North-Central Section - 50th Annual Meeting - 2016

Paper No. 14-4
Presentation Time: 2:35 PM


ROBERTSON, Dale1, HUBBARD, Laura1, LORENZ, David2 and DECICCO, Laura1, (1)Wisconsin Water Science Center, U.S. Geological Survey, 8505 Research Way, Middleton, WI 53562, (2)Minnesota Water Science Center, U.S. Geological Survey, 2270 Woodale Drive, Mounds View, WI 55112,

The Great Lakes receive nutrient and sediment inputs from many tributaries draining areas with a wide range in land cover and watershed characteristics. These inputs (loading) have caused eutrophication problems, such as algal blooms, to varying degrees and scales. Various statistical techniques have been used to quantify the loading of constituents in streams using continuously measured streamflow and, typically, very limited measurements of the specific constituent. To improve the load estimates without collecting extensive water samples in each tributary, a new load estimation technique was developed that extends the traditional load estimation models to include surrogates (continuously measured water-quality parameters such as turbidity, water temperature, specific conductance, pH, and dissolved oxygen) as potential explanatory variables in the regressions. This new continuous surrogate regression approach was used to estimate nutrient and sediment loading from 30 tributaries sampled as part of the Great Lakes Restoration Initiative. For each constituent, a single regression model was used for all sites. The final chosen models provided the “best” overall statistical fit, for each constituent, for the majority of the sites around the Great Lakes. Continuous sub-daily loads (5-minute estimates) and confidence intervals for each constituent were calculated using the U.S. Geological Survey LOADEST program, modified to include continuous surrogate variables and simulate sub-daily loads. The loads estimated with continuous surrogate data were then compared with loads estimated using traditional regression techniques (i.e., not including continuous surrogate data as explanatory variables). Based on reductions in residual variance, improved model fit, and reduced confidence intervals, the continuous surrogate regression approach improved load estimates for all constituents; however, largest improvements were found for constituents related to particulates in the water column, such as total phosphorus and suspended sediment. This new approach provides improved long-term load estimation and continuous estimates of short-term, sub-daily changes in nutrient and sediment loading.