Krista Lee 2012 




My comments

Stuffle 1996 Bathymetry from HYDICE imagery at Lake Tahoe
  • LIDAR bathymetry for seatruth
  • Modtran atmospheric correction
  • Field measurement of content in chlorophyll, then Hydrolight for estimation of attenuation coefficients
  • Bierwirth and Hamilton algotithms for computing depth
  • masks for sandy areas, for dark rock areas, and for bright rock areas: need to be calibrated and processed separately
  • "the bottom reflectance characteristics must be included in the bcalculation"

  • 1996 : this is when I first presented my results
    • Morel, Y 1996; A coral reef lagoon as seen by SPOT. Proceedings of 8th Australasian Remote Sensing Conference, Canberra (1996)
  • 1996 : in view of forthcoming US Navy's NEMO project, NPS had great expectations
  • bright sands, bright rocks and dark rocks through interactive manual masks:
    • three bottom types is not enough,
    • 4SM automatically accounts for variable botttom brightness (no need for interactive masks)
    • so that all flavors of mixed pixels are correctly modeled
Fisher 1999 
Shallow water bathymetry at Lake Tahoe from AVIRIS data
  • Similar methodology applied on the whole lake Tahoe using AVIRIS data
  • in view of forthcoming US Navy's NEMO project, NPS had great expectations
  • which were not met yet by the time NEMO disappeared
  • see also USAF's  Tactical-satellite-3    and ORS-1
  • "The ideal scenario would be one where this process is automated, but convolving of the MODTRAN and HYDROLIGHT outputs to fit the AVIRIS sensor channels, along with conversion of AVIRIS radiance data to reflectance data (and the many possible methods to do this), makes this algorithm unique for AVIRIS only (not HYDICE or the other types of hyperspectral sensors) and in itself generates errors"
  • so I understand 4SM operates the "ideal scenario"...

Clark 2005 Ikonos and Quickbird at LooeKey, Florida

  • "Naval satellite bathymetry: A performance assessment"
  • SHOALS LIDAR seatruth
  • Quoted from Camacho : The results demonstrated that the ratio method proved sensitive to bottom type.
    • It produced shallower depths over bottom types with low albedo
    • and deeper depths over bottom types with high albedo.
  • Furthermore, the ratio method failed completely in
    this study at depth less than 1 meter over sand and coral.
  • Finally, the maximum depth obtainable using the ratio method was 15 meters.
  • The incorrect outputs of depth for different substrates indicate that this technique could be potentially improved through pre-classification of bottom substrate.
  • It took NPS six years to recover, thanks to the advent of very high resolution imageries Ikonos and then Quickbird
  • This study is not available on the Internet (restricted???)
  • Fisrst attempt at Stumpf's ratio method
  • Clark's conclusions in 2005 on NOAA's log ratio method mirror closely my 4SM conclusions formulated in 2003: see 4SM presentation at slides 60 and 82
  • Camacho's comments blame it the bottom substrate
  • This is not the case :  Stumpf's ratio method (and many other methods as well) ignores superbly the first order role of the water volume reflectance Lw over dark bottoms
  • Therefore, as bottoms get darker, computed depths get more and more overestimated (just like in Bierwirth's method as reported by Stuffle)
Densham 2005  QuickBird at Looe Key, Florida
  • SHOALS Lidar seatruth
  • Atmospheric corrections
  • Deglinting
  • Conversion to Reflectance Values
  • Estimation of content in chlorophyll
  • Estimation spectral Kd using Hydrolight: this takes the following input: average chlorophyll concentration, pressure, temperature, humidity, visibility, precipitation, longitude, latitude, wavelength, solar zenith angle and satellite zenith angle
  • Stumpf's method and Jupp's DOP Stratified Genetic Algorithm
  • existing depth points are used
  • Recommandation: segment the image into bottom classes, then tune the depth model for each of these bottom classes; this shall be undertaken by Camacho 2006
  • Quoted from Camacho:  variable bottom types affected the ratio method significantly ... variable bottom type should be analyzed individually for a potential improvement to depth outputs
  • So the frail Blue/Green ratio is used to estimate/map (i) the chlorophyll content of the water and (ii) also its depth??
  • Spectral Kd may be derived from regressing linearized radiances against existing depth points in the image: why use Hydrolight at all?
  • Notice the very high level of system noise in fig. 11, 12 and 13!!!!!!!!
  • Band Math tool in ENVI: I have some experience of squeezing water column correction for Landsat images intoERDAS's  band math tool: just good for a one-off demo, as you end up with a giant macrocommand which
    • incorporates all the very many numerical specifics of your current image,
    • cannot account for the necessary -and numerous- nuances and conditions that are necessary to accomodate a real life area of interest and dataset
    • would become a real nightmare with WV2's eight wavebands
  • Computed depth is the only output
  • See 4SM results at LooeKey, using just ALI data in 2010

Camacho 2006 Quickbird at Midway atoll for bathymetry and bottom typing

  • Following shortcomings of Stumpf's ratio method observed  by Clark and by Densham:
  • Seatruth for depth from scuba dives and nautical chart
  • Conversion to radiances in physical units
  • Atmospheric correction
  • Deglinting
  • Preliminary bottom typing using  field work and ISODATA on depth-invariant water column corrected images
  • Application of Stumpf's method and existing depth points for computing depth
    • for the whole image, which yields badly underestimated depths over bright botttoms by an average of 2.5 m
    • then separately over "sand" and "coral/algae" bottoms, with just  "marginal" improvements
  • Quoted from Camacho :
  • Both bathymetric outputs (entire image and variable substrates) generated erroneous depth values over the bright sandy regions in less than 2m of water depth
  • Today any algorithm available for bathymetry derivation requires field data to scale relative bathymetry to absolute values.
  • Stumpf et al. (2003)’s algorithm does not implicitly compensate for variable bottom type and albedo as was originally concluded by its authors and postulated by Clark (2005)
  • Most importantly, new research should focus on resolving the algorithm’s inability to estimate accurate depths over shallow areas with high albedo.

  •  this is an attempt to save Stumpf's method, in respect of bad results over dark bottoms and also over deeper bottoms at Looe Key, Florida
  • this is attempted in the pristine waters of Midway atoll
  • marginal improvement is gained by processing sandy areas and coral/algae areas separately
  • Nothing here to surprise me
  • 4SM solves all of these problems, and uses only the image itself (not even its metadata).
  • The results are available for "same day" delivery

Arledge &  Hatcher 2008 QuickBird (2.8 m)  and Ikonos (4 m) at Midway atoll for bottom typing

  • ACORN was perfered over ATCOR for atmospheric correction
  • water column correction by Lyzenga's depth-invariant Ki/Kj method
  • two weeks field work
  • for bottom typing at Midway atoll by ISODATA classifier and contextual editing
  • comprehensive seatruth, user and producer accuracies
  • performances are very similar
  • cost benefits analysis
  • "This research does not show any significant improvement in classification accuracy of the QuickBird sensor over the IKONOS sensor for the highly heterogeneous coral reef environment of Midway Atoll."
See also Benfeild et al, 2007   for a much similar study case using ETM+ and QuickBird images in Panama contextual editing and object-oriented classification
  • this work aims at cost/benefits evaluation of higher spatial resolution of multispectral pixel
  • see fig 27 :  R2=0.64  Poor radiometric quality : this entails quite a high level of noise in Quickbird imagery: "deglinted " image shall still be very noisy!
  • see fig 40 : R2=0.92 Radiometric quality is much better
  • In my view : this is enough to justify the choice of Ikonos imagery, unless a 2.8 m resolution  is required
  • No bathymetry is extracted
  • see fig 31 and 44: this puzzles me
    • depth in the central lagoon by far exceed the bottom-dependent maximum penetration depth of the red band
    • therefore in the deep central lagoon, only the BluevsGreen depth invariant image is available, so that the true color composite should show shades of grey, certainly NOT color shades
  • if no threshold is enforced on the minimum reflectance in the red band, the BluevsRed and GreenvsRed depth invariant images shall keep yielding decreasing values as the actual depth increases deeper than the bottom-dependent maximum penetration depth
  • mmf..

Loomis, 2009
Aviris==> simulated WV2 at Kanehoe Bay, Hawaii
Quickbird for comparison

  • Conversion to Top-of-Atmosphere Reflectance
  • Supervised Classification for segmentation of the marine area into two classes: sand and non-sand bottom types
  • depth points from digital nautical map
  • Stumpf's ratio method applies separately to bottom type
  • tide correction
  • SHOALS LIDAR for seatruth
  • ==> yellow/green and yellow/blue ratios improve the accuracy of derived depths
  • "Comparing the green/blue ratio-derived depth values to those derived with the yellow/blue ratio, the error seems to be induced by the blue band"
  • please see 4SM at Kanehoe bay and at Waimanalo Bay
  • sand and non-sand bottom types is not enough: more classes should have been considered
  • this comparison among pairs of bands is enlightening
  • it is a shame though that transects do not display depth-invariant index along with computed depths
  • "the error seems to be induced by the blue band" : I quite agree with that
    • in my view, the water volume reflectance, which is not accounted for in Stumpf's method, is extremely important
      • water volume reflectance is quite strong in the blue band in very clear waters, much less in the green band, and just faint in a yellow band
    • therefore, the "bottom contrast" L-Lw for very dark substrates can become null and even negative for very dark bottoms
    • as Stumpf's method does not account for Lw, it entails quite strong errors over dark bottoms
    • such errors should be minimum when using the Yellow/Green pair of bands

Lee, Kim, Olsen, Kruse, 2011 "Determination of bottom-type and bathymetry using WorldView-2"


  • Quite a team work
  • Conversion to radiance (no atmospheric cottection)
  • multiple masks, no deglinting
  • Field work
  • Unsupervised and supervised classifications for classification of bottom types
  • "Between seven and eleven classes were used for every beach"
  • LIDAR bathymetry
  • A very pragmatic approach to depth estimation

  • This study is the logical follow-up of the above line of action
  • fig 9 why not deglint the image for a start?
  • fig 13 ought to display ln(L-Lw) vs Z 
    • see Philpot, and Maritorena
    • Lw in the green band is significant (think of "green coastal waters")
    • Lw in the blue range just cannot be ignored (think of "blue oceanic waters")
    • that makes quite a difference for dark bottom substrates, and explains many disappointing results reported in the past with depths badly underestimated over dark bottoms
    • I find it disturbing that the authors disregard the optical properties of the waters altogether
  • fig 16: quite interesting.
    • so Yellow depths are correct, and Green depths are ~2 m less than Yellow depths: I think that's because the color of most bottoms tend to be greenish in color
    • this is accounted for in 4SM
  • fig 17: error on depth by the green band is quite large, as observed by 4SM at Oahu
  • so, appears that they can map the bathymetry provided they can use a LIDAR coverage for calibrating each waveband over each of the 7-11 bottom classes? or did I miss the point?
  • even though their line of work is showing progress, they are far from an operational solution (so many steps involving ENVI, IDL and MS Excel), whereas 4SM is ready and all-inclusive
  • After all those years, quite surprinsing to see NPS going back to such crude ways of deriving depth
  • 4SM would not need any LIDAR data to yield both
    • water depth
    • water column corrected spectral bands, ready for bottom typing

Madden 2011 WV2 at Tampa Bay, Florida

  • USGS digital terrain model
  • Atmospheric corrections: dark pixel subtraction method
  • Calibration into water leaving reflectance
  • Stumpf's method for computing depth, calibrated using the DTM
  • Red, yellow, green and blue bands are used : this yields six ratios
  • Tide correction
  • "This study suggests that utilizing all five bands can counteract the contributions of the bottom substrate and allow for tuning of the derived bathymetry"
  • So, seems that, over time, Stumpf's ratio method won its case at NPS
  • "The basic assumption of this ratio algorithm is that the water column attenuation is the dominate factor in the amount of water leaving radiance measured"
  • may be so, although, as the bottom type gets darker, the water volume reflectance comes into significant play, mostly that of the blue bands : see 4SM presentation at slides 60 and 82
Myrick 2011 WV2 at Camp Pendelton
Coastal bathymetry using satellite observation in support of intelligence preparation of the environment
  • this uses two closely subsequent WV2 images to estimate the water depth by a wave celerity method and linear dispersion relationship for surface gravity waves
  • excellent results : "This study demonstrates that determining nearshore bathymetry from space by the wave celerity method is a viable solution for determining nearshore bathymetry of hostile or denied areas"

  • this shows that NPS wants it all!
  • no method left behind
Krista Lee 2012 WV2 at Kailua Bay
  • this study looks at a series of 39 images acquired same day "to determine the effect of acquisition angle on derived depth"
    • from looking forward at satellite elevation angle 25 degrees to looking backward at satellite elevation angle 25 degrees, through near-nadir at satellite elevation angle 78 degrees and looking away from the sun (Fig. 17)
  • after quite some mosaicking, UTM projection, radiance calibration, co-registration, etc, 39 chipsets of the same 1000x1000 subscene were processed for shallow water work
    • land, cloud, whitecaps and sunglint pixels were masked out
    • other pixels were deglinted, then used to derive a water depth by Stumpf et al's band ratio method for four pairs of visible bands (Red band was not used)
    • computed depths were then regressed against LIDAR measured depths
  • Fig 38 shows that , over the 2-8 m depth range,
    • the Blue vs Green pair  performs much better than any other pair
    • satellite elevation less than ~55 degrees makes for undesirable imaging geometries : near-nadir is best

A very usefull study
  • After more than 30 years of continued efforts, this result is not to surprise us, although it is usefull to stress that the payer/end-user must ensure that suitable images are collected/submitted for shallow water work or they'll have to face the odd consequences and not blame the particular petty method used
  • No particular conclusion is proposed as regards the performance of the deglinting process, although near-nadir viewing is known to entail minimum glint
  • No mention is made of the fact that MS1 and MS2 bands must be deglinted separately because of the "0.3 second delay between MS1 and MS2
    • lucky that both NIR1 and NIR2 are available for deglinting MS1 and MS2 bands respectively
    • what of the PAN band: how to deglint it?
  • the study is limited to 8 m of water depth
  • this study does not seem to address the problem of looking sideways, up to "standard maximum look angle of approximately 40 degrees off-nadir" either toward the sun or away from the sun
    • although we can expect that looking away from the sun must be prefered
see 4SM at Waimanalo beach, Oahu 
I want to demonstrate 4SM
on some of the above images,
for free, expenses paid.

I want to demonstrate 4SM
on USAF's Tactical-satellite-3    
and ORS-1 imageries


  • 4SM: how about with dispensing with atmospheric corrections?
  • 4SM: how about dispensing with pre-existing depth points?
  • Quite frankly, I think NPS might want to pay some attention to the 4SM approach for their consistently down-to-earth approach
  • Quick turnaround is guaranteed
  • No field data needed
Miller sept 2012

Methods Comparison  
a.  Silva and Abileah 
b.  Mustard et al. 
c.  Hochberg et al.   
d.  Hedley et al.   
e.  Lyzenga et al. 
f.  Goodman et al. 
g.  Kuster et al.  


Créer un site
Créer un site