A systematic, regional approach to predictive modeling of habitat suitability for deep-sea corals in U.S. waters

//A systematic, regional approach to predictive modeling of habitat suitability for deep-sea corals in U.S. waters
­
Loading Events
  • This event has passed.

Speaker: Brian Kinlan, NOAA National Centers for Coastal Ocean Science, Biogeography Branch

Register at: https://attendee.gotowebinar.com/register/8236288050617849092

Sponsors: This webinar is part of a NOAA Deep Sea Coral Research and Technology Program webinar series to highlight research, exploration, and management of deep-sea corals and sponges around the U.S.

Seminar POC: Heather.Coleman@noaa.gov (301-427-8650)

Abstract: Recently, predictive modeling has emerged as an essential tool to inform researchers and policy-makers involved in conservation, management, and exploration of deep-sea coral (DSC) habitats throughout U.S. waters. From 2011-2016, NCCOS and its partners have developed a series of regional-scale predictive models of habitat suitability for several taxonomic (e.g., Lophelia pertusa, Gorgonian Alcyonacea) and functional (e.g., framework-forming corals) groups. These models have resulted in a comprehensive, consistent series of predictive maps spanning four U.S. regions – Northeast/Mid-Atlantic, Southeast Atlantic, Gulf of Mexico, and Main Hawaiian Islands – with a spatial resolution of ~400 m. Multiple measures of model performance, including cross-validation statistics and novel metrics of model fit and stability, and maps of spatial uncertainty were generated to support decision-making. Maximum Entropy (MaxEnt) models were fit to coral presence records and spatial environmental predictors, including topographic, oceanographic, and geographic variables. We enhanced the standard MaxEnt approach in several ways to improve model selection, performance assessment, consistency and interpretability. We implemented a stepwise model selection process to identify models that balanced predictive power (via cross-validation statistics) with complexity (via information criteria). Using the selected models, we predicted the relative likelihood of occurrence of suitable habitat within each model grid cell. To allow consistent comparisons across coral groups and regions, we converted the standard MaxEnt “logistic” predictions, which are uncalibrated, into habitat suitability likelihood classes calibrated by a cross-validation procedure. Finally, we compared and contrasted environmental predictor relationships across coral groups and regions, yielding insights into correlates of DSC distributions at a range of spatial scales. We are presently engaged in field model groundtruthing and validation efforts, and are working on a new generation of high-resolution (~25m) models based on the most accurate field survey and seafloor mapping data available. These new models will use both presence and absence data, in combination with measures of survey effort based on area of seafloor searched, to generate probabilistic models of occurrence probability and genus-level diversity (richness) measures.

Subscribe to the OneNOAA Science Seminar weekly email: Send an email to OneNOAAscienceseminars-request@list.woc.noaa.gov with the word `subscribe’ in the subject or body. See http://www.nodc.noaa.gov/seminars/

(Brian Kinlan, NOAA National Centers for Coastal Ocean Science, Biogeography Branch)