Data-driven optimisation of multi-agency river water quality monitoring networks under high anthropogenic pressure

Abstract Integrating heterogeneous water quality data from multiple monitoring agencies into a single reproducible analytical pipeline remains an unsolved challenge in environmental science. We present the Optimised Water Quality Monitoring Framework (OWQMF), an open-source four-block pipeline that harmonises multi-source monitoring data, computes dual water quality indices with objectively derived weights using a game-theory ensemble of three objective weighting methods, produces gradient-boosting ensemble forecasts with walk-forward cross-validation, and selects the best forecasting model through multi-criteria consensus with Adaptive Conformal Prediction coverage guarantees ( $$\ge $$ ≥ 90%). Applied to 108,270 harmonised records from three contrasting South-eastern Brazilian river basins spanning four agencies (1977–2025), 18–31% of station-months receive discordant quality labels depending solely on which agency’s classification scale is applied to the same index score, rising to 23–31% once nitrogen-species substitution and objective re-weighting are also taken into account. Among officially Classe 2 designated station-months, 74.1% conceal at least one per-parameter CONAMA 357 violation—95.5% among those classified as Good (IQA 52–78). The data-driven index redistributes weight systematically away from the low-variance parameters that the regulatory schedule fixes high—pH falls from 0.120 to 0.048, total solids from 0.080 to 0.038, and temperature deviation and total phosphorus also decline—and towards the parameters that discriminate between monitoring conditions. This direction of redistribution holds in every basin and under every weight-derivation variant tested; the ordering among the up-weighted parameters is implementation-dependent and is not claimed as a result. A single fixed national weight schedule cannot express this distinction between measuring compliance and ranking condition. Ensemble forecasting achieves mean absolute errors of 2.3–4.5 IQA units at clean-catchment and main-channel stations. Portability of the weighting scheme is confirmed on the Ohio River Basin, USA ( $$n=13{,}230$$ n = 13 , 230 ); Italy ( $$n=49{,}380$$ n = 49 , 380 ); and Ireland ( $$n=13{,}380$$ n = 13 , 380 ) under WFD 2000/60/EC: IQA-equivalent medians cluster in a narrow 81.9–86.1 band while compliance rates span 12.7%–97.1%, demonstrating that composite scores systematically conceal regulatory non-conformity. Because the sub-index response curves were not regionally recalibrated, these external scores are interpreted comparatively rather than as regulatory assessments.