Specify the model calibration diagram

back to Help
A final calibration is a trade off between many parameters
  • The aim of the calibration process is to lock -or "frame"- the optical model onto observed spectral Brightest_Pixels_Line and Soil Line as tightly as possible.
  • The visual display helps a lot, so that the practitioner can keep control on the physical meaning of what he/she is doing.
    • It'll be a long way from first calibration tmnov_cal_A1.png to final calibration tmnov_final_calibration.png though! 
The following Comand Line arguments must be specified
  • -extract... :
    • 4SM first needs to read the -extract... argument
    • to ensure that it reads the proper calibration data text file
  • WLm... and WLM...: specify wavelength ranges for wavebands (not needed for SPOT, TM and ETM)
  • -cWL... specify the CoefWL values: by defaults they are all set to 0.5 (not needed for SPOT)
  • -CP... : specify a calibration pixel for BPL: this is optional
  • -KK... : specify the bWLi, bWLj and the ratio Ki/Kj for estimation of Jerlov water type
    • a Blue/Green pair is recommended, because of the great variability of the K[blue]/K[green] ratio
    • if no -CP argument is enabled, the ratio Ki/Kj provided is used for estimation of Jerlov water type
    • if a -CP argument is enabled, a new Ki/Kj ratio is computed using LsM and the calibration pixel provided: this overrides the value provided in the commandline
  • -Z........specify the conditions for computing water depth
  • -deglint... : to enable deglinting
    • BPL pixels shall first be deglinted before linearization,
    • then used for preparation of the calibration plot
The -Calibrate Comand Line argument 
  • V( v)...............enable(disable) the verbose mode
  • CodePS......various displays may be enabled/disabled: see the Syntax_for CodePS: See Syntax_Calibrate
    • B(b) enable(disable) the plot of BPL pixels
    • D(d) enable(disable) the plot of DPL
    • S(s) enable(disable) the plot of SL
      • M(D,N) specify which Soil Line is to be plotted:
        • M.........for SL_mSOIL
        • D.........for SL_DPL
        • N.........for SL_NIR
    • P(p) enable(disable) plot of all mSOIL pixels
    • Z(z) enable(disable) plot of computed depths
    • G(g) enable(disadle) the plotting of ghost pixels
  • /b4_b3_b2_b1/.......wavebands to calibrate, in order of decreasing wavelengths
  • B(b) nable(disable) the plot of bidimensional histograms of shallow pixels
    for pairs 12, 13 or 23, 14 or 24 or 34
The Calibration process
  • Starting from first calibration like tmnov_first_calibration for example:
    • K1K2 at 0.60 is obviously too low: it needs to be lowered so that the BPL match the observed BPL pixels
    • LsM are obviously too low (large stars)
    • Lw values need to be modified so that the Soil Line match the observed SL pixels
    • hump : there is an odd hump on the BPL: the first calibration data file might well be less than optimal
    • location : the practitioner can check the location of any particular calibration pixel through the Row_Line parameters in the calibration text file: for example to locate the subarea in the image which causes the hump
  • The first extracted calibration dataset might be less than optimal:
    • after a while, a better data extraction might have to be done
    • See that by just avoiding a certain area in the image (see tmnov_redpolygon),
    • the question of the hump on the BPL clarifies: no "hump" any more and a clearer choice for the ratio K1/K2for a comparison, see the calibration for the whole Caicos Bank caicostm_polygongreen    caicos_cal_full     where no hump is seen because the very bright bottom substrates which cause the hump in the tmnov image (red polygon) are present at all depths in the whole bank (green polygon)
  • LsM, La and Lsw: likely need to be adjusted gradually
  • CP: one (or more) Calibration Pixel(s) shall help a lot: the -CP argument
    • It is taken to be one of the M_1_2 pixels
    • It may be modified manually, so as to optimize the fit of the BPL onto observed BPL pixels
    • Look out for bad calibration
  • cWL: exact wavelengths are unknown:
    • wavelengths might need to be gradually adjusted through the -cWL argument,
    • so as to obtain a consistent set of K i/K jratios among all possible pairs of wavebands
  • Square inserts Z2 vs Z3, Z3 vs Z4: a good fit of computed depths is a must
    • a good fit demonstrates the internal consistency the the spectral K values
    • a good fit ensures a smooth transition in image_Z where band_3 becomes optically deep
    • pixels should display along the diagonal, slope=1, bias=0, no curved end at great depth values
  • For a N band image , the calibration process must be iterated by changing the choice of wavebands to plot in the -Calibrate/.../ b4_b3_b2_b1/... argument, so that all wavebands are calibrated by the end
Safety first: is your calibration up to date?
  • After a while, changes to the command line might have caused a difficult situation.
    • because you might have modified some parameters
    • since you last extracted the calibration parameters,
  • It is safe to
    • extract the calibration parameters again
    • check that your calibration is still valid, or make appropriate changes until satisfied
Sensitive parameters which can affect the calibration dataset are
  • -LsM... as the Soil Line by NIR is determined by LsM and La for the Red/Nir pair of bands
  • -Lw... as the Soil Line by NIR is determined by LsM and La for the Red/Nir pair of bands
  • Mask...... /Mask_3 Make sure your calibration data fit your expectation rather than some non-representative specific mask subset
  • NoSmooth: -smooth_4d/Smart+ Smoothing should be avoided when extracting calibration data
Sensitive parameters which can affect the calibration itself are
  • -WL... or -cWL...
  • -CP...
  • -Lsw... and -dLsw...
  •  -Lw... and -dLw...
 Physical consistency: upon modeling, 4SM checks that
  • water volume reflectance: normalized Lw increases steadily as wavelength decreases
    • if this is not the case, 4SM shall require you to make necessary changes before proceeding.
  • diffuse attenuation coefficient: K increases steadily as wavelength decreases
    • if this is not the case, 4SM shall require you to make necessary changes before proceeding.

The calibration data text file is extracted and written
upon the execution of the -Extract command line argument.

See for example : it contains the following sections, in that order

  • AvShallow: Average radiances at a shallow location: this point shall be displayed in the calibration plot
    • AvShallow 1 + + Vegetation 45.8 28.8 76.7
  • This is optional
    • up to 5 points may be specified
    • such points may be sampled through the -DN... argument

The BPL pixels for all pairs of bands

the format is : "M for Maximum", band i, band j, Row#, Line#, band_1, band_2, ... , band_N
  • M_003 says that this pixel of the BPL is coded 3 in mask mSE
    • from M_003 1 2 249 231 141.3 78.0 6.5         to M_003 1 2 38 581 30.5 10.0 6.6
    • from M_003 1 3 194 23 180.0 163.0 138.0     to M_003 1 3 87 500 32.0 14.0 6.0
    • from M_003 2 3 194 23 180.0 163.0 138.0     to M_003 2 3 87 500 32.0 14.0 6.0
  • etc... for the pairs  1/4, 2/4 and 3/4
  • etc... for the pairs  1/5, 2/5, 3/5 and 4/5
  • etc...
 The DPL pixels for all pairs of bands
the format is : "m for minimum", band i, band j, Row#, Line#, band_1, band_2, ... , band_N 

  • m_003 says that this pixel of the DPL is coded 3 in mask mSE   
  • from m_003 1 2 .........     to m_003 1 2 ........
  • from m_003 1 3 ..........    to m_003 1 3 ........
  • etc... for the pairs 2/3, 1/4, 2/4 and 3/4
  • etc... for the pairs 1/5, 2/5, 3/5 and 4/5
  • etc...

 Soil Line by mSOIL (mostly outdated)
  • SL_SOIL DNnir band_1, band_2, ... ,band_N
  • from SL_SOIL 135 - - - 182.0 100.0 133.0 135.0      to SL_SOIL 4 - - - 60.4 15.7 9.9 4.0

 Soil Line by DPL
  • SL_DPL DNnir band_1, band_2, ... ,band_N
  • from #SL_DPL 92 - - - 179.0 92.0 73.0 8.0        to #SL_DPL 2 - - - 55.0 16.3 8.3 2.0
Soil Line by NIR
  • SL_NIR DNnir band_1, band_2, . .. ,band_N
  • from SL_NIR 169 - - - 200.2 181.8 169.0       to      SL_NIR 28 - - - 58.0 33.0 28.0
 BiDimensional Histograms for all pairs
  • LogNbPixFullBDH, bandi, bandj, DNi, DNj, LN
    • LN is the NumberOfPixels
  • from LogNbPixFullBDH 1 2 35 10 1 ..., etc... for the pair 1/2
  • etc...
  •  LogNbPix 3 4 ... , etc... for the pair 3/4


How to Read The Calibration Diagram
The graphs in the calibration diagram
display the extracted calibration data, either natural or linearized
  • The lower graphs show calibration in natural
  • The upper graphs show calibration in linearized data.
  • For a 3-bands spectral image, this diagram .......has 4 graphs .
  • For a N-bands spectral image, this diagram only has 6 graphs .

Solid squares...............for BPLpixels
White squares..............for Ghost BPL pixels
Crosses.........................for DPL pixels
White circles................for averaged baresoils
Solid triangles
  • Symbols for extracted calibration data
    • Solid squares are used for extracted pixels that represent the Brightest Pixels M_1_2 , M_1_3 , M_2_3 , etc
    • Crosses are used for extracted pixels that represent the Darkest Pixels m_1_2 , m_1_3 , m_2_3 , etc
    • Small white circles   are used for average pixels that represent the Soil Line
    • Small white squares are used to plot "ghost" pixelsbecause BPL pixels are real pixels:
      • BPL pixels extracted for the Blue/Red pair can be plotted for the Blue/Green pair
      • BPL pixels extracted for the Green/Red pair can be plotted for the Blue/Green pair
      • BPL pixels extracted for the Blue/Green pair can be plotted for both the Blue/red and the Green/Red pairs
      • this is a good test of the BPL assumption, which assumes that the BPL pixels are not spectrally contrasted. Ideally, BPL pixels should represent neither greenish bottom substrates  nor brownish bottom substrates
  • Symbols for calibration model
    • Plain line : the mathematical models for the Brightest Pixels Line and the Soil Line are overlaid
      • this is used by the practitioner to work out a best fit visually
      • the start of the slope of the BPL is also indicated
    • Plain line : a choice of three different concepts for the Soil Line is offered:
      • either SL_SOIL
      • or SL_DPL
      • or SL_NIR
    • Solid triangles
      • if a Calibration Pixel is enabled through the -CP argument,
      • it is plotted with an indication of its computed depth
    • Large stars the spectral LsM
    • Large crosses spectral Lsw and La

Square insets 
they are scatter plots of computed depths in meters
  • each insert shows depths computed for BPL pixels shown in the corresponding graph
  • for each BPL pixel, two depths are computed: Z_based_on_band_N and Z_based_on_band_N-1
  • each insert compares Z N with Z N-1 computed on the same pixels
    • in tmnov_final_calibration_zoom3 ,
      • Z3 are depths computed based on TM 3 for M_1_3 and M_2_3 pixels
      • Z2 are depths computed based on TM 2 for M_1_3 and M_2_3 pixels
    • Bias : any bias in this display is an indication that LsM values should be modified
    • Slope : if the slope departs from diagonal , it signals some inconsistency in the choice for spectral K
    • Curved display: this display also helps to adjust Lsw values: for example,
    • A good diagonal display is the desired goal, like in tmnov_final_calibration_zoom2 , showing that consistent spectral LsM and K values have been achieved. In other words, the BPL is fully specified to satisfaction
  • the start of a choice of K i/K jratios is also displayed in the linearized graphs, from 0.0 to 1.9 with an increment of 0.1
Upper Right inset

It displays Jerlov's family of curves of diffuse attenuation coefficients 2K

tmnov_cal_Jerlov_faulty                  tmnov_cal_Jerlov_good
  • the red curve shows the intermediate water type which was interpolated to account for the observed value K[WL i]/K[WL j]
  • the two bold black dots stand for wavelengths WL iand WL j
  • smaller black dots stand for other wavelengths
  • WLmin and WLmax are also shown with vertical bars
 BPL or DPL pixels
these are real image pixels, their location is known
  • The practitioner can check the location of any particular calibration pixel through the Row_Line parameters in the calibration text file
  • M_003     1     2      249     231     141.3     78.0     6.5
    • M stands for Maximum and denotes a pixel of the BPL
    • under Mask=3
    • for the pair 1/2
  • For band 1 vs band 2 at Row_249 and Line_231,
    • Ls1=141.3,
    • Ls2=78.0,
    • Ls3=6.5
  • This allows the practitioner to check the particular location of any BPL pixel,
    • as glints, breaking waves, clouds, etc, must be avoided by all means.

 The end!!!!!

Créer un site
Créer un site