- Sam PURKIS, 21 january 2009, after viewing my report on work done in 2005: "Hi Yann, That is a fantastic presentation. Thank you. Attached are bathymetry soundings for the Dubai image. I'd be really interested to see how the seatruth fits to your prediction."
 | <== My answer: this is the first seatruth regression I got , using your depth sounding dataset and my 2005 calibration. - Well, this raises mixed feelings.
- Wavelengths were -WL480/583/665/805
- while mid-waveband is -WL480/551/665/805
- Depths were over-estimated in 2005
- seems WL[green]=583 nm is still too low
- R2=0.82 is not so bad
- RMS=1.85 m is not good!
- Depths over very dark bottoms are badly underestimated
- Many points should be thresholded out
- believe it or not: Lm was -Lm000001/00.1/01.8/0255 !!!!!
4SM.2.06 -PROCESS/31_JAN_2009/USF
-dB/PCI/dubaiIKYM/27_1_4_0/4_29/3294_3012/286947.479_2767543.207/1_1
-Origin/USF_dubai_4b.pix
-MIS/UAE/DUBAI/Ikonos/NA/USF/UTM_40_008/4_4/29_JUL_2001
-WL0480/00583/00665/0805
-M@000001/0002/0003/0004
-Lsw078.7/46.5/22.2/16.1
-La0074.8/45.4/22.2/16.1
-LsM00200/0200/0200/0200
-Lm000001/00.1/01.8/0255
-LsT00001/0001/0001/0001
-LS000500/0500/0500/0500
-SCL00137/0195/0198/0180
-KK/1_2_0.760/7.0/500_600_700
-Z+/1.18/.1/0.0/0.0/1.0/PAN_1_3/SIG_3w5/N_1/1.0000
-B/tclNE_3.0/200_100/1/1.00 -LL10
-deglintRBa/0.5/1,4_70/FN/1_22
-extractv/U16/4Max255/mBPL_1/mSOIL_21/mSHAL
-calibratev/BDSMpZ/4_3_2_1/bDH_12_23_34
-Model/mask_0 -Smooth_3d/Smart+
-RegressZZv/UTM-XYZ-sparse_GPS/0.001_0_0/1_0/ZM/2_3/0_201/0_30/* |
| Even more disturbing Regression on the 5460 depth points that are modeled using all three Blue, Green and Red bands
I seem to have a problem with the Soils Line.
4SM.2.06 -PROCESS/31_JAN_2009/USF
-dB/PCI/dubaiIKYM/27_1_4_0/4_29/3294_3012/286947.479_2767543.207/1_1
-Origin/USF_dubai_4b.pix
-MIS/UAE/DUBAI/Ikonos/NA/USF/UTM_40_008/4_4/29_JUL_2001
-WL0480/00583/00665/0805
-M@000001/0002/0003/0004
-Lsw078.7/46.5/22.2/16.1
-La0074.8/45.4/22.2/16.1
-LsM00200/0200/0200/0200
-Lm000001/00.1/01.8/0255
-LsT00001/0001/0001/0001
-LS000500/0500/0500/0500
-SCL00137/0195/0198/0180
-KK/1_2_0.760/7.0/500_600_700
-Z+/1.18/.1/0.0/0.0/1.0/PAN_1_3/SIG_3w5/N_1/1.0000
-B/tclNE_3.0/200_100/1/1.00 -LL10
-deglintRBa/0.5/1,4_70/FN/1_22
-extractv/U16/4Max255/mBPL_1/mSOIL_21/mSHAL
-calibratev/BDSMpZ/4_3_2_1/bDH_12_23_34
-Model/mask_0 -Smooth_3d/Smart+
-RegressZZv/UTM-XYZ-sparse_GPS/0.001_0_0/1_0/ZM/3_3/0_201/0_30/* ==> I've got work to do if ever this kind of atmospheric conditions becomes worthy of shallow water modeling. |
| | |
| ==> Below is the commandline : nice -20 4SM.2.06 -PROCESS/06_JAN_2009/USF
-dB/PCI/dubai/31_1_4_0/4_33/3294_3012/286947.479_2767543.207/1_1
-Origin/USF_dubai_4b.pix
-MIS/UAE/DUBAI/Ikonos/NA/USF/UTM_40_008/4_4/29_JUL_2001
-SCL00137/00195/00198/00180
-DeglintRBa/0.5/4_70/FN/1_22
-Smooth_3D/Smart+ - GlintMin is automatically set to the lowest value encountered in the NIR band under mask mSE=22
- this is GlintM[4]=15.7
- this is a good candidate for Lsw[4]
- GlintMax is automatically set to the highest value encountered in the NIR band under mask mSE=22.?
- this is GlintM[4]=23.3
- it can be seen that this value should be increased for manual deglinting
- ??the -Smooth... argument is usually not required for deglinting
- but here, without a -Smooth... argument, there is no correlation between the Blue band and other bands
- this shows how risky this activity can be, in case there would be negative bottom contrast over very dark bottoms in this environment
|  automatic "deglinting" works well
a Smart_Smoothing has been applied |
| | |
| Standard deglinting below is the commandline nice -20 4SM.2.06 -PROCESS/06_JAN_2009/USF
-dB/PCI/dubai/31_1_4_0/4_33/3294_3012/286947.479_2767543.207/1_1
-Origin/USF_dubai_4b.pix
-MIS/UAE/DUBAI/Ikonos/NA/USF/UTM_40_008/4_4/29_JUL_2001
-SCL00137/00195/00198/00180
-Lsw/79.8/47.1/22.0/15.7
-DeglintRBa/0.5/4_70/FN/1_22
-Smooth_3D/Smart+ | Standard deglinting
All you need is to
- add one -Lsw... argument taken from textfile dubai_m0_22.gli
- set your estimate for GlintMax[4]: here at 70
- specigy a -Smooth... argument if desired
|
| 
Raw color composite of BGR bands
| "deglinting"
(applied only in the central part of the image here)
removes the adjacency effect
.
A Smart_Smoothing has been applied. |
| Glint regression:
band_4 vs other bands
Regression_SG 1 4 SGslope 0.4696 SGbias 72.4578 SGr2 0.4142
Regression_SG 2 4 SGslope 1.1211 SGbias 29.5169 SGr2 0.8191
Regression_SG 3 4 SGslope 0.9683 SGbias 6.7748 SGr2 0.9572
Regression_SG 4 4 SGslope 1.0000 SGbias 0.0000 SGr2 1.0000 Glint is sampled over darkest patches==> 
| Glint regression:
band_2 vs other bands Regression_SG 1 2 SGslope 0.4964 SGbias 56.1141 SGr2 0.7101
Regression_SG 2 2 SGslope 1.0000 SGbias 0.0000 SGr2 1.0000
Regression_SG 3 2 SGslope 0.7286 SGbias -11.7788 SGr2 0.8315
Regression_SG 4 2 SGslope 0.7306 SGbias -18.0408 SGr2 0.8191 ==>masked at 22 in red in mask mSE 
|
