KappaMask (km_predict) ☁️

KappaMask, or km-predict, is a cloud detector developed by KappaZeta LTD for Sentinel-2 Level-1C and Level-2A input products. The project was funded by European Space Agency, Contract No. 4000132124/20/I-DT.

Currently, KappaMask outputs results in TIFF and PNG formats. Each pixel is classified as one of the following classes:

Class	TIFF	PNG	Description
Clear	1	66	Pixels without clouds or cloud shadows.
Cloud shadow	2	129	Pixels with cloud shadows.
Semi-transparent	3	192	Pixels with thin clouds through which the land is visible; include cirrus clouds that are on the high cloud level (5-15km).
Cloud	4	255	Pixels with cloud; include stratus and cumulus clouds that are on the low cloud level (from 0-0.2km to 2km).
Missing	5	20	Missing or invalid pixels.

KappaMask has been trained and validated with the following dataset:

• Tetiana Shtym, Olga Wold, Marharyta Domnich, Kaupo Voormansik, Mariana Rohtsalu, Joosep Truupõld, Nikita Murin, Abdullah Toqeer, Catherine Akinyi Odera, Fariha Harun, Indrek Sünter, Heido Trofimov, Anton Kostiukhin, & Mihkel Järveoja. (2022). KappaSet: Sentinel-2 KappaZeta Cloud and Cloud Shadow Masks (Version 1) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.7100327

Related publications:

• Marharyta Domnich, Indrek Sünter, Heido Trofimov, Olga Wold, Fariha Harun, Anton Kostiukhin, Mihkel Järveoja, Mihkel Veske, Tanel Tamm, Kaupo Voormansik, Aire Olesk, Valentina Boccia, Nicolas Longepe, and Enrico Giuseppe Cadau. 2021. "KappaMask: AI-Based Cloudmask Processor for Sentinel-2" Remote Sensing 13, no. 20: 4100. https://doi.org/10.3390/rs13204100

Dependencies

The following system dependencies are needed:

• micromamba 0.18 or later (https://github.com/mamba-org/mamba)
• python 3.6 or later
• cm_vsm dependencies

Due to the long environment solve times with Miniconda, we have switched to Micromamba. If you're still using Conda, Miniconda or similar, simply substitute micromamba with conda in the relevant commands below.

Setup

1. Create a micromamba environment.

    micromamba create -f environment.yml
   

2. Copy config/config_example.json and adapt it to your needs.

3. In order to run sub-tiling procedure cm_vsm should be installed (https://github.com/kappazeta/cm-vsm).

4. Make sure that your GDAL_DATA environment variable has been set, according to your GDAL version instead of the placeholder YOUR_GDAL_VERSION below:

    GDAL_DATA=/usr/share/gdal/YOUR_GDAL_VERSION
   

Input data

In the root of repository create a /data folder and copy or symlink the .SAFE product into it.

Usage

Running locally

Cloudmask inference can be run as follows:

micromamba activate km_predict
python km_predict.py -c config/your_config.json

It is possible to overwrite product_name in config file with command line argument -product

python km_predict.py -c config/your_config.json -product S2B_MSIL2A_20200401T093029_N0214_R136_T34UFA_20200401T122148

If the prediction for the same product is running multiple times and .CVAT folder is created under /data folder, it might be convenient to disable sub_tiling procedure for the next run by -t

python km_predict.py -c config/your_config.json -product S2B_MSIL2A_20200401T093029_N0214_R136_T34UFA_20200401T122148 -t

Testing locally

KappaMask tests can be run from the root of the working copy of the repository as follows:

micromamba activate km_predict
pytest

Running in Docker on AWS

By default, the KappaMask Docker image runs the km_s3 entrypoint, which expects a Sentinel-2 product title and output path in an S3 bucket. The entrypoint performs the following operations:

1. Download the Sentinel-2 product in SAFE format from a Sentinel-Hub requester-pays dataset on AWS.
2. Split the product into sub-tiles with overlap and without compression.
3. Calculate the cloud mask for the product and mosaic the sub-tiles.
4. Tile and compress the resulting mosaic and generate overviews.
5. Upload the mask and sub-tiles to and S3 bucket.

KappaMask can be run as a Docker container as follows:

1. Pull the image

    docker pull kappazeta/kappamask:v2.3
   

2. Run KappaMask for a specific Sentinel-2 product on AWS (please make sure to replace YOUR-AWS-REGION, YOUR-AWS-ACCESS-KEY, YOUR-AWS-SECRET-KEY, YOUR-S3-BUCKET with your AWS configuration, and YOUR-S2-PRODUCT-NAME with the name of the product to process)

    docker run -e AWS_REGION=YOUR-AWS-REGION -e AWS_ACCESS_KEY=YOUR-AWS-ACCESS-KEY -e AWS_SECRET_KEY=YOUR-AWS-SECRET-KEY kappazeta/kappamask:v2.3 YOUR-S2-PRODUCT-NAME s3://YOUR-S3-BUCKET/
   

For example:

docker pull kappazeta/kappamask:v2.0
docker run -e AWS_REGION=eu-central-1 -e AWS_ACCESS_KEY=A******************F -e AWS_SECRET_KEY=3**************************************I kappazeta/kappamask:v2.3 S2A_MSIL2A_20200509T094041_N0214_R036_T35VME_20200509T111504 s3://my-kappamask-experiments/output/

Running in Docker locally

KappaMask Docker image can be run locally with the km_local entrypoint. The entrypoint performs the following operations:

1. Check if the .SAFE directory exists in the /data volume, or attempt to decompress a .zip file in the volume otherwise.
2. Split the product into sub-tiles with overlap and without compression.
3. Calculate the cloud mask for the product and mosaic the sub-tiles.
4. Tile and compress the resulting mosaic and generate overviews.

KappaMask can be run as a Docker container as follows:

1. Pull the image

    docker pull kappazeta/kappamask:v2.3
   

2. Run KappaMask for a specific Sentinel-2 product locally

    docker run -v /YOUR-LOCAL-DATA-DIR/:/data kappazeta/kappamask:v2.3 YOUR-S2-PRODUCT-NAME
   

For example:

docker pull kappazeta/kappamask:v2.3
docker run -v /home/kappazeta/Documents/data/cloudmask_data/:/data kappazeta/kappamask:v2.3 S2A_MSIL2A_20200509T094041_N0214_R036_T35VME_20200509T111504

Testing in Docker

KappaMask can be tested as a Docker container as follows:

1. Pull the image

    docker pull kappazeta/kappamask:v2.3
   

2. Run KappaMask tests:

    docker run --entrypoint km_test kappazeta/kappamask:v2.3
   

Output

The predictor will generate sub-tiles masks under /prediction folder and full S2 mask under /big_image folder

Troubleshooting

Potential solutions for typical issues encountered during setup or usage.

Unable to open EPSG support file

Sentinel-2 product splitting fails with the following messages:

INFO: KMP.P: Extracting geo-coordinates.
ERROR 4: Unable to open EPSG support file gcs.csv.  Try setting the GDAL_DATA environment variable to point to the directory containing EPSG csv files.
ERROR 4: Unable to open EPSG support file gcs.csv.  Try setting the GDAL_DATA environment variable to point to the directory containing EPSG csv files.
INFO: KMP.P: Projection:
terminate called after throwing an instance of 'INFO: KMP.P: Projecting AOI polygon into pixel coordinates.
GDALOGRException'
  what():  GDAL OGR error : Failed to import spatial reference from EPSG, Generic failure
Magick: abort due to signal 6 (SIGABRT) "Abort"...

This indicates that the environment variable GDAL_DATA has not been configured correctly. This could be done in a variety of ways and the preferred method depends on your linux distribution. An export call for the variable (for example, GDAL_DATA=/usr/share/gdal/2.2) could be added to your .bashrc, .profile, etc. Alternatively, the variable could be set together with the python call, for example:

GDAL_DATA=/usr/share/gdal/2.2 python km_predict.py -c config/your_config.json

Filesystem error

Sentinel-2 product splitting fails with the following messages:

terminate called after throwing an instance of 'std::filesystem::__cxx11::filesystem_error'
  what():  filesystem error: directory iterator cannot open directory: No such file or directory [YOUR_DIRECTORY/km_predict/data/S2B_MSIL1C_20200401T093029_N0209_R136_T34UFA_20200401T113334.SAFE.SAFE/GRANULE/]
Magick: abort due to signal 6 (SIGABRT) "Abort"...

This means that km_predict cannot find the directory with the product_name specified in the configuration file. The product name in the configuration file should be provided without the .SAFE suffix.