DEVELOPING A NEW GENERATION OF LAND COVER RECONSTRUCTIONS FOR THE UPPER MIDWEST BASED ON BAYESIAN MODELS, PUBLIC LAND SURVEY DATA, AND NEOTOMA DATABASE

A central mission of Quaternary paleoecology is the accurate reconstruction of past plant taxa distributions, community composition, and vegetation structure from fossil pollen records. Such reconstructions usually are calibrated against spatial arrays of surface pollen samples and contemporary vegetation. Techniques include ordination, modern analog methods, biomization, Prentice/Sugita models of pollen transport, and, recently, Bayesian hierarchical models. All approaches must account for intertaxonomic variations in pollen productivity and dispersal and are potentially confounded by the imprint of recent land-use upon plant community structure and composition.

Here we report on current efforts to reconstruct forest composition in the upper Midwest (UM) for the last several millennia, based on the following data and tools: 1) Fossil pollen records from the public-access Neotoma Paleoecology Database, 2) a new 8km gridded dataset of settlement-era forest composition based on Public Land Survey (PLS) data, and 3) the STEPPS Bayesian hierarchical model of pollen-vegetation. PLS witness-tree data have been stitched together into settlement-era maps of relative forest composition from Minnesota to Maine and reconstructions of biomass and stem density for the UM. Recently published pollen records from the UM have been uploaded to Neotoma, with refined estimates of the settlement horizon and new bacon age models. STEPPS employs two spatial models – a conditional autoregressive (CAR) model and Gaussian Markov random field (MRF) approximation to a Gaussian process – to smooth noise across observations and interpolate estimates to locations with missing data. STEPPS is calibrated against settlement-era datasets of fossil pollen and forest composition, thereby minimizing the effects of EuroAmerican land-use. Current we are focusing on STEPPS calibration and cross-validation; next we will reconstruct forest composition in the upper Midwest for the last two millennia.