Cloud masking

Sentinel-2 cloud mask

The exclusion of clouds and cloud shadow is an important processing step which is usually done in an early pre-processing stage. The Sentinel-2 products are annotated with Quality Indicators (QI_DATA). In the Level 2A product you may find a file in the folder QI_DATA with the extension _CLD_20m.jp2 (data type: 8bit unsigned integer, spatial resolution 20m) which might be used for masking clouds. The data range is from 0 for high confidence clear sky to 100 for high confidence cloudy (for more details see L2A Product Definition Document). If you downloaded a Level 1C product, you'll find a rough dense cloud and cirrus mask (GML vector format) in the folder QI_DATA with the file name MSK_CLOUDS_B00.gml and a spatial resolution of 60m. In the vector attribute column masktype distinguishes two classes: OPAQUE (dense clouds) and CIRRUS.

• Load a multiband Sentinel-2 satellite image (Level 2-A product) Changing Raster Layer Style to display a true color and a false color composite.
• Zoom in to a cloudy part of the image and compare the extent of cloud and cloud shadow in both composites. Which composite is better for displaying clouds?

• Overlay the QI Cloud band on top of the color composites. Use the Identify Feature Tool for requesting pixel values of the QI band file inside a cloud. Which values represent clouds?
• First, Layer Properties --> Style --> Min/max value settings of the QI layer. Tick Min / max and click Apply. Then change the Render type' from Singleband grey to Singelband pseudocolor with interpolation discrete and mode equal with 10 interval classes as shown in the screenshot.
• Double click on a color field of some classes {{mitem|text=Layer --> Properties --> Style) to open the change color window. Change the opacity from 100% to 0%. Find the best threshold for clouds.

Refining cloud mask

• Create a binary cloud mask (bitmask) where clouds = 1 and non-clouds = 0. Use the Processing tool OTB --> Miscellaneous --> Band Math. Tick the single band QI layer in the Input Image list list. You may write a conditional ifelse expression using lazy operators. Type into the Expression field: (im1b1>30)?1:0. If pixel values of band 1 of multiband image 1 are larger than 30 then replace with 1 otherwise replace with 0. Mark the output layer in the TOC and click to stretch the histogram to view extents.

]

• Apply a circular majority filter with a radius of 7 pixels. Open the Processing tool SAGA --> Raster filter --> Majority filter. The input Grid is the Output image of the band math operation step. Change the Search mode to Circle with an radius of 5 pixels. Mark the Filtered Grid in the TOC and click to stretch the histogram to view extents.

• (This Processing Module does currently not work) Open the Processing tool OTB --> Feature Extraction --> BinaryMorphologicalOperation (dilate). Choose the Filtered Grid as input image. The structuring element type is ball with a radius of 7 pixels. Mark the Feature Output Image in the TOC and click to stretch the histogram to view extents.

• Convert the Feature Output Imge from raster to vector format. Raster --> Conversion --> Polygonize. Select an output file name for vector polygons.

• Change the display of the output vector polygon layer Layer properties --> Style --> Single Symbol --> Simple Fill --> Fill Style from Solid to No Brush. Change the Outline color to bright green and increase the Outline width. As background layer use a False Color Composite of the Sentinel-2 image.
• Toggle editing of the resulting vector polygon layer. Activate Select polygons by dragging a box or by single click , select all polygons with attribute DN = 0 or which are misclassified as clouds and delete them by pressing Del on your keyboard. Save your edits .

Creating cloud shadow mask

• Prepare a False color composite of the Sentinel-2 image as background layer in the canvas.
• Right click on the edited polygon layer in the TOC. Choose Export --> Save features as.... Fill the form as shown below.

• Mark the new created polygon layer in the TOC and toggle editing of the vector file. Select all polygons using or press Ctrl + A. Activate Edit --> Move feature(s) . Left click on the canvas and move the cloud polygons that shadows on the background false color composite are completely covered. Save your edits and stop editing .
• Select polygons outside the image extent of the false color composite and delete them by pressing Del on your keyboard. Save your edits and stop editing .

Applying cloud and cloud shadow mask

• Vector --> Data management tools --> Merge vector layers. as Layer to merge select the cloud and cloud_shadow polygon layers. Run.

• Create a polygon with the extent of the Sentinel-2 image: Create image extent polygon.
• Vector --> Geoprocessing Tools --> Symmetrical difference. Input layers are the image extent and the merged cloud and cloud_shadow polygons.

• Apply the output vector layer Cleaned as mask to the multiband Sentinel-2 image file. Raster --> Extraction --> Clip Raster by Mask Layer layers.

Graphical Processing Tool Model

You can use a prepared graphical processing toolbox model for cloud detection set: S2 Cloud detection.model.

1. Open QGIS and go to the Processing Toolbox; if the Toolbox is not opened go to Processing --> Toolbox to activate it. Once the Toolbox is opened go to Models --> Open existing models and load the previously downloaded model.
2. The model should appear in the Models tab. Double click on it to execute.
3. A window showing the different input and outputs should appear, just as if you were using any other geo-algorithm of the Processing Toolbox.
4. You will just need to provide a multispectral Sentinel-2 image file with 10 ordered bands (B2,B3,B4,B5,B6,B7,B8,B8a,B11,B12).

Figure A: False color composite RGB=8,4,3

Figure B: False color composite RGB=8,4,3 masked with cloud and cloud shadow