Comparative analysis of kernel-based versus ANN and deep learning methods in monthly reference evapotranspiration estimation

Abstract. Timely and accurate estimation of reference evapotranspiration (ET0) is indispensable for agricultural water management efficient use. This study aims to estimate the amount ET0 with machine learning approaches by using minimum meteorological parameters in Corum region, which has an arid semi-arid climate regarded as important centre Turkey. In this context, monthly averages variables, i.e. maximum temperature; sunshine duration; wind speed; average, maximum, relative humidity, are used inputs. Two different kernel-based methods, Gaussian process regression (GPR) support vector (SVR), together a Broyden–Fletcher–Goldfarb–Shanno artificial neural network (BFGS-ANN) long short-term memory (LSTM) models were amounts 10 combinations. The results showed that all four methods predicted acceptable accuracy error levels. BFGS-ANN model higher success (R2=0.9781) than others. GPR SVR Pearson VII function-based universal kernel was most successful (R2=0.9771). Scenario 5, temperatures including average temperature, duration inputs, gave best results. second scenario had only input BFGS-ANN, estimated having correlation coefficient 0.971 (Scenario 8). Conclusively, shows better efficacy BFGS ANNs enhanced performance ANN drought-prone regions.