As of may 2013: comments on

DG's "BathymetricCompendium.pdf"


titled "Validation Report of DigitalGlobe WV-02 Derived Water Depths in the Littoral Zone"
by Gregory Miecznik of DigitalGlobe Inc, dated May 07th 2013:

"Here is a report which I recently generated for our potential customers
It shows comparison for 5 different sites.
Let me know what you think"


Compendium: watch Digital Globe's bathymetry webinar 2013

home


 
1 - NO NEED for field data, nor for atmospheric correction
2 - this is demonstrated in this website, using a variety of hyper/multi spectral data
 
Requirements are
1 - homogeneous water body and atmosphere
2 - some coverage of optically deep water
3 - some coverage of dry land
 
Problems are
1 - the precision on estimated depth is found wanting, because the noise-equivalent change in radiance  of accessible data is too high for shallow water column correction work 
2 - radiance data should be preprocessed by the provider at level 1 in order to improve S/N ratio
3 - exponential decay: the deeper/darker the bottom, the poorer the performances
 
So
I keep digging
until suitable data
become available
 
This relates to WV2
Buck Island Reef, USVI
Marmion Marine Park, Western Australia
Princess Cays, Bahamas
Lee Stocking Island, Bahamas
La Parguera Marine Reserve, Puerto Rico
Cockburn Sound, Western Australia
Florida West Coast
Oahu, Hawaii
Bahrain
                                       
Do you know CSIRO's presentation, 2012:   26.pdf
"Bathymetry from satellites for hydrographic purposes"
by A.G. Dekker, S. Sagar, V.E. Brando, D.Hudson
(quite a team of smart people!)
see also 4SM vs ALLUT
see also 4SM where's the catch? 

see also EOMAP
 
  

Overview of the optical calibration process
Parameter retrieval  overview of the modeling process
Roadmap  we've got work to do!





Overview of the optical calibration process
DG's method
uses a variety
of software packages
4SM uses ... 4SM,
and GMT to write and display various PostScript files
on a Toshiba Satellite laptop computer
from DN to TOA radiance
 
4SM reads DNs for the current pixel
From now on, we can just as well call them reflectances
There is no need for conversion into radiance, as 4SM is a "ratio method"
from TOA radiance to total reflectance above the water surface:

This uses DG proprietary software to produce a BOA image

 
  • Various calibration processes use original DNs read from the image for the estimation of the following:
    • sea-surface glint regressions: deglinting removes sea-surface glitter, and variations of the atmospheric path radiance as well
    • Lsw a uniform spectral radiance over a body of optically deep waters for the whole deglinted scene
    • La a uniform spectral path radiance for the whole deglinted scene
    • Lw a uniform BOA spectral water volume upwelling radiance
    • 2K spectral value of the effective diffuse attenuation coefficient 
    • LM spectral BOA  radiance for the brightest type of shallow bottom type
  • These shall be used on the fly over the whole scene
    • up to now, I feel comfortable that La and Lw are estimated quite realistically, I think within comparable range of confidence if compared to other ways of estimation
  • Now with Landsat 8 OLI imagery (see USGS's webpage "Using the USGS Landsat 8 Product"), I translate La, Lw and LM into calibrated reflectance Ra, Rw and RM (range 0-1).
    • This mean that i can write out water column corrected product in units of reflectance for Landsat 8's OLI images.
    • Don't know yet if WV2 offers a similar way.

 
No in-situ measurements
are used in determination of the reflectance product
No in-situ measurements are used in determination of the BOA reflectance
  • Subtraction of estimated glint on the fly: this includes swell-modulated glint, whitecaps, boat trails, lumps of haze (and also removes the adjacency effect where atmosphere is stuffy and land is hardly vegetated a all)
  • At this stage, one obtains the DN value for the upwelling signal above the water surface for the current pixel
4SM can contribute
all this calibration stuff to DG's method


 
Uses water optical properties and bottom spectral reflectance
 data obtained either from existing literature or acquired for a specific project

Simulation of radiative transfer through water:
uses HYDROLIGHT 
to produce spectral LUTs


 
4SM is ready for modeling
possibly within just a few hours
of receiving the raw archive image
 and importing it into working database.pix

while DG personel
still have a lot of sophisticated things to do
  • either hunt for published water optical properties and libraries of bottom spectral reflectance for many (~20?) bottom type end-members,
  • or send a team of trained scientists onsite to collect such site-specific data
  • and then get through a long and delicate data reduction process
Then, once all data reduction is completed,
DG's Hydrolight specialist
comes into play to produce a series of LUTs to represent the BOA upwelling spectral reflectance for a great number of discrete bottom depths over a great number of end-member bottom types and combinations of them



Parameter retrieval : overview of the modeling process
DG's method
uses Knitro of MathLabs
 
4SM uses ... 4SM!
  • Based on a spectral matching which minimizes a difference between simulated and observed spectral reflectance above the water surface
  • This uses the spectral BOA reflectance product
  • For each shallow pixel, this requires browsing all occurences in all LUTs: that's a great many!
  • This process uses KNITRO solver for nonlinear optimization

 
  • Based on increasing depth iteratively until the watercolumn corrected spectral signature achieves an acceptable match with the spectral Soil Line at null depth that was specified in the calibration process
  • For each shallow pixel
    • read the original raw TOA DN for the current pixel
    • deglint and smart-smooth it as required
    • iterate a few ten times while increasing/optimizing Z
    • until an acceptable match is obtained
DG take pride of
  • not deglinting, other than masking alien pixels
  • not applying any smoothing
Deglinting and Smart-smoothing is essential in 4SM
I think DG should offer deglinting and smart-smoothing
of the "spectral BOA reflectance product"
  • Do DG produce a watercolumn corrected view of the scene?
4SM does at no extra cost
 
  • Do DG produce a map of bottom types?
4SM can at some extra cost
  • Can DG add the PAN band to the MULTI bandset?
4SM does
It is even becoming a must







Roadmap : we've got work to do!
I need more such nice DTM datasets: how about UKHO's PrincessCays MBES?

TCC
See SPOT at Rangiroa atoll 
for similar annular features
The ~20m wide bright annular features
might be caused by dejections of coral-grazing fishes

BUIS_Bathy_3m resampled to 2m: LIDAR
LIDAR is probably strongly smoothed
LIDAR reports those circular mounds
just ~ 2 m higher than surrounding,
with no annular moat
depth_v2_final: DG
DG depths are neither deglinted nor smoothed

depth 4SM
4SM depths are deglinted smart-smoothed
Black is 4SM deglinted-notsmoothed depth
Red is DG depth_v2_final
Black is 4SM deglinted-smoothed depth
Red is LIDAR depth
two bright sand sections
exhibit strong depth underestimation
  • It would seem to me like if an important bottom type is missing for your LUTs, so your spectral matching zoomed on a wrong  Rsim(x) fitted reflectance
  • I have tried this sort of approach myself in 4SM, starting from Z=0: with the result that the process happily stops at first opportunity on one bottom type, whereas a slightly better (or equivalent) opportunity might be found at greater depth on another bottom type
  • at 1.0 km on the white profile:
    • in 4SM: applying your estimated depth of ~13 m instead of Lidar depth ~15 m  is an incomplete watercolumn correction
      • the Green radiance is the least corrected
      • then the Blue, then the Purple
    • it would mean the bottom looks very Bright_Purple-Blue down there,
      • which is commonly seen over live corals
      • which of course is unrealistic for a sandy area
  • Selling computed depth is one thing: you have to satisfy hydrographers
  • Selling bottom typing is even more attractive: coastal monitoring is the biggest market


I think that, for DG to succeed
with shallow water work,
you should include a naturalist in your team

  • one who is a scuba diver
  • one who knows what the shallow bottoms look like,
  • one who knows it all about underwater and atmospheric optics






Progress with 4SM has been astounding lately
 since I got the attention of Ron Abileah
and access to WV2 images and some decent seatruth.


 
at Princesscays
"The bias is caused most likely
by overcompensating
the atmospheric
contribution to TOA"
That's the beauty with 4SM:
  • you don't have to start again from scratch:
    • there is no such thing as a "product" compensated for path radiance
    • just modify any parameter in the command line as required
  • then run the modeling again, possibly along a profile or inside a small window
  • you get the result in minutes
  • go back to your commandline, etc, 
  • increase or decrease any parameter, enable or disable any functionality, 
  • run a profile, run another profile, try another window, process any specific pixel, etc
  • then when satisfied, run the modeling again on the whole image
    • and mask out areas which are unsuitable
Puerto Rico
From looking at the regressions, seems to me that waters optical properties are consistently less clear than was  assumed in Hydrolight: see regression for tile 2




 
 
 
  • In 4SM I am feeling more and more uncomfortable when setting wavelengths at mid-wavebands while accessing Jerlov's table of diffuse attenuation coefficients for marine waters
    • setting WLgreen at 551 nm appears to suit most recent cases
    • as for WLyellow and WLred, they would appear to need decreasing away from mid-waveband: I noted that at PrincessCays, then again at BuckIslandReef from 0 to ~9 m
  • I've got hold of the LIDAR DTM Bathymetry.tif
    • I'd want to work on your WV2 images, including coverage of Bay of Mayaguez
Puerto Rico
"The water turbidity has the greatest effect on retrieval quality. In turbid and murky waters, the errors can be greatly inflated due to shallow penetration and reduced sensitivity to bottom
reflectance."
 
 
So, we all face the same grim reality, no  matter what!
  • Murky waters tend to increase the upwelling signal
  • This inevitably results in under-estimation of retrieved bottom depth
    • no matter what
    • in 4SM, such situation translates into conspicuously dark bottoms, which should attract the attention of a seasonned practitioner: see 4SM at Arcachon Inlet.
Puerto Rico
"Below 15 meters, depth retrievals show a consistent bias resulting in deeper values than actually measured. The bias results most likely from overcorrecting the atmospheric component."
 
With all that sophistication in DG's process,
one then must be warry of "overcorrecting the atmospheric component".

But 4SM is not exposed to the same hazard
where/if the deep water radiance Lsw is over-estimated,
and conversely if it is under-estimated.
LIDAR seatruth
at Buck Island Reef, Caribbean


 
LIDAR seatruth 
at Buck Island Reef, Caribbean

Compare with 4SM at Buck Island Reef