Fire refugia predictions for the Northwest Territories and Yukon

Data description: This data consists of 2 GEOTIFF rasters covering the majority of the Northwest Territories and Yukon, as well as northern portions of British Columbia, Alberta, Saskatchewan, and Manitoba, Canada. Projection is NAD83/Yukon Albers (EPSG:3578), resolution is 120 m. Additionally, there is a single table describing covariate names, temporal periods, sampling methods, and data sources. Data included: Rasters of predicted probability of fire refugia under average climate conditions, as well as those based on static physical setting features (e.g., terrain, soil texture, surrounding wetland proportions, etc), for the full study area. Unvegetated areas, human disturbance, and open water have been masked out. Topography_Preds_Masked_V5.tif Average_Preds_Masked_V5.tif Word document containing a table detailing the various covariates used in analyses, as well as their temporal period, sampling method, and data sources. Variable table.docx A publicly available web application, created through the Google Earth Engine App program, can be found at: https://ee-cek-nwt-fire-refugia.projects.earthengine.app/view/predicted-fire-refugia-probabilities-across-nwt-and-beyond. This app includes visualizations of each of the predictive maps. Methods Summary: We fit a series of boosted regression tree models (Elith et al. 2008) to determine the relative importance of top-down and bottom-up controls on fire refugia probability for each of 16 fire regime units (FRU, Erni et al. 2020) across the study area. Fires were sampled via randomly generated points representing 1% of fire pixels (30-m resolution), spanning 2002-2019. We extracted point (e.g., terrain, climate, fire weather, soils) and moving-window fuel attribute variables at each sample location. Fuel attribute variables were extracted using square-shaped moving windows of 300 m or 1200 m on a side. All processing and extraction of the covariates was conducted in Google Earth Engine (Gorelick et al. 2017). Fire sampling and model development was conducted using R version 4.5.1 (R Core Team 2025). Final models were used to create predictive maps of fire refugia probability in each FRU under average climate conditions, as well as those based on static physical setting features. Methods are based on and adapted from those in Kuntzemann et al. 2025: https://doi.org/10.1002/ecs2.70385. Preliminary R code for this project can be accessed via GitHub: https://github.com/CeKmann/NWT-Fire-Refugia-Code. References: Elith J, Leathwick JR, Hastie T. 2008. A working guide to boosted regression trees. Journal of Animal Ecology 77:802–813. Erni S, Wang X, Taylor S, Boulanger Y, Swystun T, Flannigan M, Parisien M-A. 2020. Developing a two-level fire regime zonation system for Canada. Canadian Journal of Forest Research:259–273. Gorelick N, Hancher M, Dixon M, Ilyushchenko S, Thau D, Moore R. 2017. Google Earth Engine: Planetary-scale geospatial analysis for everyone. Remote Sensing of Environment 202:18–27. Kuntzemann, C. E., Whitman, E., Lewis, D., Stralberg, D. (2025). Climate, topography, or fuels? Top‐down versus bottom‐up controls on fire refugia across British Columbia, Canada. Ecosphere, 16(9), e70385. R Core Team. 2025. R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna, Austria. Available from https://www.R-project.org/.