This tool called Cmask (Cirrus cloud mask) is used for cirrus cloud detection in Landsat 8 imagery using a time series of data from the Cirrus Band (1.36 – 1.39 µm).
Note: We recommend combining Cmask and another cloud detection algorithm (e.g., Fmask) to detect all kinds of clouds in Landsat Time Series (LTS) since Cmask is designed for detecting cirrus clouds solely (It cannot detect non-cirrus clouds).
See the document named 'How to use Cmask.docx' for the tutorial (within the code package)!
Data:
Example data are available at this Google Drive.
Training data will come soon.
Validation data are available at this Google Drive.
Gobal mask for places where water vapor regressor should be included in Cmask is available at this Google Drive
Please cite the following paper:
Qiu, Shi, Zhe Zhu, and Curtis E. Woodcock. "Cirrus clouds that adversely affect Landsat 8 images: What are they and how to detect them?." Remote Sensing of Environment 246 (2020): 111884. https://doi.org/10.1016/j.rse.2020.111884
The GIFs below illustrate the comparisons between Cmask and USGS Cirrus QA for all Landsat 8 images (central subset images with 500 pixels X 500 pixels) with disagreement >= 5% from 2013 to 2018. The UPPER LEFT image is a false color composite provided for perspective (SWIR1, NIR, and Red bands). The UPPER RIGHT image is the Cirrus Band TOA reflectance (Unit: X 10000). The LOWER LEFT image is the USGS Cirrus QA flag results (White color). The LOWER RIGHT image is the Cmask results (Variable threshold: Red >= 0.8, Orange >= 0.7, Yellow >= 0.6, Green >= 0.5 (Default)). Considering that any kind of clouds need to be excluded for further applications, the commission error from other non-cirrus clouds located in high altitudes (e.g., top of cumulus cloud) is not particularly harmful.
 |
 |
 |
|---|