Optical calibration, bathymetry, water column correction and bottom typing of shallow marine areas, using passive remote sensing imageries
A series of 23 Landsat ETM/TM 4-bands image of Caicos Bank, Bahamas
Images from USGS/Glovis, UTM projected

Co-registered images provided by FUGRO-NPA
return to 4SM Study Cases
Seatruth is available from BILKO package, and reprojected to UTM19 by FUGRO-NPA
a very large image: Caicos Bank is 80 km across


Most of the following was made possible
thanks to co-registered images provided by FUGRO-NPA.


caicos image menu
caicos combined depth

see also tmnov tutorial
Calibration of Landsat images : how this done


 
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
 
caicos image menu
caicos combined depth 


 

BPL and Jerlov
on narrow waveband hyperspectral images
  • LINEAR: the various linearized BPL pixels plot as straight lines over the whole depth range
  • CONSISTENT: a consistent set of Ki/Kj ratios is observed among all pairs of spectral bands i and j in the visible range
  • OPERATIONAL 2K: any of these ratios in the visible range is observed to allow for deriving spectral operational diffuse attenuation coefficient 2K from Jerlov's data
    • except in the 575-600 nm range where observed 2K values must be set at slightly lower values, because Jerlov data are provided by increment of 25 nm
  • CoefZ: computed depths are therefore calibrated in meters,
    • and only need to be multiplied by a final depth correcting factor to be derived from seatruth
 

This remarquable and clear-cut observation relies on narrow hyperspectral bands

 



 


BPL and Jerlov
on wideband multispectral images

a wideband may be considered to be comprised
of several closely spaced narrow bands

 

  • First problem: multispectral bands are very wide
    • we can't assume that the diffuse attenuation properties remain unchanged over the whole depth range
    • as the bottom depth increases, the upward flux of photons probably gets depleted of its longest wavelength component
    • as a result, the operational wavelength for wideband images would be seen to decrease as the depth increases
    • this would seem to be the case for the Red band
  • Second problem: the Green band spans well over the 575-600 nm range where gradient is strongest in Jerlov's data
  • As a result, the optical behaviour of the Green band of ETM and ALI images would be quite complex indeed
    • ETM and TM and ALI
    • IKONOS
    • SPOT no difficulty was observed
  • Third problem: illumination conditions
    • Jerlov's dataset refers to measurements at sea level, clear blue skyes, sun high in the sky
    • this is not allways the case in real life
    • scenes at high altitude: Greenland at ~1200 m Bolivia at ~3700 m
    • overcast skies, adjacency effect: Red Sea, Persian Glulf, Bahamas
    • the contribution of the skydome to the downwelling irradiance might well be the clue to our problems
  • Fourth problem: does the existence of biofilms at the bottom surface alter the spectral bottom reflectance?
  • See an illustration of the problem using ETM at RasHatibah, Red Sea
  • First problem: multispectral bands are very wide
    • we can't assume that the diffuse attenuation properties remain unchanged over the whole depth range
    • as the bottom depth increases, the upward flux of photons probably gets depleted of its longest wavelength component
    • as a result, the operational wavelength for wideband images would be seen to decrease as the depth increases
    • this would seem to be the case for the Red band
  • Second problem: the Green band spans well over the 575-600 nm range where gradient is strongest in Jerlov's data
  • As a result, the optical behaviour of the Green band of ETM and ALI images would be quite complex indeed
    • ETM and TM and ALI
    • IKONOS
    • SPOT no difficulty was observed
  • Third problem: illumination conditions
    • Jerlov's dataset refers to measurements at sea level, clear blue skyes, sun high in the sky
    • this is not allways the case in real life
    • scenes at high altitude: Greenland at ~1200 m Bolivia at ~3700 m
    • overcast skies, adjacency effect: Red Sea, Persian Glulf, Bahamas
    • the contribution of the skydome to the downwelling irradiance might well be the clue to our problems
  • Fourth problem: does the existence of biofilms at the bottom surface alter the spectral bottom reflectance?
  • See an illustration of the problem using ETM at RasHatibah, Red Sea





 


 
the BPL assumption holds for




 


Using BILKO's seatruth dataset

 We want to use the BILKO seatruth dataset at Caicos Bank, Bahamas
to try and unfold some of the complexities of underwater optics
using a time series of real life Landsat ETM images

  • Depths.xls file from Bilko dataset contains 515 depth points?
    • BILKO's dataset is UTM zone 19
  • USGS image LT50090451990326XXX05  is WGS83 UTM zone 18
  • NPA-FUGRO have reprojected this time-series  dataset  of 23-scene subsets
    • from USGS's  UTM zone 18 
    • into BILKO's UTM zone 19
  • NPA-FUGRO also verified and cleaned Bilko's original Depth.xls
    • and exported it to Depth_points_reproject.shp
      • which now contains 511 depth points
  • shp2text read it into depth_points_reproject.txt text file
    • it contains 511 depth points, formated XUTM YUTM Z
  • seatruth regression, and inspection of the results in the OpenEV display, show that
    • some odd depth point results are too close to land
    • some odd depth points results are not representative of good and consistent results in their immediate surrounding
      • in view of the 30 m pixel size on the ground
      • in view of the necessity to smooth the data prior to modeling
    • some depth points consistently yield badly underestimated depth: they are located over the steeply sloping edges of the Caicos Bank
    • some depth points are located over optically deep waters
    • some depth points simply don't seem to belong, for no reason
  • Weeding out 43 louzy points:
    • all the above depth points are weeded out of the dataset (actually "commented")
    • the cleaned seatruth dataset is named depth_points_reproject_pruned.txt
      • it now contains 468 depth points
      • this new file is then used for seatruthing

Seatruth Sounding Lines Wanted

 

Calibration : how this is done


Sea truth is wanted along these segments
 

 Estimated tide height relative to ETM_009_045_1999-09-20
(was estimated using CombinedZ)

First ETM SLCON  
ETM_009_045_1999-09-20 0.00
ETM_009_045_2001-01-28 0.54
ETM_009_045_2001-07-07 0.26
ETM_009_045_2002-02-16 0.87
ETM_009_045_2002-07-10 0.30
ETM_009_045_2003-01-18 0.89
ETM_009_045_2003-04-24 0.10
Now ETM SLCOFF  
ETM_009_045_2007-02-14 0.41
ETM_009_045_2007-05-05 0.61
ETM_009_045_2007-12-15 0.55
ETM_009_045_2007-12-31 0.45
ETM_009_045_2008-02-17 0.24
ETM_009_045_2008-04-05 1.05
ETM_009_045_2008-04-21 0.52
ETM_009_045_2008-12-17 1.54
ETM_009_045_2009-01-02 1.39
ETM_009_045_2009-01-18 0.37
ETM_009_045_2009-10-17 0.60
ETM_009_045_2009-11-18 1.31
ETM_009_045_2009-12-04 0.76
ETM_009_045_2010-03-10 0.00
ETM_009_045_2010-03-26  
ETM_009_045_2010-04-27 0.48
TM  
TM_009_045_1990-11-22  
caicos combined depth  
 



 
Better than Lyzenga's method
 
  • Lyzenga's method is still widely used
    • It needs an existing depth sounding dataset that covers faithfully the whole ranges of bottom brightness and shallow depth
    • It does not account for water volume reflectance: that's why it needs
      • segmentation of shallow areas into several bottom brightness classes, irrespective of depth
      • multiple linear regressions in order to specify Z=aX + bY + c for each class
  • With 4SM, in order to get a similar result (and also a water column corrected spectral image suitable for bottom typing),
    • no need for atmospheric correction
    • no need for calibration into units of reflectance
    • no need for image classification/segmentation
    • no need for any field data
  • With 4SM, seatruth data is only needed later on, in order to specify
    • Htide
    • CoefZ
    • ==> FinalZ = Htide+CoefZ*computedZ
  • Lyzenga's method is still widely used
    • It needs an existing depth sounding dataset that covers faithfully the whole ranges of bottom brightness and shallow depth
    • It does not account for water volume reflectance: that's why it needs
      • segmentation of shallow areas into several bottom brightness classes, irrespective of depth
      • multiple linear regressions in order to specify Z=aX + bY + c for each class
  • With 4SM, in order to get a similar result (and also a water column corrected spectral image suitable for bottom typing),
    • no need for atmospheric correction
    • no need for calibration into units of reflectance
    • no need for image classification/segmentation
    • no need for any field data
  • With 4SM, seatruth data is only needed later on, in order to specify
    • Htide
    • CoefZ
    • ==> FinalZ = Htide+CoefZ*retrievedZ




 
Nearest Neighbor Resampling, please!
 

Landsat ETM SLC OFF
Nearest Neighbor Resampling, 
PLEASE,
as this is not workable

we do not want
thhose fancy pixels