Fossil pollen assemblages provide information about vegetation dynamics at time scales ranging from centuries to millennia. Pollen-vegetation models and process-based models of dispersal typically assume stable relationships between source vegetation and corresponding pollen in surface sediments, as well as stable parameterizations of dispersal and productivity. These assumptions, however, are largely unevaluated. This paper reports a test of the stability of pollen-vegetation relationships using vegetation and pollen data from the Midwestern region of the United States, during a period of large changes in land use and vegetation driven by Euro-American settlement. We compared a dataset of pollen records for the early settlement-era with three other datasets of pollen and forest composition for two time periods: before Euro-American settlement, and the late 20th century. Results from generalized linear models for thirteen genera indicate that pollen-vegetation relationships significantly differ (p < 0.05) between pre-settlement and the modern era for several genera: Fagus, Betula, Tsuga, Quercus, Pinus, and Picea. The estimated pollen source radius for the 8 km gridded vegetation data and associated pollen data is 25–85 km, consistent with prior studies using similar methods and spatial resolutions.

Hence, the rapid changes in land cover associated with the Anthropocene affect the accuracy of ecological predictions for both the future and the past. In the Anthropocene, paleoecology should move beyond the assumption that pollen-vegetation relationships are stable over time. Multi-temporal calibration datasets are increasingly possible and enable paleoecologists to better understand the complex processes governing pollen-vegetation relationships through space and time.