GCMs are in general agreement that the effects of anthropogenic greenhouse warming will be first seen and will be largest in the polar regions, in large part due to feedbacks involving sea ice and snow cover, and the strong stability of the lower troposphere. There is a growing body of evidence pointing to significant change in the Arctic climate system over at least the past several decades. The emerging view is that at least part of the observed changes can be linked to human activities. It comes as no surprise that the Arctic community has taken great interest in global climate model simulations.
Attribution of change requires credible model simulations that quantify the natural variability of the Arctic climate system and the response of the system to changing greenhouse gas and sulfate aerosol concentrations. There are large differences between current-day GCMs in their projections of how much warming will occur, as well as the spatial patterns of change in temperature and other variables. Important in this regard is that existing models are of varying complexity. For example, some include coupling to a deep ocean, while others incorporate only a mixed layer. In turn, many models suffer from known deficiencies in their parameterizations of Arctic processes, such as interactions and feedbacks between the sea ice, atmosphere and ocean, and clouds.
Climate change projections from GCMs will be examined in Chapter 11. To lay some groundwork, we address here the ability of GCMs to simulate present climate, in part making use of results from the Atmospheric Model Intercomparison Project (AMIP). AMIP is intended to determine systematic errors in simulated present climate from uncoupled (atmosphere only) GCM simulations. Gates (1992) provides an overview of the AMIP-I effort. AMIP-I focused on climate simulations over the period 1979-88 with a series of different modes, using identical observed monthly averaged distributions of sea surface temperature and sea ice as boundary conditions for each model. Standard carbon dioxide concentrations as well as the solar constant were also specified. There were no common specifications of the land surface or of surface elevations. The decade 1979-88 was chosen to have a long simulation period for which observations are plentiful. More than 30 of the world's principal modeling groups participated in AMIP-I. This has been followed by AMIP-II, which includes more recent versions of models and improved specified lower boundary conditions, especially for sea ice. The AMIP-II simulations span the period 1979-96.
Walsh et al. (2002) provide a valuable focus on simulations of present Arctic climate based on thirteen selected AMIP-II models as well as simulations from eight different coupled models (AOGCMs) spanning the period 1961-90 (the control period for the Intergovernmental Panel on Climate Change (IPCC) simulations). These models can be considered as the late 1990s state-of-the-art. Neither sea ice nor SST are prescribed in the AOGCMs. The AMIP-II and coupled models differ greatly in their horizontal resolution and number of vertical layers. The Walsh et al. (2002) effort emphasized simulations of sea level pressure, temperature and precipitation but also included some assessment of variables such as sea ice cover, cloudiness and solar radiation at the surface.
It is obviously important to simulate the sea level pressure distribution, which is closely linked to patterns of precipitation and temperature and is the primary determinant of the wind stress and surface fluxes of heat and moisture. Based on composite means (i.e., the average of the different models) over the period of the AMIP-II and coupled model simulations, recent uncoupled and coupled models capture (with respect to NCEP/NCAR reanalysis data) the dominant observed annual mean pressure features over the Northern Hemisphere - the Icelandic and Aleutian lows, the Siberian High and weaker high pressure over the Arctic Ocean. However, as evident in difference (bias) fields with respect to the NCEP/NCAR means (model minus observed) (Figure 9.9) the model's pressures tend to be too high over the Arctic Ocean, most evident over the Eurasian side.
The surface wind field over the Arctic Ocean is hence quite different from that observed, which has serious implications with respect to the implied sea ice drift and «-
Figure 9.9 Biases (model minus observed, in hPa) of (top) the AMIP-II composite annual mean sea level pressure, and (bottom) the IPCC composite annual mean sea level pressure. Dashed contours indicate negative biases while positive biases are shown by hatching and cross-hatching (courtesy of W. Chapman, University of Illinois, Urbana-Champaign, IL, based on Walsh et al., 2002).
hence the thickness of ice comprising the Fram Strait outflow. Another conclusion from Figure 9.9 is that the bias is not attributable to coupling with the ocean. The spatial pattern of the pressure bias is essentially the same for the uncoupled and coupled models, although the biases are larger by 2-3 hPa in the coupled models. Walsh et al. (2002) suggest a potential role of the representation of topography as it influences the large-scale atmospheric circulation. In this regard, the larger biases in the coupled models are consistent with the generally coarser resolution of these models. The range between the different models in their depiction of the SLP field is greater over the Arctic Ocean than anywhere in the Northern Hemisphere. For the coupled models, across-model standard deviations exceed 6 hPa in the vicinity of the Beaufort Sea.
The AMIP-II models use energy budget computations to obtain SAT over sea ice. In most models, the energy budget computations include conduction of heat through ice of a prescribed thickness (generally 1-3 m). The surface energy budget calculation also determines the SAT over land. As a result, temperatures can differ between the different models over sea ice and land, but very little over ice-free regions, where SSTs are prescribed. Recall that neither sea ice cover nor SSTs are prescribed in the coupled models.
Figure 9.10 shows the composite model mean biases in annual mean SAT with respect to NCEP/NCAR data. Following the above discussion, biases in the AMIP-II models are small over open ocean areas. However, over the ice-covered Arctic Ocean, the model temperatures are on average 2-3° C higher than in the NCEP/NCAR reanalysis. Arctic Ocean biases in the coupled models are smaller, but on average these models are too cold (1-3 ° C) over much of northern Asia, and are too warm over the region from northwestern Greenland to Svalbard to Franz Josef land. The bias appears to be associated with having too little winter sea ice in this region. The across-model standard deviations in SAT are especially large between the coupled model simulations over the Atlantic sector (>5° C), which is again attributed to differences in winter sea ice cover. In support, the marginal ice zone corresponds closely to where the standard deviation in temperature is largest, with the warmest model in this area being that with the smallest wintertime ice extents. These conclusions must also be considered in light of uncertainties in the quality of the NCEP/NCAR temperature fields.
Corresponding biases in mean annual precipitation appear in Figure 9.11. Following from Chapter 6, evaluation of model performance is made difficult due to large uncertainties in observed precipitation, due to both gauge biases and the sparse observational network, especially over the Arctic Ocean. The evaluations in Figure 9.11 make use of a climatology developed by Russian investigators (Bryazgin, 1976; Khrol, 1996). The AMIP-II and coupled models show the same general tendency to overestimate precipitation over most of the Arctic. Both model types strongly overestimate «-
Figure 9.10 Biases (modelminus observed in ° C) of (top) the AMIP-II composite annual mean SAT and (bottom) the IPCC composite annual mean SAT. Dashed contours indicate negative biases while positive biases are shown by hatching and cross-hatching (courtesy of W. Chapman, University of Illinois, Urbana-Champaign, IL, based on Walsh et al., 2002).
60 - ■ — — — — — — — — — — — — - —
^ 50 - ■ — — — — — — — — — — — — - —
40-- — — — — — — — — — — — — - —
30 — — — — — — — — — — — — - —
20 — — — — — — — — — — — — - —
10-- — — — — — — — — — — — — — — 0 I III II
CCC CCS CNR DNM ECM J MA MGO MRI NCA NCE PNN UGA UKM tovs h95
Figure 9.12 Mean annual cycle (top panel) of monthly mean cloud cover (%) for the region north of 70° N from AMIP-II uncoupled models (thin solid lines; thick solid line is the 13-model mean) and observational estimates (dashed line from TOVS satellite data, dotted line from the surface-based observations of Hahn et al. ). The bottom shows corresponding annual mean values (from Walsh etal., 2002, bypermissionof AMS).
precipitation over Alaska and around Greenland. The only area with substantial underestimation is in the Norwegian and Barents seas region. As argued by Walsh et al. (2002), the combination of having too little precipitation in the Norwegian region and too much near southeastern Greenland implies that the North Atlantic storm track is either too weak or is placed too far west in the models. As examined seasonally, the AMIP-II models tend to produce too much cold season precipitation (November-May). For the coupled models, overestimation is greatest in winter and autumn. Most of the models (both coupled and uncoupled) tend to show peak precipitation during late summer to early autumn, roughly consistent with observations.
As a final example, we look at the model simulations of cloud cover. Poor description of cloud cover (fractional coverage and optical properties) will have first-order impacts on solar radiation receipts at the surface, with cascading effects on other components of the surface energy balance. Results for the AMIP-II uncoupled models are summarized in Figure 9.12. These are averages over the region north of 70° N. The two observational
Figure 9.11 Biases (model minus observed, in mm day-1) of (top) the AMIP-II composite annual mean precipitation, and (bottom) the IPCC composite annual mean precipitation. Dashed contours indicate negative biases while positive biases are shown by hatching and cross-hatching (courtesy of W. Chapman, University of Illinois, Urbana-Champaign, IL, based on Walsh et al., 2002).
estimates for model evaluation are based on TOVS satellite data (Schweiger et al., 1999) and surface-based observations compiled by Hahn et al. (1995). Looking first at the mean annual cycle, it is obvious that there are considerable differences even between the two observational estimates. In part, this is explained by the tendency for satellite algorithms to underestimate low-level stratus and fog. Both estimates, however, point to a general cold season minimum and warm season maximum. The annual cycle from the composite model mean seems to have about the right shape. However, monthly values are higher than observed during the cold season, with too little cloud cover from May through July, suggesting excessive downwelling solar radiation at this time. There is large scatter between the models, although performance is better than for the earlier AMIP-I runs. Large scatter is also seen for annual mean values. Not surprisingly, surface radiative fluxes differ greatly between the models, especially under cloudy skies. The implication is that the surface energy budget is likely to be problematic not just for simulations of present-day climate, but for climate change simulations (Walsh et al., 2002).
Was this article helpful?