GROUNDWATER USE IN THE OZARK PLATEAUS AQUIFER SYSTEM: CHALLENGES FOR ACCURATELY QUANTIFYING WITHDRAWAL RATES BY LOCATION

Water availability poses a serious concern because of changes in hydrologic stationarity resulting from climate change, water-quality degradation in surface and groundwater resources, and the continual increase in water demand. One of the challenges for assessing water resources is accurately quantifying components in the hydrologic budget, which includes rates of water withdrawal. Groundwater is an often overlooked resource compared to surface water, but groundwater is used widely across the United States, especially during periods of drought. In this research, groundwater withdrawals for 98 counties overlying the Ozark Plateaus aquifer system were compiled from aggregate, county-level data obtained from U.S. Geological Survey water-use databases, and site-specific water-use data obtained from state agencies. Challenges arose when merging these two datasets to address water use in various sectors, including public supply, self-supplied domestic use, agriculture (that is, irrigation and aquaculture), livestock, industry, power generation, mining, and commercial use. In 1962 (the earliest available record for water use in the study area), an estimated 76 million gallons per day of groundwater were withdrawn from the Ozark Plateaus aquifer system. In 2010, approximately 380 million gallons per day were withdrawn; an approximately 400 percent increase over 48 years. Between 1960 and 2010, the population in the counties overlying the Ozark Plateaus aquifer system increased approximately 50 percent. Population is an often-used explanatory variable for water-use rates, but did not necessarily correlate to groundwater withdrawal in the Ozark Plateaus aquifer system. The record of water use was most complete from the mid-1980s to present, but back-casting estimates of groundwater withdrawals was necessary for providing appropriate historical stress within the groundwater flow model so that future water-use scenarios can be confidently modeled. Historical groundwater-withdrawal estimates beginning in 1900 were made to fill the data gaps for the various water-use sectors using a combination of parametric and nonparametric statistical learning techniques, including ordinary least squares regression, K nearest neighbor regression, and boosted regression trees.