| Terrestrial biosphere models predict that current ecosystems will undergo major shifts due to climate change. The projected direction and magnitude of changes in plant distribution and abundance, however, remain uncertain. In particular, long-term vegetation responses to changing climates are poorly constrained in the models due to the limited temporal coverage of modern observational studies and manipulative experiments. Paleoecological data, such as fossil pollen records, extend observations into the past for studies of slower processes controlling long-term ecosystem behavior. However, paleoecological data are imperfect representations of ecosystem states in the past: they do not directly measure ecosystem properties such as vegetation composition and abundance, and they have a sparser spatiotemporal coverage comparing with modern ecological data. To make paleoecology directly useful in addressing ecological theory and modeling relevant to urgent global change issues, we need to understand how to best collect, analyze, and interpret paleoecological data, and how to apply the knowledge about long-term ecosystem behavior to assess ecological theories and to evaluate management applications. In this dissertation, I present three studies showcasing thoughtful sampling, interpretation, and integration (with ecosystem models) of paleoecological data. First, I asked the question that whether the fossil pollen data sampled for previous paleoecological studies were truly sufficient to represent the vegetation phenomenon of interest. Using the mid-Holocene hemlock decline as an example, I showed that temporal sampling density of pollen is critical for correctly identify where the hemlock-decline occurred in the sediment, and in turn, the age assignment and interpretation of the event. I also proposed general criteria for obtaining adequate sampling when studying an event in a pollen record. Second, I examined how pollen in the sediment of small lakes represents surrounding vegetation. I developed a Bayesian statistical model, which linked vegetation composition with pollen assemblages and accounted for processes such as pollen production, transport, and landscape heterogeneity. Taxon-specific parameters of pollen productivity and dispersal were estimated with uncertainties. Results also highlighted that in addition to pollen transported from the air above the canopy, the "gravity" deposition of pollen (anthers, microsporangia, and pollen clumps falling from trees growing directly above the lake) and the pollen carried by low-velocity winds in tree-trunk space under the canopy (the trunk-space source) were also important components of pollen transfer from the forest canopy to a lake. Lastly, using fossil pollen records and process-based vegetation modeling, I tested competing hypotheses on the emergence and disappearance of the late-glacial no-analog vegetation. Results from the model-data comparison demonstrated that temperature controlled the formation of the no-analog vegetation, lower atmospheric carbon dioxide concentration before the late-glacial period altered the competitive balance of taxa, and enhanced fire activity likely drove the plant communities out of the no-analog state. Collectively, this work (1) advanced the understanding of how sediment-pollen sequences represent past vegetation changes, (2) suggested the likely drivers and processes controlling late-glacial no-analog vegetation in eastern North America, and (3) provided a framework for integrating paleoecological data with vegetation modeling to investigate long-term ecological processes. |
