FORECASTING CLIMATE CHANGE EFFECTS ON GROUND WATER RECHARGE USING AN UNSUPERVISED ARTIFICIAL NEURAL NETWORK

Optimal ground-water resource management under changing climate requires knowledge of the rates and spatial distribution of recharge to aquifers. This paper presents an alternate methodology to estimate recharge from available and uncertain hydrologic, land use, and topographic information without long-term monitoring. The method was applied to twelve basins in southeastern Wisconsin where recharge observations were determined using a recession-curve-displacement technique and normalized by annual precipitation. Uncertainty was introduced and correlation preserved among these explanatory and response variables using a Monte Carlo (MC) technique. An unsupervised artificial neural network algorithm reduced dimensionality by projecting common patterns among the MC realizations onto a two-dimensional self-organized map (SOM). Fitted data vectors in the SOM were used to estimate normalized recharge ratios that compared well with the observations and published results based on a linear multivariate model. The effects of climate change on spatial recharge were estimated using the model and precipitation extremes associated with the El Niño southern oscillation. This new methodology provides an alternative approach to forecasting the effects of climate change on ground-water resources in basins with perennial streamflow.