PREDICTION OF WATER STORAGE IN IRRIGATION PONDS USING DEEP LEARNING

Abstract

In order to predict water storage with reasonable accuracy, long-term hydrological observation and information data on water intake is required. However, unlike dams, there is no such data on irrigation ponds. Therefore, in this study, we focused on a deep learning model for stock price prediction. We constructed a deep learning model based on short-term hydrological observation data, and tried to predict water storage by using this model. From the result, then, we examined how to improve the prediction accuracy. The results are as follows. In the deep learning model, the predicted value tended to be delayed by about Lead Time from the measured value. It was found that this prediction delay could be improved by adding precipitation data corresponding to the Lead Time period to the input variables of the model. In addition, it was found that the prediction accuracy of irrigation period in which the water storage changed frequently due to artificial discharge such as irrigation was inferior to non-irrigation period. Based on those results, it is inferred that to extract factors that have a great influence on the water storage during the irrigation period and to incorporate them in the model as input variables are necessary to improve the prediction accuracy in the irrigation period.