Optimizing online groundwater monitoring in industrial parks using long-term high-frequency data

Chemical industrial parks require groundwater monitoring strategies that can detect abnormal changes while limiting redundant measurements. We analyzed nearly three years of hourly data from eight online monitoring wells in a chemical industrial park. Indicator screening combined the normalized interquartile range with Spearman correlation analysis; monitoring frequency was optimized using downsampling and reconstruction-error metrics; and predictive validation employed ridge and random forest regression. Online monitoring captured sudden and persistent anomalies lasting 100-3,360 h. Ammonia nitrogen (NH₃-N), chemical oxygen demand (COD), turbidity, pH, and groundwater level were the most informative indicators. Optimized intervals ranged from 29 to 720 h and varied by well and parameter. Seasonal sensitivity analysis supported the robustness of moderately optimized intervals, and site-specific dynamic thresholds reduced excessive alarms under disturbed baseline conditions. The framework supports practical design of adaptive groundwater online monitoring and early warning in industrial parks.