Model-data fusion is increasingly employed as a way to assimilate the dynamics of real natural systems, as captured by data, into the necessarily low dimensional depiction of reality in models. Efforts such as NEON, are designed to improve the predictive ability of models of environmental change by assimilating data on ecosystem response to environmental variation collected over decades. However, the response of ecosystems to large-scale changes like land-use and climate are also controlled by even longer-term processes, such as succession and species range shifts. These slow processes are generally not depicted in models in a way that is constrained by actual data. Instead, models of long-term ecosystem change rely on abstractions, such as potential vegetation and the assumption that ecosystem dynamics in North America were stationary before European settlement.
The Paleoecological Observatory Network (PalEON) provides broad-scale reconstructions of vegetation composition and structure at decadal to millennial scales derived from historical and paleoecological data with associated estimates of uncertainty in the space-time process. Reconstructions of vegetation composition and structure at the time of settlement in the Northeast and Midwest demonstrate that estimates of potential vegetation, while broadly robust, fail to capture important fine-scale detail relevant to ecosystem models. Longer-term reconstructions of composition dynamics reveal shifts in species ranges and ecotone that are not consistent with the assumption of stationarity in current models. A remaining challenge for PalEON is developing the tools for assimilating these long-term dynamics into ecosystem models currently oriented towards the impact of faster ecological processes.