Ecological Forecasting

by

Michael C. Dietze, Colin Averill, John Foster, Kathryn Wheeler

• LAST REVIEWED: 25 September 2018
• LAST MODIFIED: 25 September 2018
• DOI: 10.1093/obo/9780199830060-0205

Introduction

Ecologists have long tried to predict how natural systems will behave in the future, or in response to manipulative treatments, for both basic and applied purposes. However, it is only recently that ecologists have begun to distinguish probabilistic forecasting from other forms of modeling. Clark, et al., “Ecological forecasts: An emerging imperative,” Science (Vol. 293 [2001], pp. 657–660) first defined ecological forecasting as “the process of predicting the state of ecosystems, ecosystem services, and natural capital, with fully specified uncertainties, and is contingent on explicit scenarios of climate, land use, human population, technologies, and economic activity” (p. 657). A key thing that sets forecasting apart from just running a model (either statistical or process-based) into the future is a shift to seeing the world probabilistically—using probability distributions to capture our imperfect understanding of the natural world and to formally include and partition numerous sources of uncertainty and natural variability. This, combined with the emphasis on forecasting real-world systems over theory, means that ecological forecasting tends to be a data-intensive and statistically focused activity (even when employing process-based models). Much of the imperative for ecologists to focus on forecasting comes from the need to respond to the multitude of environmental problems facing society and the aspiration that environmental decisions be made with the best available science in hand. Because all decision-making is ultimately based on what will happen in the future, either under the status quo or different decision alternatives, environmental decision-making ultimately depends on forecasts. Ecological forecasters try to make those forecasts, and their uncertainties, explicit. At the same time, a number of authors have argued that making ecology more predictive is also key to advancing basic science and maturing as a discipline. Generating forecasts forces us to synthesize what we already understand about a system, embodying that understanding in a quantitative model of how we think things will be different in the future, at a new location, or under different conditions. By being quantitatively precise about both our predictions and how confident we are in them, forecasts are more open to direct refutation. Forecasting into yet-to-be measured times and places also emphasizes out-of-sample validation, which is fundamentally a stronger test of our theories and reduces the risk of overfitting and post hoc rationalization. This potential to simultaneously improve basic science and increase social relevance represents a promising win-win scenario.

General Overviews

The consideration of ecological forecasting as a distinct research area is relatively new in ecology, dating to the Clark, et al. 2001 call to arms. Within the United States, this kicked off ecological forecasting programs within both the National Oceanic and Atmospheric Administration and NASA followed, a few years later, by a National Science Foundation research coordination network that produced a number of important papers, such as the overview by Luo, et al. 2011. More recently Dietze 2017 is the first book on the subject, providing an overview of concepts, methods, and case studies from a wide range of ecological subdisciplines. Focused on long-term climate change responses, much of the initial research in ecological forecasting has focused on what MacCracken 2001 defines as projections—forecasts driven by scenarios. However, Dietze, et al. 2018 argues the importance of focusing on near-term predictions (forecasts based on what is known today), which can be iteratively confronted with new observations and updated, as key to accelerating the pace of learning and increasing social relevance. Houlahan, et al. 2017 similarly argues the central importance of making predictions that can be validated to advancing ecology. Finally, as ecological forecasting has matured it has begun to develop best practices, such as those outlined by Harris, et al. 2018 and discussed in more detail in the Methods.

• Clark, J. S., S. R. Carpenter, M. Barber, et al. 2001. Ecological forecasts: An emerging imperative. Science 293:657–660.

  DOI: 10.1126/science.293.5530.657

  Widely considered the seminal article in ecological forecasting; outlines the need to anticipate environmental change and develop policy based on such forecasts.

• Dietze M. C. 2017. Ecological forecasting. Princeton, NJ: Princeton Univ. Press.

  DOI: 10.1515/9781400885459

  First and currently the only textbook on forecasting for ecologists, it synthesizes a wide range of forecasting concepts in one place and highlights how forecasting goes beyond traditional modeling. Emphasizes the Bayesian state space statistical framework, its iterative approximations (Kalman and particle filters), and uncertainty quantification. Also includes case study chapters from a wide range of ecological disciplines, decision support, and informatics.

• Dietze, M. C., A. Fox, L. Beck-Johnson, et al. 2018. Iterative near-term ecological forecasting: Needs, opportunities, and challenges. Proceedings of the National Academy of Sciences 115.7: 1424–1432.

  DOI: 10.1073/pnas.1710231115

  Lays out the central arguments, in terms of both accelerating basic research and making ecology more relevant to decision making, for focusing on near-term forecasts that are iteratively updated as new data become available. Also identifies the major challenges and opportunities across five different focal areas (data, theory and methods, cyberinfrastructure, decision support, and training, culture and institutions) and provides a roadmap for tackling them.

• Harris, D. J., S. D. Taylor, and E. P. White. 2018. Forecasting biodiversity in breeding birds using best practices. PeerJ 6:e4278.

  DOI: 10.7717/peerj.4278

  This paper outlines a set of ten “best practices” for ecological forecasting that should be required reading for anyone working in this area. The paper also provides a nice case study in multimodel hindcasting and assessment, comparing a range of time-series and species distribution models for their ability to predict breeding bird survey data.

• Houlahan J., S. McKinney, T. Anderson, and B. McGill. 2017. The priority of prediction in ecological understanding. Oikos 126:1–7.

  DOI: 10.1111/oik.03726

  Provides a strong argument about the value of forecasting for advancing basic science and the need to test quantitative, precise hypotheses. Links the lack of emphasis on prediction in ecology to the “crisis of reproducibility,” slow progress, and lack of generality of results. Provides a provocative summary of “questions that are rarely asked in ecology,” reasons ecologists don’t predict more, and how we might change the way we do ecology.

• Luo, Y., K. Ogle, C. Tucker, et al. 2011. Ecological forecasting and data assimilation in a data-rich era. Ecological Applications 21.5: 1429–1442.

  A product of the FORECAST Research Coordination Network, this paper provides a nice overview and summary of what ecological forecasting is, where it fits in the history of how we do science, and the role of models and data assimilation in ecological forecasting.

• MacCracken, M. 2001. Prediction versus projection—forecast versus possibility. WeatherZine 26:3–4.

  Formally defines the difference between a prediction (a forecast based on what is known today; e.g., weather forecast) and a projection (a forecast conditional on a scenario; e.g., Intergovernmental Panel on Climate Change climate projections).