Interplay between physical and predator landscapes affects transferability of shorebird distribution models

Coastal landscapes with extensive intertidal mudflats provide non-breeding habitat for Arctic shorebirds. Few attempts have been made to develop and test landscape-level models predicting the intertidal distribution of these birds. We modelled the distribution of a Holarctic species, Dunlin ( Calidris alpina ), at a hemispherically important non-breeding site, the Fraser River Delta, British Columbia, Canada, in seasons with different predator landscapes. We trained the models during a season when nocturnal predators were common and tested temporal transferability of the models on independent datasets when nocturnal predators were absent. Snowy Owls ( Nyctea scandiaca ) influenced Dunlin distribution and thus model transferability. After accounting for their presence, models displayed good to excellent discrimination, i.e. prediction of the instantaneous and cumulative (over low tide period) probability of mudflat use by Dunlin, in fore- and backcasting applications. Model calibration was good or else, where over-prediction was observed, the reason for the bias was identified. The distribution models may predict mudflat use by Dunlin and possibly related species given relevant data describing the intertidal landscape. The models are amenable to GIS application, describe the amount of use per hectare of the intertidal zone and can be used to determine and visualise relative and absolute suitability of intertidal areas.