Multi-state comparison of machine learning techniques in modelling reference evapotranspiration: A case study of Northeastern Nigeria

Monthly reference evapotranspiration (ET0) was predicted for Bauchi and Maiduguri stations located in the northeastern semiarid region of Nigeria. The data for 34 years (1983-2016) were used including maximum and minimum temperature, relative humidity, and wind speed. The models were developed using artificial neural networks (ANN), support vector regression and multiple linear regression (MLR). The most influential weather parameters and the best computing technique were also investigated. FAO Penman-Monteith (FAO-56-PM) is regarded as the sole method for estimating ET0, it is therefore employed in this study as the benchmark ET0. Two statistical indicators of root mean square error (RMSE) and determination coefficient (R2) were used to assess the performance of the models. The results showed that relative humidity has better performance in single input models but inclusion of wind speed can produce best performance for the 3 inputs models. However, the study revealed that ANN had the better ET0 prediction capability in both stations, and 3 inputs model with minimum temperature, relative humidity, and wind speed led to superior efficiency. The general results demonstrated that the ANN, SVR and MLR can be employed for reliable estimation of ET0 in the study stations.