Spatiotemporal analysis of water quality in Lake Titicaca using in situ physicochemical parameters and machine learning predictive models

Mamani-Quispe J.N.; Pinto-Mamani B.D.; Quispe-Alania B.; Machaca-Hallasi L.; Marquina-Araujo J.J.; Cotrina-Teatino M.A. · Sustainable Water Resources Management (2026) · DOI: 10.1007/s40899-026-01359-z

Resumen

The progressive degradation of water quality in Lake Titicaca, driven primarily by untreated wastewater discharges, diffuse agricultural runoff, and urban expansion, poses increasing risks to ecosystem integrity and public health. This study presents a hybrid framework integrating multi-campaign in situ monitoring (2011–2024), descriptive spatial mapping via Inverse Distance Weighting (IDW), and machine-learning regression to model two policy-relevant indicators: chlorophyll-a (Chl-a) and dissolved oxygen (DO). Four algorithms were benchmarked under grouped tenfold cross-validation to reduce spatial leakage: Random Forest, XGBoost, LightGBM, and an artificial neural network (MLP). XGBoost achieved the best out-of-fold performance for both targets, reaching mean R2 = 0.92 for Chl-a (testing MAPE = 10.89%) and mean R2 = 0.96 for DO (testing RMSE = 0.12 mg/L, MAE = 0.08 mg/L, MAPE = 1.25%), with small Train–Test gaps indicating good generalization. Spatiotemporal mapping and distributional analyses consistently identified persistent eutrophication pressure in Puno Bay and the Minor Lake, expressed as recurrent elevated Chl-a and localized low-DO events. Overall, the proposed workflow provides a transparent and scalable decision-support tool for monitoring design and targeted management in high-altitude Andean lakes under increasing anthropogenic pressure. © The Author(s), under exclusive licence to Springer Nature Switzerland AG 2026.

Instituciones: Universidad Nacional de Trujillo; Universidad Nacional del Altiplano; Universidad Nacional de Juliaca