Uncovering the strengths and weaknesses of an ensemble of quantile mapping methods for downscaling precipitation change in Southern Africa

Study region: Southern Africa.Study focus: We assessed quantile mapping strengths and weaknesses in climate downscaling based on global climate models projections. An ensemble of methods is used: six methods using parametric transformations, three methods with non-parametric transformations and one method with theoretical distributions transformations. Seventeen climate indices related to precipitation are applied to the locations of 15 UNESCO biosphere reserves. The methods are validated using a cross-validation procedure. The results of the projections for two different future scenarios from the Coupled Model Intercomparison Project Phase 6 are analysed and their contribution to the overall uncertainty in the climate indices is quantified. Quantile mapping stationary assumptions are also investigated.New hydrological insights on the region: The relative error from the cross-validation shows that both the non-parametric methods and the parametric methods using an exponential-type trans-formation have in general a good skill in correcting the biases. The method using theoretical distributions shows a good skill for low extremes. The analysis of the methods' uncertainty contribution reveals that the largest contributors to the overall uncertainty are the climate models whereas in some cases, the methods have the highest uncertainty share. The uncertainty contribution from the methods is higher for the high extremes compared to the low extremes. The stationary assumptions seem robust. The projections indicate a tendency of the region towards dryer conditions and intensified precipitation events. However, this trend contains strong intra -regional variations.

Automated summary

This study focuses on assessing the strengths and weaknesses of quantile mapping in climate downscaling based on global climate model projections in Southern Africa. Different methods, including parametric and non-parametric transformations, are used and validated using cross-validation. The results reveal that non-parametric methods and parametric methods using exponential-type transformation have generally good skill in correcting biases. The uncertainty contribution analysis shows that the climate models are the largest contributors to overall uncertainty, while in some cases the methods have the highest uncertainty share. The stationary assumptions of quantile mapping are found to be robust. The projections indicate a tendency towards dryer conditions and intensified precipitation events in the region, with strong intra-regional variations.