Development of a species distribution model for fin whales ( Balaenoptera physalus) within a Bayesian hierarchical framework: Implications for uncertainty

• Main Authors: Sigourney, Douglas B, Chavez-Rosales, Samuel, Conn, Paul, Garrison, Lance, Josephson, Elizabeth, Palka, Debra
• Format: Other/Unknown Material
• Language: unknown
• Published: PeerJ 2018
• Subjects: Balaenoptera physalus; Fin whale
• Online Access: http://dx.doi.org/10.7287/peerj.preprints.27424v1; https://peerj.com/preprints/27424v1.pdf; https://peerj.com/preprints/27424v1.xml; https://peerj.com/preprints/27424v1.html

Species distribution models (SDMs) have proven to be an integral tool in the conservation and management of cetaceans. Many applications have adopted a two-step approach where a detection function is estimated using conventional distance sampling in the first step and subsequently used as an offset to a density-habitat model in the second step. A drawback to this approach, hereafter referred to as the conventional species distribution model (CSDM), is the difficulty in propagating the uncertainty from the first step to the final density estimates. We describe a Bayesian hierarchical species distribution model (BHSDM) which has the advantage of simultaneously propagating multiple sources of uncertainty. Our framework includes 1) a mark-recapture distance sampling observation model that can accommodate two team line transect data, 2) an informed prior for surface availability 3) spatial smoothers using spline-like bases and 4) a compound Poisson-gamma likelihood which is a special case of the Tweedie distribution. We compare our approach to the CSDM method using a simulation study and a case study of fin whales ( Balaenoptera physalus ) off the East Coast of the USA. Simulations showed that the BHSDM method produced estimates with lower precision but with confidence interval coverage closer to the nominal 95% rate (94% for the BSHDM vs 85% for the CSDM). Results from the fin whale analysis showed that density estimates and predicted distribution patterns were largely similar among methods. Abundance estimates were also similar though modestly higher for the CSDM (4700, CV=0.13) than the BHSDM (4526, CV=0.26). Estimated sampling error differed substantially among the two methods where the average CV for density estimates from BHSDM method was approximately 3.5 times greater than estimates from the CSDM method. Successful wildlife management hinges on the ability to properly quantify uncertainty. Underestimates of uncertainty can result in ill-informed management decisions. Our results highlight the additional ...