Demography-based models of the geographic ranges of nine dominant tree species in western North America: Does a process-based approach improve models of distributions?

Background/Question/Methods

Current attempts to forecast how species' distributions will change in response to climate change suffer from being phenomenological, particularly with respect to extrapolation (predicting beyond the range of conditions a species currently experiences). Explicitly modeling a species’ demography, i.e., the set of conditions under which population growth is positive, may improve range forecasting because it should allow one to distinguish climate suitability from other processes that govern species’ occurrence on the landscape (e.g., dispersal limitation, source-sink dynamics). Integral Projection Models (IPMs) – population models similar to matrix projection models – are particularly well suited to the task of estimating the influence of climatic niche on species’ distributions, since vital rates are estimated via regression models that can incorporate environmental covariates, rather than counts. Based on >550,000 data points derived from the Forest Inventory and Analysis (FIA) for nine dominant tree species in western North America, we constructed regressions of individual tree growth and survival. Log-transformed biomass increment (or logit-transformed survival status) was modeled as a function of log-transformed biomass in the previous census, competitive environment (plot basal area) in the previous census, as well as average climate over the course of the census interval.

Results/Conclusions

We show maps of predicted growth and survival, based on regression parameters and GIS layers of the climate variables used in regression modeling. These maps are compared against 1) species distribution models (SDMs) using the same climate predictors, but occurrence data only, and 2) expert range maps. We also compare response curves from regressions of individual growth and survival vs. SDMs, since response curves are critical in determining the behavior of forecasts. Poor correspondence between predicted growth or survival and geographic distributions suggests either that 1) growth, survival (and reproduction) shape the southern vs. northern limits of species’ geographic distributions via different climatic factors, 2) censuses taken at 10-year intervals on average lack the temporal resolution to infer how vital rates respond to climate, or 3) regional and species variation in the climatic drivers of vital rates require more detailed modeling. We discuss these alternative explanations, and suggest that multilevel modeling that integrates different types of data (FIA census data, tree rings, abundance, occurrence, and/or absence data, as well as experimental data) may be critical to reconstruct biologically-reasonable response curves, scale from plots to entire geographic distributions, and generate believable forecasts of future distributions.