A major goal of ecology and evolutionary biology is to reconstruct species’ biogeographic histories, including the location of ranges through time and the magnitude and rate of range shifts during periods of global change. Typically, these questions have been addressed using disparate data and methods: genetic data coupled with model based inference; fossil remnants--including pollen coupled with pollen-vegetation models (PVMs); and contemporary specimen records coupled with species distribution models (SDMs). Each data type conveys information about different historical processes, and so independently is insufficient for confidently reconstructing species’ biogeographic histories. Here we review current approaches for inferring biogeographic trajectories using analyses of green ash (Fraxinus pennsylvanica) since the last glacial maximum (LGM). First, we apply techniques in isolation: model-based clustering analyses to detect population genetic structure; a PVM for inferring the past location of the range; and SDMs for hindcasting areas of suitable climate. Then, we present a novel, integrative approach that combines each of these methods using Approximate Bayesian Computation (ABC).
Results/Conclusions
The three techniques agree that green ash has undergone dramatic biogeographic change since the LGM. Nonetheless, genetic analysis suggests that ash was subdivided into 3 clusters during the LGM, whereas the PVM and SDMs suggest less subdivision. Unlike the genetic analysis, the PVM and SDMs are able to identify areas of purported refugia, although they disagree on the location of refugial habitats. The SDMs are also limited in their ability to infer past presence since they map locations of suitable—but not necessarily occupied—habitat. Owing to the difficulty in identifying fossil pollen to species, the PVMs are limited to inferences at the genus level. In contrast, integrating analyses using ABC builds on the strengths of each data type. Moreover, ABC allows us to infer parameters of processes that are otherwise difficult to measure: rates of short- versus long-distance dispersal, site-specific abundance through time, and biotic velocities. Taken together, these results demonstrate that analyses relying on just one technique or on informal integration risk bias due to the limitations of each method. Rather, a formal, statistical integration of distinct datasets is required to address shortcomings of each data type, provide a full accounting of uncertainty, and robustly infer species’ past biogeographic trajectories.