WV2 at Saint Croix, US Virgin Islands
8967*8805, MULTI, UTM zone 8

LIDAR seatruth of 4SM computed depths at Buck Island Reef, US Virgin Islands

home
 

 
 
 



Seatruth using LIDAR DTM
Tweaking deep water radiance
Results
Seatruth regressions
==>next to Investigation of WL and TideHeight


LIDAR seatruth DTM buis_bathy_3m.zip 
At last I was able to download and read buis_bathy_3m.zip 
from http://coastalscience.noaa.gov/datasets/ccma/biogeo/usvi_nps/2011/lidar/buis_bathy_3m.zip

I had to resample it from 3 m to 2 m resolution,
so that I can use it for seatruth of WV2 MULTI data

Buck Island Reef LIDAR DTM BUIS_Bathy_3m
resampled to 2 m ground resolution

Blind test

  • Deep water radiances are underestimated in the Blue-green range
    • this is evidenced by badly underestimated depths where Red and Yellow bands are extinct





dLsw
Increase deep water radiances in the blue-green range
this is most sensitive for the green band where modeling by green against purple and blue
 

Deep water radiances Lsw
has been increased for the Green band,
from 44.2 to 44.9
Deep water radiances Lsw
has been increased for the Green band, 
from 44.2 to 44.9
  • the improvement is quite obvious
    • it affects darkerpixels, whether darker bottoms or deeper bottoms or both
    • refer to the previous blind profile
  • in section B, deepest estimated depths are at ~35 m, over fairly dark bottoms
The problem is:
  • considering the striping which affects WV2 images
  • considering the many whitecaps at sea sueface
  • considering the natural variations of water optical quality across the scene
  • where and how to take a measurement of deep water radiances?
 The answer is:
  • Buck Island Reef belongs to the coastal water body North of St Croix
  • This water body backscatters more radiance in the blue-green range than open waters south or upper-north of St Croix
  • Even a seasonned practioner is likely to overlook this step at times
  • I should have known better
  • I should not have commited this error
  • Even an outdated and simplistic nautical chart is enough to raise alarm
 -Lsw125.1/083.7/045.1/028.1/020.2/015.2/008.5/007.7
-Lsw125.1/083.5/045.0/028.0/020.1/015.1/008.4/007.7
-Lsw125.0/083.8/045.1/028.2/020.1/015.2/008.5/007.8
-Lsw124.9/083.6/044.9/028.0/020.0/015.1/008.4/007.6
-Lsw124.9/083.7/044.9/027.9/019.9/015.0/008.3/007.5
-Lsw124.9/083.6/044.9/027.9/019.9/015.0/008.3/007.6
-Lsw125.0/083.8/044.9/027.9/019.9/015.0/008.3/007.5
-Lsw125.1/083.8/044.9/027.8/019.9/014.9/008.3/007.5
-Lsw124.9/083.6/044.8/027.7/019.7/014.9/008.3/007.5
-Lsw124.9/083.6/044.7/027.8/019.8/014.9/008.2/007.5
smoothed readings at ten deep water locations
==> 44.9 for band 3 is a reasonable average 
 


So, North of St Croix should be processed using new Lsw=44.9 for the Green band



Results

Computed depth in centimeters
4SM results as a backdrop, Lsw=44.9 for the green band
BUIS_Bathy_3m is overlayed
 
Computed depth in centimeters
Lsw=44.9 for the green band





Seatruth regressions

Difference Zcomputed - Zseatruth
RED   : ZC > ZR
BLUE : ZC < ZR



So it would appear
that a majority of estimated depths
are underestimated
ZZRegressor: Statistics of seatruth ZC-ZR
on image buckislandreefwv at BuckIslandReef, USVI
N=8.661 millions pixels  by 4SM method
HTide=0.00 Smooth=1_5 Using_bands_1_2_3_4_5 cZ=1.00
 
BLUE : ZC < ZR
 2.68% depths underestimated by more than 5.0 m
 2.47% depths underestimated by more than 3.0 m
 9.63% depths underestimated by more than 2.0 m
30.73% depths underestimated by more than 1.0 m
38.24% depths underestimated by more than 0.0 m

RED   : ZC > ZR
13.87% depths  overestimated by less than 1.0 m
 1.78% depths  overestimated by less than 2.0 m
 0.36% depths  overestimated by less than 3.0 m
 0.20% depths  overestimated by less than 5.0 m
 0.04% depths  overestimated by more than 5.0 m
 
52.11% depths are within  +-1.0 m of DTM depth
84.62% depths are within  +-2.0 m of DTM depth
94.61% depths are within  +-3.0 m of DTM depth
97.28% depths are within  +-5.0 m of DTM depth
over a total of 100.00% of computed depths
 

Seatruth regression
 


Seatruth regression
One in 25 pixel is plotted
Pixels with ZR or ZC > 30 are excluded
No tide correction has been applied

 

  • from 0 to ~8 m, modeling is based on the Red or the Yellow band
    • the slope of the scatterplot is distinctly less than 1
    • this might indicate that effective wavelengths for the Red and the Green bands should be decreased: I shall try that
  • deeper than ~8 m, modeling is based on Green band against Purple and Blue bands
    • the slope of the scatterplot is ~1.0
      • this indicates that effective wavelengths adopted for those three bands are correct: NoNeedForFieldData!
    • a majority of estimated depths are underestimated
      • this might indicate that the slope of the Soil Line should be decreased a bit: I shall try that
  • From 20 to 35 m, a bunch of computed depths stagnate at ~18 m
    • this can only be caused by the presence of locally more turbid waters; see below
 
But wait till I use WV2's  PAN band,
as things shall look even much better.
 

Average botttom brightness
BUIS_Bathy_3m is overlayed
  • Increasing Lsw to 44.9 for the Green band has cleared the area north of BUIS
  • While the area west of BUIS would require an even higher value
    • this is the cause of "the bunch of computed depths" which stagnate at ~18 m in the regression plot

This was already obvious in the blind test:
see those violet "very dark bottoms" 
with computed depths in the 16-18 m range
in image B of blind test 
I should have known better
or was I pressing too fast?
 



Créer un site
Créer un site