Progress Report of the Ecological Committee on FIFRA Risk Assessment Methods:
VII. Terrestrial Risk Assessment
On this Page
- Hypothetical Probabilistic Risk Assessment
- Multispecies Risk Assessment
- Table 1: Uncertainty Associated with Laboratory Toxicity Tests
- Table 2: Uncertainty Associated with Extrapolations
The Ecological Committee on FIFRA Risk Assessment Methods (ECOFRAM) was formed in June 1997. The Committee's purpose is to develop tools and processes within the FIFRA framework for predicting the magnitude and probabilities of adverse effects to non-target aquatic and terrestrial species resulting from the introduction of pesticides into their environment. A Terrestrial Work Group was formed to identify and discuss probabilistic methods for terrestrial assessments and develop recommendations for future use by EPA. In addition, they will identify information that must be developed in order to validate the proposed methods so as to ensure that the proposed assessment process, if adopted by EPA, supports environmental decisions that are scientifically defensible. A report of the Work Group's deliberations to date will be presented.
The Terrestrial Work Group is organized into 2 sub-groups: Terrestrial Exposure and Terrestrial Effects. The sub-groups have met together since their inception, and have approached terrestrial risk assessment in an integrated fashion. Each sub-group has developed a poster presentation for this meeting that highlights its activities. In this joint presentation, we provide a brief overview of four considerations:
- Uncertainties associated with laboratory toxicity tests
- Uncertainties associated with extrapolations to other populations of the same species, other species, other classes of vertebrates, and population- or community-level effects,
- An example of a probabilistic risk assessment for a hypothetical granular pesticide ingested by birds, and
- A multi-species approach for probabilistic risk assessment.
As a first step toward developing probabilistic risk assessments, we distinguished uncertainty associated with toxicity testing (Table 1) from uncertainty associated with extrapolations (Table 2). In both cases, we categorized the uncertainty into one of three major classes:
- Natural Stochasticity--temporal and spatial variation in environmental characteristics that affect the response of individuals, populations, or communities to human intervention;
- Parameter Error--imprecise measurements of parameters used in assessment models
- Model Error--incorrect specification of models (e.g., inappropriate variables, functional forms, or boundaries).
In addition, we commented on our ability to quantify or reduce each of the kinds of identified uncertainty.
HYPOTHETICAL PROBABILISTIC RISK ASSESSMENT
We are investigating the feasibility of developing a modular approach to probabilistic risk assessment, executable in a standard software package such as Crystal Ball or @Risk. A hypothetical example of a granular pesticide and its effect on a non-target bird species illustrates the basic principles of our approach (Figure 1). The general model has as its ecological endpoint, the probability of death to individual birds, which derives from a dose-response curve. In the general model, estimates of dose are dependent on three distributions:
- Body weights of birds (Normal with estimated parameters; (Figure 2))
- Sizes of granules (Empirical density function; ( Figure 2))
- Percent active ingredients (Assumed to be uniform)
In addition, the model assumes a Log-logistic relationship for species sensitivities (Figure 2) and a Normal distribution for slope estimates (Figure 2). In practice, any kind of distribution (e.g., theoretical, empirical) could be used as input to the model.
The output from the model is a distribution: it assigns to each probability of death a proportion (i.e., the percent of the population experiencing that likelihood of death). From that distribution, we could assess what proportion of the population has at least a 90%, 75%, or 50% likelihood of dying as a result of uptake of the pesticide (Figure 3). Moreover, we could examine the degree to which that distribution is sensitive to various parameters in the model via sensitivity analysis (Figure 4). In essence, it identifies the parameters (or uncertainties) of the model with the greatest effect on the likelihood of death. In the hypothetical example, species sensitivity has the greatest impact on estimates of risk, whereas bird weight has the least effect. Granule weight and probit slope of the LD50 relationship have intermediate effects on the distribution of risk.
MULTISPECIES RISK ASSESSMENT
We are investigating an approach to risk assessment that uses all relevant single- species toxicity data, and illustrate it with a hypothetical example for a granular pesticide (Figure 5). In particular, we illustrate the relationship between rank-percentile of species sensitivity (14 black circles) and the log of the number of granules per LD50 per bird. Points on the best-fit linear regression (solid line) provide estimates of the proportion of species protected at a given level of dose (light gray circle, 10% of species protected; dark gray circle, 50% of species protected). Comparisons involving the consumption curve for all species (bold dashed line) or the curve for the worse case scenario (species with the greatest consumption rate; dashed line) provide an additional perspective on risk.