Assessment of nutrient load estimation approaches for small urban streams in Durham, North Carolina
This cooperative study between the City of Durham Public Works Department, Stormwater Division and U.S. Geological Survey evaluated whether alternate monitoring strategies that incorporated samples collected across an increased range of streamflows would improve nutrient load estimates for Ellerbe and Sandy Creeks, two small, highly urbanized streams in the City of Durham, North Carolina. Water-quality and streamflow data collected between January 2009 and December 2020 were used to develop instream nutrient-load models using the U.S. Geological Survey R-LOADEST program. This study compared model results from two sampling scenarios: routine monthly (fixed frequency) sampling combined with targeted high-streamflow sampling (scenario A), and fixed frequency sampling only (scenario B). Calibration diagnostic results were used to select the final, or most optimal, models. Most final models included seasonality terms to compensate for intra-annual variability in the data. Storm-runoff samples provided better definition at higher streamflows and improved the overall concentration versus flow relations for all constituents, except nitrate + nitrite. Uncertainties in the nutrient load estimates were lower and less variable for the scenario A tests compared to the scenario B tests. Five time steps representing 12-, 9-, 7-, 6-, and 5-year subsets of the overall dataset were used to examine the effect of prediction period length on the computed loads and uncertainties. In focusing on the scenario A results, nutrient loads tended to be higher for the shorter time steps. These shorter time steps also produced higher errors, or uncertainty, in the load estimates compared to longer time steps. Evaluations of annual nutrient loads during 2016–20 indicated that the most consistent load estimates and tightest confidence intervals were obtained for longer 12- and 9-year time steps. Estimated loads were more variable and uncertain when based on the shorter 6- and 5-year time steps. The degree of uncertainty (standard error of prediction) in the nutrient load estimation results was influenced by sampling approach, calibration time step, and hydrologic characteristics during the model period of interest.