A hybrid framework integrating quantile regression and isolation forest for reservoir water level anomaly detection in early warning applications

Reservoir water level monitoring and early warning are critical components of flood control and disaster mitigation in river basins, as they can effectively prevent dam failures or downstream flooding. However, existing reservoir monitoring systems generally suffer from response delays and ineffective early warnings when dealing with sudden anomalies in water level readings (such as those caused by extreme hydrological conditions, instrument malfunctions, or localized damage). This study proposes a method for the automatic identification and early warning of abnormal water level changes that integrates the quantile regression model and the isolated forest algorithm (QR-IForest framework). Taking two series-connected reservoirs in the Yonghan River Basin in Guangdong Province as the study subjects, the method performs anomaly detection on real-time water levels based on data from actual monitoring stations in the reservoir area. The results indicate that: (1) The fused method achieves higher accuracy in anomaly detection than either method alone, with a precision rate of 92.24%, a recall rate of 0.8992, and an F1-score of 0.9106. The true positive rate (TPR) for anomaly detection reaches 99.15%, while the true negative rate (TNR) reaches 89.92%. (2) Simulation tests were conducted for two different scenarios (heavy rainfall and river channel blockage). The results show that the model’s anomaly detection capability can provide rapid, early warnings, detecting anomalies as early as 2 hours before an incident occurs. (3) This integrated model can automatically classify detected anomalies. When identifying anomalies related to monitoring equipment or those caused by deviations in hydrological patterns, the detection accuracy exceeds 85%. The method developed in this study provides a reliable basis for efficiently identifying anomalies in water level monitoring data from small and medium-sized reservoirs, thereby ensuring the safe management and operation of these reservoirs.