Interpolated depth to water table (groundwater) maps for the continental United States

DATA: This is a collection of depth to water table maps with uncertainty estimates for the continental United States for years 1989 and 2019. Data used to create these maps were obtained from the National Ground-Water Monitoring Network (NGWMN). Data included 14,351 sites and 17,632,047 observations for the years 1989-2019. To improve our inference a set of auxiliary variables proven to have a relation with depth to water table were included. We paired point estimates of depth to water table data with environmental data, as well as terrain variables derived out of a base digital elevation model (DEM) created by NASA at a 1x1km resolution. Climatic layers (temperature, precipitation, and snow melt equivalent) for 1989-2019 were obtained from Daymet (Version 4), which provides a continuous grid of historical monthly and annual weather data, with a 1x1km spatial resolution (Thornton et al., 2020). Out of the DEM, primary (slope, aspect) and secondary terrain attributes (curvatures, upslope contributing areas) were used to calculate a compound topographic index (CTI). MODELING FRAMEWORK: Water table depth analyses were conducted using a three-step interpolation approach: 1) we utilized gradient boosted regression trees (GBRT) to make predictions, 2) we used kriging interpolation on GBRT residuals to reduce bias from spatial autocorrelation, to incorporate a spatial correlation structure and to create uncertainty maps, and 3) we then combined the GBRT and kriging predictions for the final map. This method is equivalent to a Universal Kriging, where in our case, we evaluated the trend using GBRT. Model metrics were calculated for the training (80% of the data) and validation (20% of the data) datasets to evaluate overall performance. *** Uncertainty is greater surrounding the 1989 interpolations due to a lower number of observations.