Models incorporating non-stationarity improve detection of climate-driven range shifts in odontocetes

Aim: Climate change is causing distributional shifts in many species
globally. Identifying and anticipating these shifts is critical to
understanding ecosystem impacts and implementing successful management
strategies. Species distribution models (SDMs) are useful tools often
employed to describe current and changing habitat use, particularly for
marine predators. However, most SDMs assume the statistical relationships
between species and their environment are temporally static, which may not
be true. We examined how incorporating temporal variability improved SDM
performance and estimated range shifts for six odontocete species. We used
a high-performing model to quantify changes in odontocete distribution
over a 24-year period. Location: Waters of the United States, east coast,
from Florida to Nova Scotia. Methods: We utilized nearly 1.4 million
kilometers of line transect survey data collected from 1997-2020 along the
east coast of the United States to evaluate changes in the distribution of
six odontocete species. We assessed six model specifications of generalize
additive models that varied in the extent of temporal and spatial
variability incorporated. Results: We found that the best performing model
specifications included temporally dynamic species-environment
relationships and temporally dynamic spatial terms. These model
specifications identified significant poleward range shifts in all species
for which we had sufficient data across their range. In contrast, model
specifications which only included static terms performed poorly and
identified limited or no spatial shifts. Main conclusions: These results
advance our predictive capabilities from static species-environment
relationships for marine predators and demonstrate the importance of
carefully considering assumptions and model specifications when modeling
changes to distributions. The odontocete range shifts we identified are
likely to have substantial ecosystem impacts, and the framework we present
offers a diagnostic approach for modeling and identifying range shifts in
other wide-ranging species.