©2002 CRC Press LLC Structuring Population- Based Ecological Risk Assessments in a Dynamic Landscape Christopher E. Mackay, Jenee A. Colton, and Gary Bigham CONTENTS 11.1 Introduction 11.2 Ecological Risk Assessment Model 11.2.1 Risk Model Parameterization 11.2.1.1 Mercury Concentration in Fish P ([ C ] i | F i ) 11.2.1.2 Heron Exposure Rate ( P (ER)) 11.2.1.3 Spatial Function 11.2.1.4 Toxicity Response Function 11.3 Population-Based Risk Characterization 11.3.1 Population Modeling for the Great Blue Heron 11.3.2 Characterization of Population Dynamics 11.4 Discussion References 11.1 INTRODUCTION Ecological risk assessment is not a purely scientific endeavor. Rather, it is most commonly applied as an exercise in regulatory compliance intended to illustrate objectively demonstrable harm (or lack thereof) as the result of identified human activities. Legal guidance is limited to very broad directives requiring the protection of the environment from harm. Unfortunately, the definition of harm can be con- tentious. Risks to individual plants or animals are the easiest types of impacts to identify. However, within the context of providing protection to the environment, the magnitude of individual impacts may very well represent inconsequential 11 ©2002 CRC Press LLC events. Therefore, the characterization of harm may be misrepresented if limited solely to the individual. Simple and transparent (i.e., easy to understand and repli- cate) methods should be developed for application in a regulatory context to expand the scope of ecological risk assessments to the level of spatially defined subpopu- lations and populations, in order to determine whether an activity or activities in question actually represent an unacceptable risk within the legal context of envi- ronmental protection. This chapter examines a probabilistic method of population-level ecological risk assessment. It is intended for application within the regulatory context of a remedial investigation under the Comprehensive Environmental Response, Compensation and Liability Act (CERCLA). It describes a risk assessment in terms of impact on population dynamics. In this case, risk was modeled for great blue herons ( Ardea herodias ) exposed to methylmercury from fish taken as prey from an inland lake in the northeastern United States. Although the lake and its surrounding environs were used to parameterize both the risk model and the later-discussed heron population dynamics model, it was necessary to assume higher mercury concentrations in fish and other media than those measured in the lake to demonstrate certain attributes of both models. Therefore, the lake and the results of this assessment are hypothet- ical. This being the case, the general approach described herein is relevant to marine systems as well as freshwater systems. To describe the impact of this hypothetical exposure, two separate but intercon- nected models were developed. The first was a probabilistic risk assessment used to estimate the proportional population impact resulting from methylmercury exposure. The second was a population dynamics model to describe the risk estimates in terms of their impacts on the stability of the exposed heron subpopulation. The goal is to provide an approach that may be applied within a regulatory context to better illustrate the results of an ecological risk assessment in terms that may be quantita- tively applied to evaluate environmental protection. 11.2 ECOLOGICAL RISK ASSESSMENT MODEL The objective of the assessment component of this analysis was to express the impact of individual exposures to a toxic contaminant in terms of the risk to receptor populations. To determine the overall impact of all potential individual responses, it was necessary to quantify the probability of every possible response occurring within the exposed population. The paradigm used to characterize risk is the same one first proposed by the American Society for Testing and Materials (ASTM) 1 where risk ( r ) is expressed as the ratio of the exposure rate ( e ) to the expected exposure-dependent response ( T ): (11.1) In ecological risk assessment, the underlying assumption is that there is no risk of an adverse impact if the rate of exposure of a receptor is less than a defined r e T ϭ ©2002 CRC Press LLC response threshold. However, if the exposure exceeds the threshold, as indicated by an r value greater than 1, then a risk exists that the response ascribed to T will occur. If it is assumed that the risk to be characterized is the result of the exposure of a receptor to a toxic contaminant ( C ) that is found in an environmental medium F , then the risk paradigm can be expressed as a model where the risk of a receptor is estimated as the product of its exposure rate (ER) and the contaminant concentration in the medium of exposure ([ C ] F ), divided by the dose-dependent response threshold for C ( T C ). Hence, a generic risk model can then be expressed as follows: (11.2) If the risk is to be characterized for a group or population of individuals, then neither the exposure rate of the receptor, nor the contaminant concentration of the medium, nor the dose-dependent response of the receptor is an absolute value. Each variable parameter possesses a distribution of possible values within the exposed population, and from observations of the relative frequency of occurrence, a prob- ability function for risk can be discerned. The relation between a parameter value and its probability of occurring is referred to as a probability density function. To represent this, the variables must be expressed as probability density functions (generically denoted as D ( x ), where x represents the independent variable for the function), which is the integral of potential occurrences of all possible parameter values for the exposed population (generically denoted as P ( x ), where x represents the parameter for which the probability is expressed). Hence, the generic risk model can be expressed as follows: D ( r ) = D (ER) × D ([ C ] F ) × D ( T C ) –1 (11.3) Solving the probability function for risk ( D ( r )) can be accomplished either by convolution or faltung (denoted as D × D ) or by simulation using Monte Carlo techniques. When using Monte Carlo techniques, the probability of risk is determined using the pooled estimated probabilities associated with the parameters of the model over a range of potential exposure situations, A , for all members of the exposed population, N , as follows: (11.4) The risk is now described for all individuals of population N as a function of the probability for all concentrations of contaminant C within medium F , the prob- ability for all the possible exposure rates of the receptor to medium F within A simultaneous situations, and the probability of all possible dose-dependent responses to C by the receptors. Hence, the risk function is now defined as a probability density function D ( r ) for the characterization of risk. r ER C[] F ϫ T C ϭ Dr() P ER a C[] F () a ϫ T C a 1ϭ A Α nd n 1Ϫ N Ύ ϭ ©2002 CRC Press LLC 11.2.1 R ISK M ODEL P ARAMETERIZATION The generic risk model derived above was parameterized to predict the risk to a population of great blue herons exposed to methylmercury in fish from a lake. Most of the eastern and southern shore of the lake has been developed for either urban or suburban uses. The western shore remains largely undeveloped and provides habitat to numerous avian and mammalian species. Its shallow sloping banks and moderate bank cover make it excellent heron foraging habitat, and herons are commonly observed during the warm-weather seasons. The exposure received by a population of receptors is not simply related to the distribution of the contaminant ( C ) in the entire medium, but rather to the concen- tration of C in the constituents of medium F that the receptors contact directly. In situations where such distinctions may be made within a medium, the probability for exposure to any concentration C can now be made dependent upon the probability of exposure to a subcomponent of F ( F i of known [ C ]) across the range of all possible exposures ( F i to F i ) as follows: P ([ C ]) = P ([ C ] i | F i ) a (11.5) The relationship P ([ C ] i | F i ) is the probability of encountering a specific con- taminant concentration [ C ]; and is based on the probability of the exposure of the receptor to an identifiable subcomponent of the medium F i . (See Chapter 4 for further explanation of such conditional probabilities.) This may represent empirical distributions such as time spent in a specific location, or may be used to distinguish between the probability of ingesting specific prey items. Methods for the derivation of these probability density functions can be found in Efron and Tibshirani. 2 11.2.1.1 Mercury Concentration in Fish P ([ C ] i | F i ) The modeled medium of exposure ( F ) for the great blue heron was fish. The distribution of mercury concentrations used in the risk model was based on the likelihood observed for the distribution of mercury concentrations in individual fish ( F i ) from the lake, and the likelihood that a fish would be preyed upon by a heron. Likelihood, in this context, is defined as a past probability (i.e., observed) based on reported distributions. The mercury distribution was determined empirically from data collected from the lake in 1992. The probability that a sampled fish, containing a known concen- tration of mercury ( P (Hg)) would be prey for the great blue heron was determined as the product of the likelihood ( L ) of the heron selecting that size of fish (prey), and the likelihood of that species of fish being available from the lake (available). Therefore, the probability that any individual heron would ingest a fish represented by size and species by a specific sampled fish can be expressed as follows: P ([Hg]| F ) = ([Hg]| L (prey) × L (available)) (11.6) ©2002 CRC Press LLC The likelihood that a fish would be prey for the heron was determined from empirical observations reported by Alexander. 3 The great blue heron’s predominant prey is fish ranging from 3 to 33 cm in length. Proportional dietary content based on fish size was reported from the survey to be 8, 40, and 52% for fish 3 to 7, 7.1 to 14, and 14.1 to 33 cm, respectively. The sampled fish were ranked according to size. Sampled fish outside the range of 3 to 33 cm were excluded. The remainders were classed into three cohorts based on the above size ranges and L (prey) for each sampled fish was then determined as follows: (11.7) P c is the proportion that the cohort represents in the heron’s diet, and n c is the number of sampled fish in that cohort. The likelihood that a sampled fish is available as prey for the great blue heron is dependent on the abundance of that species in the lake. Because the fish samples were not random with regard to fish species, abundance within the sample cannot be assumed to be representative. However, overall abundance statistics were available from lake surveys. Therefore, the L (available) for any fish in the sampled group was deemed proportional to the abundance of that species, within each size cohort, throughout the entire lake ( A t ). Only fish species typically available to the heron were considered. Deepwater species that are not available for prey were excluded, as were fish determined to be either too large or too small to constitute heron prey. To control for bias in the sample due to disproportional representations of fish species in the sample set, the L (available) was made inversely proportional to the species abundance within the cohort ( A c ). Therefore, L (available) can be mathematically defined as follows: (11.8) Assuming that size selection by the heron was independent of species abundance in the lake, the product of these two likelihoods defines the probability of selection (P(F i )). Therefore, the probability of the heron’s exposure to a given concentration of mercury could be derived as follows: (11.9) The estimate of the probability density function across all potential prey fish (D([Hg] i |F i )) was determined by bootstrapping (with replacement) the sampled mer- cury data using the individual P(F i ) values as the metric of probability for selection. Each mercury observation was assigned a probability of occurrence based on the above likelihood. The mercury concentrations were then selected with replication L prey() P c n c ϭ L available() A t A c ϭ P Hg[]F i ()Hg[] P c A t ϫ N c A c ϫ i ϭ ©2002 CRC Press LLC based on the probability assigned to derive a probability distribution of potential exposures. A more detailed discussion of this method is available in Chapter 24 of Efron and Tibshirani. 2 The frequency of selection was tracked and used to derive the probability density function (D([Hg] i |F i ), illustrated in Figure 11.1. 11.2.1.2 Heron Exposure Rate (P(ER)) The dietary intake rate (IR) for the great blue heron may also be described as a probability density function. Unfortunately, there is rarely a sufficient record of empirical observations to develop an adequate distribution for this parameter directly. However, since the dietary requirements of the heron are related to its energy demands, it is possible to model the dietary requirements based upon a metric for which adequate distributions are available. Kushlan 4 developed an allometric equa- tion, specifically for wading birds, by regressing a series of observed dietary intake rates against the paired body masses (BW) for the birds. An estimation for the distribution of body masses for the great blue heron population has been developed by Henning et al. 5 By aggregating reported data on the body masses of great blue herons from the northeastern United States, it was found that the distribution of this parameter conformed to a normal distribution with an average mass ( BW ) of 2300 g with standard deviation (s BW ) of 670 g, and a minimum and maximum body mass of 1600 and 3000 g, respectively. The model was truncated to disallow values greater than or less than the minimum and maximum parameters. The variance used remained unchanged. By substituting the allometric relationship, the exposure rate function was parameterized based on the distribution in the heron’s body mass as follows: (11.10) FIGURE 11.1 Reverse cumulative probability density function for mercury exposure con- centrations for the great blue heron from lake fish. IR BW()0.00925 BW 1.64Ϫ ϫ gDW gBW dayи ϭ D IR() D 0.00925 N BW s BW ,() 1.64Ϫ ϫ()ϭ ©2002 CRC Press LLC 11.2.1.3 Spatial Function The great blue herons found in this area are migratory. Although they do nest in this region, they winter in the lower Mississippi Valley, the Gulf Coast, and the Southern Atlantic seaboard. 6 Therefore, herons are not present on the lake for a large part of a year. Within its breeding range, the great blue heron may be either colonial or solitary, depending on its location and situation. Herons in this area tend to be solitary nesters, and will establish and defend a nesting territory. 6 Great blue herons are most likely to be found hunting near their nesting sites, but may range as far as 24 km during daily feeding forays. 7,8 Because there are two considerations affecting the foraging behavior of the exposed heron population, one for migration and one for local foraging use, two spatial functions had to be developed to control for the heron’s feed locations. The first function was modeled as a temporal parameter (D(t)) that was used to describe the heron’s location in its migratory cycle where P(t) = 1 represents 100% residence around the lake, and P(t) = 0 represents 100% residence somewhere else along the migration route. The second distribution was a spatially explicit parameter (D(a)) that was used to describe the probable foraging patterns for the great blue heron subpopulation that relies on the lake as part of its food source while in residence around it. This also was parameterized in a similar manner where P(a) = 1 repre- sented complete dietary reliance on the lake and P(a) = 0 represented complete dietary reliance on other locations within the defined foraging range. The generic risk model was therefore structured as follows: (11.11) The temporal probability, P(t), was parameterized based upon regional observa- tions. Observations indicate that herons arrive in the area around the lake between days 46 and 90, and depart on winter migration between days 258 and 273. With lack of data to the contrary, it was assumed that the dates of arrival and departure of any given individual were independent. For illustration purposes, the distributions for both arrival and departure dates were assumed to be represented by a skewed triangular distribution with the mode defined at the earliest and latest 10th and 90th percentiles for arrival and departure, respectively. By solving for the residency time (departure date minus arrival date) using Monte Carlo techniques, the resulting residence time was found best to fit a beta distribution of alpha 38.72, beta 3.93, scaled to 227 days and truncated at 168 days (Figure 11.2). The habitat use function, D(a), was modeled based on the great blue heron’s bioenergetics. Flight to and from any location within the foraging area would require an energy output proportional to its distance from the nest or colonial roost. Asso- ciated with this expenditure is also a loss in potential foraging time equivalent to the time en route. This relationship can be expressed as follows: Dr() Pt() Pa() t P ER i C[] F () i ϫ() ta i 1ϭ I Α ϫ a 0ϭ A Α ϫ t 0ϭ T Α nd n 1Ϫ N Ύ DT C () ϭ ©2002 CRC Press LLC ⌬ E (x, y) = (W (x, y) – W (0, 0) ) × (t (0, 0) – t Flight ) – (W Flight × t Flight ) (11.12) The net benefit to the heron by foraging at any given location (denoted as (x, y)) is expressed as the net energy availability at (x, y), ( ⌬ E (x, y) ), and is proportional to the total power (i.e., energy per unit time foraging) that may be derived at the location (W (x, y) ) relative to the power derived if the heron had not traveled, but foraged in the immediate vicinity of the roost or nest (W (0, 0) ). The amount of time available to the heron to forage at location (x, y) is equal to the amount of foraging time available at location (0, 0), minus the time necessary to commute to and from (x, y) (t Flight ). It is also necessary to consider the energy expended in commuting between (0, 0) FIGURE 11.2 Temporal estimate of heron residence in the vicinity of the lake. ©2002 CRC Press LLC and (x, y). This is determined as the product of the power requirement for flight (W Flight ) and the duration of the commute, t Flight . This is subtracted from the net energy difference between (0, 0) and (x, y) to derive the net energy available. Since the ultimate goal is to develop a probability density function based on the relative benefit of one location over another, another energy term is defined (E H ) that relates the net energy benefit at any given point (x, y) relative to that at (0, 0) (E (0, 0) ). The probability density function may now be expressed as a proportional function of the relative energy availability at any point (x, y) with its differential radial distance from (0, 0) (dr) as follows: (11.13) The distribution of this function is illustrated in Figure 11.3. The probability density function D(E H ) is now a spatially explicit metric that describes the probability of a heron being present at any point (x, y), based solely on the bioenergetic advantage relative to location (0, 0). This may now be applied to a measure of habitat quality at (x, y) relative to (0, 0) to predict the likelihood of a heron’s presence based on the overall advantage that a heron would derive by foraging at that location. Habitat quality is a site-specific parameter and dependent upon estimates of the quantity of available forage fish and the quality of the local environment as adequate foraging habitat. The great blue heron is a wading bird that can utilize a variety of freshwater and marine habitats. It is found in areas of shallow water that have firm substrates and high concentrations of small fish. 6 Great blue herons forage in lakes, rivers, brackish marshes, lagoons, coastal wetlands, tidal flats, and sandbars, as well as wet meadows and pastures. 4,6 For the purposes of this assessment, potential heron habitat was defined as any shoreline or riverbank within 24 km of the lake, with a FIGURE 11.3 The function of the relative energetic benefit EH (x, y) with distance from the origin (0, 0). E (0, 0) and E (x, y) are assessed as uniform within the radius of 24 km. DE H () E ⌬ xy,() E 00,() rd r = 0 23 km Ύ ϭ ©2002 CRC Press LLC wading depth less than 50 cm for any water body whose minimum dimension was greater than 2 m. Habitat quality was also assumed to be proportional to relative habitat density. Habitat within urban areas was deemed unsuitable. Prey availability was determined from available state surveys. For habitats where survey results were unavailable, prey abundance was estimated based on comparisons with similar sur- veyed water bodies based on size and location. To determine the relative habitat quality, all potential foraging locations were identified and mapped relative to location (0, 0) to a radius of 24 km. These were then grouped into octants and segmented into 1 / 20 ths of the total foraging radius (Figure 11.4). Habitat quality (H (x, y) ) was expressed as the product of the density of appropriate habitat (d H(x, y) ) within the segment and the average prey abundance (a (x, y) ) relative to the prey abundance at point (0, 0) (a (0, 0) ) as follows: (11.14) The distribution of habitat density relative to the requirements of the great blue heron is provided in Figure 11.5. The bioenergetics model describes the probability of a heron being at location (x, y) based on the potential advantage of commuting from (0, 0) to (x, y). The spatial habitat quality model describes the potential advantage of a heron being at location (x, y) based on the availability of prey and the quality of the habitat. When these two are combined as follows, the product is a measure of probability that describes the likelihood of a heron foraging at any location within the prescribed foraging range (P(q (x, y) )): FIGURE 11.4 Spatial distribution of great blue heron habitat in the vicinity of the lake. Circle represents the 24-km radius assumed for potential habitat use. Shaded areas represent urban regions excluded as potential habitat. Segments represent octants segmented radially in 1 / 20 th of the total radius. H xy,() d Hxy,() d H 00,() a xy,() a 00,() ϫϭ [...]... Environmental Conservation, Albany, 1995 11 U.S EPA, Mercury Study Report to Congress, U.S EPA-452-R-9 6-0 01c, U.S Environmental Protection Agency, Washington, D.C., 1996 12 Chlorine Institute, Environmental Fate and Toxicity of Mercury, Final Report, Alliance 5-0 5 1-0 01, The Chlorine Institute, Washington, D.C., 1992 13 Eisler, R., Mercury Hazards to Fish, Wildlife, and Invertebrates: A Synoptic Review,... greater than 1, and thus was assumed to have received an exposure sufficient to incur adverse toxicological effects At this stage of the risk characterization, the risk was evaluated separately for reproductive effects and lethality end points (Figure 11. 8) The evaluation indicated that 9.4% of the population would be at risk for reproductive effects, and 4.4% of the same population would be at risk for mortality... continues, risk managers can expect better evaluations of threats and hazards to wildlife in terms that are easier to conceptualize and that can be quantitatively applied directly to management goals These improvements in risk communication will allow direct management at the population level, which in turn will permit risk managers to better allocate limited resources to environmental protection and restoration... found that a reproductive failure rate of 11. 8% (which corresponds to a lethal impact of 5.3%) would result in the exposed subpopulation no longer being self-sustaining and therefore requiring immigration from the regional non-exposed subpopulation in order to fulfill the carrying capacity (Figure 11. 12) Distributions for these estimates are illustrated in Figure 11. 13 When the model was executed in an... potential impact determined on 60-year runs to ensure steady state Evaluation of emigration (Equation 21) was determined simultaneously within each of the timestep determinations Runs were started with the assumptions that P1 equaled Kp and that F1 and F2 equaled 0 Equilibrium between the cohorts was achieved within five time steps Results from the probabilistic risk assessment were applied directly... on their relative sizes This was modeled as follows: ©2002 CRC Press LLC X (t Ϫ 1) X ( t ) ϭ X ( t Ϫ 1 ) ϫ ( 1 Ϫ M x ) Ϫ I L ϫ - ; where X ʦ { P 1, F 1, F 2 } (11. 23) Α X (t Ϫ 1) 11. 3.2 CHARACTERIZATION OF POPULATION DYNAMICS Figure 11. 11 illustrates the comparative impact on recruitment productivity for the exposed subpopulation of great blue heron as a result of exposure to mercury... preference and motivation to forage Since methylmercury is a cumulative toxicant, and the aberrant behaviors manifested would be expected to affect individual survival in the wild, this was assumed to represent a lethal concentration Based on an intake rate of 0.97 kg fresh weight/kg body weight/day (as reported by Bouton16), a lethal TRV of 0.098 mg/kg body weight was derived 11. 3 POPULATION-BASED RISK. .. could not sustain the P1 cohort at Kp, and therefore resulted in population declines over time Relative effects at various levels of impact on the exposed subpopulation are illustrated in Figure 11. 14 Results from the closed-model analyses indicate that there is no threat of extinction, so long as the proportional impact is less than 11. 8% reproductive failure and, correspondingly, 5.3% lethality The... probabilistic risk assessment may be evaluated for a specific species The great blue heron possesses a relatively low reproductive rate However, it is relatively longlived and highly mobile, with populations consisting of a large number of interacting subpopulations In the example provided here, it was demonstrated that for the great blue heron, an impact greater than 11. 8% reproductive failure and 5.3%... and the results tracked to develop the probability density function for area use by the heron subpopulation relative to the proportion of the lake to which they will be exposed The algorithm used was as follows: 1 D(a) ϭ Ύ ( A Site P ( q ) ) d A Site (11. 16) A Site = 0 The probability density function for this relation is illustrated in Figure 11. 7 and was used as the distribution for P(a) in the risk . Population- Based Ecological Risk Assessments in a Dynamic Landscape Christopher E. Mackay, Jenee A. Colton, and Gary Bigham CONTENTS 11. 1 Introduction 11. 2 Ecological Risk Assessment Model 11. 2.1. Function 11. 3 Population-Based Risk Characterization 11. 3.1 Population Modeling for the Great Blue Heron 11. 3.2 Characterization of Population Dynamics 11. 4 Discussion References 11. 1 INTRODUCTION . an ecological risk assessment in terms that may be quantita- tively applied to evaluate environmental protection. 11. 2 ECOLOGICAL RISK ASSESSMENT MODEL The objective of the assessment component