Model-Based Essential Fish Habitat Definitions for Gulf of Alaska Groundfish Species
The 1996 reauthorization of the Magnuson-Stevens Fishery Conservation and Management Act (MSFCMA) mandates that the National Marine Fisheries Service (NMFS) identify habitats essential for managed species and conserve habitats from adverse effects of fishing and other anthropogenic activities. Essential Fish Habitat (EFH) is defined under the act as ‘those waters and substrates necessary to fish for spawning, breeding, feeding or growth to maturity.’ As part of this mandate, EFH descriptions for all species listed under a Fisheries Management Plan (FMP) in Alaska waters are needed. In addition, these descriptions are routinely revisited under a 5-year cycle that reviews and updates EFH information (including species descriptions) with new data and research.
Essential fish habitat descriptions consist of maps of EFH and text descriptions. In Alaska, most EFH descriptions for groundfish have been limited to qualitative statements on the distribution of adult life stages. These are useful, but could be relatively easily refined both in terms of spatial extent and life history stage using species distribution models (SDMs) and data available from a variety of NOAA sources. Distribution models have been widely used in conservation biology and terrestrial systems to define the potential habitat for organisms of interest (e.g., Delong and Collie 2004, Lozier et al. 2009, Elith et al. 2011, Sagarese et al. 2014). Recently SDMshave been developed for various groundfishes in the eastern Bering Sea (McConnaughey and Syrjala 2009, Laman et al. 2017), and for coral and sponge species in the Gulf of Alaska and Aleutian Islands (Rooper et al. 2014, Sigler et al. 2015, Rooper et al. 2016).
Species distribution models themselves can take a number of forms, from relatively simple frameworks such as generalized linear or additive models to more complex methods including computer learning methods (boosted regression trees, maximum entropy models, and random forest models) or multi-stage models. The models can be used to predict potential habitat, abundance, or probability of presence, but they all have some features in common:1) the underlying data consistof some type of independent variables (predictors) and a dependent response variable (presence, presence/absence or abundance), 2) raster maps of independent variables are used to predict a response map, and 3) confidence bounds aroundthe predictions and partitioning of the data can be used to produce test statistics useful for evaluating the model. The outputs of SDMsare raster maps that can show the predicted abundance of a species at each of the raster cells. This type of product is useful for EFH descriptions, as it lends itself to producing maps of areas of high abundance or hotspots of distribution and the models themselves can be used to generate the required text descriptions.