Precipitation forecasting using random forest over an ecuadorian andes basin

Abstract

Precipitation is vital for sustaining water supplies, agriculture, and ecosystems, directly impacting human life and environmental health. Accurate forecasting of precipitation is crucial for effective water resource management, disaster preparedness, and agricultural planning, helping mitigate risks and optimize hydrological resource. Accurately forecasting precipitation becomes challenging in regions marked by diverse influencing factors and significant spatial variations, as observed in the tropical Andes. Precipitation in the tropical Andes is associated to various sea surface temperature (SST) indices, such as ONI, NIÑO 1+2, TNI, and atmospheric variables, including temperature, wind, and humidity. The SST indices could better explain seasonal precipitation anomalies and the atmospheric variables could better explain faster changes in atmospheric conditions associated to precipitation. This study aims to forecast cumulative 30 days moving average precipitation in the Paute Basin River using the Random Forest technique to forecast 30, 60, and 90 steps ahead, utilizing a variety of predictors. The precipitation product used was IMERG Late Run. Global climates indices were used as predictors. Also, Atmospheric variables were used to build regional drivers associated with precipitation in the study area through Pearson correlation. Five distinct models were developed, gradually incorporating predictors such as past rainfall (ILP), global climate indices (GCI), regional climate drivers (RCD), and combinations of GCI+RCD and GCI+RCD+ILP. The results indicate that the most effective model is GCI+RCD+ILP, achieving R2 (MAE) values of 0.92 (9.38 mm/cumulative 30 days), 0.93 (9.85 mm/cumulative 30 days), and 0.94 (9.68 mm/cumulative 30 days) for 30, 60, and 90 steps ahead, respectively. The model could be used for water resource management, and the mitigation of impacts associated with droughts and floods and the optimization of hydrological resource. Finally, used this model for evaluate the accuracy in flows and hydropower generation is crucial for its use of operativity work