A novel red-edge-based anomalous water detection index and diagnostic framework

Frequent water anomalies pose serious threats to human habitats. Owing to the complexity of aquatic environments and the diversity of anomaly types, satellite remote sensing still faces challenges. In this study, the anomalous water index (AWI) and the advanced water turbidity index (AWTI), were developed using Sentinel-2's red edge band. Based on these parameters, we constructed a two-step decision-tree-based diagnostic framework (ADF) consisting of detection and classification components. The proposed indices and framework were compared with existing spectral indices and supervised learning algorithms in eight global study areas. The results showed that AWI effectively identified various water anomalies, whereas the AWTI distinguished turbid water from algal water. Both indices yielded F1 and kappa values exceeding 0.89, outperforming the existing indices. ADF achieved comparable performance to both the random forest and support vector machine algorithms, with F1 scores of approximately 0.9. The ADF exhibited strong time series monitoring performance across four cases, with kappa values of 0.87–0.94, indicating high accuracy and few false positives. In conclusion, this study provides a solid theoretical basis for detecting and classifying water anomalies without prior knowledge. Future work will extend the method to more anomaly types and regions, enhancing threshold generalization for global monitoring.