© 2002 by CRC Press LLC CHAPTER 15 Conclusions and Recommendations Robert A. Pastorok and Lev. R. Ginzburg Population, ecosystem, and landscape models are generally mechanistic models that predict eco- logical state variables by using mathematical relationships to represent components and processes in environmental systems. In many cases, the state variables estimated by these models are relevant endpoints for ecological risk assessment (e.g., organism abundance or biomass, recruitment, or population growth rate). Thus, these models are directly applicable in a risk characterization for assessing the significance of estimated risks initially expressed in terms of individual-level end - points. In this mode, ecological models aid in translating risks for individual-level endpoints to more relevant endpoints at higher levels of biological organization. In contrast with population and higher-level models, toxicity-extrapolation models are generally nonmechanistic, statistical models that attempt to extrapolate as precisely as possible the toxicity of a chemical from one endpoint to another, from one species to another, or across a whole community of species (e.g., by formulating species-sensitivity distributions). Toxicity-extrapolation models may be applied in an effects assessment to support the use of a population, ecosystem, or landscape model. Currently, many ecological risk assessments are limited by a failure to consider population-, ecosystem-, or landscape-level endpoints. The typical hazard quotient approach compares an expo - sure estimate to a toxicity threshold for some individual-level endpoints such as organism survival, growth, or reproduction. Some authors interpret the hazard quotient for individual-level endpoints to infer population-level risks. This approach often leads to an overestimation of risk at the population level but in some cases may lead to an underestimation of population risk. For example, in their summary of toxicity test data with both individual-level endpoints and population param - eters, Forbes and Calow (1999) found that the population growth parameter was more sensitive than individual-level endpoints in 5 of 99 cases when the lowest-observed-effect concentrations were compared among endpoints. When the percentage change in an endpoint was used as the measure of effect, 66 of 81 cases were found in which the relative change in the population growth parameter was less than the change in the most sensitive individual-level endpoint, two cases were found in which the percentage changes were equal; and 13 cases were found in which the relative change in the population growth parameter was greater than the change in the most sensitive individual-level endpoint. Our review of ecological models shows that population and ecosystem modeling have been applied successfully in past ecological risk assessments, especially for toxic chemical issues 1574CH15.fm Page 211 Tuesday, November 26, 2002 6:36 PM © 2002 by CRC Press LLC (Barnthouse et al. 1986; Bartell et al. 1992; Spromberg et al. 1998; Spencer et al. 1999) and for population viability analysis in conservation biology (Boyce 1992; Burgman et al. 1993; Norton 1995; Brook et al. 2000). Although their use is not presently widespread in the assessment of toxic chemical contamination, ecological models can contribute new perspectives and enhance the value of risk assessments in support of environmental management. Increasingly, ecotoxicol - ogists are calling for an evaluation of population-level effects (Barnthouse et al. 1986; Emlen 1989; Sibley 1996; Caswell 1996; Barnthouse 1998; Forbes and Calow 1999; Snell and Serra 2000; Kuhn et al. 2000). Forbes and Calow concluded that the basic population growth parameter, r, is “a better measure of response to toxicants than individual-level effects because it integrates potentially complex interactions among life-history traits and provides a more relevant measure of ecological impact.” Population models are currently a cost-effective approach for addressing many risk assessment issues (Ferson et al. 1996; Barnthouse 1998). Indeed, incorporation of simple, scalar population modeling into screening-level ecological risk assessments could greatly improve the results of such assessments; in some cases, they could help to avoid unnecessary expenditures on a higher-tier risk assessment. Examined over a hierarchical spectrum of endpoints, modeling of populations and metapopulations gives the highest combination of ecological rele - vance and tractability (Figure 15.1). We therefore recommend that population modeling be included in most ecological risk assessments either at the screening level or in detailed assessments. We recommend scalar population models, life-history models, and metapopulation models for widespread use in ecological risk assessment. Scalar population models are still very crude, but they address the essential elements of risk to the basic ecological unit (i.e., a group of individuals of the same species constituting an interacting population). Although they ignore biological com - plexity, they are tractable and widely accepted, and their use in screening-level ecological risk assessments is clearly valuable. These and other population models can be implemented in a full probabilistic mode. In particular, we recommend the development of stochastic differential equation models and stochastic discrete-time models. Life-history models are more realistic than scalar population models, and they are essential in cases in which the receptors of concern form an age or stage class within the population. Matrix models in particular have a long history of use, and their behavior is well understood. Macroecology databases (e.g., allometric relationships and other summary statistical approaches) should be developed further to enable simple scalar population models and life-history models to be used efficiently in ecological risk assessments. Metapopulation models are particularly relevant for addressing physical disruption of habitats because they account for the effects of habitat fragmentation. Because physical habitat features, including fragmentation patterns, can greatly affect the response of a population to toxic chemicals (Spromberg et al. 1998), we recommend the development of such models for use in ecological risk assessments of toxic chemicals. In particular, metapopulation models should be linked to GIS platforms to integrate fate-transport analysis with an evaluation of ecological effects. Individual-based population models have not generally been applied to chemical risk issues, so most available models do not incorporate functions or rules to account for the effects of toxic chemicals. In some cases, these models are being developed further to address toxic chemical issues (e.g., as part of the ATLSS approach to evaluate mercury contamination in the Everglades), but their transference to other systems is questionable. One can implicitly model chemical effects by running different scenarios with individual-based models and changing organism dispersal, fecun - dity, and growth parameters to account for toxicity. However, the use of individual-based models is presently limited because of the high level of effort necessary to develop such models for specific species and sites. Although ecosystem models provide valuable conceptual tools for analyzing ecological systems subjected to stress, they are expensive to develop and apply to particular risk issues (Figure 15.1; Table 14.1 ). Ecosystem models have been developed to fulfill two basic roles. First, ecologists have used them as descriptive constructs to evaluate the sensitivity of ecosystems to specific environmental parameters. Second, they have been developed as predictive tools to evaluate envi - 1574CH15.fm Page 212 Tuesday, November 26, 2002 6:36 PM © 2002 by CRC Press LLC ronmental management alternatives. Within either of these contexts, ecosystem models require substantial development in specific cases to provide predictions with the level of precision often desired for decision-making. Koelmans et al. (2001) discussed several general problems with complex models, including the lack of a general theory concerning the structural detail needed for accurate description of ecosystem dynamics, the difficulty of coupling representations of different trophic levels that function on a wide range of space and time scales, and the risk of obtaining a good model fit with the wrong parameter setting. Moreover, uncertainties increase in integrated models such as IFEM and AQUATOX, which superimpose toxicological effects models for food webs on representations of contaminant transport and fate. Koelmans et al. (2001) recommended that use of integrated fate and effects models of ecosystems in a predictive mode be restricted to forecasting the short-term effects of acutely toxic chemicals until further progress can be made in understanding and simulating the dominant processes in complex systems. Ecosystem and landscape models are best utilized as heuristic tools for understanding basic ecological processes and for identifying sources of uncertainty in predictive outcomes. Some models, such as IFEM, incorporate Monte Carlo analysis to yield probabilistic expressions of risk as their final output. Other computational methods such as neuronets, Bayesian analysis, and maximum likelihood predictors have been introduced recently into ecosystem models to allow the results to be expressed in terms of probability. This step has allowed much more transparent treatment of uncertainty. Unfortunately, most of the models utilizing these new techniques are highly experimental and would not be directly applicable in current ecological risk assessment. Aquatic ecosystem models are generally better developed than terrestrial ecosystem models (aside from the forest-gap models, which have been as well researched as the aquatic models). In the near future, ecosystem and landscape models are likely to be applied to only the most complex sites and issues. We recommend the following activities to further the development of ecological models for use in chemical risk assessment: • Educating environmental managers and risk assessors on the application of ecological models through workshops and the development of ecological modeling courses offered via the Internet • Developing several generic models of populations or ecosystems for routine use in registering new chemicals • Modifying available, spatial models of populations (i.e., RAMAS GIS and VORTEX) so the effects of toxicants are explicitly represented • Developing software and guidance for use of stochastic scalar abundance models in ecological risk assessments of toxic chemicals • Integrating metapopulation models with ecosystem models and with landscape models • Linking available fate and transport models with selected ecological models • Applying selected models in site specific test cases Figure 15.1 Relevance and tractability of ecological models in relation to endpoints. 1574CH15.fm Page 213 Tuesday, November 26, 2002 6:36 PM © 2002 by CRC Press LLC Further information about the recommended ecological models can be obtained from the Internet web sites listed in Table 15.1 . Although our review emphasizes user-friendly software that is specifically designed with built- in optional equations to support population and ecosystem modeling, all of the model types discussed in this report can also be implemented using generalized modeling software such as STELLA (High Performance Systems 2001), ModelMaker (ModelKinetix 2001), and MATLAB (MathWorks 2001). In all applications, selection of the most appropriate model is the primary concern; the specific software used to implement the model is a secondary concern. Moreover, development of generic models for use in applications such as registering new chemicals should not be seen as an excuse for relying on a fixed set of general models. If such models are developed, they should be regularly evaluated and updated on the basis of current reviews of available approaches. In most cases, a single model is insufficient for all applications, even when a generic approach is appropriate. Several complementary models (e.g., within population and ecosystem or landscape categories) should be developed for each purpose so potential users have options depend - ing on their specific assessment needs. Table 15.1 Selected Ecological Effects Software for Risk Assessment Model Type Web Site Populations and Metapopulations RAMAS ® GIS www.ramas.com VORTEX pw1.Netcom.com/~rlacy/vortex.html Food Web Populus www.cbs.umn.edu/populus RAMAS ® Ecosystem www.ramas.com Ecosystem AQUATOX www.epa.gov/ost/models/aquatox CASM www.cadmusgroup.com IFEM www.cadmusgroup.com Landscape ATLSS http://atlss.org/ LANDIS http://landscape.forest.wisc.edu/Projects/LANDIS_overview/landis_overview.html JABOWA www.naturestudy.org/ http://eco.wiz.uni-kassel.de/model_db/mdb/jabowa.html 1574CH15.fm Page 214 Tuesday, November 26, 2002 6:36 PM . estimated risks initially expressed in terms of individual-level end - points. In this mode, ecological models aid in translating risks for individual-level endpoints to more relevant endpoints at. cost-effective approach for addressing many risk assessment issues (Ferson et al. 1996; Barnthouse 1998). Indeed, incorporation of simple, scalar population modeling into screening-level ecological. species constituting an interacting population). Although they ignore biological com - plexity, they are tractable and widely accepted, and their use in screening-level ecological risk assessments