Environmental context, parameter sensitivity, and structural sensitivity impact predictions of annual‐plant coexistence

Abstract

Predicting the outcome of interactions between species is central to our current understanding of diversity maintenance. However, we have limited information about the robustness of many model‐based predictions of species coexistence. This limitation is partly because several sources of uncertainty are often ignored when making predictions. Here, we introduce a framework to simultaneously explore how different mathematical models, different environmental contexts, and parameter uncertainty impact the probability of predicting species coexistence. Using a set of pairwise competition experiments on annual plants, we provide direct evidence that subtle differences between models lead to contrasting predictions of both coexistence and competitive exclusion. We also show that the effects of environmental context dependency and parameter uncertainty on predictions of species coexistence are not independent of the model used to describe population dynamics. Our work suggests that predictions of species coexistence and extrapolations thereof may be particularly vulnerable to these underappreciated founts of uncertainty.

Widespread analytical pitfalls in empirical coexistence studies and a checklist for improving their statistical robustness

“…There are a variety of mathematical functions describing how an individual's reproductive output is affected by various forms of competition. Choosing among many similar, alternative competition models is a longstanding problem in ecology (Ayala et al, 1973;Law & Watkinson, 1987;Martyn et al, 2021), but is a requirement of most approaches for predicting coexistence (Cervantes-Loreto et al, 2023), since all coexistence metrics and their associated uncertainties are contingent on the model itself being a reasonable representation of reality. Each population's monoculture equilibrium and their invasion growth rates when the competitor is at that equilibrium are quantities central to MCT's coexistence metrics.…”

“…This simple model assumes a saturating form of density dependence (Figure 1) from which it is convenient to derive niche and fitness differences. Specifically in the context of MCT-which relies on extrapolating fitted models to equilibrial abundances in order to estimate invasion growth rates-reasonable alternative model structures can lead to drastically different inferences of predicted coexistence (Abrams, 2022;Cervantes-Loreto et al, 2023). An example of this using simulated data is shown in Figure 3.…”

Experimental validation of ecological coexistence theory to forecast extinction under rising temperatures

“…Fitting was based on counts of adult females with a total of 1,223 observed generation-to-generation population transitions for D. pallidifrons and 592 for D. pandora . Model fitting and selection used a Bayesian framework and followed recently established best practices 34,35 (see methods). The selected population model for both species used a simplified Brière thermal performance function 36 and a Beverton-Holt competition model with temperature-dependent competition exerted by D. pallidifrons : where N is the number of individual females of species i in generation g, T g is the temperature in generation g, a, T min and T max are parameters of the thermal performance function defining the low-density growth rate, and α terms are competition coefficients.…”

Widespread analytical pitfalls in empirical coexistence studies and a checklist for improving their statistical robustness

“…This simple model assumes a saturating form of density dependence (Figure 1) from which it is convenient to derive niche and fitness differences. Specifically in the context of MCT-which relies on extrapolating fitted models to equilibrial abundances in order to estimate invasion growth rates-reasonable alternative model structures can lead to drastically different inferences of predicted coexistence (Abrams, 2022;Cervantes-Loreto et al, 2023). An example of this using simulated data is shown in necessary to select a model based on precedent and acknowledge the dependence of the results on these choices.…”

“…Handling model uncertainty is a significant challenge across scientific fields (Simmonds et al, 2022). In an MCT context, error propagation into final coexistence predictions has been used to inform the results (Cervantes-Loreto et al, 2023;Hart et al, 2019;Hess et al, 2022;Mordecai, 2013;Terry et al, 2021). Bowler et al (2022) demonstrates particularly well how a failure to propagate uncertainty can result in misleading conclusions about coexistence.…”