Hybrid Approaches for Irrigation Optimization Based on Weather Forecast: a Study on Reference Evapotranspiration Prediction in Beni Mellal

Accurate prediction of Reference Evapotranspiration (ET0) is vital for optimizing irrigation, thereby facilitating efficient water management and agricultural planning. This study compares three distinct methods for predicting ET0 using the FAO Penman-Monteith (FAO-PM), leveraging daily weather data collected over a span of 38 years, from 1984 to 2022. The first approach involves predicting ET0 directly based on actual ET0 values, while the second hybrid approach uses Recurrent Neural Networks (RNN) to predict Net Radiation, Temperature, Wind speed, and Dew Point Temperature. These predicted values are then utilized in the FAO-PM equation to calculate ET0 (RNN-FAO-PM). The third approach is another hybrid method that combines RNN for predicting the weather parameters, followed by the application of a well-trained Random Forest (RF) model that uses the predicted weather parameters as features to predict ET0 (RNN-RF). The performance of each method is evaluated using various metrics, including Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and R-squared (R2) values for both training and testing datasets. The results of this study reveal that the hybrid approaches demonstrate comparable performance for long-term prediction of ET0 of the period Spanning from 2020 to 2022 (3 years). These hybrid approaches slightly outperform the RNN method when applied solely on the ET0 time series. This finding contributes to the research in the area of water resource management, specifically in the context of irrigation optimization. It provides valuable insights that can inform agricultural decision-making in the Beni Mellal region, enabling more efficient and effective use of water resources for irrigation purposes.