Abstract
Fossil pollen assemblages are an important tool to reconstruct vegetation variability at centennial and millennialscales. However, in the Cerrado biome of Brazil, the influence of the local vegetation, the heterogeneity of the landscape and the accuracy of pollen identification leads to a wide margin of error in interpretation. In this study, we present a method of quantitative reconstruction of past vegetation cover (forest, savanna, and grassland) that accounts for these different aspects. Our method is based on analysis of 47 surface samples collected in the southern Cerrado biome (east-central Brazil). First, we defined criteria that can be used to identify the local ecosystem (lake, marsh, palm swamp, or gallery forest) represented by the pollen assemblages. The pollen assemblages were then compared to data extracted from vegetation maps, resulting in the assessment of three pollen score equations related to each of the vegetation types (forest, savanna, and grassland). Based on these scores, we developed linear regression models that can be used to reconstruct the proportion of each type of land cover from pollen assemblages. We compared the results obtained for three different areas (respectively with 300 m, 500 m, and 1 km radius) to evaluate the size of the pollen source area for the Cerrado sites. We apply our models to two fossil pollen records to illustrate their use and application. Our method for the quantitative reconstruction of past land cover in the Cerrado biome is a new tool to investigate climate drivers and the longterm effects of fire in this Neotropical savanna.