New framework brings mortality risk to life in mechanistic ecology models

Mortality is the ultimate currency of natural selection, yet in the computational models that ecologists use to understand how populations work, death is often handled with surprising crudeness. A new perspective article by Steven F. Railsback and Bret C. Harvey, published in the journal Individual-based Ecology, argues that this simplicity is a bottleneck for the field, and it offers a practical mathematical toolkit for representing mortality risk in mechanistic, individual-based models with far greater realism and flexibility.

The authors, both longtime developers of individual-based models in which simulated organisms behave as autonomous agents, point out that mortality risk is one of the most complex components of individual fitness. Real animals face multiple kinds of danger simultaneously: predators, starvation, harsh physical conditions, disease. Each of these risks can change with habitat, with time of day or season, with an individual’s own state such as size or energy reserves, with the activity the individual is currently performing, and even with the status of the population as a whole. A young, hungry fish foraging in open water at dusk faces a very different probability of death than a well-fed adult resting in dense cover, and a good model should be able to capture that difference.

In practice, however, risk is frequently represented very simply in ecological models, often as a single constant mortality rate or a crude function of one or two variables. The literature offers little practical guidance on how to model the variation in risk that matters most for behavior and fitness. Railsback and Harvey draw on decades of modeling experience to identify five characteristics that desirable risk representations should have. First, survival probability should be able to vary with multiple variables describing individuals, habitat, and other entities in the model. Second, the relationship between survival and any specific variable should be addable, removable, or modifiable without re-fitting the entire model. Third, those relationships should be able to take different mathematical forms, including continuous nonlinear functions and discrete values. Fourth, the relationships should be easy to understand and easy to fit to many kinds of data or assumptions. Fifth, and perhaps most pragmatically, the whole risk representation should be calibratable by adjusting only a single parameter.

That last requirement may sound modest, but it addresses one of the chronic headaches of individual-based modeling. Complex models typically contain many uncertain components, and calibration becomes nearly impossible when every improvement to one part forces the re-estimation of parameters everywhere else. A mortality module that can be tuned with one knob, while still responding flexibly to many variables, gives modelers a way to keep complex simulations tractable and testable.

To achieve these goals, the authors first review the terminology and conventions ecologists have traditionally used when modeling risk, then lay out a general mathematical framework. The core of the framework is a concept they call survival increase functions. Each such function relates survival probability to exactly one variable, such as habitat cover, body size, activity level, or the density of predators. Crucially, each function can have a different mathematical form tailored to the evidence available for that particular variable, and each can be based on different kinds of information, from field observations to experimental data to theoretical reasoning.

The individual functions are then combined into a single survival probability value for each simulated individual at each time step. The combination is designed so that the resulting probability remains well behaved even when many variables act at once, and so that a single calibration parameter scales the overall strength of the risk representation. This modular architecture means a modeler can improve the representation of, say, predation risk from herons without disturbing how the model handles winter starvation, then recalibrate the whole thing by adjusting one number.

The approach is illustrated in the article with examples showing how multiple survival increase functions interact to produce complex, realistic patterns of mortality risk across habitats, times, and individual states. Such patterns are exactly what drive the risk-avoidance behaviors that individual-based models are often built to explore. When a simulated animal can weigh the survival cost of foraging in a risky patch against the energetic benefit, the fidelity of the underlying risk representation directly determines whether the model’s predictions about habitat choice, growth, and population dynamics are credible.

One of the most valuable contributions of the paper is its discussion of how survival increase functions can be evaluated and supported by evidence. The authors describe a spectrum of methods, ranging from general field observations of where and when animals die, to controlled field experiments that isolate the effect of a single variable, to mechanistic knowledge and data on the processes driving survival, and even to conceptual models of those mechanisms when quantitative data are scarce. This flexibility matters because mortality is notoriously difficult to study directly; predation events are rarely observed, and dead animals are seldom found. A framework that can absorb evidence of many kinds, and make its assumptions explicit and modular, gives ecologists a way to build defensible risk representations even from imperfect knowledge.

The broader significance of the work lies in what it enables for individual-based ecology as a discipline. Models that represent how individual behavior and fitness depend on risk are central to questions about habitat selection, predator-prey dynamics, responses to environmental change, and the management of threatened populations. By providing a clear mathematical structure with desirable properties spelled out in advance, Railsback and Harvey offer a practical standard that other modelers can adopt, critique, and refine. The framework has already proven practical in their own modeling projects for representing the complex effects of multiple variables on survival probability.

For a field in which the difference between a crude constant mortality rate and a well-structured risk function can determine whether a model illuminates reality or misleads, this article is both a critique of current practice and a constructive blueprint for doing better. It invites ecologists to treat mortality risk not as an afterthought bolted onto the end of a simulation, but as a first-class component of individual fitness, designed with the same care and modularity as the behavioral and energetic processes that surround it.

Subject of Research: Mortality risk representation in mechanistic individual-based ecology models

Article Title: Representing mortality risk in mechanistic models

Article References: Representing mortality risk in mechanistic models. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: mortality risk, individual-based ecology, survival probability, mechanistic models, survival increase functions, model calibration, fitness, risk-avoidance behavior, ecological modeling, Steven F. Railsback, Bret C. Harvey, Representing