Ecological Forecasting

This emerging approach in ecology emphasizes the need to predict the future state of ecosystems, natural capital, and ecosystem services, and, that while ecological systems are complex, there are strategies to improve the skill of ecological models, borrowed from numerical weather forecasting (iterative, near-term confrontation between model predictions and in-coming data) and using hierarchical Bayesian statistics (the ability to fuse different sources of data and characterize uncertainty). With an Early Career Award from the NSF Macrosystems Biology program, we have been applying these principles to the problem of forecasting future tree growth. Our paper published in Global Change Biology demonstrates (a) the fusion of tree-ring and forest inventory measurements (with a Bayesian state space model) to gain traction on the complexity of tree growth, (b) validation of competing models of tree growth with measurements of tree size that were withheld from model fitting, and (c) parsing of the different causes of forecast uncertainty that point to avenues for further model improvement – and it received the 2022 Ecological Forecasting Outstanding Publication Award from the Ecological Society of America (photo).

Another line of work using tree-ring data to project future tree growth revealed the limits of space-for-time substitution. Based on >30,000 growth ring-width time series for Douglas-fir (Pseudotsuga menziesii) across its very large geographic distribution, we found it grows at a higher rate (produces wider growth rings) at warmer locations, at the same time that trees grow less in a warmer-than-average year (consistently across almost all sites and seasons). We’ve argued that the positive response to spatial variation in temperature arises from both fast and slow ecological processes (including evolutionary adaptation), whereas the negative response to time-varying temperature reflects fast ecological processes (i.e, the plasticity of genotypes exposed to different conditions in different years). Hence, using the growth response of a tree at a warmer location to predict how a tree at a cooler location today will grow in a warmer future relies upon evolution and/or dispersal being instantaneous…and is misleading. These ideas (about the timescales of ecological processes that influence different kinds of data, and the implications for ecological forecasting) have been very nicely captured independently by Peter Adler and colleagues in this paper.

My history in “ecological forecasting” began with a 2003 NSF-funded training workshop on hierarchical Bayesian statistics held at Duke University, with instructors Jim Clark, Alan Gelfand, Kent Holsinger, and others. Out of this came a series of papers on hierarchical Bayesian population viability analysis with Kent Holsinger (Evans et al. 2008, 2010, 2012), which were featured as a chapter in Mike Dietze’s 2017 book Ecological Forecasting. I have offered training in ecological forecasting in the form of one unofficial graduate seminar (fall, 2016) and one official graduate seminar (fall, 2019).