|
Download EDA/feature.md from kabiriitb/rockglacierNET: direct link, hf CLI and curl.
- Browser
- Download file 2.99 kB
-
https://huggingface.co/kabiriitb/rockglacierNET/resolve/main/EDA/feature.md
- Command line
-
hf download hf://kabiriitb/rockglacierNET/EDA/feature.md
-
curl -L -o feature.md https://huggingface.co/kabiriitb/rockglacierNET/resolve/main/EDA/feature.md
2.99 kB
Feature Stack Specifications
This document outlines the internal structure and metadata of the multi-band GeoTIFF files extracted from Google Earth Engine for the Rock Glacier Detection project.
File Format & Metadata
- Format: Multi-band GeoTIFF (
.tif) - Coordinate Reference System (CRS):
EPSG:32645(WGS 84 / UTM zone 45N) - Spatial Resolution:
10.0meters per pixel - Spatial Extent: Variable per rock glacier. The bounds strictly cover the bounding box of the rock glacier polygon plus a precise
200mbuffer on all sides.
Band Mapping (12-Band Stack)
Every exported feature file (e.g., feature_0000.tif) contains exactly 12 aligned bands. Because GeoTIFFs often strip band names during Earth Engine export, this index mapping is required for loading the data into PyTorch/TensorFlow.
| Band Index | Feature Name | Data Source | Description |
|---|---|---|---|
| 1 | B2 |
Sentinel-2 (SR) | Blue Band |
| 2 | B3 |
Sentinel-2 (SR) | Green Band |
| 3 | B4 |
Sentinel-2 (SR) | Red Band |
| 4 | B8 |
Sentinel-2 (SR) | Near Infrared (NIR) |
| 5 | B11 |
Sentinel-2 (SR) | Shortwave Infrared (SWIR) 1 |
| 6 | B12 |
Sentinel-2 (SR) | Shortwave Infrared (SWIR) 2 |
| 7 | NDVI |
Derived (S2) | Normalized Difference Vegetation Index (B8 - B4) / (B8 + B4) |
| 8 | NDWI |
Derived (S2) | Normalized Difference Water Index (B3 - B8) / (B3 + B8) |
| 9 | NDSI |
Derived (S2) | Normalized Difference Snow Index (B3 - B11) / (B3 + B11) |
| 10 | Elevation |
Copernicus DEM | Elevation in meters (GLO-30) |
| 11 | Slope |
Derived (DEM) | Terrain slope in degrees |
| 12 | Aspect |
Derived (DEM) | Terrain aspect in degrees |
Temporal Processing
- Timeframe: 3-year aggregate (
2021-01-01to2024-01-01). - Cloud Masking: Strict pixel-level masking using the Sentinel-2
QA60band (removing both opaque and cirrus clouds). - Compositing: A
.median()reducer is applied across the 3-year stack to ensure a perfectly clear, snow-free (summer-biased), and cloud-free representation of the surface.
Actual Dataset Values (Sample from feature_0000.tif)
Below are the actual numbers (Min, Max, and Mean) calculated directly from the extracted pixels inside feature_0000.tif:
| Band Index | Feature Name | Min Value | Max Value | Mean Value |
|---|---|---|---|---|
| 1 | B2 |
0.1843 | 0.7763 | 0.5603 |
| 2 | B3 |
0.1761 | 0.7384 | 0.5447 |
| 3 | B4 |
0.1900 | 0.7294 | 0.5364 |
| 4 | B8 |
0.1899 | 0.6998 | 0.5119 |
| 5 | B11 |
0.1028 | 0.4263 | 0.2215 |
| 6 | B12 |
0.0954 | 0.3540 | 0.1999 |
| 7 | NDVI |
-0.0783 | 0.1143 | -0.0220 |
| 8 | NDWI |
-0.1570 | 0.0973 | 0.0292 |
| 9 | NDSI |
-0.1862 | 0.6793 | 0.4111 |
| 10 | Elevation |
4643.9253 | 5007.4976 | 4790.8872 |
| 11 | Slope |
0.1379 | 57.5960 | 22.2821 |
| 12 | Aspect |
0.0065 | 359.9010 | 165.6444 |