| |The underwater light field is complex: 4SM is not a "push-button" magic black box for dummies The prospective "operator" must definitely grow up
into a seasoned "practitioner"
- If the imagery is suitable, a knowledgeable and seasoned 4SM practioner shall obtain satisfaction:
- a first order atmospheric correction of the image,
- a thoroughly deglinted image,
- the calibration of spectral operational attenuation coefficient K in m -1
- the water column correction of the shallow areas of the spectral image (~= a low-tide view),
- along with an estimate of the depth in meters at each shallow pixel, which is pretty close to actual depth,
- without the need or use of any field data .
- If the imagery is not suitable, 4SM shall produce an output anyway: garbage in, garbage out.
- If the practioner is not suitable, 4SM shall produce an output anyway.
the deep water radiance Lsw
- Unsuitable deep water radiance Lsw values shall result in unacceptable output
- Deglinting: in many cases, it is necessary to remove the "glint" from the imagery to the extent possible
- This also efficiently removes the haze that surrounds dense clouds
- This also efficiently removes the atmospheric adjacency effect where the landmass is bright and poorly vegetated.
- Unfortunately though, this tends to damage radiances of very shallow bottoms
- Hopefully the image, or part of it, is now suitable for shallow water modeling. Alternately , that image must be discarded
- The outcome is a set of spectral deep water radiances Lsw which are applicable to the whole sub-area under study
- A set of spectral LsM values : this represents the brightest shallow substrate which exists inside the sub-area under study
- A set of spectral La values : this intends to represent the atmospheric path radiance
- Spectral LsM and spectral La values make up a spectral radiometric model of pixels at null depth: Z=0
the Brightest Pixels Line (BPL)
- Ratios K i/K j for all pairs of wavebands i and j are estimated: the slope of the linearized BPL is the ratio Ki/Kj
- Their internal consistency as this system of ratios is required and verified: we found that they need to fit Jerlov's optical classification of marine and coastal water types worldwide.
the atmospheric path radiance La
- Combined with Lsw, the Soil Line is used to estimate the atmospheric path radiance La
- This achieves for a first order atmospheric correction of the imagery (cf the "dark pixel" assumption)
- this does not correct for the atmospheric adjacency effect, though.
the color of optically deep water Lw
- The BOA deep water radiance is then estimated as Lw=Lsw-La
- This is the color of the optically deep water column, which is also called water volume reflectance.
| | the two-ways spectral effective attenuation coefficients K
- The ratio K i/K j for one pair of visible bands i and j, along with Jerlov's data, are used to approximate effective spectral K val
- Spectral K values provide a consistent estimation of the optical properties of the clearest waters that are present inside the sub-area under study.
- This last statement means that results obtained for areas that have less clear waters may be erroneous: this shows in the bottom reflectance results.
- 4SM operates the simplified radiative transfer equation on each shallow water pixel, through an iteration of inverse modeling steps.
- the spectral water-leaving radiance is then reverse-modeled into a spectral shallow "water-column corrected" bottom signature that is compatible with the Soils Line assumption.
- this is achieved by increasing the depth Z until spectral LB is deemed acceptable.
- This yields an estimate of both
- the shallow "water column corrected" bottom spectral signature
- the shallow water depth
- Tide correction: as an option, the estimated depth is then corrected for Height of Tide.
- Seatruth: the estimated depth is then multiplied by a final Depth Correcting Factor to be derived from some existing sea truth.
4SM then operates a fully supervised bottom typing scheme
| that need to be given appropriate values in order to alleviate the most conspicuous artifacts in the output: |
- Threshold on inconsistent high radiance (high -saturated- radiance is bad data)
- Threshold on inconsistent low radiance (low radiance is bad data)
- Threshold on low radiance (bottom contrast Ls-Lsw is too faint)