Optical calibration, bathymetry, water column correction and bottom typing of shallow marine areas, using passive remote sensing imageries

A CASI image of Prince Edward Island, Northshore, Nova Scotia, Canada
CASI data 2000, courtesy of HDI, Halifax, Canada, pixel size 2m, UTM
Seatruth 2001, courtesy of Gavin Manson, BIO, Halifax, Canada

  • 4SM work done in 2001 as a blind test,
    • then in february 2002 using the seatruth data,
    • CASI was new to everyone around, image had to be reprocessed for radiometric and geometric distrotions
    • 4SM was still a newborne application to hyperspectral data!
  • This dataset lacks coverage of optically deep waters.
    • This possibly caused  the bad underestimation of depths over dark bottomed troughs.
  • Still quite an achievement back in those days.
  • Soon after, Manson and colleagues were assigned a new position in Alaska later in 2002: I did'nt hear of him since.

The best seatruth in town
See work done february 2012

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

The following material refers to CASI Line 1 above


Sand waves along the northern coast 
of Prince Edward Island, 

in the Gulf of St Laurence, Canada

images from Gavin Manson


4SM CASI bathymetry
 image from Gavin Manson
  • This was a blind test in 2001
  • CASI coverage lack optically deep waters
    • ?this was the cause of the bad artifact seen in profile 6700

4SM CASI bathymetry
 Seatruth profile by Gavin Manson

  • This was a blind test in 2001
  • Optical calibration is quite good indeed, apart for the trough (see below):
    • no field data were used in this blind test

From: Manson, Gavin <gmanson@nrcan.gc.ca>
To: 'ygmorel@attglobal.net' <ygmorel@attglobal.net>
Cc: 'Herb Ripley' <herb.ripley@hdi.ns.ca>; 'William Jones' <bill.jones@hdi.ns.ca>
Subject: Progress and plans etc.
Date: Friday, December 21, 2001 5:45 PM


Thought I should check in with you regarding progress on our end. As you may know, we had a meeting a few weeks ago with Herb and Bill, as well as Gary Bugden et al. to discuss progress and directions for our CASI data. We have a tentative agreement with HDI to have the north shore imagery properly radiometrically corrected and georeferenced (cost estimates are pending - we don't have a whole lot of money so hopefully they'll be relatively low).

When/if this is done we'll get the files to you and, if you're willing, try
again from a correct starting point.
I am in British Columbia until Jan 7
and will continue pursuing this when I return.

I am assuming this will all go ahead and have prepared an abstract
(attached) for a meeting in Toronto in June. (International geoscience
geoscience and remote sensing symposium 2002
http://www.igarss02.ca/Authors/CallForPapers.html). It is too late to ask
you this for the initial abstract as I need to send it immediately (sorry
for not allowing comments!) but do you have an affiliation or company name
for future reference? I have tried to be somewhat vague about our final CASI
results as our work is not complete yet.

Have a merry christmas, talk again in the new year


These results
prompted a review
of the calibration/modeling
in february 2002, using the 
which is illustrated below

4SM CASI bathymetry
 Seatruth regression for 11 profiles by Gavin Manson


4SM results (february 2002)

This exercise is a clear proof
that the calibration of the model
using Jerlov's data
is trustworthy,
provided the effective wavelengths
are precisely specified

below : image_Z, image_B, and image-BSC for CASI Line_1


February 2002

  • This dataset lacks a real coverage of optically deep waters

  • Sea truth was acquired one year after CASI imagery in this highly dynamic environment.

  • Optical modeling results reported above by Manson et al. were obtained prior any field data was communicated to Y. Morel.


The best 4SM seatruth in town
work done 14 february 2002,
using the new radiometrically and geometrically corrected CASI image
and the complete seatruth dataset

Lsw5 = 42

Lsw5 = 43

Lsw5 = 44

band 5 is 577 nm
So it would appear that a bad choice of deep water radiances
which are nowhere to be sampled in the image
resulted in a a bad underestimation of the depth
over the dark bottomed troughs 
in the blind test

Please, Gavin, consider some alterations to your paper!

My advice would have been :

  • first try Landsat to get a feeling of the scene
  • then use multispectral satellite images: they'll be good enough for bathymetry purpose
  • as CASI does not carry any significant advantage fir such a sandy environment
  • but, I know, in those days, going CASI was the ultimate craze!!



 By application of an input K[799nm]=4.56 m -1 in the -KK argument of the command line , 
a CoefZ=1.33 is proposed by the calibration process in calibration diagram ns-E calibration.
The details are as follows:

  • Left alone, initial calibration through Jerlov's water type yields K[799nm]=6.075 m -1:

  • Left alone, this initial calibration yields a sea truth regression for Line6100 of ZC=0.02+0.75*ZR (N=167; r 2=0.92; RMS=0.94 m) ns-E RegressZZ Line6100 1.

  • This suggests that a value of  CoefZ=1/0.75=1.33 is in order.

  • ==> The application of a value of CoefZ=1.33 brings about an excellent fit for Line6100: ZC=0.03+1.00*ZR (N=167; r 2=0.92; RMS=0.36 m), as well as for most of the whole seatruth dataset  ns-E RegressZZ ALL.

  • We can now derive an operational value for K[799nm]=6.075/1.33=4.56 m -1, which is virtually twice the absorption coefficient for pure water at this wavelength.

  • Therefore we can accept  that K[799nm]=4.56 m -1 is a very satisfactory operational value.

  • Using this last value as a seed value to the complete set of Ki/Kj ratios observed in the image, we can then derive the whole series of spectral operational K values for the whole CASI bandsetting.

  • This suggests that, using published values for the absorption coefficient  of pure water in the Near InfraRed region of the solar spectrum, it might turn out that a quasi-absolute optical calibration of shallow water optical modeling may be achieved without the use of any field data.

True color composite                                                  Water column corrected true color composite

  • The sea truth dataset was acquired in summer 2001,  one year after the CASI imagery. 
  • Major storms have affected the scene between those two data collection exercises,
    • which have affected the location and the shape of the sand waves.
  • Sand waves appear to rest on a bed of red-rich dark bottom deposit.


  • In sea truth regression for all_lines ns-E RegressZZ ALL ,
  • In the sea truth profiles for the blind test ns-E_ProfileZZ ,

    • the computed depths (colored profile) across the dark patch ns-E_dark_bottoms

    • appears to have been badly underestimated (Line6400 to Line6900),

    • as the trough between sand waves is shown to be flat and shallow at approx 3.5 m in the computed depths

    • while sea truth (black profile) shows depths ranging from 4 to 6 m one year later.

  • These "faulty" pixels exhibit radiance levels well above deep water radiances in the 650-670 nm range: ns-E_650nm

    • This radiance level is ~7 DNs above deep water radiance in 8-bits scaled bands at 650 nm.

  • Maximum depth of bottom detection for bright sands does not exceed ~5-4 m over this wavelength range,

    • even much less for a dark bottom.

  • Applying a depth 1-2 m deeper would result in unacceptable extremely bright computed bottom reflectances

    • whereas it is clear that the bottom  exhibits distinctly dark features: see ns-E_RGB where a raw RGB view is shown on left, and a "water column corrected" RGB view is shown on the right.

  • Note in this last image that most bright bottoms exhibit a greenish computed bottom spectral signature.

    • either because the bottom has really changed since sea truth data acquisition

    • or because of the growth of algae/seaweeds , in which case the top of the canopy is detected instead of the true bottom.

  • It is therefore likely that genuine bottom detection  actually occurs at 650 nm



This shows that the water-column corrected radiances are available for thematic classification

  •  Scatter plot of spectral bottom reflectance inside the red polygon above

    • Left :   TOA raw radiances Ls[490] versus Ls[547].

    • Right: BOA normalized water column corrected radiances LB[490] versus LB[547].


Water column corrected spectral bottom signatures
ready for bottom ytping

Five BOA normalized water column corrected spectral bottom reflectance signatures are illustrated 
above a backdrop of true color composite of BOA water column corrected radiances

  • for 10 spectral bands ranging from ch_46=490 nm to ch_55=690 nm

  • for   5 distinct targets (black, red, blue, green, yellow).

Band setting of the CASI
North_Shore : for the 19 channel file here is the band listing:

   1 [16U] 435.1nm+/- 15.8nm (rows 261-277) DNSRU:1000.000000
   2 [16U] 462.7nm+/- 12.2nm (rows 248-260) DNSRU:1000.000000
   3 [16U] 490.4nm+/- 10.4nm (rows 234-244) DNSRU:1000.000000
   4 [16U] 522.1nm+/- 12.3nm (rows 216-228) DNSRU:1000.000000
   5 [16U] 547.4nm+/-  7.7nm (rows 205-212) DNSRU:1000.000000
   6 [16U] 565.2nm+/-  4.9nm (rows 197-201) DNSRU:1000.000000
   7 [16U] 577.5nm+/-  7.7nm (rows 189-196) DNSRU:1000.000000
   8 [16U] 607.7nm+/-  7.7nm (rows 173-180) DNSRU:1000.000000
   9 [16U] 624.7nm+/-  5.8nm (rows 165-170) DNSRU:1000.000000
  10 [16U] 650.4nm+/- 10.6nm (rows 149-159) DNSRU:1000.000000
  11 [16U] 665.6nm+/-  4.9nm (rows 144-148) DNSRU:1000.000000
  12 [16U] 682.7nm+/-  3.0nm (rows 136-138) DNSRU:1000.000000
  13 [16U] 690.3nm+/-  4.9nm (rows 131-135) DNSRU:1000.000000
  14 [16U] 702.7nm+/-  7.8nm (rows 123-130) DNSRU:1000.000000
  15 [16U] 715.1nm+/-  4.9nm (rows 118-122) DNSRU:1000.000000
  16 [16U] 735.2nm+/-  5.9nm (rows 107-112) DNSRU:1000.000000
  17 [16U] 745.7nm+/-  4.9nm (rows 102-106) DNSRU:1000.000000
  18 [16U] 800.1nm+/-  9.7nm (rows 71-80) DNSRU:1000.000000
  19 [16U] 849.8nm+/-  9.7nm (rows 45-54) DNSRU:1000.000000

not used
not used
not used
not used
not used
not used
not used
not used

Work done 26-27 february 2012

Raw TCC : quite conspicuous 

Raw FCC : no glint

NIR band at 850 nm

Water column corrected

  • all water column corrected shallow bottoms are very dark


Water column corrected, Normalized

  • this enhances the spectral bottom signature,
    • in view of bottyom typing
  • ?Shallow bottoms appear to have a greenish signature

Computed depth

Average bottom brightness

  • all water column corrected shallow bottoms are very dark
  • shalow sands in the inlet are the same as along the SW coast 
  • bottoms just in front of the inlet are extremely dark
    • ?this extends all the way along the NE coast
  • Now, ?I see no sign that the waters in the inlet would be less clear

Calibration 11_8_6_1

  • Jerlov water type is half way between            Coastal 1 and 2
  • shallow pixels are very dark
  • Knir is confirmed at 4.5 m-1

BiDimensional histograms


  • shallow pixels are the tiny group with distinct exponential decay shape, with radiances hardly exceeding 100 in all bands
  • it is seen that shallow bottoms are very dark, with radiances hardly exceeding 100 in all bands,
  • while the 100-255 range depicts exclusively land pixels and wave breakers


Work in progress:
Seatruth 2012