fied as the
d.
DEL
an observed
and bottom
(3)
s (Stumpf et
(4)
nes that the
form for the
it using two
rent bottom
ice between
r the scene.
G)
sh multiple
rresponding
y corrected,
, where Raci
., 2000). In
ast as many
y area, an
estimated.
(6)
under the
re uniform
fluenced by
tance. This
er the depth
nagery data
ild produce
During the last three decades several bathymetry applications
were accomplished based on the above model. Two or more
bands of low or high resolution passive images were tested in an
effort to remove errors due to bottom and/or water quality
differences producing quite satisfying results (Lyzenga, 1985;
van Hengel and Spietzer, 1988; Papadopoulou and Tsakiri-
Strati, 1998; Hatzigaki et al, 2000; Stumpf et al, 2003;
Lyzenga, 2006; Bramante et al., 2010; Liu et al., 2010; Lyons et
al., 2011).
4. DATA AND PRE-PROCESSING
4.1 The multispectral imagery and echo sounding data
The depth estimation concerns the coastal area of Nea
Michaniona, Thessaloniki, in the northern part of Greece. The
sea bottom changes smoothly and the water is clear. The
shallower parts are covered with dense sea grass while the
deeper area is sandy.
Figure 1. The study image area (R:4, G:3, B:2)
The imagery data set included the eight (8) bands of
Worldview-2 multispectral image. The image was acquired in
16 June 2010 with spatial resolution of 2m. Despite the water
clarity, the depths estimation was constrained by image noise
that sun glint caused by appearing sparsely in a great part of
image scene. The available data were georeferenced to UTM
(zone 34) system and WGS84. The study area included only the
water region of the image (fig.1). From now on, the 8 bands of
the image will be symbolized as: band 1 (coastal), 2 (blue), 3
(green), 4 (yellow), 5 (red), 6 (red-edge), NR1 (first near-
infrared) and NIR2 (second near-infrared).
The linear bathymetric model was calibrated using echo
sounding data. The survey of the bottom was accomplished
through 719 measurements of depths (from 3.5 m to 15.0 m)
and GPS corresponding horizontal positions on a calm sea
surface. The echo sounding device was a CODEN CVS106 and
the GPS pair of dual frequency receivers was the model system
300 of Leica. The internal accuracy of depth measurments
reached 10cm. The horizontal position was determined using
the kinematic method (Tziavos, 1996, Andritsanos et al, 1997,
Fotiou and Pikridas, 2006) with a final accuracy of 5-6cm. The
all data process was performed using manufacturer processing
software and horizontal coordinates were georeferenced to the
system of the multispectral data.
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B8, 2012
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia
4.2 Imagery data pre-processing
The conversion from radiometrically corrected image pixels to
spectral reflectance (Updike and Comp, 2010) was realized
prior to deglinting process and atmospheric correction for every
band. The given equation required the absolute radiometric
calibration factor and the effective bandwidth for a certain band
that were available in the image metadata file.
The technique of Hedley et al. (2005) was implemented on the
‘glint’ image bands towards the correction of sun-glint effect.
Three image samples with size 50x50 pixels were carefully
selected from glinted areas at different locations on the image.
The critical at this point was the definition of the proper band
combination of NIR (two bands) and visible (six bands)
available bands that would be involved in linear regression
model. Experimental results demonstrated that there was a
strong linear relationship among the ‘new’ bands, i.e. band 1,
band 4 and band 6 with the NIR2, and among the ‘traditional’
bands, band 2, band 3 and band 5 with the NIR1. Thus the de-
glinting process was twofold, one for each set of images. As
soon as the regression slope was defined for every band
combination, the equation (1) was used to determine the
deglinted pixel.
The atmospheric correction through the subtraction of the dark
pixel value followed the glint correction. In order to avoid
negative differences between the image pixels and the dark
pixel value, the histogramme of every band was examined and a
cut-off at its lower end was spotted. The value corresponding to
this cut-off was considered as the dark pixel value (Benny and
Dawson, 1983). A very small proportion of pixels had values
less than the dark pixel value but this fact did not affect the
correction procedure. For the implementation of the linear
bathymetric model (eq. 6) the natural logarithm of the corrected
pixel values was calculated.
5. DEPTH ESTIMATION
5.1 Depth estimation
The linear model was firstly implemented over the total study
area for the corrected bands of the Worldview-2 image. Bands
1, 2, 3, 4 and 5 were used. For this particular step of the study
as well as for all the following steps, band 6, NIR1 and NIR2
were excluded as their spectral information is generally conside-
red insignificant for bathymetry applications. Using 250 control
points with known depth, the linearity between the depth
(dependent variable) and corresponding pixel values
(independent variables) of every band was firstly tested. The
scatterplots showed that the relationship between depth and
band values was not linear. The linearity was affected by small
pixel values existing in swallow water area due to the presence
of sea-grass. Together the high spatial image resolution accents
the differences in bottom types since the detailed and clear
information. Thus, the study area was separated by optical
interpretation into three different areas according to their
bottom type: area A where the bottom is sandy (depths about
6.0 m to 15.0 m), area B where the bottom is mostly sandy with
sparsely distributed sea grass (depths about 2.5 m to 6.0 m) and
area C where the bottom is densely covered with sea grass.
(depths about 2.0 m to 6.0 m). For every area and every band a
new dark pixel value was defined as described in $4.
