Sample count, method of predicting tolerance limits, and confounding among covariates affect species sensitivity distributions derived from survey data.
Uložené v:
| Názov: | Sample count, method of predicting tolerance limits, and confounding among covariates affect species sensitivity distributions derived from survey data. |
|---|---|
| Autori: | Stemmler, Katlyn M.1 (AUTHOR) katlynmgardner@gmail.com, Hawkins, Charles P.1 (AUTHOR) chuck.hawkins@usu.edu |
| Zdroj: | Freshwater Science. Sep2025, Vol. 44 Issue 3, p301-317. 17p. |
| Predmety: | *CONFOUNDING variables, *SAMPLE size (Statistics), *WATER quality, *ENVIRONMENTAL exposure, *ENVIRONMENTAL databases, *ECOLOGICAL risk assessment, *ECOLOGICAL models |
| Abstrakt: | Species sensitivity distributions (SSDs) rank the sensitivities of species to a stressor and are used to set water-quality criteria protective of aquatic life. SSDs have been traditionally constructed from laboratory bioassays but are increasingly being derived from ecologically realistic survey data. For survey-derived SSDs, the sensitivity of each taxon is estimated as the stressor concentration above which the taxon is inferred to be nonviable—for example, the 95th percentile (XC95) of stressor values derived from a weighted cumulative frequency distribution (WCFD). These XC95 values are then used to estimate the hazardous concentration that a small percentage of taxa (e.g., 5%) cannot tolerate (HC05). However, at least 3 issues can potentially affect the robustness of survey-derived HC05 estimates: incomplete detection of taxa presence, the method used to estimate XC95 concentrations, and confounding of the stressor of interest with covariates. We used a combination of simulation, correlation, modeling, ordination, and propensity analyses applied to 2 large survey datasets (United States Environmental Protection Agency [USEPA] National Rivers and Streams Assessment and West Virginia [WV] Department of Environmental Protection) to assess the sensitivity of salinity SSDs to these issues. The simulation analysis showed that HC05 estimates declined 2-fold with increasing sample count: 408 (n = 50) to 198 (n = 10,000) μS/cm. Correlation analysis, random forest regression, and ordination analysis applied to the USEPA data showed that 32 to 66% of the spatial variation in specific conductivity at regional and national scales was associated with temperature, pH, nutrient concentrations, and substrate variables, indicating possible confounding. When applied to the spatially dense WV data, WCFD-based estimates of XC95 produced HC05 estimates 2.2-fold smaller than produced by general additive models (GAMs; 298 vs 658 μS/cm). When we used propensity analysis to control for confounding in the WV data, the adjusted WCFD-based HC05 value was slightly smaller than the unadjusted value (277 vs 298 μS/cm), but the adjusted GAM-based HC05 value was larger than the unadjusted value (807 vs 658 μS/cm). These results indicate that the performance of survey-derived SSDs is sensitive to methodological choices and that these SSDs must be used with caution when establishing regulatory criteria. [ABSTRACT FROM AUTHOR] |
| Databáza: | Academic Search Index |
Nájsť tento článok vo Web of Science | Abstrakt: | Species sensitivity distributions (SSDs) rank the sensitivities of species to a stressor and are used to set water-quality criteria protective of aquatic life. SSDs have been traditionally constructed from laboratory bioassays but are increasingly being derived from ecologically realistic survey data. For survey-derived SSDs, the sensitivity of each taxon is estimated as the stressor concentration above which the taxon is inferred to be nonviable—for example, the 95th percentile (XC95) of stressor values derived from a weighted cumulative frequency distribution (WCFD). These XC95 values are then used to estimate the hazardous concentration that a small percentage of taxa (e.g., 5%) cannot tolerate (HC05). However, at least 3 issues can potentially affect the robustness of survey-derived HC05 estimates: incomplete detection of taxa presence, the method used to estimate XC95 concentrations, and confounding of the stressor of interest with covariates. We used a combination of simulation, correlation, modeling, ordination, and propensity analyses applied to 2 large survey datasets (United States Environmental Protection Agency [USEPA] National Rivers and Streams Assessment and West Virginia [WV] Department of Environmental Protection) to assess the sensitivity of salinity SSDs to these issues. The simulation analysis showed that HC05 estimates declined 2-fold with increasing sample count: 408 (n = 50) to 198 (n = 10,000) μS/cm. Correlation analysis, random forest regression, and ordination analysis applied to the USEPA data showed that 32 to 66% of the spatial variation in specific conductivity at regional and national scales was associated with temperature, pH, nutrient concentrations, and substrate variables, indicating possible confounding. When applied to the spatially dense WV data, WCFD-based estimates of XC95 produced HC05 estimates 2.2-fold smaller than produced by general additive models (GAMs; 298 vs 658 μS/cm). When we used propensity analysis to control for confounding in the WV data, the adjusted WCFD-based HC05 value was slightly smaller than the unadjusted value (277 vs 298 μS/cm), but the adjusted GAM-based HC05 value was larger than the unadjusted value (807 vs 658 μS/cm). These results indicate that the performance of survey-derived SSDs is sensitive to methodological choices and that these SSDs must be used with caution when establishing regulatory criteria. [ABSTRACT FROM AUTHOR] |
|---|---|
| ISSN: | 21619549 |
| DOI: | 10.1086/737157 |