X-axis:
ferences
easured
timated
depths
Y-axis:
»quency
etween
% of the
s performed
he area. The
| (8) and its
satisfied by
4, although
53X, (8)
1ms of the
) and 4.
depths. Six
t is 76% of
ces between
s vary from
l to 0.17 m
's estimated
:quation (8)
sufficiently
p
550
in area B.
among the
1 factor ana-
d therefore,
onent (PC1)
was used. A new simple regression analysis took place. The
independent variable was the natural logarithm of the first
principal component values. The regression was performed on
75 points and the final valid model of 55 points is given by
equation (9):
z = -0.031 + 2.01(InPC1) (9)
X-axis:
Absolute differences
between measured and
estimated depths
Y-axis:
Frequency
Figure 5. The histogramme of absolute differences between
measured and estimated depths (area B). All of the
differences are under 0.4m.
The statistical parameters of the regression models in areas A, B
and C (for the parameters see Mallows, 1973, Myers, 1990,
Stevens, 2002) are presented in table 1.
R| R | R,| DW | VIF Ri C.
094] 088] 087 | 185 | <=1,7 | 0954 | 4
097] 093] 093] 18] <=3.1 | 0965 5
0.88] 0.78] 0.77 | 1.55 | -—— 0.939 2
a= |»
Tablel. The statistical parameters
The model was tested with 178 points of known depths. The 48
of them lie outside the confidence zone while the 130 that is
73% of the total lie inside it (fig.6). The absolute differences
between known depths and estimated depths at these points vary
from 0.04 m to 0.93 m (fig. 7) with a mean value equal to 0.24
m and a standard deviation equal to 0.37 m. The zone's
estimated depths vary from 2.0 m to 5.8 m. According to
statistical parameters and tests a very sufficient performance of
the model was remarked.
predicted z - -0,032*2,.017"PC1
Figure 6. The graphic expression of confidence zone in area C.
The symbolisms are given in figure 2.
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
Absolute differences
between measured and
estimated depths
Y-axis:
Frequency
Figure 7. The histogramme of absolute differences between
measured and estimated depths (area C). 67% of the
differences are under 0.5m.
6. CONCLUSIONS
The linear bathymetric model was applied on the image after the
sun glint removal and atmospheric correction. The image was
integrated with the available echo sounding and GPS data for
the calibration of the model as well as for the analysis of the
corresponding depths in the area of interest. The presence of sea
grass in a part of the study area and the high resolution of the
image affected the linear relationship between water reflectance
and depth and hindered the implementation of the model on the
whole image. Thus, the water area was divided into three parts:
an area with sea grass (depths about 2.0 m to 6.0 m), a mixed
area with sea grass and sand (depths about 2.4 m to 6.0 m) and
a sea grass-free area (depths about 6.0 m to 15.0 m). Bands 1, 2,
3, 4, and 5 of the image were used in the linear model. The
outcomes of the statistical analysis indicated that the model
provided very good results for the mixed and sea grass-free
area, unlike the ‘sea grass’ area where the first principal
component was used instead of the five image bands. In all
areas the majority of the estimated depths (73-76 %), differed
adequately from the soundings. The model in the mixed and the
sea grass-free area was mainly influenced by the green band.
The contribution of the blue band in these two areas was
significant but less than the contribution of the green. The red
band had a significant contribution only in the sea grass-free
area that is in depths >= 6.0 m. The coastal and yellow band
satisfied only the linear model of the mixed area and their
contribution, although it was statistically significant, was very
small. To conclude, the green band proved to be the most
effective for bathymetry applications. The blue band contributed
less while the red band participated only in the sea-grass free
area. In general the bathymetric model involving the imagery
data of high spectral and spatial resolution produced fairly
accurate results. However a thorough statistical analysis was
required to optimize the selection of the appropriate spectral
bands.
7. REFERENCES
Andritsanos V.D., C. Pikridas, D. Rossikopoulos, I.N. Tziavos,
A. Fotiou, 1997. Depth represantation in closed sea areas, la-kes
and rivers using an echo sounder and the global positioning
system (GPS). In Proccedings of the 4th National Carto-
graphic Conference: Cartography and Maps for Success and
Environment Protection, HCS, Kastoria, Greece (in greek).
Bramante J., Raju D. K. and S.T. Min, 2010. Derivation of
bathymetry from multispectral imagery in the highly turbid
waters of Singapore’s south islands. A comparative study.
