Constraining pollen-based estimates of forest cover in the Amazon: A simulation approach

Abstract

Pollen dispersal and deposition (PDD) modelling has been instrumental in reconstructing historical vegetation in temperate regions, but its application has been limited in the tropics where there is greatest uncertainty in past land cover change. Here, we apply PDD modelling to Amazonian savanna and forested ecosystems. Empirical pollen data from lakes situated in southwestern Amazonia were used to calibrate the PDD model for a two-component landscape of forest and non-forest. The PDD model was then used to simulate pollen assemblages for different combinations of landscape arrangements (the multiple scenario approach) that reflect possible anthropogenic and climate-driven forest cover change in the late-Holocene. We show that pollen records from large Amazonian lakes vary greatly in their sensitivity to forest loss depending on the baseline forest cover. Lakes in landscapes containing >80% forest will detect small reductions (5% of total cover), but this sensitivity degrades rapidly with forest cover loss. There are a wide range of uncertainties in pollen reconstructions from mosaic and ecotonal landscapes. In forest-savanna mosaics, large reductions of forest cover could be undetectable through the pollen record. In ecotonal landscapes, the relationship between forest cover and its representation in the pollen record rapidly weakens with increasing distance from the forest boundary. Further application of PDD modelling in combination with the multiple scenario approach can address the uncertainties in pollen-based reconstructions of past land cover in the tropics, but require further investment and development.