Impacts of regional characteristics on improving the accuracy of groundwater level prediction using machine learning: The case of central eastern continental United States

Abstract

Study region: Central eastern continental United States. Study focus: Groundwater level prediction is of great significance for the management of global water resources. Recently, machine learning, which can deal with highly nonlinear interactions among complex hydrological factors, has been widely applied to groundwater level prediction. However, previous studies mainly focused on improving the simulation performance in specific regions using different machine learning methods, while this study focused on the impacts of regional characteristics on improving the accuracy of groundwater level prediction using machine learning. New hydrological insights for the region: A gated recurrent unit (GRU) neural network was built for groundwater level simulation in 78 catchments in the study region, and principal component analysis was used to cluster a variety of catchment hydrological variables and determine the input variables for the GRU model. Detrended fluctuation analysis was applied to analyze the autocorrelation of groundwater level in each catchment. This study further explored the influences of the hydrogeological properties of different catchments and the autocorrelation of groundwater levels on machining learning simulations. The results showed that the GRU model performed better in regions where hydrogeological properties could promote more effective responses of groundwater to external changes. Moreover, a negative correlation between the simulation performance of machine learning and the autocorrelation of the groundwater level was found.

Keywords

• Machine learning
• Groundwater level
• Detrended fluctuation analysis
• Principal component analysis
• Regional characteristics
• Central eastern continental United States