Using an Artificial Neural Network to forecast groundwater levels following the removal of a large dam, Milltown Montana Ashley Marks

Fifty percent of the world’s population depends upon groundwater as their main source of drinking water (Hirata et al., 2007). Scarcity of groundwater clearly affects the entire world. One quarter of the world’s people live in areas characterized by physical water scarcity, making competition for water resources intense (International Water Management Institute (IWMI), 2006; World Water Council, 2008). Tools that forecast groundwater levels have been progressively developed over time, from the Boussinesq equation in 1871 to present day. However, complex three dimensional numerical flow models are the standard for determining groundwater behavior in most settings. These often require excessive field work, data collection, expense, and computational expertise. Artificial Neural Networks (ANNs) have been successfully used in other disciplines as a more practical and cost effective alternative for predicting outcomes dependant on multiple, complex, varying inputs. This research investigates the utility of ANNs to forecast groundwater levels from common data acquired on national data bases. Around Missoula in west central Montana, groundwater levels play an important role especially in the East Missoula and Turah areas, since groundwater levels were recently affected by the removal of the 28 ft Milltown Dam. The dam had impounded contaminated sediments which were polluting the Clark Fork River and nearby wells. Prior to dam removal engineers lowered the reservoir by 12 feet to examine the submerged portion of the dam. Water levels declined in wells during this initial drawdown and local citizens reported dry wells. This prompted a one million dollar well replacement response by the EPA to proactively protect water supplies in the 500+ domestic wells proximal to the reservoir. ANN’s can be an invaluable tool for forecasting groundwater behavior and have been successful for predicting groundwater levels within a foot of observed levels in several Milltown wells.