Advanced Machine Learning Techniques for Hydrological Prediction of Lake Titicaca

Abstract

Global climate change is contributing to significant water reduction, evident in Lake Titicaca, where water levels have declined markedly in recent years. This study employs ten advanced machine learning algorithms to improve the prediction accuracy of water level behavior in the Lake Titicaca basin. The models utilize year, month, day, precipitation, maximum temperature, and minimum temperature as input variables. Accurate predictions are crucial for informing multidisciplinary planning related to agricultural and livestock production and for developing contingency strategies against droughts and floods. Historical data were sourced from the SENAMHI ENAFER pier station in Puno. Comparative evaluation indicates that the Gradient Boosting Regressor algorithm achieves superior performance based on multiple metrics: MAE = 0.00035279, MSE = 0.00000247, RMSE = 0.0015725, Coefficient of Determination R² = 0.99999706, and Cross-Validation (average RMSE = 0.00199047). This research underscores the potential of machine learning algorithms in hydrology, offering valuable insights applicable to basin modeling worldwide.