WorldView 2 image Gulf of Laganas, Zakynthos Island, Greece  8114*6968 image courtesy of Emmanouil Oikonomou, Technological Institute of Athens, Feb 11th 2016 work done in february 2016 the "PAN solution" uses the PANCHRO band, 4x4 sum-aggregated and coregistered bands 1, 2, 3, Panchro, 4, 5, 6, 7 and 8        become       bands 1, 2, 3, 4, 5, 6, 7, 8 and 9 home

Optical calibration for MULTI bands mBPL=1 vs mBPL=2

Optical calibration for MULTI bands

certainly a "not for beginners" study case
 First look at the 0-5 m depth range under Mask_4 Brown: land areas are flagged in special effects mask any scattered "sea" pixel inside land areas must be forced to land Green polygons mask_1: then, radiometric artifacts over "sea" areas must be eliminated upon extraction of calibration data: such as boats and their wakes Mask_4 is the red area for some reason, things only become convincing when restricting the extraction of calibration data to pixels flaged at Mask=4 (red color). Deglinting of sea pixels is mandatory upon extraction of calibration data, in order to remove adjacency effect which is quite strong here. One stands a null chance of grasping an acceptable optical calibration for this scene, unless the above precautions have been identified and applied cautiously: certainly a "not for beginners" study case First look at the 0-5 m depth range under Mask_4 Special effects mask See AutoCAL under mask_4 Calibration diagram for bands 2, 3, 6 and 7 X[2] vs X[3]: we apply Kblue/Kgreen=0.95 although data do not clearly support this choice X[2_3] vs X6: this diagram, which covers the 0-5 m depth range, clearly shows that Kblue/Kred ~= Kgreen/Kred, in other words Kblue~=Kgreen this confirms Kblue/Kgreen=0.95 This at least applies to the 0-5 m depth range, but obviously does not apply over the 5-30 m depth range under mask_4 Calibration diagram for bands 2, 5, 6 and 7 Over the 0-5 m depth range, the scenario is quite clear and straightforward : we observe a fairly tight fit of the BPL pixels to the Ki/Kj optical model for a OII+0.9 water type of Jerlov for bands 2, 5, 6 and 7 this is obtained by applying a ratio Kgreen/Kblue=0.95 , then deriving spectral K values for all visible bands strictly from Jerlov's data. Brightest shallow pixel LsM -LsM/248.5/267.1/209.4/204.2/206.6/208.0/146.8/125.6/118.0 TOA for Laganas beach Clearly, the brightest shallow bottoms over the 0-5 m depth range are very well estimated. this is ~0.25 times the brightness of coral reef sands Brightest shallow pixel LsM -LsM/248.5/267.1/209.4/204.2/206.6/208.0/146.8/125.6/118.0 TOA for Laganas beach We shall assume that this is also the case over the 5-30 m depth range: this is at the heart of 4SM assumptions i.e that the brightest shallow bottoms are present at all depths, if only as small  isolated patches. Then pay attention to the 5-30 m depth range under Mask_4 Then pay attention to the 5-30 m depth range under Mask_4 Calibration diagram for bands 2, 3, 6 and 7   the optical model now includes spectral diffuse 2K value for all MULTI bands at 1100 discrete depths over the 0-50 m depth range here this describes homogeneous OIII water from 0 to 5 m, then clearing progressively to OIB from 5 to in excess of 30 m. Now we must figure out what is going on over the 5-30 m depth range: see  diagram [X2] vs [X3] we shall now assume that waters become clearer as depth increases, from 5 m down to ~30 m and find a way to account for this assumption in the optical model   For this we use calibration pixels -CP... in the command line by which to force the model into fitting the ovserved data so that the ratio Kblue/Kgreen increases steadily from Kblue/Kgreen=0.95 at 5 m to      Kblue/Kgreen=0.54 at ~30 m -CP/210.9/195.9/123.8/077.2/064.7/051.7/029.5/019.0/010.4_5.00m -CP/200.1/170.0/085.8/063.0/055.5/049.0/029.5/019.0/010.4_20.00m   This amounts to assume that the water type changes progressively from OIII at 5 m to OIB at 30 m this is compatible with the common experience by scuba divers: waters tend to get clearer as one swims away from the beach! Reference to coral reef sands -LsM/423.5/553.4/511.3/544.8/566.3/586.5/426.0/379.3/374.1 TOA in the image for coral reef sand Reference to coral reef sands See that coral reef sands are ~3 times brighter than even the brightest shallow bottoms at Laganas. Calibration diagram for bands 2, 3, 6 and 7 optical model applied to an average coral reef sand Calibration diagram for bands 2, 4, 6 and 7 optical model applied to an average coral reef sand La and Lw Then get an estimation of La: atmospheric path radiance and Lw: water volume reflectance (backscatter) La and Lw Then get an estimation of La: atmospheric path radiance and Lw: water volume reflectance (backscatter) Calibration diagram for bands 2, 3, 6 and 7 Land areas have been excluded so that the bidimensional diagrams represent only marine pixels over the whole depth range Lsw: TOA spectral deep water radiance has been estimated when specifying deglinting conditions. We now need to estimate Lw: BOA spectral water volume reflectance (backscatter). So that spectral atmospheric path radiance may now be specified as path radiance La=Lsw-Lw this is the "dark pixel assumption" for a first order atmospheric correction (assuming that glint -and/or adjacency effect- has been removed efficiently). This is conveniently done using the calibration diagram by adjusting the -Lw... parameter so that the curvature of the Soil Line parallels that of bidim histogram pixels at null depth knowing that, for clear blue waters, the backscatter Lw is negligeable over the RED-NIR range Lw is low but significant over the GREEN range Lw increases dramatically over the BLUE range

Optical calibration for PAN band
PAN band is in channel 4
of my 4SM work database laganas_wvmp.pix

certainly a "not for beginners"
see PAN response for WV2
For a 4SM application,
the PAN response should be stronger in the Blue-Green range:

this would reduce the system noise,
and yield more reliable and stable results.
 As the optical water type has been evaluated at discrete depth intervals over the whole depth range, the spectral diffuse attenuation coefficient for all of Jerlov's visible wavelengths is now specified at 1100 depth intervals for all MULTI bands. This is sort of a LUT which describes diffuse attenuation properties of a homogeneous OIII water from 0 to 5 m, then clearing progressively to OIB from 5 to in excess of 30 m. Using the response curve of the PAN channel of the WV2 sensor, one may now compute KPAN   for each of these 1100 depths. The optical model now includes spectral diffuse 2K value for all WV2 bands, including PAN, at 1100 discrete depths over the 0-50 m depth range Calibration diagram for bands 2, 4, 6 and 7 X[2] vs X[4]: see that the model using the PAN response curve is quite close to yield a very good fit Calibration diagram for bands 2, 4, 6 and 7 with reference to average coral reef sand

March 2nd 2016    AutoCALIBRATION    March 2nd 2016
 AutoCAL After considerable investigation of this scene, I realized that a few artifacts in the sea area, which had not been masked, were still interfering with the extraction of calibration data, thus preventing a clear appreciation of the calibration diagram. I diagnosed this by inspecting the laganas_wvmp_m0.cal text file: all BPL pixels are referenced with their row/line location. Odd pixels turned out to be artifacts, i.e. objects at the sea surface which should have been masked out. It just happened that the area under mask_4 did not have any remaining adverse artifact!   "on the fly" The AutoCAL procedure goes all the way from creating the database, then importing the raw 16U data, then getting glint regressions then using shapefiles to create the special effects mask in channel_10, then extracting the calibration data from the whole scene (except unwanted areas masked at 1) into the text file laganas_wvmp_m0.cal , then formating the script calibration_AutoCal.sh to displaying the laganas_wvmp_AutoCal_8_6_3_2_m0.eps calibration diagram. AutoCAL laganas_wvmp_8_6_3_2_m0_AutoCal This diagram represents the whole scene,  -instead of just the area masked at 4. In the X[2_3] vs X[6] diagram, it is quite clear that Kblue/Kred~=Kgreen/Kred. In other words Kblue/Kgreen~=1 in the 0-5 m depth range of the RED band. This denotes a OII+0.9 water type of Jerlov. In this AutoCAL procedure, the ratio Kblue/Kgreen=0.5 by default, just because a value is needed for the procedure to execute. The manual calibration that follows now needs to apply a realistic value for Kblue/Kgreen. If the BPL assumption holds, all BPL pixels in the X[2] vs X[3] diagram represent the same brightest bottom substrate. This leads us to assume that waters get progressively clearer as the depth increases deeper than ~5 m.
 "on the fly" nice -19  ./4SM.5.03 -Process    -Origin/Oikonomou                                 -DB/laganas_wvmp/50_3S_9S_0s/9_54/8114_6968/484.615_4181.115/1_1                             -Mis/Greece/Zakynthos/WV02/MultiPan/PIX/bOA/UTM_34_008/0.002_0.002/22_AUG_2014                     -MakePIX/Import/scale/AutoCAL    @-Prepare/Scale                                     -Import/v/dTM/data*laganas_data/dbnc_0_0s_9S_0s/R8114_L6968/Origin_484.614_4181.116_0_0/chIn1,9/chOut54,62 -LS/005000/05000/05000/05000/05000/05000/05000/05000/05000                         -M/@000001/@0002/@0003/00004/00005/00006/00007/00008/00009                         -SCL/00050/00035/00060/00060/00080/00060/00100/00110/00100                         -deglint/vRbaD/q/DegTol1.00/F1/L9/N1/FN/mDEGLINT1/GlintM70                         -extract/v/rawBDH/FullBDH/NIRband8/NIRmax255/mapBPL/mSOIL21/mBPL2                     -AutoCAL/print_2_3/mapBPL/MakeNewMask/Land_5_1.05_8/GetGlint/GlintM70/clouds                 #First create laganas_wvmp.pix working database, then import the 16U data #Then use masks to get glint parameters, and to create special effect mask in channel_10 #Then extract calibration data textfiles laganas_wvmp_m0.cal and  laganas_wvmp_m0.bdh #Then format calibration script calibration_AutoCal.sh #Then run calibration_AutoCal.sh #Then display calibration diagram laganas_wvmp_AutoCal_8_6_3_2_m0.eps #Then format a complete command line AutoCal.txt #Then stop

 mBPL=1 to 20 all marine pixels are accepted -Extract/v/raw...............SOIL21/mBPL1 mBPL=2 to 20 marine pixels masked at 1 are excluded -Extract/v/raw...............SOIL21/mBPL2 Zoom on X[2] vs X[3] mBPL=1 to 20 Most confusing see how artefacts feature Zoom on X[2] vs X[3] mBPL=2 to 20 Things get clearer! see that artifact pixels have disappeared