1. Introduction
Aquatic vegetation is an integral component of fluvial and coastal systems, where it alters the velocity field and consequently sediment transport (Bennett et al. Reference Bennett, Wu, Alonso and Wang2008; Nepf Reference Nepf2012; Vargas-Luna, Crosato & Uijttewaal Reference Vargas-Luna, Crosato and Uijttewaal2015). An understanding of how vegetation modifies sediment dynamics is critical, as sediment transport drives the evolution of riverbeds and coastlines, and influences the distribution of nutrients and pollutants (Gacia & Duarte Reference Gacia and Duarte2001; Schulz & Peall Reference Schulz and Peall2001; Schulz et al. Reference Schulz, Kozerski, Pluntke and Rinke2003; Huai et al. Reference Huai, Li, Katul, Liu and Yang2021). Moreover, by promoting sediment deposition and reducing erosion, vegetation enhances riverbank stability and creates habitats that support biodiversity (Sand-jensen Reference Sand-Jensen1998; Gurnell Reference Gurnell2014; O’Briain et al. Reference O’Briain, Corenblit, Garófano-Gómez and O’Leary2024). Bernhardt et al. (Reference Bernhardt2005) noted that annual expenditures on river restoration in the USA exceed $1 billion, with many projects using vegetation to control sediment dynamics. Given its influence on landscape evolution and ecosystem health, quantifying sediment transport in vegetated environments remains a central and persistent challenge. An important control on sediment transport is the bed shear stress.
For beds without vegetation, which we call bare beds, bed shear stress predicts sediment entrainment and transport rate (van Rijn Reference Van Rijn1984; García Reference García2013). However, for beds with vegetation, the controls on sediment transport are more complex. While some studies show that bed shear stress remains the controlling parameter (Jordanova & James Reference Jordanova and James2003; Kothyari, Hashimoto & Hayashi Reference Kothyari, Hashimoto and Hayashi2009; Armanini & Cavedon Reference Armanini and Cavedon2019), other studies have shown that near-bed turbulence plays a more important role in sediment transport (Yang & Nepf Reference Yang and Nepf2018; Shan et al. Reference Shan, Zhao, Liu and Nepf2020; Zhao & Nepf Reference Zhao and Nepf2021; Yang et al. Reference Yang, Zhang, Qi and Li2025; Zhang et al. Reference Zhang, Lu, Wang, Li, Li, Guo, Yu and Wang2025). These contrasting observations are due, in part, to the lack of good measurements and models for bed shear stress over rough beds in the presence of vegetation. Because the presence of vegetation alters the boundary layer thickness and bed shear stress, classic characterisations from bare bed conditions cannot be applied within vegetation (Yang, Kerger & Nepf Reference Yang, Kerger and Nepf2015; Conde-Frias et al. Reference Conde-Frias, Ghisalberti, Lowe, Abdolahpour and Etminan2023). An accurate prediction for bed shear stress in the presence of vegetation, and in particular for rough beds, remains a critical knowledge gap.
Models for bed shear stress in vegetated conditions can be categorised into three main approaches. First, in the drag partitioning method, the bed shear stress is calculated by subtracting an estimated vegetation drag from the measured total flow resistance (Jordanova & James Reference Jordanova and James2003; Kothyari et al. Reference Kothyari, Hashimoto and Hayashi2009). This method is prone to large uncertainties, because the vegetation drag is typically an order of magnitude larger than the bed shear stress, such that small uncertainties in the estimated vegetation drag can lead to significant errors in the bed shear stress. In addition, for mobile beds, bedforms introduce additional form drag that further complicates the drag partitioning (Le Bouteiller & Venditti Reference Le Bouteiller and Venditti2015).
The second approach estimates bed shear stress from the structure of the bottom boundary layer (BBL). Yang et al. (Reference Yang, Kerger and Nepf2015) proposed this approach for smooth beds, hypothesising that the boundary layer thickness is reduced by the vegetation-generated turbulence to a scale set by the stem diameter. Building on this, Etminan, Ghisalberti & Lowe (Reference Etminan, Ghisalberti and Lowe2018) improved upon Yang’s model by casting the prediction of bed shear stress directly in terms of the boundary layer thickness, rather than the stem diameter. More recently, Conde-Frias et al. (Reference Conde-Frias, Ghisalberti, Lowe, Abdolahpour and Etminan2023) empirically demonstrated an inverse relationship between boundary layer thickness and near-bed turbulent kinetic energy, providing a way to predict bed shear stress from the near-bed turbulence. While the aforementioned studies focused on smooth beds, Aliaga & Aberle (Reference Aliaga and Aberle2025) considered a rough bed. For canopies of low density, they showed that the velocity profile remains logarithmic within the BBL, so that bed shear stress could be estimated following the conventional bare bed methods. However, this will fail within canopies of higher density, for which the velocity profile within the BBL is not logarithmic, but instead follows a power-law function (Xu et al. Reference Xu, Ma, Chen, Zhong, Xu, He and Nepf2026). Therefore, a general model for predicting bed shear stress in vegetated conditions over rough beds does not yet exist and is the focus of this paper.
A third approach for predicting bed shear stress utilises the phenomenological theory of turbulence. Grounded in the work pioneered for bare beds by Gioia & Bombardelli (Reference Gioia and Bombardelli2001), this framework builds on the idea that bed shear stress is the manifestation of momentum exchange between the bulk flow and the bed, which can be described in terms of the interaction of two distinct eddy scales: the large, energy-containing eddies that carry the bulk momentum towards the bed, and the small, near-bed eddies that directly transfer this momentum to the bed. This theory has been used to represent the stress between a submerged layer of roughness elements (such as a submerged canopy) and the overflowing layer, which enabled the prediction of the flow resistance (Huthoff, Augustijn & Hulscher Reference Huthoff, Augustijn and Hulscher2007; Huthoff Reference Huthoff2012). Further, Lu, Cheng & Wei (Reference Lu, Cheng and Wei2021) extended the model to emergent vegetation with rough beds by redefining the characteristic scales of the large eddies for vegetated conditions. Specifically, they argued that the characteristic length scale of the large eddies is governed not by the channel’s hydraulic radius, but by a vegetation hydraulic radius:
$R_{v}=0.25\pi d(1-\phi )/\phi$
(Cheng & Nguyen Reference Cheng and Nguyen2011), with
$\phi =n\pi d^{2}/4$
the solid volume fraction of vegetation within the canopy volume, and
$n$
the stems per bed area. While this work provided a valuable advance, a conceptual challenge limits its applicability at low vegetation densities. Specifically, as vegetation density (
$n$
) approaches zero,
$R_{v}$
approaches infinity – a physically unrealistic scenario.
The goal of this paper is to address the limitations described above by developing a more complete framework for predicting BBL thickness and bed shear stress in vegetated channels that covers both smooth and rough beds. While natural aquatic plants may be flexible or have more complex morphology (e.g. foliage), which can alter the vertical profiles of velocity and turbulence over depth, this study isolates the fundamental mechanisms using a simplified canopy consisting of homogeneously distributed, emergent and rigid cylinders. First, a force balance is used to derive a theoretical prediction for the BBL thickness. Second, we extend the turbulence model proposed by Gioia & Bombardelli (Reference Gioia and Bombardelli2001) to vegetated conditions by redefining the characteristic length and velocity scales of the large-scale eddies. The outcome is a unified model for BBL thickness and bed shear stress that describes the full range of bare to vegetated conditions for both smooth and rough beds. Section 2 details the theoretical derivation of BBL thickness and bed shear stress models. Section 3 compares the models to data compiled from previous studies. Section 4 discusses the physical mechanisms and limitations of the proposed models. Finally, § 5 presents the main conclusions.
2. Theory
2.1. The BBL model
2.1.1. Double-averaged Navier–Stokes equations
Flows through arrays of obstacles, such as aquatic vegetation, are both spatially heterogeneous due to the obstacles, and typically turbulent, thus unsteady. As a result, a double-averaging procedure must be applied to the Navier–Stokes equations to describe the macroscopic flow behaviour (López & García Reference López and García2001; Nikora et al. Reference Nikora, McEwan, McLean, Coleman, Pokrajac and Walters2007; Yang et al. Reference Yang, Kerger and Nepf2015). First, temporal averaging decomposes an instantaneous quantity
$\zeta$
into time mean
$\overline{\zeta }$
and fluctuating
$\zeta '$
components (
$\zeta =\overline{\zeta }+\zeta '$
). The averaging time is chosen to filter out turbulent fluctuations. Second, spatial averaging over a horizontal plane is performed on the time-mean quantities
$\overline{\zeta }$
and over a region encompassing multiple vegetation elements, yielding the double average
$\langle \overline{\zeta }\rangle$
and spatial deviation
$\overline{\zeta }''$
, such that
$\overline{\zeta }=\langle \overline{\zeta }\rangle +\overline{\zeta }''$
. The overbar denotes time averaging, angle brackets
$\langle \rangle$
denote spatial averaging, a single prime (
$'$
) denotes temporal fluctuations from the time mean, and a double prime (
$''$
) denotes spatial deviations of time-averaged quantities from their spatial mean. The double average removes stem-scale variations in the water surface slope generated by the emergent cylinders.
In an emergent canopy of uniform porosity, the double-averaged Navier–Stokes equations are
in which
$u_{i}=[u, v, w]$
are the velocity components in the
$x_{i}=[x, y, z]$
directions, corresponding to the streamwise (parallel to the bed), spanwise and vertical directions, respectively. Here,
$p$
is the pressure,
$\rho$
is the fluid density,
$g_{i}$
is the component of gravitational acceleration in the
$x_{i}$
direction,
$F_{i}$
is the vegetation drag per fluid volume acting in the
$i$
th direction, and
$\tau _{\textit{ij}}^{\textit{disp}} , \langle \tau _{\textit{ij}}^{\textit{Rey}}\rangle, \langle \tau _{\textit{ij}}^{v\textit{is}}\rangle$
are the dispersive stress, spatially averaged Reynolds stress and spatially averaged viscous stress, defined as
in which
$\nu$
is the kinematic viscosity. For steady, uniform flow within an emergent canopy of homogeneously distributed cylinders, we can assume that
$\langle \overline{u}\rangle , \tau _{\textit{ij}}^{\textit{disp}} , \big\langle \tau _{\textit{ij}}^{\textit{Rey}}\big\rangle, \big\langle \tau _{\textit{ij}}^{v\textit{is}}\big\rangle$
are uniform in
$x$
and
$y$
, and that
$\langle \overline{v}\rangle \approx 0$
and
$\langle \overline{w}\rangle \approx 0$
, such that (2.1) can be simplified to
The vegetation drag
$F_{x}$
can be represented by a quadratic law (e.g. Nepf Reference Nepf2012)
in which
$C_{v}=0.5C_{\!D}a/(1-\phi )$
is the canopy drag parameter, defined as the inverse of the canopy drag length
$L_{v}=1/C_{v}$
(Belcher, Jerram & Hunt Reference Belcher, Jerram and Hunt2003). Here,
$a$
is the frontal area per unit volume of canopy, and
$C_{\!D}$
is the drag coefficient of a single stem. For cylindrical stems of diameter
$d$
,
$a=4\phi /(\pi d)$
and
$C_{v}=2\phi C_{\!D}/[\pi d(1-\phi )]$
. The gravitational term
$g_{x}=\textit{gs}_{b}$
represents the component of gravitational acceleration parallel to a bed with slope
$s_{b}$
with respect to a horizontal plane. The pressure gradient is
in which
$s_{s}$
is the water surface slope with respect to a horizontal plane. Defining the total shear stress as
$\tau _{xz}^{\textit{total}}=\tau _{xz}^{\textit{disp}}+\big\langle \tau _{xz}^{\textit{Rey}}\big\rangle +\langle \tau _{xz}^{v\textit{is}}\rangle$
, (2.3) becomes
2.1.2. The BBL thickness
In a channel with emergent vegetation, the water column can be divided into two layers. In the upper layer, which we call the vegetation layer, the momentum balance is dominated by vegetation drag, and the streamwise velocity is vertically uniform. In the near-bed layer, called the BBL, the bed shear stress makes a significant contribution to the momentum balance. Assuming steady, uniform flow, we integrate (2.6) across the vegetation layer (
$\delta \leq z\leq h$
) and the BBL (
$0\leq z\leq \delta$
), in which
$h$
and
$\delta$
represent the water depth and BBL thickness, respectively. In this study, the top of the BBL (
$z=\delta$
) is defined as the height at which the total shear stress becomes negligible. The integration yields
The total stress at the water surface
$\tau _{xz}^{\textit{total}}|_{z=h}$
is zero. The total stress at the bed is the bed shear stress
$\tau _{xz}^{\textit{total}}|_{z=0} = \tau _{\textit{bed}}$
. The term
$\tau _{xz}^{\textit{total}}|_{z=\delta }$
denotes the total stress at the top of the BBL, which is set to zero to be consistent with the definition of
$\delta$
. In the upper layer,
$\langle \overline{u(z)}\rangle$
is uniform in
$z$
and under the assumption that
$\delta \ll h$
, is equal to the pore velocity
$u_{p}=Q/[Bh(1-\phi )]$
, in which
$Q$
is the volumetric discharge, and
$B$
is the channel width. The velocity profile within the BBL is described by a shape factor
$\lambda = {\int _{z=0}^{z=\delta } \langle \overline{u (z)} \rangle ^{2}\mathrm{d}z} /(\delta {u_{p}^{2}})$
, with
$\lambda \lt 1$
. With these assumptions, (2.7a
) and (2.7b
) can be expressed as
In the upper layer (2.8a
), the hydrostatic pressure (
$\textit{gs}_{s}$
) is balanced by vegetation drag (
$C_{v}{u_{p}}^{2}$
). In the BBL (2.8b
), the hydrostatic pressure (
$\textit{gs}_{s}$
) is balanced by the sum of vegetation drag (
$\lambda C_{v}{u_{p}}^{2}$
) and bed shear stress (figure 1
a). The hydrostatic pressure (
$\textit{gs}_{s}$
) is the same in both layers. Assuming that the shape factor
$\lambda$
is a constant, balancing (2.8a
) and (2.8b
) predicts the scale of the BBL thickness,
(a) Schematic diagram showing the vertical distribution of the time and horizontally averaged streamwise velocity, the different vortex scales, and the force balance in the upper vegetation layer (light blue) and lower BBL (grey) within an array of rigid, emergent vegetation. Vegetation stems of diameter
$d$
are shown with vertical green bars. The stem diameter
$d$
and BBL thickness
$\delta$
are not shown to scale, but were chosen for visual clarity. In most conditions,
$\delta$
is smaller than
$d$
. (b) Characteristic length scales of the large, energy-containing eddy (
$\delta$
) and near-bed eddy (
$\eta$
) in vegetated flow over a smooth bed. (c) Characteristic length scales of the large, energy-containing eddy (
$\delta$
) and near-bed eddy (
$d_{s}$
) in vegetated flow over a rough bed.

which is the vertical extent over which the bed friction influences the velocity profile. As vegetation drag increases, the influence of the bed shear stress is confined to a thinner layer, i.e. smaller
$\delta$
.
2.2. Bed shear stress model
Gioia & Bombardelli (Reference Gioia and Bombardelli2001) (GB2001) used turbulence theory to derive a physics-based prediction of bed shear stress for hydraulically smooth and fully rough bare beds, providing a theoretical foundation for the empirical Blasius and Manning–Strickler laws. These regimes represent the asymptotic limits of a general phenomenological theory later proposed by Gioia & Chakraborty (Reference Gioia and Chakraborty2006), which reproduced the trends across the entire Nikuradse friction factor chart, and which has recently been extended to predict jet-induced scour (Bombardelli, Palermo & Pagliara Reference Bombardelli, Palermo and Pagliara2025). In the present study, given the limited availability of bed shear stress data in vegetated flows, we restrict our analysis to the limits of hydraulically smooth and fully rough regimes. This subsection first revisits the GB2001 model for bare bed conditions and then extends this framework to vegetated environments, developing a unified model that describes the transition from bare to vegetated beds.
2.2.1. Bed shear stress for bare bed conditions
For a channel of sufficient width to depth ratio, which can be approximated as two-dimensional, the bed shear stress
$\tau _{\textit{bed}}$
equals the turbulent momentum flux at
$z=0$
:
GB2001 described this momentum flux in terms of the eddy scales contributing to the streamwise (
$u'$
) and vertical (
$w'$
) turbulent fluctuations:
Here,
$V$
is the velocity scale of the large, energy-containing eddies of scale
$L$
, which carry the streamwise momentum, reflected in the streamwise velocity fluctuation (
$u'$
), and
$u_{r}$
is the characteristic velocity of the near-bed eddies of scale
$l_{r}$
that produce the wall-normal velocity fluctuations (
$w'$
) that facilitate momentum transfer to the bed.
(1) Hydraulically rough bed without vegetation. As described in GB2001, for a bare bed (without vegetation), the larger, energy-containing eddies have velocity scale
$V$
proportional to the bulk current velocity
$u_{c}$
and length scale
$L$
proportional to the hydraulic radius
$R$
, which is simply the water depth with our two-dimensional assumption. The near-bed eddies, which contribute the wall-normal velocity fluctuations, have size comparable to the roughness elements
$r$
(typically the grain size
$d_{s}$
). Eddies much larger than
$r$
are geometrically constrained from generating significant wall-normal velocity at the bed, while eddies much smaller than this do not contribute as much to momentum transfer. Thus
$l_{r}\sim r$
. The characteristic velocity of near-bed eddies,
$u_{r}$
, is determined by applying Kolmogorov’s inertial subrange scaling (Kolmogorov Reference Kolmogorov1941; Gioia & Bombardelli Reference Gioia and Bombardelli2001). Specifically, for
$r\gg \eta$
, with
$\eta$
the Kolmogorov scale, the characteristic velocity
$u_{r}$
is related to the eddy size
$l_{r}$
and the turbulent energy dissipation rate per unit mass
$\varepsilon$
(Kolmogorov Reference Kolmogorov1941):
For steady, uniform flow, it is reasonable to assume that
$\varepsilon$
equals the rate of turbulent energy production per unit mass by the energy-containing eddies,
$P$
. The energy per unit mass of the energy-containing eddies scales with
$V^{2}$
, and their characteristic time scale
$T\sim L/V$
, such that
Equating (2.12) and (2.13), GB2001 obtained the scaling
Substituting this into (2.11) gives
Substituting the characteristic scales
$l_{r}\sim r ,\ L\sim R,\ V\sim u_{c}$
, the scaling relationship for bed shear stress over rough bare beds is
This formulation is consistent with the classical Manning–Strickler law in rough channels (Strickler Reference Strickler1923; Katopodes Reference Katopodes2019).
(2) Hydraulically smooth flow without vegetation. In the case of a hydraulically smooth bed, the large-scale eddy is unchanged:
$L\sim R$
and
$V\sim u_{c}$
. However, the near-bed eddies cannot be governed by roughness geometry, which is absent or dynamically unimportant in hydraulically smooth flow. Instead, consistent with the framework of Gioia & Chakraborty (Reference Gioia and Chakraborty2006), the near-bed eddies are governed by the dissipative range of the turbulent energy spectrum. Consequently, the near-bed eddy scales with the Kolmogorov microscale,
$l_{r}\sim\eta$
, with characteristic velocity
$u_{r}$
given by the Kolmogorov velocity scale
$u_{\eta }$
. Specifically,
Again, under the assumption that
$\varepsilon \sim P\sim V^{3}/L$
(2.13),
$l_{r}$
and
$u_{r}$
can be expressed as:
By substituting (2.20) into (2.11), or equivalently, substituting (2.19) into (2.15), the bed shear stress for a hydraulically smooth bed becomes
Finally, substituting the large-eddy characteristics
$L\sim R$
and
$V\sim u_{c}$
into (2.21) yields the scale dependence
in which
${\textit{Re}}_{c}=u_{c}R/\nu$
is the bulk Reynolds number. Equation (2.22) is consistent with the Blasius law for turbulent flow over smooth surfaces (Chow Reference Chow1988).
2.2.2. Bed shear stress in vegetated flow
We adapted the framework of GB2001 to a channel with rigid, emergent vegetation in a uniform distribution by accounting for the impact of vegetation on the characteristic length and velocity scales for both the large-scale and near-bed eddies. For simplicity, we considered a wide channel, for which the hydraulic radius
$R$
reduces to the water depth
$h$
. The flow is assumed to be fully developed turbulent flow within the double-averaged framework.
(1) Hydraulically rough bed with vegetation. In a rough channel with vegetation, three eddy scales are present (figure 1
a). First, eddies generated by individual stems have a characteristic size set by the stem diameter
$d$
in sparse vegetation, and by the stem spacing in dense vegetation (Tanino & Nepf Reference Tanino and Nepf2008). However, these eddies are suppressed near the bed, since their generation is restricted by the presence of the solid boundary. Second, eddies generated by bed roughness scale with the grain diameter
$d_{s}$
(figure 1
c). Similar to bare beds (§ 2.2.1), these grain-scale eddies produce the wall-normal velocity fluctuations near the bed, such that
$l_{r}\sim d_s$
. Finally, the eddies generated within the BBL, with length scale
$\delta$
, are the large, energy-containing eddies within the framework of GB2001, such that
$L\sim \delta$
. The characteristic velocity of the large-scale eddies,
$V$
, depends on the vegetation density, with two possible regimes.
(a) Low vegetation density regime. If vegetation density, and thus vegetation drag, is small enough (small
$C_{v}$
), then the BBL (2.9) is large enough for the mean and turbulent velocity structure to resemble a bare bed channel. The large-eddy size is set by the boundary layer thickness (
$L\sim\delta$
). Since the large eddies are driven by the bulk flow momentum, their velocity
$V$
scales with the pore velocity
$u_{p}=Q/[Bh(1-\phi )]$
. By additionally choosing
$l_{r}\sim d_{s}$
for a rough bed, (2.15) produces an expression for bed shear stress:
In the limit approaching zero vegetation density (
$\phi \rightarrow 0$
), the pore velocity
$u_{p}$
approaches the current velocity
$u_{c}$
, the BBL thickness
$\delta$
approaches the water depth
$h$
, and (2.23) smoothly transitions to the rough, bare bed formulation (2.16).
To apply (2.23), we must also predict the BBL thickness
$\delta$
. By combining (2.23) with (2.9), we obtain scaling relationships for both the bed shear stress and BBL thickness in terms of vegetation drag coefficient
$C_{\!D}$
, stem volume fraction
$\phi$
, sediment size
$d_{s}$
, and pore velocity
$u_{p}$
:
\begin{equation}\delta \sim d_{s}^{1/4}C_{v}^{-3/4}\sim \left[\frac{\pi d_{s}^{1/3}d(1-\phi )}{2\phi C_{\!D}}\right]^{3/4}\quad\text{rough, low-density regime}.\end{equation}
(b) High vegetation density regime. As vegetation density and drag increase (larger
$C_{v}$
in (2.9)), the BBL thins. In this case, the large-scale eddies defined by the BBL,
$L\sim \delta$
, are confined more closely to the bed. It was hypothesised that under these conditions, the characteristic velocity of these eddies,
$V$
, will be reduced from the pore velocity
$u_{p}$
and be determined instead from an energy budget only within the BBL, i.e. within a control volume extending from the bed (defined at the roughness crest,
$z=0$
) to the top of the BBL (
$z=\delta$
). The large-scale eddies within the BBL are generated by the velocity gradient in this layer, so that their energy is derived from the work done by the mean flow against bed friction (
$u_{p}\tau _{\textit{bed}}$
). The rate of turbulent energy production within the BBL is
in which
$A$
is the bed area, and
$M=\rho A\delta (1-\phi )$
is the fluid mass within the control volume. From (2.13), the velocity scale of the large eddies is
$V\sim (\textit{PL})^{1/3}$
. Substituting
$P$
as defined in (2.26) yields the characteristic velocity of the large-scale eddies in the BBL:
Finally, substituting this expression for
$V$
and the characteristic length scales for rough bed conditions (
$L\sim \delta$
and
$l_{r}\sim d_{s}$
) into the stress formulation (2.15) leads to
from which we have
Combining the scaling in (2.29) with the expression for
$\delta$
(see (2.9)), we obtain the functional dependencies for
$\tau _{\textit{bed}}$
and
$\delta$
in the high vegetation density regime:
(2) Hydraulically smooth bed. For a hydraulically smooth bed, the scales of the energy-containing eddies are the same as for a rough bed (see previous section). Specifically,
$L\sim \delta$
, and
$V\sim u_{p}$
for low-density vegetation, and
$V\sim (u_{p}\tau _{\textit{bed}}/\rho )^{1/3}$
for high-density vegetation. Because the bed roughness has no dynamic significance in hydraulically smooth conditions, the length and velocity scales of the near-bed eddies are the Kolmogorov length and velocity scales, as described for a hydraulically smooth bare bed (GB2011). Specifically,
$l_{r}\sim \eta$
and
$u_{r}\sim u_{\eta }$
. Substituting these into the stress formulation (see (2.21)), we derive scaling relationships for bed shear stress in smooth hydraulic conditions for the low-density (2.32) and high-density (2.33) emergent vegetation:
where
${\textit{Re}}_{\delta }=u_{p}\delta /\nu$
is the BBL Reynolds number. In the limit of zero vegetation density, for which
$u_{p}\rightarrow u_{c}$
and
$\delta \rightarrow h$
, (2.32) reduces to the bare bed formulation for smooth channels (see (2.22)), consistent with the classical Blasius law. Ali & Dey (Reference Ali and Dey2022) also found
$\tau _{\textit{bed}}\sim {\textit{Re}}_{\delta }^{-1/4}$
in a spatially developing boundary layer,
$\delta (x)$
. The emergence of the same scaling exponent reinforces the universality of the momentum transfer mechanism on smooth boundaries. Finally, by combining the scales defined in (2.32) and (2.33) with the scale for
$\delta$
(see (2.9)), we obtain dependencies for the low-density and high-density vegetation regimes, respectively.
For the smooth, low-density regime:
For the smooth, high-density regime:
The vegetation density defining the transition between the low- and high-density regimes will be explored using data collected from several previous studies.
3. Comparison with experimental data
This section compares the scaling relations to datasets obtained from previous studies. Section 3.2 discusses the transition between low- and high-density vegetation regimes. Finally, § 3.3 determines the scale coefficients and evaluates the model’s performance.
Summary of datasets for bed shear stress in vegetated flows.

3.1. Datasets
As summarised in table 1, data were collected from several previous studies, each of which used a cylinder array as model vegetation. There are 54 cases for smooth bed conditions from Yang et al. (Reference Yang, Kerger and Nepf2015) and Conde-Frias et al. (Reference Conde-Frias, Ghisalberti, Lowe, Abdolahpour and Etminan2023), and 60 cases for rough bed conditions from Ishikawa, Sakamoto & Mizuhara (Reference Ishikawa, Sakamoto and Mizuhara2003) and Aliaga & Aberle (Reference Aliaga and Aberle2024, Reference Aliaga and Aberle2025). These datasets cover a wide range of vegetation density:
$\phi = 0.0025{-}0.25$
for smooth beds, and
$\phi = 0.00049{-}0.032$
for rough beds. The rough beds include grain diameters
$d_{s} = 0.0018$
and 0.0065 m. Aliaga & Aberle (Reference Aliaga and Aberle2024) used an artificial grass mat composed of 0.026 m stems at density 492 800 stems
$\text{m}^{-2}$
, for which they measured a bed friction factor
$C_{\!f}=0.0109$
. An equivalent grain size was determined using two empirical equations:
In (3.38), the relationship
$n=0.047d_{s}^{1/6}$
was used (Strickler Reference Strickler1923; Katopodes Reference Katopodes2019). These formulations are valid for hydraulically rough flow over a uniform bed, which was consistent with the conditions in the Aliaga & Aberle (Reference Aliaga and Aberle2024) dataset. Specifically, using the friction velocity derived from the reported bare bed stress (
$u_{*}=\sqrt{\tau _{\textit{bed}}/\rho }$
) and roughness height
$k_{s}$
approximated by the height of the artificial grass (0.026 m), the shear Reynolds number (
${\textit{Re}}_{*}=u_{*}k_{s}/\nu$
) was consistently above 500, far exceeding the critical value 70 for the fully rough regime (Soulsby Reference Soulsby1997). Using the bare bed friction factor and water depth (0.22 m) reported in Aliaga & Aberle (Reference Aliaga and Aberle2024), the equivalent grain sizes
$d_{s}$
extracted from (3.38) and (3.39) were 0.027 m and 0.021 m, respectively. The average of these two estimates,
$d_{s}=0.024$
m, agrees remarkably well with the physical height of the artificial grass stems (0.026 m), demonstrating the robustness of this estimation method. The value
$d_{s}=0.024$
m was used in all subsequent analyses of this study.
The definition of the BBL thickness varied between studies. Yang et al. (Reference Yang, Kerger and Nepf2015) defined the BBL thickness
$H_{v}$
(called the linear-stress layer in their study) as the height above the bed at which the parabolic portion of the velocity profile ends. In contrast, Conde-Frias et al. (Reference Conde-Frias, Ghisalberti, Lowe, Abdolahpour and Etminan2023) defined the BBL thickness
$\tilde{\delta }$
as the height above the bed at which the velocity first reaches its depth-uniform value. The majority of compiled data used the Conde-Frias et al. (Reference Conde-Frias, Ghisalberti, Lowe, Abdolahpour and Etminan2023) definition. In the present study, the BBL thickness, denoted as
$\delta$
, was conceptually defined as the height above which the total shear stress can be considered negligible. While these definitions will not yield identical values for the BBL thickness, it is a reasonable assumption that they are physically related and close to one another. Therefore, for the purpose of comparing our model with the experimental data, we assumed
The canopy drag parameter
$C_{v}$
(table 1) was estimated from geometric parameters provided in each study and the cylinder drag coefficient,
$C_{\!D}$
, estimated from the formulation of Cheng & Nguyen (Reference Cheng and Nguyen2011):
in which the vegetation Reynolds number is defined as
3.2. Transition between low- and high-density vegetation regimes
The vegetation density defining the transition between the low- and high-density regimes was determined by assuming that the model equations should connect smoothly between regimes. To collapse cases across the wide range of pore velocity, the bed shear stress was normalised by the pore velocity, forming a bed friction factor
$C_{\!f}=\tau _{\textit{bed}}/(\rho u_{p}^{2})$
. For convenience, the resulting friction factors are presented together with the BBL model below.
For the hydraulically rough bed:
\begin{equation} \quad C_{\!f}=\left\{\begin{array}{ll}\!\! \alpha _{r1}\left(\frac{d_{s}}{\tilde\delta}^{1/3}\right),&\text{low-density regime},\\[6pt] \!\!\alpha _{r2}\,\frac{d_{s}}{\tilde\delta},&\text{high-density regime}. \end{array}\right. \end{equation}
\begin{equation} \tilde{\delta }=\left\{\begin{array}{ll}\!\! \beta _{r1}d_{s}^{1/4}C_{v}^{-3/4},&\text{low-density regime},\\[3pt] \!\!\beta _{r2}d_{s}^{1/2}{C_{v}}^{-1/2}, &\text{high-density regime}. \end{array}\right. \end{equation}
\begin{equation} \quad C_{\!f}=\left\{\begin{array}{ll}\!\! \gamma _{r1}\left(C_{v}d_{s}\right)^{1/4},&\text{low-density regime},\\[3pt] \!\!\gamma _{r2}\left(C_{v}d_{s}\right)^{1/2},&\text{high-density regime}. \end{array}\right. \end{equation}
For the hydraulically smooth bed:
\begin{equation} C_{\!f}=\left\{\begin{array}{ll}\!\! \alpha _{s1}{\textit{Re}}_{\tilde{\delta }}^{-1/4},&\text{low-density regime},\\[3pt] \!\!\alpha _{s2}{{\textit{Re}}_{\tilde{\delta }}}^{-3/5},&\text{high-density regime}. \end{array}\right. \end{equation}
\begin{equation} \tilde{\delta }=\left\{\begin{array}{ll}\!\! \beta _{s1}\left(\frac{\nu }{u_{p}}\right)^{1/5}C_{v}^{-4/5},&\text{low-density regime},\\[3pt] \!\!\beta _{s2}\left(\frac{\nu }{u_{p}}\right)^{3/8}C_{v}^{-5/8},&\text{high-density regime}. \end{array}\right. \end{equation}
\begin{equation} C_{\!f}=\left\{\begin{array}{ll}\!\! \gamma _{s1}\left(\frac{C_{v}\nu }{u_{p}}\right)^{1/5},&\text{low-density regime},\\[3pt] \!\!\gamma _{s2}\left(\frac{C_{v}\nu }{u_{p}}\right)^{3/8},&\text{high-density regime}. \end{array}\right. \end{equation}
The BBL Reynolds number is
${\textit{Re}}_{\tilde{\delta }}=\tilde{\delta }u_{p}/\nu$
, and
$\alpha , \beta , \gamma$
are empirical coefficients. The transition between low- and high-density regimes was defined by the value of vegetation density (
$\phi$
), captured in
$C_{v}=2\phi C_{\!D}/[\pi d(1-\phi )]$
, at which
$C_{\!f}$
and
$\tilde{\delta }$
yielded the same values for both regimes. For hydraulically rough beds, this required
Similarly, the transition for a smooth bed was defined by
Experimental data were used to define the transition values of
$d_{s}C_{v}$
and
${C_{v}\nu }/{u_{p}}$
in (3.49) and (3.50), respectively. Because not all datasets included measurements of the BBL thickness (
$\tilde{\delta }$
), we focused on the friction factor (
$C_{\!f}$
).
Figure 2 plots the bed friction factor
$C_{\!f}$
against
$d_{s}C_{v}$
for rough beds and
$\nu C_{v}/u_{p}$
for smooth beds. For rough bed cases (figure 2
a), a shift occurred around
$d_{s}C_{v}\approx 0.0013$
. Specifically, for
$d_{s}C_{v} \lt 0.0013$
, the data followed the low-density regime (solid line in figure 2
a), and for
$d_{s}C_{v} \gt 0.0013$
, the data followed the high-density regime (dashed line in figure 2
a). The data from Ishikawa et al. (Reference Ishikawa, Sakamoto and Mizuhara2003) exhibit some scatter for fixed values of
$d_{s}C_{v}$
, which was likely due to residual velocity dependence not captured by the fully rough flow assumption. However, the model successfully captured the primary relationship between the friction factor and both vegetation parameters and sediment grain size (combined as
$d_{s}C_{v}$
). For smooth bed conditions, nearly all data fell along the high-density regime (dashed line in figure 2
b), corresponding to
$C_{\!f}=(C_{v}\nu /u_{p})^{3/8}$
. Lacking sufficient cases in the low-density regime (solid line in figure 2
b), it was not possible to define the transition point for smooth bed conditions.
Comparison of measured friction factors
$C_{\!f}$
to models in low-density (solid lines) and high-density (dashed lines) regimes, for (a) rough beds and (b) smooth beds. Symbols *,
and
represent data reported in previous studies, as noted in the legends. In (b), ‘num’ and ‘exp’ denote numerical and experimental data from Conde-Frias et al. (Reference Conde-Frias, Ghisalberti, Lowe, Abdolahpour and Etminan2023).

3.3. Model performance
As shown in figure 2(b), data were unavailable for the low-density, smooth bed regime (
$\alpha _{s1} , \beta _{s1} , \gamma _{s1}$
), leaving nine empirical coefficients that could be calibrated ((3.43)–(3.48)). The calibration procedure is described in Appendix A, with the final values and their 95 % confidence intervals (CI) summarised in table 2. The entire compiled dataset was used for calibration to ensure that the coefficients were constrained by the widest possible range of flow, bed and vegetation conditions, which maximised the robustness of the calibrated model coefficients. Consequently, the comparisons below represent the model’s calibration performance and are not a validation. Due to the scarcity of data, the acquisition of new data to facilitate a validation remains a necessary future step.
Summary of calibrated model coefficients.

Using the calibrated coefficients, the predicted BBL thickness (
$\tilde{\delta }$
) had good agreement with the measured data for both rough and smooth bed conditions (figure 3). The model also produced good predictions for the dimensionless friction factor (
$C_{\!f}$
) and the friction velocity (
$u_{*}=u_{p}\sqrt{C_{\!f}}$
). The prediction of
$C_{\!f}$
from measured
$\tilde{\delta }$
((3.43), (3.46)) had strong agreement with measured
$C_{\!f}$
for both rough beds (figure 4
a) and smooth beds (figure 5
a). The low uncertainty in the coefficients (table 2) supported the key theoretical assumption that
$\tilde{\delta }$
is the characteristic length scale of large eddies. Furthermore, the prediction of
$C_{\!f}$
directly from the external parameters
$d_{s} , C_{v} , u_{p}$
(see (3.45), (3.48)) also had good agreement with the measured values (figures 4(b) and 5(b)), consistent with the low uncertainties of the fitted coefficients (table 2). This demonstrated the practical utility of the model to predict bed friction without prior knowledge of the BBL thickness. Finally, the predicted and measured friction velocity (
$u_{*}$
) had good agreement across all cases (figures 4(c,d) and 5(c,d)).
Taken together, (3.43)–(3.48) form a unified model that describes the continuous change in bed shear stress and BBL thickness from bare to vegetated conditions for both smooth and rough beds. The change in model regimes with increasing vegetation density, represented by the dimensionless
$C_{v}d_{s}$
, is shown for a rough bed in figure 6. The transition for smooth beds is not presented, as all available data for smooth beds fell in the high-density regime.
Agreement between measured BBL thickness
$\tilde{\delta }_{Mea}$
and predicted BBL thickness
$\tilde{\delta }_{\textit{Pre}}$
under (a) rough bed conditions, and (b) smooth bed conditions. Symbols
and
represent data reported in previous studies, as noted in the legends; ‘num’ and ‘exp’ denote the numerical and experimental data of Conde-Frias et al. (Reference Conde-Frias, Ghisalberti, Lowe, Abdolahpour and Etminan2023).

Rough bed conditions: comparison of measured and predicted values of the bed friction factor (
$C_{\!f}$
) and the friction velocity (
$u_{*}=u_{p}\sqrt{C_{\!f}}$
). (a,c) Predictions from (3.43), using measured BBL thickness. (b,d) Predictions from (3.45), using parameters
$d_{s} \it{,} C_{v} \it{,} u_{p}$
. Symbols *,
and
represent data reported in previous studies, as noted in the legend.

Smooth bed conditions: comparison of measured and predicted values of the bed friction factor (
$C_{\!f}$
) and the friction velocity (
$u_{*}=u_{p}\sqrt{C_{\!f}}$
). (a,c) Predictions from (3.46), using measured BBL thickness. (b,d) Predictions from (3.48) using parameters
$d_{s} , C_{v} , u_{p}$
. Symbols *,
and
represent data reported in previous studies; ‘num’ and ‘exp’ denote the numerical and experimental data of Conde-Frias et al. (Reference Conde-Frias, Ghisalberti, Lowe, Abdolahpour and Etminan2023).

Unified model for BBL thickness
$\tilde{\delta }$
and bed friction factor
$C_{\!f}$
across different regimes for a rough bed. The background colours denote the bare bed (yellow), low-density vegetation (green) and high-density vegetation (blue) regimes. The vertical dashed lines mark the transitions between these regimes. The solid black curve represents the unified model (3.43)–(3.45), including the specific formulations for each regime. Circles with error bars represent the experimental data, which support the theoretical model dependencies.

Three regimes are highlighted by different background colours in figure 6. The yellow region corresponds to extremely weak vegetation drag (
$C_{v}d_{s}\rightarrow 0$
), such that the BBL thickness and bed shear stress are essentially the same as for a bare bed, with
$\tilde{\delta }=h$
. To facilitate the visualisation of this quasi-bare-bed regime, water depth
$h=10d_{s}$
was assumed. The green and blue regions represent the low- and high-density vegetation regimes, respectively. The circles are experimental data. For cases with identical vegetation density, stem diameter and grain size, but different channel velocity, the circle represents the average, and the error bar indicates the variation over different velocities. The transitions between the three regimes are marked by vertical dashed lines in figure 6.
The first transition, from bare bed to the low-density canopy, is at
$C_{v}d_{s}=0.00012$
, the value at which
$\tilde{\delta }=h$
(3.44). For
$\tilde{\delta }=h$
, (3.43) reduces to
$C_{\!f}=(0.0213\pm 0.0014)(d_{s}/h)^{1/3}$
, which is consistent with the empirical friction factor for bare bed,
$C_{\!f}=0.022(d_{s}/h)^{1/3}$
(Strickler Reference Strickler1923). The second transition, from low-density to high-density vegetation, is at
$C_{v}d_{s}=0.00128$
, as defined in figure 2(a). At this point,
$\tilde{\delta }=1.7d_{s}$
. After the transition from bare bed to low-density vegetation, the BBL thickness decreases rapidly with increasing
$C_{v}d_{s}$
. Correspondingly, the bed friction factor
$C_{\!f}$
increases in proportion to
$(C_{v}d_{s})^{1/4}$
. In contrast, in the high-density regime, the BBL thickness decreases more gradually with
$C_{v}d_{s}$
, but since the BBL is much thinner in this regime, the friction factor increases more quickly with vegetation density, with
$C_{\!f}\sim (C_{v}d_{s})^{1/2}$
.
4. Model summary and limitations
It is useful to review the physical mechanisms captured by the model that govern the BBL thickness and bed friction. First, consider the BBL thickness
$\tilde{\delta }$
, which defines the vertical extent of the bed’s influence relative to the influence of the vegetation drag. Consistent with this definition,
$\tilde{\delta }$
scales with the ratio of bed shear stress to vegetation drag (2.9), and
$\tilde{\delta }$
decreases as canopy drag
$C_{v}=2\phi C_{\!D}/[\pi d(1-\phi )]$
increases. For rough beds, a larger sediment size
$d_{s}$
increases the size of the near-bed eddies, which consequently enhances the momentum transfer towards the bed, and thus the bed shear stress (2.15). Because the bed shear stress increases with sediment size, the BBL thickness also increases with sediment size (3.44). In contrast, for smooth beds, the size of the near-bed eddies and boundary layer is controlled by the pore velocity. Specifically, as the pore velocity
$u_{p}$
increases, the turbulent energy passed through the inertial subrange also increases, which reduces the size of the near-bed Kolmogorov eddies and the BBL thickness (3.47). This trend is consistent with Etminan et al. (Reference Etminan, Ghisalberti and Lowe2018), who also observed that the BBL thickness over smooth vegetated beds decreased with increasing pore velocity, and it is also consistent with a bare smooth boundary – e.g. for the Blasius boundary layer,
$\tilde{\delta }\sim u^{-1/2}$
.
Next, consider the bed friction factor
$C_{\!f}=\tau _{\textit{bed}}/u_{p}^{2}$
. From (2.15),
$C_{\!f}=(V/u_{p})^{2}{} (l_{r}/\tilde{\delta })^{1/3}$
. A higher friction factor will primarily result from an increase in the near-bed eddy size
$l_{r}$
or a decrease in the BBL thickness
$\tilde{\delta }$
. For rough beds,
$l_{r}\sim d_{s}$
, and
$\tilde{\delta }$
decreases with increasing vegetation drag (
$C_{v}$
), such that
$C_{\!f}$
increases with increasing grain size (
$d_{s}$
) and increasing canopy drag (3.45). The dependencies are stronger in the high-density regime,
$C_{\!f}\sim {C_{v}}^{1/2}{d_{s}}^{1/2}$
, compared to the low-density regime,
$C_{\!f}\sim {C_{v}}^{1/4}{d_{s}}^{1/4}$
. This is because for the low-density regime,
$V/u_{p}$
is a constant, but for the high-density regime,
$V/u_{p}$
increases with increasing canopy drag. Specifically, an increase in
$C_{v}$
or
$d_{s}$
boosts the energy production rate within the boundary layer (
$P$
, see (2.26)), which increases the large-eddy characteristic velocity
$V$
(i.e.
$C_{v}d_{s}\uparrow \Rightarrow \tau _{\textit{bed}}\uparrow \Rightarrow P\uparrow \Rightarrow V\uparrow$
), which in turn amplifies the bed shear stress. Note that for rough beds,
$C_{\!f}$
had no dependence on channel velocity (3.45), which is consistent with the bare bed, hydraulically rough regime. For smooth beds,
$C_{\!f}$
depends on the channel velocity (3.48). Similar to smooth bare beds, for smooth vegetated beds, an increase in
$u_{p}$
reduces the near-bed eddy size (
$\eta$
) and velocity (
$u_{\eta }$
), leading to a lower
$C_{\!f}$
(see (3.48)). For vegetated beds,
$C_{\!f}$
additionally depends on the vegetation drag, because the boundary layer, which sets the scale of the energy-containing eddies, decreases as vegetation drag (
$C_{v}$
) increases. The dependency on
$C_{v}$
and
$u_{p}$
is stronger in the high-density regime (
$C_{\!f}\sim (C_{v}\nu /u_{p})^{3/8}$
) than in the low-density regime (
$C_{\!f}\sim (C_{v}\nu /u_{p})^{1/5}$
), due to the increase in the large-eddy characteristic velocity
$V$
as the boundary layer thins, which was described above for the rough boundary regime.
Finally, we consider the limitations of the model. The model assumes that the large eddies,
$L$
, scale with the BBL thickness (
$L\sim \tilde\delta$
) and that the near-bed eddies,
$l_{r}$
, scale with the sediment grain size (
$l_{r}\sim d_{s}$
) for rough beds or the Kolmogorov microscale (
$l_{r}\sim \eta$
) for smooth beds. A fundamental requirement is that the large-scale eddies be larger than the near-bed eddies, i.e.
$L\gt l_{r}$
. For smooth beds, this condition is always satisfied, since the Kolmogorov microscale
$\eta$
is always small compared to the boundary layer. However, for rough beds, the condition requires
$\tilde\delta \gt d_{s}$
. For the experimental data used in this study, the ratio
$\tilde{\delta }/d_s$
ranged from 0.74 to 8.4, with the majority of data (82 %) satisfying
$\tilde\delta \gt d_{s}$
. The few cases for which this condition was not met (
$\tilde\delta \lt d_{s}$
) occurred in two specific situations: first, at the highest vegetation density tested (
$\phi =3.2\,\%$
and
$C_{v}d_{s} = 0.0067$
), for which the BBL was the thinnest; second, for the cases with the largest equivalent roughness (
$d_{s}=0.024$
m), where the measured BBL thickness (0.022 m) was slightly smaller than the roughness scale. Notably, even for these cases, the model prediction of bed shear stress was reasonable, as shown by the data points at the highest
$C_{v}d_{s}$
in figure 6(b). Finally, the framework for predicting BBL thickness and bed shear stress was developed for homogeneously distributed, rigid, emergent vegetation. The degree to which this framework applies to real vegetation, for which flexibility and more complex morphology may alter near-bed turbulence, remains an open topic for future study.
5. Conclusion
Building on a phenomenological theory for turbulent stress and a two-layer force balance analysis, this study developed a new theoretical framework for bed shear stress and bottom boundary layer (BBL) thickness in vegetated channels for both rough and smooth beds. A key feature of the model was the identification of the BBL thickness, rather than the flow depth, as the scale parameter for the energy-containing eddies that contribute to the momentum flux at the bed, determining the bed shear stress. The framework provides a unified description of how bed shear stress and BBL thickness continuously transition from bare bed to vegetated conditions. In the limit of zero vegetation density, the bed shear stress model reduces to the classic bare bed formulation, and the BBL thickness equals the water depth. The models were compared to 114 measurements from previous studies over both smooth and rough vegetated beds.
Acknowledgements
The authors gratefully acknowledge insightful and constructive discussions with Y. Lu during the course of this research.
Funding
This research was sponsored by the National Natural Science Foundation of China (grant nos 12272349, 52339005), the HPC Centre of ZJU (Zhoushan Campus), and the Truth-Seeking Flying Eagle Programme of Zhejiang University.
Declaration of interests
The authors report no conflict of interest.
Data availability statement
All data used for this study can be found in the published literature, as summarised in table 1.
Appendix A. Fitting coefficients
The complete model framework (3.43)–(3.48) contains 12 empirical coefficients. No data were available for the low-density smooth regime, which prevented the calibration of three coefficients (
$\alpha _{s1} , \beta _{s1} , \gamma _{s1}$
). This left nine coefficients to be determined. These coefficients are not independent, as (3.45) and (3.48) can be derived from (3.43)–(3.44) and (3.46)–(3.47), respectively. The following relationships must hold:
Additionally, assuming a smooth transition between low- and high-density regimes, for the rough beds (figure 2 a), we have
Therefore, for the nine coefficients (
$\alpha _{r1} , \alpha _{r2} , \beta _{r1} , \beta _{r2} , \gamma _{r1} , \gamma _{r2} , \alpha _{s2} , \beta _{s2} , \gamma _{s2}$
), there are four internal relationships (see (A1a
)–(A1c
) and (A2)), leaving only five independent coefficients to be determined.
First, the rough-bed datasets were classified into low- and high-density regimes based on the transitional criterion for rough beds (
$d_{s}C_{v}\approx 0.0013$
) identified in figure 2(a). The nine experimental runs from Ishikawa et al. (Reference Ishikawa, Sakamoto and Mizuhara2003) have
$\phi =0.00805$
and
$d_{s}C_{v}\approx 0.0013$
, which is very close to the transition between regimes. Therefore, these cases were used in calibrating both regimes. The nine coefficients were calibrated by maximising the sum of the coefficients of determination (
$R^2$
) for bed friction factor, friction velocity and BBL thickness, thereby achieving the best overall fit to the observed data (table 2). In addition, the 95 % confidence intervals for each coefficient were estimated using a bootstrap sampling method.
























