NEON: Enabling Ecological Forecasting

Background/Question/Methods The National Ecological Observatory Network (NEON) will be an NSF-sponsored research facility for the study of long-term, large-scale ecological change. NEON's science mission is to enable understanding and forecasting of the impacts of climate change, land use change and invasive species on continental-scale ecology by providing infrastructure and information to support research in these areas. Ecological forecasting, a modeling and analysis activity is central to NEON because of the Grand Challenges and NEON's derived mission, which involves understanding and predicting across a span of environmental challenges. The science vision that led to NEON's conception involved advancing the field's ability to quantitatively predict, and not just develop retroactive explanations. Ecological forecasting represents a quantitative prediction that is critical for documenting and advancing scientific understanding and is also useful in societal application of knowledge. Ecological forecasting includes two closely related activities. The first activity is similar to a weather forecast and addresses the question: what is the most likely future state of an ecological system? The second activity involves the what-if question: what is the most likely future state of a system given a decision today?

Results/Conclusions While qualitative forecasts may be made using models derived from theory, quantitative forecasts in complex dynamical systems require estimates of the state of the system and include parameters that must be estimated empirically. The limitations of linking forecasting to short-term or episodic data collection arise as a consequence of the lack of stationarity that exists in dynamic ecological systems. Iterative or cyclic forecasting provides a powerful approach that-in a general way-accommodates the lack of stationarity. In cyclic forecasting, a model is initialized with observations, integrated forward to produce a forecast, compared again to observations, re-initialized, and again integrated forward. A model developed over a single forecast cycle tends to explore a small subregion of the solution space, whereas models that are developed iteratively through updating can characterize a much larger region of the solution space. Iterative/cyclic forecasting can reveal patterns of error that are not evident in a single forecast cycle. For example, a model may perform well at low population densities but fail as higher densities are reached. NEON must collect and make available data on a regular schedule to enable iterative comparison of model predictions and observations, leading to an orderly forecast evaluation/update/improvement cycle.