Assessment of current situation, and validation of 4SM

"What is your assessment of the current state of the art and your algorithms in particular?" (sept 9th 2011)

For me, the "current state of the art" is

• After over 30 years of generous funding,
  • the basics for a simplified radiative transfer equation have been very well documented (by Philpot then Maritorena), along with the risks one has to take/assume prior to "operating" this model.
  • no one should indulge in ignoring the water volume reflectance, lest they'll have to pay the price in the form of badly under-estimated depths over darkest bottoms.
• The situation is in a mess.
  • many pretend to be quite happy with their own even more simplified version of that old story, and still find peer reviewers to buy their story. One can even find champions of Neural networks, PCA, Multiple linear regressions, Levenberg-Marquardt, Ratios of logarithms, End members libraries, you name it, not bothering a bit about the laws of physics in the background.
  • a distinguished few go all the way into accounting for all the physics : we'll meet them in a few decades for any operational application.
  • no one actually speaks about NRL's solution using hyperspectral data in orbit : not really an affordable "dual-use" technology?
  • this is still an efficient way of raising R&D funds, if only from US Navy, NASA, ESA, data providers,...
• All this shows
  • how badly the establishement wants to see a practical/affordable generic solution to come forward.
  • how badly the establishement has been abused by countless pretenders with not much of a background in underwater optics.

"and your algorithms in particular?" (sept 9th 2011)
For me, the 4SM suite of tools solves quite a number of pending aspects

• 4SM reads the raw image data in the form of plain DNs.
• 4SM proceeds with an  "on the fly"/automatic process for
  • estimating sun/sky glint regressions,
  • estimating the spectral deep water reflectance,
  • estimating a spectral Soil Line, a concept well documented by RS agronomers
  • extracting calibration dada
  • estimating Ki/Kj, a concept well documented by Lyzenga and followers
  • estimating spectral K, a concept well documented by Jerlov, then Kirk
• The practitionner then proceeds with a fully supervised fine-tuning of the calibration parameters (better brush up yours basics of underwater optics)
• A practical way of estimating an effective spectral atmostheric path radiance is built in: all computations use the image's native radiances in units of... DNs.
• This in turn allows for estimating spectral Lw (should I say ballpark): the effective water volume reflectance is a key stone of the simplified radiative transfer equation.
• Measuring Ki/Kj is really quite easy in most cases, provided bright sands are present at various depths,  if only as isolated pixels.
• Effective two-ways spectral K are derived from Jerlov's data
  • this requires a precise value of the operational wavelength
  • comments by potential "big wig" wouldbe bloggers like Andre Morel, Stephane Maritorena, William Philpot, Curtis Mobley, Arnold Dekker, Deepak Mishra, Tim Malthus are invited.
• Upon backward inversion of the model, Z is increased until
  • some kind of acceptable fit of the spectral water column corrected radiances is reached
  • against some version of the Soil Line observed in the image.
• 4SM writes out a DTM in meters of the shallow water area, as well as spectral water column corrected DNs ready for bottom typing.
• 4SM includes shallow bottom typing using water column corrected radiances
• All this in one single day in one single environment, using only OpenSource system calls and OpenEV as the image viewer.
• Problems arise when modeling closer to limit conditions , like negative bottom contrast, plumes of muddy waters: over 15 years of experience are imbedded in the 4SM code to try and alleviate these difficulties. More experience welcome.
• All this ahead of any field data/work:
  • Final_Z = Tide_Correction + CoefZ * Computed_Z
  • Tide_Correction and CoefZ to be derived from some seatruth data
  • Labels for bottom types to be derived from some field data

"Has there been enough validation of your algorithm to interest commercial users? Would it be useful to do more?" (sept 9th 2011)

• Apart from Tommy Lindell, and later Gavin Manson, no one ever took up the challenge of venturing into uncharted waters to speak out their views about 4SM or its products, other than with the sheer intent of putting it down, as they are not the ones to shine - or so I suppose.

Partagez sur les réseaux sociaux

Catégorie

Autres publications pouvant vous intéresser :

Commentaires :