Satellite Data-Driven Deep Learning Approach for Monitoring Groundwater Drought in South Korea

Abstract

Due to the effect of climate change on the hydrological cycle process, the severity and frequency of drought have increased. Typically, drought begins with meteorological drought, after which it propagates to agricultural and hydrological drought. Thus, it is essential to investigate the process involved in the drought propagation from meteorological to groundwater drought. In this study, we investigated groundwater drought by calculating the standardized groundwater level index (SGI) using predicted groundwater storage changes (GWSC) based on satellite data-driven deep learning models. The GWSC was predicted using two deep learning models (the convolution neural network-long short term memory (CNN-LSTM) and LSTM), and the results were validated using in situ observation data. In addition, the SGI was compared to meteorological, agricultural, and hydrological drought indices based on remote sensed data, and the drought propagation was analyzed. This study revealed the potential of satellite data-driven deep learning models for assessing groundwater droughts, which is important for the development of multi-scale drought monitoring systems.