Monitoring design for assessing compliance with numeric nutrient standards for rivers and streams using geospatial variables

Abstract

Elevated levels of nutrients in surface waters are among major human and environmental health concerns. Increases in nutrient concentrations in surface waters have been linked to urban and agricultural development of watersheds across the United States. Recent implementation of numeric nutrient standards in Colorado has prompted a need for greater understanding of human impacts on nutrient levels at different locations within a watershed and for how upstream influences affect the monitoring needs of specific locations. The objectives of this research are (i) to explore the variability of annual nutrient concentration medians under varying levels of upstream anthropogenic influences, (ii) to explore the variability of the standard deviation of nutrient concentrations under varying levels of upstream anthropogenic influences, and (iii) to develop a mathematical expression for approximating the number of samples required for estimating nutrient medians in the context of compliance with numeric standards. This analysis was performed in the Cache La Poudre (CLP) River watershed, which provides a gradient of anthropogenic influences ideal for studying water quality impacts. Multiple linear regression (MLR) models were used to explain the relationship of the median and lognormal standard deviation of nutrient concentrations in the CLP River, i.e., Total Kjeldahl Nitrogen (TKN), nitrate (NO3-N), total nitrogen (TN), and total phosphorous (TP) to upstream point and non-point sources of nutrients and general hydrologic descriptors. The number of samples required annually at monitoring locations is predicted based on an equation for determining sample size using relative error of a dataset which accounts for the difference between the median and standard for a lognormal population. MLR models for annual medians performed better for TN (R2 = 0.86) than TP (R2 = 0.90) despite high coefficients of multiple determination. Anthropogenic predictor variables, which characterize upstream urban and agricultural impacts on nutrient concentrations, were sufficient for describing variation of median concentrations between monitoring sites. A general hydrologic predictor was sufficient for characterizing variability of annual medians between years. The preferred MLR for all of the nutrient parameters uses inverse distance weighted WWTP and AFO capacities with annual mean daily discharge as a hydrologic predictor. The percent land use is equivalent to nutrient point source parameters (i.e., number of WWTPs and AFOs) for predicting median nitrogen concentrations in the watershed, though urban and agricultural land use predictors cannot be employed in the same model due to high multicollinearity. Little value is gained in the MLR models by including capacity of point sources in the predictive variables. For TP, a parameter which describes the variability of medians between years was not found, thus limiting the applicability of the model. The MLR models were less successful for predicting lognormal standard deviation of nutrients due to limited datasets. However, for robust datasets, high R2 values were found for TN and TP (0.80 and 0.73, respectively) based on anthropogenic predictors and annual rainfall. Overall, the MLR approach was appropriate for predicting median nutrient concentrations and lognormal standard deviations in the study watershed. Anthropogenic variables and general hydrologic descriptors were sufficient predictive parameters for the MLR models. Results of the application of an expression derived for predicting annual required samples indicate that sampling requirements to meet a 95% confidence level are lower than the current regulatory monthly sampling requirement. The required number of samples for reporting compliance at a 95% confidence level substantially varied among sampling sites depending on the difference between annual median of the nutrient of concern and its numeric standard. When the median is within 20% of the standard, the required number of samples rapidly increases from several samples per year to hundreds of samples per year. A comprehensive monitoring plan that targets sampling to sites near the standard with limited sampling elsewhere will optimize sampling resources and increase confidence level of the results.