Evaluation and inter-comparison of Global Climate Models’ performance over Katonga and Ruizi catchments in Lake Victoria basin
Regional impact assessments of climate change on hydrological extremes require robust examinations of climate model simulations. The climate models may satisfy mean statistics but fail to reproduce extreme quantiles which are crucial for applications of climate change impact analysis on water resources. Through statistical analysis, this paper evaluates and inter-compares the performance of Global Climate Model (GCM) simulations for their ability to predict changes in hydrological extremes for given locations or catchments in the Nile basin. Two catchments were considered: Katonga and Ruizi catchments in the Lake Victoria basin. Models that differ significantly from the observed extremes were considered unreliable for impact assessments on hydrological extremes. A graphical approach (rainfall quantile/frequency analysis), which allows for easy spotting of discordant models, in combination with several statistics, was used to evaluate 18 GCM control simulations against observed rainfall data. Standard deviation, coefficient of variation and root mean squared error (about the mean) of the observed rainfall, were used to derive error margins against which GCM simulations were evaluated. Model results outside the error margins were considered inconsistent with the observed rainfall. Model inter-comparison was also carried out for the rainfall change projections till the 2050s and 2090s through analysis of perturbations and percentage changes based on A1B, A2, and B1 SRES scenarios. It is noted that the GCM outputs are more consistent in reproducing rainfall signatures at annual aggregation level than at monthly aggregation levels with tendency of overestimation of the rainfall depths but with significant variation among different GCM simulations. The GCMs perform better in reproducing rainfall frequency with higher return periods compared with lower return periods. Most of the GCMs perform better for the wet months than the drier months. The GCMs CGCM3.2a, CM3.O, CM4.1, PCM1, CGCM3.1T47, MIROC3.2.HIRES, CCSM3.0 and FGOALS, are the most inconsistent with the observed rainfall for both catchments. Good performing models are MK3.5, MK3.0, ECHAM5, CM2.1U.H2 and CM2.0. In general, most GCMs perform poorly for both catchments. This signals the need for significant improvements in the rainfall modelling of the climate models for the study region. There is no strong evidence to suggest that GCM performance improves with higher spatial resolution. Models which are highly inconsistent with other models in reproducing the observed rainfall are not necessarily inconsistent with other models in the future projections. Differences in projections for the A1B, B2, and B1 scenarios were found to be smaller than the differences between the GCM simulations.