Pre-Processed Dual-Pol Sentinel-1 SAR Dataset for Machine Learning-Based Burned Area Mapping
Description
This dataset consists of pre-processed dual-polarization Sentinel-1 SAR data prepared for machine learning-based burned area mapping in the Palisades, Los Angeles region. The data includes VV and VH backscatter from ascending and descending orbits, as well as derived GLCM texture features computed for both pre- and post-fire periods. Additionally, it provides the difference Normalized Burn Ratio (dNBR) for preparing refernce map and an RGB composite for visualization. The code for burned area mapping using machine learning with this dataset is available at: https://github.com/rabinatwayana/SAR-Burnt-Area-Mapping-ML.
Files
Steps to reproduce
Six dual-polarized IW GRD SAR images for both ascending and descending acquisition modes, on the dates 2024-12-09, 2024-12-21, and 2025-01-02 (before the event), and the dates 2025-01-14, 2025-01-26, and 2025-02-07 (after the event) were downloaded from the Alaska Satellite Facility Data search portal (https://search.asf.alaska.edu/). The ESA Sentinel Application Platform (SNAP) version 11 was used to perform the pre-processing of Sentinel-1 image datasets, following the standard workflow (https://doi.org/10.3390/ECRS-3-06201). Ascending and descending image sets were processed separately. The steps followed are described briefly below. 1. A subset of the area of interest was extracted. 2. The orbit file was applied to improve geolocation accuracy by incorporating updated satellite position and velocity information. 3. The thermal noise removal was implemented. 4. The border noise removal was applied to remove low-intensity noise and invalid data on scene edges. 5. Radiometric calibration was executed to convert digital pixel values into sigma nought values. 6. All six images were co-registered by using the co-registration tool with settings resampling type (bilinear interpolation), initial offset method (product geolocation), and output extent (minimum). 7. Single product speckle filtering using Lee sigma filter, window size of 7×7, and an output window of 3×3 was applied. 8. Range-Doppler Terrain Correction was performed using SRTM 1Sec HGT as a Digital Elevation Model (DEM), bilinear interpolation as a method for resampling, default CRS (WGS 84), pixel spacing of 10m and without masking out areas with missing elevation. 9. The GLCM texture properties were also computed for both VV and VH polarizations separately using a 5×5 window and a probabilistic quantizer set to 8 quantization levels. 10. The final products (backscatter and GLCM) were exported in .geotiff format and loaded into QGIS. 11. Finally, the result was resampled into 10m and reprojected into UTM zone 11N (EPSG:32611) to maintain spatial accuracy and consistency for regional analysis. For reference purposes, Sentinel-2 Level 2A analysis-ready images, acquired on 2025-01-02 (before the event) and 2025-01-12 (after the event), have been downloaded from the Copernicus browser. Pre-processing of the labels space data to generate a reference map. For that, the Differenced Normalized Burn Ratio (dNBR) was calculated according to: dNBR=(NBR_post_fire) - (NBR_pre_fire) ......(Equation 1) where NBR_pre_fire and NBR_post_fire account for the Normalized Burn Ratio for pre-fire and post-fire events computed from near-infrared (NIR) and shortwave infrared (SWIR) bands of Sentinel 2 with the following equation: NBR=(NIR-SWIR)/(NIR+SWIR)...........(Equation 2) The NIR (Band 8) and SWIR (Band 12) bands have differing spatial resolutions of 10 meters and 20 meters, respectively. Therefore, resampling of Band12 was performed to a 10-meter resolution using bilinear interpolation.
Institutions
- Universitat Salzburg