11.2.2 Skill of Models in Simulating Present and Past Climates

There are biases in the simulations of African climate that are systematic across the MMD models, with 90% of models overestimating precipitation in southern Africa, by more than 20% on average (and in some cases by as much as 80%) over a wide area often extending into equatorial Africa. The temperature biases over land are not considered large enough to directly affect the credibility of the model projections (see Supplementary Material Figure S11.21 and Table S11.1).

The ITCZ in the Atlantic is displaced equatorward in nearly all of these AOGCM simulations. Ocean temperatures are too warm by an average of 1°C to 2°C in the Gulf of Guinea and typically by 3°C off the southwest coast in the region of intense upwelling, which is clearly too weak in many models. In several of the models there is no West African monsoon as the summer rains fail to move from the Gulf onto land, but most of the models do have a monsoonal climate albeit with some distortion. Moderately realistic interannual variability of SSTs in the Gulf of Guinea and the associated dipolar rainfall variations in the Sahel and the Guinean Coast are, by the criteria of Cook and Vizy (2006), only present in 4 of the 18 models examined. Tennant (2003) describes biases in several AGCMs, such as the equatorward displacement of the mid-latitude jet in austral summer, a deficiency that persists in the most recent simulations (Chapter 8).

Despite these deficiencies, AGCMs can simulate the basic pattern of rainfall trends in the second half of the 20th century if given the observed SST evolution as boundary conditions, as described in the multi-model analysis of Hoerling et al. (2006) and the growing literature on the interannual variability and trends in individual models (e.g., Rowell et al., 1995; Bader and Latif, 2003; Giannini et al., 2003; Haarsma et al., 2005; Kamga et al., 2005; Lu and Delworth, 2005). However, there is less confidence in the ability of AOGCMs to generate interannual variability in the SSTs of the type known to affect African rainfall, as evidenced by the fact that very few AOGCMs produce droughts comparable in magnitude to the Sahel drought of the 1970s and 1980s (Hoerling et al., 2006). There are exceptions, but what distinguishes these from the bulk of the models is not understood.

The very wet Sahara 6 to 8 ka is thought to have been a response to the increased summer insolation due to changes in the Earth’s orbital configuration. Modelling studies of this response provide background information on the quality of a model’s African monsoon, but the processes controlling the response to changing seasonal insolation may be different from those controlling the response to increasing greenhouse gases. The fact that GCMs have difficulty in simulating the full magnitude of the mid-Holocene wet period, especially in the absence of vegetation feedbacks, may indicate a lack of sensitivity to other kinds of forcing (Jolly et al., 1996; Kutzbach et al., 1996).

Regional climate modelling has mostly focused on southern Africa, where the models generally improve on the climate simulated by global models but also share some of the biases in the global models. For example, Engelbrecht et al. (2002) and Arnell et al. (2003) both simulate excessive rainfall in parts of southern Africa, reminiscent of the bias in the MMD. Hewitson et al. (2004) and Tadross et al. (2006) note strong sensitivity to the choice of convective parametrization, and to changes in soil moisture and vegetative cover (New et al., 2003; Tadross et al., 2005a), reinforcing the view (Rowell et al., 1995) that land surface feedbacks enhance regional climate sensitivity over Africa’s semi-arid regions. Over West Africa, the number of Regional Climate Model (RCM) investigations is even more limited (Jenkins et al., 2002; Vizy and Cook, 2002). The quality of the 25-year simulation undertaken by Paeth et al. (2005) is encouraging, emphasizing the role of regional SSTs and changes in the land surface in forcing West African rainfall anomalies. Several recent AGCM time-slice simulations focusing on tropical Africa show good simulation of the rainy season (Coppola and Giorgi, 2005; Caminade et al., 2006; Oouchi et al., 2006).

Hewitson and Crane (2005) developed empirical downscaling for point-scale precipitation at sites spanning the continent, as well as a 0.1° resolution grid over South Africa. The downscaled precipitation forced by reanalysis data provides a close match to the historical climate record, including regions such as the eastern escarpment of the sub-continent that have proven difficult for RCMs.