Abstract

Precise prediction of reference evapotranspiration (ET₀) plays an essential role for successful water resource management and agricultural practices. This study aimed to assess the effect of semiarid climate extremity in the prediction of ET0 in Mallam Fatori, Borno state, Nigeria using empirical and Machine Learning (ML) methods. Selection of dominant inputs are important for accurate prediction involving ML. Therefore, using past4 software, Pearson correlation based feature selection technique was applied to select the best inputs. Thereafter, four empirical prediction models, which include Hargreaves-Samani (HS), Modified HS (MHS), Ritchie (RT) and Makkink (MK) and three ML models namely; Support Vector Machines (SVM), Boosted Regression Trees (BRT), and Gaussian Process Regression (GPR) based were employed for the study purpose. Meteorological data for 34 years from 1984 – 2022 including monthly minimum, maximum and mean air temperatures (Tmin, Tmax and Tmean) (oC), relative humidity RH (%), monthly average wind speed U2 (m/s), monthly precipitation P (mm/month), surface pressure SP (kPa), specific humidity SH (g/kg), and solar radiation RS (kWh/m2/day) were used as inputs. The FAO-56 Penman-Monteith (PMF-56) method approved as the standard method for estimating ET0 was employed as benchmark for comparison purposes. The results showed that being an area subjected to high temperatures, Tmax was found to be the most dominant parameter. For empirical models, MHS led to better performance with root mean square error (RMSE) value of 0.1728. However, the ML models outperformed the empirical models across all modelling scenarios by up to 92%. SVM edged a little bit other ML models with superior performance having determination coefficient (R²) value of 0.9741 in the validation phase. Generally, this study demonstrated the potential of ML models to predict ET₀ effectively even in semiarid extreme climate and thus, leading to better water resource management and agricultural practices.

Key words: Boosted Regression Trees (BRT), Reference Evapotranspiration, FAO-56 Penman-Monteith, Gaussian Process Regression (GPR), Hargreaves-Samani (HS), Support Vector Machines (SVM).