Trait-Based Individual Models Extend Chemical Risk Assessment to Data-Poor Species

Ecological risk assessment of chemicals has long faced an uncomfortable paradox: the species that regulators most need to protect are often the species about which least is known. Standardized laboratory tests rely on a handful of hardy, easily reared organisms, yet the animals and plants actually living in contaminated rivers, fields, and forests number in the thousands, many of them poorly characterized biologically. A growing body of work argues that the way out of this impasse is to stop thinking about species primarily as taxonomic labels and start thinking about them as bundles of measurable traits—body size, feeding rate, reproductive strategy, habitat preference—that determine how chemicals reach them, how sensitive they are, and how quickly their populations can recover.

The framework is set out in a recent review of trait-based and individual-based modeling for ecological risk assessment across species. It describes how assessments of chemicals commonly follow a stepwise, tiered approach, beginning with simple standardized studies on a few standard species and progressing toward more detailed evaluations that better reflect real-world ecological conditions. More advanced assessments may consider more realistic patterns of chemical exposure, studies conducted under semi-natural or natural conditions, and models that combine experimental results with other information to explore how effects vary across environmental conditions. At this stage the focus often shifts from effects on individual organisms to effects on populations of particular species, using models that describe population dynamics and, less frequently, interactions among species in ecological communities.

This progression increases ecological relevance but introduces substantial challenges. Although focal species may be clearly defined in some cases, higher-tier assessment often needs to consider a wide range of potentially exposed species, many of which are poorly characterized, so the species of greatest regulatory or ecological concern are frequently those for which the least data exist. Because traits describe properties of individuals, the review argues, individual-based models provide the most natural mechanistic framework for translating trait variation among individuals and species into population-level outcomes that regulators can act on, providing a structured way to extrapolate from data-rich standard test species to the data-poor focal species that matter most.

Traits, in this context, are any species properties that are morphological, such as wingspan or body shape; physiological, such as feeding and metabolic rates; or ecological, such as habitat preference—preferably linked to a function like feeding preferences or feeding guilds. An appropriate combination of such traits can provide a functional representation of a species for a defined assessment purpose, without requiring a complete species-specific dataset. Trait-based approaches have been applied to improve prediction of risk in aquatic species, to identify patterns in communities under various pressures, and to link species to ecosystem functioning, giving them a track record that extends well beyond regulatory toxicology.

Trait-based thinking is not entirely new to ecotoxicology. The vulnerability concept introduced by van Straalen in 2003 has been applied to identify vulnerable aquatic species in pesticide assessments, bee species, and wildlife exposed to metals, and has been used more broadly to develop a framework for trait-based ecotoxicology. In that approach, vulnerability-related traits are defined in advance—considering sensitivity, exposure, and population resilience—and expert opinion is often used to derive rankings of vulnerable species. A complementary approach groups species by trait similarity without predefining vulnerability traits, using established life-history categories, for example in freshwater mussels, or, where no prior grouping exists, cluster analyses to identify representative species for risk assessment. Selecting data-rich species as proxies for wider groups becomes far more defensible when a preceding trait-based analysis confirms that the chosen representatives truly stand in for the species they are meant to protect.

The modeling side of the framework comes in two flavors: community models built on trait combinations, and single-species population models capable of representing multiple species within one common structure. In both cases, species-specific differences are captured through variation in trait values—that is, model parameters—while the underlying model structure stays fixed. This consistency is more than an aesthetic preference. Comparative studies indicate that model structure can substantially influence model outcomes and the management implications derived from them, so applying a common structure facilitates comparisons among species and helps isolate the key traits driving differences in population dynamics and responses to stressors.

One instructive example involves four cyprinid fish species listed as endangered or threatened under the US Endangered Species Act. The model simulates population dynamics across the four species, with particular emphasis on variation in life-history characteristics. Sensitivity-related traits were parameterized using data from the fathead minnow, another cyprinid, meaning that differences in modeled responses among the focal species arose solely from differences in their life-history traits. That design allowed the study to reveal the influence of individual traits on population responses to stress and recovery, identify vulnerable species, highlight critical data gaps, and provide an extrapolation framework for pesticide-related risks to population persistence of the listed species.

Solitary bees illustrate the approach at its most ambitious. More than 20,000 solitary bee species occur globally, yet sufficient data to develop models exist for only a subset, despite important similarities across species’ life histories and life cycles. SolBeePop, an individual- and trait-based model, currently represents seven ground- and cavity-nesting species using a common structure, with species-specific differences captured through parameter values. The model has been evaluated against independent data for the red mason bee, Osmia bicornis, and the leaf-cutter bee, Megachile rotundata, demonstrating that the implemented structure can realistically reproduce the dynamics of two different species. That validation increases confidence in predictions for untested species—one of the key uncertainties in applying models to pesticide risk assessment.

Chemical sensitivity data remain the stubborn bottleneck. Within chemical risk assessment, effects data are available primarily for standard test species, and testing nonstandard or focal species is generally discouraged on ethical and logistical grounds. Information on sensitivity for focal species must therefore either be informed by standard test species, as in the cyprinid example, or extrapolated using newer tools such as quantitative structure–activity relationship models that integrate biological trait information. Notably, toxicokinetic–toxicodynamic models, among the most widely used models in ecological risk assessment, already embody a form of trait-based reasoning: they typically assume a shared generic toxicity pathway across species while representing interspecific differences through parameter values governing uptake, elimination, damage dynamics, recovery, and effect thresholds. Sensitivity traits can then be combined with traits driving exposure, such as spatial behaviors and dietary preferences, extending predictions from laboratory conditions to realistic field settings as assessments incorporate greater ecological relevance.

Two converging trends are steadily eroding the data problem. First, remote sensing, other sensor-based approaches, and environmental DNA methods now generate records of species occurrence and, in some cases, trait-relevant information, while the global expansion and standardization of citizen-science initiatives have produced large volumes of biodiversity data from a growing corps of active naturalists. Second, artificial intelligence methods can leverage these datasets to evaluate hundreds of features and identify those with the greatest predictive value. Integrating biological traits, including taxonomic relatedness, with conventional QSARs has improved the predictive accuracy of species sensitivity, and the machine-learning approach enables extrapolation across both species and chemicals, yielding predictions for species–chemical combinations never included in model training. The review notes that this increased predictive capability can ultimately contribute to more accurate and protective risk assessments while reducing the amount of animal testing required—though the reliability of these data-intensive tools depends strongly on the quality and representativeness of the underlying data, so advances in data generation and tool development must progress in parallel.

The outlook is one of disciplined extrapolation rather than blind faith. Trait-based models should be parameterized to represent real species or biologically plausible species profiles rather than arbitrary trait combinations, supporting credible extrapolation from well-studied test species to ecologically relevant but less-studied ones. Where data are scarce, a trait-based approach can still bridge gaps across related or functionally similar taxa and identify representative species likely to provide protection against unacceptable adverse effects for a wider group—but such extrapolation introduces uncertainty, making rigorous testing and validation against independent data essential, which remains challenging where population-level responses for nonstandard species are limited. Machine-learning approaches can maximize the utility of existing data while systematically identifying key drivers and remaining data gaps, guiding targeted, hypothesis-driven data generation. Individual-based ecology, by explicitly capturing individual-level toxicity, individual variability, and life-history and ecological traits, offers the coherent framework needed to connect mechanistic understanding and data-driven predictions to the population- and community-level outcomes on which long-term protection goals—and the management decisions that flow from them—ultimately depend.

Subject of Research: Trait-based and individual-based modeling for extrapolating chemical ecological risk assessment across species

Article Title: From species traits to protection goals: Individual-based modeling for ecological risk assessment across species

Article References: Galic, N. (2026). From species traits to protection goals: Individual-based modeling for ecological risk assessment across species. Individual-based Ecology, 2, Article e216492. https://doi.org/10.3897/ibe.2.216492

Image Credits: AI Generated

DOI: 10.3897/ibe.2.216492

Keywords: ecological risk assessment, trait-based modeling, individual-based models, SolBeePop, solitary bees, QSAR, machine learning, TKTD models, pesticides, population dynamics, species traits, biodiversity data