APPLYING DATA MINING TECHNIQUES TO MODEL HYDRAULIC HEAD IN THE EDWARDS AQUIFER

As climatic conditions change, observation wells in the highly heterogeneous Edwards Aquifer respond with varying degrees of hydraulic head fluctuations. The complex hydrogeology of karst aquifers impedes hydraulic head prediction accuracy. To increase hydraulic head prediction accuracy, this research applies the data mining techniques of Hierarchical Clustering and Artificial Neural Network (ANN) to detect and model emerging patterns under various hydrologic conditions, using the Palmer Hydrologic Drought Index as a guide. The clusters identified during the Hierarchical Clustering analysis are used as input to the ANN, thus providing highly representative data for prediction.

The results of this research have many possible applications. Missing hydraulic head values for observation wells can be filled in with a high degree of accuracy. In addition, for locations where is there is no data, hydraulic head values can be spatially predicted based on the known relationships between observation wells which respond similarly under various hydrologic conditions. Further, this methodology allows exploration of the presence or absence of hydrogeological constraints near an observation well by using synthetic time-series which contains information about the constraint being modeled. These relationships can be confirmed by inspecting hydraulic head time-series which contains information about the Edwards Aquifer hydrogeology that is understand. Combining these applications, it is possible to build a spatial-temporal, three-dimensional model of the Edwards Aquifer hydraulic head.