Pollen records are extensively used to reconstruct past vegetation dynamics and study the underlying drivers and processes. New work is developing the statistical techniques needed to accurately represent both data and process uncertainties. Recent advances in paleoecoinformatics (e.g. Neotoma Paleoecology Database and European Pollen Database), Bayesian age-depth models, process-based pollen-vegetation models (PVMs), and Bayesian hierarchical modeling have advanced paleovegetation reconstructions to a point where multiple sources of uncertainty can be incorporated into reconstructions, which enables new hypotheses to be tested and more rigorous integration of paleovegetation data with terrestrial ecosystem models.
Several process-based PVMs have been developed, notably LOVE/REVEALS, and STEPPS, a Bayesian hierarchical model. The two differ structurally, but the effects of these differences on model predictions and behavior have not been explored. Key differences include the specific pollen dispersal kernels and inclusion of modeled spatial and temporal persistence of vegetation in STEPPS, and inclusion of lake size in LOVE/REVEALS. Additionally, STEPPS estimates all parameters from spatial datasets of pollen and vegetation and explicitly represents parameter and process uncertainty. Here we compare the behavior and parameter sensitivities of these PVMs, using pre-settlement pollen and vegetation datasets from the northeastern US.
Results/Conclusions
Both STEPPS and LOVE/REVEALS reproduce general vegetation patterns of the northeastern US, with hardwoods dominated by oak in the southeast, hardwoods dominated by beech in the northwest and forests dominated by spruce and fir in the northeast. Pollen dispersal varies between LOVE/REVEALS and STEPPS with STEPPS allowing pollen dispersal over larger distances. Comparative analyses highlighted the effect on PVMs of the maximum distance of pollen dispersal allowed, an underconstrained parameter that is specified by the analyst. For STEPPS, parameterization of pollen dispersal is broadly consistent between calibration studies in the upper Midwest and NEUS, while estimates of pollen productivity are consistent with values found in the literature. Based on these calibration analyses, work is now underway to reconstruct with uncertainty changes in forest composition for the last 2000 years in the NEUS. These reconstructions are intended to be used for testing and refining terrestrial ecosystem model simulations of forest dynamics at timescales of centuries to millennia.