A health-score-based framework for quality assessment and spatial reconstruction of rainfall monitoring data in reservoir watersheds

Abstract

Reliable rainfall records are essential for flood forecasting, reservoir operation and water resource management, yet automatic rain-gauge networks often contain persistent fixed values, long data gaps, isolated spikes and short peaks without realistic recession behavior. We present a model-driven framework for diagnosing and reconstructing rainfall monitoring data from the 100-station Meishan Reservoir network in Anhui, China. The framework couples rule-based anomaly detection with a station health score that summarizes data quality and guides donor selection for inverse distance weighting reconstruction. Across the network, 21 stations had health scores below 60 and were classified as poor or critical quality. Under the operational evaluation protocol, the health-score-filtered IDW method produced an RMSE of 0.281 mm, an MAE of 0.182 mm, an NSE of 0.979 and a PBIAS of − 3.82%. The reported aggregate comparison gave IDW the lowest RMSE and MAE and the highest NSE among the evaluated methods, although method coverage differed for the longest gaps. The results provide a practical basis for quality screening and gap reconstruction in reservoir rainfall monitoring.