Quantitative assessment of groundwater controls across major US river basins using a multi-model regression algorithm

• We evaluate spatial variability in the physical controls of groundwater behavior.
• We introduce a novel clustering algorithm that links predictors to responses.
• Our analysis encompasses the majority of eight major river basins in the US.
• Water table depth is most strongly related to location within the basin and slope.
• Groundwater flux is most strongly related to hydraulic conductivity and recharge.

Spatial patterns in the physical controls of groundwater depth and flux are assessed quantitatively using results from a first of its kind, integrated groundwater surface water simulation over the majority of the contiguous US. We apply a novel, k-regression algorithm to the simulated system to simultaneously identify spatial subsets of grid cells with similar relationships between explanatory variables and groundwater metrics while quantifying behavior using multiple linear regression. The combination of this statistical approach with the results of a large-scale, high-resolution groundwater simulation allows us to evaluate the ability to represent complex groundwater behavior with simple linear models across an unprecedented range of climates and physical settings. In almost all of the eight major basins considered, we identify at least some areas where the coefficient of determination for the linear regression model is larger than 0.7, and in many cases this is achieved for more than 50% of the total basin area. In general, we show that water table depth is most strongly related to location within a basin and slope, while conductivity and recharge are more important predictors for groundwater flux metrics. Results also illustrate spatial variability in these relationships; further demonstrating the historic difficulty in developing spatially contiguous classifications of groundwater behavior. This work highlights the potential to combine new statistical techniques with integrated hydrologic models to help improve our understanding of complex heterogeneous systems.