Extreme value combination prediction of summer–autumn drought–flood abrupt alternation in the upper Hanjiang River based on multiple machine learning ensembles

Abstract Summer-autumn drought–flood abrupt alternation (DFAA) event in the upper Hanjiang River significantly impacts regional water resources regulation and flood control safety. This study aims to improve seasonal prediction of the summer-autumn modified standardized drought-flood abrupt alternation index (MSDFAI). Key influencing factors were identified using the maximal information coefficient. Three machine learning models-random forest (RF), support vector machine (SVM), and long short-term memory (LSTM))-were optimized using particle swarm optimization (PSO) and evaluated via expanding window cross-validation. Nine ensemble members were constructed by combining three models and three factor schemes. To mitigate the systematic underestimation of extreme values in ensemble predictions, an extreme-value combination strategy was proposed to improve prediction performance. Results show that the extreme-value combination achieves a test-period NSE of 0.62 and RMSE of 0.17, outperforming the ensemble mean (NSE 0.56, RMSE 0.18), Bayesian model averaging (NSE 0.36, RMSE 0.22), best single member (NSE 0.48, RMSE 0.20), quantile regression (NSE 0.33, RMSE 0.23), and stacking (NSE 0.12, RMSE 0.256), with particularly pronounced improvements particularly in extreme years. This study provides an effective technical approach for seasonal DFAA prediction in the upper Hanjiang River.