PCA, CCA, and ANN Modeling of Climate and Land-Use Effects on Stream Water Quality of Karst Watershed in Upper Green River, Kentucky

Abstract

Climate and land-use impacts on stream water quality in a karst watershed were studied to determine how they might be useful to model stream water quality parameter (SWQP) distribution from diffuse source pollution. This was achieved using principal component analysis (PCA), canonical correlation analysis (CCA), and artificial neural networks (ANNs) for the Upper Green River Watershed, Kentucky. Correlation analysis and eigenvalue analysis gave rise to what are known as effective input parameters for the subsequent ANN modeling. The results of ANN modeling indicated that although land-use contributions were better correlated when they were combined with temperature and precipitation for some of the SWQPs, most of the SWQP distribution trends were better predicted using the field data of temperature and precipitation at the watershed scale. This was in contrast to the findings of the many nonkarst watersheds where SWQPs showed a significant dependence on land use alone according to the literature. This, in essence, implies that the land-use influences are secondary, and precipitation and temperature affect the Upper Green River karst watershed’s SWQPs primarily, whereas for nonkarst watersheds the land use influences SWQPs primarily from diffuse source pollution.