Abstract
Quantitative understanding of vegetation dynamics over timespans beyond a century remains limited. In this regard, the pollen-based reconstruction of past vegetation enables unique research opportunities by quantifying changes in plant community compositions during hundreds to thousands of years. Critically, the methodological basis for most reconstruction approaches rests upon estimates of pollen productivity and dispersal. Previous studies, however, have reached contrasting conclusions concerning these estimates, which may be perceived to challenge the applicability and reliability of pollen-based reconstruction. Here we show that conflicting estimates of pollen production and dispersal are, at least in part, artifacts of fixed assumptions of pollen dispersal and insufficient spatial resolution of vegetation data surrounding the pollen-collecting lake. We implemented a Bayesian statistical model that related pollen assemblages in surface sediments of 33 small lakes (<2 ha) in the northeastern United States, with surrounding vegetation ranging from 101 to >105 m from the lake margin. Our analysis revealed three key insights. First, pollen productivity is largely conserved within taxa and across forest types. Second, when local (within a 1-km radius) vegetation abundances are not considered, pollen-source areas may be overestimated for some common taxa (Cupressaceae, Pinus, Quercus, and Tsuga). Third, pollen dispersal mechanisms may differ between local and regional scales; this is missed by pollen-dispersal models used in previous studies. These findings highlight the complex interactions between vegetation heterogeneity on the landscape and pollen dispersal. We suggest that, when estimating pollen productivity and dispersal, both detailed local and extended regional vegetation must be taken into account. Also, both deductive (mechanistic models) and inductive (statistical models) approaches are needed to better understand the emergent properties of pollen dispersal in heterogeneous landscapes.