Reliability Assessment and Statistical Calibration of SAI Ensemble Projections in Southeast Asia

Stratospheric Aerosol Injection (SAI) has been widely investigated as a potential Solar Radiation Management (SRM) strategy to offset global warming, with ensemble-based Earth System Model simulations such as the Geoengineering Large Ensemble Simulation (GLENS) providing key evidence for its climatic impacts. However, the reliability of these ensemble projections, particularly at regional scales, remains insufficiently assessed. This study presents a joint evaluation of the projection skill and statistical calibration of GLENS ensemble outputs over Southeast Asia, focusing on precipitation and near-surface temperature variables. First, ensemble skill is assessed against ERA5 reanalysis using rank histograms, confidence interval coverage, Continuous Ranked Probability Score (CRPS), and Brier Score. Results show that raw GLENS projections are systematically underdispersive and biased, with overly narrow uncertainty ranges that frequently fail to capture observations. Projection skill exhibits strong regional contrasts, with poorer precipitation performance over the Maritime Continent and weaker temperature skill over mainland Southeast Asia. These deficiencies indicate overconfident ensemble behavior and limit the direct usability of raw GLENS outputs for impact assessment and decision support. To address these limitations, Bayesian Model Averaging (BMA) is applied as a probabilistic post-processing method to calibrate monthly mean temperature projections. BMA substantially reduces systematic bias, corrects ensemble dispersion, and improves probabilistic reliability across most countries. Post-calibration CRPS values consistently decrease, and predictive distributions better represent observed variability. Overall, the combined results demonstrate that while GLENS captures large-scale climatic signals of SAI, statistical calibration is essential to reduce uncertainty and obtain reliable regional projections. This study highlights the importance of integrating ensemble verification and calibration to support robust interpretation of SRM impacts in climate-sensitive regions such as Southeast Asia.