MEPS 413:163-183 (2010)  -  DOI: https://doi.org/10.3354/meps08696

ABSTRACT: Generalized linear and generalized additive habitat models were used to predict cetacean densities for 10 species in an 818000 km² area off California. The performance of models built with remotely sensed oceanic data was compared to that of models built with in situ measurements. Cetacean sighting data were collected by the Southwest Fisheries Science Center on 4 systematic line-transect surveys during the summer and fall of 1991, 1993, 1996, and 2001. Predictor variables included temporally dynamic, remotely sensed environmental variables (sea surface temperature and measures of its variance) and more static geographical variables (water depth, bathymetric slope, and a categorical variable representing oceanic zone). The explanatory and predictive power of different spatial and temporal resolutions of satellite data were examined and included in the models for each of the 10 species. Alternative models were built using in situ analogs for sea surface temperature and its variance. The remotely sensed and in situ models with the highest predictive ability were selected based on a pseudo-jackknife cross validation procedure. Environmental predictors included in the final models varied by species, but, for each species, overall explanatory power was similar between the remotely sensed and in situ models. Cetacean–habitat models developed using satellite data at 8 d temporal resolution and from 5 to 35 km spatial resolution were shown to have predictive ability that generally met or exceeded models developed with analogous in situ data. This suggests that the former could be an effective tool for resource managers to develop near real-time predictions of cetacean density.

KEY WORDS: Cetacean density · Habitat modeling · GAM · GLM · California Current · Remote sensing · Whale · Dolphin · Porpoise