Long short-term memory neural network for irrigation management: a case study from Southern Alabama, USA

Abstract

The metabolism and growth of vegetation are highly dependent on the changes in soil water content. Irrigation scheduling and application of water at the right time and rate are a key aspect for precision irrigation. In this study, the Long Short-Term Memory (LSTM) Neural Network model was studied to predict irrigation prescriptions for 1, 3, 6, 12 and 24 h in advance. Training data for LSTM were collected from a precision irrigation study conducted in Alabama, USA. The prediction estimation of irrigation prescription used soil matric potential data measured within two contrasting soil types. Performance of the LSTM models were evaluated by comparing neural network parameters and prediction capability by soil type. The optimal learning algorithm for each case was also determined. The LSTM Neural Network showed good prediction capabilities for both soil types, with R2 ranging between 0.82 and 0.98 for one hour ahead prescription and getting smaller as prediction time increases. The irrigation rate prediction was verified by actual observations that demonstrate the suitability of the machine learning technique as a decision-support tool for irrigation scheduling.