The Use of Zero-Inflated Generalized Additive Models to Predict Aggregate Functional Group Distributions for the Gulf of Mexico

Development of the Atlantis Gulf of Mexico ecosystem model (Atlantis-GoMex) is currently underway at the University of South Florida in conjunction with NOAA's Integrated Ecosystem Assessment for that region. Spatially explicit ecosystem models of all types require an initial allocation of biomass on varying spatial scales which can be problematic due to the lack of comprehensive stock assessments for more than a handful of commercially valued species. Previous spatial modeling efforts have relied on a variety of methods to distribute the biomass of their respective taxa including expert opinion, similarity matrices, and outside modeling.  A generalized additive modeling (GAM) approach was used to estimate functional group biomass over 64 irregular polygons throughout the Gulf of Mexico. Two GAM models are developed respectively describing the probability of occurrence at all sampling locations and the population density at non-zero locations.  These are combined using the Delta method to yield an overall abundance.  The GAM models utilize zero-inflated Gamma distributions to characterize catch per unit effort rates from the Southeast Area Monitoring and Assessment Program (SEAMAP).   Regional climatology from the National Oceanographic Data center was used to predict seasonal functional group biomass throughout the Gulf of Mexico, including Mexican and Cuban waters.