A hybrid rule-learning framework for identifying anomalies in high-frequency groundwater-level data

High-frequency groundwater monitoring is critical for effective water-resource management, yet anomaly identification in high-frequency monitoring data remains challenging because of strong temporal variability, heterogeneous regional characteristics, and diverse anomaly patterns. In this study, we proposed a hybrid rule-learning (HRL) algorithm framework for identifying anomalies in groundwater level monitoring data from multiregional porous aquifers. The HRL algorithm framework integrated 15 candidate anomaly-detection rules, enhanced via window processing, variational mode decomposition, and first-order differencing to derive rules for three common anomaly types: local outliers, long-term flatlines without change, and short-term sharp fluctuations. HRL used a stepwise optimization algorithm (SWOA) to select optimal regional rules by maximizing a scoring function: TP-α*FP-β*FN. The framework was validated using 210 monitoring stations across 14 major groundwater resource zones from the China Groundwater Monitoring Network. The results show that HRL achieves precision and recall rates exceeding 80% on most datasets, with an average F1 score of 89.4%, outperforming five baseline methods and a simple ensemble method. When α≥3 and β=1, HRL exhibits high accuracy, robustness and algorithmic transferability. This method enables efficient, rapid anomaly labelling of vast monitoring data and substantially reduces manual annotation requirements.