precipitation is characterized by a pronounced seasonality: while the winter arctic precipitation shows tendencies similar to the annual means (Fig. 7b), the summer precipitation does not demonstrate any systematic change through the twentieth century (Fig. 7c). In this respect, the seasonality of the precipitation trend is consistent with the seasonality of the greenhouse warming of the models (Chapman and Walsh 2007). The precipitation trends for the period in the individual model runs do not reveal a systematic correspondence with the early/midtwentieth-century warming of the Arctic, even though some of the models that included natural forcings were able to reproduce such a warming at some time in the early/middle century (Wang et al. 2007). The study by Wang et al. utilized ensembles of twentieth-century simulations from the different models. Our analysis and others have shown that, at least for variables such as temperature and precipitation, the intermodel variance is considerably greater than the intra (within ensemble) variance. Nevertheless, all the nine models that both include

11 JUNE 2007 K A T T S O V E T A L. 581 FIG. 8.P E linear trends (mm day 1 per century) in all model ensemble members: maximum (MAX) and minimum (MIN) values for each model ensemble, and values for the ensemble means (Mean) over the entire Arctic Ocean terrestrial watershed for natural forcings and have more than one 20C3M simulations (see Table 1) show positive trends of Arctic precipitation in their ensemble means (not shown here). This can be regarded as an indication of a role that natural forcings may play in the midtwentieth-century arctic warming. Groisman and Easterling (1994) present data showing an increase of precipitation over northern Canada (poleward of 55 N) since For the period since 1960, the gauge-adjusted and basin-averaged precipitation data of Serreze et al. (2003) have little discernible indication of trends in the annual means over the Ob, Yenisey, Lena, and Mackenzie basins. However, summer precipitation over the Yenisey Basin shows a decrease of 5% 10% over the four decades since The variations of P in these regions appear to be largely associated with variations the atmospheric circulation. The analysis of the simulated trends over the four river basins also did not reveal any systematic features in either the annual mean or in summer precipitation, with two partial exceptions: the Yenisey and the Lena Basins annual mean precipitation amounts show positive trends in almost all individual model ensemble means. The time series of observed discharge of the great Siberian rivers (the Ob, the Yenisey, and the Lena) start in the mid-1930s ( edu/v3.0). Thus, they are long enough to be useful for evaluation of model performance in reproducing the evolution of P E through the greater part of the twentieth century. Unfortunately, while some model runs encouragingly agree with the observational time series for individual river basins, the successful simulation of the area-averaged annual mean P E evolution is not a robust feature of any model or any basin. An attempt to correlate even strongly smoothed time series of simulated P E and observed discharge time series resulted in unsystematic intraensemble correlations widely ranging from large positive values (up to 0.8 for the Yenisey) to negative values (the figures are not shown here). A more robust feature of the temporal variability of P E over the Arctic Ocean terrestrial watersheds appears to be the generally positive trends of this variable over the last third of the twentieth century. Figure 8 shows the linear trends simulated in all ensemble members from each model for the period over the entire terrestrial watershed draining into the Arctic Ocean. All but one ensemble means show positive trends, in some cases exceeding 0.1 mm day 1 per century. 5. Projections for the twenty-first century The most recent analysis of the twenty-first-century climate projections for the Arctic, including the Arctic Ocean freshwater budget components, was presented as a part of ACIA (Kattsov and Källén 2005; Walsh 2005). In that assessment, two scenarios B2 and A2 were considered in simulations with five models of the

12 582 J O U R N A L O F H Y D R O M E T E O R O L O G Y VOLUME 8 TAR generation. This section provides similar information for the IPCC AR4 models and adds two more scenarios, A1B and B1, which to our knowledge have not been analyzed specifically for the Arctic with an ensemble of TAR models. Figure 9 shows the geographical distributions of projected annual mean precipitation change for the A2, A1B, and B1 scenarios by the mid- (left) and late (right) twenty-first century relative to the baseline period The projected changes are averaged across the 13 models available for all three scenarios. The Arctic Ocean and terrestrial arctic regions of North America and Eurasia are the areas where the percentage increase in simulated precipitation is largest. For the A2 scenario, the projected increases of annual mean precipitation by vary in the Arctic from less than 5% in the Atlantic sector to locally above 40% in the central Arctic, and even higher in northeastern Greenland. The corresponding changes for the B1 scenario over the Arctic Ocean are generally less than 30%. The differences between the three different scenarios are comparatively small in the first half of the twenty-first century. Toward the end of the twenty-first century the differences between the scenarios increase but do not exceed the intermodel scatter. The general increase in high-latitude precipitation with global warming is a robust and qualitatively well understood result from climate change experiments. With increasing temperature, the atmospheric circulation s transport of moisture from lower to higher latitudes increases, leading to an increase in precipitation in the polar areas where the local evaporation is relatively small. For the Arctic Ocean north of 70 N, the 17 models with available A2 scenarios project an increase in annual precipitation by varying from 16% [Meteorological Research Institute Coupled General Circulation Model version (MRI-CGCM2.3.2)] to 57% [the Met Office Hadley Centre Global Environmental Model version 1 (UKMO-HadGEM1)] (Fig. 10). The 17-model mean change is 33%, which is much higher than the global mean (4.5%). In support of the argument that the increase of precipitation over the Arctic Ocean is driven largely by changes in moisture flux convergence, we note that the percentage increase of precipitation is largest in winter and smallest in summer (Figs. 11 and 12a) (even though evaporation is largest in summer). In this respect, the seasonality of the increases of precipitation and temperature are generally similar, consistent with an increase of the atmospheric water vapor and associated transport. The increases of arctic precipitation are also generally largest in the models with the largest warming, both in the Arctic and globally, consistent with Kattsov and Källén (2005). While over the Arctic Ocean precipitation is projected to increase in all seasons, such behavior is not the case for terrestrial watersheds. For example, for the Ob Basin, 8 of the 17 models in the A2 simulations project a decrease in summer precipitation by the end of the twenty-first century (Fig. 12b). The differences in precipitation change between the projections are affected by model differences and by noise associated with internal variability. To investigate the role of the latter factor alone, the annual/seasonal precipitation changes in the twenty-first century (A1B scenario) in each individual model were compared with the interannual variability (standard deviation) of precipitation in the 20-yr baseline period ( ) in all 20C3M ensemble simulations with that model. The signal-to-noise ratio is estimated by dividing the projected increase by the standard deviation, and averaging this ratio across the models (Fig. 13a). The model-averaged ratios begin to exceed the factor of 2 only in the second half of the twenty-first century in the annual means and in winter and fall, and mostly over the ocean. As an alternative depiction of the signal-to-noise ratio of the projected change of precipitation, Fig. 13b illustrates the effect of model differences on the precipitation change projections. The model-mean precipitation change is divided by the intermodel standard deviation of the projected changes. While geographical distributions and structures of the two ratios shown in Figs. 14a and 14b differ significantly, both figures suggest that in the Arctic, local precipitation increases may be very difficult to discern from natural variability in the next several decades. (See Kattsov and Källén 2005 for a relevant discussion.) Annual mean P E over the Arctic terrestrial watersheds, and accordingly the river discharge into the Arctic Ocean, increase in all scenarios. Figure 14 compares the changes of annual mean area-averaged P E in the A2, A1B, and B1 scenarios by the relative to over the Arctic Ocean (70 90 N); all Arctic Ocean terrestrial watersheds; and the Ob, Yenisey, FIG. 9. Geographical distributions of projected annual mean P change for the twenty-first century (13 models available for all scenarios): (left) and (right) vs (%) all models (single ensemble members): (a) B1, (b) A1B, and (c) A2.

13 JUNE 2007 K A T T S O V E T A L. 583

14 584 J O U R N A L O F H Y D R O M E T E O R O L O G Y VOLUME 8 FIG. 10. Annual mean P changes (%) from 1980 to 1999 by in A2 scenario: global and over the Arctic Ocean (70 90 N). Lena, and Mackenzie basins. The strongest relative increase in the river discharge is projected for the Lena (33% by the end of the twenty-first century in the A2 scenario); the weakest relative change is shown by the Ob with its 14% for A2 (compared, e.g., to the Lena s 20% for the weak B1 scenario). The projected increase of the total freshwater discharge into the Arctic Ocean by the end of the twenty-first century ranges from about 14% in B1 to more than 25% in A2. The changes of P E over the Arctic Ocean within N are projected to range from 13% (B1) to about 28% (A2). However, having in mind the difference in the area of the Arctic Ocean itself within the 70 N latitude and the area of its terrestrial watersheds, the absolute FIG. 11. Geographical distributions of seasonal changes (%) of P for A2 (17 models) in the two time slices ( and ): (a) DJF ; (b) JJA ; (c) DJF ; and (d) JJA

15 JUNE 2007 K A T T S O V E T A L. 585 FIG. 12. P seasonal changes (%) by (A2): (a) over the Arctic Ocean (70 90 N) and (b) the Ob watershed. value of the increase of the direct freshwater input from the atmosphere over the Arctic Ocean is only about half as large as the additional freshwater mass projected to be received from the surrounding land territories. 6. Conclusions Evaluation of the IPCC AR4 model simulations of the arctic hydrologic variables is complicated by uncertainties in the corresponding observational estimates. This is especially a challenge when the twentiethcentury variability and trends are addressed. Compared to the previous (TAR) generation of AOGCMs, there are some indications that models as a class have improved in simulations of the arctic precipitation; in particular, the model-mean bias is well within the range of uncertainty of the observational estimates, and the number of models reproducing the key characteristics (mean, seasonality, trends) of arctic precipitation has increased since TAR. At the present time, it is too soon to specify what particular improvements in model physics or/and numerics and resolution are responsible for this progress. Hopefully, ongoing analysis and, more importantly, controlled experiments with model resolution, cloud parameterizations, treatment of land surface processes, sea ice parameterizations, and other process formulations will help to sort out the possible causes in the coming years. The more complete documentation and assessment of recent model improvements in the IPCC s upcoming Fourth Assessment Report will likely serve as a stimulus for such experiments to address the reasons for the recent improvements. In spite of the observational uncertainties, it appears that the models still tend to oversimulate area-averaged precipitation over major river basins draining into the Arctic Ocean. Geographically, the AR4 model-mean precipitation biases in the Arctic and sub-arctic have retained their major patterns, which are at least partly attributable to the insufficiently resolved local orography, as well as to biases in large-scale atmospheric circulation and sea ice distribution. The P E over major terrestrial watersheds, which provides a measure of the river discharge into the Arctic Ocean, generally is also slightly oversimulated. However, it is not apparent that these biases and the precipitation biases are crucial for the credibility of the model projections of the twenty-first century arctic and global climate. Some agreement has been demonstrated by the models in reproducing positive trends of Arctic precipitation over the entire twentieth century and the latetwentieth century precipitation, as well as positive trends of area-averaged P E over the entire terrestrial watershed of the Arctic Ocean. All the nine models using natural forcings and having more than one twentieth-century simulation showed positive trends in their ensemble mean annual precipitation over the Arctic Ocean in the first third of the twentieth century. This feature may be attributed to the influence on the arctic precipitation of the natural forcings included in those simulations. With this exception, the evolution of the simulated arctic precipitation and Siberian river discharges during the first two-thirds of the twentieth century do not show systematic trends. When viewed in the aggregate, the IPCC AR4 AOGCMs are qualitatively consistent in projections of the Arctic Ocean freshwater components change in the twenty-first century. The precipitation over the Arctic Ocean and its watersheds is projected to gradually increase through the twenty-first century, showing much faster relative increase compared to the global mean changes. The arctic precipitation changes have a pronounced seasonality, with the strongest relative increase in winter and fall, and the weakest in summer. The river discharge into the Arctic Ocean increases for all scenarios from all major river basins considered. For

16 586 J O U R N A L O F H Y D R O M E T E O R O L O G Y VOLUME 8 FIG. 13. Changes in annual/seasonal mean precipitation (%) in the AR4 A1B projections by (left) early, (middle) middle, and (right) late twenty-first century. (a) The ratio between the model mean change from 1980 to 1999 and the standard deviation, obtained in all runs with all models for the period ; (b) the ratio between the model-mean change from 1980 to 1999 and the intermodel standard deviation, obtained for the corresponding changes from 1980 to 1999 in each individual model (single ensemble member for each individual model).

17 JUNE 2007 K A T T S O V E T A L. 587 FIG. 13. (Continued). each scenario, across-model scatter in projected changes is substantial, but smaller than the scatter between the different model climates in the baseline period. Even the differences between two single-model projections driven with different emission scenarios (e.g., A2 and B1) are generally less than the range of corresponding changes obtained in the model-ensemble simulations driven with the same scenario.

18 588 J O U R N A L O F H Y D R O M E T E O R O L O G Y VOLUME 8 Energy. We thank the three anonymous reviewers for their valuable comments. REFERENCES FIG. 14. Annual mean P E (discharge) change in A2, A1B, and B1 scenarios by over different regions (watersheds): the Arctic Ocean (70 90 N); all Arctic Ocean terrestrial watersheds; the Ob; the Yenisey; the Lena; and the Mackenzie. With regard to the overall freshwater budget of the Arctic Ocean, a salient finding is that the projected increase of river discharge exceeds the increase of the total amount of precipitation over the Arctic Ocean (at least within N), by as much as a factor of 2. This finding points to the importance of realistic simulations of the terrestrial surface moisture budget, particularly the evapotranspiration and runoff components. Both these hydrologic budget components have substantial observational uncertainties at present. As soils warm and thaw, the role of changing permafrost will further complicate the moisture fluxes at the terrestrial surface. More comprehensive evaluations of the terrestrial surface moisture fluxes, both simulated and observed, are therefore a priority for assessments of the robustness of the projected changes of the hydrologic budget of the Arctic. Acknowledgments. This study was supported by the U.S. National Science Foundation via the International Arctic Research Center of the University of Alaska Fairbanks (Subaward UAF of OPP ), the Japan Agency for Marine-Earth Science and Technology, the Russian Foundation for Basic Research (Grant ), and INTAS (Grant ). We acknowledge the international modeling groups for providing their data for analysis, the Program for Climate Model Diagnosis and Intercomparison (PCMDI) for collecting and archiving the model data, the JSC/ CLIVAR Working Group on Coupled Modelling (WGCM) and their Coupled Model Intercomparison Project (CMIP) and Climate Simulation Panel for organizing the model data analysis activity, and the IPCC WG1 TSU for technical support. The IPCC Data Archive at Lawrence Livermore National Laboratory is supported by the Office of Science, U.S. Department of Arzel, O., T. Fichefet, and H. Goosse, 2006: Sea ice evolution over the 20th and 21st centuries as simulated by the current AOGCMs. Ocean Modell., 12, Bogdanova, E. G., B. M. Ilyin, and I. V. Dragomilova, 2002: Application of a comprehensive bias-correction model to precipitation measured at Russian North Pole drifting stations. J. Hydrometeor., 3, Bryazgin, N. N., 1976: Yearly mean precipitation in the Arctic region accounting for measurement errors (in Russian). Proc. Arct. Ant. Res. Inst., 323, Chapman, W. L., and J. E. Walsh, 2007: Simulations of Arctic temperature and pressure by global coupled models. J. Climate, 20, Colony, R., V. F. Radionov, and F. J. Tanis, 1998: Measurements of precipitation and snow pack at Russian North Pole drifting stations. Polar Rec., 34, Cullather, R. I., D. H. Bromwich, and M. C. Serreze, 2000: The atmospheric hydrologic cycle over the Arctic Basin from reanalyses. Part I: Comparison with observations and previous studies. J. Climate, 13, Forland, E. J., and I. Hanssen-Bauer, 2000: Increased precipitation in the Norwegian Arctic: True or false? Climatic Change, 46, Gates, W. L., 1992: AMIP: The Atmospheric Model Intercomparison Project. Bull. Amer. Meteor. Soc., 73, Goodison, B. E., P. Y. T. Louie, and D. Yang, 1998: WMO solid precipitation measurement intercomparison: Final report. WMO Tech. Doc. 872, 212 pp. Groisman, P. Y., and D. R. Easterling, 1994: Variability and trends of total precipitation and snowfall over the United States and Canada. J. Climate, 7, Houghton, J. T., L. G. Meira Filho, B. A. Callander, N. Harris, A. Kattenberg, and K. Maskell, Eds., 1996: Climate Change 1995: The Science of Climate Change. Cambridge University Press, 572 pp., Y. Ding, D. J. Griggs, M. Noguer, P. J. van der Linden, X. Dai, K. Maskell, and C. A. Johnson, Eds., 2001: Climate Change 2001: The Scientific Basis. Cambridge University Press, 944 pp. Karl, T., 1998: Regional trends and variations of temperature and precipitation. The Regional Impacts of Climate Change: An Assessment of Vulnerability, R. T. Watson, M. C. Zinyowera, R. H. Moss, and D. J. Dokken, Eds., Cambridge University Press, Kattsov, V. M., and J. E. Walsh, 2000: Twentieth-century trends of Arctic precipitation from observational data and a climate model simulation. J. Climate, 13, , and, 2002: Reply. J. Climate, 15, , and E. Källén, 2005: Future changes of climate: Modelling and scenarios for the Arctic region. Arctic Climate Impact Assessment, C. Symon, L. Arris, and B. Heal, Eds., Cambridge University Press, Khrol, V. P., Ed., 1996: Atlas of Water Balance of the Northern Polar Area. Gidrometeoizdat, 81 pp. Korzun, V. I., Ed., 1978: World Water Balance and Water Resources of the Earth. UNESCO Press, 663 pp.