4SM on Landsat 8 OLIP at Caicos Bank Bahamas

scene LC80090452013133LGN01 May 13th 2013
Work done august 2013
The data
Deglinting
Optical calibration
Calibration of the Panchromatic band
Modeling: breaking new ground
Retrieved depth
DTM seatruth RMSE=0.74m
Bottom reflectance
home

DTM sea truth: RMSE=0.52 m for the whole image
No need for field data
Tide height 0.9 m
2K is correct for the Blue and Green bands
This DTM gives me a valuable opportunituy

Seatruth regression Z4SM vs ZDTM+0.9m RMSE=0.52 m Tide correction applied: 0.9 m	Difference ZDTM+0.9 - Z4SM Tide correction applied: 0.9 m see legend
4SM retrieved depth see legend	DTM by Harris-Ellis resampled to 30 m GSD and scaled to centimeters see legend
NO NEED FOR FIELD DATA	NO NEED FOR ATMOSPHERIC CORRECTION

The data

Seven bands are used: Purple, Blue, Green, PAN, Red, NIR and SWIR1

Panchromatic band, sum-aggregated to 30 m,
exhibits excellent radiometric quality

TCC
Landsat OLI exhibits excellent radiometric quality

Deglinting

Glint regression based on SWIR1 band SWIR1 is quite convenient for deglinting	Glint regression based on NIR band
Deglinted TCC Deglinting is far from perfect: I need to dig harder: lots of clouds, lots of surface glitter	Deglinting along Profile Red

**Bottom detection by OLI's red band**
Deglinted Red band	Red band at 655 nm in this image exhibits bottom detection in excess of 10 m, as evidenced using BILKO's field dataset This means that operational K_red for remote sensing radiance is distinctly lower in this scene than diffuse attenuation coefficient for downwelling irradiance Kd_red of Jerlov. Either we need to shift WL_redat an operational wavelength much shorter than mid-waveband, which would only reach down to ~6 m over very bright bottoms. I suppose sensor designers would object loudly! Or , as I realized using HYPERION in 2014, operational 2K_red for remote sensing radiance is affected by viewing geometrical conditions and/or atmospheric optical properties. I suppose HYDROLIGHT designers would agree loudly! See comment by Jerlov.

Optical calibration
is under tight control, thank to the optimization process:
Water type OIB+0.7 of Jerlov
Spectral reflectance of the brightest bottoms:
0.272 0.343 0.439 0.462 0.548 0.678 0.634
on the scale 0-1, as calibrated using the _MTL.txt textfile

Calibration diagram for bands Blue, Green, Red and NIR [2] vs [3]: excellent fit of GREEN against BLUE	Calibration diagram for bands Coastal, Blue, PAN and Red [1&2] vs [4]: excellent fit of COASTAL and BLUE against PAN
Calibration diagram for bands Blue, PAN, Red and NIR [2] vs [4]: excellent fit of PAN band against the BLUE	Calibration diagram 2K for Coastal band must be increased for bands Coastal, Blue, Green and Red [1&2] vs [3]: excellent fit of COASTAL and BLUE against GREEN
Rw: water volume reflectance 0.018 0.017 0.003 0.001 0.000 0.000 0.000	Ra: path radiance 0.106 0.076 0.042 0.037 0.023 0.011 0.003
Spectral operational 2K This calibration yields the following: 2K/0.104/0.104/0.190/....../0.637/4.447/18.00 2K_blue =0.104 m^-1 2K_green=0.190 m^-1 2K_red =0.637 m^-1 well below pure water at 655 nm! 2K_nir ~=0.45 m^-1 is a customary for NIR *Final_Z = CoefZZ -Tide_to_chart_datum** I hope that CoefZ=1.00!	2K_PAN The effective diffuse attenuation coefficient of the PAN band decreases progressively from very high values at shallow depths to much lower values at depth It is necessary to account for it in the optical calibration of the PAN band in a very precise way, based on the Landsat 8 PAN response curve

Modeling: breaking new ground

Using the PAN band is a great step forward for water column correction.

LEFT

This shows in yellow that the PAN band was used for a very large majority of shallow pixels,
while deeper pixels colored in green are modeled using purple, blue against green.

Yellow: modeling by bands 1, 2 and 3 against 4

Green: modeling by bands 1 and 2 against 3

RIGHT

For a comparison, we also show the more general case,
which takes advantage of the whole visible range.

Red: modeling by bands 1, 2, and 3 against 5

Yellow: modeling by bands 1, 2 and 3 against 4

Green: modeling by bands 1 and 2 against 3

Water column correction
Performance of the PAN band reaches ~20 m over bright bottoms in this OIB water type of Jerlov

4SM retrieved depth see legend	BOA TCC of water column corrected bottom reflectance rosy hues: possibly some "whiting"
PAN solution and RED solution This plot compares Z4 vs Z5 along Profile Red Z4 uses Purple, Blue, Green against PAN Z5 uses Purple, Blue, Green, and PAN against RED Deglinting and Smart-Smoothing applied	GREEN solution and PAN solution This plot compares Z3 vs Z4 along Profile Red Z4 uses Purple, Blue, Green against PAN Z3 uses Purple, Blue against Green Deglinting and Smart-Smoothing applied
Waypoints A1 and B1: the RED solution underestimates the retrieved depth where some "whiting" or suspended particle load comes into play at waypoints A1 and B1 Waypoint A2: the PAN solution appears to underestimate retrieved depth over very dark bottoms see waypoint A2	Waypoint A2: the GREEN solution appears to underestimate even more badly the retrieved depth over very dark bottoms see waypoint A2

Bottom typing
As of august 2016, 4SM operates a generic bottom typing protocol,
based on the spectral angle mapping.

SAM classified image SA=(Rwcc_coastal+Rwcc_blue)/(2*Rwcc_green) Now that water column corrected radiances are converted to BOA spectral reflectances, bottom typing is performed in a generic way, which is applicable in 4SM to any Landsat 8 or WV2 image Then BOA bottom type signatures can be extracted using the SA-classified image
SAM classified image zoom	TCC BOA deglinted zoom
SA=0.684 SA=(Rwcc_coastal+Rwcc_blue)/(2*Rwcc_green) The calibration is optimized so as to achieve a "flat" bottom type signature for the brightest bottom type in this scene. This involves a very fine-tuning of the -LsM... parameter	SA=0.475 SA=(Rwcc_coastal+Rwcc_blue)/(2*Rwcc_green)

Afficher la suite de cette page