Use of cluster and principal component analyses to profile areas in California where ground water has been contaminated by pesticides

An empirical approach to profiling areas of ground water contamination by pesticides was devised that did not rely upon determining the level of vulnerability between land areas and that did not assume any particular pathway for ground water contamination. Climatic and soil data were obtained for 1-square mile sections of land in California where pesticide residues had been found in well water samples and the detection was attributed to legal agricultural applications. These sections were designated as known contaminated (KC) sections. Climate and soil data were also obtained for sections which lacked either well sampling data or a positive pesticide detection. These sections were designated as candidate sections. Statistical procedures were used to cluster groups of KC sections first with respect to climate characteristics and then with respect to soil characteristics. Principal components analysis (PCA) was used to construct a statistical profile of soil variables for each cluster of KC sections. A method based on the PCA was developed to compare the similarity of soil profiles derived for each KC section cluster to individual candidate sections. Since the profiling scheme was based only on data from KC sections, candidate sections that did not match any KC cluster profile could only be considered dissimilar to contaminated sections, receiving a status of not-classified. This profiling method is flexible and it can be revised to incorporate updated well sampling information.