NaN Rrs values for inland water lakes

[quangtu]

Hello,   
      I'm doing research involving the Tonlé Sap lake of Cambodia. I'm trying to use l2gen with aer-opt -2 for MSI  but keep getting NaN rhow/Rrs values on many water bodies.
      I have found out that the main problem is because of high SWIR reflectance value, due to direct sun glint on the water surface. Thus, I want to significantly raise the threshold for the non-water masking to retrieve the lake pixels. However, I cannot find the option to adjust the threshold. Can anyone please suggest how to do it?

Thank you for your time!
Tu

[OB.DAAC - SeanBailey]

Tu,

There are a number of issues that could be impacting your retrievals. 

1)  If high turbidity or glint is present, it may be that the cloud mask is being set.  By default, l2gen does not process pixels flagged as cloud.  The cloud threshold is determined by two options:
cloud_thresh (float) (alias=albedo) (default=0.027) = cloud reflectance
   cloud_wave (float) (default=865.0) = wavelength of cloud reflectance test
For MSI, those are set to:
cloud_wave=2200
cloud_thresh=0.018

Increasing cloud_thresh will reduce the number of pixels flagged as cloud.  If you'd rather not do an iterative process to determine that value, you can either set it rather high (e.g. 0.1) or process the scene with maskcloud=0.  If you output the cloud_albedo product, you can select a threshold by visual inspection.

2)  If you're using the l2gen defaults, the landmask is likely too coarse for the MSI resolution and may be masking regions of the lake.  Since we don't have a high resolution landmask for your region of interest, you can turn off the landmask during l2gen processing (it will slow things down a bit, as every pixel will be treated as if it were water):
land=$OCDATAROOT/common/landmask_null.dat 

If you include rhos_865 (or rhos_nnn to get Rayleigh corrected surface reflectance for all bands),  in your l2prod output, you can use that band to derive a landmask to apply.

3) It is not surprising that the standard method aer_opt=-2 does not work well for MSI over Tonlé Sap Lake. The MSI bands are much broader and have a lower signal to noise than typical ocean color sensors.  This makes the atmospheric correction (really the aerosol retrieval) process less robust.  You can either turn off the aerosol removal (aer_opt=-99, or try one of the alternative aerosol options:
       -4: Multi-scattering with fixed model pair
           (requires aermodmin, aermodmax, aermodrat specification)
       -5: Multi-scattering with fixed model pair
           and iterative NIR correction
           (requires aermodmin, aermodmax, aermodrat specification)
       -6: Multi-scattering with fixed angstrom
           (requires aer_angstrom specification)
       -7: Multi-scattering with fixed angstrom
           and iterative NIR correction
           (requires aer_angstrom specification)
       -8: Multi-scattering with fixed aerosol optical thickness
           (requires taua specification)
These, however, require a bit of a priori knowledge of the aerosols - and the -4/-5 options require knowledge of what our cryptic aerosol model names mean!  

Hopefully, this gives you some ideas about how to proceed.

Regards,
Sean