Climate change affects the abundance and distribution of species worldwide. Poikilothermic animals comprise most species on Earth and are extremely sensitive to changes in environmental temperatures. Predicting species responses to climate change when temperatures exceed the bounds of observed data is fraught with challenges. Here, we combine empirical observations of species abundance and environmental conditions across the landscape with laboratory-derived data on the physiological response of poikilotherms to changes in temperature to predict species geographical distributions and abundance in response to climate change. We show that predicted changes in distributions, local extinction, and abundance of cold, cool, and warm-adapted species vary substantially when physiological information is incorporated into model predictions. Failure to account for species-specific physiological constraints can lead to under-estimates of local extinction for cold-adapted species near the edges of their climate niche space and overoptimistic predictions of warm-adapted species under a warming climate.
This software involves files to fit the physiologically-guided abundance (PGA) model using two different approaches to accommodate uncertainty in the parameters that describe thermal response curves. The first approach, is where we incorporate uncertainty in thermal response curves through the Bayesian framework by assigning prior distributions to thermal response curve parameters based on the literature. Independent normal prior distributions are assigned to ToptT_{opt}Topt and CTmaxCT_{max}CTmax using the literature-derived means and standard deviations reported in SI Appendix, Table S3. We use numerical integration to incorporate the uncertainty in ToptT_{opt}Topt and CTmaxCT_{max}CTmax as an alternative to using the joint posterior of all parameters since that approach is too heavily weighted by the likelihood and overwhelms the information from the literature when using temperature data that doesn't span the range of values for which the temperature curves are estimated. Specifically, we randomly sample ToptT_{opt}Topt and CTmaxCT_{max}CTmax from their prior distributions, while ensuring ToptT_{opt}Topt is less than CTmaxCT_{max}CTmax, providing a random realization of the thermal response curve. For each realization of ToptT_{opt}Topt and CTmaxCT_{max}CTmax, we fit the Bayesian model and obtain samples from the posterior distribution of all other model parameters with these values fixed. Numerical integration is obtained by repeating this process 100 times and aggregating the posterior distributions. This allows for the uncertainty in thermal response curves based on the literature to be propagated through to uncertainty in the other parameters and predictions. The R script for fitting the model for each of the three species (bluegill, yellow perch, and cisco) is named 'fish_predictions_species_name_pga_BMA.R, where 'species_name is bluegill, yellow perch, or cisco. The corresponding stan file for fitting the model is 'fish_prediction_BMA.stan'.
The second approach is to use informative prior distributions to account for uncertainty in thermal performance parameters ToptT_{opt}Topt and CTmaxCT_{max}CTmax. Currently these are given very informative priors, which could be relaxed if suitable data existed. There is an R script for each of the three species (bluegill, yellow perch, and cisco) named 'fish_predictions_species_name_pga.R, where 'species_name is bluegill, yellow perch, or cisco. The corresponding stan file for fitting the model is 'fish_prediction.stan'.
There is also a script to fit the naive model ('fish_predictions_naive.R' and fish_prediction_naive.stan) and a PGA model using simulated data ('PGA_sim.R' and fish_prediction_sim.stan). The data for fitting the models can be found at https://hdl.handle.net/11299/228403.
Newer versions of this source code may be available, please check here to ensure you are using the latest/appropriate version