Climate change is altering the marine environment at a global scale, with some of the most dramatic changes occurring in Arctic regions. These changes may affect the distribution and migration patterns of marine species throughout the annual cycle. Species distribution models have provided detailed understanding of the responses of terrestrial species to climate changes, often based on observational data; biologging offers the opportunity to extend those models to migratory marine species that occur in marine environments where direct observation is difficult. We used species distribution modelling and tracking data to model past changes in the non-breeding distribution of thick-billed murres Uria lomvia from a colony in Hudson Bay, Canada, between 1982 and 2019. The predicted distribution of murres shifted during fall and winter.

The largest shifts have occurred for fall migration, with range shits of 211 km west and 50 km north per decade, compared with a 29 km shift west per decade in winter. Regions of range expansions had larger declines in sea ice cover, smaller increases in sea surface temperature, and larger increases in air temperature than regions where the range was stable or declining. Murres migrate in and out of Hudson Bay as ice forms each fall and melts each spring. Habitat in Hudson Bay has become available later into the fall and earlier in the spring, such that habitat in Hudson Bay was available for 21 d longer in 2019 than in 1982. Clearly, marine climate is altering the distribution and annual cycle of migratory marine species that occur in areas with seasonal ice cover.

See paper for full methods describing data processing: https://doi.org/10.3354/meps13890

meps13890_SDM.RDS:
This is an R data file containing the species distribution model described in the paper. It is a train class object created with the caret package (version 6.0-84) and the ranger package (version 0.11.2) using (R version 3.6.1). 

meps13890_sdm_model_train_data.csv: 
Training data used in developing the species distribution model. Includes used and pseudo absence locations. Used locations are all 100 possible tracks calculated using the probGLS algorithm. Absence locations are randomly generated points paired with each possible location. Data include extracted environmental predictors at each location. Data fields are described in 

meps13890_sdm_model_test_data.csv: 
Training data used in developing the species distribution model. Includes used and pseudo absence locations. Used locations are all 100 possible tracks calculated using the probGLS algorithm. Absence locations are randomly generated points paired with each possible location. Data include extracted environmental predictors at each location.

meps13890_data_description.csv:
Data descriptions for all columns in meps13890_sdm_model_train_data.csv and meps13890_sdm_model_test_data.csv

meps13890_SDM_predictions:
Geotiff file with all predicted probability of occurrence values at 3-day intervals from 1982-2019, for the area 115W to 5W and 30N to 80N. Make sure to keep the .xml file, which contains layer names (dates), with the .tif file.