An Evapotranspiration Forecast Model for Food and Cash Crops Productionusing Long Short-Term Memory Network

Achieving self-sufficiency in food production remains a key priority for the Nigerian government, withsignificant progress made in increasing yields of staple crops such as rice, maize, and cassava. However, optimizing water resources remains a critical challenge for sustainable agriculture. Evapotranspiration (ET) the combined process of water evaporation from soil and plant transpiration plays a crucial role in efficient irrigation planning and water resource management. Traditional ET estimation methods require complex mathematical models, which often struggle with accuracy due totheir limited ability to capture intricate temporal patterns and dependencies in environmental data. This study develops a deep learning-based mid-and-long-term evapotranspiration forecasting model using Long Short-Term Memory (LSTM) networks. Unlike conventional models, LSTMs excel at capturinglong-range dependencies in time-series data, making them well-suited for ET prediction. Historical evapotranspiration data from Ogwashi-Uku, Southern Nigeria, spanning January 3, 2023, to December 22, 2023 (daily forecast), and January 1, 2003, to December 1, 2024 (monthly forecast), were used for model training and evaluation. The experimental results demonstrate high predictive accuracy, with a Mean Squared Error (MSE) of 0.0034, Root Mean Squared Error (RMSE) of 0.0583, and Mean Absolute Error (MAE) of 0.0433, leading to an overall model accuracy of 95.68% for daily evapotranspiration and MSE of 0.0005, RMSE of 0.0222, and MAE of 0.0182 for monthly evapotranspiration.