Impact of Two Variants of Quantile Mapping Bias Correction on CMIP6 ‐Based Projections of Extreme Precipitation Over Southern Africa

This study evaluates two variants of quantile distribution mapping (QDM) bias‐correction techniques applied to CMIP6 model ensemble projections: (1) coefficient of variation conserving (C‐QDM) and (2) non‐conserving quantile mapping (NC‐QDM). Both methods were applied to daily precipitation from 11 CMIP6 models using the CHIRPS dataset as reference. We examined how these approaches influence projections of key extreme precipitation indices under three emission scenarios (SSP1‐2.6, SSP2‐4.5, SSP5‐8.5) for the austral summer (DJF) across near (2026–2050), mid (2051–2075), and far (2076–2100) future periods relative to 1990–2014. Both methods consistently project increases in consecutive dry days (CDD) across most of Southern Africa, suggesting heightened drought risk, particularly under SSP5‐8.5. In contrast, wet‐day persistence shows method‐dependent responses: C‐QDM generally projects stronger reductions in consecutive wet days (CWD) while NC‐QDM indicates declines in wet‐day frequency (RR1) with comparatively weaker reductions in CWD. Heavy precipitation and intensity indices are projected to increase substantially under higher‐emission scenarios, notably over eastern Southern Africa and Madagascar sub‐regions. By the late 21st century, CDD increases by 0.88 days per decade under C‐QDM and 1.16 days per decade under NC‐QDM, demonstrating stronger amplification of drying under NC‐QDM. In contrast, Rx1day increases more strongly under C‐QDM (6.44 mm per decade) than under NC‐QDM (5.17 mm per decade), while R1mm decreases by 0.38 and 1.30 days per decade under C‐QDM and NC‐QDM, respectively. The SDII increases by 0.5 and 1.1 mm per decade under C‐QDM and NC‐QDM, respectively, indicating stronger intensification of precipitation under NC‐QDM. C‐QDM generally produces more moderate and spatially coherent changes, particularly for persistence‐related indices, whereas NC‐QDM projects larger increases in both wet and dry extremes relative to the historical baseline. These differences arise from their treatment of distributional variability: C‐QDM constrains the mean and coefficient of variation using a method‐of‐moments formulation, which stabilises variability, whereas NC‐QDM relies on direct distribution fitting, allowing greater flexibility and amplification in the upper tail. Overall, the findings demonstrate that while the direction of climate change signals is robust across methods, the magnitude and characteristics of projected extremes are sensitive to both emission pathways and bias‐correction choices. The results highlight the importance of using multiple bias‐correction approaches to better constrain projections and support climate‐resilient water resource management, agricultural planning, and disaster risk reduction in Southern Africa.