Quantifying pollen-vegetation relationships to reconstruct ancient forests using 19th-century forest composition and pollen data. (1st April 2016)

• Record Type: Journal Article
• Authors: Dawson, Andria; Paciorek, Christopher J.; McLachlan, Jason S.; Goring, Simon; Williams, John W.; Jackson, Stephen T.
• Is Part Of: Quaternary science reviews. Volume 137(2016)
• Page Start: 156
• Page End: 175
• Publication Date: 2016-04-01
• Subjects: Pollen -- Fossil -- Sediment -- Vegetation -- Forest -- Bayesian -- Modelling -- Expert elicitation -- Dispersal -- Calibration -- Prediction; Geology, Stratigraphic -- Quaternary -- Periodicals; Stratigraphie -- Quaternaire -- Périodiques; 551.79
• DOI: 10.1016/j.quascirev.2016.01.012
• Languages: English
• ISSNs: 0277-3791

Abstract

Abstract: Mitigation of climate change and adaptation to its effects relies partly on how effectively land-atmosphere interactions can be quantified. Quantifying composition of past forest ecosystems can help understand processes governing forest dynamics in a changing world. Fossil pollen data provide information about past forest composition, but rigorous interpretation requires development of pollen-vegetation models (PVMs) that account for interspecific differences in pollen production and dispersal. Widespread and intensified land-use over the 19th and 20th centuries may have altered pollen-vegetation relationships. Here we use STEPPS, a Bayesian hierarchical spatial PVM, to estimate key process parameters and associated uncertainties in the pollen-vegetation relationship. We apply alternate dispersal kernels, and calibrate STEPPS using a newly developed Euro-American settlement-era calibration data set constructed from Public Land Survey data and fossil pollen samples matched to the settlement-era using expert elicitation. Models based on the inverse power-law dispersal kernel outperformed those based on the Gaussian dispersal kernel, indicating that pollen dispersal kernels are fat tailed. Pine and birch have the highest pollen productivities. Pollen productivity and dispersal estimates are generally consistent with previous understanding from modern data sets, although source area estimates are larger. Tests of model predictions demonstrate the ability of STEPPS to predict regional compositional patterns. Graphical abstract: Highlights: We develop and calibrate STEPPS, a Bayesian hierarchical pollen-vegetation model. Calibration uses settlement-era datasets, minimizing signal of Euroamerican land use. Bayesian modelling enables quantification of uncertainty in pollen processes. Best-fitting pollen dispersal kernels are fat-tailed. Pollen source area may be larger than previously thought.