Modeling the pollen and vegetation relationships for comparing simulated vegetation with fossil pollen data

Background/Question/Methods

Fossil pollen records can be used to infer past vegetation dynamics. Quantifying the relationships between pollen assemblages and vegetation composition can (i) improve the theoretical understanding of how pollen can be used as an index of vegetation composition, and (ii) facilitate quantitative comparison between model-simulated past vegetation and fossil pollen records. We built a Bayesian statistical model to characterize the relationship between pollen composition and vegetation composition. Modern pollen samples were collected from surface sediments of 33 small lakes in the northeastern US, and were paired with quantitative estimates of vegetation composition surrounding each lake. We partitioned the effect of vegetation on pollen into “local” and “background” contributions. We used vegetation survey data within a 1-km radius around the lakes to estimate the local pollen contribution, and USFS Forest Inventory and Analysis (FIA) data to inform the background pollen contribution. We estimated pollen productivity and dispersal parameters for different taxa to identify and correct for distortions and biases in the relationship between pollen percentages and composition of source vegetation.

Results/Conclusions

Among the 13 common taxa in the northeastern US, the productivity parameters are highly variable; e.g. Betula, Pinus and Quercus appear to be the highest pollen producers, and they are ten times or more productive than Acer. Uncertainties in the pollen productivity estimates vary among taxa, partially due to the taxonomic resolution of pollen. The taxon-specific pollen dispersal parameters also differ significantly among taxa; e.g. most Fraxinus pollen dispersal is within 4 km, whereas most Pinus pollen dispersal can occur over distances up to 65 km. For the poorly dispersed taxa, such as Larix, the dispersal parameters could not be estimated well. In general, local vegetation within the 1-km radius contributes more than 50% to the pollen composition. The ratio of local vs. background contributions varies among regions of different forest types. The local contribution is lower in the coniferous and northern hardwoods forests, and higher in the oak-dominated forests. We are applying the fitted model to convert a simulated Holocene vegetation time-series from the northeastern US into a pollen time-series, in an effort to identify the discrepancy between the simulated vegetation (driven by paleoclimate simulations) and the actual fossil pollen record.