Generalizability of machine learning models and empirical equations for the estimation of reference evapotranspiration from temperature in a semiarid region

Abstract The Penman-Monteith equation is recommended for the estimation of reference evapotranspiration (ETo). However, it requires meteorological data that are commonly unavailable. Thus, this study evaluates artificial neural network (ANN), multivariate adaptive regression splines (MARS), and original calibrated Hargreaves-Samani (HS) temperature (PMT) equations daily ETo using temperature. Two scenarios were considered: (i) local, models calibrated/developed evaluated from individual weather stations; (ii) regional, pooled several stations independently in each one. Local also outside calibration/training station. Data 9 used. PMT outperformed HS, but after local or regional calibrations, they performed similarly. locally machine learning exhibited higher performances than their versions. had generalization capacity, with a more stable performance between stations. better evaluated. When comparing ANN HS equation, mean RMSE reduced 0.96 to 0.87 0.84 0.73, scenarios, respectively. MARS similarly, slight advantage ANN.