Simulations of Arctic Temperature and Pressure by Global Coupled Models

William L. Chapman Department of Atmospheric Sciences, University of Illinois at Urbana–Champaign, Urbana, Illinois

Search for other papers by William L. Chapman in
Current site
Google Scholar
PubMed
Close
and
John E. Walsh Department of Atmospheric Sciences, University of Illinois at Urbana–Champaign, Urbana, Illinois

Search for other papers by John E. Walsh in
Current site
Google Scholar
PubMed
Close
Full access

Abstract

Simulations of Arctic surface air temperature and sea level pressure by 14 global climate models used in the Fourth Assessment Report of the Intergovernmental Panel on Climate Change are synthesized in an analysis of biases and trends. Simulated composite GCM surface air temperatures for 1981–2000 are generally 1°–2°C colder than corresponding observations with the exception of a cold bias maximum of 6°–8°C in the Barents Sea. The Barents Sea bias, most prominent in winter and spring, occurs in 12 of the 14 GCMs and corresponds to a region of oversimulated sea ice. All models project a twenty-first-century warming that is largest in the autumn and winter, although the rates of the projected warming vary considerably among the models. The across-model and across-scenario uncertainties in the projected temperatures are comparable through the first half of the twenty-first century, but increases in variability associated with the choice of scenario begin to outpace increases in across-model variability by about the year 2070. By the end of the twenty-first century, the cross-scenario variability is about 50% greater than the across-model variability. The biases of sea level pressure are smaller than in the previous generation of global climate models, although the models still show a positive bias of sea level pressure in the Eurasian sector of the Arctic Ocean, surrounded by an area of negative pressure biases. This bias is consistent with an inability of the North Atlantic storm track to penetrate the Eurasian portion of the Arctic Ocean. The changes of sea level pressure projected for the twenty-first century are negative over essentially the entire Arctic. The most significant decreases of pressure are projected for the Bering Strait region, primarily in autumn and winter.

Corresponding author address: William L. Chapman, Department of Atmospheric Sciences, University of Illinois at Urbana–Champaign, Urbana, IL 61801. Email: chapman@atmos.uiuc.edu

Abstract

Simulations of Arctic surface air temperature and sea level pressure by 14 global climate models used in the Fourth Assessment Report of the Intergovernmental Panel on Climate Change are synthesized in an analysis of biases and trends. Simulated composite GCM surface air temperatures for 1981–2000 are generally 1°–2°C colder than corresponding observations with the exception of a cold bias maximum of 6°–8°C in the Barents Sea. The Barents Sea bias, most prominent in winter and spring, occurs in 12 of the 14 GCMs and corresponds to a region of oversimulated sea ice. All models project a twenty-first-century warming that is largest in the autumn and winter, although the rates of the projected warming vary considerably among the models. The across-model and across-scenario uncertainties in the projected temperatures are comparable through the first half of the twenty-first century, but increases in variability associated with the choice of scenario begin to outpace increases in across-model variability by about the year 2070. By the end of the twenty-first century, the cross-scenario variability is about 50% greater than the across-model variability. The biases of sea level pressure are smaller than in the previous generation of global climate models, although the models still show a positive bias of sea level pressure in the Eurasian sector of the Arctic Ocean, surrounded by an area of negative pressure biases. This bias is consistent with an inability of the North Atlantic storm track to penetrate the Eurasian portion of the Arctic Ocean. The changes of sea level pressure projected for the twenty-first century are negative over essentially the entire Arctic. The most significant decreases of pressure are projected for the Bering Strait region, primarily in autumn and winter.

Corresponding author address: William L. Chapman, Department of Atmospheric Sciences, University of Illinois at Urbana–Champaign, Urbana, IL 61801. Email: chapman@atmos.uiuc.edu

1. Introduction

Arctic surface air temperatures and sea level pressures are intimately linked in a suite of ocean–ice–atmosphere interactions. These interactions are complex and involve a number of feedbacks that can obscure diagnoses of climate change and make the projection of climate change particularly challenging in the Arctic. The albedo–temperature feedback, for example, is widely thought to amplify warming in the Arctic due to global greenhouse gas increases (Kattsov and Källen 2005; Serreze and Francis 2006).

The potential for an amplified Arctic response to increasing greenhouse gases has brought increased scrutiny to the region’s climate in the past decade. In addition, investigations of some Arctic climate parameters have shown large changes in the recent decades, fueling the fire of the Arctic climate focus (Polyakov et al. 2002a, b; Karl 1998). The Arctic Climate Impact Assessment (ACIA; Hassol 2004), for example, reports that winter surface air temperatures in the Arctic have risen +3° to +4°C over the past 50 yr. Meanwhile, Arctic sea level pressures have decreased over the past several decades over the Arctic Ocean (Walsh et al. 1996). Northern Hemisphere sea ice extent has decreased by 15%–20% in summer over the past half-century (Hassol 2004) and the three most recent summers have had record or near-record minimum ice extents in the Arctic (Stroeve et al. 2005). Given these intriguing recent trends and the potential for enhanced climate change, there is a clear need for accurate representation of Arctic processes in global climate models (GCMs) if model projections of the future climate in the Arctic are to be credible.

In late 2004 and early 2005, a suite of state-of-the-art GCM simulations were executed with common greenhouse gas forcing conditions in support of the Intergovernmental Panel on Climate Change (IPCC) Fourth Assessment Report (AR4). The IPCC AR4 is scheduled for publication in 2007. Historic simulations forced by observed greenhouse gas concentrations for the twentieth century and climate projections for the twenty-first century using three greenhouse gas concentration scenarios were performed by modeling centers around the world and the output was archived for evaluation by climate scientists and policy makers. The IPCC GCM output archive provides a unique opportunity to critically evaluate the “state of the art” of large-scale climate modeling in the Arctic, and bring those evaluations to bear on the climate change projections from these models.

The objectives of this evaluation include the following:

  1. a synthesis of the Arctic surface air temperatures and sea level pressures from a recent set of simulations performed by coupled GCMs;

  2. an assessment of the ability of these state-of-the-art models to simulate historic means and variability of these primary atmospheric variables for late-twentieth-century greenhouse gas forcing conditions;

  3. a synthesis of projections of surface air temperature and sea level pressures for the twenty-first century, and a quantitative comparison of the projected changes of temperature and pressures to the models’ internal variability and the across-model variance.

Our focus on surface air temperature and sea level pressure (SLP) is motivated by two considerations. First, as fundamental variables of state, surface air temperature and SLP characterize the near-surface thermal (temperature) and dynamical (SLP) climate. Second, the observational databases of these variables include suites of in situ measurements that are far more comprehensive, relative to the spatial scales of variability, than are databases of variables such as precipitation, cloudiness, and even wind. These in situ observations, in turn, are assimilated into operational analyses and reanalyses such as the one used here for evaluation of the models (section 2).

2. Model and observational data

Output from 14 IPCC AR4 GCMs listed in Table 1 represent the largest subset of model output for which surface air temperature and SLP data are available for the twentieth-century simulations and twenty-first-century projections forced by several IPCC greenhouse gas concentration scenarios. Simulations of the twentieth century use greenhouse gas concentrations and, in some cases, estimated sulfate aerosols (Table 1; see discussion in Wang et al. 2007). The projected greenhouse gas concentrations for the three twenty-first-century scenarios IPCC Special Report on Emission Scenarios (SRESA2, SRESA1B, and SRESB1) are described in detail in Nakićenović et al. 2000.

In many cases, modeling centers provide multiple ensemble member simulations for their twentieth-century simulations. We utilize all available ensemble members in our analysis of temperature and SLP variability for the late twentieth century, but in cases where composite weighting and variability statistics need to be preserved, we use only the first ensemble member. These exceptions are noted in the text.

The 40-yr European Centre for Medium-Range Weather Forecasts (ECMWF) Re-Analysis (ERA-40) directly assimilates observed air temperature and SLP observations into their reanalysis product. The ERA-40 is one of the most consistent and accurate gridded representations of these variables available, and therefore, a logical choice for observational analyses against which we validate the model biases of late-twentieth-century surface air temperatures and SLP. (Data and documentation for the ERA-40 can be found online at http://www.ecmwf.int/research/era/Products.)

To facilitate GCM intercomparison and validation against the reanalysis data, all monthly fields of GCM temperature and SLP are interpolated to the common 2.5° × 2.5° latitude–longitude ERA-40 grid. Our intercomparison of historic means use monthly, seasonal, and annual means for a late-twentieth-century 20-yr period (1981–2000). Projected changes of surface air temperature and SLP are based on three twenty-first-century time slices: 2010–29, 2040–59, and 2070–89.

3. Simulation of the present climate

a. Surface air temperatures

Surface air temperatures available for intercomparison include output from 14 GCMs and from the ERA-40 reanalysis for the 20-yr “present climate” time slice: 1981–2000. We compare surface air temperatures for this period against the corresponding observation-based annual, seasonal, and monthly means from the ERA-40 reanalysis. Figure 1 illustrates 1981–2000 mean surface air temperatures for the winter [December–February (DJF)], spring [March–May (MAM)], summer [June–August (JJA)], and autumn [September–November (SON)] seasons from the ERA-40 reanalysis (top row), a composite of temperatures from 14 IPCC AR4 GCMs (middle row), and the GCM–observation differences (bottom row). To the extent that the ERA-40 reanalysis accurately portrays the observed present-day climate for 1981–2000, the 14-GCM composite–ERA-40 differences represent the composite surface air temperature biases of the 14 GCMs. Across all seasons, the spatial patterns of mean temperatures simulated by the GCMs are very similar to those of the ERA-40 and the resulting biases are small. It would be difficult to argue that the GCMs are not simulating seasonal mean temperatures accurately in the Arctic from this side-by-side comparison. One possible exception is a lack of detail in the GCM composites over the major topographic features. The coarser grid resolutions of the GCMs and the averaging of the GCM fields over 14 data sources likely cause the smoothing.

Details of the seasonal surface air temperature biases show the model’s average surface air temperatures are 1°–2° too cold over much of the domain. The largest regional biases occur over the Barents Sea where GCM temperatures average 8°–12°C cooler in winter and 6°–8°C cooler in spring than the corresponding observed surface air temperatures. These extreme temperature biases are caused primarily by an oversimulation of the models’ sea ice extent in the Barents Sea when compared to observations (Arzel et al. 2005). The presence of sea ice in the models permits the intrusion of polar continental air masses over the Barents Sea, insulated from the effects of the relatively warm underlying ocean by the exaggerated sea ice cover. A near-constant influx of relatively warm Atlantic surface water combined with a divergent surface wind stress field keeps the Barents generally ice free in the present (observed) climate. The lack of negative biases in the GCM-simulated temperatures near the Gulf Stream and North Atlantic currents implies that the Barents Sea ice bias has its origin in ocean dynamics (currents) and/or divergent surface wind stresses rather than in local thermodynamic forcing by the atmosphere.

The Barents Sea cold bias is a fairly robust feature among the GCMs. Simulated temperatures are too cold in 12 of the 14 GCMs, and are apparent as the most prominent feature in the composite annual temperature bias in Fig. 2. Only the Max Planck Institute (MPI) ECHAM5 and the National Center for Atmospheric Research (NCAR) Community Climate System Model version 3(CCSM3) simulate temperatures that are warmer than observed in the Barents Sea. Most of the other large cold biases are found near sea ice–covered areas.

Some of these extreme biases occur over the marginal ice zones [e.g., NCAR Parallel Climate Model version 1(PCM1), the Commonwealth Scientific and Industrial Research Organisation Mark version 3.0 (CSIRO Mk3.0) the Centre National de Recherches Météorologiques (CNRM) Climate Model version 3 (CM3)] pointing to an oversimulation of seasonal sea ice in these models. Others occur over the central pack ice [e.g., the Canadian Centre for Climate Modelling and Analysis (CCCMA) Coupled GCM version 3.1 (CGCM3.1) and the Meteorological Research Institute (MRI) Coupled GCM version 2.3.2A (CGCM2.3.2A)], raising the possibility that these models undersimulate the winter fraction of leads and open water in the central Arctic.

Historically, flux adjustment algorithms used by some models have helped to constrain simulations from deviating far from observed climate means. As recently as the IPCC Third Assessment, almost half of the GCMs utilized some form of flux adjustment. In the current suite of 14 GCMs, only two models, the CCCMA CGCM3.1 and MRI CGCM2.3.2A, retain limited flux adjustment in their design. Nevertheless, we compare output from the models using flux adjustment separately from those that do not whenever possible. In the case of patterns of GCM surface air temperature biases, Fig. 2 shows that the two flux-adjusted models have some of the largest negative biases over the Arctic Ocean. While these regional biases are not necessarily representative of those globally, they are, nevertheless, counterintuitive. (Seasonal and monthly means of these biases and the full suite of flux adjusted versus nonflux-adjusted bias comparisons are available online at http://igloo.atmos.uiuc.edu/IPCC/.)

The Geophysical Fluid Dynamics Laboratory (GFDL) was the only modeling center to provide output to IPCC AR4 from two versions of the same model and they were successful in significantly reducing surface air temperature biases from GFDL Climate Model version (CM2) to the updated GFDL CM2.1. While the largest biases in GFDL CM2 occur over regions of highly varying terrain, the resolution of the two GFDL simulations is identical, so upgrades in model physics and parameterizations appear to be responsible for the improved simulations. (Further diagnoses of the biases of the individual GCMs is facilitated by browsing the archive of monthly and seasonal surface air temperature biases provided for each model online at http://igloo.atmos.uiuc.edu/IPCC/.)

Walsh et al. (2002) report Arctic biases from a suite of five IPCC Third Assessment Report (TAR) GCMs (Houghton et al. 2001), all of whose centers have provided results from updated GCMs to the IPCC AR4 archive. The availability of the output from the IPCC TAR models (circa 2000, available online at http://igloo.atmos.uiuc.edu/ACIA/), and the current AR4 suite of models affords the opportunity to directly compare biases from the previous suite of five GCMs to biases from corresponding GCMs updated and included in the AR4 suite. Figure 3 shows the change in magnitude of the surface air temperature biases from the TAR GCMs to the AR4 suite (AR4 − TAR). Decreases (greens and blues) indicate regions where biases relative to ERA-40 have been reduced in the updated (AR4) GCMs; yellows and reds indicate higher biases for the updated GCMs. Two broad areas of improvement are noted in the annual averages: 1) Barents and Norwegian Seas and 2) the Sea of Okhotsk. The contributions to the annual mean bias changes for both these features come primarily from the winter and spring months. Pockets of notable decreases of apparent skill in the updated models include the high terrain areas of Alaska and northwest Canada for all seasons. While the picture of improvement in model simulations of surface air temperature over the Arctic is mixed in this analysis, it is important to note that this conclusion is clouded by the fact that three of the five models used in the IPCC TAR have removed flux adjustment techniques during the model “upgrades” to the IPCC AR4 versions. Four of five GCMs used flux adjustment in the TAR while only one of those same models continue to use flux adjustment in the AR4. The significant improvements in simulations that have allowed the modeling centers to forego flux adjustment techniques in their latest model incarnations are not reflected in these comparisons.

A summary of the individual model performance in simulating surface air temperatures over the challenging Arctic Ocean area can be seen in the area-averaged annual and seasonal root-mean-square (RMS) errors for each model (Fig. 4). Mean winter (DJF) rms errors average 3 times larger than corresponding summer (JJA) values. RMS errors are second largest in autumn (SON), third largest in spring (MAM), and smallest in summer (JJA). The annual mean rms errors range from about 2°C for the MPI ECHAM5, NCAR CCSM3, and GFDL CM2.1 models to as high as 7°C for the CSIRO Mk3.0 model. Skill in simulating winter surface air temperatures over the Arctic Ocean varies considerably across the models while summer simulations are more consistent. The range of rms errors for summer is 1.5°–3.75°C, while the range for winter values is about 3 times as great (2.5°–11°C). The MPI ECHAM5 and the NCAR CCSM3. GCMs are notably good performers over all seasons. The annual mean RMS errors for the flux-adjusted models are more than 2°C higher than those for the nonflux-adjusted GCMs. This counterintuitive relationship is maintained throughout all four seasons.

To complement our assessment of the GCM climate means, we compare the variability of the late-twentieth-century GCM simulations to corresponding reanalysis fields. Figure 5 includes 14-model composite annual and seasonal plots of the interannual variability expressed as standard deviations of the 20 annual and seasonal means, one for each year of 1981–2000 (column 2). The first column contains the corresponding annual and seasonal interannual variability of surface air temperatures from the ERA-40 reanalysis. There is considerable agreement between the modeled and observed interannual variability both in magnitude and spatial pattern of the variations and the seasonality of the variability is also well simulated. Key attributes of the observed variability are also seen in the GCM output: 1) maximum variations over high-latitude landmasses during winter and 2) minimum year-to-year variability over summer sea ice. (Monthly plots and variability statistics for each individual GCM can be found online at http://igloo.atmos.uiuc.edu/IPCC/.)

The variability in the biases shown in Fig. 2 hints at the significant underlying variability between the model simulations of annual and seasonal temperature means. The across-model variability of annual and seasonal surface air temperatures shown in column 3 in Fig. 5 are computed as standard deviations of 14 values, one for each GCM. The across-model variability is generally larger than the interannual variability for most seasons, but the key features of the spatial patterns are similar. The largest winter, spring, and autumn variability, both interannual and across model, occurs over regions of high interannual sea ice variability. These same regions exhibit high model-to-model sea ice variability. The presence, or lack of, sea ice during the winter, spring, and autumn months has large impacts on the localized temperature variations, both from one year to the next and from one model to another. Conversely, the presence of sea ice during summer constrains the temperatures to the melting point and severely limits variability, both interannually and across models.

We compare the across-model variability of the 12 GCMs not utilizing flux adjustment in their formulation with the across-model variability derived by including all models (flux adjusted and nonflux adjusted). One might speculate that adding flux-adjusted models would decrease the across-model scatter for simulated surface air temperatures by including models constrained to a predetermined state. In fact, including the two large biased, flux-adjusted models in the across-model scatter actually increases the across-model variability in this archive of GCM output.

In section 4, we compare GCM projections for three twenty-first-century 20-yr time slices. To place the projected temperature changes into the context of the models internal variability we utilize the “preindustrial control simulations” by dividing the 300–1000-yr-long control simulations into 15–50 unique 20-yr time slices. These control simulations are performed using constant greenhouse gas and aerosol concentrations from the preindustrial era (circa 1850). The composite internal variability defined as the 14-model mean of standard deviations of the 15–50 20-yr mean annual and seasonal temperatures are shown in the right-hand column in Fig. 5. This measure of variability, which can be thought of as the standard deviation of 15–50 unique 20-yr ensemble control runs, is considerably smaller than both the across model variability and the interannual variability for all seasons. The spatial patterns, however, are similar to those of the interannual variability.

Finally, we measured the ability of the GCMs to form a consensus by comparing the across-model standard deviations of surface air temperature from the IPCC AR4 GCMs with those from the previous IPCC TAR suite of GCMs. The across-model variability for the annual means is generally smaller for the IPCC AR4 GCMs than for the previous generation GCMs (not shown). Winter differences indicate lower across-model variability in the most recent GCM output, but slightly higher variability for the summer months. The summary of the analysis of the changes of biases and across-model standard deviations from the IPCC AR4 GCMs to previous generation models is that more recent simulations are generally better in winter and more consistent across models. Summer errors, however, are slightly larger and the across-model agreement is generally worse.

b. Sea level pressure

Realistic simulations of SLP fields are imperative if GCMs are to be used as meaningful climate diagnostic tools. A GCM’s SLP distribution is a key determinant to its near-surface temperature and precipitation distributions, and any biases in the model’s SLP gradients will adversely affect wind stresses and surface fluxes in the coupled model. The SLP gradient field in the Arctic is particularly important because near-surface winds are primary drivers of sea ice transport and the resulting sea ice concentration and thickness distributions. Biases in simulated Arctic SLP patterns can have global effects by altering the advective transport of sea ice, and therefore, freshwater, into the North Atlantic. Significant variations in North Atlantic surface freshwater have been shown to reduce convection, deep-water formation, and ultimately, reduce the ocean’s global thermohaline circulation (Dickson et al. 2000).

While the GCMs capture the first-order seasonal SLP patterns over the northern high latitudes, differences in the details between the GCM composite means and the observed SLP fields are more apparent than for the surface air temperatures (Fig. 6). The observed (ERA-40) North Atlantic storm track, for example, extends east through the Barents Sea and into the Kara Sea, whereas the GCM-simulated storm track ends in the Barents Sea. For the winter mean, the truncation of the North Atlantic storm track results in a 14-GCM composite-observed SLP bias greater than +9 mb in the Barents Sea. The shortened storm track broadens and shifts the center of the anticyclone in the Arctic Ocean. The well-documented Beaufort high circulation in the Arctic is the main driver for sea ice transport from the north coast of Asia, eastward via the “Transpolar Drift Stream,” into the North Atlantic through Fram Strait. The northward shift and broadening of the simulated versus observed Beaufort high has subtle but significant impacts on the sea ice advection pattern and, therefore, the ice thickness and concentration distributions in the Arctic. In the simulations, sea ice is transported farther west along the north coast of Asia and into the Kara and Barents Seas before being ejected into the North Atlantic. Also, strong low pressure systems are known to advect warm water into the Barents Sea and to mix waters locally, thereby aiding in keeping this region sea ice free. Fewer simulated storms entering this region in the GCM simulations (more anticyclonic flow) have the opposite effect (i.e., advection of sea ice into the region and a contribution to the models’ cold bias in the Barents region).

As evidence to support the above hypothesis, we note that the individual GCMs simulating the largest positive SLP biases (Fig. 7) over the Barents Sea are the same models that produce the largest negative temperature anomalies over the Barents Sea [GFDL CM2, the Goddard Institute for Space Studies Model E-R (GISS-ER), MRI CGCM2.3.2A, the U.K. Met Office (UKMO) Third Hadley Centre Coupled Ocean–Atmosphere GCM (HadCM3), and CSIRO Mk3.0], and vice versa (NCAR CCSM3). A circumpolar ring of negative SLP biases around 55°–60°N latitude in the winter, spring, and autumn seasons surrounds the positive SLP bias in the Barents Sea. This ring of negative biases appears to approximately balance the surplus represented by the Barents Sea positive bias.

Simulated SLPs in summer are also higher by 3–5 mb than the observed SLP pattern, primarily over the observed sea ice–covered portion of the Arctic Ocean and also over Greenland. Contrary to the other three seasons, the maximum positive bias in summer occurs nearer the North Pole, rather than the Barents Sea.

In addition to the effects on the Arctic sea ice distribution, the Barents Sea SLP bias likely affects the local surface heat and moisture exchanges, cloud cover and cloud radiative effects, large-scale circulation patterns, and precipitation amounts and types in the Barents region. Kattsov et al. 2007 report a general overestimate of precipitation throughout much of the Arctic by these IPCC models when compared with observed precipitation datasets. However, their analysis shows an underestimate of the precipitation amounts in the Barents and Kara Seas, likely caused in large part by the positive SLP bias described here.

Like the surface air temperature bias, the positive SLP bias in the Barents Sea is a robust feature among the models. Figure 7 shows that the bias occurs in the 14-GCM composite annual SLP field (upper left) and in 13 of the 14 individual GCMs. Only the NCAR CCSM3.0 differs in this regard, with generally negative SLP biases over the entire Arctic and North Atlantic Oceans and adjacent land areas.

Figure 8 shows the change in magnitude of the SLP biases from the IPCC TAR GCMs to the present suite (AR4–TAR). As in the temperature bias changes (Fig. 3), decreases (greens and blues) indicate regions where biases from ERA-40 have been reduced in the new-generation GCMs; yellows and reds indicate higher biases for the updated GCMs. Winter biases in the North Pacific have improved dramatically in the new-generation models. A more seasonally persistent change, however, is that the biases are now considerably smaller over the eastern Arctic Ocean in the annual mean and in all four seasons. The GCMs, while still overestimating the SLP in the Barents Sea by a considerable amount, are not as biased as they were in previous incarnations of the models. Again, these metrics of model improvement are only part of the story. They should be factored into the improvements attained by three of the five modeling centers in permitting them to forego flux-adjustment techniques in the AR4 incarnations of their respective models.

It has recently been hypothesized by Bitz (cf. Walsh et al. 2005) that smoothed topography, especially over Greenland, in the relatively coarse-resolution GCMs may lead to reduced cyclogenesis downstream (i.e., near the Barents Sea). If true, the resolution issue may be to blame for the positive SLP biases, excess sea ice, and hence, the surface air temperature biases in the Barents Sea. One notable “improvement” in the IPCC AR4 model versions is a significant enhancement of the spatial resolution of the simulation grids for several of the five GCMs used in the AR4–TAR comparison. The improvements in the GCM simulations of SLP downstream of Greenland in the models with higher spatial resolution are consistent with the conjecture that the SLP biases are a manifestation of the smoothed Greenland topography of the GCMs. As further evidence of this link between spatial resolution and SLP bias in the Barents Sea, we evaluated a correlation of +0.24 between atmospheric model grid cell area and corresponding model annual SLP bias in the Barents Sea for the 14 GCMs in this analysis. While these linkages are not clearly definitive, collectively they provide evidence that the representation of the Greenland topography and its interaction with the model physics may be a good launching point for potential model reforms.

Figure 9 shows the corresponding area-weighted RMS errors for SLP over the ocean areas: 70°–90°N. While it is not always true that the models with the smallest temperature errors also simulate SLP accurately, it is worth noting that the GFDL CM2.1, CNRM CM3, and the MPI ECHAM5 GCMs perform much better than average by both these metrics. In fact, the MPI ECHAM5 GCM outperforms all other GCMs for both the SLP and surface air temperature RMS errors. The NCAR CCSM3 GCM, on the other hand, while outperforming most models for temperature RMS errors, has some of the largest annual and seasonal SLP RMS errors. SLP RMS errors are largest in winter and autumn, and smallest in summer, but this is not nearly as consistent across models as for the temperatures. The NCAR CCSM3 summer RMS errors are more than nearly double the RMS errors for winter. Annual SLP RMS errors range from 2 mb (MPI ECHAM5) to almost 9 mb (NCAR CCSM3), while the across-model range of winter SLP RMS errors is comparable to the across-model range of summer SLP RMS errors.

Figure 10 shows 14-GCM composite annual and seasonal distributions of interannual variability of SLP, expressed as standard deviations of the 20 annual and seasonal means, one for each year of 1981–2000 (second column). The left column contains the fields of annual and seasonal interannual variability of SLP from the ERA-40 reanalysis. While the agreement between the modeled and observed interannual variability is not as close both in magnitude and spatial pattern as those for air temperatures, the main features and seasonal variations of the observed variability are captured by the models. Key features of the observed variability that are also seen in the GCM fields include the following: 1) maximum variations during winter and a minimum in summer and 2) relative maxima in variability east of Greenland and the North Pacific in all seasons.

The magnitudes of the simulated variability are less than observed over the North Atlantic and the eastern Arctic Ocean, while they are slightly larger in the Pacific sector. The lack of simulated SLP variability near Iceland and to the east is consistent with the conclusion that the models are not only underrepresenting the frequency of Icelandic low pressure systems, but their eastward migration is not as extensive as observed low pressure systems. (Monthly plots and SLP variability statistics for each individual GCM can be found online at http://igloo.atmos.uiuc.edu/IPCC/).

The across-model standard deviations of annual and seasonal SLP shown in the third column in Fig. 10 are generally larger than the interannual variability statistics. The largest across-model variability is confined to the Arctic and North Atlantic Oceans in all seasons with winter, spring, and autumn having comparable magnitudes and summer somewhat smaller magnitudes. The general increase in across-model variability with latitude contrasts with the interannual variability that has its maxima confined to the climatological Atlantic and Pacific storm tracks. Greenland stands out as having the highest across-model SLP variability in the domain. The unique topography of Greenland and the widely varying spatial resolutions of the different GCMs combine to ensure that Greenland is represented quite differently from one model to the next. Also, the extrapolation of surface pressure to SLP is further affected by differences in simulated air temperatures near the Greenland surface.

As with temperatures, the internal variability of 20-yr means of SLP simulated by the preindustrial control runs is much smaller than both the across-model and interannual variabilities. The procedure for measuring this internal variability is analogous to that described for the temperatures (section 3a). The spatial patterns of internal variability more closely resemble those of the interannual variability than they do the across-model variability with maxima located near both the Icelandic (Atlantic) and Aleutian (Pacific) storm tracks.

4. Projections for the twenty-first century

a. Projected surface air temperature

Three different sets of prescribed greenhouse gas forcing for the twenty-first century are used in the IPCC AR4 GCM projections, representing a range of emission scenarios: SRESB1, SRESA1B, and SRESA2. Time series of two centuries of GCM-simulated and projected temperature change area averaged for 60°–90°N are plotted in Fig. 11. The twentieth-century annual mean temperature anomalies (black), relative to 1980–99 means, are plotted for single ensemble members of each of the 14 GCMs listed in Table 1. Twenty-first-century-projected annual temperature changes are also plotted for first ensemble members from each of the models for the three greenhouse gas forcing scenarios: (SRESB1 in blue, SRESA1B in green, and SRESA2 in red). While all models simulate warming in the Arctic for every scenario beyond the year 2030, the ranges of the projected changes vary widely by forcing scenario. By the end of the twenty-first century, temperature changes projected by the SRESB1 scenario range from +1° to +5.5°C, the SRESA1B range is from +2.5° to +7.0°C, the SRESA2 range is from +4.0° to +9.0°C.

The spread in the twenty-first-century-projected temperatures shown in Fig. 11 illustrates that uncertainties in projected temperatures vary from one model to the next and across all greenhouse gas forcing scenarios. To assess the relative impact of these two sources of uncertainty, we evaluated 1) the intermodel variability for each scenario, expressed as centered 11-yr running means of across-model standard deviations of annual temperature change; and 2) the cross-scenario variability, expressed as standard deviations of projected temperature change across the three scenarios. The evolution of these two measures of variability is plotted as time series in Fig. 12. The across-model and across-scenario uncertainties in the projected temperatures are comparable through the first half of the twenty-first century, but increases in variability associated with the choice of greenhouse gas scenario begin to outpace those of the across-model variability by about year 2070. By the end of the twenty-first century, the cross-scenario variability is about 50% greater than the across-model variability.

We have evaluated the projected surface air temperature change for three 20-yr time slices of the twenty-first century by plotting projected changes in temperature relative to 1981–2000 means for annual, seasonal and monthly means. (Plots are available online at http://igloo.atmos.uiuc.edu/IPCC for the 3 IPCC greenhouse gas forcing scenarios, 14 GCMs, and the 14-GCM composites.) For the rest of this evaluation, we summarize results only for the IPCC SRESA1B scenario, which is considered the “middle-of-the-road” forcing scenario (see Fig. 11).

Annual and seasonal temperature changes for the 2070–89 time slice from projections made by the 14 GCMs and the 14-model composite are presented in Fig. 13. The annual composite shows warming over most of the domain. The greatest warming is found over the Arctic Ocean and adjacent landmasses with the maximum warming (+6.5°C) near the Barents Sea. Eleven of 14 models show the maximum warming in the Barents Sea. Interestingly, this is the same location of the greatest GCM ERA-40 biases (section 3a). The implication is that while the GCMs overestimate the sea ice in the Barents Sea for the late twentieth century, a modest increase in greenhouse gas forcing rectifies the sea ice and associated surface air temperature biases in the Barents Sea. The minimum warming (slight cooling in several models) occurs in the North Atlantic Ocean just south of Greenland and Iceland, where deep-ocean mixing occurs.

The 14-GCM composite and individual GCM winter plots also show the warming maximum in the Barents Sea and other marginal sea ice zones, with the exception of the NCAR CCSM3 GCM, which projects maximum warming over the central Arctic. The winter pattern of warming appears consistent with a reduction in the Northern Hemisphere sea ice area. Regions of the Arctic Ocean with sea ice in the present climate but absent in the projections experience temperature changes of +10° to +18°C during winter under this scenario.

The magnitude of the Arctic Ocean warming in autumn is similar to that of the winter months. In fact, nearly half of the models [the Model for Interdisciplinary Research on Climate 3.2, medium-resolution version (MIROC3.2medres), MRI CGCM2.3.2A, the Institute of Numerical Mathematics Coupled Model version 3.0 (INM-CM3.0, Russia), L’Institut Pierre-Simon Laplace Coupled Model version 4 (IPSL CM4), and HadCM3] project autumn temperature changes exceeding those of winter. Again, the mechanism for these extreme temperature changes is sea ice related. The models project significant melting of the perennial sea ice cover during the summer months, together with a delay in the refreezing date, causing extensive open-water areas prevalent during a significant portion of the autumn. The present-day climate has sea ice cover over most of the Arctic Ocean during the autumn months. Given the lack of solar radiation at these high latitudes during the autumn months, surface temperatures cool rapidly without an ice-free open ocean constraining the surface temperatures to near freezing. The temperature change pattern for the summer months is quite consistent across all models with very little temperature change projected over the Arctic Ocean and relatively mild (+2° to +4°C) warming over the continents.

The internal variability and the across-model scatter of the surface temperatures in the Arctic are large compared to the rest of the earth, especially over land. To place the projected temperature changes into the context of these separate types of variability, we construct two signal-to-noise ratios defined as the mean projected changes for a given time slice divided by measures of 1) the internal variability of the model and 2) the across-model standard deviation (of the annual or seasonal mean temperature change). The IPCC GCM archive provides “control run” output for simulations spanning 300–1000 yr using greenhouse gas concentrations from the preindustrial era. These simulations permit the evaluation of internal variability of the models as standard deviations of 15–50 unique 20-yr annual, seasonal, and monthly means from the extended preindustrial time series. The 14-GCM composite annual and seasonal ratios of projected changes to corresponding model variability are shown in Fig. 14a, in which we highlight surface air temperature changes exceeding two standard deviations of the preindustrial control climate by a transition from the pale yellow to deeper orange color in the images. (The results for each GCM can be found online at http://igloo.atmos.uiuc.edu/IPCC/, but we present a summary of the annual and seasonal 14-GCM composites of these signal-to-noise ratios here in Fig. 14.)

Projected temperature changes exceeding two standard deviations appear in Fig. 14a in the 2010–29 time slice over large portions of the Arctic. The absence of significant changes for some regions and seasons in the early twenty-first century is attributable to either small projected changes, as in the case of the western North Atlantic, or large internal model variability, as in the case of the Alaska region. The annual and autumn composites show the greatest warming for this time slice, but all seasons include large regions with significant warming relative to the baseline climate. The annual composite shows the largest signal-to-noise ratio in the 2070–89 time slice, with temperature changes projected to exceed 16–18 standard deviations over much of the Arctic Ocean. The impacts of expressing the projected temperature as these ratios are illustrated in two examples. 1) While winter mean temperatures in Alaska are projected to increase 2°–3°C (among the largest projected land area changes) by 2010–29, the simulated (and observed) temperature variability in this region is large enough that the projected changes do not exceed the 2σ threshold. 2) In contrast, the relatively mild projected changes (+1°C) over the Arctic Ocean in summer easily exceed the 2σ threshold by 2040–59.

Addressing the projected temperature change signal relative to the across-model variability, signal-to-noise ratios of projected changes divided by the across-model standard deviations for the corresponding time slice are displayed in Fig. 14b. While these ratios do not reach the magnitudes of the ratios in Fig. 14a, the annual mean temperature projections exceed two standard deviations over half the Arctic Ocean in the first time slice (2010–29). Portions of each of the seasonal changes also exceed the 2σ threshold in these early projections. By the 2040–59 period, the annual mean warming is larger over all of the Arctic Ocean and nearly all of the Arctic land areas by two standard deviations.

b. Projected sea level pressure

Projected changes of annual and seasonal mean change of SLP during 2070–89 are plotted in Fig. 15 for 14 GCMs and the 14-model composite. The composite annual SLP change is negative for the entire Arctic north of 60°N by 0 to −4 mb. The broad SLP reduction is robust among the GCMs with all 14 models showing a somewhat consistent pattern of change. Many models project slight SLP increases over the North Atlantic, and a few project slight increases near the Gulf of Alaska.

Perhaps the most striking feature of the seasonal mean SLP projected changes is the winter decrease in SLP centered over the Bering Strait and extending north into the Chukchi Sea and south into the Bering Sea. Twelve of the 14 models have relative minima of varying magnitudes located over the Bering Strait in winter and all of the models project decreased SLP in this region. Perhaps coincidentally, the three best-performing models (MPI ECHAM5, GFDL CM2.1, and CNRM CM3), in terms of temperature and SLP biases (from sections 3a and 3b), project the largest SLP decreases over the Bering Strait region (all with decreases exceeding 6 mb) in winter. This feature is likely a manifestation of a reduction in sea ice and an associated shift in the surface temperature gradient pattern. Accordingly, Aleutian lows are projected to follow a more northerly storm track into the west coast of Alaska with increasing frequency, rather than along the Aleutians. In this scenario there would be significant impacts on the region’s sea ice distribution, as well as temperature advection, precipitation distribution, and coastal erosion in Alaska and the Anadyr Plateau. The projected SLP decrease is also present to a lesser extent in autumn; however, it is absent in spring and replaced by slight pressure increases in summer.

We note that the pattern of circulation change projected by the models is not confined to the surface. A pattern of change consistent with the SLP change is noted in the projected 700-mb geopotential height fields and, to a lesser extent, the 500-mb height fields. The composite winter 500-mb heights projected for 2040–59 (included in the Web archive) show a general pattern of increasing heights across the Arctic domain. Superimposed on this mean increase is a pattern of generally lower heights over northeast Siberia and higher heights over Alaska. This wave couplet change is the dominant pattern of change for this season and period. The gradient of 500-mb height change in the Pacific hemisphere is consistent with an increase in surface low pressure frequency and/or intensities near the Bering Strait.

Composite signal-to-noise plots for the SLP projection time slices are shown in Fig. 16. These ratios based on internal variability (Fig. 16a) and across-model variability (Fig. 16b) are constructed analogously to the temperature ratio fields in Fig. 14. By both measures of variability, the decrease of pressure in the Bering Strait/Chuckchi Sea exceeds two standard deviations in the annual mean by 2040–59 and, in the case of the internal variability based ratio, exceeds six standard deviations by 2070–89. The salient message from these ratio plots is that the SLP values in the Bering Strait region bear watching for early signals of climate change in the Arctic even though the region’s variability is high and the model uncertainties are large.

5. Conclusions

The results of our assessment of the IPCC AR4 surface air temperatures and sea level pressures in section 3 emphasize the means and variability of the late-twentieth-century simulations and their biases from corresponding observational reanalysis fields. In section 4 we document the projected temperature and pressure changes from the baseline simulations and place the changes into the context of their internal and across-model variability. Primary conclusions include the following.

  • Simulated composite GCM surface air temperatures for 1981–2000 are generally 1°–2°C colder than corresponding observations with the exception of a cold bias maximum of 6°–8°C in the Barents Sea. The Barents Sea bias, most prominent in winter and spring, occurs in 12 of the 14 GCMs and corresponds to a region of oversimulated sea ice.

  • Surface air temperature biases from the IPCC AR4 suite of models are generally comparable to those from the previous generation models (IPCC AR3). Regionally, simulations of air temperature have improved in the most recent models over the Barents Sea and the Sea of Okhotsk, but have degraded over the mountainous regions of northwest North America.

  • Winter model biases of temperature for the Arctic Ocean, expressed as RMS errors from observational reanalysis fields, average 3 times larger than those for summer, but the across model scatter of the winter errors are 4 times larger than the corresponding summer errors. The MPI ECHAM5 and the NCAR CCSM3 GCMs outperform the other models for most seasons.

  • Patterns of variability of observed and simulated surface air temperatures agree very well across all seasons. Model-derived variations in sea ice extent appear to be the major contributor to 1) the patterns of natural variability of surface air temperature and 2) the across-model scatter of the surface air temperature simulations in the cold seasons. Sea ice acts to exaggerate interannual and across model variability in the cold season(s) while inhibiting the same temperature variability during the melt season.

  • While simulations of SLP are generally better than in the previous generation of GCMs, positive biases are still found in the Barents and Kara Seas and are consistent with a continued truncation of the North Atlantic storm track. The resulting surface wind stresses associated with these circulation biases are apparently large enough to result in a substantial oversimulation of the extent of sea ice in the region.

  • The GCMs project a twenty-first-century warming in the Arctic that is largest in the cold seasons, although the rates of the warming vary considerably among the models and across greenhouse gas forcing scenarios. The across-model and across-scenario uncertainties in the projected temperatures are comparable through the first half of the twenty-first century, but by the end of the twenty-first century, the cross-scenario variability is about 50% greater than the across-model variability.

  • All 14 GCMs project decreases of sea level pressure for the twenty-first century over essentially the entire Arctic. The most significant decreases of pressure are projected for the Bering Strait region, primarily in autumn and winter.

Individual modeling centers can use the temperature and SLP bias summaries here and the individual detailed biases in the Web archive to isolate specific model difficulties and potential enhancements. One issue common to almost all models, however, is the weaker and shortened storm track into the Barents Sea. The SLP errors in the Barents Sea are evident in both the model mean and variability biases. If this single model bias were rectified, the simulation improvements in SLP would likely spill over to significant improvements in other simulation characteristics. The extension of the simulated storm track would trigger warmer advection and divergent wind forcing events in the Barents Sea region. The resulting surface forcing conditions would rectify the oversimulation of sea ice in the region, further improving the surface temperature simulations. It is not clear from this analysis whether the deficient SLP fields are caused by dynamical issues or by misplaced thermodynamic gradients such as ocean surface boundary forcing. Given the unique topography of the Greenland ice sheet, however, a good first attempt at rectifying the Arctic biases might include experiments with different topographical representations of Greenland in the models.

Finally, while the internal and across-model variabilities are large in the Arctic relative to the rest of the globe, the associated signal-to-noise ratios indicate that the climate change signal is large enough in some regions to be verifiable over the next several decades. According to these simulations, a surface air temperature warming signal should be detected first in the central Arctic Ocean and the Barents Sea during the cold season and the near the climatological ice edge during the summer months. Climate change signals in the form of SLP decreases should be first detected in the Bering Sea/northeast Siberia region. Toward the middle of the twenty-first century, the coastal impacts on the Bering Strait region due to the projected northward shift in the storm track will be significant and the implications for patterns of precipitation change and modified surface wind stress on ocean and ice warrant a detailed investigation.

Acknowledgments

This work was supported by the National Science Foundation through Grants ATM-0332081 and ARC 0520112 to the University of Illinois and Grant OPP-0327664 to the International Arctic Research Center, University of Alaska.

REFERENCES

  • Arzel, O., T. Fichefet, and H. Goosse, 2005: Sea ice evolution over the 20th and 21st centuries as simulated by the current AOGCMs. Ocean Modell., 12 , 401415.

    • Search Google Scholar
    • Export Citation
  • Dickson, R. R., and Coauthors, 2000: The Arctic Ocean response to the North Atlantic Oscillation. J. Climate, 13 , 26712696.

  • Hassol, S. J., 2004: Impacts of a Warming Arctic: Arctic Climate Impact Assessment. Cambridge University Press, 139 pp.

  • Houghton, J. T., Y. Ding, D. J. Griggs, M. Noguer, P. J. van der Linden, X. Dai, K. Maskell, and C. A. Johnson, 2001: Climate Change 2001: The Scientific Basis. Cambridge University Press, 881 pp.

    • Search Google Scholar
    • Export Citation
  • Karl, T., 1998: Regional trends and variations of temperature and precipitation. The Regional Impacts of Climate Change: An Assessment of Vulnerability, R. T. Watson et al., Eds., Cambridge University Press, 412–425.

    • Search Google Scholar
    • Export Citation
  • Kattsov, V., and E. Källen, 2005: Future changes of climate: Modelling and scenarios for the Arctic region. Arctic Climate Impact Assessment (ACIA), Cambridge University Press, 1042 pp.

    • Search Google Scholar
    • Export Citation
  • Kattsov, V., J. E. Walsh, W. L. Chapman, V. A. Govorkova, T. V. Pavlova, and X. Zhang, 2007: Simulation and projection of Arctic freshwater budget components by the IPCC AR4 global climate models. J. Hydrometeor., in press.

    • Search Google Scholar
    • Export Citation
  • Nakićenović, N., and Coeditors, 2000: IPCC Special Report on Emission Scenarios. Cambridge University Press, 599 pp.

  • Polyakov, I., and Coauthors, 2002a: Trends and variations in Arctic climate system. Eos, Trans. Amer. Geophys. Union, 83 , 47. 547548.

    • Search Google Scholar
    • Export Citation
  • Polyakov, I., and Coauthors, 2002b: Observationally based assessment of polar amplification of global warming. Geophys. Res. Lett., 29 , 18781881.

    • Search Google Scholar
    • Export Citation
  • Serreze, M. C., and J. A. Francis, 2006: The Arctic amplification debate. Climatic Change, 76 , 241264.

  • Stroeve, J. C., M. C. Serreze, F. Fetterer, T. Arbetter, W. Meier, J. Maslanik, and K. Knowles, 2005: Tracking the Arctic’s shrinking ice cover: Another extreme September minimum in 2004. Geophys. Res. Lett., 32 .L04501, doi:10.1029/2004GL021810.

    • Search Google Scholar
    • Export Citation
  • Walsh, J. E., W. L. Chapman, and T. L. Shy, 1996: Recent decrease in sea level pressure in the central Arctic. J. Climate, 9 , 480486.

    • Search Google Scholar
    • Export Citation
  • Walsh, J. E., V. M. Kattsov, W. L. Chapman, V. Govorkova, and T. Pavlova, 2002: Comparison of Arctic climate simulations by uncoupled and coupled global models. J. Climate, 15 , 14291446.

    • Search Google Scholar
    • Export Citation
  • Walsh, J. E., S. Vavrus, and W. L. Chapman, 2005: Workshop on modeling of the Arctic atmosphere. Bull. Amer. Meteor. Soc., 86 , 845852.

    • Search Google Scholar
    • Export Citation
  • Wang, M., J. E. Overland, V. Kattsov, J. E. Walsh, X. Zhang, and T. Pavlova, 2007: Intrinsic versus forced variation in coupled climate model simulations over the Arctic during the twentieth century. J. Climate, in press.

    • Search Google Scholar
    • Export Citation

Fig. 1.
Fig. 1.

(top row) Observed and (middle row) simulated seasonal surface air temperatures averaged over the period 1981–2000. (bottom row) Model-observed differences (biases) are also shown.

Citation: Journal of Climate 20, 4; 10.1175/JCLI4026.1

Fig. 2.
Fig. 2.

Model-observed annual mean surface air temperature biases for 14 GCMs of the IPCC AR4 and (top left) the 14-model composite bias.

Citation: Journal of Climate 20, 4; 10.1175/JCLI4026.1

Fig. 3.
Fig. 3.

Change in magnitude of temperature biases from the IPCC TAR to the AR4. Smaller biases (AR4–TAR) shown in green and blue represent improvements, yellows and reds indicate a decrease in observational correspondence of the five GCMs.

Citation: Journal of Climate 20, 4; 10.1175/JCLI4026.1

Fig. 4.
Fig. 4.

Annual, winter (DJF), spring (MAM), summer (JJA), and autumn (SON) area-averaged model-observation RMS temperature errors for ocean areas 70°–90°N.

Citation: Journal of Climate 20, 4; 10.1175/JCLI4026.1

Fig. 5.
Fig. 5.

Annual and seasonal standard deviations of surface air temperature for the period 1981–2000 are shown for (left column) observed and (second column) 14-GCM composites. (third column) The corresponding annual and seasonal across-model scatter and (right column) internal model variability from the preindustrial control simulations are also plotted.

Citation: Journal of Climate 20, 4; 10.1175/JCLI4026.1

Fig. 6.
Fig. 6.

(top row) Observed and (middle row) simulated seasonal SLPs are averaged over the period 1981–2000. (bottom row) Model-observed differences (biases) are shown.

Citation: Journal of Climate 20, 4; 10.1175/JCLI4026.1

Fig. 7.
Fig. 7.

Same as in Fig. 2, but for SLP biases.

Citation: Journal of Climate 20, 4; 10.1175/JCLI4026.1

Fig. 8.
Fig. 8.

Change in magnitude of SLP biases from the IPCC TAR to the AR4. Smaller biases (AR4–TAR) shown in green and blue represent improvements, yellows and reds indicate a decrease in observational correspondence of the five GCMs.

Citation: Journal of Climate 20, 4; 10.1175/JCLI4026.1

Fig. 9.
Fig. 9.

Same as in Fig. 4, but for SLP errors.

Citation: Journal of Climate 20, 4; 10.1175/JCLI4026.1

Fig. 10.
Fig. 10.

Same as in Fig. 5, but for SLP.

Citation: Journal of Climate 20, 4; 10.1175/JCLI4026.1

Fig. 11.
Fig. 11.

Simulated and projected annual mean surface air temperature, expressed as departures from 1981–2000 means, by 14 global climate models for the twentieth- and twenty-first centuries. Projections use three greenhouse gas forcing scenarios: IPCC SRESB1 (blue), IPCC SRESA1B (green), and IPCC SRESA2 (red).

Citation: Journal of Climate 20, 4; 10.1175/JCLI4026.1

Fig. 12.
Fig. 12.

Across-model variability, expressed as 11-yr running standard deviations of annual surface air temperature for three greenhouse gas forcing scenarios: IPCC SRESB1 (blue), IPCC SRESA1B (green), and IPCC SRESA2 (red). Also plotted is the across-scenario variability (black), expressed as the 11-yr running standard deviation of surface air temperatures across the three IPCC scenarios.

Citation: Journal of Climate 20, 4; 10.1175/JCLI4026.1

Fig. 13.
Fig. 13.

Projected surface air temperature change (from 1981–2000 means) for the 2070–89 time slice for 14 GCMs and the 14-GCM composite.

Citation: Journal of Climate 20, 4; 10.1175/JCLI4026.1

Fig. 14.
Fig. 14.

The 14-GCM composite ratios of annual and seasonal projected temperature change for three time slices: (left) 2010–29, (middle) 2040–59, and (right) 2070–89 with (a) corresponding annual and seasonal standard deviations of 15–50 20-yr means from the control simulations, and (b) the across-model standard deviations for the corresponding time slices.

Citation: Journal of Climate 20, 4; 10.1175/JCLI4026.1

Fig. 15.
Fig. 15.

Projected SLP change (from 1981–2000 means) for the 2070–89 time slice for 14 GCMs and the 14-GCM composite.

Citation: Journal of Climate 20, 4; 10.1175/JCLI4026.1

Fig. 16.
Fig. 16.

The 14-GCM composite ratios of annual and seasonal projected SLP for three time slices: (left) 2010–29, (middle) 2040–59, and (right) 2070–89 with (a) corresponding annual and seasonal standard deviations of 15–50 20-yr means from the control simulations, and (b) the across-model standard deviations for the corresponding time slices.

Citation: Journal of Climate 20, 4; 10.1175/JCLI4026.1

Table 1.

IPCC AR4 models assessed in this study. (Detailed model documentation is available online at http://www-pcmdi.llnl.gov/ipcc/model_documentation/ipcc_model_documentation.php)

Table 1.
Save
  • Arzel, O., T. Fichefet, and H. Goosse, 2005: Sea ice evolution over the 20th and 21st centuries as simulated by the current AOGCMs. Ocean Modell., 12 , 401415.

    • Search Google Scholar
    • Export Citation
  • Dickson, R. R., and Coauthors, 2000: The Arctic Ocean response to the North Atlantic Oscillation. J. Climate, 13 , 26712696.

  • Hassol, S. J., 2004: Impacts of a Warming Arctic: Arctic Climate Impact Assessment. Cambridge University Press, 139 pp.

  • Houghton, J. T., Y. Ding, D. J. Griggs, M. Noguer, P. J. van der Linden, X. Dai, K. Maskell, and C. A. Johnson, 2001: Climate Change 2001: The Scientific Basis. Cambridge University Press, 881 pp.

    • Search Google Scholar
    • Export Citation
  • Karl, T., 1998: Regional trends and variations of temperature and precipitation. The Regional Impacts of Climate Change: An Assessment of Vulnerability, R. T. Watson et al., Eds., Cambridge University Press, 412–425.

    • Search Google Scholar
    • Export Citation
  • Kattsov, V., and E. Källen, 2005: Future changes of climate: Modelling and scenarios for the Arctic region. Arctic Climate Impact Assessment (ACIA), Cambridge University Press, 1042 pp.

    • Search Google Scholar
    • Export Citation
  • Kattsov, V., J. E. Walsh, W. L. Chapman, V. A. Govorkova, T. V. Pavlova, and X. Zhang, 2007: Simulation and projection of Arctic freshwater budget components by the IPCC AR4 global climate models. J. Hydrometeor., in press.

    • Search Google Scholar
    • Export Citation
  • Nakićenović, N., and Coeditors, 2000: IPCC Special Report on Emission Scenarios. Cambridge University Press, 599 pp.

  • Polyakov, I., and Coauthors, 2002a: Trends and variations in Arctic climate system. Eos, Trans. Amer. Geophys. Union, 83 , 47. 547548.

    • Search Google Scholar
    • Export Citation
  • Polyakov, I., and Coauthors, 2002b: Observationally based assessment of polar amplification of global warming. Geophys. Res. Lett., 29 , 18781881.

    • Search Google Scholar
    • Export Citation
  • Serreze, M. C., and J. A. Francis, 2006: The Arctic amplification debate. Climatic Change, 76 , 241264.

  • Stroeve, J. C., M. C. Serreze, F. Fetterer, T. Arbetter, W. Meier, J. Maslanik, and K. Knowles, 2005: Tracking the Arctic’s shrinking ice cover: Another extreme September minimum in 2004. Geophys. Res. Lett., 32 .L04501, doi:10.1029/2004GL021810.

    • Search Google Scholar
    • Export Citation
  • Walsh, J. E., W. L. Chapman, and T. L. Shy, 1996: Recent decrease in sea level pressure in the central Arctic. J. Climate, 9 , 480486.

    • Search Google Scholar
    • Export Citation
  • Walsh, J. E., V. M. Kattsov, W. L. Chapman, V. Govorkova, and T. Pavlova, 2002: Comparison of Arctic climate simulations by uncoupled and coupled global models. J. Climate, 15 , 14291446.

    • Search Google Scholar
    • Export Citation
  • Walsh, J. E., S. Vavrus, and W. L. Chapman, 2005: Workshop on modeling of the Arctic atmosphere. Bull. Amer. Meteor. Soc., 86 , 845852.

    • Search Google Scholar
    • Export Citation
  • Wang, M., J. E. Overland, V. Kattsov, J. E. Walsh, X. Zhang, and T. Pavlova, 2007: Intrinsic versus forced variation in coupled climate model simulations over the Arctic during the twentieth century. J. Climate, in press.

    • Search Google Scholar
    • Export Citation
  • Fig. 1.

    (top row) Observed and (middle row) simulated seasonal surface air temperatures averaged over the period 1981–2000. (bottom row) Model-observed differences (biases) are also shown.

  • Fig. 2.

    Model-observed annual mean surface air temperature biases for 14 GCMs of the IPCC AR4 and (top left) the 14-model composite bias.

  • Fig. 3.

    Change in magnitude of temperature biases from the IPCC TAR to the AR4. Smaller biases (AR4–TAR) shown in green and blue represent improvements, yellows and reds indicate a decrease in observational correspondence of the five GCMs.

  • Fig. 4.

    Annual, winter (DJF), spring (MAM), summer (JJA), and autumn (SON) area-averaged model-observation RMS temperature errors for ocean areas 70°–90°N.

  • Fig. 5.

    Annual and seasonal standard deviations of surface air temperature for the period 1981–2000 are shown for (left column) observed and (second column) 14-GCM composites. (third column) The corresponding annual and seasonal across-model scatter and (right column) internal model variability from the preindustrial control simulations are also plotted.

  • Fig. 6.

    (top row) Observed and (middle row) simulated seasonal SLPs are averaged over the period 1981–2000. (bottom row) Model-observed differences (biases) are shown.

  • Fig. 7.

    Same as in Fig. 2, but for SLP biases.

  • Fig. 8.

    Change in magnitude of SLP biases from the IPCC TAR to the AR4. Smaller biases (AR4–TAR) shown in green and blue represent improvements, yellows and reds indicate a decrease in observational correspondence of the five GCMs.

  • Fig. 9.

    Same as in Fig. 4, but for SLP errors.

  • Fig. 10.

    Same as in Fig. 5, but for SLP.

  • Fig. 11.

    Simulated and projected annual mean surface air temperature, expressed as departures from 1981–2000 means, by 14 global climate models for the twentieth- and twenty-first centuries. Projections use three greenhouse gas forcing scenarios: IPCC SRESB1 (blue), IPCC SRESA1B (green), and IPCC SRESA2 (red).

  • Fig. 12.

    Across-model variability, expressed as 11-yr running standard deviations of annual surface air temperature for three greenhouse gas forcing scenarios: IPCC SRESB1 (blue), IPCC SRESA1B (green), and IPCC SRESA2 (red). Also plotted is the across-scenario variability (black), expressed as the 11-yr running standard deviation of surface air temperatures across the three IPCC scenarios.

  • Fig. 13.

    Projected surface air temperature change (from 1981–2000 means) for the 2070–89 time slice for 14 GCMs and the 14-GCM composite.

  • Fig. 14.

    The 14-GCM composite ratios of annual and seasonal projected temperature change for three time slices: (left) 2010–29, (middle) 2040–59, and (right) 2070–89 with (a) corresponding annual and seasonal standard deviations of 15–50 20-yr means from the control simulations, and (b) the across-model standard deviations for the corresponding time slices.

  • Fig. 15.

    Projected SLP change (from 1981–2000 means) for the 2070–89 time slice for 14 GCMs and the 14-GCM composite.

  • Fig. 16.

    The 14-GCM composite ratios of annual and seasonal projected SLP for three time slices: (left) 2010–29, (middle) 2040–59, and (right) 2070–89 with (a) corresponding annual and seasonal standard deviations of 15–50 20-yr means from the control simulations, and (b) the across-model standard deviations for the corresponding time slices.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 1190 437 102
PDF Downloads 630 179 23