Daily Prediction and Multi-Step Forward Forecasting of Reference Evapotranspiration Using LSTM and Bi-LSTM Models

Precise forecasting of reference evapotranspiration (ET0) is one the critical initial steps in determining crop water requirements, which contributes to reliable management and long-term planning world’s scarce sources. This study provides daily prediction multi-step forward ET0 utilizing a long short-term memory network (LSTM) bi-directional LSTM (Bi-LSTM) model. For predictions, model’s accuracy was compared that other artificial intelligence-based models commonly used forecasting, including support vector regression (SVR), M5 model tree (M5Tree), multivariate adaptive spline (MARS), probabilistic linear (PLR), neuro-fuzzy inference system (ANFIS), Gaussian process (GPR). The outperformed comparison based on Shannon’s entropy-based decision theory, while PLR proved be lowest performers. Prior performing multi-step-ahead ANFIS, sequence-to-sequence (SSR-LSTM), LSTM, Bi-LSTM approaches were for one-step-ahead past values time series. results showed sequence ascending order terms accuracies > SSR-LSTM ANFIS LSTM. provided (5 day)-ahead next step. According results, reasonably accurate acceptable multi-step-forward with relatively lower levels errors. In final step, generalization capability proposed best (LSTM predictions forecasting) evaluated new unseen data obtained from test station, Ishurdi. performance assessed three distinct datasets (the entire dataset first second halves dataset) derived between 1 January 2015 31 December 2020. indicated deep learning techniques Bi-LSTM) achieved equally good performances as training station dataset, developed. research outcomes demonstrated ability developed generalize capabilities outside station.