In this study, the southern African climate response to increasing amounts of greenhouse gases is investigated, based on the dataset of a 150-yr climate change experiment following the IPCC Special Report on Emissions Scenarios marker scenario B2 (SRES-B2) performed with the coupled ARPEGE/OPA/GELATO general circulation model (GCM). The method of canonical correlation analysis (CCA) is adopted to validate the ability of the GCM to simulate the present-day climate over the southern African region and project the late-summer rainfall change over South Africa at the end of the 21st century. The model validation shows that the ARPEGE/OPA/GELATO GCM is able to capture the observed link between rainfall over South Africa and adjacent sea-level pressure (SLP), despite the existence of some systematic errors. The structure and variability of SLP are reproduced by the GCM in a realistic way. The major controlling mechanism of rainfall over South Africa can be identified in the GCM. The projection of rainfall indicates a drying trend during the 21st century over most parts of South Africa, in particular the central interior. Compared to present-day climatology, the overall late-summer rainfall will decrease by 8.2% by the end of 21st century as derived from GCM grid-point output, and by 16.1% from the downscaling model.
KEY WORDS: Climate change · Model evaluation · Statistical downscaling · Canonical correlation analysis · South AfricaResale or republication not permitted without written consent of the publisher Clim Res 28: 109-122, 2005 ever, there is evidence of increased interannual variability in recent decades over the region (Richard et al. 2001), which fuels further concern about that evolution.General circulation models (GCMs) are the most fundamental tool for studying climate change. Whereas GCMs do a creditable job on the large scale, they have much lower skill at the regional scale. However, it is at these higher resolution scales that climate change information is most needed. A possible approach to bridging the scale gap is downscaling, which uses dynamical or statistical models to relate largescale information from GCMs to regional parameters (Karl et al. 1990, Giorgi & Mearns 1991, Joubert et al. 1999, Zorita & Von Storch 1999. Considering that the application of regional climate models is still in its infancy and computationally very demanding, statistical downscaling is a practical means to address the immediate needs of the southern African region.Applying statistical downscaling methods to climate change analysis is becoming quite popular (Von Storch et al. 1993, Cui et al. 1995, Busuioc et al. 2001). However, downscaling has received relatively little attention in Africa. and Hewitson & Joubert (1998, available at www.egs.uct.ac.za/fccc/) have applied the statistical downscaling method of artificial neural networks (ANNs) to generate climate scenarios for South Africa. In their study, the climate scenario used is 2 × CO 2 which is an idealized situation. With new scenario experiments av...