The CMIP6-downscaled CORDEX-Southeast Asia (SEA) ensemble: evaluation and benchmarking for megacities of SEA

Abstract. A 21-member ensemble of regional climate simulations has been produced for Southeast Asia (SEA) by dynamically downscaling Coupled Model Intercomparison Project Phase 6 (CMIP6) Global Climate Models (GCMs) under the World Climate Research Programme’s Coordinated Regional Climate Downscaling Experiment (CORDEX). The ensemble was generated by several modelling institutes using three regional climate models (RCMs) with eight distinct model configurations, resulting in a total of 62 simulations spanning the historical period and multiple future emissions scenarios. Model performance for mean, daily maximum/minimum temperature, and precipitation was evaluated against multiple observations at annual, seasonal, and daily time scales over SEA and its two subregions: Mainland and Maritime Continent (MC). Despite large observational uncertainties in precipitation intensity, the CMIP6 CORDEX-SEA ensemble captures the spatial and seasonal rainfall distribution reasonably well but tends to substantially overestimate observed rainfall. Wet biases, evident in about two-thirds of the models, are regionally and seasonally heterogeneous and larger over monsoon-dominated regions and seasons (e.g., MC during November–April and the Mainland during May–October). All RCMs showed widespread, statistically significant cold biases in daily mean temperature, which were largest during boreal winter, over the Mainland, and in simulations that have significant wet biases. These cold biases primarily arise from the models’ underestimation of daily maximum temperature. The MC remains a challenging region since models struggle to accurately capture the spatial variability of rainfall and the internal variability of temperature. A standardised benchmarking framework was applied to precipitation and temperature, which ultimately identified 15 historical simulations that met our a priori model performance expectations. Analysing the range of future projections and model independence shows that simulations from the same RCM family exhibit similar bias structures, highlighting the importance of RCM setup and the selection of statistically independent models. From this process, eight simulations spanning three RCM configurations were selected for further kilometre-scale dynamical downscaling over megacities of SEA.