You are here:
Adaptation of a Weighted Regression Approach to Evaluate Water Quality Trends in an Estuary
Citation:
Beck, M. AND Jim Hagy. Adaptation of a Weighted Regression Approach to Evaluate Water Quality Trends in an Estuary. CERF 2015, Portland, OR, November 08 - 12, 2015.
Impact/Purpose:
This abstract is for an invited presentation in the 'making data work' session at CERF 2015.
Description:
To improve the description of long-term changes in water quality, we adapted a weighted regression approach to analyze a long-term water quality dataset from Tampa Bay, Florida. The weighted regression approach, originally developed to resolve pollutant transport trends in rivers, allows model coefficients to vary between water quality and explanatory variables and can more clearly resolve the trends attributable to multiple drivers of ecosystem response. The model resolved changes in chlorophyll-a (chl-a) time series collected from 1974 to 2012 at seasonal and multi-annual time scales while considering variation associated with changes in freshwater influence. Separate models were developed for each of 4 Bay segments to evaluate spatial differences in patterns of long-term change. Observed trends in chl-a coincided with a decrease in nitrogen loading to Tampa Bay since the 1970s. Although median chl-a has remained constant in recent decades, model predictions indicated that variation has increased for upper Bay segments and that low biomass events occur less often in the lower Bay segment. These previously undocumented changes illustrate the utility of the model that may suggest recent changes in eutrophication status that warrant further evaluation. The dynamic relationships observed between chl-a and freshwater inputs also suggested that the environmental drivers of primary production varied across the time series, which is consistent with historical shifts in point versus non-point sources of pollution and effects of extreme flow events. As such, results from our analyses provide additional resolution of water quality trends in Tampa Bay that has not been possible with traditional, fixed coefficient, regression models. The approach could easily be applied to other systems with long-term datasets and we are currently developing a software package for the R statistical platform.