Comparison of Artificial Neural Networks and Random Forest for Reference Evapotranspiration Estimation and Its Application to the Water Balance of the Pucapuquio Irrigation System, Huancayo, Peru

Martinez Y.C.V.; Romero G.J.V.; Terrel E.J.Y.; Terbullino G.F.M. · SSRG International Journal of Civil Engineering (2026) · DOI: 10.14445/23488352/ijce-v13i7p110

Resumen

Climate change causes rising temperatures, changes in precipitation and other climatological variables that are related to reference Evapotranspiration (ET₀). The estimation of ET₀ is crucial for irrigation scheduling and water resource management. In this sense, the Penman-Monteith (PM) method is considered the most precise for ET₀ calculation, but the use of the method is often constrained by limited climatological data. In Peru, the PISCOeo_pm dataset provides ET₀ values for the period 1981–2016 that apply the PM method and presents a baseline to train machine learning models that can use limited data to extend ET₀ estimates to recent years and make precise calculations of water demand requirements for irrigation systems. In this research, a random forest model and a multilayer perceptron neural network are developed using PISCOeo_pm data and then used to estimate recent ET₀ values. Daily estimation performance and climatological consistency revealed that the MLP is superior to the RF model. The MLP-estimated ET₀ values for the 2017–2025 period is then used to calculate and update the water demand in the Pucapuquio irrigation system located in the district of Pucará, in Huancayo province, Peru. The results revealed that water demand has increased in recent years and that the previous water demand for the critical month of August, originally 20,769 m³, has increased to 24,150 m³ under current climatological conditions. Potential mitigation strategies are presented that include increasing irrigation application efficiency or reducing cultivated areas by 15% for critical months, until more robust solutions are implemented, such as evaluating supplementary water sources that can potentially satisfy the new demand or the implementation of more efficient irrigation systems.

Instituciones: Universidad Continental