Sentinel 1 pre-processing pipeline using SNAP GPT 7.0

NB: This file is auto-generated using pip freeze > requirements.txt

We are using Graph Processor Tool (GPT) to process the sentinel-1 datasets.

We use Copernicus Open Access Hub to download the metadata available for the images. Change your username and password on .env file variable COPERNICUS_USERNAME and COPERNICUS_PASSWORD

script: download_image_metadata.py

We use Alaska Satellite Facility (ASF) Distributed Data Archive Center (DAAC) to download the sentinel-1 SAR scenes. This blog post has detailed instructions on how you can go about creating an account and downloading via interface the required scene. We modified the python scripts provided by the ASF DAAC to download batch scenes especially utilizing the metadata collected in 1.

All the images are saved in dir images.

script: download_image.py

For co-registration of images, we need slaves images to co-registered against. Update the database with the slaves images where column slave is changed to True. Use script 1. and 2. to get the metadata and download the slave images.

Example of footprint of set of slaves images we used.

The slaves are processed using process_slaves.py and the processed slaved are stored in dir slaves. The slaves are processed with step a-f of 5.

The processing steps with the parameters value are:

a. Apply Orbit File

b. Thermal Noise Removal

i.  removeThermalNoise: true
ii. reIntroduceThermalNoise: false

c. Border Noise Removal

i.  borderLimit: 1000
ii. trimThreshold: 0.5

d. Calibration

i.   outputSigmaBand: false
ii.  outputGammaBand: false
iii. outputBetaBand: true

e. Multilook

i.   No. of Range Looks (nRgLooks): 1
ii.  No. of Azimuth Looks (nAzLooks): 1
iii. outputIntensity: true
iv.  grSquarePixel: true

f. Radiometric Terrain Flattening

i.   demName: SRTM 1Sec HGT
ii.  demResamplingMethod: BICUBIC_INTERPOLATION
iii. oversamplingMultiple: 1.5
iv.  additionalOverlap: 0.1

g. DEM Assisted Co-registration

NB: The slaves list are obtained by selecting those images processed in 4. that intersects with the scene currently processed.

i.   demName: SRTM 1Sec HGT
ii.  demResamplingMethod: BICUBIC_INTERPOLATION
iii. resamplingType: BICUBIC_INTERPOLATION
iv.  tileExtensionPercent: 100
v.   maskOutAreaWithoutElevation: true

h. Speckle Filtering

i.   filter: Refined Lee

i. Range-Doppler Terrain Correction

i.   demName: SRTM 1Sec HGT
ii.  nodataValueAtSea: false
iii. pixelSpacingInMeter: 10
iv.  pixelSpacingInDegree: 8.983152841195215E-5

The processed images are then subset to get the master images and exported to individual polarisation bands in the dir output

This is accomplished by process.py The path to snappy is provided in .env as SNAPPY_PATH. snappy is used to read the bandnames of the files processed before co-registration, but the actual processing is done via SNAP GPT.

Specify the Google Cloud Storage path on GCS_PATH and the GEE Path on GEE_ASSET_PATH env variable.

The post-processing includes using GDAL to scale the output files by factor of 10000 and converting to unsigned integer 16 bit so save storage space in the GEE. The GDAL path is specified with GDAL_PATH variable.

The files are then copied to Google Cloud Bucket with gsutil installed and pushed to Earth engine with the manifest upload. The upload includes VV and VH or relevant polarisation of the same scene as different bands with needed metadata obtained in 1. as properties of the image.