surface reflectance and segmentation

[Tristan Postadjian]

Hi,

I am currently working on Louisiana marshes, on the coast.

The pictures I use come from Landsat 8. I would like to achieve a segmentation on an area covered by several pictures.

I have 3 pictures I need to merge so I can compute the segmentation with segutils.runShepherdSegmentation .

I'd rather like the process being fast and efficient, so I cropped the pictures before segmentation according to a kml file in QGIS.

The area of interest was composed of 3 pictures from April-May 2015. But one of them was unusable because of the clouds that mask almost all the pictures.

So I took a 2014 picture (same period) to replace it. No clouds on the three of them.

I am assuming that the segmentation needs to be done on reflectance, that's why I used the below arcsi.py command :

arcsi.py -s ls8 -f KEA --stats -p RAD DOSAOT SREFSTDMDL \
-o ./Outputs/LC80230402014106LGN00 --tmpath ./tmp \
--atmosimg ./data/WorldAtmosphereParams.kea --aeroimg ./data/WorldAerosolParams.kea \
--dem ./data/srtm_louisiana.kea  --minaot 0.05 --maxaot 0.6 \
-i ./Inputs/LC80230402014106LGN00/LC80230402014106LGN00_MTL.txt

The thing is that the 2 pictures from 2015 have surface reflectances that match when merging (cannot see the border between them), but the 2014 surface reflectance picture is far different from the first ones. Once the pictures are computed in surface reflectance pictures, I thought they should be very close in terms of colour, etc.

Then, I chose to change to only 2014 pictures so I could check if it was a diachronic issue. I only needed a second picture additionally to the first one (I used before).

The one I picked had some clouds on it, but very few.

I do not know anything about the AOT I should use in this area, then I preferred to estimate it.

Anyway, the problem is that the resulting surface reflectance pictures covering the wanted area are very different in terms of appearance. One looks better, with details, but the other is very colorless. This last image is the cloudy on e, but I don't know if it's the problem.

Is the command can handle partially cloudy pictures ? If so, what should I do ?
It seems that it impacts the resulting surface reflectance.

Thanks,

Tristan Postadjian