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, feb 11th 2011
DTM datum
  • Seatruth dataset is a digital terrain model (DTM), where depths in centimeters are tide-corrected to Lowest Astronomic Tide (LAT) datum

Using the DTM: decimation
Because there is such an abundance
of seatruth depth points
4SM may need to decimate
  • Increment=1: 4SM first tries to use all depth points available, eventually subject to a mask condition
  • Increment=2: if the above produces too many points which overflow the allocated size of array variables, 4SM then decimates 1 in 2 rows, 1 in 2 lines
  • Increment=3...4SM then decimates 1 in 3 rows, 1 in 3 lines
  • and so on...
  • increment=7
    • see All pixels with both ZR and ZC: N=6722
  • and so on...
Above is the raw regression
  • Tide correction must then be estimated and applied
  • CoefZ correction must then be estimated and applied

  • Htide=100 cm: Z_Recorded=Z_Recorded+Htide
    • this brings the DTM depths to local tide datum at the time of imaging
    • local Htide is estimated at 100 cm  - instead of reported 150 cm over LAT datum at the time of imaging
    • its application brings the offset of the regression to 0
  • CoefZ=0.80: Z_Computed=CoefZ*Z_Computed
    • a value of CoefZ=0.80 is observed, subject to wavelengths adopted here at mid-wavebands
    • its application brings the slope of the regression to 1.0
    • see Plot under RED_mask=3
  • Wavelengths: the above was obtained using wavelengths at mi-wavebands.
  • Operational wavelengths: see in the table below that 2K values are increased by a factor of 0.84 (quite close to observed 0.80!) by setting wavelengths at more appropriate values while seemingly maintaining acceptable physical consistency with both Jerlov's data and WV2 response curves
    • Using this new set of wavelengths, the slope of the regression would be very close to 1
    • More on this aspect through more seatruth using more wideband imageries
Regression line
  • a good regression assumes an even representation of depth points over the whole depth range
    • like in Plot under RED_mask=3: Slope=1
  • an uneven representation of depth points over a limited depth range may result in an outrageous regression
    • like in Plot under BLUE_mask=2

Plot under RED_mask=3

Away from live dredging sites
Plot under RED_mask=3
  • Using the following
    • Htide=100 cm
    • CoefZ=0.80
  • This plot yields
    • slope=1.00
    • offset=0
    • R2=0.74
    • RMSE=0.26 m
This should be the WV2 performance
over all areas worthy of shallow water modeling
in this OIII water type
i.e. well away of areas
disturbed by ongoing dredging activities
The DTM lacks depth points
in the 0-3 m depth range
Plot under BLUE_mask=2

Plot under BLUE_mask=2
  • Part of the points are consistent with the previous regression
  • A nasty concentration of points at ~3.5 m causes the regression to go astray
  • The DTM lacks depth points in the 0-3 m depth range

General plot
of all pixels with both ZR and ZC

General plot
of all pixels with both ZR and ZC

This plot maps the difference
DZ  =  ZC*CoefZ   -   (ZR+Htide)
in centimeters
Brightest areas are
within the DZ range  -100 to 100 cm
with R2=0.74 and RMSE~=26 cm

Foul waters
yield foul results

Conclusions on WV2
  • the "NIR 2" band at 908 nm shows poor correlation of the glint signal with other bands: I could not use it
  • the "NIR 1" band at 831 nm is extremely usefull
    • for segmenting land from water
    • for deglinting
    • for delineating the Soils Line
  • the "RedEdge" band at 724 nm is usefull for detailing very shallow areas in the 0-2 m depth range
  • the "Yellow" band #4 at 608 nm saved our day with bottom detection down to 7.8 m over bright bottoms in these waters
    • as 2K values in the blue-green range are too close to allow water column correction
    • and bottom detection in the red band at 659 nm is only 5.8 m
  • the "Coastal" band #1 at 427 nm is of good radiometric quality, certainly quite usefull in clearer waters
  • The radiometric quality at a 2 m ground resolution is outstanding: most appreciated for shallow water work as it allows
  • deeper bottom detection because it allows  lower threshold Lm values
  • more reliable results over very dark areas if/where negative bottom contrast is rife
  • Foul waters: this being said, no improvement of the image quality can counter foul waters, whatever the cause
  • PANCHRO: I wish I could test the use of a 2 m decimated co-registered PANCHRO band, along with the above multispectral bands
Conclusions on modeling
  • OIII water type, and Coastal types beyond, are a difficult case, because there is hardly any color separation in the blue-green range of bands
    • this means that water column correction may only be operated as long as the Yellow band exhibits bottom detection, i.e. ~ 7.3 m over very bright bottoms and much less over dark bottoms (and quite less if the Yellow band is not available)
    • even if the waters were not marred by dredging activities, it would have been impossible to model deeper areas even though all 3 bands in the blue-green range exhibit bottom detection down to ~15 m
    • In this regard, an additional band ca 590 nm in between the Green and the Yellow bands would be most profitable
  • 11 bits data: the S/N ratio of this data is very good indeed. This allows the practioner to lower the threshold values, i.e. to increase the bottom penetration in a quite significant way
  • RMSE=0.26 m in Plot under RED_mask=3
    • increasing steadily with depth
    • 0.26 m on average: this is very good for the 3-6 m depth range (certainly much better than with an ETM image)
    • would clearly be even lower over a 0-6 m depth range
  • Htide=1 m  instead of 1.5 m under Red_mask: this is annoying
    • does the 1.5 m value refer precisely to Port of Bahrain?
  • Lsw: LUCKY I could find an area that seems free of suspended sediment where to measure a usable set of deep water radiances Lsw
  • Glint: very high ground resolution data usually exhibit sufficient sea-surface glint effects, even over very calm waters, to allow for
    • deglinting
    • setting Lsw

See on ftp site:
It contains
BahrainDZ.tif    ZC*0.80 - ZR+100   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

Créer un site
Créer un site