Recreating species’ biogeographic histories over the last 21,000 years using an integrated species distribution model–pollen-density model

Background/Question/Methods

Recreating species’ biogeographic histories–locations of past glacial refugia, rates and routes of subsequent dispersal, and abundance–remains a challenge for understanding the past and present distribution of life on Earth. Two primary means of uncovering these histories rely on fossil pollen deposits and contemporary specimen records. Fossil pollen can be analyzed with pollen-density models (PDMs) which use spatial relationships to interpolate locations of past vegetation, while specimen records can be combined with climate data in species distribution models (SDMs) to estimate locations of past climatically favorable habitats. Each type of analysis has its own strengths and weaknesses. Pollen offers a deep-time record of occurrence, but is usually only identifiable to the genus level, can disperse far from its source, and is subject to depositional issues. SDMs can be used to estimate the location of past climatically suitable habitat, but this approach assumes that present-day occurrences adequately sample the fundamental niche, and that species have unrestricted dispersal. A formal integration of model types is needed to address shortcomings of each method and offer robust estimates of species’ historical trajectories. Here, we present results based on a novel, Bayesian, integrated PDM-SDM for trees in the genus Fraxinus (ash) in eastern North America.

Results/Conclusions

Models based on single data types (just pollen or just occurrences) demonstrate the tendencies of each method of analysis. We find, for example, that the stand-alone PDM predicted the refuge for the genus was close to the glacial margin then moved progressively northward, which is commensurate with the higher level of sampling opportunities (i.e., lakes) in northern areas. In contrast, the SDMs predicted more southerly but mostly distinct refugia, and subsequent expansion tracking climatic envelopes without demographic or dispersal lags. The integrated PDM-SDM relies on a joint-likelihood approach where abundance of each species at a given time and place is modeled as a latent state (unobserved, but estimable) variable. Variants of the integrated model enable species-level inferences while using genus-level pollen data, and for shared or distinct spatial relationships of each data type. The integrated model better accounts for dispersal limitations while constraining occupancy of climatically-suitable habitats by each species to the genus-level distribution. The integrated model also shares information across species to help account for truncation of the niche in the present, thereby providing a more robust estimation of climatic niche tolerances of each species.