Abstract: In this study, we aim to simulate the groundwater level in irrigation area by utilizing sequential observation data from ground-based and remote sensing platforms. A deep learning model based on a multi-layer GRU network was developed for this purpose. Precipitation, soil moisture, historical measurements of groundwater levels, and Sentinel-2 remote sensing observations were used as simulation factors. Furthermore, groundwater level simulation experiments and result analyses are conducted for two irrigation areas located in North and South China. The experimental findings reveal that the groundwater level simulation model, which incorporates the synergy between sky-earth observations and deep learning, effectively establishes the intrinsic relationship between external environmental factors and groundwater levels within the irrigation area. The model is apparently superior to comparative models that merely considers groundwater level as it demonstrates impressive simulation performance, exhibits applicability in diverse geographical environments, and holds promising potential for practical applications. It can provide valuable decision-making support for crop cultivation and water resources management in irrigation area.

Key words: groundwater level simulation, deep learning, sky-earth cooperation, irrigation area, remote sensing