Predicting extinction risks under climate change: coupling stochastic population models with dynamic bioclimatic habitat models

Keith, David A., et al. “Predicting extinction risks under climate change: coupling stochastic population models with dynamic bioclimatic habitat models.” Biology Letters 4.5 (2008): 560-563.

Keith et al. set out to determine how climate change will affect landscape scale extinction risk for a variety of species. They note that habitat suitability (HS) models have been used in the past to predict organism responses to climate change. These models are trained on the current relationship between species presence and climate variables and used to predict future species ranges on the basis of climate projection models. These changes in species ranges are not, however, easily translatable into species extinction risks and the HS method leaves out demographic processes that could lead to extinction. They propose a marriage of these HS models with stochastic population models in order to more accurately assess extinction risk through time. Before briefly describing the methods I’d like to mention that the paper could definitely use a supplement to more specifically describe their models (it’s possible that a supplement exists but it is not accessible through the Biology Letters website). Three plant species were used for this project “A, widespread range undergoing large contraction (represented by Protea neriifolia); B, restricted range contracting at the margins but with HS increasing in the core (Leucadendron laureolum); and C, restricted range under-going shift and fragmentation (Leucadendron levinsianus).” They use generalized additive models fit to presence-absence records and five climatic and 3 substrate variables. GAMs were calibrated using a 70-30 training-testing split of the data, AIC based variable selection and the AUC on test data. Distributions for species were fit for 2000, 2030, and 2050 and a full time series was produced using linear interpolation.

These HS maps are used to define the carrying capacity (K) of populations on the landscape in two alternative ways (1) incorporating both habitat suitability and local habitat area or (2) just habitat area. Stochastic population models were constructed for two different life-history patterns (obligate seeding and resprouting) for responding to periodic fire (oh yeah there is fire included in these systems this part of the methods is the first mention of it and it’s about to become important). Spatially explicit stage-structured models which incorporate periodic fire (mean interval of 8 or 14 years), seed dispersal, environmental and demographic stochasticity and stage specific density dependence. 1000 simulations were run for each combination of range type, life-history type, fire regime, and carrying capacity type for the 50 year length of the climate model.

Generally probability of extinction was dependent on a number of factors, including distribution pattern, life history, evenness of density-dependence effects across life stages and fire regime. The most general conclusion was that populations persisted better under 14 year fire cycles than under 8 year cycles. The widespread but contracting distribution (A) was most negatively affected by climate change with longer fire intervals helping the seeder more than the sprouter life-history type. For the other two distribution types (B and C) climate change increased persistence as compared to a static model under 8 year fire regimes but decreased it under 14 year regimes (this seems to run contrary to the statement that 14 year cycles were generally better for persistence but they don’t present any figures on the comparison to the static model so I can’t follow up on that). Dispersal (within biologically reasonable ranges) did not have any effect on population viability. These results suggest complex interactions between habitat suitability, life history traits, and disturbance determine population viability across a landscape, which seems quite reasonable. They also motivate the incorporation of demographic processes into other forecasting Species Distribution Models. Unfortunately I genuinely don’t think I could begin to replicate the methods used in this paper (and am not sure I fully understand all of them) due to a high level of complexity and a lack of detail.