A review of pollen-based quantitative vegetation reconstruction models

Abstract

Quantitative vegetation reconstruction aims to provide information about past vegetation cover. One method is using pollen diagrams that are supposed to reflect the relative abundance of taxa in the surrounding vegetation. However, the relationship between pollen and vegetation is affected by production and dispersal bias. Therefore, a need for methods arose that could correct for these biases. To quantify the effect of production bias, estimates for pollen productivity (PPEs) were developed. Secondly, to account for dispersal bias, accurate pollen dispersal models that described the movement pattern of pollen grains with different physical characteristics were needed. This review first presents an overview of the development of dispersal and deposition models from the basic idea of the R-value model by Margaret Davis to the Extended R-value model by Colin Prentice that was further developed by Shinya Sugita. It continues with a reconstruction of the development of PPEs and dispersal models that should correct for production and dispersal bias is given. Lastly, three current quantitative vegetation reconstruction approaches (the Landscape Reconstruction Algorithm (LRA), the Multiple Scenario Approach (MSA) and the Extended Downscaling Approach (EDA)) are compared in the context of a theoretical intermediate scale landscape reconstruction in the Netherlands for the Lateglacial and the Holocene. While the LRA has been often applied on regional-continental scales and sporadically on local scales, the MSA and EDA have been only applied on local scale reconstructions. The LRA reconstructs regional landscapes based on the pollen assemblages of large lakes (REVEALS) and local landscapes on the pollen assemblages of small lakes (LOVE). The MSA and EDA reconstruct landscapes based on the input of abiotic landscape parameters and the simulation of theoretical pollen assemblages that are then compared to the empirical assemblage. The MSA can yield multiple likely landscape scenarios, whereas the EDA only reconstructs one scenario using an optimization key. In the light of a potential landscape reconstruction of the Dutch Lateglacial and Holocene, application of the LRA could be impractical due to the low availability of large lakes (100-500ha). The availability of high-quality abiotic landscape information makes the application of the MSA, and the EDA seem quite suitable, but both methods have not yet been applied on intermediate scales and scaling up may be bothered by high simulation times, especially for the MSA.