UW Aquatic & Fishery Sciences Quantitative Seminar

Jim Thorson

Operations Research Analyst, NOAA Northwest Fisheries Sceince Center

Know Thy Stochastic Process: The key to generality in statistical ecology

Abstract

Methods to study ecology are as diverse as the questions that ecologists seek to answer. One growing niche for ecological students is in “statistical ecology”, where researchers use statistical models to link ecological theory with available data. In this talk, I argue that general advances in statistical ecology are possible via a strong background in “stochastic processes” (i.e., basic mechanisms for change that can be predicted using one or few variables). To illustrate this claim, I discuss three ongoing collaborations using simple stochastic models for survival, evolution, and synchrony. I first use a survival model to estimate when a fished stock will be first assessed (or remain unassessed) in the United States. This analysis suggests that assessed stocks differ systematically from stocks that remain to be assessed, in particular by having greater landings or price. I next use a multivariate model for the evolution of life-history traits to estimate mortality, maturity, and growth for all 33,000 fishes. This analysis shows that the ratio of mortality and growth is not a “life history invariant” but instead varies systematically based on the timing of maturity. Finally, I use a stochastic model of population synchrony to estimate the relative contribution of multiple species and locations to portfolio effects in seven marine ecosystems. This analysis identifies a large decrease in portfolio effects in the Celtic Sea after 2000, enabled by the strong spatial synchrony for fishes in that ecosystem. I conclude by discussing avenues for future advances in statistical ecology. These include climate-impact assessments for marine ecosystems, tests for stability in community dynamics, and simultaneous analysis of planned and opportunistic data. Throughout, I seek to show how simple statistical mechanisms can be used to analyse processes involving different scales, taxa, and questions.