Estimation of reference evapotranspiration from temperature data: A comparison between conventional calculation and artificial intelligence techniques in a warm-sub-humid region

Abstract

Reference evapotranspiration (ETo) is an agro-meteorological parameter of great importance for many areas of study such as geotechnics, climatology and hydrology, where its greatest importance falls in the calculation of the crop’s evapotranspiration (ETc). In this study, using only temperature data, the performance of three artificial intelligence models and two conventional equations to predict the reference evapotranspiration (ETo) was evaluated in a warm sub-humid climate in México. The artificial intelligence models evaluated were: support vector machines (SVM), Gene Expression Programming (GEP) and XGBoost, and the conventional models were those by Hargreaves-Samani and Camargo. The performance of the models was evaluated according to the statistical indexes: Mean Absolute Error (MAE); Root Mean Square Error (RMSE); Coefficient of Determination (R²), and Mean Bias Error (MBE). Confidence intervals were constructed for each statistical index using the technique of bootstrap resampling with the purpose of evaluating their uncertainty. The results show that among the conventional models evaluated, the equation by Camargo obtained a better performance in the estimation of ETo compared to the equation by Hargreaves. Regarding the artificial intelligence models, the SVM model obtained the best performance among the techniques evaluated. In general, it is recommended to use the SVM model to estimate the ETo values since it outperforms the other techniques.