Spatial and temporal comparison of CORDEX-SEA CMIP5-based downscaling experiments to ERA5 across Southeast Asia

Abstract

Simulating precipitation remains a significant challenge for climate models. This study assesses the performance of nine CORDEX-SEA CMIP5-based downscaling experiments, representing the highest resolution available for Southeast Asia, against ERA5 reanalysis data. Performance is evaluated across spatial and temporal dimensions using Taylor’s skill score. To compare rainfall patterns among SEA subregions in the models and ERA5, k-means clustering is proposed as a more appropriate method than those used in previous studies. Spatially, results indicate that the ECE_b and GFDL_b models tend to overestimate precipitation compared to ERA5, particularly over the Indonesian Maritime Continent during boreal winter (DJF) and northern Southeast Asia during summer (JJA). Conversely, the IPSL_b model consistently simulated lower precipitation levels. CNRM_a demonstrated the closest resemblance to ERA5, followed by NorESM1_d and HadGEM2_a. Temporally, all models showed lower performance scores, with NorESM1_d and GFDL_b achieving the highest results. Notably, several models captured heightened precipitation in the mountainous regions of Papua and Sumatra. For trend evaluation, annual rainfall trends are simulated more reliably than monthly variations, though localized serial correlation significantly alters trend detection in specific regions. The k-means clustering analysis revealed inconsistencies in cluster label assignments among different datasets, highlighting the need for post-processing to reorder labels for comparison. These findings provide critical insights for selecting appropriate rainfall models for future climate projection analysis across Southeast Asia.