Data accompanying "Pre-fire Vegetation Conditions and Topography Shape Burn Mosaics of Siberian Tundra Fire Scars"

This dataset contains the data used in the burned area classificaiton and statistical analysis in:

Rietze et al. (in prep.) - Pre-fire Vegetation Conditions and Topography Shape Burn Mosaics of Siberian Tundra Fire Scars

All code to preprocess, analyze and visualize this data can be found under https://github.com/nrietze/SiberiaFires.

Folder structure:

This dataset contains three major components:

The feature layers like training polygons used for the burned area classification and areas of interest.

The burned area maps produced from 3 m PlanetScope imagery.

The raster layers used in the statistical analysis, e.g., pre-processed digital elevation models and burned area products.

└───geodata
├───feature_layers
│ ├───aoi_wv
│ ├───burn_polygons
│ │ └───planet
│ └───training_polygons
└───raster
├───arcticDEM
├───burned_area
│ └───planet
├───landsat
├───predictors
└───water_area
└───planet

Detailed description of content:

Feature layers

planet_masks.shp
Manually delineated polygons to mask out undetected clouds in the Berelech and Lapcha sites.

aoi_wv/aois_analysis.geojson
Square areas of interest for data cropping and selection.

burn_polygons/planet/rough_burn_perimeter_{fire scar name}.shp
6 shapefiles (1 per fire scar) that contain the fire perimeter generated in ArcGIS.

training_polygons/training_polygons_burn_area.shp
Training polygons for all fire scars used for the burned area classification.

Raster data

arcticDEM/aoi_{fire scar name}_dem_v3_utm.tif
Elevation data (ArcticDEM v3) reprojected to UTM 55N and cropped to the square areas of interest of each fire scar. 

burned_area/ba_descals_landsat_2020_utm_shifted.tif
Landsat-based burned area from Descals et al. (2022), contains data from "Tile 17" only. Dara was reprojected to UTM 55N align with the Landsat-8 Collection-2 Level-2 grid. (burned class (2019) = 29, burned class (2020) = 30, unburned = 0)

burned_area/N75E145_burn_class_UTM_55N.tif
Landsat-based burned area from Wei et al. (2022). Dara was reprojected to UTM 55N align with the Landsat-8 Collection-2 Level-2 grid. (burned class = 255, unburned = 0)

burned_area/planet/{fire scar name}_burned_area_top5TD.tif
Binary PlanetScope-based burned area (this study) for each fire scar based on random forest classifiers using the top 5 predictors ranked by transformed divergence (TD).  (burned class = 2, unburned = 1)

landsat/LC08_L2SP_115010_20200608_20200824_02_T1_{spectral or quality band}.TIF
Landsat-8 Collection-2 Level-2 pre-fire image from 8 June 2020. B4 = RED, B5 = NIR, ST_B10 = LST

landsat/LC08_L2SP_116010_20200615_20200824_02_T1_{spectral or quality band}.TIF
Landsat-8 Collection-2 Level-2 pre-fire image from 15 June 2020. B4 = RED, B5 = NIR, ST_B10 = LST

predictors/{fire scar name}_predictors_30m.tif
Raster maps of model predictors & the response burned_fraction for the ZOIB model (exported in "ZOIB_model.R" before running the model). Raster bands are named.

water_area/{fire scar name}_Landsat_mask.tif
Landsat-8 binary (water = 2, clear pixels = 1) water mask for each fire scar based on the Quality assessment rasters.

water_area/planet/{fire scar name}__water_area_top5TD.tif
[deprecated] 3 m resolution water areas classified from PlanetScope imagery. Not used for the analysis.

Using this data:

Clone the Github repository before downloading this data and insert the contents of this dataset into the empty "data" folder from the Github repo. 

Important: Please move the entire "geodata" folder into the "data" folder from the code repo.

If you use this data, please cite as follows:

Rietze, N., Heim, R. J., Troeva, E., Schaepman-Strub, G. & Assmann, J. J. (in prep.). Data accompanying "Pre-fire Vegetation Conditions and Topography Shape Burn Mosaics of Siberian Tundra Fire Scars" [Data set]. Zenodo. https://doi.org/10.5281/zenodo.12650945

Abstract (from manuscript):

The fire season of 2020 in Siberia set a precedent for extreme wildfires in the Arctic tundra. Large fires burned in the carbon-rich permafrost landscape, releasing vast amounts of carbon, and changing land surface processes by burning vegetation and organic soils. However, little is known about the mosaics of burned and unburned patches formed by tundra fires and the underlying processes that generate them. In this study, we investigated six fire scars in the northeastern Siberian tundra using high-resolution PlanetScope imagery (3 m) to map burned fraction within the scars. We then used Bayesian mixed models to identify which biotic and abiotic predictors influenced the burned fraction. We observed high spatial variation in burned fraction across all tundra landforms common to the region. Current medium-resolution fire products could not capture this heterogeneity, thereby underestimating the burned area of fire scars by a factor of 1.1 to 4.4. The heterogeneity of the burn mosaic indicates a mix of burned and unburned patches, with median unburned patch sizes being smaller than 180 to 324 m². Pre-fire land surface temperature, vegetation heterogeneity and topography predicted burn fraction in our analysis, matching factors previously shown to influence large-scale fire occurrence in the Arctic. Future studies need to consider the fine-scale heterogeneity within tundra landscapes to improve our understanding and predictions of fire spread, carbon emissions, post-fire recovery and ecosystem functioning.

Acknowledgements (from manuscript):

N.R. was supported through the TRISHNA Science and Electronics Contribution (T-SEC), ESA PRODEX Trishna T-SEC project (PEA C4000133711). Field work and vegetation sample processing were conducted in the scope of State Assignment of the Ministry of Science and Higher Education of the Russian Federation (Project AAAA-A21-121012190038-0), using the equipment of the Centre for collective use of Federal Research Centre "Yakut Scientific Centre" (grant no. 13.TsKP.21.0016). We would like to thank Planet Labs fo free access to PlanetScope imagery. We would like to thank Tim Gyger for helpful discussions regarding our statistical analysis. The authors declare no competing interests.