COVARIATE BUILDER AND MACHINE LEARNING MODELING SOFTWARE TO PREDICT GROUNDWATER LEVELS FOR THE MISSISSIPPI RIVER VALLEY ALLUVIAL AQUIFER

Abstract

Declining groundwater levels of the surficial Mississippi River Valley alluvial aquifer (MRVA) within the Mississippi Alluvial Plain (MAP), located in the south-central United States, have necessitated quantification of available groundwater resources to help inform future sustainability. Agriculture, public supply, and domestic wells rely on the MRVA as a source for irrigation and drinking water use. Statistical models can be used to quantify changes to groundwater resources and are advantageous to both potentiometric-surface maps and numerical groundwater-flow models because statistical models can be developed quickly and leverage large-scale networks of discrete and continuous groundwater observations at the regional scale. Applied statistical models hold promise to synthesize groundwater levels complementary to potentiometric maps and groundwater-flow models and statistical models can interpolate between wells to fill in spatial data gaps. The United States Geological Survey (USGS) MAP regional water availability study has created two software packages written in the R-programming language to predict monthly-mean groundwater levels quickly and quantify uncertainty of the predictions without developing sophisticated, time-intensive groundwater-flow models and potentiometric-surface maps. The covMRVAgen1 software package constructs covariates, such as hydrogeologic framework, global climate indices, and surface water data, bound to existing monthly groundwater levels, and feeds the results into the cubMRVAgen1 software package to train a Cubist machine-learning statistical model. The statistical model predicts monthly water levels through space and time (January 1980 through December 2019) and new insights to relative covariate importance in groundwater-level estimation are readily attained. The software are designed to run out-of-the-box and provide rapid estimates of changing water levels in an aquifer. Information gained from model outputs can be fed into groundwater-flow models, other machine-learning models and used by water-resource managers to make informed decisions regarding groundwater availability in their area.