A new integrated framework for spatio-temporal anomaly detection in groundwater observation network

Ensuring the reliability of groundwater observation networks is critical for accurate hydrogeological assessments and sustainable water resource management. However, spatio-temporal anomalies in monitoring data can distort interpretations and compromise groundwater models. This study introduces a novel integrated framework for detecting and classifying anomalous observations in the Shiraz aquifer, Iran, using time series data from 28 piezometers. For the first time, this framework combines Multiple Linear Regression (MLR), Kriging interpolation, and Short-Time Fourier Transform (STFT) to evaluate spatial consistency, identify temporal abnormalities, and quantify observational uncertainty. Results reveal that several piezometers exhibit weak correlations with neighboring wells, reflecting distinct hydrogeological conditions or measurement inconsistencies. By integrating Machine Learning, geostatistical, and spectral analyses into a composite anomaly scoring system, the framework enhances confidence in distinguishing reliable from unreliable monitoring data. The findings underscore the importance of systematic anomaly detection in groundwater observation networks, offering practical guidance for improving data quality assessment, groundwater modeling accuracy, and long-term resource management strategies.