Re: [GRASS-user] i.atcorr with Landsat 8

[Micha Silver]

Hi Nikos: Maybe you have some additional tips for me on this problem using i.atcorr with Landast 8. On 05/23/2015 04:26 PM, Nikos Alexandris wrote: So, do I need to read the Min and Max radiance for each band separately from the MTL file, and use those as the range= parameter? Micha, I now see what the concern is. If I understand it correctly, in the case of spectral radiance (default input type for i.atcorr) as input, I would simply use the min, max values of the images themselves. For reflectances (-r flag for i.atcorr) I think I would use [0, 1.0]. In addition, (as I have done in the past) one can flatten, all values > 1.0 to 1.0 (as they normally are --should be, imho-- erroneous large outliers). I have to look closer at what I did in my script (see: https://github.com/NikosAlexandris/i.landsat.atcorr/blob/master/i.landsat.atcorr.py#L371) and fix it (currently broken). But, I do read the min and max values from the images that are going to be processed. I looked thru your script and tried several combinations of "range=..." and "rescale=..." parameters. First I read the min/max values of all bands (r.info -r) and put these with each band's raster name into a text file for easy access in a script. Then I tried: 1) rescale=0,1 : result was indeed values between 0-1 (reflectance as expected) but the histgrams were all wrong. Almost all pixels got value << 0.005. All values from 0.01 - 1.0 were only a very small number of pixels. If I tired to get an RGB composite from bands 4,3,2 it looked pychedelic. 2) rescale set to the same range as each raster's min/max values. In this case some of the resulting histograms looked fine. And if I display the RGB composite of bands 4,3,2 after i.atcorr I get nice bright images, with even better contrast than the toar raster (as expected). However values are *not* between 0-1.  So can you guess what is correct? Should I just "normalize" the values resulting from (2) above by manually rescaling to 0-1? Any tips are appreciated. Best, Micha At least in the case of Landsat8 imagery, the script worked well (before it went broken :-p). The default of "0,255" seems to be too small. I agree. Furthermore, I think the current default 8-bit range for the output spectral reflectances (i.atcorr *always* delivers reflectances) is (was) like a standard for which many modules would simply do their job while they would fail in case of floating point numbers or larger ranges. I think we should communicate and maybe ask for it to change that the "normal" range for reflectance is zero to one. And depending on how rich was the input, simply rescaling to 8-bit comes with cost of loosing fine radiometric details.

_______________________________________________
grass-user mailing list
grass-user@lists.osgeo.org
http://lists.osgeo.org/mailman/listinfo/grass-user