Neural network soil moisture model for irrigation scheduling

Abstract

Real-time irrigation scheduling systems attempt to eliminate crop water stress and achieve a high yield at harvest through the control of soil moisture. Artificial intelligence algorithms are possibly to learn the soil moisture dynamics in the soil-plant-atmosphere system and then are embedded into a low-cost controller to generate appropriate irrigation schedules. In this study, a neural network (NN) model was proposed to learn from a process-based agricultural systems model, the Root Zone Water Quality Model (RZWQM2) in predicting the root zone soil moisture during the crop growing season. Climatic data, rooting depth and hesternal soil moisture are set as inputs of the NN model to predict intraday soil moisture in different layers. Conditions with and without water supply are modeled separately to achieve a higher accuracy. The NN-based irrigation scheduling method (NN method), triggers irrigation when the predicted soil moisture drops to a level defined by the product of management-allowed depletion multiplied by the depth of available water to plant. Irrigation quantity was set to replenish the root zone soil water content to field capacity. NN method was compared with the reported water stress (WS) method based on RZWQM2. The results reveal that though the constructed NN model well-predicted soil moisture changes during the main crop season with minor errors, but the error was larger at lower soil moisture thereby decreased the scheduling efficiency. An NN ensemble model was tested and shown to improve the precision and robustness of soil moisture prediction as well as the scheduling performance in view of water conservation and yield maintenance. Combined with the NN ensemble model and adjusted lowest soil moisture for triggering irrigations, the NN ensemble-based irrigation scheduling method achieved a performance not better than that of the RZWQM2-WS method but exceeded the method based on evapotranspiration and water balance by up to 20%. The constructed NN ensemble model and NN ensemble-based irrigation scheduling method could act as an alternative to predicting soil moisture and obtaining efficient irrigation scheduling.