Fitting demographic models of forest distributions to both patterns and processes in forest inventory data
Demographically-Driven Distribution Models (DDDMs) represent a new and promising framework for linking spatial variation in vital rates (growth, mortality, and recruitment) and population dynamics to species’ broader geographic distributions. By modelling variation in demographic rates against environmental covariates, DDDMs can enable a process-based understanding of how separate components of individual performance govern variation in abundance across species’ ranges, as well as the limits of their geographic distributions. These approaches hold particular promise for forest communities, for which population and community dynamics can be characterized well with size-structured models based on height-structured competition for light.
Forest DDDMs may be parameterized from data through either of two complementary approaches. Firstly, demographic models can be fit directly from observations on the processes themselves (i.e., growth, mortality, and recruitment observed from repeat surveys of individuals across tree ranges). Secondly, demographic parameters can be inferred indirectly from stand chronosequences (i.e., by matching predicted species abundance over time and space to plots of different ages sampled across a broad region). By comparing and combining models that are fit to underlying demographic processes versus emergent patterns in abundance across tree ranges, we may seek a rich, process-based understanding of the dynamics of forest distributions.
We used data from Forest Inventory and Analysis plots across the eastern US to develop demographic models of tree growth, mortality, and recruitment for six plant functional types (PFTs), then implemented the resulting demographic models in simulations of joint stand-level population dynamics in 1° grid cells across the region. When the demographic models were parameterized purely from tree-level demographic data, we found that the emergent geographic distribution of each PFT approximated its actual distribution and reproduced qualitative successional patterns. Some lack-of-fit was apparent as most PFT distributions were predicted to be larger than they actually are. We next applied a hybrid approach where predicted demographic rates were re-scaled at the level of individual grid cells to provide an improved fit to the plot-level abundance of each PFT at different times since stand establishment (10-150 years). The resulting models yielded excellent fits to the abundance, geographic distribution, and successional dynamics of each PFT. Comparisons of demographic models under the two approaches showed that the improved fit to abundance patterns was achieved largely by increasing mortality, rather than through systematic changes to growth or recruitment rates. We discuss implications in terms understanding demographic controls over forest distributions in space and time.