Combined models of species distribution and population dynamics are emerging as a powerful tool to predict the impacts of multiple ecosystem threats, such as climate change, land use change, altered fire regimes, and invasive species. Population predictions from these models require many parameters and assumptions, potentially compounding uncertainty in results. Computing numerical sensitivities of model predictions to various sources of uncertainty can help to identify which assumptions are most critical. Using the endangered San Diego thornmint (Acanthomintha ilicifolia) as a case study, we estimated the sensitivities of long-run population predictions to different parameter settings and ecological assumptions.
Results/Conclusions
Uncertainty about habitat suitability predictions, due to the choice of species distribution model (Random Forest versus MaxEnt), contributed the most to variation in model predictions about long run populations. Choice of climate model (PCM versus GFDL) and land use change scenario contributed less to changes in population predictions, with population model choice (scalar versus matrix model) being the least important. A sensitivity analysis of population model parameters showed that only very large changes in growth and survival rates led to large changes in population predictions.