Deep Learning-based prediction of groundwater level fluctuation with hydrologic datasets of climate change scenarios in paddy fields agriculture area, Hongseong, Korea

Abstract:

Paddy fields generally require huge amount of water during growing season even they are influenced by drought. Hongseong is paddy-dominant agricultural area and water supply for rice growing is high priority of water distribution planning. Groundwater demands for paddy fields increase dramatically when drought comes. Therefore, predicting groundwater level fluctuation by climate change is important to estimate readily available groundwater resources under water supply emergency by drought. Since the Intergovernmental Panel on Climate Change (IPCC) has been founded, many assessment reports including climate change scenarios have been published. Hydrologic datasets such as surface runoff, evapotranspiration, and groundwater recharge using water-balance model have been derived from climate change scenarios and opened to public for application. Korea Meteorological Administration (KMA) provides predictive climate datasets of the Korean Peninsula in varying resolution. This study uses the datasets from KMA to develop an artificial neural network forecasting groundwater level fluctuation according to future climate change. Hydrologic datasets are usually time-series form and LSTM (Long Short Term Memory), one of deep learning algorithms, shows high performance to predict time-series data. A LSTM model has been structured to predict groundwater level fluctuation in Hongseong area under climate change. The model reflects hydrologic characteristics of the area and climate change. The results of the prediction showed that vulnerability of groundwater resources influenced by climate change occurs in future.

Acknowledgement: This work was supported by Korea Environment Industry & Technology Institute (KEITI) through Demand Responsive Water Supply Service Program, funded by Korea Ministry of Environment (MOE) (RE201901099).