Prediction of soil moisture based on Extreme Learning Machine for an apple orchard

Accurately grasp the dynamics of soil moisture enable one to react accordingly to the crop water demand to a certain extent, hence forecasting soil moisture is very important in scheduling of irrigation and allocating water in irrigation area rationally and efficiently. Soil moisture is affected not only by previous time sequence of soil moisture, but also by weather factors. In this paper, a novel method, which integrates weather factors with time series of soil moisture, to predict soil moisture is proposed. In order to improve the prediction accuracy and effectiveness, Extreme Learning Machine (ELM) which is a simple, reliable, efficient single-hidden layer feed-forward neural networks with fast learning speed and good generalization ability, is employed to simulate the relationship between future soil moisture and the above mentioned impact factors. Based on large-scale datasets obtained from the Dookie apple orchard in Victoria, Australia, the results show that the model can accurately predict the future trends of soil moisture and this can be a useful decision support for future irrigation scheduling. In contrast to Support Vector Machine (SVM) which is the conventional method of soil moisture forecasting, ELM has higher prediction accuracy and greater forecast range with fast learning speed.