Optical calibration, bathymetry, water column correction and bottom typing of shallow marine areas, using passive remote sensing imageries
Busy? 4SM in 10 lines
Review of some papers collected on the net

This review shows how hot this area of R&D still is!  

return to 4SM Study Cases

As it turns out, this review shows that most of those
go through formal atmospheric correction and  calibration  to reflectance 4SM does not
are mostly stuck with either Lyzenga's model or Stumpf's model 4SM innovates
need existing depth sounding to yield water depth 4SM does not
don't even mention water column correction and bottom typing 4SM does it all

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
I keep digging
until suitable data
become available

log-linear bathymetric inversion approach
Lyzenga+depth 2006
Conger, Hochberg, Fletcher & Atkinson 2006
Arce's thesis in Puerto Rico 2005
Sagawa et al
Nurlidiasari et al : bottom typing of coral reefs in Indonesia 2005
Legleiter OBRA
SHOM 2012    


Lyzenga+depth in 2006
 This paper in 2006, by the founding father of the most widely used algorithms with two other big wigs of the US accademic scene,
  • elaborates on various aspects of the simplified radiative transfer model
  • and attempts to improve the estimation of shallow water depth
    • Though reading
  • GREAT presentation of seatruth correlations
    • using eight IKONOS images in clear waters and excellent LIDAR seatruth datasets which are used for the calibration of model parameters
Authors claim that, thanks to rather obscure refinements, their new "algorithm corrects for a range of variations in both water attenuation and bottom reflectance"
  • Estimation of water attenuation properties is still conditional to the use of an adequate seatruth dataset
  • The result is still just an estimation of water depth, and the estimation of spectral bottom reflectance is not even mentioned

Fig. 15 of Lyzenga et al.
log(Lsgreen-Lswgreen) vs log(Lsblue-Lswblue)


  • "Clearly, there is a wide variation of the signals within each of these depth intervals, which could be caused by variations in either the water optical properties or the bottom composition "
  • I say: Nope! Accounting for the fact that Lwblue >> Lwgreen explains nicely the curved display of these clouds of linearized pixels



Fig. 15 is flipped for comparison with 4SM illustrations
log(Lsblue-Lswblue) vs log(Lsgreen-Lswgreen)


Variable diffuse attenuation properties

  • what worries the authors is the curved aspect of displays of pixel clouds
  • authors claim that the curved fit observed is caused by some obscure variations in the diffuse attenuation properties of the water which they now undertake to estimate and account for...Hmmm!


Spectral water volume reflectance

  • in 4SM, we have observed that the curved aspect of isobath lines in such a scatter plot is because Lwblue>>Lswgreen
  • Nothing to do, therefore, with alleged/observed variations in the diffuse attenuation properties of the water
    • Fig. 13 shows them to be "uniform",
    • as is the case with all faithfull images we have investigated todate in 4SM
  • Please refer to optical_modeling_slide47
  • Please refer to optical_modeling_slide79

The authors' method

  • is quite confusing from an operational point of view
  • require adequate seatruth dataset for calibration
  • uses just the blue and green pair of bands
  • only allow for "depth-invariant" images of the variations of bottom reflectance to be produced
  • still do not account for the water volume reflectance!!!!!!!!!! in spite of an all-out review of all aspects of the model


No sweat then: although it uses the same simplified physical approach, the 4SM approach

  • is much simpler on an operational point of view
  • does account for water volume reflectance Lw
  • does not require any seatruth dataset,
  • can use all bands with significant bottom detection (instead of just the blue and green pair)
  • delivers
    • computed depths which should compare favorably with those in this paper
    • spectral K i.e. water quality
    • spectral water-column-corrected bottom reflectance
I ought to look into the sun angle aspect though, as it might explain the alleged shift in wavelength I seem to observe.

Conger, Hochberg, Fletcher & Atkinson 2006
"A New Method of Decorrelating Remote Sensing Color Bands from Bathymetry in Optically Shallow Waters"

Nice reading, keeps track of physics behind
  • "The goal is to remove the variability due to depth while maintaining variability due to bottom composition"
  • Innovates from Lyzenga's method
  • Needs existing bathymetric data for calibration : "this method requires a depth value for each pixel to be rotated" : uses a Quickbird image in Hawaii and SHOALS Lidar bathymetry
  • "There is also potential that our pseudo-color bands may be calibrated to absolute reflectance through techniques such as the empirical line method, thus allowing application of spectral classifiers built using libraries of in situ reflectance data"

I say: use 4SM
  • you won't need any existing depth information
  • for most shallow pixels, your output shall be
    • a depth value
    • N spectral values for 2Kd (water quality)
    • N spectral values of La, the atmospheric path radiance
    • N spectral values of Lw, the water volume reflectance
    • N water column corrected spectral bands in units of DNs at the base of the atmosphere , that you can convert into units of reflectance if you so need , ready for bottom typing
    • and a deglinted image , whether BOA or TOA

Jeanette Arce's thesis 2005

"Remote sensing of benthic habitats in SouthWestern Puerto Rico"   

  • Preprocessing of Ikonos image
    • atmospheric correction using dark pixel subtraction
    • sunglint correction using Hochberg's technique
  • Preprocessing of Hyperion image
    • destriping
    • atmospheric correction using ACORN
    • deglinting using Lee's "750 Normalizing" technique
    • followed by supervised classifications
  • Water column correction using Lyzenga's technique, followed by supervised classification

This spectrum for corals (right) is remarkable,
appears locally in water column corrected data at Caicos Bank



Sagawa et al


Easy reading,
pragmatic approach to water column correction,
quite simplistic though


  • Bottom typing: authors only want to do bottom typing over seagrass
  • Lw: first time I see someone noting that the water volume reflectance is not accounted for in Lyzenga' method
    • L = Ls + ar exp( − K g Z )
    • authors do not mention atmospheric correction from TOA radiance to BOA radiance
  • Please compare their model with ours:
    • TOA: Ls= La + Lw - Lw/ exp (2KZ) + LB/ exp(2KZ)
    • BOA: L =         Lw - Lw/ exp (2KZ) + LB/ exp(2KZ)
  • Jerlov and effective K: authors could have resorted to Jerlov's data in order to derive operational K, and spared the burden of using seatruth data. Particularly so as they use an existing DTM in addition to side scan sonar data
  • What they do is use Z and spectral K at the current pixel, so that they can solve for spectral bottom reflectance ar
    • just like Malthus and Karpouzli whose paper in 2003 authors seem to have missed: how sad
  • "bathymetry data accuracy is of primary concern"
  • Their method does not work in the 0-2 m depth range: how sad!
    • not because of internal reflections,
    • but rather because they need to include Lw in their TOA model or apply atmospheric corrections to the image data
  • Blue vs Green: so they do not use the Red band at all
  • Posidonia on sand: that's an easy binary task indeed


NURLIDIASARI et al   water column correction and bottom typing

Work done in 2004 at ITC, Enschede, the Netherlands, on a QuickBird image
  • elaborate process to calibrate the data into BOA reflectance
  • elaborate process to estimate spectral deep water radiance, based on field work
  • elaborate process to estimate ratios Ki/Kj for Lyzenga's algorithm for water column correction
    • variance and covariance
    • Kblue/Kgreen~=0.74, quite neat
    • Kgreen/Kred  ~=0.60, lousy
  • Linearization : at the base of the atmosphere Xi = ln(Li)         should actually be            Xi = ln(Li - Lwi)
  • Depth-invariant indexes
  • Bottom typing for 6 classes, using only the Blue/Green depth-invariant image
    • this is equivalent to shades of grey, from brightest to darkest
  • Water column correction improves the acuracy of the bottom typing by 22% for patchy seagrass, patchy reef, sand and algae
With 4SM
  • Image preparation and full processing is quite simple in comparison
    • no need for field data
    • no need for conversion into BOA calibrated radiances
    • a same-day process
  • Bayhymetric map and Bottom typing map are prepared ahead of any field work
    • Computed depths still need some seatruthing   Zfinal= bias + slope*Zcomputed
      • bias accounts for the tide height correction
      • slope accounts for real atenuation coefficients in the Blue  and Green bannds
    • Bottyom typing map only needs
      • possibly binning of classes
      • a label for each class
  • Field work much easier,
    • using pre-classified image and detailed bathymetric map


Legleiter : the OBRA approach 2009
  • River bathymetry using Hyperspectral images
  • Calibrate a optimal band ratio using existing depths : Optimal Band Ratio Analysis
  • An extension of the "linear approach"