One of the principal challenges within the context of global climate change is that of developing plausible regional-scale projections of change, for both the impacts community, and for policy development. Two approaches are currently prevalent: empirical downscaling, and regional climate models (RCMs). However, in both cases the forcing is derived from GCM simulation results, and in each case assumes viability of the GCM forcing. Evaluation of the GCM fields is thus essential in order to characterize the uncertainty associated with any regionalized climate change projection.

The characteristics of three GCM simulations are compared in terms of the boundary forcing variables used to drive a RCM; the HadAM3, CSIRO Mk2, and ECHAM4. Using Self Organizing Maps (SOMs), the models are compared firstly with NCEP ranalysis fields, and secondly with one another. The frequency distribution of the atmospheric variables forming the boundary fields of the RCM domain over sub-equatorial Africa are initialy compared. The biases and differences between models and the NCEP reanalysis are then discussed in context of the southern Africa climate system, and the implications for climate change and the downscaling of regional climate projections.