|
Image preparation
| 8924x18001
0.5x0.6 m footprint - 4SM accomodates this correctly
| | Mask I imported your following masks into channel_5 of my working database shirahoQBPS_20070702.pix - cloud-shadow_mask
- land_mask
- water_mask
Green tone is to exclude non-lagoon areas from optical calibration | | Zero - The NIR band has many pixels at ZERO:
- this can't be: path radiance over cloud shadow areas is ~5
- I recoded them to 1, so that they not map as no-data pixels
| | Dry reef - For sake of lagoon modeling, reef crest and fringing reef areas
- which are NOT covered by water here at low tide,
- are forced into water mask
- and shall be assigned a retrieved depth of 10 cm
- so that they shall be included for bottom typing
| PAN and NIR bands - the PAN band is not available in this dataset: too bad!
- the NIR band is available: I seem to rememcer that pan-sharpening did not allow that?
| Deglinting of this PAN-sharpenned image is most problematic Effective wavelengths  | - This has been a recurrent aspect of the 4SM practice over the years
- From looking at the QB response curve, choosing an effective wavelength for each MULTI band is not straightforward, as response curves are distinctly peaked away from mid-wavebands.
- Wavelengths at mid-waveband do not allow for a comfortable calibration, and most importantly yield retrieved depths that appear to be distinctly deeper than depths retrieved from Landsat 8 OLIP image of Shiraho Reef.
- Wavelengths at peak response make me feel a lot better, and quite importantly yield retrieved depths that appear to be compatible with depths retrieved from Landsat 8 OLIP image of Shiraho Reef.
| Operational wavelengths are set to
WLblue=490.0 nm WLgreen=550.0 nm WLred=615.0 nm | Optical calibration The scene is quite small : 4.5x11 km, with a narrow lagoon-reef system - The calibration data only sample the reef moat and crest: that's the 0-5 m depth range in Blue , as the GREEN areas are excluded.
- deeper than that, over the outer reef slope, deglinting would be absolutely necessary, but is impractical.
- This is the first opportunity for me to calibrate a pan-sharpened BGRN image
- Gram-Schmidt sharpening transformation, and cubic convolution resampling, by Collin et al, to parallel GE derived image
- this image is very noisy
- not being able to deglint is... questionable
- estimation of path radiance, deep water reflectance and water volume reflectance is... questionable
- Nonetheless, the calibration diagrams below are fairly straightforward to interpret
- operational wavelengths are set to WLblue=490.0 nm WLgreen=550.0 nm WLred=615.0 nm
- Kblue/Kgreen=0.75 appears to fit the data over the 0-5 m depth range
- this a water type OII of Jerlov, with 2Kblue=0.133 m-1 2Kgreen=0.178 m-1 2Kred=0.610 m-1
| 
Calibration diagram for bands 1, 2, 3 and 4 | Profiling Smart smoothing hardly improves the result Plain smoothing would negate the interest of pan-sharpening altogether!!! 
Modeling along profile_red
NO deglinting applied
NO smoothing applied - this plot shows retrieved depth and water column corrected reflectances for the Blue, Green and Red bands
| 
Modeling along profile_red
NO deglinting applied
NO smoothing applied - this plot shows
- in red: depths retrieved by the 3 bands case, i.e. Blue and Green against Red
- in black: depths retrieved by the 2 bands case, i.e. Blue against Green
| Modeling NO deglinting applied NO smoothing applied After all, the value of this exercise is to test the interest of PAN-sharpening water column correction and for bottom typing: As long as the calibration diagram shows that things are in order for the main part, water column correction and for bottom typing should be satisfactory, and retrieved depths might need correction using suitable field depths.
Retrieved depth Zoom | 
Normalized water column corrected TCC | 
TCC Zoom | 
Normalized water column corrected TCC Zoom | |
