NEGATIVE rhos_ (bit flag 3 value 8) for entire scene Sentinel-2 DSF

[beder]

Hi,

With the following settings (attached as files), nearly all of the scene is masked with bit flag 3 (value 8) for negative RHOS.

Using DSF tiled with sunglint correction on.

Could you help me understand what the problem is?

Thank you.

[quinten]

Hi Bryan

It is not uncommon for ACOLITE/DSF processing to produce negative reflectance for clearer waters, at the longer wavelength range 650-900 nm. The clear water signal is very low in this spectral range and even small errors in the atmospheric correction may cause the reflectance to become (slightly) negative. The glint correction can further increase the occurrence of negatives in those wavelengths.

If these slight negatives in the red-NIR are no concern to your application you can change the wavelength range of negative masking:
l2w_mask_negative_wave_range=400,650

For your scene the reflectance are only very slightly below zero, and with this setting I get the following results:

S2A_MSI_2019_01_09_16_16_21_T17RMH_L2R_rgb_rhos.jpg (180.98 KiB) Viewed 122515 times

S2A_MSI_2019_01_09_16_16_21_T17RMH_L2W_chl_oc3.jpg (123.67 KiB) Viewed 122515 times

You could disable negative masking altogether, but it may be safer to keep the check for negative blue-green bands:
l2w_mask_negative_rhow=False

I should note that there are obvious bottom reflectance problems in the OC3 algorithm outputs here.

I hope this is useful!

Quinten

[beder]

Hi Quinten,

Thank you for the reply.

In fact, the slight negative reflectances in the red and NIR are a significant concern. In light of this, what would you recommend?



-Bryan

[quinten]

Hi Bryan

If these negatives are of significant concern I would recommend not using the ACOLITE/DSF results for this scene. Perhaps it is worth exploring other algorithms such as POLYMER, C2RCC, SeaDAS/l2gen, which may be more suited for this type of target. If you do so, please let me know what you find!

Quinten

[beder]

Thank you for the follow up.

Could DSF without glint correction or the exponential method be viable options?


-Bryan

[quinten]

Hi Bryan

DSF without glint seems to also give some negatives and uncorrected glint. EXP has its own drawbacks (especially in the blue: https://doi.org/10.1016/j.rse.2019.03.010) but seems to provide no negatives in this case.

Quinten

[beder]

Hi Quinten,

Would the alternative glint method included in acolite be of any benefit in this case?

-Bryan

[quinten]

Hi Bryan

In this image it seems the main issue is the overestimation of the aerosol optical thickness. You could modify the processing by providing a (lower) aerosol optical thickness compared to the default estimate, e.g. I found reasonable results for this image specifically with:

Code: Select all

dsf_fixed_aot=0.12
dsf_residual_glint_correction=True
dsf_aot_estimate=fixed

S2A_MSI_2019_01_09_16_16_21_T17RMH_L2W_rgb_rhow.png (849.15 KiB) Viewed 122192 times

Of course for other images the dsf_fixed_aot=0.12 may not be appropriate. Additional tests are needed really to understand why the default DSF is overestimating the aerosol.

I hope this helps!

Quinten

[beder]

Hi Quinten,

Thank you for investigating further. Much appreciated.

Do you have any advice as far as what tests should be conducted to examine why the AOT is being overestimated?

Our effort is to use acolite as part of a routine processing system where products are derived from acolite corrected S2 imagery with minimal human intervention.

-Bryan

[quinten]

Hi Bryan

I meant this more for myself, from a development perspective. The DSF (without glint correction) should not give negatives by design, unless if the TOA reflectance is lower than the computed Rayleigh. This is not the case here, and there is quite a bit of wiggle room to select a lower AOT and avoid negatives. Here the default gives an estimate of around 0.24 while 0.12 gives a more realistic result. Perhaps the DSF+glint correction should be done in a combined/iterative manner.

Quinten