Application of artificial intelligence for nutrient estimation in surface water bodies of basins with intensive agriculture.

This article focuses on developing machine learning (ML) models to estimate nitrogen and phosphorus nutrient concentrations related to eutrophication in surface waters of Sinaloa, Mexico, using readily accessible physicochemical parameters such as pH, dissolved oxygen, conductivity, water temperature, and total suspended solids. The study compares several ML methods—including multilayer perceptron (MLP), k-nearest neighbor (KNN), convolutional neural network (CNN), and random forest (RF)—and finds that the RF model, optimized via hyperparameter tuning and data normalization techniques, provides the most accurate nutrient estimations with coefficients of determination (R²) above 0.90 for total nitrogen, Kjeldahl nitrogen, ammonia nitrogen, total phosphorus, and orthophosphate. The models were trained on data from 2012–2019 and validated with independent data from 2020–2022, demonstrating robustness across different surface water types and climatic seasons without relying on spatial or temporal variables. While these ML-based estimations are not substitutes for laboratory analyses, they offer a cost-effective, rapid screening tool for environmental monitoring and decision-making related to eutrophication management.