thoughts on terrain shadows

[dcar1s0n]

Hello! Thanks for your help so far. I am still working on a fjord region in Greenland and I am trying to deal with the mountain shadows that start to become problematic in September. I've attached an example of the turbidity where shadows show up as regions of low turbidity.

I am thinking about performing a terrain correction in SNAP with Sen2corr and then continuing with acolite. I guess I could use L1R from Sen2corr as input?

Any other ideas?

Thanks,
Dan

[quinten]

Hi Dan

Cast shadows are annoying indeed!

I am not familiar with the terrain correction available in Sen2Cor, and am unsure whether it includes a cast shadow correction. Shadow correction is perhaps not the best idea - the irradiance and hence target radiance in the cast shadows is very low, so the uncertainty on the corrected product would be very high. Atmospheric correction for water is already quite difficult due to the (generally) low signal from the water. Correcting a cast shadow on water is even more difficult since you have an even lower signal, and have to estimate the diffuse and direct illumination contributions.

Even without considering technical hurdles, e.g. Sen2Cor does not produce files compatible with ACOLITE, and generates L2 surface reflectances, I would not be confident in getting good results at all.

Your safest bet would be to mask the shadows, this can be done with the DEM and sun position. This is currently not implemented in ACOLITE, but I can take a look at whether this is feasible to add. This would make satellite remote sensing of your fjords more limited in space and time compared to other regions.

Quinten

[dcar1s0n]

Hi Quinten,

Thanks! I am looking into options to mask the shadows. IdePix (SNAP) with the Copernicus 30m Global DEM does a pretty good job at identifying mountain shadows, but I can only apply it to the L1C images. If necessary, I could run IdePix separately and then apply the mountain shadow mask to the L2W products from acolite. Clunky, but doable.

Dan

[dcar1s0n]

Actually I don't think IdePix is a good solution for mountain shadows as it seems to limit detection to terrestrial pixels. Hmmm....

[quinten]

Hi Dan

It could be that IdePix uses a hillshade algorithm, which kind of works for detecting mountain slopes, but does not cast shadows. When you opened this topic I did try a hillshade algorithm on the SRTM DEM but this only identified slopes... We'll need a more complete raytracing solution for casting shadows.

I believe your region also was not covered by the SRTM, so I had to register for accessing the Copernicus global DEM. This took a few days to complete and by then the momentum was a bit lost. My account is now activated and I will try and figure out the access to the DEM.

Quinten

[dcar1s0n]

Hi Quinten,

Any progress on this? I have the Copernicus 30 m DEM in my SNAP auxdata folder, if you need it.

Thanks!

Dan

[quinten]

Hi Dan

I have not had the time to look further into this! It will be at least another few weeks - I am currently on field work with limited internet access.

Quinten

[dcar1s0n]

Hi Quinten,

I was able to mask the shaded areas using the TanDEM-X 90 m DEM and the matlab hillshade function: https://de.mathworks.com/matlabcentral/ ... -hillshade

I'm sure there's a more elegant solution, but this does the job for now.

Ciao,
Dan

[quinten]

Hi Dan

Thanks for the update. Happy to hear the hillshade algorithm worked for you! I will probably come back to this at some point, as a shadow casting algorithm may perform better than just a hillshade.

Looking forward to seeing your results!

Quinten