SenSeIv2 Cloud Mask

SEnSeI (Spectral ENcoder for Sensor Independence) is a deep learning algorithm, designed to deliver highly accurate cloud and cloud shadow. Unlike traditional rule-based methods, SEnSeI learns complex spectral and spatial patterns, allowing it to reliably identify thin clouds, thick clouds, and shadow regions even in challenging atmospheric or lighting conditions.

Sentinel-2 True Color
SEnSeI 4-class mask: clear skies (green), thick clouds (purple), thin clouds (light blue, Shadows (red)

Input products

  • Sentinel-2 L1C
  • Sentinel-2 L2A
  • Landsat 8/9 – products must contain bands from both instruments OLI and TIRS  

Input parameters

Parameter nameDescription
output_stylea. 4-class: pixel values represent:
0 - Clear Sky,
1 - Thick Clouds, 
2 - Thin Clouds,
3 - Cloud Shadows
b. cloud-noncloud: pixel values represent: 
0 - non-clouds (classes 0 and 3 from the 4-class option),
1- clouds (classes 1 and 2 from the 4-class option)
c. valid-invalid: pixel values represent:
0 - valid (classes 0 from the 4-class option),
1 - invalid (classes 1, 2, and 3 from the 4-class option)  
strideA size of the moving window to be used in inference step in pixels,
smaller value might give better results but increases processing times and price.
The value shall be one of 64, 128, 256, 384 (default), 512.
categoriseTRUE: Returns categorical values based on output_style Range:
[0–3 or 0–1 depending on mode]
FALSE: Returns probability values (%)
Range: [0–100]
output_storagePRIVATE: results delivered to your S3 bucket (S3 keys required).
TEMPORARY: results available for download.  Download link will be valid for 14 days.

Output

Output is delivered as a Cloud Optimized GeoTIFF (COG) file.

Pixel values represent cloud classification depending on selected parameters.

  • NoData value: 255
  • Spatial resolution:
    • 10 m (Sentinel-2)
    • 30 m (Landsat)

References:

  1. SenSeIv2 on GitHub: https://github.com/aliFrancis/SEnSeIv2?tab=readme-ov-fileGihut
  2. A. Francis, "Sensor Independent Cloud and Shadow Masking With Partial Labels and Multimodal Inputs," in IEEE Transactions on Geoscience and Remote Sensing, vol. 62, pp. 1-18, 2024, Art no. 5405018, doi: 10.1109/TGRS.2024.3391625.
  3. A. Francis, J. Mrziglod, P. Sidiropoulos and J. -P. Muller, "SEnSeI: A Deep Learning Module for Creating Sensor Independent Cloud Masks," in IEEE Transactions on Geoscience and Remote Sensing, vol. 60, pp. 1-21, 2022, Art no. 5406121, doi: 10.1109/TGRS.2021.3128280.