A data-driven methodology for calculating key eutrophication parameters in water systems

Eutrophication, driven by human-induced nutrient enrichment, primarily nitrogen and phosphorus promote algal blooms, reduces dissolved oxygen and adversely impacts aquatic biota across diverse systems, including lotic, lentic, and coastal waters. Traditional assessment methods rely on laboratory analyses of chlorophyll, nitrogen, and phosphorus, with costs in CONAGUA-accredited laboratories ranging from approximately USD $75 to $200 per sample, plus transport expenses of USD $15–40 per shipment. These approaches are expensive, time-consuming, and often vary by water body typology. Moreover, heterogeneous datasets encompassing diverse aquatic environments introduce computational challenges that require advanced processing. This study proposes a deterministic artificial intelligence model to estimate total nitrogen and phosphorus from low-cost, in situ measurable parameters, offering generalizability across multiple surface water bodies without temporal constraints. Data preprocessing included augmentation, class balancing, normalization, and stratification to address unbalanced categories and ensure robust performance. Several supervised machine learning algorithms were implemented within a Monte Carlo simulation, and hyperparameters were optimized via Bayesian methods. The models achieved high predictive accuracy (R2 > 0.97) in validation tests across four Pacific states (Sinaloa, Jalisco, Michoacán, Nayarit). Independent datasets from two additional states (Sonora and Oaxaca) confirmed the model's generalization capacity, maintaining robust performance despite slightly lower accuracy.