All Climate Events
Webinar: Evaluating near-term forecast skill of species distribution models to novel ocean conditions: guiding expectations for species distribution model projections
Thursday, 28 July 2022, 12:00
Thursday, July 28, 2022. 12:00 PM. Webinar: Evaluating near-term forecast skill of species distribution models to novel ocean conditions: guiding expectations for species distribution model projections. Kathy Mills and Andrew Allyn, Gulf of Maine Research Institute. Sponsored by US Northeast Climate-Fisheries Seminar Series. More information here. Register here.
Abstract: Correlative species distribution models are commonly used to provide predictions of species occurrence patterns under future environmental conditions, offering essential information for managing the challenges arising from climate-driven species distribution shifts. Although widely applied, we still have much to learn about the predictive skill of these models. We helped fill this gap using a simulation study to evaluate model near-term forecast skill to novel ocean conditions by leveraging recent climate change responses in two different large marine ecosystems: the Northeast U.S. Continental Shelf and the California Current. Within each ecosystem, we simulated resident and seasonally-migrating species archetypes given known species-temperature response curves. Using data from 1985-2004, we fit boosted regression tree distribution models and then used the fitted models to forecast monthly species distributions to the 2005-2020 holdout testing data. For each of the forecast timesteps, we calculated a suite of prediction skill statistics. Additionally, we summarized how novel the forecast conditions were relative to the conditions the model learned under using Hellinger's Distance values. Generally, forecast skill declined with increasing novelty of forecast target condition. However, the seasonally-migrating species archetype presented an interesting contrast to this general pattern as forecast skill remained high and even slightly increased when forecasting to months in the summer and fall despite increasing forecast target novelty. These results build our theoretical understanding of species distribution model forecast skill and provide guidance for distribution projection efforts, highlighting how underlying system dynamics and species behavior may interact to create unexpected patterns in model forecast skill.
Bio(s): Andrew Allyn is a quantitative research associate at the Gulf of Maine Research Institute in Dr. Kathy Mills Integrated Systems Ecology Lab and a PhD candidate at the University of Massachusetts Amherst. Within Dr. Mills' lab, Andrew leads the species distribution modeling efforts and is particularly interested in using models to forecast species distribution and abundance under future environmental conditions.