What spatial statistical model is best for predicting fisheries by catch risk?

Start: 09/11/2017 - 4:15pm
End  : 09/11/2017 - 5:15pm

Applied Math Seminar

Brian Stock (HMC Math Bio '09; UC San Diego)


Bycatch (i.e. catch of at least some non-targeted species) is an omnipresent problem in commercial and recreational fisheries. High bycatch rates can reduce the efficiency and sustainability of fisheries, but even extremely low bycatch rates can be a problem for protected or rebuilding species. Given these economic and environmental concerns, the fishing community would be well served by tools that predict, and ultimately help avoid, bycatch. I will demonstrate the ability of a new, computationally efficient spatial statistics method, Gaussian Markov Random Fields (GMRFs) implemented in R-INLA software, to produce bycatch risk maps using two large U.S. fisheries observer datasets. I compare the GMRF approach with two other species distribution model frameworks, generalized additive models (GAMs) and random forests (RF), and show how the models' performance differs for species with a broad range of bycatch rates, from leatherback sea turtles (0.7%) to blue sharks (96%) in the Hawaii longline fishery, and yelloweye rockfish (0.3%) to Pacific halibut (29%) in the West Coast groundfish trawl fishery.

I will conclude by highlighting other research opportunities at the intersection of applied math/statistics and fisheries science.

Emmy Noether Rm, Millikan 1021, Pomona College