Quantifying Uncertainty in Spatio-temporal Forest Composition Changes Inferred from Fossil Pollen Records

Thursday, 18 December 2014: 5:15 PM
Andria Dawson1, Christopher J Paciorek1, Jason S McLachlan2, Simon J Goring3, John W Williams3 and Stephen T Jackson4, (1)University of California, Berkeley, CA, United States, (2)University of Notre Dame, Notre Dame, IN, United States, (3)University of Wisconsin Madison, Madison, WI, United States, (4)USGS Southwest Research Climate Science Center, Tucson, AZ, United States
Understanding past compositional changes in vegetation provides insight about ecosystem dynamics in response to changing environments. Past vegetation reconstructions rely predominantly on fossil pollen data from sedimentary lake cores, which acts as a proxy record for the surrounding vegetation. Stratigraphic changes in these pollen records allow us to infer changes in composition and species distributions. Pollen records collected from a network of sites allow us to make inference about the spatio-temporal changes in vegetation over thousands of years. However, the complexity of the relationship between pollen deposits and surrounding vegetation, as well as the spatially sparse set of fossil pollen sites are important sources of uncertainty. In addition, uncertainty arises from the carbon dating and age-depth modelling processes.

To reconstruct vegetation composition including uncertainty for the Upper Midwestern USA, we build a Bayesian hierarchical model that links vegetation composition to fossil pollen data via a dispersal model. In the calibration phase, we estimate the relationship between vegetation and pollen for the settlement era using Public Land Survey data and a network of pollen records. In the prediction phase, parameter estimates obtained during the calibration phase are used to estimate latent species distributions and relative abundances over the last 2500 years.

We account for additional uncertainty in the pollen records by: allowing expert palynologists to identify pre-settlement pollen samples to be included in our calibration data, and through the incorporation of age uncertainty obtained from the Bayesian age-depth model BACON in our prediction data.

Resulting spatio-temporal composition and abundance estimates will be used to improve forecasting capabilities of ecosystem models.