WorldView 2 image Princess Cays, Eleuthera island, Bahamas
  November 11th 2011, 2 m ground resolution
  "(c) DigitalGlobe, Inc. All Rights Reserved"
"Includes copyrighted material of DigitalGlobe, Inc., All Rights Reserved"


This dataset from Digital Globe under Data Exchange Agreement
courtesy of Gregory MIECZNIK: watch his bathymetry webinar

This page presents the most recent advance of 4SM
as regards the use of the PAN band
along with the MULTI bands of a WV2 image
  • Wed, 26 Sep 2012 12:14:38, Gregory advises that image   is   available on ftp site
  • Fri,    28 Sep 2012 12:53:30, Yann      advises that results are available on ftp site
    • That's almost a same day service
  • On Oct 3rd 2012, Gregory sends out his preliminary seatruth regressions
September 2012


Adjacency effect

MULTI calibration and seatruth regressions
The good and the bad
The MULTI command line script
Download preliminary depth
===> Much
===> September 2014

MULTI calibration
MULTI profiles

MULTI+PAN calibration
MULTI+PAN profiles
Command Line MULTI+PAN
Download MULTI+PAN results


This studycase particularly addresses
the quest for effective wavelengths
of wide multispectral wavebands
for use with Jerlov's data
  • No need for field data for hyperspectral data: as of 2012, I knew from previous work that
    • once the ratio Ki/Kj is estimated for any two visible bands i and j is estimated from the current image,
    • and knowing that Ki/Kj=(Ki/Kk)/(kj/Kk) must be satisfied for all sets of visible bands i, j and k, 
    • then, by specifying effective wavelengths at mid-waveband of CASI hyperspectral fairly narrow bands (except between 575 and 600 nm), 
    • the correct spectral values of the two-ways diffuse attenuation coefficient K may be derived, which results in correct retrieved depth without the need or use of ant field data.
  • No need for field data for SPOT data: surprisingly, this remarkable fact was first established using SPOT images.
  • But for TM, ETM, IKONOS, and WV2, I experienced much confusion in this regard
    • all have widebands, 
    • and I did not have access to suitable ground truth
The problem of the 575-600 nm range
  • the absorption coefficient for pure water exhibits a steep gradient in this range
  • the green waveband of multispectral images usually includes this range
How does a waveband behave re: operational attenuation coefficient 2K?
  • a wide waveband typically covers 30-40 nm, and Jerlov's data are provided at 25 nm steps
  • photons at longer wavelengths may be expected to be faster attenuated out of the upwelling radiance, in such a way that the effective wavelength of that waveband shifts from mid-waveband to a longer value for most of Jerlov's water types
    • even more so for the green band which includes part of the above-mentioned steep gradient range
  • this fact actually totally rules the behaviour of a panchromatic band
    • which starts behaving as a NIR band at very shallow optical depths and ends behaving as a blue band at great optical depth

So the question is:
how to specify the effective wavelength of multispectral bands in 4SM??
  • to answer this question,
    • a precise groundtruth dataset must be available
    • an image of exceptional quality is desirable, and waters must be clear and homogeneous
    • the main purpose pertains to the green band, but all visible bands must be considered the same way
None of the previous attempts succeeded
With the prospect of first class groundtruth by Digital Globe
the following was done:
  • september 28th: first run a rapid/rough first calibration,
    • and submit the results for retrieved depth
  • then inspect and investigate seatruth evidence
  • then only investigate ways of getting things right
    • while keeping an eye on the physics behind
    • until seatruth evidence suggest success

Full image is  8-bands, 7200*12515

A beautifull image

very low noise, apart from banding
some faint glint
and adjacency effect

This processing was achieved
within 24 hours
of downloading the image from DG's ftp site.

It is not perfect:
where only bands 1, 2 and 3 exhibit bottom detection, computed depth can easily be underestimated by several meters where bottoms are green.

OIB water type of Jerlov.
Modelling down to ~25 m.


tide height on Nov 11th 2011
is approx 0.50 m above MLLW
on the east coast

The Eleuthera tides on the Exuma Sound side (west coast) of the island are approximately 2 hours behind the tides on the Atlantic side (east coast) of Eleuthera. When rising, ocean tide will be higher on the east coast of Eleuthera then on the west coast.

Banding is bad
TCC deglinted enhanced
note the banding over deep waters

TCC natural

Adjacency effect

Adjacency effect at 908 nm
  • It extends ~600 m away from waterline
  • Profile_adjacency is shown, it is 1.2 km long

Deglinting removes adjacency effect to satisfaction

  • band_4 at "570 nm" is the Panchromatic band

Sept 28th 2012: preliminary MULTI calibration
Sept 28th: first calibration
Sept 28th: first calibration
all wavelengths at mid-wavebands
2K    0.091 0.074 0.142 0.535 0.799 3.500 4.490 8.900
Notice the lousy fit for bands 4 and 5


Please, download TIF image 20 MB
of computed depth in centimeters
16 bits signed from my GoogleDrive


Seatruth by Digital Globe October 3rd 2012
 on Sept 28th preliminary work

DG_Morel-Comparison Sept 28th 2012
DG's depths vs DTM       4SM's depths vs DTM

This image, compiled by Gregory Miecznik,
compares results obtained

Sept 28th 2012 MULTI: the good and the bad
On purpose
  • leaving wavelengths at mid-wavebands and assuming neutral spectral bottom reflectance was done on purpose, knowing that seatruth against a good DTM shall show where things need attention.
    • seeing it in your above plot was quite rewarding for me, indeed.
  • I had advised you on my website that this should likely be the case.
    • this makes this this "demo" to DG all the more convincing.
    • it gives you a chance to grasp quickly some of the important fundamentals of 4SM modeling.
  • it was clear to me, and it shows in the results,  that
  • and I had advised you of that on website
  • anyway, this is only the begining: more clear-cut WV2 studies cases are needed.
28 sept 2012: West looks Good
Not knowing where the seatruth DTM is located,
I had managed to get good looking results
at the above sub-window

28 sept 2012: East looks Bad

 The MULTI command line script
Sept 28th 2012
nice -20  4SM.4.09  -Process  -Origin/DigitalGlobe            
-B/tclNe/cLM1  @-NDR_50_zDTM                        
-extract/v/rawBDH/NIRband8/NIRmax5.0/FullBDH/mSOIL21/mBPL2 /ext_1_7200_1_12515/mask_3        

Créer un site
Créer un site