Floodplain Heterogeneity and Meander Migration

Floodplain heterogeneity and meander migration

MOTTA Davide Department of Civil and Environmental Engineering University of Illinois at Urbana-Champaign, Urbana, Illinois, USA E-mail: [email protected]

ABAD Jorge D. Department of Civil and Environmental Engineering University of Pittsburgh, Pittsburgh, Pennsylvania, USA E-mail: [email protected]

LANGENDOEN Eddy J. US Department of Agriculture, Agricultural Research Service National Sedimentation Laboratory, Oxford, Mississippi, USA E-mail: [email protected]

GARCIA Marcelo H. Department of Civil and Environmental Engineering University of Illinois at Urbana-Champaign, Urbana, Illinois, USA E-mail: [email protected]

ABSTRACT: The impact of horizontal heterogeneity of floodplain soils on rates and patterns of meander migration is analyzed with a Ikeda et al. (1981)-type model for hydrodynamics and bed morphodynamics, coupled with a physically-based bank erosion model according to the approach developed by Motta et al. (2011). We assume that rates of migration are determined by the resistance to hydraulic erosion of the soils, which is described by an excess shear stress relation. This relation uses two parameters characterizing the resistance to erosion: critical shear stress and erodibility coefficient. The spatial distribution of critical shear stress in the floodplain is generated on a regular grid with varying degree of randomness to mimic natural settings and the corresponding erodibility coefficient is computed with a relation derived from field-measured pairs of critical shear stress and erodibility. Centerline migration and associated statistics for randomly-disturbed distribution based on the distance from the valley axis are compared for sine-generated centerline using the Monte Carlo method. Two relevant parameters are identified: (i) the standard deviation of the critical shear stress distribution, which is an indicator of the local soil heterogeneity, determines centerline skewness and its variability, and can lead to development of downstream skewness and complex planform features, while not significantly affecting lateral migration; (ii) the cross-valley increase in soil resistance, which mainly constrains rates of lateral migration also affecting skewness. The Monte Carlo approach, applied this time to the case of a natural river alignment and purely random floodplain-soil distribution, shows that migrated centerlines present larger variability the coarser is the scale of the floodplain heterogeneity (third key parameter for describing the effect of floodplain heterogeneity on migration) and the increase in variability with the heterogeneity scale is less then linear. Finally, when the approach for meander migration based on hydraulic erosion is compared to the “classic” approach based on excess velocity at the outer bank, the former approach produces more variability and shape complexity for equal stochastic variability of the corresponding governing parameters, because of the presence of a threshold for bank erosion missing in

the case of classic approach for migration.

1 INTRODUCTION

A great quantity of two-dimensional (2D) depth-averaged analytical models have focused on the impact of hydrodynamics and bed morphodynamics on the migration of meandering rivers (Ikeda et al., 1981; Blondeaux and Seminara, 1985; Johannesson and Parker, 1989; Sun et al., 2001a; Zolezzi and Seminara, 2001) and on the degree of planform-shape complexity that can be obtained (Seminara et al., 2001; Frascati and Lanzoni, 2009). Less attention has been given to the complexity resulting from the heterogeneity of the floodplain soils, the quantification of its importance, and the identification of the important parameters. Past studies have been all based on the concept of rate of bank erosion linearly proportional, through a migration coefficient, to near-bank excess velocity (Howard, 1992, 1996; Sun et al., 1996, 2001b; Posner and Duan, 2011), according to the method independently introduced by Hasegawa (1977) and Ikeda et al. (1981). Howard (1992), Howard (1996), and Sun et al. (1996) studied the development of heterogeneities in the floodplain caused by river migration and their influence on the future development of the channel. In particular, Sun et al. (1996) focused on the change in erodibility associated to different sedimentary environments such as point bar deposits, floodplain deposits, and oxbow lake deposits, and were able to reproduce the formation of meander belts for long-term simulations. Sun et al. (2001b) went a step further and simulated the development of size distribution as channel migrates. Posner and Duan (2011) compared a deterministic model, adopting a constant migration coefficient, with a stochastic model, where the instantaneous migration coefficient at each bank location is a random variable satisfying either uniform or normal distribution, and compared the two approaches against experimental data, finding that the stochastic approach yields more realistic predictions of meandering-planform evolution. In terms of planform shapes, Sun et al. (1996) claimed that the typical meander wavelength is determined mainly by hydraulic factors (flow in the channel and inclination of the underlying floodplain) and is independent of the difference in the erodibilities of sedimentary deposits. Perucca et al. (2007), who focused on the role played by vegetation on meandering river morphodynamics and the feedback between riparian vegetation dynamics and meandering dynamics, showed the possible occurrence of peculiar meander shapes that do not show the usual marked upstream skewness, when the bank erodibility is linked to the biomass density. In this paper we discuss how soil floodplain distribution modulates planform shapes and rates of migration. The model by Motta et al. (2011) is used, which allows for focusing on the horizontal heterogeneity of the physical parameters for hydraulic erosion, which are erodibility and critical shear stress. Motta et al. (2011)'s model quantifies the migration of meandering streams through a physically- and process-based method that relates channel migration to the streambank erosion processes of hydraulic erosion and mass failure. Hence, channel migration depends on measurable soil properties, natural bank geometry, and both vertical and horizontal heterogeneity of floodplain soils. In general, the processes responsible for bank retreat are classified as follows: hydraulic (fluvial) erosion, and cantilever, planar, rotational, and piping streambank failure (Pizzuto and ASCE Task Committee on Hydraulics, Bank Mechanics, and Modeling of River Adjustment, 2008). As shown by Constantine et al. (2009), using field measurements, and by Motta et al. (2011), through modeling, mass failure mechanisms like planar failures, although important, can be indirectly represented by modifying the parameters for hydraulic erosion to quantify rates of migration, at least for the case of vertical homogeneity of the bank material. Hydraulic erosion requires that the local boundary shear stress exceeds the critical value to initiate motion of the soil particles. From a modeling perspective, the rate of lateral hydraulic erosion E* (with dimensions of length over time) for each bank-material layer is modeled using an excess shear stress relation, typically used for fine-grained materials (but also applicable to non-cohesive materials)

******* EM=(ττcc−1) = k( τ− τ) (1)

* * where M is the erosion-rate coefficient (with dimensions of length over time), τ c is the critical shear * * * * * stress, k is the erodibility, and τ is the shear stress acting on the bank. M , τ c, and k are all site-specific. The critical erosional strength denotes the cohesion strength provided by interparticle forces of attraction or repulsion acting at the microscopic level, including electrostatic forces, Van der Waals forces,

hydration forces, and biological forces (Papanicolaou et al., 2007). Although Eq. 1 appears simple, in practice it is necessary to define the erodibility parameters and the boundary shear stress τ*. These are all highly variable, which explains why observed rates of fluvial erosion range over several orders of magnitude (Darby et al., 2007). * * * Besides the complexity related to the experimental quantification of the parameters τ c and k (or M ), there is that of their spatial distribution in the floodplain, which depends on the soil distribution that is the result of the development of the fluvial system in time, through processes of channel deposition because of lateral migration and overbank deposition formed by vertical accretion. Anomalies in the modern floodplain can be the remnant of the evolution over geologic time scales. Therefore, depending on when the computer modeling of meander migration begins, the spatial distribution of soil erodibility may vary. However, a general characterization of a typical modern floodplain can be made, with a general decrease of sediment size and consequent decrease in erodibility away from the channel. It is also recognized that cutoff deposits generated by a meandering river tend to confine it to a strip in the flood basin, known as meander belt (Allen, 1965; Sun et al., 1996). As regards the shear stress acting on the river bank, it is the sum of the stress resulting from drag on bedforms and the stress acting on the actual boundary (Constantine et al., 2009). Therefore, in Eq. 1, τ* has to be interpreted as effective shear stress. The present paper, after briefly illustrating the meander-migration model and the numerical techniques for the representation of floodplain-soil distribution and the generation of the erodibility pattern, evaluates the impact of horizontal heterogeneity of floodplain soils on meander-migration rates and patterns, using a stochastic approach for comparing configurations characterized by different degrees of randomness and spatial scale of variation.

2 MODEL OF MEANDER MIGRATION

The RVR Meander platform (Motta et al., 2011) merges and extends the functionalities of the first version of RVR Meander (Abad and Garcia, 2006) and CONCEPTS (Langendoen and Alonso, 2008; Langendoen and Simon, 2008; Langendoen et al., 2009), and is composed by a hydrodynamics/bed morphodynamics module, a bank erosion module, and a migration module. Hydrodynamics and bed morphodynamics are computed according to the linear model illustrated by Garcia et al. (1994), which is a slightly modified version of Ikeda et al. (1981)’s and Johannesson and Parker (1985)’s models. It provides an analytical solution of the 2D St. Venant Equation, written in streamwise and transverse coordinates

*** ** 1 **∂∂UUC ** gH ∂τ UV++ UV =− −s (2) 111+nC**∂∂ s * n *+ nC ** + nC **∂ s *ρ D * *** ** 1 **∂∂VVC *2 ∂ Hτ UV+− Ug=−−n (3) 11+nC**∂∂ s * n *+ nC **∂ n *ρ D * UD** VD ** * 1 ∂∂( ) ( ) C ** ++VD =0 (4) 11+nC**∂ s * ∂ n *+ nC **

Superscript star indicates values with dimensions, s* and n* are streamwise and transverse coordinates, U* and V* are streamwise and transverse depth-averaged velocities, g is the acceleration of gravity, H* is * * * water stage, τ s and τ n are streamwise and transverse bed shear stresses, ρ is water density, D is water depth, and C* is the local curvature. The bed morphology is described assuming that the transversal bed slope is proportional to the local curvature through a constant of proportionality A named scour factor (Ikeda et al. (1981)’s assumption)

* η1 ** * =−AC n (5) Dch

* where η 1 is the bed elevation deviation from the bed elevation of the channel centerline at a particular * cross section and D ch is the reach-averaged water depth for uniform flow. An analytical solution of the linearized Eqs. 2-4, for low curvature and constant discharge Q and channel width 2B*, is obtained for

streamwise and transverse velocity and water depth, expressed as sum of their uniform-flow reach-averaged value (subscript “ch”) plus a perturbation (subscript “1”), which varies in both streamwise and transverse directions. The solution depends on upstream and local curvature, sinuosity (which affects * * the slope of the channel), half-width to depth ratio β = B /D ch, reach-averaged Froude number Fch and friction coefficient Cf,ch, and scour factor A. We report here just the solution for streamwise depth-averaged flow velocity

Usn( ,1) =+ U1 ( sn ,) (6) s −−as'''222 as as Usna11( ,') =++( ne) naCs ' 34( ) ae ' Cseds( ) (7) ( ∫0 ) and streamwise bed shear stress

***2*2 τρsfch=+CUU, V (8)

* * * * * * * * where the normalizations U=U /U ch, s=s /B , n=n /B , and C= B C are made, the parameters a’1, a’2, a’3, 2 and a’4 are function of β, F ch, Cf,ch, A, and upstream boundary conditions on velocity and curvature (see Garcia et al. (1994) and Motta et al. (2011) for details), and the friction coefficient Cf,ch is calculated, following Engelund and Hansen (1967), as

−2 * ⎛⎞⎛⎞Dch C fch, =+⎜⎟6.0 2.5ln (9) ⎜⎟⎜⎟2.5d * ⎝⎠⎝⎠s

* where d s is a measure of the bed grain size. Here the friction coefficient is assumed to be constant * everywhere in the channel for a given channel planform configuration (i.e., a given D ch). Two alternative methods are implemented in the RVR Meander platform for the computation of bank erosion and centerline migration: with the first method (migration-coefficient method, named MC method hereafter) the outer bank displacement rate, equal to the centerline displacement rate (so that the channel width keeps constant), is proportional to the excess velocity at the outer bank through a dimensionless migration coefficient (Hasegawa, 1977; Ikeda et al., 1981):

*** REUU=0 ( bA−) (10)

* * R is the rate of migration (with dimensions of length over time), U A is the cross-sectionally-averaged * velocity, and U b is the depth-averaged near-bank velocity. The dimensionless migration coefficient E0 is usually obtained via calibration against historic channel centerlines. With the second method (physically-based method, named PB method hereafter), the centerline displacement rate is based on the modeling of the physical processes responsible for bank retreat (Motta et al., 2011). As mentioned, just the hydraulic erosion is here considered, which is expressed by Eq. 1. All applications shown in this paper adopt the PB method, except for a case when MC and PB methods are compared for heterogeneous floodplain.

3 NUMERICAL REPRESENTATION OF THE FLOODPLAIN AND GENERATION OF THE PATTERN OF ERODIBILITY

3.1 Numerical representation of the floodplain

Floodplain is represented as a rectangular grid, characterized by constant grid spacing Δx* and Δy* in the * * * * * x and y direction respectively. At each grid node, the parameters M (or k ) and τ c (for PB method) or * * * E0 (for MC method) are assigned. The value V of those parameters at an arbitrary location (x ,y ) is computed by inverse-distance weighting interpolation (Shepard, 1968):

4 Vxywxy***,, ** ∑ iiii( ) ( ) Vxy***, i=1 ( ) = 4 (11) wxy**, ∑ i ( ) i=1 with

** 11 wxyi , ==pp ( ) **** ⎛⎞**22 ** (12) dx( ,,, y xii y) ( ) ⎜⎟(xx−ii) +( yy−) ⎝⎠

* * where (x i,y i) are the coordinates of one of the four vertices of the grid cell containing the arbitrary point (x*,y*), d is the distance of the arbitrary point from one of the four vertices, and p is a positive real number. * * * p is here assumed equal to one. As the river migrates in the floodplain, the parameters τ c, k , M , and E0 at right and left bank are computed with Eq. 11 according to their position in the floodplain. The values are recomputed at the beginning of each migration time step. The extension of the floodplain grid has to be set large enough to contain the possible migration of the centerline, and an algorithm verifies that the boundary of the grid is not crossed by the channel (in that case the simulation stops, because no extrapolation is carried out).

3.2 Generation of the pattern of erodibility

* The horizontal distribution of critical shear stress τ c (or E0, in case the MC method is adopted), defined at the floodplain-grid nodes, can be generated in two different ways. The first corresponds to purely random distribution, referred to as PR distribution hereafter; the second is a randomly-disturbed distribution based on the distance from the valley axis, or RD distribution. * In the PR case, τ c values are generated assuming a normal distribution (µ,σ), where µ and σ are mean and standard deviation respectively, using the Box-Muller transform (Box and Muller, 1958) in the polar form (Bell, 1968; Knopp, 1966). Because of physical consistency, whenever a negative normal number is generated, it is re-generated until it is positive. In the RD case, once defined the valley axis as a straight line in the form

yMxQ**=+ (13) the local mean critical shear stress increases exponentially with the distance l from the valley axis

τ * Pa= aebl[ m] ( c [ ])mean (14) where a is the mean critical shear stress at the valley axis, and b is a coefficient that governs the variation of the mean critical shear stress in the cross-valley direction. The choice of the exponential form, besides being among the simplest, is somewhat analogous to what derived analytically by Pizzuto (1987) to model cross-valley grain size trends associated to overbank flows and to what adopted by Sun et al. (1996) for the increase in time of the erodibility in oxbow lakes because of silt infilling (observe that time and distance from valley axis can be thought as correlated). At each floodplain-grid node, the actual value of critical shear stress is then generated using a Gaussian distribution characterized by mean given by Eq. 14 and a certain assigned standard deviation σ. * Regardless the way a value τ c is assigned at a floodplain-grid node (PR or RD approach), the corresponding value of erodibility k* is found using the widely used expression developed by Hanson and Simon (2001), obtained for channels in the loess areas of the midwestern USA (Iowa, Nebraska, and Mississippi) and based on in situ jet-testing measurements

−0.5 k*3 cm Ns0.2 * Pa ⎣⎦⎡⎤( ) = (τ c [ ]) (15) that is equivalent to

0.5 *6*− M[ m s] =0.2⋅ 10 (τ c [ Pa]) (16)

* * * In both PR and RD cases, values of τ c and k (or M ) outside the floodplain-grid nodes are computed by interpolation using Eq. 11.

4 TEST CASES

We evaluated the impact of horizontal heterogeneity of floodplain soils on rates and patterns of meander migration for sine-generated and natural initial channel centerlines. Table 1 reports the run parameters, including, for convenience, the value of Manning's roughness coefficient corresponding to the value of the friction coefficient. Because reach-averaged Froude number, friction coefficient, and half-width to depth ratio change over time as the reach-averaged depth varies with changing sinuosity, Table 1 reports the values corresponding to a straight channel characterized by same width and inclined at the valley slope. A stochastic approach was followed, based on the Monte Carlo method, that rely on repeated generation of the distribution of the erodibility properties in the floodplain, according to a certain PR or RD distribution, and computation of the corresponding migrated centerline. In this way, a set of migrated centerlines is obtained for a particular PR or RD floodplain-soil distribution. Observe that, because only hydraulic erosion is considered here, vertical banks were assumed, because bank shape has no effect on rates of migration. On the contrary, bank shape would have an impact on rates of migration in case mass failure mechanisms, such as planar failure, were taken into consideration. However, as observed by Motta et al. (2011), the effect can be indirectly accounted for by modifying the * * resistance-to-erosion parameters τ c and k , at least for the case of vertically-homogeneous soil. For both sine-generated and natural cases, we considered relatively short time scales (few decades), that is periods before cutoff occurrence. In this case, all planform complexity is caused by floodplain heterogeneity, which is the focus here, whereas cutoff would introduce additional complexity, here not considered.

* 2 Table 1 Test case run parameters. (Q is discharge; 2B is channel width; S0 is valley slope; F0 is squared Froude number for straight channel characterized by valley slope; Cf0 is friction coefficient for straight channel characterized by valley slope; β0 is half-width to depth ratio for straight channel characterized by valley slope; A is the transverse bed slope parameter (scour factor); and n0 is the Manning's roughness coefficient.) Q 2B* S F 2 C A n Case Shape 0 0 f0 β 0 0 [m3/s] [m] [m/m] [-] [-] [-] [-] [-] 1 Sine-generated 20.0 30.0 0.0005 0.026 0.0194 12.44 5.0 0.046 Mackinaw 2 46.2 38.0 0.0009 0.109 0.0082 17.06 5.0 0.030 River

4.1 Sine-generated channel

For the sine-generated case, focus is on meander migration in RD configurations, the quantification of lateral migration and bend skewness, and the identification of the governing parameters in heterogeneous floodplains.

4.1.1 Sine-generated channel: description

The centerline of a sine-generated channel is expressed as

⎛⎞2π s θθ= o sin ⎜⎟ (17) ⎝⎠Λ

where θ is the angle between centerline and valley axis, θ0 is its value at the crossover point, and Λ is the length of the channel centerline over one meander wavelength. We considered here a 2,040-meter long (2,000-meter long sinuous section with 20-meter long straight

entrance and exit sections) and 30-meter wide channel with Λ = 250 m and θ0 = 55°, which corresponds to a relatively low sinuosity of 1.27. 200 equally-spaced nodes describe the initial centerline, yielding a node spacing of about 10.2 m. The channel contains 17 bends. We simulated the centerline evolution for a 20-year period employing a time step of 0.2 years. Table 1 lists the various simulation parameters.

4.1.2 Sine-generated channel: statistics for migrated centerline

Following the approach described by Posner and Duan (2011), the valley axis allows for identifying bends (lobes) based on the intersections of the centerline with the valley axis. For each lobe, statistics can be computed using the (x*,y*) coordinates of the centerline nodes. While Posner and Duan (2011) used such statistics to compare simulated and measured migrating bends, here the statistics allow describing and comparing the migrated-centerline variability associated to different floodplain-soil configurations. * In particular, we considered the mean value Meany* of the coordinate y , which quantifies lateral * migration, and the skewness Skewx* of the coordinate x , where a positive (negative) value of Skewx* represents upstream (downstream) skewness. Figure 1 shows an example of evaluation of Meany* and Skewx* for all migrated-centerline lobes, except the most upstream (Lobe 1) and downstream (Lobe 17) excluded hereafter in the analysis to focus on the central bends, obtained for PR distribution of critical shear stress with µ = 6 Pa and σ = 1 Pa. The valley axis is in this case characterized by y* = 34.4 m.

100 0.3

80 0.2 60

40 0.1 20 x* [m]

y* 0 0 Skew

Mean -20 -0.1 -40

-60 -0.2 -80

-100 -0.3 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Lobe (c) Lobe (d) Figure 1 Sine-generated curve: example of migrated-centerline lobes (Figure b) for a certain PR floodplain-soil distribution (Figure a) and corresponding values of Meany* (Figure c) and Skewx* (Figure d). Flow is from left to right. In Figure b, the initial centerline is in red, the migrated centerline in black.

4.1.3 Sine-generated channel: meander migration in RD configurations

Table 2 reports the parameters of four Monte Carlo simulations for RD configuration, characterized by different values of b (indicator of the cross-valley increase of soil resistance) and σ (indicator of the local randomness of soil property). We selected a grid-cell size Δx* = Δy* of 60 m, which is equal to two times the channel width, almost two times the amplitude of the initial centerline, and about three times the wavelength of the initial centerline measured along the valley axis; moreover, the value of the grid-cell size is large enough to produce very appreciable variability of the migrated centerlines (see Section 4.2.2 for in-depth analysis of the effect of the grid-cell size). One of the rows of the floodplain grid coincides with the axis of the initial centerline at y* = 34.4 m. Because of the shape of the initial centerline, several simulations were discarded for floodplain-soil configurations producing migrated centerlines characterized by non-physical loops at the downstream end.

Therefore, 1000 floodplain-soil generations were used for each of the four Monte Carlo simulations for sine-generated curve, in order to obtain a sufficiently large number of migrated centerlines for computing meaningful metrics. Note also that a Monte Carlo approach will never completely represent an entire statistical population; on the other hand, a sufficient number of runs can reasonably represent it. Figure 2 shows, for each Monte Carlo simulation, an example of realization of floodplain-soil spatial distribution and all the migrated centerlines. We generally observe, for higher value of σ, a higher variability of migrated centerlines (compare Figures 2a,b (configurations RD1 and RD2) Vs. Figures 2c,d (RD3 and RD4)) and, for higher value of b, less lateral migration and a more pronounced tendency towards upstream skewness (Figures 2a,c (RD1 and RD3) Vs. Figures 2b,d (RD2 and RD4)).

Table 2 Sine-generated curve: parameters for RD Monte Carlo simulations. Grid size Configuration M [-] Q [m] a [Pa] b [1/m] [-] [m] axis axis σ RD1 60.0 0.0 34.4 6.0 0.001 1.0 RD2 60.0 0.0 34.4 6.0 0.003 1.0 RD3 60.0 0.0 34.4 6.0 0.001 0.5 RD4 60.0 0.0 34.4 6.0 0.003 0.5

Figure 2 Sine-generated curve: example of realization of floodplain-soil spatial distribution and all 20-year migrated centerlines for each of the four RD Monte Carlo simulations, characterized by different values of parameters b and σ. (a) Configuration RD1, (b) RD2, (c) RD3, and (d) RD4 (see Table 2). Flow is from left to right. The initial centerline is in red, migrated centerlines in black.

A box plot can be used to graphically visualize minimum, 25th percentile, median, 75th percentile, and maximum of both Meany* and Skewx* values, as shown in Figure 3. Note that, for convenience, the absolute values of Meany* are actually plotted in Figures 3a,c,e,g and 4a,b. In Figure 4 we compare, for the four RD configurations and for each lobe, the median values of Meany* (Figure 4a), the difference th th between 75 and 25 percentiles of Meany* (Figure 4b), the median value of Skewx* (Figure 4c), and the th th difference between 75 and 25 percentiles of Skewx* (Figure 4d).

100 0.8 RD1 (low b , high σ ) RD1 (low b , high σ ) 90 0.6 80 0.4 70 60 0.2 x* [m]

y* 50 0.0

40 Skew -0.2 Mean 30 -0.4 20 -0.6 10 0 -0.8 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Lobe (a) Lobe (b) 100 0.8 RD2 (high b , high σ ) RD2 (high b , high σ ) 90 0.6 80 0.4 70 60 0.2 x* [m]

y* 50 0.0

40 Skew

Mean -0.2 30 -0.4 20 -0.6 10 0 -0.8 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Lobe (c) Lobe (d) 100 0.8 RD3 (low b , low σ ) RD3 (low b , low σ ) 90 0.6 80 0.4 70 60 0.2 x* [m]

y* 50 0.0

40 Skew

Mean -0.2 30 -0.4 20 -0.6 10 0 -0.8 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Lobe (e) Lobe (f) 100 0.8 RD4 (high b , low σ ) RD4 (high b , low σ ) 90 0.6 80 0.4 70 60 0.2 x* [m]

y* 50 0.0

40 Skew

Mean -0.2 30 -0.4 20 -0.6 10 0 -0.8 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Lobe (g) Lobe (h) Figure 3 Sine-generated curve: box plots of Meany* and Skewx* for the four RD Monte Carlo simulations, characterized by different values of parameters b and σ (see Table 2).

55 20 RD1 (low b, high sigma) 18 [m] RD2 (high b, high sigma)

50 y* RD3 (low b, low sigma)

[m] 16 RD4 (high b, low sigma) y* 45 14

40 12 10 35 8

30 percentile of Mean 6 th RD1 (low b, high sigma) 4 RD2 (high b, high sigma) - 25

Median value of Mean 25 RD3 (low b, low sigma) th 2

RD4 (high b, low sigma) 75 20 0 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Lobe (a) Lobe (b) 0.40 0.16

0.35 x* 0.14 x* 0.30 0.12

0.25 0.10

0.20 0.08

0.15 0.06 percentile of Skew percentile th 0.10 RD1 (low b, high sigma) 0.04 RD1 (low b, high sigma) Median value of Skew RD2 (high b, high sigma) - 25 RD2 (high b, high sigma) 0.05 RD3 (low b, low sigma) th 0.02 RD3 (low b, low sigma) RD4 (high b, low sigma) 75 RD4 (high b, low sigma) 0.00 0.00 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Lobe (c) Lobe (d) th th Figure 4 Sine-generated curve: median value and difference between 75 and 25 percentiles for Meany* and Skewx* corresponding to the four RD Monte Carlo simulations characterized by different values of parameters b and σ (see Table 2).

From Figures 3 and 4, if b (indicator of cross-valley increase of soil resistance) increases, the value of Meany* (indicator of lateral migration) generally decreases: this can be observed by comparing Figure 3a (RD1) with Figure 3c (RD2), or Figure 3e (RD3) with Figure 3g (RD4), or from Figure 4a, where the median values of Meany* are lower for higher b. As regards Skewx* (indicator of planform shape), high b values produce a more enhanced tendency toward upstream skewness, which is visible especially from Figure 4c: this can be explained as the result of the constrained lateral migration.

If σ (indicator of the local randomness of soil property) increases, Meany* values are not significantly affected on average (Figure 4a); on the other hand, the possibility of getting downstream (negatively)-skewed lobes increases (compare Figures 3b (RD1) and 3d (RD2) with Figures 3f (RD3) and 3h (RD4)) and both Meany* and Skewx* present a wider spectrum (Figures 4b and 4d).

4.2 Natural channel

For the natural channel case, focus is on the effect of the scale of heterogeneity, conditions for development of downstream-skewed bends and generally complex planform shapes, and the comparison between MC and PB stochastic approaches.

4.2.1 Natural channel: description

We tested different floodplain configurations for a reach on the Mackinaw River in Illinois, USA. The reach is located in Tazewell County about 15 kilometers upstream of the junction of the Mackinaw River with the Illinois River between the towns of South Pekin and Green Valley. The 1951 historic centerline was taken as starting centerline. An equidistant set of 350 centerline nodes with a spacing of 11.4 m describes the initial channel centerline. Channel width is 38 m and valley slope is 0.0009 m/m (Motta et al., 2011). The mean sinuosity of the study reach is 1.34. As assumed by Motta et al. (2011), we used a model discharge of 46.2 m3/s, that is, from an analysis of the discharge record between 1922 and 1956 at the USGS station 05568000 near Green Valley, between the average value (20.9 m3/s) and the maximum value (61.6 m3/s) of the mean annual streamflow over the period. The upstream boundary for modeling was set in a straight

reach. Because the velocity distribution is not known there, a uniform profile in the transverse direction was assumed for simplicity. Further, we used a 2 nd-order Savitzky-Golay filter with an averaging window of 5 nodes applied every 10 iterations.

4.2.2 Natural channel: effect of heterogeneity scale

Monte Carlo simulations were run for different values of grid-cell size (Δx*, Δy*) with PR configuration * * characterized by µ(τ c) = 9.0 Pa and σ(τ c)= 1.0 Pa (Table 3). The mean value is the same adopted for the Mackinaw River by Motta et al. (2011) for simulations in homogeneous floodplain and recently confirmed by jet erosion tests in situ. Because floodplain-soil values at grid nodes are random while values in between nodes are obtained through interpolation (therefore they are deterministic), the grid-cell size is an indicator of the spatial scale of randomness of the floodplain-soil properties. Figure 5 shows the results for grid-cell size values respectively equal to 1, 2, and 3 times the channel width. 500 floodplain-soil generations were considered for each of the three cases, which is a number or realizations that was visually judged as sufficient to describe the variability of the migrated centerlines.

Table 3 Natural channel: PR Monte Carlo simulations characterized by different scale of heterogeneity. Configuration Δx* = Δy* [m] Δx*/2B* = Δy*/2B* [-] 2W*/2B* [-] PR1 38.0 1.0 2.12 PR2 76.0 2.0 3.10 PR3 114.0 3.0 3.59

From Figure 5 we see that lower grid-cell size is associated to lower migrated-centerline variability. In quantitative terms, the ratio was computed between the area occupied by all migrated centerlines (whose perimeter was manually digitized) and the length of the migrated centerline obtained for homogeneous * * -8 3 floodplain characterized by mean floodplain-property values (τ c = 9 Pa, k = 6.7 · 10 m /(Ns)). This ratio with dimensions of length, indicated as 2W*, was then divided by the channel width 2B* to give a dimensionless indicator of the migrated-centerline variability. From Table 3 we observe that 2W*/2B* increases for increasing grid-cell size less than linearly. This implies, from a practical perspective, that a finer-scale campaign of characterization of the floodplain soil is, as one could reasonably expect, associated to less uncertainty on the predicted migrated centerline. On the other hand, the increase of 2W*/2B* with the scale of heterogeneity is less than linear and presents, at least in this case, a tendency toward a plateau: our stochastic methodology may therefore provide a tool for determining the suitable spatial density for field characterization of floodplains.

Figure 5 Continues in the next page

Figure 5 Natural channel: example of realization of floodplain-soil spatial distribution and all 40-year migrated centerlines for each of the four PR Monte Carlo simulations, characterized by different values of grid-cell size Δx* = Δy*. (a) Configuration PR1, (b) PR2, and (c) PR3 (see Table 3). Flow is from right to left. The initial centerline is in red, migrated centerlines in black.

4.2.3 Natural channel: complexity of planform shapes

While the belts occupied by all migrated centerlines in Figure 5 show upstream skewness, typically developed by meanders, a few cases are present where more complex shapes are obtained. Considering for instance the set of PR2 simulations (Table 3 and Figure 5b), we could identify downstream-skewed bends (Figure 6a,b), symmetric compound loops (Figure 6c), and compound bends (Figure 6d). In general, large scales of heterogeneity are more likely to produce downstream-skewed and, in general, complex planform features, given the higher migrated-centerline variability that can be obtained (see Section 4.2.2).

Figure 6 Natural channel: examples of downstream-skewed bends (L1 and L2), compound loops (L3 and L4), and compound bend (L5). Flow is from right to left. The initial centerline is in red, migrated centerlines in black.

4.2.4 Natural channel: comparison of MC and PB approaches for heterogeneous floodplain

MC and PB stochastic migration approaches were compared by considering PR distributions of the * * parameters E0 (for the MC approach) and τ c or k (for the PB approach) characterized by the same ratio between their standard deviation σ and mean value µ (Table 4). In this way it was possible to compare the effect of the spatial variability of those parameters on migrated centerlines. For the MC approach, the -7 * mean value 5 · 10 for E0 was taken from Motta et al. (2011). In the case of the PB approach, when τ c values were randomly generated at the floodplain-grid nodes, the corresponding values of k* were * * computed with Eq. 15, whereas, when k values were randomly generated, the corresponding values of τ c were computed with the inverse of Eq. 15. A grid size Δx* = Δy* = 2B* = 38 m was used and a period of 20 years was simulated. * * From Figure 7 we observe that the spatial variability of parameters τ c or k , which govern PB migration, reflect in greater variability of migrated centerlines than in the case of E0, which governs the MC

migration. In other words, the PB approach intrinsically produces more centerline variability as result of floodplain-resistance variability than the MC approach. Quantitatively, using the indicator 2W*/2B* * * introduced in Section 4.2.2, the variables τ c and k respectively produce about 340 and 595% of the migrated-centerline variability associated to the variable E0 (Table 4). This can be explained by observing that, while the MC approach always predicts a “smooth” migration pattern because it does not account for any threshold for erosion (Eq. 10), the PB approach using hydraulic erosion does (Eq. 1, where erosion * occurs for shear stress at the bank greater than the critical shear stress τ c) and depends on two parameters * * instead of one, τ c and k , even if correlated by Eq. 15.

Table 4 Natural channel: PR Monte Carlo simulations for comparison of effects of the spatial variability of the * * parameters E0, τ c, and k . Parameter µ σ σ/µ [−] 2W*/2B [-] -7 -8 E0 [-] 5 · 10 5.6 · 10 1/9 0.43 * τ c [Pa] 9.0 1.0 1/9 1.47 k* [m3/(Ns)] 6.7 · 10-8 7.4 · 10-9 1/9 2.56

Figure 7 Natural channel: example of realization of floodplain-soil spatial distribution and all 20-year migrated centerlines for Monte Carlo simulations characterized by (a) PR distribution of E0 (MC migration method), (b) PR * * distribution of τ c (PB migration method); (c) PR distribution of k (PB migration method). Flow is from right to left. The initial centerline is in red, migrated centerlines in black.

5 DISCUSSION

We developed a methodology for assigning numerically on a regular grid the floodplain-soil properties in terms of migration coefficient E0 (when using the classic MC approach for migration) or parameters for * * hydraulic erosion τ c and k (for PB approach, of main interest here), combining random number generation at grid nodes and interpolation in between nodes. Two different spatial patterns were analyzed, purely random (PR) and randomly disturbed (RD), where, in the latter case, a general pattern with increasing resistance to erosion was assumed away from the valley centerline. We considered two application cases, corresponding to sine-generated and natural channel, for which we conducted several numerical tests (Monte Carlo simulations) based on repeated generations of floodplain

configuration and computation of the corresponding migrated centerlines and associated metrics. In the case of sine-generated channel, focus was put on RD configurations, and two metrics, Meany* and Skewx*, were used to respectively quantify lateral migration and planform orientation of bends. Two governing parameters were found, which indicate the local randomness of soil resistance to erosion (σ) and the cross-valley increase of soil resistance (b). The parameter σ mainly governs the bend shape, and larger values increase the possibility of occurrence of downstream-skewed bends and in general allow for a wider spectrum of bend skewness. Roughly speaking, b controls lateral migration; however, deeper analysis revealed that bend skewness is affected too, being higher b values associated to more pronounced upstream skewness of bends. For a natural channel, migrated-centerline variability as function of heterogeneity scale was evaluated by defining a dimensionless metric (2W*/2B*) that quantifies the “spread” of the migrated centerlines corresponding to a Monte Carlo simulation for PR floodplain configuration characterized by a certain grid-cell size. The analysis showed that lower grid-cell size, i.e. lower scale of heterogeneity, corresponds to lower migrated-centerline variability and that the increase of migrated-centerline variability with heterogeneity scale is less than linear and seems to tend toward a plateau. In presence of heterogeneous floodplain, even using a simple model for hydrodynamics and bed morphodynamics like that adopted here, downstream-skewed lobes and complex planform shapes such as symmetric compound loops and compound bends can be obtained, especially for large scale of heterogeneity and natural-channel alignment. Finally, comparison of stochastic MC and PB simulations showed that the PB approach for migration is intrinsically associated to greater migrated-centerline variability as result of floodplain-resistance spatial variability.

6 CONCLUSIONS

In this paper we performed an analysis of the impact of floodplain-soil heterogeneity on rates and patterns of migration, considering horizontal heterogeneity and the mechanism of hydraulic erosion. We showed, using a simple Ikeda et al. (1981)-type model for hydrodynamics and bed morphodynamics, that floodplain-soil complexity can greatly contribute to planform complexity of meandering rivers and produce downstream-skewed features, which an Ikeda et al. (1981)-type model cannot produce in homogeneous floodplain. The simulations performed allowed for highlighting three key parameters that characterize the effect of floodplain heterogeneity on channel planform complexity, respectively quantifying the local randomness of soil resistance, the cross-valley increase of soil resistance, and the spatial scale of heterogeneity. Future work will include the analysis of mass failure mechanisms in heterogeneous floodplains, with the goal of addressing the impact of vertical soil heterogeneity and bank-profile shape on rates and patterns of migration. Moreover, it will be interesting to couple the described procedure for channel migration in heterogeneous floodplain with a model for hydrodynamics and bed morphodynamics that considers the free response of the bed sediments, by fully coupling flow, sediment transport, and bed morphodynamics, such as the second-order model by Johannesson and Parker (1989) or the fourth-order model by Zolezzi and Seminara (2001). The analysis would allow for assessing how the planform complexity due to the hydrodynamics compares to that caused by bank processes and floodplain heterogeneity. As shown by Camporeale et al. (2007) both second and fourth-order models can give rise, or wash away, multilobed planimetries in homogeneous floodplain, and Seminara et al. (2001) already showed that, without invoking cutoff processes, spatial heterogeneity of bank erodibility or human constraints, in super-resonant regime meander shape displays alternating upstream- and downstream-skewed loops, as well as multiple loops. Finally, moving to long-term simulations will allow for analyzing the coupled effect of cutoff processes and floodplain heterogeneity on meander patterns.

7 ACKNOWLEDGMENTS

This research was supported by an agreement from the U.S. Department of Agriculture, Forest Service, Pacific Southwest Research Station and using funds provided by the Bureau of Land Management

through the sale of public lands as authorized by the Southern Nevada Public Land Management Act. This work was performed under Specific Cooperative Agreement No. 58-6408-8-265 between the Department of Civil and Environmental Engineering at the University of Illinois at Urbana-Champaign and the U.S. Department of Agriculture, Agricultural Research Service, National Sedimentation Laboratory. Eric Waschle Prokocki is also acknowkledged for the interesting discussions on floodplain-soil patterns.

REFERENCES

Abad, J.D., Garcia, M.H. 2006, RVR meander: A toolbox for re-meandering of channelized streams. Computers and Geosciences 32 (1), 92-101. Allen, J.R.L. 1965, A review of the origin and characteristics of recent alluvial sediments. Sedimentology 5, 89-191, 1965. Bell, J. 1968, Algorithm 334: Normal random deviates. Communications of the ACM 11 (7). Blondeaux, P., Seminara, G. 1985, A unified bar-bend theory of river meanders. Journal of Fluid Mechanics 157, 449-470. Box, G.E.P., Muller, M.E. 1958, A Note on the Generation of Random Normal Deviates. The Annals of Mathematical Statistics 29 (2), 610-611. Camporeale, C., Perona, P., Porporato, A., Ridolfi, L. 2007, Hierarchy of models for meandering rivers and related morphodynamic processes. Reviews of Geophysics 45, 1-28. Constantine, C.R., Dunne, T., Hanson, G.J. 2009, Examining the physical meaning of the bank erosion coefficient used in meander migration modeling. Geomorphology 106 (3-4), 242-252. Darby, S.E., Rinaldi, M., Dapporto, S. 2007, Coupled simulations for fluvial erosion and mass wasting for cohesive river banks, Journal of Geophysical Research 112, Art. No. F03022. Engelund, F., Hansen, E. 1967, A Monograph on Sediment Transport in Alluvial Streams. Danish Technical Press, Copenhagen. Frascati, A., Lanzoni, S. 2009, Morphodynamic regime and long-term evolution of meandering rivers. J. Geophys. Res. 114, art. no. F02002, doi:10.1029/2008JF001101. Garcia, M.H., Bittner, L., Nino, Y. 1994, Mathematical modeling for meandering streams in Illinois: a tool for stream management and engineering. Hydraulic Engineering Series No. 43, Civil Engineering Studies report of the Department of Civil Engineering, University of Illinois at Urbana-Champaign, Urbana, Illinois. Guo, J., Julien, P.Y. 2005, Shear stress in smooth rectangular open channel flows, Journal of Hydraulic Engineering 131, 30-36. Hanson, G.J., Simon, A. 2001, Erodibility of cohesive streambeds in the loess area of the midwestern USA. Hydrologic Processes 15(1), 23-38. DOI: 10.1002/hyp.149. Hanson, G.J., Cook, K.R. 2004, Apparatus, test procedures, and analytical methods to measure soil erodibility in situ. Applied Engineering in Agriculture 20 (4), 455-462. Hasegawa, K. 1977, Computer simulation of the gradual migration of meandering channels. In Proceedings of the Hokkaido Branch, Japan Society of Civil Engineering, 197--202. (in Japanese) Howard, A., 1992. Modelling channel evolution and floodplain morphology. In Floodplain Processes, edited by M. G.Anderson,D. E.Walling, and P.D. Bates, pp. 1562, John Wiley, New York. Howard, A. 1996, Modelling channel migration and floodplain development in meandering streams. In Lowland Floodplain Rivers: Geomorphological Perspectives, edited by P.A. Carling and G.E. Petts, pp. 241, JohnWiley, New York. Ikeda, S., Parker, G., Sawai, K. 1981, Bend theory of river meanders. Part 1. Linear development. Journal of Fluid Mechanics 112, 363-377. Johannesson, H., Parker, G. 1985, Computed simulated migration of meandering rivers in Minnesota. Project Report No. 242. Minneapolis, Minnesota. Johannesson, H., Parker, G.. 1989, Linear theory of river meanders. In River Meandering, Water Resources Monograph, Vol. 12, 181--212, edited by S. Ikeda and G. Parker. American Geophysical Union, Washington, DC. Knopp, R., 1969. Remark on algorithm 334 [G5]: normal random deviates. Communications of the ACM 12 (5). Langendoen, E.J., Alonso, C.V. 2008, Modeling the evolution of incised streams. i: model formulation and validation of flow and streambed evolution components. Journal of Hydraulic Engineering 134 (6), 749-762. Langendoen, E.J., Simon, A. 2008, Modeling the evolution of incised streams. ii: streambank erosion. Journal of Hydraulic Engineering 134 (7), 905-915. Langendoen, E.J., Wells, R.R., Thomas, R.E., Simon, A., Bingner, R.L. 2009, Modeling the evolution of incised streams. iii: model application. Journal of Hydraulic Engineering 135 (6), 476-486. Mosselman, E. 1998, Morphological modelling of rivers with erodible banks. Hydrological Processes 12 (8), 1357-1370. Motta, D., Abad, J.D., Langendoen, E.J., Garcia, M.H. 2011, A simplified 2D model for meander migration with physically-based bank evolution. Geomorphology. In Press.

Papanicolaou, A.N., Elhakeem, M., Hilldale, R. 2007, Secondary current effects on cohesive river bank erosion. Water Resources Research 43, Art. No. W12418. Perucca, E., Camporeale, C., Ridolfi, L. 2007, Significance of the riparian vegetation dynamics on meandering river morphodynamics, Water Resources Research 43, W03430, doi:10.1029/2006WR005234. Pizzuto, J.E. 1987, Sediment diffusion during overbank flows. Sedimentology 34, 301-317. Pizzuto, J.E., ASCE Task Committee on Hydraulics, Bank Mechanics, and Modeling of River Adjustment 2008, Streambank erosion and river width adjustment. ASCE manual of practice 110, M. H. Garcia, ed., ASCE, Reston, Va., Chap. 7, 387-438. Posner, A.J., Duan, J.G. 2011, Numerical Simulation of River Meandering: Deterministic or Stochastic. Geomorphology. In press. Rinaldi, M., Mengoni, B., Luppi, L., Darby, S.E., Mosselman, E. 2008, Numerical simulation of hydrodynamics and bank erosion in a river bend. Water Resources Research 44 (9), art. no. W09428. Seminara, G., Zolezzi, G., Tubino, M., Zardi, D. 2001, Downstream and upstream influence in river meandering. Part 2. Planimetric development Journal od Fluid Mechanics 438, 213-230. Simon, A., Thomas, R.E., Klimetz, L. 2010, Comparison and experiences with field techniques to measure critical shear stress and erodibility of cohesive deposits. 2nd Joint Federal Interagency Conference, Las Vegas, NV, June 27 - July 1, 2010 Shepard, D. 1968, A two-dimensional interpolation function for irregularly-spaced data. Proceedings of the 1968 ACM National Conference. pp. 517-524. doi:10.1145/800186.810616 Sun, T., Meaking, P., Jossang, T., Schwarz, K. 1996, A simulation model for meandering rivers. Water Resources Research 32 (9), 2937-2954. Sun, T., Meakin, P., Jossang, T. 2001a, A Computer Model for Meandering Rivers with Multiple Bed Load Sediment Sizes 1. Theory. Water Resources Research 37 (8), 2227-2241. Sun, T., Meakin, P., Jossang, T. 2001b, A Computer Model for Meandering Rivers with Multiple Bed Load Sediment Sizes 2. Computer Simulations. Water Resources Research 37 (8), 2243-2258. Zolezzi, G., Seminara, G. 2001, Dowstream and upstream influence in river meandering. Part 1. General theory and application to overdeepening. Journal of Fluid Mechanics 438, 183-211.