Improving crop-specific groundwater use estimation in the Mississippi Alluvial Plain: Implications for integrated remote sensing and machine learning approaches in data-scarce regions

Abstract

Study region: The Mississippi Alluvial Plain (MAP) in the United States (US). Study focus: Understanding local-scale groundwater use, a critical component of the water budget, is necessary for implementing sustainable water management practices. The MAP is one of the most productive agricultural regions in the US and extracts more than 11 km3/year for irrigation activities. Consequently, groundwater-level declines in the MAP region pose a substantial challenge to water sustainability, and hence, we need reliable groundwater pumping monitoring solutions to manage this resource appropriately. New hydrological insights for the region: We incorporate remote sensing datasets and machine learning to improve an existing lookup table-based model of groundwater use previously developed by the U.S. Geological Survey (USGS). Here, we employ Distributed Random Forests, an ensemble machine learning algorithm to predict annual and monthly groundwater use (2014–2020) throughout this region at 1-km resolution, using pumping data from existing flowmeters in the Mississippi Delta. Our model compares favorably with the existing USGS model, with higher R2 (0.51 compared to 0.42 in the previous model), and lower root mean square error (RMSE) and mean absolute error (MAE)— 0.14 m and 0.09 m, respectively in our model, compared to 0.15 m and 0.1 m in the previous model. Therefore, this work advances our ability to predict groundwater use in regions with scarce or limited in-situ groundwater withdrawal data availability.

Keywords

• Groundwater
• Remote sensing
• Machine learning
• Regression
• Irrigation
• Estimation