Early warning forecasting of river ecological flow in a typical Karst basin, southwest China: Leveraging baseflow segmentation and interpretable machine learning

The Chengbi River Basin, a typical karst basin in Southwest China. Establishing an effective ecological flow early warning mechanism is crucial for water ecology protection and water resource management in a karst basin. Therefore, we combined the Boughton-Chapman digital filter baseflow separation technique with the Tennant method to estimate the ecological flow thresholds. We constructed an interpretable ecological flow early warning system using Shapley Additive Explanations and the CatBoost model and proposed a suitable early warning mechanism for the Chengbi River Basin. The potential contribution of input features to the output was deeply analyzed. The ecological flow early warning system was successfully applied to the Chengbi River Basin, and the accuracy was evaluated by the classification of warning levels. The results showed that the monthly baseflow frequency in the study area that satisfies the ecological flow condition was 30 % (1.95 m³/s) in the dry season and 30 % (20.43 m³/s) in the wet season. The accuracy of this early warning system exceeded 85 % in both the dry and wet seasons. The karst landscape in the Chengbi River Basin delayed the effect of precipitation on baseflow by up to 8 days. The early warning system had high ecological flow prediction accuracy and interpretability, providing a scientific basis for water ecology protection. • A novel ecological flow early warning system for the karst basin achieves a high forecasting accuracy of over 85 %. • Ecological flow thresholds for a karst region are determined, considering seasonal variations. • The time lag of rainfall impacts on ecological flows in the Chengbi River karst areas is about 8 days.