Seatruth
Discrepancies
Downloads
My whishes


Seatruth
U. of Queensland DTM is used as a seatruth dataset
Data collection over two days, not tide corrected:
 "The data is not tide corrected and consists of stitched flight lines taken at slightly different times,
the linear discontinuity seen across the right of the image is probably due to small tidal variation between flight line times.
Small gaps in the image are due to gaps in coverage between the flight lines"
  -RegressZZ/v/p/DTM_1/0.001_0_0/1.000_0.0/ZR/6_14/0_201/0_50_50/*     

X axis: DTM depths are used as a seatruth dataset Y axis: 4SM retrieved depth	Seatruth regression   as there has been no calibration against any field data in 4SM, Final Z = coefZ * Computed_Z + Htide DTM DTM model was carefully calibrated using depth sounding lines. DTM depths are truncated at ~20 m?? DTM depths are not corrected for tide effects. For this regression one in 400 pixels was used Regression most pixels are in the 0-3 m depth range. from looking at this correlation, a coefZ=1.075 would appear to be required: Final Z = 1.075 *Computed_Z
Z4SM - ZDTM in centimeters, in channel_73 Profile Black is shown (see below) Red tones where  over-estimated  depths Blue tones where under-estimated depths severe and orderly discrepancies are observed: this is not caused by noise. this is inherent to the way depths are computed in 4SM: very contrasted bottom substrate signatures are ill-corrected by the 4SM process. this problem is now solved for Landsat 8 and WorldView 2 imageries through the "PAN solution". in this respect, I have great expectation in developping the "PAN solution" for hyperspectral data.	Legend  ZComputed is Z4SM ZRecorded   is ZDTM
ZZRegressor: Statistics of seatruth ZC-ZR on image heron_20020702 at HeronIsland, Australia N=23.895 millions pixels   by Morel's 4SM method HTide=0.00 Smooth=1_5 Using_bands_1_2_3_4_5_6_7_8_9_10_11_12_13_14 cZ=1.00  1.96% pixels with depth underestimated by more than -5.0 m  1.43% pixels with depth           ""                -4.0 m  2.09% pixels with depth           ""                -3.0 m  2.57% pixels with depth           ""                -2.0 m  3.16% pixels with depth           ""                -1.0 m 59.32% pixels with depth           ""                -0.0 m   24.84% pixels with depth  overestimated by more than  0.0 m  1.27% pixels with depth            ""                1.0 m  0.79% pixels with depth            ""                2.0 m  0.63% pixels with depth            ""                3.0 m  0.53% pixels with depth            ""                4.0 m  1.40% pixels with depth            ""                5.0 m   84.16% of computed depths are within  +-1.0 m of DTM depth 88.59% of computed depths are within  +-2.0 m of DTM depth 91.95% of computed depths are within  +-3.0 m of DTM depth 94.67% of computed depths are within  +-5.0 m of DTM depth over a total of 100.00% of computed depths

Discrepancies
The discrepancies call for some explanation!

Profile Black Note that the DTM data is truncated at ~20 m Note the ~3 m high "mound" from 6 km to 8 km it is conspicuous, it shows badly as a gap in the map of retrieved depth there is no need for the DTM reference for a trained practioner to identify it on a profile or upon inspecting the map of retrieved depths.	Profile Black, band 8 disabled just by disabling band 8, the problem is alleviated but it is not solved the solution would require a smart optimization involving the knowledge gained at shallower depths such optimization potentially can replace the use of a spectral library. I already tried to operate this optimization in 4SM, with limited success though, as my crude process tends to stop at the first opportunity which is not allways the best: I keep digging: see below!
	Nothing wrong with the calibration of band 8 Calibration diagram for bands 3, 8, 13 and 17 i.e. 479, 564, 643, and 705 nm
band 8 solution at pixel  R4709 L372 4SMZ=14.31 m               DTMZ=17.43 m So band 8 solution yields 4SMZ=14.31 m.	band 9 solutionat pixel  R4709 L372 4SMZ=18.46 m               DTMZ=17.43 m So band 9 solution yields 4SMZ=18.46 m. It exhibits a strong/narrow reflectance peak in band 8: There is the cause of observed discrepancies: a very green bottom signature in band 8 at 564 nm.
Before: band 8 enabled	After: band 8 disabled

Downloads
Heron Reef data provided courtesy of the Center for Spatial Environmental Research at the University of Queensland.
Dataset provided by HyPhoon.com.
Modeling with band 8 disabled (Dec 21rst 2013)
4SM retrieved depth is in heron_20020702_Z.001
For your convenience,
these files are now in ENVI format,
with the same content
 

Importing those PCIDSK format files Subject to the attached HeronReef_DataUserAgreement, You may import these files into your favorite image processing format Or you may import inividual channel files as "raw raster" 6949 columns  5174 lines 8 bits unsigned or 16 bits signed (beware of byte swap) As of December 27th, the georeferencing of these files should be correct, as the seatruth datasets now overlay correctly -this is the southern hemisphere, Eh!!

heron_20020702_result_20131219_Z.zip  at https://drive.google.com/file/d/0B6EgWTY7EkNxcmNnWDFQbTRLb1k/edit?usp=sharing contains the following (54 MB):  71908252  2013-12-19 19:37   heron_20020702_Z.001           Z4SM depth in centimeters      6949*5174 S16 71908252  2013-12-19 19:38   heron_20020702_Z.002           ZDTM depth in centimeters      6949*5174 S16 71908252  2013-12-19 19:38   heron_20020702_Z.003           Z4SM-ZDTM in centimeters      6949*5174 S16       73216  2013-12-19 19:36   heron_20020702_Z.pix             file header     203811  2013-12-16 18:41   HeronReef_DataUserAgreement_v000.pdf

heron_20020702_result_20131219_WCC.zip  at https://drive.google.com/file/d/0B6EgWTY7EkNxZ0VsSGVINFEwcFU/edit?usp=sharing contains the following (280 MB):  35954126  2013-12-19 19:51   heron_20020702_WCC.001     water column corrected band   1     6949*5174 U8  35954126  2013-12-19 19:51   heron_20020702_WCC.002     water column corrected band   2     6949*5174 U8  35954126  2013-12-19 19:51   heron_20020702_WCC.003     water column corrected band   3     6949*5174 U8  35954126  2013-12-19 19:51   heron_20020702_WCC.004     water column corrected band   4     6949*5174 U8  35954126  2013-12-19 19:51   heron_20020702_WCC.005     water column corrected band   5     6949*5174 U8  35954126  2013-12-19 19:51   heron_20020702_WCC.006     water column corrected band   6     6949*5174 U8  35954126  2013-12-19 19:51   heron_20020702_WCC.007     water column corrected band   7     6949*5174 U8  35954126  2013-12-19 19:51   heron_20020702_WCC.008     water column corrected band   8     6949*5174 U8  35954126  2013-12-19 19:51   heron_20020702_WCC.009     water column corrected band   9     6949*5174 U8  35954126  2013-12-19 19:51   heron_20020702_WCC.010     water column corrected band 10     6949*5174 U8  35954126  2013-12-19 19:51   heron_20020702_WCC.011     water column corrected band 11     6949*5174 U8  35954126  2013-12-19 19:51   heron_20020702_WCC.012     water column corrected band 12     6949*5174 U8  35954126  2013-12-19 19:51   heron_20020702_WCC.013     water column corrected band 13     6949*5174 U8  35954126  2013-12-19 19:51   heron_20020702_WCC.014     water column corrected band 14     6949*5174 U8        84480  2013-12-19 19:44   heron_20020702_WCC.pix       file header     203811  2013-12-16 18:41   HeronReef_DataUserAgreement_v000.pdf

My whishes, as of december 2013

After all those years, this has been an uplifting experience. First time I processed a CASI dataset was in 2000: Gezirat Siyul Reef in the Red Sea. Last time I processed a CASI dataset was in 2006: Seychelles. Since then, I processed several HYPERION datasets.

I whish to access the raw Heron Island CASI mosaic dataset: that is balanced and geocorrected, before atmospheric correction and any glint removal, full dataset, not just this subset window. Is this possible/reasonable?

I wish to develop a "PAN solution" for hyperspectral data

next to using a CASI Panchromatic band (march 2014)