The effects of landcover change on pollen-vegetation relationships in the Upper Midwest
Pollen assemblage data are widely used to estimate past vegetation community composition and to study plant responses to past climate change. However, this type of reconstruction assumes a stable relationship between pollen and vegetation, and little research has been done to confirm this assumption.
It has been known for decades that the widespread land clearance associated with Euro-American settlement massively altered forest composition and the pollen assemblages produced by these forests, but it was unknown whether the relationships between forest and pollen assemblage composition were themselves altered. Using vegetation surveys and pollen core data from the Upper Midwest (Michigan, Minnesota, and Wisconsin), we test hypotheses regarding the stability of pollen-vegetation relationships from the time of Euro-American settlement to the present, a time period that encompasses significant changes in landcover, climate, and fire regime. We use R Statistical Software to construct a zero-inflated, beta-distributed generalized additive model of pollen-vegetation relationships, using three large-scale ecological databases: pollen data from the Neotoma Paleoecological Database, from which we extract pollen samples immediately prior to settlement horizons; records of presettlement vegetation from the Public Land Survey System; and modern forest composition data from the U.S. Forest Service’s Forest Inventory and Analysis Program.
For some taxa (ash, basswood, elm, fir, maple, poplar, and tamarack), pollen-vegetation relationships show no significant change between the settlement era and present, suggesting that for these taxa, pollen-based estimates are transferable. However, pollen-vegetation relationships shift significantly (p < 0.05) for several taxa: beech, birch, hemlock, oak, pine, and spruce. The application of contemporary pollen-vegetation relationships to reconstruct past ecosystems from pollen data will under-predict the presence of some taxa (beech, birch, oak, pine, hemlock, and spruce), while possibly over-predicting the presence of others (ash, basswood, and maple). These modeling distortions imply a need for careful consideration of pollen-based estimates of past vegetation, and suggest that, when possible, pollen-vegetation calibration datasets in North America should be built on historical vegetation surveys conducted prior to the major period of Euro-American landuse. Given that historical datasets are not available for many parts of the world, a useful future direction for these historical and contemporary vegetation datasets is to estimate the errors and uncertainties associated with using contemporary pollen-vegetation relationships to make broader inferences about past shifts in forest composition.