| |
The Upper Floridan Aquifer (UFA) receives nitrogen loads from a variety of sources varying through space and time, the fate of which is influenced by a range of hydrogeologic and biogeochemical factors. Mechanistic models of nitrogen fate and transport are difficult to construct because of the complicated karst geology and high potential for conduit flow. As an alternative, we developed a statistical model using spatially explicit geologic, land use, and nutrient loading data to predict occurrence of elevated aquifer nitrate concentrations. We retrieved nitrate monitoring data for groundwater wells within the vicinity of Silver Springs, including public water system (PWS) wells, private drinking wells, and monitoring wells from multiple government agencies. The final dataset consisted of approximately 3,000 wells in the 574,000 acre springshed that had been sampled for nitrate within the last 13 years. We used recursive partitioning, applying the random forest technique to identify important predictive variables and construct decision trees. Nitrogen loading, recharge to the UFA, and depth to the UFA were determined to be important predictors of elevated nitrate concentration in the aquifer. To investigate temporal effects of evolving land use, we regressed imputed N loading, based on periodic land use mapping, against groundwater N concentrations. Wells demonstrating significant positive or negative nitrate trends were identified and additionally examined for influence of local land use. The current modeling framework provides a promising approach for describing karst areas, where the Floridan aquifer is minimally confined and there is a tight spatial and temporal link between land surface activities and aquifer contamination. |