Application of Multivariate Statistical Methodology to Model Factors Influencing Fate and Transport of Fecal Pollution in Surface Waters

Abstract

The increasing number of polluted watersheds and water bodies with total maximum daily loads (TMDLs) has resulted in increased research to find methods that effectively and universally identify fecal pollution sources. A fundamental requirement to identify such methods is understanding the microbial and chemical processes that influence fate and transport of fecal indicators from various sources to receiving streams. Using the Watauga River watershed in northeast Tennessee as a model to better understand these processes, multivariate statistical analyses were conducted on data collected from four creeks that have or are expected to have pathogen TMDLs. The application of canonical correlation and discriminant analyses revealed spatial and temporal variability in the microbial and chemical parameters influencing water quality, suggesting that these creeks differ in terms of the nature and extent of fecal pollution. The identification of creeks within a watershed that have similar sources of fecal pollution using this data analysis approach could change prioritization of best management practices selection and placement. Furthermore, this suggests that TMDL development may require multiyear and multisite data using a targeted sampling approach instead of a 30-d geometric mean in large, complex watersheds. This technique may facilitate the choice between watershed TMDLs and single segment or stream TMDLs.

Canonical Variable Selection for Ecological Modeling of Fecal Indicators

“…Canonical correlation analysis has been used previously to identify factors influencing fate and transport of fecal pollution, and this approach can identify dominant trends within heterogeneous watersheds (Hall et al, 2014). Continuous monitoring of this system for more than 10 yr (Hall et al, 2014, unpublished results) has established that alkalinity, hardness, and BOD 5 are typically low with low variability during dry periods and areas are not influenced by urban and industrial point sources. This suggests that variation in these parameters is influenced by runoff entering Sinking Creek.…”

“…Better understanding about the integration and interaction of FIOs and enteric pathogens in microbial communities can aide in understanding sources and community interactions that influence fate and can provide insight to improve the efficacy of fecal indicators to predict human health risks (Cloutier et al, 2015;Crowther et al, 2003). Canonical correlation analysis has been used previously to identify factors influencing fate and transport of fecal pollution, and this approach can identify dominant trends within heterogeneous watersheds (Hall et al, 2014). Continuous monitoring of this system for more than 10 yr (Hall et al, 2014, unpublished results) has established that alkalinity, hardness, and BOD 5 are typically low with low variability during dry periods and areas are not influenced by urban and industrial point sources.…”

“…Although point sources, leaking wastewater lines, and failed septic tanks are possible sources, our efforts and efforts by the city wastewater department have not identified any large failures that would explain the trends in these water quality parameters. Previous use of multivariate statistical models (Hall et al, 2014) to identify clusters of sites with similar water quality trends suggested that runoff is a large input. This suggestion is also supported by the local geology, terrain, and soil type.…”

Development of Multiple Regression Models to Predict Sources of Fecal Pollution

“…This result suggests that prior determination of the chemical and microbial water quality variables that are most associated with degraded water quality as they pertain to land use patterns in one stream are similar to those variables contributing to degraded water quality throughout the entire watershed. This result highlights the combined usefulness of multivariate statistical analyses such as canonical discriminant and multiple regression analyses (Hall et al, 2014).…”

“…The shortcomings of conventional methods of source identification suggest that alternative water quality monitoring program design and data analysis methods are needed to protect surface water resources. This study's authors' previous research suggests that the use of multivariate statistical analyses may improve understanding the influence of spatial and temporal variability on fecal pollution (Hall et al, 2014). This approach for identifying the water quality variables associated with fecal pollution from specific sources may be more successful at predicting water quality than more common data analysis methods, including multiple regression analysis.…”