Habitat characterization allows prediction of dolphin distributions in response to oceanographic processes and can be used to understand and predict effects of anthropogenic disturbances. Many habitat models focus on contemporary dolphin occurrence and environmental predictor data, but time-lagged oceanographic data may increase a model's predictive power due to ecological successional processes. Using hourly occurrence of Risso's dolphin Grampus griseus clicks and 2 types of Pacific white-sided dolphin Lagenorhynchus obliquidens clicks in autonomous passive acoustic recordings, we investigate the importance of time-lagged predictor variables with generalized additive models. These models relate dolphin acoustic activity from recordings at 6 sites in the Southern California Bight between August 2005 and December 2007 to oceanographic variables including sea surface temperature (SST), SST coefficient of variation (CV), sea surface chlorophyll concentration (chl), chl CV, upwelling indices, and solar and lunar temporal indices. The most consistently selected variables among the trial models evaluated during cross-validation were SST (100% of models) and SST CV (80%) for Risso's dolphin clicks; solar indices (100%) and SST and SST CV (60% each) for Pacific white-sided type A (PWS A) clicks; and SST CV (100%), solar indices (100%) and SST (80%) for Pacific white-sided type B (PWS B) clicks. Best predictive models for Risso's dolphins and PWS A clicks included time-lagged variables, suggesting the importance of ecological succession between abiotic variables and dolphin occurrence, while best models of PWS B clicks were for current conditions, suggesting association with prey-aggregating features such as fronts and eddies.
KEY WORDS: Risso's dolphin · Grampus griseus · Pacific white-sided dolphin · Lagenorhynchus obliquidens · Habitat model · Generalized additive model · Passive acoustic monitoring · Southern California BightResale or republication not permitted without written consent of the publisher Mar Ecol Prog Ser 423: 247-260, 2011 and cost constraints. The inability to sample during night and rough weather conditions may affect the accuracy or precision of resulting models. Additionally, visual survey data are limited to times when cetaceans are conspicuous at the surface. Models based on these data may not represent 'prime' habitat (i.e. habitat that is important to feeding or breeding, and hence the animals' overall fitness) if the animals are more easily observed when resting or transiting to such key areas (Hamazaki 2002). Alternate sampling methods, such as passive acoustic monitoring, may provide a better representation of cetacean presence and activity in prime habitats and thereby improve model accuracy and precision.The accuracy and precision of habitat models is also dependent on the environmental data upon which they are built. Habitat models developed for predictive purposes should include environmental variables that are easily accessible and available across regions and times of int...