Académique Documents
Professionnel Documents
Culture Documents
Journal of Hydrology
journal homepage: www.elsevier.com/locate/jhydrol
a r t i c l e
i n f o
Article history:
Received 14 October 2008
Received in revised form 18 March 2009
Accepted 25 March 2009
s u m m a r y
Continuous surface of river bathymetry (bed topography) is typically produced by spatial interpolation of
discrete point or cross-section data. Several interpolation methods that do not account for spatial trend in
river bathymetry produce inaccurate surfaces, thus requiring complex interpolation methods such as
anisotropic kriging. Although isotropic methods are unsuitable for interpolating river bathymetry, issues
that limit their use remain unaddressed. This paper addresses the issue of effect of spatial trend in river
bathymetry on isotropic interpolation methods. It is hypothesized that if the trend is removed from the
data before interpolation, the results from isotropic methods should be comparable with anisotropic
methods. Data from six river reaches in the United States are used to: (i) interpolate bathymetry data
using seven spatial interpolation methods; (ii) separate trend from bathymetry; (iii) interpolate residuals
(bathymetry minus trend) by using all seven interpolation methods to get residual surfaces, (iv) add the
trend back to residual surfaces; and (v) compare resulting surfaces from (iv) with surfaces created in (i).
Quantitative and qualitative comparison of results through root mean square error (RMSE), semi-variograms, and cross-section proles show that signicant improvement (as much as 60% in RMSE) can be
accomplished in spatial interpolation of river bathymetry by separating trend from the data. Although
this paper provides a new simple way for interpolating river bathymetry by using (otherwise deemed
inappropriate) isotropic methods, the choice of trend function and spatial arrangement of discrete
bathymetry data play a vital role in successful implementation of the proposed approach.
2009 Elsevier B.V. All rights reserved.
Introduction
River bathymetry (bed topography) plays a critical role in
numerical modeling of ow hydrodynamics, sediment transport,
ecological and geomorphologic assessments. Conventional way of
measuring river bathymetry is through cross-sectional surveys
where ground proles are collected at certain locations along the
river depending on available resources, river morphology and end
use of the data. A recent technological development in bathymetry
measurement includes the use of boat-mounted SONAR (Sound
Navigation And Ranging) devices such as single or multi-beam
echosounder in conjunction with global positioning system (GPS)
to give a series of (x, y, z) bathymetry points (Vermeyen, 2006;
Rogala, 1999). Although the spatial resolution of bathymetry points
collected through echosounding techniques can be much better
compared to cross-sections, these data still represent a discrete
sample of a continuous bathymetric surface. Continuous mapping
of shallow river bathymetry over large areas through air-borne
techniques is also an active area of research these days (Hilldale
and Raff, 2008; Legleiter et al., 2004; Marcus et al., 2003), but for
deeper rivers discrete eld data are still the best way for creating
* Tel.: +1 765 494 2176; fax: +1 765 494 0395.
E-mail address: vmerwade@purdue.edu
0022-1694/$ - see front matter 2009 Elsevier B.V. All rights reserved.
doi:10.1016/j.jhydrol.2009.03.026
170
nite element mesh). Therefore, the accuracy of bathymetric surfaces represented in 2D/3D models is dependent upon the ability
of interpolation methods in making accurate predictions at unmeasured locations using discrete data. Recent studies have shown that
commonly available interpolation methods such as triangulation,
inverse distance weighting (IDW), splines or kriging, which assume
isotropy in data, yield inaccurate river bathymetric surface (Goff
and Nordfjord, 2004; Merwade et al., 2006). The isotropic assumption in most spatial interpolation methods ignores the trend in river bathymetry that is linear (bed slope) in ow direction and
nonlinear (cross-sectional shape) across ow direction. As a result,
methods that account for river ow direction and topographic
trend are recommended for interpolating discrete bathymetry
data.
Additional constraints imposed by river ow direction and
topographic trend limit the choice of methods available for interpolating river bathymetry to only a few specialized ones such as
anisotropic kriging, or custom modications of existing isotropic
methods (e.g., elliptical IDW by Merwade et al., 2006; anisotropic
IDW by Tomczak, 1998). Although anisotropic methods provide
better results compared to isotropic methods, ways to improve
the results from application of simple isotropic methods for interpolating river bathymetry remain unexplored. Specically the effect of separating existing trends in river bathymetry before
interpolating discrete points is not studied. Such a study can: (a)
potentially overcome the limitations of isotropic methods, thus
making them widely applicable for interpolating river bathymetry;
(b) demonstrate the effect of trend on interpolating river bathymetry; and (c) provide information on accuracy that can be gained by
Fig. 1. Bathymetry dataset for: (a) Brazos (B) River; (b) Ohio (O) River; (c) King Ranch (KR) reach of Guadalupe River; (d) Rainwater (R) reach of Guadalupe River; (e) Sulphur
(S) River; and (f) Kentucky River (K).
171
Brazos
King Ranch
Rainwater
Sulphur
Kentucky
Ohio
River characteristics
Bathymetry data
Length
(km)
Mean width
(m)+
Slope
Substrate type
Total points
Points/50
50 m2 area
Mean
(m)
Std. Dev.
(m)
Min.
(m)
Max.
(m)
7.1
1.6
3.2
1.4
7.2
30
105
42
44
33
106
465
3.1
1.4
1.74
4.22
7.98
4.48
0.03
0.01
0.08
0.04
0.02
0.001
Sandy
Gravel/bedrock
Gravel/bedrock
Silty/Sandy clay
Gravel/bedrock
Cohesive/noncohesive sediment
37288
7602
14955
7732
66682
20554
116
274
243
407
224
4
10.42
143.46
144.8
72.06
142.8
111.97
1.43
0.64
1.16
1.48
2.46
1.81
0.11
141.66
140.85
68.9
138.11
106.99
13.25
144.96
146.48
76.3
150.8
116.45
along cross-sections, alternate cross-sections are selected for validation. Overall, all datasets except Kentucky River are split such
that 70% of bathymetry points are included in testing dataset and
30% in validation dataset. For Kentucky River, the testing and validation datasets are split to have 50% points each (alternate crosssections).
Mapping of data in (s, n) coordinate system
River bathymetry data are collected and stored using Cartesian
(x, y) coordinates, but use of these data in this coordinate system
can introduce issues related to meandering nature of the channel.
For example, computing distance between two points along a river
using (x, y) coordinates will not give the actual ow length between
these points for a meandering river. To overcome such issues,
channel-tted coordinate system dened by an s-axis along the
ow direction and n-axis across the ow direction (perpendicular
to s-axis) is widely used for rivers (e.g., Johannesson and Parker,
1989; Nelson and Dungan, 1989; Ye and McCorquodale, 1997).
The s-axis can be aligned with either river banks or centerline. In
this study, the geometric centerline of the channel is treated as
the s-axis with upstream end of the river as its origin (s = 0). Similarly, looking downstream, all n coordinates are negative on the
left hand side of s-axis, and vice versa. Plotting of river in (s, n)
coordinates straightens the river so data can be treated with respect to the ow direction as shown in Fig. 2. A GIS procedure
developed by (Merwade et al., 2005) is used for mapping bathymetry in (s, n) coordinates for all six reaches.
Fitting trend to bathymetry points
Mapping of bathymetry in (s, n) coordinates makes the linear
trend (bed slope) a function of s coordinate, and nonlinear trend
(cross-sectional shape) a function of n coordinate. Besides using
the n coordinate, the nonlinear trend is modeled by using other
physical attributes such as channel width, depth and meandering
curvature (see e.g., Deutsch and Wang, 1996; James, 1996;
Legleiter and Kyriakidis, 2008). It is assumed that the quality of
trend surface, and eventually variance in residuals (measurements
trend) will depend on the selected trend function. In addition,
whether a trend function is applied locally (separate function
formulations for individual local areas) or globally (one function
for the entire reach) can also produce different results. Several
techniques such as power functions, polynomials, splines and
probability density function (pdf) can be used to t a trend to
rivers cross-sectional shape. A detailed review of these techniques
including their application in tting river cross-sections can be
found in Merwade and Maidment (2004).
To assess the effect of trend function in spatial interpolation,
three techniques (two local and one global) are employed in this
study. Local techniques include cubic polynomial and a combination of two beta pdf (Eq. (1)), and the global technique includes a
cubic polynomial function.
172
n
s
-n
(a)
(b)
Fig. 2. (s, n) Coordinate transformation for King Ranch dataset. Bathymetry points in: (a) (x, y) coordinates and (b) (s, n) coordinates.
Tz ff xja1 ; b1 f xja2 ; b2 g k w;
where, T(z) is the trend function, w is the channel width, k (0 < k < 1)
is a scaling factor, and a1, a2, b1, and b2 are the parameters of beta
pdf given by Eq. (2)
f xja; b
1
xa1 1 xb1 ; 0 < x < 1a > 0;
Ba; b
b>0
Cubic polynomial is selected for its simplicity and easy availability in many software programs for tting spatial trends. A combination of beta pdf is used because comparison of several
techniques listed earlier for tting cross-sectional trend has shown
that a combination of two beta pdf produce best results (Merwade
and Maidment, 2004). The objective is not to advocate any one
technique over others for tting trends, but to see how different
techniques affect spatial interpolation results of bathymetry
points. For local polynomial trend tting, a circular search neighborhood is dened to include 50 points, and for beta pdf, the neighborhood is dened to include points covering a channel length
equal to the average channel width. The neighborhoods for polynomial and beta pdf are dened (by trial and error) such that the
points included in a neighborhood should be able to dene a complete river cross-section. All parameters in cubic polynomials and
beta functions are optimized to minimize the root mean square error (RMSE, Eq. (5)) between predictions and observations.
Spatial interpolation
Spatial interpolation is preformed for each training dataset in
(s, n) coordinates for two variables: zi and ei where, zi is measured
bathymetry at any point i, and ei ( = zi T) is corresponding residual after tting the trend function. Seven spatial interpolation
techniques are used in this study. These are IDW, regularized
spline (RS), spline with tension (TS), topogrid (TG), natural neighbor (NN), ordinary kriging (OK) and ordinary kriging with anisotropy (AK). A review of these techniques for interpolating
watershed topography and river bathymetry can be found in Chaplot et al. (2006) and Merwade et al. (2006), respectively, but a
brief description and corresponding equations of each technique
are presented in Table 2. All these techniques are commonly used
in many disciplines including hydrology through several commercial and public domain software programs.
Assessment of trend surfaces
Trend surfaces are assessed for similarity and their ability to
best describe the trend in the data. Such an assessment will help
to understand how much effect a trend surface can have on nal
interpolated bathymetry surface (trend + interpolated residual).
Assessment of trend surfaces is performed by using percentage of
Table 2
Spatial interpolation methods.
Method
Description
Equations
IDW
i
z RN
i1 ki zi ; ki RN
Tension spline
1
d2
1
i1 d2
i
RR
y
2
2
fx2 fy2 fxx
fxy
fyy
dxdy;
Regularized spline
A minimum curvature (smoothest) function [I(f)] is dened that passes through a set
of observations. R = 2D euclidian space; T(x, y) = local trend; R(r,rj) = basis function;
/ = tension weight; K0 = modied Bessel function of zero order; c = 0.577215
Third and higher order derivatives are added to I(f)
Topogrid
Natural neighbor
Similar to IDW, but weights are computed based on area. pi and qi are the Thiessen
areas surrounding zi excluding and including z*, respectively
ki pi qj =pi
Ordinary kriging
Similar to IDW, but weights are assigned based on distance and spatial correlation.
The objective is to minimize r2 to get unbiased estimate with minimum variance
Ordinary kriging that takes data anisotropy into account
Anisotropic kriging
If
R2 /
f x; y RN
i1 t i Rr; r i Tx; y;
h
i
d/
Rr; r j 2p1/2 ln 2j c K 0 dj /
RNi1 ki zi
173
Pn
zi ^zi
SST 1 Pi1
;
ni1 zi z2
where z is the mean. A semi-variogram plot is a plot of semivariance (Eq. (4)) as a function of distance between observation
points as shown in Fig. 5. In spatially correlated data such as river
bathymetry, semi-variance is smaller for nearer points, and vice
versa. Therefore, semi-variance increases with distance between
measurement points, until a threshold is reached in the distance
of separation. This threshold is called the range as shown in
Fig. 5b. The semi-variance corresponding to the range is called the
sill, and semi-variance corresponding to zero separation distance
on a semi-variogram plot is called the nugget. Semi-variogram can
be used to compare the spatial distribution of semi-variance captured by different trend and interpolated surfaces. Moreover, tting
of semi-variogram in different directions will show the extent to
which the trend in the data is captured by a trend-tting function.
For example, if the trend in the bathymetry is captured, the semivariograms of residuals in all direction should be identical.
cij h
Ezi zj 2
;
2
"
RMSE
n
1X
zi ^zi 2
n i1
#1=2
;
Results
Assessment of trend surfaces
To keep the terminology simple, beta trend surface will be referred as BT, local polynomial trend as LT and global polynomial
trend as GT. Application of traditional techniques such as Students
t-test and SST for comparing trend surfaces show some interesting
results (Tables 3 and 4). For example, the mean of BT, LT and GT are
not signicantly different for Kentucky, the mean for BT and LT are
Table 3
Students t-test results (a = 0.05) between beta (BT), local polynomial (LT) and global
polynomial (GT) trend surfaces for all six reaches (B = Brazos; KR = King Ranch;
RW = Rainwater; S = Sulphur; K = Kentucky; and O = Ohio).
BT
BT
LT
GT
K, KR
R, S, K
LT
GT
K, KR
R, S, K
B, KR, R, K, O
B, KR, R, K, O
Table 4
SSE results for beta (BT), local polynomial (LT) and global polynomial (GT) trend
surfaces for all six reaches.
Brazos
King Ranch
Rainwater
Sulphur
Kentucky
Ohio
BT
LT
GT
82.05
79.61
90.23
80.22
94.48
79.52
83.85
75.77
84.82
82.95
87.46
81.59
44.89
19.41
60.08
50.15
56.75
51.62
174
Fig. 3. Trend surfaces for King Ranch Reach: (a) bathymetry surface for the entire reach; (b) close-up view of the surface in (a); (c) trend surface using beta function; (d) trend
surface using local polynomial; and (e) trend surface using global polynomial.
145.0
144.5
Z
BT
LT
GT
144.5
BT
LT
GT
Elevation (m)
144.0
144.0
143.5
143.5
143.0
143.0
142.5
142.5
142.0
0
10
20
30
40
50
10
20
30
40
50
Distance (m)
a)
b)
Fig. 4. Cross-sections for King Ranch Reach at: (a) x location in Fig. 3b; (b) y location in Fig. 3b (Z represents the base surface; BT = beta trend; LT = local polynomial; and
GT = global polynomial).
Similarly semi-variograms of residuals show that BT and LT capture about 75% of semi-variance in the bathymetry (residual sill
is approximately 0.1 compared to 0.4 for bathymetry); whereas
residuals from GT produce a semi-variogram that looks similar to
observed bathymetry in terms of range and sill, suggesting that
no or very little variance is captured. Similarity of residual semi-
variograms from BT and LT in omni-direction and s-direction suggest that most of the trend is captured by these functions. Although
overall results from King Ranch semi-variograms are applicable to
all datasets, semi-variograms from Brazos that had relatively better GT surfaces is also presented for comparison (Fig. 6). Unlike
other datasets that had similar semi-variograms for BT and LT,
175
0.5
0.5
Z
BT
LT
GT
0.4
0.3
0.3
0.2
0.2
0.1
0.1
LT
GT
Range
Nugget
Sill
0.4
BT
0.0
0.0
0
50
100
150
200
50
100
a)
150
200
b)
0.5
0.5
BT
LT
BT
GT
0.4
0.4
0.3
0.3
0.2
0.2
0.1
0.1
0.0
0
50
100
150
200
LT
GT
0.0
0
50
100
c)
150
200
d)
Fig. 5. Semi-variograms of bathymetry, trend and residual surfaces for King Ranch: (a) bathymetry and trend surfaces in all directions; (b) bathymetry and trend surfaces in
s-direction; (c) residuals in all directions; and (d) residuals in s-direction (Z represents the base surface; BT = beta trend; LT = local polynomial; and GT = global polynomial).
2.5
2.5
2.0
2.0
1.5
1.5
1.0
1.0
0.5
0.5
BT
LT
Z
LT
300
GT
0.0
0.0
100
200
300
400
500
100
200
a)
400
BT
GT
500
b)
1.8
1.8
1.5
1.5
1.2
1.2
0.9
0.9
0.6
0.6
0.3
0.3
BT
LT
BT
GT
0.0
0
100
200
c)
300
400
500
LT
GT
300
400
0.0
0
100
200
500
d)
Fig. 6. Semi-variograms of bathymetry, trend and residual surfaces for Brazos River: (a) bathymetry and trend surfaces in all directions; (b) bathymetry and trend surfaces in
s-direction; (c) residuals in all directions; and (d) residuals in s-direction (Z represents the base surface; BT = beta trend; LT = local polynomial; and GT = global polynomial).
176
0.6
0.6
0.5
0.5
0.4
0.4
0.3
0.3
0.2
0.2
0.1
0.1
0.0
0.0
IDW
NN
RS
TS
TG
OK
AK
IDW
NN
Brazos
RS
TS
TG
OK
AK
King Ranch
0.7
0.8
0.6
0.7
0.5
0.6
0.5
0.4
0.4
0.3
0.3
0.2
0.2
0.1
0.1
0.0
0.0
IDW
NN
RS
TS
TG
OK
AK
IDW
NN
Rainwater
RS
TS
TG
OK
AK
TG
OK
AK
Sulphur
1.0
3.0
2.5
0.8
2.0
0.6
1.5
0.4
1.0
0.2
0.5
0.0
0.0
IDW
NN
RS
TS
TG
Kentucky
OK
AK
IDW
NN
RS
TS
Ohio
LT + interpolated residuals
BT + interpolated residuals
GT + interpolated residuals
Fig. 7. RMSE (m) results for spatial interpolation techniques (BT = beta trend; LT = local polynomial; and GT = global polynomial).
Brazos
King Ranch
Rainwater
Sulphur
Kentucky
Ohio
(a)
(b)
(c)
(d)
0.41
0.40
0.45
0.52
1.83
0.54
0.32 (22)
0.34(17)
0.41 (10)
0.49 (6)
0.73 (60)
0.39 (28)
0.36(14)
0.36(11)
0.41 (10)
0.40 (25)
1.18(35)
0.42 (22)
0.38 (9)
0.37 (9)
0.43 (6)
0.41 (22)
1.33 (27)
0.45(16)
Table 6
Mean percentage reduction in RMSE for each technique compared to interpolation
including trend: (b) BT + interpolation of corresponding residuals; (c) LT + interpolation of corresponding residuals; and (d) GT + interpolation of corresponding residuals.
IDW
Natural neighbor
Regular spline
Tension spline
Topogrid
Ordinary kriging
Anisotropic kriging
(b)
(c)
(d)
39.22
13.02
9.85
20.41
24.74
23.91
17.12
30.52
13.00
6.07
19.29
17.74
17.18
20.33
23.39
10.94
3.03
15.59
13.28
14.01
14.27
177
Fig. 8. Interpolated surfaces for area surrounding Y in Fig. 3b for King Ranch: (a) Base surface; (b) IDW of z including trend; (c) beta trend + IDW residuals; (d) local
polynomial + IDW residuals; (e) global polynomial + IDW residuals; and (f) natural neighbor interpolation of z including trend.
178
Elevation (m)
145.0
a
144.0
143.0
142.0
10
20
30
40
50
Distance (m)
Fig. 9. Cross-section at location X for surfaces in Fig 8: (a) Base surface; (b) IDW of z
including trend; (c) beta + IDW residuals; (d) local polynomial + IDW residuals; and
(e) global polynomial + IDW residuals.
Fig. 10. Interpolated bathymetry surfaces with natural neighbor for Brazos and Kentucky: (a)/(d) Base surface; (b)/(c) interpolation including trend; and (c)/(f) interpolation
of residuals + trend.
179
12
150
(a)
(b)
(c)
(d)
(e)
147
11
(f)
10
144
141
138
8
0
20
40
60
80
Brazos
20
40
60
80
100
Kentucky
Fig. 11. Cross-section proles at location P and Q for Brazos and Kentucky, respectively for surfaces in Fig. 10: (a)/(d) Base surface; (b)/(c) interpolation including trend; and
(c)/(f) interpolation of residuals + trend.
as sensitive to bathymetric trend) compared to other methods produced identical surfaces for all techniques (including/excluding
trends) for King Ranch. Although this may suggest that a robust
technique is unaffected by how the interpolation is performed
(including/excluding trends), this is not true for all datasets as described in the next paragraph. Only one surface created by natural
neighbor interpolation that includes trend is presented in Fig. 8f for
King Ranch whose IDW counterpart is 8b. The difference in surfaces is directly reected in cross-section proles presented in
Fig. 9.
The robustness of natural neighbor as seen for the King Ranch
dataset is not applicable for all datasets. For example, cross-sections from interpolated surfaces including and excluding trends
using natural neighbor interpolation for Brazos and Kentucky show
signicant differences in results (Figs. 10 and 11). The surfaces created by excluding trend are smoother compared to surfaces created by interpolation including trend. The overall representation
of bathymetric surface created by excluding trend is more representative of measured bathymetry compared to surfaces created
by interpolation of points including trend. The difference in nal
results is more distinct and considerably better for the Kentucky
dataset that includes sparsely located bathymetry points along
cross-sections.
Discussion and conclusions
There is a growing need to obtain accurate bathymetry data for
use in multi-dimensional river models (e.g., Dutta et al., 2007; Shen
and Diplas, 2008; Yue et al., 2008). Typically, river bathymetry data
are collected as cross-sections or discrete points, which are interpolated to get elevation at computational nodes in river models.
Because of the presence of spatial trends in river bathymetry, commonly used isotropic interpolation methods that do not account
for spatial trends yield inaccurate results. The objective of this
study was to use discrete bathymetry data from six study areas
to investigate the effect of trend on spatial interpolation methods
for creating continuous surfaces. Although recent studies have
advocated the use of anisotropic interpolation techniques (e.g.,
Merwade et al., 2006; Tomczak, 1998), the present study shows
that isotropic methods can yield improved results if the trend in
the data is excluded during spatial interpolation. The trend from
the bathymetry data should be excluded at two levels. First, the
data should be mapped using ow oriented (s, n) coordinates to remove the spatial heterogeneity in the orientation of river slope
with respect to Cartesian coordinates. Second, the cross-sectional
trend must be removed from the bathymetry by using a nonlinear
function. After removing the trend, the residuals can be interpolated to create a surface to which the trend is added back to get
180
References
Booker, D.J., 2003. Hydraulic modelling of sh habitat in urban rivers during high
ows. Hydrological Processes 17 (3), 577599.
Buttner, O., 2007. The inuence of topographic and mesh resolution in 2D
hydrodynamic modelling for oodplains and urban areas. European
Geosciences Union 2007 Geophysical Research Abstracts 9 (08232).
Carrivick, J.L., 2006. Application of 2D hydrodynamic modelling to high-magnitude
outburst oods: an example from Kverkfjoll, Iceland. Journal of Hydrology 321
(14), 187199.
Chaplot, V. et al., 2006. Accuracy of interpolation techniques for the derivation of
digital elevation models in relation to landform types and data density.
Geomorphology 77 (12), 126141.
Crawford, C.A.G., Hergert, G.W., 1997. Incorporating spatial trends and anisotropy in
geostatistical mapping of soil properties. Soil Science Society of America Journal
66 (1), 298309.
Crowder, D.W., Diplas, P., 2000. Using two-dimensional hydrodynamic models at
scales of ecological importance. Journal of Hydrology 230 (34), 172191.
de Jalon, D.G., Gortazar, J., 2007. Evaluation of instream habitat enhancement
options using sh habitat simulations: case-studies in the River Pas (Spain).
Aquatic Ecology 41 (3), 461474.
Deutsch, C.V., Wang, L.B., 1996. Hierarchical object-based stochastic modeling of
uvial reservoirs. Mathematical Geology 28 (7), 857880.
Dutta, D., Alam, J., Umeda, K., Hayashi, M., Hironaka, S., 2007. A two-dimensional
hydrodynamic model for ood inundation simulation: a case study in the lower
Mekong river basin. Hydrological Processes 21 (9), 12231237.
Gard, M., 2005. Variability in ow-habitat relationships as a function of transect
number for PHABSIM modelling. River Research and Applications 21 (9), 1013
1019.
Goff, J.A., Nordfjord, S., 2004. Interpolation of uvial morphology using channeloriented coordinate transformation: a case study from the New Jersey shelf.
Mathematical Geology 36 (6), 643658.
Hilldale, R.C., Raff, D., 2008. Assessing the ability of airborne LiDAR to map river
bathymetry. Earth Surface Processes and Landforms 33 (5), 773783.
Horritt, M.S., Bates, P.D., Mattinson, M.J., 2006. Effects of mesh resolution and
topographic representation in 2D nite volume models of shallow water uvial
ow. Journal of Hydrology 329 (12), 306314.
Hutchinson, M.F., 1989. A new procedure for gridding elevation and stream line
data with automatic removal of spurious pits. Journal of Hydrology 106, 211
232.
Hutchinson, M.F., 1993. Development of a continent-wide DEM with applications to
terrain and climate analysis. In: Goodchild, M.F. et al. (Eds.), Environmental
Modeling with GIS. Oxford University Press, New York, pp. 392399.
James, L.A., 1996. Polynomial and power functions for glacial valley cross-section
morphology. Earth Surface Processes and Landforms 21 (5), 413432.
Johannesson, H., Parkerm, G., 1989. River Meandering. In: Ikeda Syunsuke, Parker
Gary (Ed.). Water Resources Monograph, vol. 12, Washington, DC, pp. 181
213.
Khosronejad, A., Rennie, C.D., Neyshabouri, S.A.A.S., Townsend, R.D., 2007. 3D
numerical modeling of ow and sediment transport in laboratory channel
bends. Journal of Hydraulic Engineering-ASCE 133 (10), 11231134.
Lampe, D.C., Morlock, S.E., 2007. Collection of bathymetric data along two reaches of
the Lost River within Bluespring Cavern near Bedford, Lawrence County,
Indiana, July 2007. US G.S. Karst Interest Group Proceedings, Bowling Green,
Kentucky, May 2729, 2008: US G. S. Scientic Investigations Report 20085023, 142 p. <http://pubs.water.usgs.gov/sir>2008-5023.
Lane, S.N., Hardy, R.J., Elliott, L., Ingham, D.B., 2002. High-resolution numerical
modelling of three-dimensional ows over complex river bed topography.
Hydrological Processes 16 (11), 22612272.
Lee, K.T., Ho, Y.H., Chyan, Y.J., 2006. Bridge blockage and overbank ow simulations
using HEC-RAS in the Keelung River during the 2001 Nari Typhoon. Journal of
Hydraulic Engineering-ASCE 132 (3), 319323.
Legleiter, C.J., Kyriakidis, P.C., 2008. Spatial prediction of river channel topography
by kriging. Earth Surface Processes and Landforms 33 (6), 841867.
Legleiter, C.J., Roberts, D.A., Marcus, W.A., Fonstad, M.A., 2004. Passive optical
remote sensing of river channel morphology and in-stream habitat: physical
basis and feasibility. Remote Sensing of Environment 93 (4), 493510.
Marcus, W.A., Legleiter, C.J., Aspinall, R.J., Boardman, J.W., Crabtree, R.L., 2003. High
spatial resolution hyperspectral mapping of in-stream habitats, depths, and
woody debris in mountain streams. Geomorphology 55 (14), 363380.
Martin, Y., 2003. Evaluation of bed load transport formulae using eld
evidence from the Vedder River, British Columbia. Geomorphology 53 (12),
7595.
Mekonnen, M.A., Dargahi, B., 2007. Three dimensional numerical modelling of ow
and sediment transport in rivers. International Journal of Sediment Research 22
(3), 188198.
Merwade, V.M., Maidment, D. R., 2004. geospatial description of river channels in
three dimensions. University of Texas at Austin Center for Research in Water
Resources Online Report 04-8. <http://www.crwr.utexas.edu/reports/2004/
rpt04-8.shtml>.
Merwade, V.M., Maidment, D.R., Hodges, B.R., 2005. Geospatial representation of
river channels. Journal of Hydrologic Engineering 10 (3), 243251.
Merwade, V.M., Maidment, D.R., Goff, J.A., 2006. Anisotropic considerations while
interpolating river channel bathymetry. Journal of Hydrology 331 (3-4), 731
741.
181