MOD35 cloud mask vs MOD09GA cloud state

[gelieza]

I am a new user of MODIS data and I work with MOD09GA.061 accessed via Earth Engine (https://developers.google.com/earth-engine/datasets/catalog/MODIS_061_MOD09GA) to construct a local cloud climatology.

In the user guide of MOD09 collection 6, it is stated that the cloud_state bits 0-1 of SDS state_1km_1 are "read from MOD35" while bit 10 of state_1km_1 is computed by an internal cloud algorithm. However, looking at the documentation of MOD35_L2 (https://atmosphere-imager.gsfc.nasa.gov/products/cloud-mask/format-content), bits 1-2 contain the cloud mask and are labelled very differently from the cloud_state bits in MOD09. The labels in MOD35_L2 are "confident cloudy", "probably cloudy", "probably clear" and "confidently clear", whereas the labels in MOD09GA are "clear", "cloudy", "mixed", "not set, assumed clear" for the values 00, 01, 10 and 11, respectively.

How are the cloud_state values in MOD09GA related to the cloud mask in MOD35? How are MOD09GA cloud state values computed and where can I find documentation on this?

Thank you very much for your help!

[LP DAAC - lien]

Hi,
Thanks for you interest in the MODIS SR products. We don’t have official documentation of the rationale used to create the internal algorithm QA layer, but we have reached out to the NASA LDOPE team to see if they can provide some documentation about this.
As you said State bits 0-1 are propagated from the MODIS Atmosphere groups MOD35 product, but it is not an exact replication of the cloud mask. In the MOD35 cloud mask bit 0 is 0=determined, 1=not determined. So rather than create another bit for that it was rolled in with the others. Their intention was never to replicate the MOD35, but just use it for their purposes.
These two different cloud QAs are really intended for 2 different types of users of the MOD09 products. The MOD09 Surface Reflectance products are probably the only MODIS Land products that clouds do not disqualify a pixel from being considered “Good Quality”.
The Quality Assurance team put it this way:
“First of all, in MOD09 cloudy doesn’t mean bad quality, it is only an atmospheric state and the retrieved reflectance is the reflectance of the cloud surface and is good. For VI or BRDF or for any other land product it doesn’t make sense to retrieve the (VI/BRDF etc) parameters using cloudy observation because that is not the reflectance of the surface/land cover. So, if these parameters are retrieved using cloudy observation, then obviously they are bad – so cloudy or high aerosol is considered as bad quality surface reflectance for VI/BRDF/LAI_fpar etc, but not bad in a surface reflectance product (MOD09) by itself.

The product is just the reflectance from the observation - if the instrument can see the land it is surface reflectance of cloud clear surface which could be from any land cover or ocean or inland water body. If it is cloudy then it is the reflectance from the cloud. If it is not cloudy, then the algorithm will retrieve AOT and then correct for aerosol. If AOT cannot be retrieved, then also there will be correction using climatology. So atmospheric correction is done pretty much everywhere.”

This rationale is what leads to the 2 different Quality layers. The Surface Reflectance team considers the Internal Cloud QA (Bit 10) the preferred QA for the Surface Reflectance data products which uses cloud/snow/aerosol/fire/shadow etc. as determined by the reflectance algorithm.

So, if a project is using the MOD09 itself then bit 10 is generally recommended. However, the MOD09 Surface Reflectance is used in virtually every other Land product (with the exception of thermal products like fire and LST) so for that purpose cloud free data is generally desired so the Cloud mask derived from the MOD35 is often used.

As soon as we get more information from the developers, we will be sure to let you know.
Thanks,
Brett

[LP DAAC-EDL - lien]

The relevant bits in MOD35 (Cloud Mask, band 1) are these:



0 Cloud Mask Flag 0 = Not determined

1 = Determined



2, 1 Unobstructed FOV Quality Flag 00 = Cloudy

01 = Uncertain

10 = Probably Clear

11 = Confident Clear



The state_qa bits are set from this as follows — if bit 0 is set to 1 ('determined') and neither bit 1 or 2 is set, the state_qa pixel is flagged as 'cloudy'. But if bit 1, only, is set ('uncertain'), the pixel is flagged as 'mixed'.

[victor_grosser]

Thanks Brett for the detailed clarification.

If I understand correctly, the choice between MOD09 (bit 10) and MOD35 depends on the context:

For direct use of MOD09 Surface Reflectance products, internal QA (bit 10) is preferred as it integrates reflectance-based detection including cloud, snow, aerosol, shadow, and fire.

But for higher-level products or external applications where cloud-free data is essential, the MOD35 cloud mask offers more conservative cloud filtering.


Also interesting to note that even when AOT cannot be retrieved, atmospheric correction is still performed using climatology — so coverage is quite robust across different surface types (ocean, inland water, land).

Appreciate the heads-up about upcoming developer updates. Looking forward to more insights!

Victor Grosser - The Netherlands

[LP DAAC - lien]

Hi Victor,
Yes, that is all correct. One thing you might want to be careful about is using the MOD09 products for reflectance over water. That is not recommended, the atmospheric correction and surface reflectance retrieval algorithms are no ideal for water bodies due to the inherently low level of water leaving
radiances and consequently very low SNR. Quantitative remote sensing retrievals of water column constituents would require different algorithms. There will be a lot of negative values.
Thanks,
Brett