Data-Driven Machine Learning Approach for Water Quality Anomaly Detection and NPS-FM Non-Compliance Risk Prediction: A Study With Five New Zealand Rivers

New Zealand’s National Policy Statement for Freshwater Management (NPS-FM) 2020 sets legally enforceable water quality limits, yet compliance is assessed retrospectively, and no operational system exists to predict which sites carry the greatest risk of future non-compliance. This study presents a machine learning pipeline that detects abnormal water quality events and quantifies NPS-FM non-compliance risk, applied to five South Island rivers of New Zealand spanning four land cover types. Anomaly detection combined Isolation Forest method with seasonal interquartile range thresholds, with observations flagged by both methods classified as confirmed anomalies. A weighted risk score (0–100) was computed from NPS-FM exceedance rate, confirmed anomaly rate, 90th percentile severity, and Mann–Kendall trend slope. This risk score is then validated using temporal hold-out, literature benchmarking, and seasonal pattern analysis. The stability of the applied pipeline is assessed by using alternative anomaly detection algorithms and by varying the quality-control filter. Risk scores followed a clear land cover gradient: the urban site scored 34.12 (High risk), with 96% of total nitrogen and 72% of E. coli readings exceeding NPS-FM limits, while native vegetation sites scored 14.24–16.24 (Medium risk). A paired catchment comparison isolated land use as the dominant driver, with the pasture site recording total nitrogen concentrations 0.3–0.5 mg/L above the upstream native site. Multi-indicator anomaly co-occurrence was 10.6–14.0 times higher than expected under random noise, and temporal hold-out validation passed at four of five sites. Site rankings and risk grades were unchanged when One-Class SVM or Local Outlier Factor was substituted for Isolation Forest. This research produces interpretable, indicator-level risk scores distinguishing episodic events from persistent contamination, consistent with established land use signatures. The framework is transferable to other regulatory contexts where long-term monitoring data and threshold-based standards exist.