You are on page 1of 9

Comput Geosci (2013) 17:661–669

DOI 10.1007/s10596-013-9347-1

ORIGINAL PAPER

Hydrographic data modeling methods for determining


precise seafloor topography
Nedim Onur Aykut · Burak Akpınar · Ömer Aydın

Received: 24 May 2012 / Accepted: 13 March 2013 / Published online: 27 March 2013
© Springer Science+Business Media Dordrecht 2013

Abstract Acoustic and light detection and ranging are 1 Introduction


the recent methods used in hydrographical surveying.
Depth values depending on X and Y horizontal coor- Depth values of hydrographical data which are the base
dinates are measured in both methods. While process- for the digital maps such as electronic navigation charts
ing the hydrographical data, all data with different are determined by using single-beam echo sounder
density are interpolated and modeled for determining (SBES), multibeam echo sounder (MBES), and air-
the seafloor model using different interpolation tech- borne light detection and ranging (LIDAR) methods
niques. In this study, effects of different surface mod- [1]. Depth values are measured with different density
eling methods are investigated. Data obtained from using these methods. Digital elevation models (DEM)
single-beam echo sounder (SBES) are modeled using are used for determining the bathymetric model. For
inverse distance, kriging, local polynomial, minimum determining the model, the most important issue is the
curvature, moving average, nearest-neighbor, and De- selection of interpolation method. The interpolation
launay interpolation methods. Interpolation results are method directly affects the accuracy of DEM.
compared with the multibeam echo sounder data which Yanalak [2] tested the different interpolation meth-
were collected on the same area for determining the ods on five different simulated areas and obtained
accuracy of modeling methods. Depending on the max- the more accurate results using the minimum curva-
imum total vertical uncertainty values in hydrographic ture method. Triangulated irregular network (TIN)-
survey standards, the best results were determined by structured kriging method instead of global kriging
using the kriging method. The Delaunay, minimum method is recommended in [3]. Babak and Deutsch [4]
curvature, and inverse distance methods can be used for stated that the inverse distance interpolation method
modeling the SBES data in shallow waters. is more preferable than kriging; Kholgi and Hosseini
[5] used the kriging method for groundwater level es-
Keywords Seafloor topography · Bathymetric timation. Brouns et al. [6] stated that the most suitable
modeling · Acoustic depth measurement · method for hydrographical studies is the Delaunay tri-
Digital elevation model angulation method.
In this study, SBES and MBES data were collected
on different areas located in the Gulf of Gemlik in the
Sea of Marmara. These areas have coast and shallow
N. O. Aykut (B) · B. Akpınar · Ö. Aydın
Department of Geomatic Engineering, Yıldız Technical
water characteristics. Sea bottom topography of the ar-
University, Davutpaşa, Esenler, 34210 Istanbul, Turkey eas has different characteristics even though they are in
e-mail: noaykut@gmail.com the same region. SBES data were interpolated using the
B. Akpınar inverse distance, kriging, local polynomial, minimum
e-mail: burakpinar@gmail.com curvature, moving average, nearest-neighbor, and De-
Ö. Aydın launay methods. As the results of the interpolations, the
e-mail: oaydin@yildiz.edu.tr obtained digital bathymetric model was compared with
662 Comput Geosci (2013) 17:661–669

MBES data, and the differences between the model and the traditional equipment used on hydrographic sur-
the original MBES data were analyzed. veys worldwide [8]. Furthermore, MBES surveys pro-
duce large data sets where several measurements are
available for every square meter in coastal waters.
2 Acoustic depth measurement methods When real-time processing of MBES data is required,
such as pipeline construction projects, MBES data out-
Acoustic depth measurement systems measure the put density may be decreased while maintaining a re-
elapsed time that an acoustic pulse takes to travel from alistic bathymetric model [10]. The bathymetric model
a generating transducer to the waterway bottom and is defined as “a digital representation of the topogra-
back. If the velocity of sound propagation in the water phy (bathymetry) of the seafloor by coordinates and
column is known, along with the distance between the depths” [11]. Modeling of the terrain surface from
transducer and the reference water surface, the cor- three-dimensional data is called digital terrain model-
rected depth can be computed by the measured travel ing (DTM) [2]. This term is widely used as digital eleva-
time of the pulse which is transmitted and receipted by tion model (DEM) in the USA, digital height model in
the transducers [7]. SBES may have transducers either Germany, and digital ground model in the UK; digital
with a single transducer piece or an array. MBES have terrain elevation model is used and introduced by the
transducer arrays built up from several elements based US Geological Survey [12]. To keep the original survey
on a fan-shaped transmission pulse directed towards points or to reflect the original measurements in DEM,
the seafloor. Several beams are electronically formed, one of the popular solutions is TIN based on linear
using signal processing techniques, with known beam interpolation method [3, 6]. The triangular grid-based
angles [8]. Depth is calculated by the cosine of the beam DEM is generally used for a scattered data pattern,
angle multiplied with the range to the seabed for each while the rectangular grid-based DEM is used for either
transmitted pulse. regular or scattered data patterns. Transformation of
Basic principle of SBES and MBES methods is the data from a regular to a scattered or a scattered
shown in Fig. 1. The differences between the methods to a regular pattern is always possible by the use of
are the measured area and the survey time. Sweep sys- a suitable interpolation algorithm. Transformation of
tems or MBES have better seafloor coverage and less scattered data to regular data (on the corners of regular
survey time. rectangles) is generally known as “gridding,” used for
modeling of seafloor [13].

3 Interpolation methods of bathymetric data 3.1 Nearest-neighbor interpolation

With the development of multibeam sonars and other The nearest-neighbor interpolation is the simplest in-
high-resolution swath mapping systems in concert with terpolation method. The depth of an interpolation
advances in positioning systems and computer process- point is assumed to be equal to the depth of the nearest
ing power, the ability to map the seafloor has funda- reference point. The depth value is determined by aver-
mentally changed [9]. Despite these new technologies, aging the depths of the nearest neighbors weighted by
single-beam echo sounders still remain, for the present, the inverse distances from the position of the gird node

Fig. 1 Same seabed profile


measured by SBES and
MBES methods
Comput Geosci (2013) 17:661–669 663

or constructing a polynomial equation (Eq. 1) where α i


are constants [14].

f (x, y) = ao + a1 x + a2 y + a3 x2 + a4 xy + a5 y2 + · · · .
(1)
Costa et al. [15] used nearest-neighbor algorithm
to resample bathymetric mosaic from 1 × 1 to 5 × 5
LIDAR data in 7.8–53.4 m of water depth.

3.2 Kriging interpolation

Kriging determines the weights wi for the prediction of


a depth zo = w1 z1 + · · · wn zn at the location p0 given
depth observations z1 ,· · · ,zn at the locations p1 ,· · · , pn
and given a covariance function that returns a covari-
ance value as a function of horizontal distance between
the observations [10], named as simple kriging. Or-
dinary kriging is a linear weighted-average technique
which is unbiased in regard to expected value of resid-
uals [5].

Z (xo ) = i=1 wi Z (xi ).
n
(2)
Fig. 2 Experimental region
Pineda et al. [16] conducted a study in the Yucatan
Peninsula region where kriging was the most suitable
method for mapping the spatial distribution of depth For determining the geoid undulation at a specific
values at regional scale with depths at approximately location, statistical methods including polynomial in-
0–3,000 m. Rakhmatullaev et al. [17] used the kriging terpolation and least-squares collocation have usually
algorithm to calculate the reservoir volume from bathy- been employed to interpolate the data [19].
metric data in Akdarya reservoir between 477 and
495 m of depth. Forney et al. [18] said that kriging was
3.4 Inverse distance interpolation
selected as the best method for interpolating the ocean-
ographic data collected during the Southwest Fisheries
Inverse distance interpolation is a nearest-neighbor in-
Science Center surveys. Kriging allows patterns of vari-
terpolation technique that combines both the neighbor-
ability in the data to be used directly through the fitted
hood and the gradual change of the trend surface [16].
variogram model.
An inverse distance weighted interpolation is defined
3.3 Polynomial interpolation

This method is used to express the seafloor surface with


a polynomial function. The expression of a surface with
the dth order degree polynomial is given in Eq. 3 [13].
 
z (x, y) = d
k=0
k i
j=k−i aij x yi . (3)
i=0

Here, aij is the constant of the polynomial, n is the


degree number of the polynomial, and i, j are positive
integer values. In order to determine a second-degree
polynomial, at least six reference points are required,
and the polynomial is expressed in Eq. 4 [13].

z (x, y) = a00 + a01 y + a10 x + a20 x2 + a11 xy + a02 y2 .


(4) Fig. 3 Bathymetric model of area 1
664 Comput Geosci (2013) 17:661–669

Table 1 Statistical values of Inverse Kriging Local Minimum Moving Nearest Delaunay
different modeling methods
distance polynomial curvature average neighbor
on area 1
Maximum 2.206 1.476 3.271 4.088 7.920 6.677 1.955
Minimum −2.424 −1.692 −6.926 −1.950 −28.338 −4.690 −1.466
Mean 0.110 0.096 0.431 0.084 −4.086 0.252 0.217
Std. dev. 0.469 0.270 0.952 0.367 8.072 1.457 0.380

as a spatially weighted average of the sample values utilizes the Delaunay triangulation which is a unique
within a search neighborhood calculated in Eq. 5. triangulation, satisfying the empty circle property
 which means that the circle passing through any three
Z ∗ (u) = i=1 λi Z (ui ) .
n
(5) vertices of a Delaunay triangle does not contain any
other data points in its interior [3]. An important fea-
In Eq. 5, u is the estimation location; ui , (i = 1,..,n)
ture of the Delaunay triangulation is to obtain optimal
are the locations of the sample points within the search
equilateral triangles [13]. The most common interpola-
neighborhood; Z ∗ (u) is the estimated depth value, n
tion method on triangles is linear interpolation and is
is the number of sample points; λi , i = 1,. . . ,n are the
expressed in Eq. 7.
weights assigned to each sample point; and Z ∗ (ui ), i =
1,. . . ,n are the conditioning depth data at sample points.
The weights are determined in Eq. 6 [4]. z = a00 + a10 x + a01 y. (7)
 
1
p
di
λi =    (6) The constants a00 , a10 , and a01 are calculated using
n 1 three corner points of the triangle. The z value for an
i=1 p
di
interpolation point is calculated in Eq. 7 using x and y
di are the Euclidian distances between estimation loca- coordinates.
tion and sample points, and exponent p is the power or If it is required to keep original survey points in
distance exponent value. DTM, the popular solution is a triangulated irregular
network. The Delaunay triangulation algorithms have
3.5 Delaunay triangulation interpolation been used for multibeam echosounder data processing
in the study of Brouns et al. [6]. Most hydrographic
A popular local interpolation technique is the TIN softwares use the Delaunay triangulation method for
based on linear interpolation method. The TIN method interpolating the depth data.

Fig. 4 Standard deviations of


different modeling methods
on area 1
Comput Geosci (2013) 17:661–669 665

method. It is used to interpolate a point by making use


of a number of reference points nearby.
n
i=1 zi
Z0 = (10)
n
where n is the total number of reference points used for
the averaging operation and zi (i ∈ 1, n) is the height of
the ith reference point [21]. This is a simple averaging.
It means that it does not matter how close a reference
is to the interpolation point.

4 Experimental test

In this study, SBES and MBES data were collected on


Fig. 5 Bathymetric model of area 2 three different areas located in the Gulf of Gemlik,
in the Sea of Marmara. Figure 2 shows these areas in
the same region. These areas have coast and shallow
water characteristics. Sea bottom topography of the
3.6 Minimum curvature interpolation areas has different characteristics even though they are
in the same region. These three areas in the same region
A gridding method commonly called minimum curva- were selected for examining the results of the interpo-
ture is widely used in earth sciences for interpolating lation methods in different areas which have different
the gravity values on the corners of a regular grid by characteristics. The first and the second areas have
using scattered gravity data and finally for producing a the similar characteristics of shallow water depth but
contour map. The method interpolates the data to be have different dimensions. The third area was selected
gridded with a surface having continuous second deriv- because of its coast water characteristics.
atives and minimal total squared curvature [13, 20]. The The depth values of these areas are less than 100 m
algorithm of the minimum curvature method is based and 1b degree areas in IHO S44 standards [22] are con-
on the numerical solution of the modified bi-harmonic sidered. SBES data on these areas were interpolated
differential equation expressed in Eq. 8. using the inverse distance, kriging, local polynomial, min-
  imum curvature, moving average, nearest-neighbor,
∂2z ∂2z
C (z) = + 2 dxdy. (8) and Delaunay methods, and the bathymetric model
∂ x2 ∂y of the seafloor was determined. Differences between
The z function which minimizes the total squared SBES data and MBES data on the same area were
curvature must satisfy Eq. 9 where the differential is determined using Eq. 11, and the standard deviation
solved with the use of boundary conditions [13]. of these differences were determined using Eq. 12.
d(i), Z(i), σ , and n values refer to differences, depth
∂4z ∂s4 z ∂4z values, standard deviation, and measurement number,
+ 2 2 2 + 4 = 0. (9) respectively, in Eqs. 11 and 12.
∂x 4 ∂y x ∂y
d (i) = Z (i)MBESData − Z (i)model (11)
3.7 Moving average interpolation
n l − di 2
One of the point-based interpolation methods is mov- i=1 d
σ = (12)
ing averaging, which is normally seen as a smoothing n

Table 2 Statistical values of Inverse Kriging Local Minimum Moving Nearest Delaunay
different modeling methods
distance polynomial curvature average neighbor
on area 2
Maximum 1.891 1.813 2.054 1.771 7.908 2.599 1.944
Minimum −1.564 −1.526 −9.766 −1.711 −7.527 −2.007 −1.626
Mean 0.018 0.004 0.047 −0.006 0.129 0.008 0.020
Std. dev. 0.279 0.272 0.427 0.282 2.577 0.350 0.286
666 Comput Geosci (2013) 17:661–669

Fig. 6 Standard deviations of


different modeling methods
on area 2

The bathymetric model of area 1 (800 m × 400 m) Depth values of 153,872 were used for determin-
determined using MBES data is given in Fig. 3. The ing the differences between MBES data and mod-
mean depth value for area 1 is 37.671 m. eled SBES data on area 3. Statistical values of these
d(i) differences in Eq. 11 were determined by us- differences and standard deviations of different mod-
ing 77929 MBES and modeled SBES data on area 1. eling methods are given in Table 3 and Fig. 8, respec-
Maximum, minimum, mean and standard deviations of tively. Depending on the standard deviations in Fig. 8, it
these differences which were determined by using eight can be said that the best results on area 3 were obtained
different modeling methods are given in Table 1. A by using the kriging and inverse distance methods.
graphical representation of the standard deviation of Overall, results on three different areas are given in
these differences with respect to modeling methods is Fig. 9. Depending on the results in Fig. 9, the kriging
given in Fig. 4. Depending on the standard deviations, it method has the minimum standard deviation on three
can be said that the best results on area 1 were obtained different areas. It can be said that the best results were
by using the kriging, minimum curvature, and Delaunay
methods, respectively.
The bathymetric model of area 2 (2000 m × 700 m)
determined using MBES data is given in Fig. 5. The
mean depth value for area 2 is 55.133 m.
Depth values of 167,605 were used for determin-
ing the differences between MBES data and mod-
eled SBES data on area 2. Statistical values of these
differences are given in Table 2. A graphical represen-
tation of the standard deviation of these differences
with respect to modeling methods is given in Fig. 6.
Depending on the standard deviations, it can be said
that the best results on area 2 were obtained by using
the kriging, inverse distance, minimum curvature, and
Delaunay methods respectively.
The bathymetric model of area 3 (300 m × 300 m)
determined using MBES data is given in Fig. 7. The
mean depth value for area 3 is 33.980 m. Fig. 7 Bathymetric model of area 3
Comput Geosci (2013) 17:661–669 667

Table 3 Statistical values of Inverse Kriging Local Minimum Moving Nearest Delaunay
different modeling methods
distance polynomial curvature average neighbor
on area 3
Maximum 1.286 1.208 1.359 2.132 1.089 2.450 2.723
Minimum −1.624 −1.544 −3.634 −1.993 −10.539 −1.890 −2.032
Mean −0.024 0.002 −0.098 0.061 −2.650 0.020 −0.068
Std. dev. 0.309 0.270 0.474 0.516 2.329 0.471 0.459

Fig. 8 Standard deviations of


different modeling methods
on area 3

Fig. 9 Standard deviations of


different modeling methods
on three different areas

Table 4 The processing time Area Inverse Krigging Local Minimum Moving Nearest Delaunay
(in second) of each
(m2 ) distance polynomial curvature average neighbor
interpolation method
Area 1 320,000 0.44 4.34 0.29 0.05 0.05 0.02 0.03
Area 2 1,400,000 0.36 1.70 0.39 0.06 0.11 0.03 0.08
Area 3 90,000 0.27 1.58 0.33 0.17 0.05 0.03 0.02
668 Comput Geosci (2013) 17:661–669

Fig. 10 Cost of interpolation


methods

determined by using the kriging, minimum curvature, mum curvature, inverse distance, and Delaunay mod-
inverse distance, and Delaunay methods. eling methods. These standards were achieved also on
The cost of each interpolation method was deter- area 2 and area 3 by using the local polynomial and
mined by the time passed during interpolation process, nearest-neighbor methods. It can be said that the mov-
given at the Table 4. Depending on Table 4, kriging ing average modeling method has no sufficient accuracy
gives the best performance for seafloor modeling but for modeling the seafloor in this region.
has the longest processing time. The nearest-neighbor
and Delaunay interpolation methods have the shortest
processing time as it is seen in Fig. 10.
Depending on IHO S44 hydrographical survey stan- 5 Conclusions
dards for 1a and 1b degree areas [11], the maximum
total vertical uncertainty values (Fig. 11) for 10, 20, 40, Despite that the multibeam echosounder systems have
60, 80, and 100 m were determined as 0.52, 0.56, 0.72, 100 % coverage of seafloor, single-beam echo sounders
0.93, 1.15, and 1.39 m, respectively (Fig. 9). still remain at the present as the traditional equipment
By considering Figs. 9 and 10, the hydrographical used on hydrographic surveys worldwide. The accuracy
survey standard for 1a and 1b degree areas given in of the results obtained by SBES method is directly
IHO [11] can be achieved by using the kriging, mini- related to the accuracy of the bathymetric modeling

Fig. 11 Maximum total


vertical uncertainty values for
different depths
Comput Geosci (2013) 17:661–669 669

method. The accuracy of the bathymetric model is 8. International Hydrographic Organization (IHO): Manual on
important for hydrographical maps. Hydrography, 1st edn. Publication M-13, International Hy-
drographic Bureau, Monaco (2005)
In this study, effects of the different interpolation
9. Mayer, L.A., Paton, M., Gee, L., Gardner, J.V., Ware, C.:
methods in different areas especially in shallow waters Interactive 3-D visualization: a tool for seafloor navigation,
were investigated. Depending on the maximum total exploration and engineering. Proc. IEEE Oceans. 2, 913–920
vertical uncertainty values in IHO S44, the best results (2000)
10. Bottelier, P., Briese, C., Hennis, N., Lindenbergh, R., Pfeifer,
were determined by using the kriging method both in
N.: Distinguishing features from outliers in automatic kriging-
the coast and the shallow waters in this region. The based filtering of mbes data: a comparative study. In: Geo-
Delaunay, minimum curvature, and inverse distance statics for Environmental Applications. Springer, Berlin
methods can be used for modeling the SBES data in (2005)
11. International Hydrographic Organization (IHO): IHO Stan-
shallow waters by considering the IHO S44 standards.
dards for Hydrographic Surveys, 5th edn. Special Publication
Depending on the literature search, it can be said that No:44, International Hydrographic Bureau, Monaco (2008)
the results are acceptable when they were compared 12. Ojha, T.P.: Magnetostratigraphy, topography and geology of
with the results of other studies in different regions. the Nepal Himalaya: a GIS and paleomagnetic approach.
Doctor of Philosophy Thesis, The University of Arizona
In general, the kriging method is the most suitable for
(2009)
interpolating the single-beam data. 13. Yanalak, M., Baykal, O.: Digital elevation model based vol-
ume calculations using topographic data. J. Surv. Eng. 129(2),
Acknowledgement The authors wish to acknowledge the assis- 56–64 (2003)
tance of the Turkish Navy Office of Navigation, Hydrography 14. Du, C.: An interpolation method for grid-based terrain mod-
and Oceanography for collecting the SBES and MBES data. eling. Comput. J. 39(10), 837–843 (1996)
15. Costa, B.M., Battista T.A., Pittman S.J.: Comparative evalu-
ation of airborne LiDAR and ship-based multibeam SoNAR
bathymetry and intensity for mapping coral reef ecosystems.
References Remote Sens. Environ. 113, 1082–1100 (2009)
16. Bello-Pineda, J., Hernández-Stefanoni, J.: Comparing the
1. Aykut, N.O.: Using the multibeam echosounder systems performance of two spatial interpolation methods for creat-
in shallow water hydrographic surveying and error budget ing a digital bathymetric model of the Yucatan submerged
analysis. Ph.D Thesis, Yildiz Technical University, Istanbul platform. Pan-Am. J. Aquat. Sci. 2(3), 247–254 (2007)
(2009) 17. Rakhmatullaev, S., Marache, A., Huneau, F., Coustumer, P.,
2. Yanalak, M.: Effect on gridding method on digital terrain Bakiev, M., Motelica-Heino, M.: Geostatistical approach for
model profile data based on scattered data. J. Comput. Civ. the assessment of the water reservoir capacity in arid regions:
Eng. 17(1), 58–67 (2003) a case study of the Akdarya reservoir, Uzbekistan. Environ.
3. Felus, Y.A., Saalfeld, A., Schaffrin, B.: Delaunay triangula- Earth. Sci. (2011) 63, 447–460 (2010)
tion structured kriging for surface interpolation. Surv. Land 18. Forney, K.A., Ferguson, M.C., Becker, E.A., Fiedler, P.C.,
Inf. Sci. 65(1), 27–36 (2009) Redfern, J.V., Barlow, J., Vilchis, I.L., Ballance, L.T.:
4. Babak, O., Deutsch, C.V.: Statistical approach to inverse Habitat-based spatial models of cetacean density in the east-
distance interpolation. Stoch. Environ. Res. Risk Assess. 23, ern Pacific Ocean. Endanger. Species Res. 16, 113–133 (2012)
543–553 (2009) 19. Kavzoglu, T., Saka, M.H.: Modelling local GPS/levelling
5. Kholghi, M., Hosseini, S.M.: Comparison of groundwater geoid undulations using artificial neural networks. J. Geod.
level estimation using neuro-fuzzy and ordinary kriging. En- 78, 520–527 (2005)
viron. Model. Assess. 14, 729–737 (2009) 20. Smith, W.H.F., Wessel, P.: Gridding with continuous curva-
6. Brouns, G., Wulf, A.D., Constales, D.: Delaunay triangula- ture splines in tension. Geophysics 55(3), 293–305 (1990)
tion algorithms useful for multibeam echosounding. J. Surv. 21. Li, Z., Zhu, Q., Gold, C.: Digital Terrain Modelling: Prin-
Eng. 129, 2 (2009) ciples and Methodology. CRC, Florida, ISBN: 0415324629
7. USACE: Engineering and Design Hydrographic Surveying. (2004)
U.S. Army Corps of Engineers Washington. DC 20314–1000, 22. IHO: IHO Standards for Hyrdographic Surveys, 5th edn.
Publication No:1110-2-1003 (2002) Special Publication No:44, Monaco (2008)

You might also like