a 7-bands 9318*8807 pixels WorldView 2 image Dec 10th 2010

ground resolution 2 m
"(C) COPYRIGHT 2011 DigitalGlobe, Inc., Longmont CO USA 80503.  DigitalGlobe and the DigitalGlobe logos are trademarks of DigitalGlobe, Inc.  The use and/or dissemination of this data and/or of any product in any way derived there from are restricted. Unauthorized use and/or dissemination is prohibited."
The data
Calibration and results SeaTruth, march 3rd 2011
work done February 7th 2011 and March 3rd 2011

SeaTruth, march 3rd 2011
New seatruth
New seatruth under mask 3

New seatruth under mask 3
  • The previous seatruth is biased
    • Lsw: deep water radiance in the Yellow band was overestimated: I had to reduce it by 0.7 DN: from 36.0 to 35.3
    • MaxDepth threshold: imposing a MaxDepth threshold forced even more bias: this feature is disabled in 4SM from now on
    • Biased DTM dataset: it not safe to test the modeling results using a field dataset that (1) lacks a fair representation of the 0-4 m depth range, and (2) has an uneven representation in the 4-6 m depth range:
      • the regression line ZC=offset+slope*ZR can go astray
      • using it to set HTide=-offset and CoefZ=1/slope is risky, unless the seatruth dataset ensures a fair representation of over the whole depth range
    • the practioner must exercise caution
    • ensure that the seatruth dataset is worthy
    • and resort to common sense when inspecting the results when specifying HTide and CoefZ
  • As it turns out, setting HTide=1.5 m and slope=0.96 through visual inspection is just as good: this is shown below
Last seatruth
Last seatruth under mask 3

Last seatruth under mask 3
<<==  HTide=1.5 m as prescribed
<<==  coefZ=0.96 for good visual fit
<<==  and forget about the regression line
<<== and forget about wavelengths
  • It only takes a small amount of bias for the offset to be mis-estimated:
    • first we seemed to get 1.0 m
    • and then we seemed to get 1.78 m...
  • The height of tide over LAT is reported to be 1.5 m
    • So it seems reasonable at this stage to assume 1.5 m,
    • and set CoefZ=0.96 so as to place the cloud of points in a suitably diagonal position (a very small correction, hardly needed actually)
    • and forget about the regression line
  • I need to harden the regression process!
  • One way around this problem might be an optimization scheme which would aim at
    • maximizing the R2 figure
    • minimizing the RMSE figure.
  • In this respect, the display of  both ZR and ZC along a depth sounding profile provides an alternate way to appreciate the offset and slope values that result in an optimized fit of ZC with ZR: I need to implement that feature for the particular case of a DTM

<<==This plot maps the difference
DZ  =  ZC*0.96   -   (ZR+150)
in centimeters

Brightest areas are
within the DZ range  -50 to 50 cm,
with R2=0.92 and RMSE~=30 cm
  • R2 has improved markedly from 0.74 to 0.92
  • RMSE has increased, due to many more points modeled following the decrease of Lswyellow

nice -20 4SM.4.03 -Process/producer/02_Mar_2011 -OriginNPA       
-KK2_4_0.3800/Knir4.500/Bm400_BG520_GR585_RN700 /mask_3           
-Model/mask_2        @-E4869/8391/1/3249    @-E4869/8391/1000/3249       

TCC of
water column corrected bottom reflectances

There is an excellent potential
for bottom typing
See on ftp site:
It contains
BahrainDZ.tif    ZC*0.96 - ZR+150   in cm
BahrainZC.tif    ZComputed          in cm
BahrainZR.tif    your ZRecorded DTM
foul.shp    outer foul waters
m2.shp        BLUE polygons
m3.shp        RED polygon
import.shp    Area imported from your mosaic
As it turns out,
  • CoefZ=0.96 is peanuts, actually hardly worth mentioning
  • as we lack depth points in the 1-4 m depth range, we can't really conclude
    • other than by saying "Could be that 4SM is really "self-calibrated" when using WV2 data, as it proved to be when using SPOT data, when using IKONOS data, and then when using  CASI data"
    • assessing this shall require more reliable WV2 images and seatruth datasets
All this has been a lot of work. March 3rd 2011
That was my first WV2 image.
A hard case at that!

I had to chase some loopholes in 4SM
It's in the box now.

Créer un site
Créer un site