Zone-specific selection of CMIP6 models using relative importance metrics for projecting precipitation extremes in South Kivu, eastern DRC

The intensification of precipitation extremes threatens hydro-ecological stability in topographically complex regions such as the Albertine Rift, yet global climate model (GCM) projections there are limited by structural uncertainties. We present a multi-stage framework to evaluate historical and future extreme precipitation (ten ETCCDI indices) across six agroecological zones (AEZs) of South Kivu, eastern DRC. From eight CMIP6 GCMs, we isolated regime-specialist models using Relative Importance Metrics (RIM), the Compromise Programming Index (CPI), and Jenks Optimized Classification (JOC), then apply Quantile Delta Mapping (QDM) and Bayesian Model Averaging (BMA) under SSP2–4.5, SSP3–7.0, and SSP5–8.5. Model fidelity is zone- and index-specific: MPI-ESM1–2-HR excelled for orographic intensity in high-altitude zones, while MRI-ESM2–0 in lowland and rain-shadowed zones. Several AEZs showed significant historical declines (1983–2014) in precipitation indices (e.g., decline of −4.5 mm yr −1 for PRCPTOT in the EHAZ), which reverse under high-emission pathways by the late 21st century, with PRCPTOT trends up to + 18.1 mm yr −1 and R95Ptot trends up to + 7.1 mm yr −1 . The THMAZ evolved into a hotspot of hydro-climatic whiplash combining longer dry spells with intensified convective extremes. The Time of Emergence (ToE) for SDII was as early as 2024 in the EMAZ and THMAZ, and that of RX1day was ~ 2027–2029 in the EHAZ. In contrast, most intensity and volume indices showed no emergence in the high-variability lowland and transitional zones. Provincial-scale averaging masks this sub-regional vulnerabilities: the THMAZ R95pTOT trend (+7.06 mm yr⁻¹) is more than twice the provincial mean (+3.2 mm yr⁻¹). These results provide a physical baseline for infrastructure and early-warning systems in this climate-exposed region.