General circulation model projections are inadequate for impact studies due to their coarse resolution and inherent biases. Therefore, there is a need to accurately correct these biases and examine the abilities of these correction techniques in replicating the observed climate change signals. Over the Komadugu-Yobe Basin, this study investigates the performance of univariate empirical and parametric quantile mapping, as well as multivariate bias-correction (BC) techniques, using; N-dimensional probability density function transform (MBCN), Spearman rank correlation dependency (MBCR) and Pearson correlation dependency (MBCP) in correcting the biases in some selected CMIP5-Coordinated Regional Climate Downscaling Experiment model temperature projections based on the historical period (1975–2005) and the future (2020–2050) for annual, dry and wet periods under two emission scenarios (RCP 4.5 and 8.5). The temporal temperature variability of the BC outputs is further assessed. In correcting the temperature distribution, both univariate BC perform well for all seasons, while MBCN performs best during the dry season. The latitudinal-time cross-section result shows that high temperatures are mismatched either by the years of occurrence or by the latitudes at which they occurred. The univariate BC techniques and MBCN performed best in replicating the observed monthly variability. There is a positive temperature trend in most parts of the basin, however, with increased magnitudes in the future. Overall, the bias-corrected model ensemble mean performs better in replicating the trend in temperature, while the multivariate BC methods correct the joint dependence structure between them modelled variables, thereby providing a general-purpose methodology to the climate community.
Read full abstract