CADDIS Volume 4: Data Analysis
- Controlling for
- Analyzing Trait Data
- Propensity Score
Species Sensitivity Distributions
- Where can I obtain species sensitivity distributions?
- How do I use species sensitivity distributions in causal analysis?
- Helpful tips
Authors: P. Shaw-Allen, G.W. Suter II
Species Sensitivity Distributions (SSDs)
Species sensitivity distributions (SSDs) are models of the variation in sensitivity of species to a particular stressor (Posthuma et al. 2002). SSDs are generated by fitting a statistical or empirical distribution function to the proportion of species affected as a function of stressor concentration or dose. Traditionally, SSDs are created using data from single-stressor laboratory toxicity tests, such as median lethal concentrations (LC50s; see Figure 1).
Typically, data from laboratory toxicity tests are used to develop SSDs. Two sources of toxicity data are U.S. EPA's ECOTOX database, which provides single chemical toxicity information for aquatic and terrestrial life exposed in the lab or under controlled field conditions, and the Environmental Residue-Effects Database , which provides data and plots associating body residues of single chemicals with responses of aquatic and terrestrial life exposed in the lab or under controlled field conditions.
Volume 3: Examples & Applications includes a gallery of generic SSDs for metals. These SSDs were generated using laboratory toxicity test results from ECOTOX. The same method can be used to generate SSDs for other chemicals. CADDIS's SSD Generator produces SSDs by fitting the most commonly applied distribution, the log-probit, to toxicity data. An alternative is the Dutch program ETX .
You can create a case-specific SSD by selecting data from the literature or a database that will generate a site-relevant model. For example, since pH and hardness are important determinants for the speciation and toxicity of metals, data selected for a metals SSD should have pH and hardness similar to the site in question. If tests have been performed with local water (e.g., to derive a water-effects ratio), their results may be added to the modeled data set. Note that in general, SSDs should not be derived from fewer than five species (van Leeuwen 1990).
SSDs can be used to generate predictions that may be confirmed by site data and to quantify stressor-response relationships. The interpretations discussed below, like any application of laboratory toxicity data to the field, depend on a reasonable concordance of physical, chemical, and biological conditions between the laboratory and field.
SSDs can be used to generate verified predictions using data from the case. SSDs reveal the relative sensitivities of species or other taxonomic categories. They may be used to generate predictions that certain taxa should be affected while others should be unaffected if a particular stressor is the cause of impairment. If an analysis of site data shows the predicted pattern of relative sensitivities, the prediction can be considered to be verified. Note that, to be considered a prediction, the relative sensitivities revealed by an SSD must be novel and the confirmatory data from the case must be prompted by the prediction. If the SSDs are simply consistent with the impairment (e.g., the impairment is low EPT taxa richness and the SSD shows that insects from those taxa are sensitive) no explicit prediction is verified by the data.
SSDs provide evidence for the stressor-response relationships from laboratory studies line of evidence. SSDs are useful for determining whether contamination and effects in an impaired community are consistent with those in laboratory toxicity tests. By comparing site data with an SSD, they can indicate whether potentially harmful concentrations of the chemical of interest occur at the site, the magnitude of effects expected to occur at those levels, and the certainty with which an assessor may apply this information. Like all other methods of relating laboratory effects to the field, SSDs must relate the nature and magnitude of the laboratory effects to field responses. Most SSDs are based on LC50 values (i.e., concentrations at which half of the organisms die in short-term exposures) whereas most biological surveys measure the presence and the relative abundance of different species. Species may not be observed if their probability of occurrence in a sample is very low, or if they have been locally extirpated. The relationship between 50% mortality and the probability of extirpation is related to the life history traits of a species. One species may be unable to withstand even relatively small impacts on survival, growth, or reproduction; a different species may be able to persist despite episodes of 50% mortality if its reproduction rate or immigration rate is sufficiently high. Still, many chemical exposure-response curves are steep, and chemical exposures in the environment may produce more severe effects than laboratory exposures because they may be sustained or recur. For these reasons, excursions above an LC50 may well result in extirpation and be reflected in biological survey results.
Biological observations at individual sites can be compared with SSDs by expressing the biological observations as the proportion of species that have been affected at the site. First, you compare the number of species at the site to the expected number of species at that site, given habitat characteristics or the number of species in local reference sites. This comparison generates an observed/expected proportion of species, which is comparable to the inverse of the proportion of affected species (Y-axis) values of an SSD. The value of the stressor variable at the site (i.e., the X-axis variable of the SSD), is then used to compare the observed biological response with the magnitude of response predicted by the SSD.
For example, the proportion of species affected by cadmium at site A (Figure 2) is closely predicted by the model indicating that the stressor-response relationship from laboratory tests supports the candidate cause. The proportion of species affected at sites B and E are higher than predicted by the SSD model. This evidence may weaken the case for the candidate cause as the sole cause of the impairment. The proportion of species affected at sites C and D are lower than that model would predict. These results may occur when a chemical's bioavailability is low (e.g., metal ions may be bound to suspended particles), when less toxic forms of the chemical occur at the site (e.g., trivalent versus hexavalent chromium), when populations adapt or acclimate to the stressor, or when adapted species replace sensitive species. Results similar to site C may suggest that the proportion of species affected may not be an appropriate measure of impairment, because the magnitude of the measured effect is small.
Although confidence intervals are informative, they do not include uncertainty due to extrapolating from the laboratory to the field or due to variability among communities, and so they should not be used to decide whether a point is close enough to the line to be in agreement. SSD comparisons with site data are most informative when high quality site data are available, site conditions are similar to laboratory conditions with respect to relevant physical and chemical variables, and the laboratory response is relevant to the field response.
An application of laboratory toxicity data to the field depends on a reasonable concordance of physical, chemical, and biological conditions in the laboratory and field. The data used to generate SSDs may not accurately represent toxic effects at a particular site and we recommend that an experienced aquatic ecotoxicologist help in interpreting the SSD models.
SSDs are most easily interpreted at their extremes. If the concentrations of a chemical at an impaired site are predicted by the SSD to affect all or nearly all species, the model is consistent with the candidate cause if impairment includes effects on many species. The case for the candidate cause is weakened if the concentrations of the chemical at the site where the impairment occurs are predicted to affect no species, or only a very few.
SSDs should be generated using data for the same type and level of effect.
The types of taxa selected for an SSD may span several taxonomic groups, fall within a single group (e.g., arthropods or salmonids), or share certain habitat preferences (e.g., coldwater or warmwater taxa). It also may be helpful to distinguish between early and late life stages of a species, as they frequently differ in sensitivity.
As with all analyses of secondary data, we recommend that primary sources be consulted to ensure the quality and relevance of influential data. Influential data include values that particularly influence the fitted function or that are important to the inference because they represent especially relevant site characteristics (e.g., affected taxa, stressor concentrations).