11.6.2 Skill of Models in Simulating Present Climate

In the Central America (CAM) and AMZ regions, most models in the MMD have a cold bias of 0°C to 3°C, except in AMZ in SON (Supplementary Material Table S11.1). In southern South America (SSA) average biases are close to zero. The biases are unevenly geographically distributed (Supplementary Material Figure S11.25). The MMD mean climate shows a warm bias around 30°S (particularly in summer) and in parts of central South America (especially in SON). Over the rest of South America (central and northern Andes, eastern Brazil, Patagonia) the biases tend to be predominantly negative. The SST biases along the western coasts of South America are likely related to weakness in oceanic upwelling.

For the CAM region, the multi-model scatter in precipitation is substantial, but half of the models lie in the range of –15 to 25% in the annual mean. The largest biases occur during the boreal winter and spring seasons, when precipitation is meagre (Supplementary Material Table S11.1). For both AMZ and SSA, the ensemble annual mean climate exhibits drier than observed conditions, with about 60% of the models having a negative bias. Unfortunately, this choice of regions for averaging is particularly misleading for South America since it does not clearly bring out critical regional biases such as those related to rainfall underestimation in the Amazon and La Plata Basins (Supplementary Material Figure S11.26). Simulation of the regional climate is seriously affected by model deficiencies at low latitudes. In particular, the MMD ensemble tends to depict a relatively weak ITCZ, which extends southward of its observed position. The simulations have a systematic bias towards underestimated rainfall over the Amazon Basin. The simulated subtropical climate is typically also adversely affected by a dry bias over most of south-eastern South America and in the South Atlantic Convergence Zone, especially during the rainy season. In contrast, rainfall along the Andes and in northeast Brazil is excessive in the ensemble mean.

Some aspects of the simulation of tropical climate with AOGCMs have improved. However, in general, the largest errors are found where the annual cycle is weakest, such as over tropical South America (see, e.g., Section 8.3). Atmospheric GCMs approximate the spatial distribution of precipitation over the tropical Americas, but they do not correctly reproduce the temporal evolution of the annual cycle in precipitation, specifically the mid-summer drought (Magaña and Caetano, 2005). Tropical cyclones are important contributors to precipitation in the region. If close to the continent, they will produce large amounts of precipitation over land, and if far from the coast, moisture divergence over the continental region enhances drier conditions.

Zhou and Lau (2002) analyse the precipitation and circulation biases in a set of six AGCMs provided by the Climate Variability and Predictability Programme (CLIVAR) Asian-Australian Monsoon AGCM Intercomparison Project (Kang et al., 2002). This model ensemble captures some large-scale features of the South American monsoon system reasonably well, including the seasonal migration of monsoon rainfall and the rainfall associated with the South America Convergence Zone. However, the South Atlantic subtropical high and the Amazonia low are too strong, whereas low-level flow tends to be too strong during austral summer and too weak during austral winter. The model ensemble captures the Pacific-South American pattern quite well, but its amplitude is generally underestimated.

Regional models are still being tested and developed for this region. Relatively few studies using RCMs for Central and South America exist, and those that do are constrained by short simulation length. Some studies (Chou et al., 2000; Nobre et al., 2001; Druyan et al., 2002) examine the skill of experimental dynamic downscaling of seasonal predictions over Brazil. Results suggest that both more realistic GCM forcing and improvements in the RCMs are needed. Seth and Rojas (2003) performed seasonal integrations driven by reanalyses, with emphasis on tropical South America. The model was able to simulate the different rainfall anomalies and large-scale circulations but, as a result of weak low-level moisture transport from the Atlantic, rainfall over the western Amazon was underestimated. Vernekar et al. (2003) follow a similar approach to study the low-level jets and report that the RCM produces better regional circulation details than does the reanalysis. However, an ensemble of four RCMs did not provide a noticeable improvement in precipitation over the driving large-scale reanalyses (Roads et al., 2003).

Other studies (Misra et al., 2003; Rojas and Seth, 2003) analyse seasonal RCM simulations driven by AGCM simulations. Relative to the AGCMs, regional models generally improve the rainfall simulation and the tropospheric circulation over both tropical and subtropical South America. However, AGCM-driven RCMs degrade compared with the reanalyses-driven integrations and they could even exacerbate the dry bias over sectors of AMZ and perpetuate the erroneous ITCZ over the neighbouring ocean basins from the AGCMs. Menéndez et al. (2001) used a RCM driven by a stretched-grid AGCM with higher resolution over the southern mid-latitudes to simulate the winter climatology of SSA. They find that both the AGCM and the regional model have similar systematic errors but the biases are reduced in the RCM. Analogously, other RCM simulations for SSA give too little precipitation over the subtropical plains and too much over elevated terrain (e.g., Nicolini et al., 2002; Menéndez et al., 2004).