Paul A. Conrads¹ and Edwin A. Roehl, Jr.^2
1USGS South Carolina Water Science Center, Columbia, SC, USA
²Advanced Data Mining, LLC Greenville, SC, USA

To develop accurate empirical models of water levels in the Everglades, it is necessary to group continuous gaging stations that exhibit similar hydrologic responses. Artificial neural networks (ANN) are a non-linear multivariate empirical modeling technique that has produced good results in other hydrologic systems. Given the behavioral discontinuities of various “compartments” in the Everglades, and that ANN models are generally ill suited to synthesizing discontinuous functions, a divide-and-conquer approach was needed to the segment disparate behaviors. A time-series clustering algorithm was applied to a 5-year dataset of water-level hydrographs of the Everglades Depth Estimation Network (EDEN) data to subdivide data into classes having similar behaviors. The hydrographs of all the stations were cross-correlated to produce a matrix of Pearson coefficients. Each row and column represented a different gaging station and its behavioral similarity to each of the other gaging stations. The rows were then clustered using the k-means algorithm. The number of classes was determined by the sensitivity of the mean square error to k. Inspection showed that the members of a class were similar, and dissimilar to those in other classes. Not surprisingly, there were gradations of similarity class-to-class. An important side benefit of time series clustering is that it identifies redundant data, largely answering the questions of, “Which monitoring stations should be moved and where should they be moved to?” Cascaded ANN sub-models were developed for each class to optimally model behaviors that evolve on different time scales.

Contact Information: Paul A. Conrads, USGS South Carolina Water Science Center, Stephenson Center – Suite 129, 720 Gracern Road, Columbia SC, 29210 USA, Phone: 803.750.6140, Fax: 803.750.6181, Email: pconrads@usgs.gov

(This abstract is from the 2006 Greater Everglades Ecosystem Restoration Conference.)

Back to Project Homepage