Relationships among environmental factors and stream water ion yields of watersheds in the United States

Abstract

Understanding relationships among environmental factors and water quality constituents is very important for ion yield prediction, water quality management, and pollution prevention strategies. In this study, 13 stream water constituents - total dissolved solids, dissolved oxygen, pH, total nitrogen, orthophosphate, phosphorus, calcium, magnesium, sodium, potassium, sulfate, chloride, and bicarbonate from 47 watersheds in the United States were analyzed in relation to the environmental factors, including bedrock types (limestone, sandstone, and crystalline), land cover types, average annual precipitation, population density, watershed slope, and watershed area. Two regression analyses were applied, multiple linear regression using the "stepwise" method of model selection and principal component regression (PCR) followed by a prediction and validation process. Geographic information system (GIS) techniques were used to delineate watersheds, handle data, and examine the spatial distribution of the predicted ion yields. The results of this study indicated that there was a strong exponential relationship among environmental factors and selected stream water ion yields. The most important factor affecting stream water ion yields was average annual precipitation, which either appeared uniquely in the multiple regression models or appeared in company with other factors, accounting for 12 to 94 percent of the variance of most annual ion yields. Bedrock type was the second importance for stream water ion yields. Some average annual ion yields of limestone watersheds are 2 to 5 times larger than those of sandstone or crystalline watersheds, including total dissolved solids, bicarbonate, magnesium, and calcium. In addition, ion yields in limestone and sandstone watersheds are more closely related to the environmental factors than those in crystalline watersheds. Population density was the third most important factor, which accounted for 26 to 61 percent (in company with other factors) of the variance of ion yields of total dissolved solids (sandstone), sodium (limestone), magnesium (sandstone), potassium (limestone, sandstone), chloride (limestone), and bicarbonate (sandstone). Land cover was less important. The effects of land cover were significant in company with other factors and accounted for 15 to 45 percent of the variances in the predictions of ion yields in limestone and sandstone watersheds, including orthophosphate, phosphorus, magnesium, and sulfate (limestone watersheds), total dissolved solids, sodium, and bicarbonate (sandstone watersheds). Watershed slope was an ineffective predictor, significant only in limestone watersheds and accounted (in company with other factors) for 12 to 42 percent of the variance of ion yields of sodium, potassium, magnesium, and dissolved oxygen. Watershed area was not significant in any ion yield regression analysis. In this study, multiple linear regression (MLR) provided simpler models but larger R square, smaller standard error of estimate, and could detect more significant equations than principal component regression (PCR). The results of this study can be helpful to understand effects and importance of environmental factors on water quality, to predict average annual stream water ion yields, especially for watersheds where water chemistry data are not available, and to evaluate the explanation of the predicted results contributed by each environmental factor. The methods of this study may also be useful for those who want to develop their own models to predict water ion yields.