Evaluating Nex-GDDP CMIP6 Models for Extreme Wet and Dry Events Over Indonesia

Abstract

This study evaluates the capability of 19 downscaled CMIP6 models from NASA's NEX-GDDP dataset in simulating hydroclimatic extremes across Indonesia. Using the high-resolution MSWEP dataset as a reference for the historical period 1981–2014, we assessed three precipitation-based ETCCDI indices: maximum 1-day rainfall (RX1DAY), total precipitation from very wet days (R95PTOT) and consecutive dry days (CDD). The evaluation was performed seasonally (DJF, MAM, JJA, SON) using bias, RMSE and spatial correlation metrics. We also explored four ensemble strategies: MEAN, MEDIAN, and skill-based subsets (BST_MEAN, BST_MEDIAN), based on five top-performing models ranked by Taylor Score. Results indicate that NEX-GDDP models tend to underestimate wet extremes (RX1DAY, R95PTOT) and overestimate dry spells (CDD), particularly during monsoonal periods. Seasonal disaggregation reveals that model performance is generally stronger during the dry seasons (JJA, SON), with lower RMSE and higher correlation, while wet seasons (DJF, MAM) show larger discrepancies. Skill-based ensemble strategies marginally improve performance, but no configuration performs optimally year-round. Biases are also spatially heterogeneous, reflecting limitations in representing sub-grid processes such as convection and orographic rainfall. MSWEP shows significant RX1DAY increases in MAM–JJA, while NEX-GDDP trends remain weaker and spatially limited. These findings highlight the importance of season-specific and index-specific evaluation in ensemble construction. For operational climate services in Indonesia, such as early warning systems and risk assessments, reliance on raw model output without further refinement could misinform decision-making. Our results underscore the need for tailored post-processing and regional calibration to enhance the utility of downscaled climate projections in Southeast Asia.