WorldView 2 image Florida West Coast
December 29th 2011, 8875*14996, 2 m ground resolution
MULTI + PANchromatic
bandset 1, 2, 3, PAN, 4, 5, 6, 7, 8     becomes      1, 2, 3, 4, 5, 6, 7, 8, 9
This dataset from Digital Globe, under a "Data Exchange Agreement"
  • This is breaking new grounds
  • This is my 5th WV2 image, quite a difficult one at that
  • Princes Cays is the only staightforward image I processed so far
  • I'm learning and testing so much: quite fascinating
  • How do I estimte deep water Lsw radiance when there is no optically deep waters in the image?
  • Accounting for the water becoming much clearer outwards is not quite simple in 4SM!
  • Only through time and a lot of seatruth shall WV2 images become routine; but FIRST: keep away from hevily glinted images or adverse imaging geometries
  • Until then, I need seatruth to tell good from bad 4SM practises
work done oct 30th 2012

Beautiful image: no glint !
Very low noise: no need for any smoothing!!!!


  • This image is clearly OFF LIMITS for operating of the simplified radiative transfer model, which assumes homogeneous waters: KEEP OUT or FACE THE CONSEQUENCES

NIR1 at 831 nm
  • Note the absence of suspended particle turbidity
  • Note the banding
  • Banding in RED and NIR1 bands is 1024 pixels wide
  • Therefore, deglinting based on NIR1 band shall transfer this noise to Red, Green aand Blue bands: this is BAD
    • I insist that, for shallow water work, such banding ought to be corrected for by DG at level 1 pre-processing prior to geocorrection
  • Banding does not  seem to affect 908, 724, 595 and 427 nm bands

Red band

Yellow band

  Green band

  Blue band

  Purple band


Oct 30th 2012

Purple and PAN channels are disabled: not used for now
more on that later on

Z in dm

Z in cm
Water column corrected
RUN of 4SM.4.10 started on Sun Oct 28 22:38:01 2012
nice -20  4SM -Process    -Origin/DigitalGlobe                               
-B/TclNe/cLM1 0.7                                           
-Model/v/runWV2/mask_3    -LL1                  

Purple and Panchro bands are disabled
At 4-5 m, coastal water type C5
appears to yield to type C1 and clearer

Quite a complex situation here!

Waters optical structure
Watertype Z in m Ki/Kj 2K1 2K2 2K3 K4 2K5 2K6 2K7 2K8 2K9
C1+0.74 0 1.388 0.896 0.481 0.346 - 0.486 0.668 3.5 4.49 8.9
C1+0.74 4.5 1.388 0.896 0.481 0.346 - 0.486 0.668 3.5 4.49 8.9
O2+0.08 10 0.749 0.173 0.134 0.179 - 0.474 0.668 3.5 4.49 8.9
O2+0.08 12.02 0.749 0.172 0.134 0.179 - 0.473 0.668 3.5 4.49 8.9
O2+0.08 32.75 0.749 0.172 0.134 0.179 - 0.473 0.668 3.5 4.49 8.9
K_JERLOV - - 0.173 0.134 0.179 - 0.477 0.76 - - -
  • Water type along the coastline is seen to be Coastal type 2
  • Deeper than ~4.5 m, it is seen to yield to Oceanic type II
  • Sounds familiar??
Band 1 2 3 4 5 6 7 8 9
WL 427 477.5 546 570 595.8 641 724 831 908
16U_LsM 407 534 870 18339 754 535 1290 1250 1473
16U_Lsw 271 244 192 2698 110 57 75 30 25
16U_La 245 222 188 2698 110 57 75 30 25
16U_Lw 426 22  4 0 0 0 0 0 0

16 bits values for radiance parametrs used
I'd be interested to compare with yours for Lsw, La and Lw


seatruth as of NOV 8th 2012
Gregory's seatruth Yann's seatruth
NO smoothing applied at all
NO deglinting
PANchro band not used at all
PURPLE band not used at all
Gregory said: "Attached are comparison files between my and your results. "
  • I'm impressed by both 4SM's accuracy and precision figures
  • I expect to qualify for a copy of your FloridaWestCoast seatuth DTM.
  • I need to understand why my 0-6 m computed depths are ~+1 m too deep: quite distubing.
  • Unless that's the ~TideHeight over depth datum?
    • FinalZ=CoefZ*ComputedZ -TideHeight
  • If TideHeight~=1 m, then I need to see how to increase the 6-12 m depth range by ~1 m:
    • easy: by making it slightly greener in the calibraion diagram: call that "tweaking" the Soil Line!
Gregory said" "Also, I do not have any sun-glint correction."
I did not operate any deglinting either, as there is no glint in FloridaWestCoast image
Gregory said: "My results are more noisy than yours since I did not implement any smoothing filter as yet."
 I did not enable any smoothing either, as the FloridaWestCoast MUTLI data is really low noise: not needed, and also, I wanted our results to be comparable.
  • I'm not surprised though, as my experience tends to show that using "endmembers spectral signature" entails quite a reduction of nature's diversity and of phenologic statuses of the living thing down below.
  • then, when applying a set of endmember signatures, there is a risk the  algorithm shall pick up the first match, starting a Z=0 towards MaxDepth, although there might be a better choice at deeper depth: would seem to be the case in your regression: how about starting at MaxDepth towards Zero

Comments on WV2 images
Exceptional radiometric quality
  • PrincessCays and FloridaWestCoast are of outstanding radiometric quality.
  • Shallow water work obviously shall demand top radiometric quality, and keep away from heavily glinted images or face the consequences.
  • This needs to be investigated and exposed in full before any effort by data vendors to advertise WV2 for shallow water work.
  • Still, I am concerned that deglinting transfers system noise from NIR bands 7 and 8 to other bands, in particular the banding

Inter-bands correlations
  • There is a time gap between acquisition of bands 2-3-5-7 and  the other four bands.
  • (There may even be yet another time gap between PAN and MULTI?)
  • Then I must ask: what is the consequence of time gaps on the inter-band correlation, which is at the heart of any ratio method for bathymetry? Sure enough, it can't help!
see waimanalowv4.php


Créer un site
Créer un site