In this paper, we address an important problem of understanding the impacts of shale-gas development on the groundwater in the state of Pennsylvania, U.S.A., with a focus on the methane leakage from shale-gas wells. The problem is highly challenging because methane concentrations in water are influenced by a large number of factors, many of which are unobserved. Further, the correlation between the methane values and the impacting factors exhibits strong spatial heterogeneity. As a result, simply applying existing data analysis tools would result in poor performance in terms of predicting the methane concentrations values using the available data of the impacting factors, and more importantly, provide little insight into the actual impacts of shale-gas development. In this paper, we take a different approach to develop a simple spatially local ensemble model which is able to explore the local correlations in the data while enabling domain experts to provide timely feedback on its performance. Through experiments on real environmental datasets from multiple sources, we demonstrate the effectiveness of the proposed method.