USING FUZZY C-MEANS CLUSTERING AND PARTIAL LEAST SQUARES REGRESSION TO EVALUATE SOURCES FOR GEOCHEMICAL CHANGES AND CHLORIDE ENRICHMENT IN A CALCAREOUS FEN LOCATED IN AN URBAN ENVIRONMENT

Fuzzy C-Means Clustering (FCM) and Partial Least Squares Regression (PLSR) are multivariate methods that have been used in a variety of geochemical and physical settings. FCM has been widely utilized to identify spatial or chemical patterns within a dataset. PLSR is used to create predictive models for variables, based on the response from other variables. These two methods were applied to a long-term water chemistry dataset (n = 763 samples) from a fen to identify potential sources of chemical changes, and to evaluate if those sources can be distinguished by their variances in chloride (Cl) with respect to other analytes.

Bluff Spring Fen (BSF) Nature Preserve is located in the Chicago suburb of Elgin, IL. The site contains calcareous fens and is bordered by Gifford Lake (GL) to the east and an underground mine and business park to the south. GL receives both clean construction debris and roadway runoff, while the mine extracts dolomitic bedrock and infiltrates mine water directly into BSF via an infiltration trench. Water-quality monitoring began at BSF in 2002 to evaluate site dynamics in response to mine development and operation, and adjacent site changes that could affect hydrology and geochemistry.

Five clusters were identified by FCM based on average water chemistry. Clusters 1, 2, and 5 were associated with land uses occurring outside of the preserve boundary. Cluster 1 was associated with GL and the eastern portion of the monitoring network, with increased sodium (Na) and Cl, which are attributed to the input of roadway runoff directly into GL. Cluster 2 was associated with the mine water discharge, with increased levels of strontium (Sr), fluoride (F), barium (Ba), and boron (B). Cluster 5 was associated with the central portion of BSF, which exhibited elevated levels of Na, Cl, Sr, and F suggesting mixing of waters both from the mine and GL.

The R² values of the PLSR models for Cl from Clusters 1 and 2 were 0.92 and 0.96, respectively. While both models exhibited accuracy in predicting Cl, the analytes and their predictive relationships to Cl were different, suggesting that changes in Cl observed in the fen can be traced to these distinct sources. Overall, this study has shown that FCM and PLSR can be used together to identify potential urban sources of chemical influence and to track observed changes in a complex geologic system.