A Novel Hyper-Resolution Water Table Depth Product for the Contiguous US

Abstract

Groundwater is our largest freshwater reservoir, playing a key role in supplying drinking water, guaranteeing food security, supporting biodiversity, and sustaining surface water bodies. While we have water table depth (WTD) observations at approximately one million wells over the contiguous US (CONUS), WTD data are sparse at the city or individual farm level, where local decisions are often made. To address the challenge, we introduce a novel WTD product for the CONUS, that consists of hyper-resolution (1 arcsec, ~30 m) long-term mean WTD estimates from a random forest model trained on available WTD observations. Uncertainty assessment is also provided. The input data to the random forest model include annual mean precipitation and temperature, elevation, distance to stream, soil texture, and other geology-related data. The model implicitly learns pumping from the WTD observations used for training, and thus the WTD product accounts for human interference with groundwater. Compared to coarser-resolution WTD data, it provides better estimates for groundwater storage and the proportions of very shallow and very deep aquifers over the CONUS. The 1-arcsec WTD product represents our most accurate estimate of accessible freshwater for the CONUS to date, useful for sustainable freshwater management, groundwater depletion studies, and hydrological modeling improvement. Since the CONUS covers many different hydrogeological settings, the random forest model trained for the CONUS may be transferrable to other regions with a similar setting and limited observations. We plan to extend the study globally, with the initial effort focused on transferring groundwater knowledge between the CONUS and Denmark.