Resampling of SWIR bands

[quinten]

The SWIR bands on Sentinel-2 are relatively noisy, and thus resampling them before processing is recommended. The resampling feature can be accessed in the advanced settings, by selecting "Smooth bands used for A/C" on the "Resampling" tab.

ACOLITE_20160520.0_advanced_resampling_crop.png (4.32 KiB) Viewed 28996 times

You can set the size of the resampling in "Smoothing window". A masking of bright non-water targets is performed before the resampling and the "Mask dilation" allows to add an additional buffer around this mask. Pixels in this buffer will not be included in the resampling window, but this has no effect on the final product mask. The option "Fill mask with median reflectance" will fill the masked (and dilated) area with the median reflectance, in an attempt to further reduce the impact of remaining unmasked bright pixels.

The figure below illustrates different processing settings over a part of Lake Balaton. (a) shows the Rayleigh corrected RGB composite, (b) and (c) show the results with no resampling for a fixed and variable aerosol type, (d) shows the results using a variable aerosol type after resampling. The grey area shows the original non-water mask. Taken from Vanhellemont Q. & Ruddick K. (2016) ACOLITE For Sentinel-2: Aquatic Applications of MSI imagery. Open Access Proceedings

balaton_smoothing.png (705.93 KiB) Viewed 28996 times

All bands are converted to 10 m resolution, and thus the smoothing window and mask dilation numbers correspond to 10 m pixels even though the native resolution of the bands may differ. So the above example will perform a 320 m smoothing and a 160 m dilation around masked pixels. Note that larger smoothing windows will have a longer processing time.

[marcelapereira]

Dear Quinten,
I am working with the SWIR atmospheric correction on S2A images on different inland waters.
They have small dimensions and height between 100 and 100 meters above sea level, for this reason I used differents DEMs. I attached an example of the result.
My question is, Why I do not have a retrieval in the mostly of center of the lake?.
The configuration is:

• Use per pixel variable epsilon

• Smooth bands used for A/C

• Smoothing windows by default and higher and lower value)

• Threshold by default and higher and lower value)

Thank in advance

Marcela

[quinten]

Hi Marcela

This seems like it could be masked because negative reflectances were retrieved in the middle of the lake. By default pixels with any negative bands will be masked, you can disable this in Advanced Settings > Masking.

From the RGB composite I'm guessing that in this case the NIR bands might be (slightly) negative. Due to the low particulate matter concentration in the middle of the lake I would expect a near zero signal in the NIR.

It would be helpful if you could provide the scene identifier and your region of interest so I could have a closer look.

Quinten

[tivincent]

Hello,

I am a very new user of acolite.
In the actual version (20180419), i could not find the "Advanced settings" and any information about it.

I am using S2-L1C data in order to compute chla and spm concentrations, and my resulting maps are very noisy and I would like to test some different parameters...

Thanks !
Vincent

[quinten]

Hi Vincent

Do you experience the noise also in the DSF outputs? If you could share your ROI and scene identifier I could have a look at the cause of the noise.

The resampling cannot currently (v. 20180419.0) be activated in the distributed package, and likely it will be added in a future release. You can at the moment process Sentinel-2 at 10, 20 and 60 m resolution (s2_target_res in the settings file), which may help with the noise.

I removed the Advanced Settings tab in the Python version, as it was too much work for me to add all the options in the GUI, especially since they are still changing a bit between versions. Maybe some version of this tab will come back in the future, but for now I prefer to keep adding features rather than spend time on the GUI. Most users at some point switch to command line processing anyway, where the settings do have to be set in separate settings file anyway.

Quinten

[tivincent]

Hello Quinten,

Thanks for your answer.
I am using the command line processing. All the advanced Settings could be inserted in the "settings file" ?

Here is in attachment some figures computed with Acolite, with Sentinel-2 data over the Berre Lagoon (South of France).

S2A_MSI_2017_11_16_10_33_11_T31TFJ_rhos.jpg (175.74 KiB) Viewed 27404 times

S2A_MSI_2017_11_16_10_33_11_T31TFJ_t_nechad.png (396.85 KiB) Viewed 27404 times

S2A_MSI_2017_11_16_10_33_11_T31TFJ_chl_re_gons740.png (531.98 KiB) Viewed 27404 times

Thanks !
Vincent
PS : I have used the default value of each parameter of acolite

[quinten]

Hi Vincent

I think you have two kinds of noise in your example.

1) The white masked areas are caused by masking in ACOLITE, likely due to negative reflectances or high SWIR reflectance. In the current version (ACOLITE 20180611.0) you can disable the masking of the water products by setting l2w_mask_water_parameters=False in your settings file. (Currently I track all bands for the negative reflectance, but should probably exclude the SWIR channels.)

2) The Gons chlorophyll product is here quite noisy, likely due to the low red/red-edge reflectance ratios. Some masking is performed to exclude pixels where the algorithm would perform badly, but this masking is not perfect, and can currently not be disabled or modified.

The noise in the turbidity product noise could probably be improved at native resolution, but for the Gons algorithm pixel binning may be required. Your scene is quite late in the year and has a fairly low sun zenith angle (63°) which could also impact the image quality.

Quinten