Profile RED | - RED profile is "seatruth" depth retrieved by U. of Queensland
- a tide height of 0.30 m was added
- BLACK profile is 4SM retrieved depth
|
Not using the PAN band Not using the PAN band: - all shallow pixels are modeled by band_j against all_seven_band_1_to_band_7
- see that , as depth increases , band j shifts
- from band_15 (brown)
- to band_14 (orange)
- to band_13 (red)
- ...
- through to band_7 (green)
- see that , deeper than ~17 m, "seatruth" depths are extremely noisy and confined to ~18 m:
- this is an artifact of the ALUT method: Hedley et al write: "Hence the maximum retrievable
depth is 20 m" | Using the PAN band  Using the PAN band: - all shallow pixels are modeled by band_PAN against all_seven_band_1_to_band_7
- see that , as depth increases , band j remains the PAN band 16 (light grey)
|
SEATRUTH - A tide height of 0.30 m was added to the "seatruth" depths, based on this kind if profile
- This is the only "use" of the "seatruth" dataset
Hedley et al used tide corrected recorded depth points for control of their model: "with tide-corrected sonar depth transects used for ground-truth" | SEATRUTH - The depth profiles are virtually identical
- this is most remarkable, as it tends to validate both 4SM and ALUT methods
- this indicates that the calibration of the PAN band is properly mastered
- But, in the very shallow depth range,
- there is still room for improvement
- I want to exclude bands beyond 670 nm from the PANHRO band agregation
|
Using Jerlov's data for calibration with hyperspectral images - This is a strong confirmation that , using Jerlov's data along with an narrow_bands hyperspectral image just yields the correct values for the spectral operational diffuse attenuation coefficient 2K
- This is also a confirmation that the 2K values between 575 and 600 nm, 2K values must be lowered significantly:
- -0.02 m-1 for band_8 at 564 nm
- -0.02 m-1 for band_9 at 574 nm
- -0.07 m-1 for band_10 at 596 nm
- -0.03 m-1 for band_15 at 675 nm ?
- No proper calibration may be achieved unless this tweaking is accounted for
- This confirms what was established in the early days of using CASI datasets
| No need for field data for calibration when using hyperspectral images - This is a strong confirmation that , using Jerlov's data along with an narrow_bands hyperspectral image just yields
- Therefore : no need for field data when using narrow_band images
- The reason why this is not really true for WV2 or OLI data is :
- these are very_wide_band images
|