Developing a foundation for eco‐epidemiological assessment of aquatic ecological status over large geographic regions utilizing existing data resources and models

Abstract

Eco-epidemiological studies utilizing existing monitoring program data provide a cost-effective means to bridge the gap between the ecological status and chemical status of watersheds and to develop hypotheses of stressor attribution that can influence the design of higher-tier assessments and subsequent management. The present study describes the process of combining existing data and models to develop a robust starting point for eco-epidemiological analyses of watersheds over large geographic scales. Data resources from multiple federal and local agencies representing a range of biological, chemical, physical, toxicological, and other landscape factors across the state of Ohio, USA (2000-2007), were integrated with the National Hydrography Dataset Plus hydrologic model (US Environmental Protection Agency and US Geological Survey). A variety of variable reduction, selection, and optimization strategies were applied to develop eco-epidemiological data sets for fish and macroinvertebrate communities. The relative importance of landscape variables was compared across spatial scales (local catchment, watershed, near-stream) using conditional inference forests to determine the scales most relevant to variation in biological community condition. Conditional inference forest analysis applied to a holistic set of environmental variables yielded stressor-response hypotheses at the statewide and eco-regional levels. The analysis confirmed the dominant influence of state-level stressors such as physical habitat condition, while highlighting differences in predictive strength of other stressors based on ecoregional and land-use characteristics. This exercise lays the groundwork for subsequent work designed to move closer to causal inference.

Eco-epidemiology of aquatic ecosystems: Separating chemicals from multiple stressors

“…The mixture toxic pressure of pesticides (evaluated for 159 compounds) was highly correlated with a GIS-derived percentage agricultural land use. Further details on the outcomes of the preliminary analyses are provided by Kapo et al (2014).…”

“…Monitoring data were collected for Ohio (a state covering seven ecoregions) between 2000 and 2008 (Kapo et al, 2014), starting with 3000 sites, of which 1917 sites had data for all parameters. Mixture toxic pressures were determined for (separately) pesticides (n = 156), "down-the-drain" consumer product chemicals (n = 7, household product constituents), pharmaceuticals (n = 49), estrogens (n = 3) and conventional pollutants (n = 11, metals, ammonia, and nitrite), based on measured and modelled concentrations.…”

“…The Weight of Evidence/Logistic Regression (WOE/LR) method was applied here to investigate mixture impacts using the more recent Ohio (2000Ohio ( -2008 data set. The latter data set is a vast expansion of the Ohio (1990Ohio ( -1995 data set, examining both direct impacts to flowing water (i.e., local catchment) and for entire upstream areas (Kapo et al, 2014). As illustration, the differences between the older and the newer Ohio data set are: (1) number of variables, after quantifying mixture toxic pressures: from 18 to 32; (2) the numbers of chemicals considered in the mixture toxic pressure variables: from 11 to 226; (3) the numbers of sites, from 1572 to 3000; and (4) the number of sites with complete data from 695 to 1917.…”

“…These methods focus directly on ecosystems and species occurring in field conditions that are exposed to mixtures and multiple stresses under field conditions, while encompassing inter-species interactions. We apply a suite of methods in exposure and effects assessment, as described in the case studies, and earlier introduced by De Zwart et al (2006) and Kapo et al (2014).…”

“…The typical steps in data collection and management consist of the collection of monitoring data, the imputation of abiotic and biotic parameters, the expansion of the data set with additional landscape and land use data, adding information of scale (like proximity of land use to the water bodies), addition of mixture toxic pressure proxies (Kapo et al, 2014), and screening of the resulting data set. The mixture toxic pressures were based on measured or predicted concentrations (e.g., via the iSTREEM® model (www.iStreem.org)).…”