Multidimensional analysis of climate-induced streamflow variability using CMIP6 data and advanced modeling techniques

The climate driven streamflow changes in Thailand’s TC-Basin are assessed with CMIP6 GCMs, machine learning (Random Forest, XGBoost) and Bayesian Model Averaging (BMA). For precipitation (R2 = 0.90, RMSE = 2.26–2.37 mm) and temperature (R2 = 0.82–0.91, RMSE = 0.78–1.43 °C), RF/XGBoost achieved high accuracy. For precipitation downscaling, GFDL-ESM4 was suitable, and CNRM-CM6-1 was best for temperature. The BMA ensembles improved reliability (precipitation: R2 = 0.84, RMSE = 4.24–5.45 mm; temperature: R2 = 0.91, RMSE = 0.78–1.17 °C). Annual precipitation increased to 96,151.50 mm under SSP585, and maximum temperatures were 33.66 °C. Wet season (June–October) increases by 9–10% (higher flood risks), dry season (March–April) decreases by 40–43% (water scarcity), and cool, dry season flows increase by 33–56% are projected. Variation of wind speed was moderate (R2 = 0.88, RMSE = 1.80–3.03 m/s), and humidity was stable (R2 > 0.89). This framework combines ML, GCMs, and BMA to quantify hydroclimatic shifts under climate extremes.