Because of the slow pace of terrestrial ecosystem processes, including the slow generation time, growth rate, and decomposition rate of trees, the impact of changing climate and disturbance on forests plays out over hundreds of years. For this reason, centennial scale projections of terrestrial ecosystem models are used to anticipate the trajectory of forest response to environmental change. Modelers would like to have data on how forests have changed at regional scales and over hundreds of years to help assess such projections. A rich assemblage of relevant paleoecological data has been collected, but they have not been synthesized into a form that can be incorporated into broad-scale modeling efforts. Funding provided will support the establishment of a paleoecological observatory network (PALEON) to address this challenge. PALEON is an interdisciplinary team of paleoecologists, environmental statisticians, and ecosystem modelers with the goals of producing rigorous and robust reconstructions of forest change from the Atlantic to the Great Plains over the past 2,000 years, and then using these reconstructions to validate and improve the predictions of terrestrial ecosystem models.
PALEON has identified the integrated analysis of paleoecological data with statistical and mechanistic modeling as a key challenge for improving research capacity for anticipating the future of environmental change. For this reason, PALEON incorporates interdisciplinary training and community building into all aspects of the PALEON mission. In addition to focused working groups, PALEON works with relevant disciplinary communities to develop common approaches to data collection, analysis, and experimental protocols to ensure that long-term data can be seamlessly integrated into macroscale ecosystem analyses. Interdisciplinary training of post-doctoral fellows and graduate students, including a summer short course, will ensure that the next generation of researchers thinks naturally at the spatial and temporal scales relevant to understanding the broad scale impact of changing climate and land-use disturbance.
Ecosystems are changing at the scale of continents due to natural processes and human impacts, such as land-use and climate change. To help society anticipate and plan for these changes, we increasingly rely on models that forecast the state of the earth under the conditions expected in the future. These ecosystem models are remarkably robust at capturing the functioning of the planet at large scales, but they are currently more adept at capturing rapid processes, like the response of forests the year after a drought, then they are at capturing slow processes, like the response of regional vegetation to a changing fire regime imposed by years of drought. The obvious reason that ecosystem models are clumsy at depicting slow ecological processes is that most of the data that underlie these models are from rapid processes that are easier to observe. However, there is a large and growing body of scientific information about slow processes that has not been integrated into models: historical survey data documenting millions of trees back to the Colonial period; tree rings, which show the growth response of trees after disturbance and in response to changing climate; a menagerie of sedimentary data (isotopes, charcoal, lipids, fossil pollen, testate amoebae, etc) that correlate with climate, wildfire, and vegetation change over millennia. Two technical breakthroughs allow us to incorporate these paleoecological and historical data into predictive ecosystem models for the first time. First, statistical breakthroughs allow us to translate these obscure proxies into more familiar ecosystem terms: We can statistically estimate the composition of forest trees from networks of fossil pollen data. We can statistically estimate changing fire regimes from fossil charcoal in sediments. Second, because these standard ecosystem properties are now measured with uncertainty, we can assimilate the information about changing ecosystem state into our forecasting models. Thus, the models we use to forecast the impact of global change can now be improved with empirical estimates of the slow ecosystem processes that have eluded modelers in the past. The PaleoEcological Observatory Network (PalEON) is an effort to build a team of paleoecologists, statisticians, and ecosystem modelers to improve ecological forecasts by incorporating slow processes into models. In our Macrosystems Phase 1 stage, PalEON1 built this team; compiled networks of paleoecological data across New England and the Midwest; developed statistical tools and used them to model ecosystem change across millennia; and, finally, demonstrated the importance of these slow changes to societally relevant forecasts of ecosystem change. As an example, we know little about the distribution of species at the time of Euroamerican settlement (the Colonial era in New England and the early 19th century in the Midwest). But the "original" vegetation of the United States is important for benchmarking conservation and resource management, and for understanding the impact of subsequent human land-use. Figure 1 shows the PalEON estimate of the distribution of American beech at the time of Euroamerican settlement produced by a statistical model of hundreds of thousands of trees in surveys of original forests across 15 states. Because the abundance of American beech has been strongly reduced by climate, land-use, and disease since the industrial revolution, these estimates have great value for guiding forest conservation and management. Empirical estimates of the abundance of Eastern all tree taxa also allow us to evaluate the performance of the ecosystem models underlying predictions of climate change. The "CMIP5" model comparison is a recent effort to compare the performance of dozens of ecosystem models run at the global scale. We found that none of these models captured the distribution of pre-industrial forests correctly. Because forests are linked to the atmosphere through the carbon cycle and biophysics, biases in modeled vegetation could potentially have an aggregate impact on climate estimation. Figure 2 shows that, for many models, compensating local errors resulted in predictions of net primary productivity (forest growth) that were not biased at the regional scale, but errors in forest simulation had significant impact on productivity and other properties relevant to climate at regional scales in a subset of models. Last year, PalEON funding was extended to a full five year Macrosystems Phase 2 stage (PalEON2). The results illustrated here will feed forward to a more complete integration of models and long term data in PalEON2 and we will produce publicly accessible estimates of vegetation, fire, productivity, and climate at large scales over the last two millennia. As we developed our novel interdisciplinary team, we have emphasized the training of young scientists to integrate the new tools of Bayesian statistics and data assimilation with paleoecological data analysis. This integrated training takes place across PalEON laboratories, and also beyond the PalEON team in training courses that we run. Throughout PalEON, we have communicated our approach and new findings through social and traditional media and through more traditional scientific conferences and publications.
