Abstract
Streamflow projections under climate and land-use change rarely state which results are robust to model uncertainty. This study developed a hybrid QSWAT-ANN workflow with a built-in verification system of four criteria, namely model performance (C1), signal direction (C2), spatial ranking (C3), and reference-series sensitivity (C4), for the Huai Luang River Basin (HLRB) in Northeast Thailand. It used six CMIP6 trajectories (three climate models under SSP2-4.5 and SSP5-8.5) and a single land use pathway to 2070. The corrected model passed C1 with overall NSE of 0.80 and 0.79. Ensemble precipitation increases of 10.5–13.7% were amplified into annual streamflow increases of 32.0–49.9%t, concentrated in the monsoon season, while paddy rice decreased by 55.1% toward cash crops. Within-trajectory Standardized Streamflow Index (SSFI) signals agreed in direction in at least four of six trajectories in all cases (C2 passed); the GCM model, ACCESS-CM2 was the only dissenting trajectory, with an explainable cause. The Drought Hazard Index (DHI) and Flood Hazard Index (FHI) classifications of 31 sub-basins gave rank agreement (Spearman p) of 0.545–0.646 and 0.570–0.811, passing the 0.60 threshold in three of six cases (C3). Screening still identified non-overlapping robust cores of eight (DHI) and ten (FHI) sub-basins at agreement of at least five of six trajectories. C4 confirmed complete rank preservation (p = 1.000), while index magnitudes shifted by typically 1.2–1.5 SD and more than 7 SD in extreme drought months. The workflow identified signal direction, seasonal structure, and upper-tier spatial cores as robust enough for Huai Luang reservoir management and sub-basin drought and flood targeting, whereas absolute index magnitudes should not be used as design values. This framework transfers directly to other monsoon basins.