A MCDM-based framework for selection of general circulation models and projection of spatio-temporal rainfall changes: A case study of Nigeria

Abstract

Abstract A multi-criteria decision-making approach was used for the selection of GCMs for Nigeria based on their ability to replicate historical rainfall estimated using three entropy-based feature selection methods namely, Entropy Gain (EG), Gain Ratio (GR), and Symmetrical Uncertainty (SU). Performances of four bias correction methods were compared to identify the most suitable method for downscaling and projection of rainfall using the selected GCMs. Random forest (RF) regression was used for the generation of the multi-model ensemble (MME) average of projected rainfall. The ensemble projections for each of the representative concentration pathways (RCPs) 2.6, 4.5, 6.0, and 8.5 were computed and compared with global precipitation climatology centre (GPCC) historical rainfall of Nigeria to assess the percentage changes in annual rainfall with 95% level of confidence at different ecological zones for three future periods 2010–2039, 2040–2069, and 2070–2099. Quantile regression was used to assess the changes in seasonal rainfall at 95% confidence interval over the present century. The results revealed that MRI-CGCM3, HadGEM2-ES, CSIRO-Mk3-6-0 and CESM1-CAM5 are the most suitable GCMs for the projection of rainfall in Nigeria. The linear scaling method was found as the most suitable approach for downscaling of rainfall in terms of all the statistical indices used. It was found to downscale rainfall with normalized root mean square error (NRMSE) in the range of 30.7–44.0%, while Nash-Sutcliff efficiency (NSE) was between 0.81 and 0.91, and modified coefficient of agreement (md) was between 0.82 and 0.88. Projection of rainfall showed no significant change in Nigeria over the century under RCP 2.6, 4.5 and 6.5, while RCP 8.5 showed a decrease in the last part of the century (2070–2099). The seasonal changes in rainfall showed an increase in rainfall in the range of 0–20% in most parts of the north. The methodology in this study can reduce the uncertainty inherent in climate change projection and produce better projection of possible spatial and temporal changes in annual and seasonal rainfall.