Bathymetry and water column corection
World View 2 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
Investigation of Wavelengths and Tide Height, etc

home
 
 
 
 

Seatruth
 

image wZ


 

The "RED solution" from 0 to ~5 m: wZ=5
Get things straight when modeling by Red against Purple, Blue, Green and Yellow: this is wZ=5
The goal is to get a nice diagonal fit for the 0-5 m scatter of depth points                   

Smoothing is OFF
RMSE=0.66 m (0-5m, N=18311)

1 in 25 DTM grid points are investigated
wZ=5 for 18311 of those points without smoothing

  • Deglinting is ON
  • Smoothing is OFF
  • WLred   is at mid-waveband
  • Lswred   is increased from 19.7 to 20.5
  • Lmred   stays at 2
  • Tide height is set to 0.6 m, and tide correction is applied to Zcomputed

Smoothing is ON
RMSE=0.49 m (0-5m, N=18024)

1 in 25 DTM grid points are investigated
wZ=5 for 18024 of those points with smoothing

  • Deglinting is ON
  • Smoothing is ON
  • WLred   is at mid-waveband
  • Lswred   is increased from 19.7 to 20.5
  • Lmred   stays at 2
  • Tide height is set to 0.6 m, and tide correction is applied to Zcomputed
  • A small number of scattered depth points report Zcomputed depths less than 3 m where Zrecorded depths are in excess of 4 m
  • this is commonly caused by locally turbid waters or cloud shadow
  
 






The "YELLOW solution" from 0 to ~9 m: wZ=4
Get things straight when modeling by Yellow against Purple, Blue, and Green: this is wZ=4
                  The goal is to get a nice diagonal fit for the 0-9 m scatter of depth points                                                        

Smoothing is OFF
RMSE=1.00m (0-9m, N=43062)

1 in 25 DTM grid points are investigated
wZ=4 for 43062 of those points without smoothing

  • Deglinting is ON
  • Smoothing is OFF
  • WLyellow   is at 603.6 nm
  • Lswyellow   is increased from 28.0 to 29.2
  • Lmyellow   stays at 1.5
  • Tide height is set to 0.6 m, and tide correction is applied to Zcomputed

Smoothing is ON
RMSE=0.85m (0-9m, N=42890)

1 in 25 DTM grid points are investigated
wZ=4 for 42890 of those points with smoothing

  • Deglinting is ON
  • Smoothing is OFF
  • WLyellow   is at 603.6 nm
  • Lswyellow   is increased from 28.0 to 29.2
  • Lmyellow   stays at 1.5
  • Tide height is set to 0.6 m, and tide correction is applied to Zcomputed






The "GREEN solution" from 0 to ~50 m: wZ=3
Get things straight when modeling by Green against Purple and Blue: this is wZ=3
                     The goal is to get a nice diagonal fit for the 0-50 m scatter of depth points                                                        

Smoothing is OFF
RMSE=2.24m (N=346299, 0-50m)

1 in 25 DTM grid points are investigated
wZ=3 for 346299 of those points without smoothing

  • Deglinting is ON
  • Smoothing is OFF
  • WLgreen   is at 550.9 nm
  • Lswgreen  stays at 44.9
  • Lmgreen   stays at 0.5 
  • Tide height is set to 0.6 m, and tide correction is applied to Zcomputed

Smoothing is ON
RMSE=1.80m (N=345339, 0-50m)

1 in 25 DTM grid points are investigated
wZ=4 for 345339 of those points with smoothing

  • Deglinting is ON
  • Smoothing is OFF
  • WLgreen   is at 550.9 nm
  • Lswgreen  stays at 44.9
  • Lmgreen   stays at 0.5
  • Tide height is set to 0.6 m, and tide correction is applied to Zcomputed
 
coral heads
  • coral heads are deglinted and smart-smoothed correctly
  • coral heads are so dark that yellow and red bands are extinct: they are modeled by the 3 bands case: wZ=3
    • their computed depth is commonly overestimated by several meters
    • this shows in the above regression plot in the 4-10 m depth range
  • I can't think of a way to counter that in 4SM



The "PAN solution"
unfortunately, I have not been given the PAN band





Profile RED

4SM Blind test
for a reference


Blind test on profile_red: reference
Red profile is LIDAR seatruth depth
Black profile is Zcomputed by 4SM

4SM Final
is much improved

Profile_red: final
Red profile is LIDAR seatruth depth
Black profile is Zcomputed by 4SM

Reduce RMSE
The challenge is:
how to get to final
without the use of any seatuth data

Reduce RMSE using 4SM
As confirmed by this seatruth exercise,
the following should help

  • WL (this is outdated: see Hyperion at Fakarava)
    • increase WLgreen away from mid-waveband
    • decrease WLyellow away from mid-waveband
    • keep  WLred at mid-waveband
  • Lsw
    • check on screen for radiance at extinction of shallow bottom
    • tune according to depth results
  • Turbid
    • watch for extremely dark watercolumn corrected reflectances: they are likely to signal that Lsw should be increased, and mask out areas with locally turbid waters altogether
  • Deglint
    • apply deglinting, mask out non-deglinted bright-cloud pixels if at all possible
  • Aliens
    • mask out cloud/shadows manually
    • mask out all alien pixels manually
  • Smooth
    • apply smart-smoothing
 
 



Créer un site
Créer un site