Quantifying forest range shifts and biotic velocities since the Last Glacial Maximum in the upper midwestern US

Terrestrial ecosystems play an important role in Earth systems processes, yet we still do not understand how they respond to changes in climate. These ecosystem processes operating at multiple timescales; fast processes occur at sub-annual timescales, and slow processes, driven by changes in forest composition and structure, occur over decadal and longer timescales. Slow processes are rarely directly observed from instrumental data, yet are critical to understanding the stability of the terrestrial biosphere over the coming decades. Networks of paleoecological data, particularly sedimentary pollen data, offer our strongest observational constraint on long-term vegetation dynamics and underlying processes and feedbacks.

We reconstruct maps of forest composition from the last glacial maximum to pre-Euro-American settlement. To do this, we use a network of fossil pollen records - the most reliable paleoecological proxy for forest composition. We link the fossil pollen records to public land survey forest composition using a Bayesian hierarchical spatio-temporal model which accounts for key processes including pollen production and dispersal. The model is calibrated using data from the pre-settlement time to minimize anthropogenic impacts. Process parameters are estimated in the calibration phase, and are subsequently used in the prediction phase to generate spatially explicit maps of species composition across the upper Midwestern US over the last 18000 years, with robust uncertainty estimates.

Estimates of forest composition and uncertainty improve the spatio-temporal resolution of our previous understanding of past forest change in the upper midwestern US. These estimates are used to identify: (i) regions that experienced large change (taxon or community level); (ii) regions that were stable; and (iii) taxon biotic velocities. Finally, these novel spatio-temporal composition estimates will be used to improve the forecasting capabilities of ecosystem models.