Community dynamics under environmental change: How can next generation mechanistic models improve projections of species distributions?
Abstract
Environmental change is expected to shift the geographic range of species and communities. To estimate the consequences of these shifts for the functioning and stability of ecosystems, reliable predictions of alterations in species distributions are needed. Projections with correlative species distribution models, which correlate species’ distributions to the abiotic environment, have become a standard approach. Criticism of this approach centres around the omission of relevant biotic feedbacks and triggered the search for alternatives. A new generation of mechanistic process-based species distribution models aims at implementing formulations of relevant biotic processes to cover species’ life histories, physiology, dispersal abilities, evolution, and both intra- and interspecific interactions. Although this step towards more structural realism is considered important, it remains unclear whether the resulting projections are more reliable. Structural realism has the advantage that geographic range shifting emerges from the interplay of relevant abiotic and biotic processes. Having implemented the relevant response mechanisms, structural realistic models should better tackle the challenge of generating projections of species responses to (non-analogous) environmental change. However, reliable projections of future species ranges demand ecological information that is currently only available for few species. In this opinion paper, we discuss how the discrepancy between demand for structural realism on the one hand and the related knowledge gaps on the other hand affects the reliability of mechanistic species distribution models. We argue that omission of relevant processes potentially impairs projection accuracy (proximity of the mean outcome to the true value), particularly if species range shifts emerge from species and community dynamics. Yet, insufficient knowledge that limits model specification and parameterization, as well as process complexity, increases projection uncertainty (variance in the outcome of simulated model projections). The accuracy–uncertainty-relation reflects current limits to delivering reliable projections of range shifts. We propose a protocol to improve and communicate projection reliability. The protocol combines modelling and empirical research to efficiently fill critical knowledge gaps that currently limit the reliability of species and community projections.
What the paper shows and why it matters
(AI-generated)
Environmental change is expected to shift where species live, and decision-makers need projections of these shifts that can be trusted. Correlative species distribution models link records of where a species occurs to abiotic (non-living) conditions. Critics note that such models leave out biotic feedbacks, which prompted process-based (mechanistic) models of dispersal, life history, evolution and interactions between species. The paper takes on an assumption behind this step: that added structural realism makes projections more reliable, even though the ecological knowledge needed to specify and parameterise the processes is available for few species.
The paper argues that process-based species distribution models cannot yet deliver reliable projections of range shifts for many species, because the realism they call for and the knowledge available pull in opposite directions. It is an opinion paper with no analysis of its own. It separates accuracy, how close the average projection is to the true value, from uncertainty, how much simulated outcomes vary. Omitting a relevant process, such as dispersal, can impair accuracy. Including processes that are poorly known raises uncertainty, and the stochastic, nonlinear dynamics of interacting species add inherent variability. A schematic example, a species with medium dispersal that depends on another species, shows a correlative model that, assuming no dispersal, comes closer to the true shift for the wrong reason, while a mechanistic model stays inaccurate until the interaction is added, and conveys wider uncertainty than the correlative one. Its way out is a five-step protocol: collect theory, expert knowledge and evidence; parameterise the relevant processes; assess and partition uncertainty; alternate model improvement with empirical data collection aimed at the gaps that matter most; and communicate the model, its alternatives and the uncertainty.
The paper reads the heavy data demand of process-based models as a reflection of the knowledge needed for reliable projections, not as a defect. It nonetheless holds that such models should include the relevant biotic processes, because they are the framework in which these processes can be added and tested. They are also tools for locating the gaps that limit reliability, so that targeted empirical work can close them, and the paper advises stating the shortcomings of projections openly, since overly optimistic claims of reliability are risky in decision-making. Limits remain: the protocol is proposed, not tested, in the paper.