A time series of Landsat 8 images
at San Lorenzo Channel, Baja California


LANDSAT 8 OLIP 
 
Using the Panchromatic band for water column correction
to derive water depth and spectral bottom signature:

Landsat 8 OLIP bandset used for this work

Purple=1Blue=2Green=3PAN=4Red=5NIR=6 and SWIR1=7
Collaboration with 
Fabio Favoretto, Ph.D Student, Coralline Algae Ecology
Grupo Interdisciplinario de Ciencia Ambiental, Universidad Autonoma de Baja California Sur
Carretera al sur km 5.5  | La Paz
favorettofabio@gmail.com
Using pan sharpened images in this study
Pan sharpening  using Rstudio with Brovey method

work done october-november 2016
Procedure/flowchart for operating  4SM.7.00

home
 
 

To Fabio Wed, 7 Dec 2016
 
  • 15m GSD:
    • I have been using the Landsat 8 PAN band for quite a while, resampling it to 30 m GSD
    • I owe it to you that I at last could play with pan-sharpened 15m GSD Landsat 8 images
    • Then I just realized this: why pansharpenen whereas I can enjoy using the 15 m PAN band as it is with resampled 15 m MULTI bands
      • after all, "sharpening" means nothing more than applying some (nasty?) "filter" to MULTI bands: see that mysterious "focal" R_function
    • This way I have the benefit of 15m GSD:
      • TRUE, because depth is computed by optimizing the resulting water column corrected spectrum with the Soil Line
      • I do this by increasing retrieved depth Z until the ratio (WCCcoastal+WCCblue)/(2*WCCpan) finally fit the Soil Line to satisfaction
      • This is what I have done at 30m GSD see http://www.watercolumncorrection.com/4sm-study-cases.php#OLI
      • So moving from 30 to 15 m GSD is no big deal
    • The problem is that MULTI bands are just duplicated in row and in column
      • one 30m MULTI pixel is split into four identical 15m pixels,
      • although this is unfair because this data has gone through bicubic interpolation:
      • see a MTL file:  "RESAMPLING_OPTION = "CUBIC_CONVOLUTION""
      • this results in very good results,
      • provided a very good co-registration has been secured in the first place,
      • except that the results look "blocky"
      • and need some sort of smoothing
      • see http://www.watercolumncorrection.com/slcoli-20140107-15m-2.php#blocky
    • The solution
      • I had to try sharpening for good
      • I now think that I don't need it
      • just light smoothing of the results does the job for now
    • Then I have some ideas about not just duplicating the MULTI pixels,
      • but applying a smart "filter" while doing so
      • after all, all these pixels have gone through bicubic interpolation (USGS)
      • so a smart "filter" should account for the immediate environment of the current pixel while duplicating: there are several filters out there that I should look at
      • " combined by weighted average according to distance"
      • the result potentially shall be: no "blocky" aspect anymore, and therefore no need for smoothing 15 m data
  • SAM and more
  • will then be possible to “finish the classification”?
  • CombinedDepth
    • beware: applying ZDTM results in enhancing the bias in the results: it shows in WCC bands!
    • while applying Z4SM yields a much less biased retrieved depth and WCC bands:
      • that's of course only where local water properties don't play tricks!




The SAM problem
Once the shallow areas in a scene have been correctedd for the diffuse attenuation effects of the water column,
  • it is desired to proceed to some flavour of bottom typing
For this purpose, various sources/methods can be used.
  • NDR the Normalized Diference Index, by Collin et al.
  • SAM the Spectral Angle Mapper, by Kruse et al
  • Brightness B, as the average of Blue reflectances
  • etc
The Spectral Angle is defined as the ratio Blue/Green of reflectances,
  • and converted to radians
In 4SM, the Spectral Angle is computed as the ratio (Coastal+Blue)/(2*Green)
of water column corrected reflectances.
  • it is not converted to radians,
  • values are mapped in a U8 channel as SA*100 , then displayed on screen using a color table SAM2.
  • this is "un-comfortable"
Coral sands have the following reflectances for OLIP bands (from Maritorena)
0.550 is estimated for the Panchromatic band
0.310, 0.380, 0.440, 0.550, 0.560, 0.670, 0.700
This yields SAcoralsand=0.784
so, in theory, this should be the maximum SA value for shallow substrates
My own estimation from processing images in the Bahamas Caicos  LSI
are quite close to Maritorena's estimation
Now, how do I set a color table for displaying images of SA?
SAM2 color palet is for CoralSand reflectances

Now, The brightest shallow bottom at San Lorenzo Channel yields SA~=0.59
This is at the intertidal sand flats of Bahia San Gabriel (figure 4 of Halfar et al, 2000)

where low sandy dunes may be assumed to be
devoid of any greenish material (do you concur?)

Type10 signature below is used as a local reference for SAM
It is sampled at a depth of ~1,52 m
Calibration of the Soil Line for this scene is fine tuned so as to obtain this result.

  
  
 

RAW: 100*SA                                                     RESCALED: 100*SA*0.75/0.59
true SA                                                                     normalized to CoralSand

    
 RAW: 100*SA                                                     RESCALED: 100*SA*0.75/0.59
true SA                                                                     normalized to CoralSand 
using the same color table SAM2 for both images
So: how can I help you?
Which do you prefer?
Is this useful to your purposes?
Where do we go from there?
"The advantage of SAM is that it uses only the direction of the spectra, and not their length, and hence the method is insensitive to the unknown gain factor, and all possible illuminations are treated equally."
"This method is insensitive to illumination since the SAM algorithm uses only the vector direction and not the vector length"


 



Créer un site
Créer un site