Consistent Cloud Classification Errors in Landsat 8/9 C2L1 Surface Reflectance QC Files Since January

[varishtg]

Hello,

I’ve been monitoring several regions using Landsat 8 and Landsat 9 Collection 2 Level 1 Surface Reflectance (C2L1 SR) data. I derive Land Surface Temperature (LST) from these datasets and use the Quality Control (QC) file that accompanies the SR product to identify and filter out cloudy pixels.

Since January 2025, I’ve noticed a recurring issue: the QC file often flags scenes as cloudy, even though they appear clear in the True Color Image (TCI). This misclassification has been consistent across multiple regions and has resulted in the unnecessary exclusion of good-quality data from my analysis.

To illustrate the issue, I’ve attached example images where the TCI clearly shows a cloud-free scene, yet the QC flags indicate cloud contamination.

I would appreciate help with the following:

Is this a known issue with the C2L1 SR product's cloud flags?

Are there recommended workarounds or alternative cloud masks to improve accuracy?

Should I consider additional thresholds or post-processing to correct the QC output?

Thank you in advance for your support.

[LP DAAC - dgolon]

Hello @varishtg Thanks for writing in. I am discussing this with the internal Landsat team here at EROS and will follow up when I have additional details. Thanks -- Danielle

[LP DAAC - dgolon]

Hello @varishtg Please see this known issue related to the SR QA: https://www.usgs.gov/landsat-missions/landsat-collection-2-surface-reflectance

The cloud and cloud shadow indicators in the Surface Reflectance data product are known to report erroneous conditions in areas where temperature differentials are either too large or too small. For example, a warm cloud over extremely cold ground may not calculate enough difference in temperature to identify the cloud. Conversely, residual ice surrounded by unusually warm ground can potentially be identified as cloud.

We are discussing how to improve this issue in a future collection.

Thanks -- Danielle