3.1. Mean flow quantities

To investigate the flow above and within the canopy for different values of $Ca$, we start looking at the mean profiles of the velocity and of the Reynolds stresses, along with relevant derived quantities. In the following, averaging in time and along the homogeneous directions is denoted with angle brackets, while fluctuations with respect to such a mean are marked with an apostrophe. Our analysis is based on $100$ flow fields for each value of $Ca$, regularly collected over $50$ bulk time units. The flow statistics do not show any appreciable variation upon computation with half of the fields, thus confirming that they are converged.

The mean profiles of the streamwise velocity $u$, shown in figure 3(a), exhibit two distinct inflection points: an outer inflection point ($y_{out}$, square symbol) generated by the drag discontinuity at the average canopy tip position and an inner inflection point ($y_{in}$, triangle symbol) closer to the wall, where the inflected velocity profile connects to the wall boundary layer. Below the outer inflection point the inner flow is reminiscent of that attained in an anisotropic porous medium (Rosti, Brandt & Pinelli Reference Rosti, Brandt and Pinelli2018b), while immediately above that the outer flow is similar to a turbulent mixing layer (Raupach et al. Reference Raupach, Finnigan and Brunei1996). Yet, canopy flows can also be considered instances of obstructing substrates for the assessment of outer-layer similarity (Chen & García-Mayoral Reference Chen and García-Mayoral2023). Fitting a logarithmic profile to the mean velocity well above the canopy is in fact a convenient numerical expedient to simplify its parametrisation for modelling purposes, even though it does not satisfactorily represent the physical features of the flow. We therefore compute the virtual origin of the outer flow, $y_{vo}$, imposing the matching with a canonical logarithmic profile and notice that, for all the cases of interest, it lays well between the two inflection points (figure 3b), thus confirming that we are in a dense canopy regime (Monti et al. Reference Monti, Omidyeganeh, Eckhardt and Pinelli2020,  Reference Monti, Nicholas, Omidyeganeh, Pinelli and Rosti2022). The canopy becomes more compliant to the flow as $Ca$ increases, hence, the mean position of its tip as well as all the other relevant points are monotonously shifted downward, maintaining their relative order; the mean streamwise velocity at those points, instead, exhibits a non-monotonous trend. At the canopy tip, in particular, it first undergoes a slight increase due to the reduction of the filaments drag, later decreasing again as the effect of the downward shift becomes dominant. The positions of all points reach a plateau for high values of $Ca$, where the vertical stacking of the deflected filaments poses a lower bound to the thickness of the inner flow region.

Figure 3. Mean profiles of the streamwise velocity (a) for different values of $Ca$ and associated relevant points (inset and (b)). In (b) we show the position of the relevant points for different values of $Ca$, maintaining the vertical scale unchanged with respect to that of (a).

As expected, on scaling the velocity profile of the inner flow with the friction velocity computed at the wall, $u_{\tau }^{in}=\sqrt {\tau _w/\rho _f}=\sqrt {\nu \partial \langle u\rangle / \partial y \vert _0}$, the typical trend ${\langle u\rangle }/{u_{\tau }^{in}} = y u_{\tau }^{in} / \nu$ is recovered close to the bottom wall (figure 4a). Instead, on scaling the velocity profile of the outer flow with the friction velocity computed at the virtual origin, $u_{\tau }^{out}=\sqrt {\nu \partial \langle u\rangle / \partial y \vert _{y_{vo}}- \langle u'v' \rangle \vert _{y_{vo}}}$, we confirm good agreement with a logarithmic profile (figure 4b) of the form

(3.1)\begin{equation} \frac{\langle u\rangle}{u_{\tau}^{out}} = \frac{1}{\kappa} \log\left( \frac{(y-y_{vo})u_{\tau}^{out}}{\nu}\right)+B-{\rm \Delta} u^{+}_{out}, \end{equation}

where $\kappa =0.41$ and $B=5.2$, while ${\rm \Delta} u^{+}_{out}$ denotes the friction function accounting for the mean velocity shift in the outer flow due to the presence of the canopy (or wall roughness, as in Jiménez Reference Jiménez2004). As noted by Monti et al. (Reference Monti, Nicholas, Omidyeganeh, Pinelli and Rosti2022), to whom we compare our results, ${\rm \Delta} u^{+}_{out}$ exhibits an exponential trend with the driving pressure gradient $\mathrm {d}P/\mathrm {d}\kern 0.06em x$ (figure 4c), once made dimensionless upon the height of the channel above the virtual origin. We acknowledge a minor deviation from their data, that we impute to the different shape of our filaments and to their movement. Differently from ours, in fact, the study of Monti et al. (Reference Monti, Nicholas, Omidyeganeh, Pinelli and Rosti2022) only concerned rigid canopies: in particular, at a chosen value of the Reynolds number ($Re_b=6000$, higher than ours), they varied the canopy solidity $\lambda$ by changing the inclination of the filaments.

Figure 4. Plot (a) reports the first five computed points of the mean velocity profiles, made dimensionless upon the friction velocity at the wall, $u_{\tau }^{in}$, against the wall distance. Plot (b), instead, reports the mean velocity profiles made dimensionless upon the friction velocity at the virtual origin, $u_{\tau }^{out}$, against the wall distance shifted by $y_{vo}$, within the region where the scaling holds. In (b), for each case, we also report the logarithmic profile computed with (3.1) as a dash-dotted grey line. The inner scaling yields good overlapping of the different profiles with a quasi-linear trend over the first grid points off the wall while, with the outer scaling, the profiles collapse on the analytical predictions. Finally, in (c) we show the exponential trend of the friction function ${\rm \Delta} u^{+}_{out}$ appearing in (3.1) with respect to the driving pressure gradient, $\mathrm {d}P/\mathrm {d}\kern 0.06em x$, and compare it with the numerical data of Monti et al. (Reference Monti, Nicholas, Omidyeganeh, Pinelli and Rosti2022), who studied rigid canopies with different inclinations.

The diagonal components of the Reynolds stress tensor (in figures 5 and 6a) increase monotonously moving away from the wall in the canopy region, and peak at a position variable with $Ca$. They therefore decrease towards the centre of the channel until the no-penetration condition becomes relevant and damps the fluctuations of the wall-normal velocity, enhancing those of the wall-parallel components because of continuity. All the peaks move towards the wall as $Ca$ increases and the filaments get more deflected by the action of the fluid, but the effect appears to saturate for the highest values of $Ca$. The fluctuations of the wall-normal and spanwise velocity components, shown in figure 5, always reach their peak value above the canopy, highlighting the intense turbulent activity caused by the unstable shear layer at the drag discontinuity. The maximum in the fluctuations of the streamwise velocity (reported in figure 6a), instead, lays close to the canopy tip for the lowest values of $Ca$ and moves slightly above that for the highest values. This picture is compatible with the existence of high and low streamwise velocity regions at the canopy tip, induced by the overlying Kelvin–Helmholtz-like instability. Those velocity structures alternatively deflect the filaments and penetrate the upper region of the canopy; nevertheless, for the most flexible cases, the filaments are significantly bent forward and, therefore, shield the inner flow, causing a slight shift up of those structures with respect to the canopy tip. The streamwise mean momentum equation imposes

(3.2)\begin{equation} \mathrm{d}\tau/\mathrm{d}y = \mathrm{d}P/\mathrm{d}x, \end{equation}

where the total shear stress writes $\tau =\rho \nu \,\mathrm {d}\langle u\rangle /\mathrm {d}y - \rho \langle u'v'\rangle + D_c$, where $D_c$ is the canopy drag. Denoting with $\tau _w$ the total shear stress at the wall, the sum of the three components constituting $\tau$ is therefore constrained by $\tau (y)=\tau _w(1-y/H)$. Observing the separate contributions to $\tau$ normalised by $\tau _w$, in figure 6(b) we note that the viscous shear stress ($\rho \nu \, \mathrm {d}\langle u\rangle /\mathrm {d}y$, plotted with dash-dotted lines) has two local maxima, one at the wall and one at the canopy tip, while it is negligible elsewhere. Inside the canopy, the stress is dominated by the drag contribution $D_c$ (plotted with dashed lines), which vanishes when moving out of it, giving way to the turbulent shear stress ($- \rho \langle u'v'\rangle$, plotted with continuous lines), which dominates the outer region as in conventional turbulent channel flows. We once again notice how the transition from the inner to the outer regions occurs closer to the wall for growing values of $Ca$ due to the deflection of the filaments; nevertheless, the variation of their flexibility affects also the intensity of the shear layer above them. The viscous shear stress exhibits a sharp peak at the canopy tip for the lowest values of $Ca$, while for the highest values, the peak is less pronounced and spans a wider vertical span. Indeed in this case, the shear layer is less definite and penetrates more into the canopy due to the filaments motion. Integrating (3.2) in the wall-normal direction, we note that the sum of the different integral components of the shear stress decreases as $Ca$ increases and plateaus for the most flexible cases (as in figure 7), always matching the streamwise pressure gradient needed to drive the flow at a constant value of $Re_b=5000$. The component coming from the viscous shear remains small and practically constant across all the cases, while those induced by the turbulent shear and by the canopy drag significantly decrease. We impute the depletion of the former to a lower level of turbulent activity, associated to a weaker shear layer, while the latter is mainly reduced by the compliant nature of the filaments and the consequent reduction of the frontal canopy area.

Figure 5. Fluctuations of the wall-normal (a) and spanwise (b) velocity components for different values of $Ca$. The positions of the canopy tip (identified with the outer inflection point) are denoted by vertical dashed lines.

Figure 6. Fluctuations of the streamwise velocity component (a) and shear stress balance (b) for different values of $Ca$. In (a) the positions of the canopy tip (identified with the outer inflection point) are denoted by vertical dashed lines. In (b) the total shear stress (black line) normalised by the wall shear stress is given by the sum of the turbulent shear stress (continuous lines), the viscous shear stress (dash-dotted lines) and the canopy drag (dashed lines), as described in the main text.

Figure 7. Decomposition of the driving streamwise pressure gradient $\mathrm {d}P/\mathrm {d}\kern 0.06em x$ into the contributions of the viscous and turbulent shear stresses along with the canopy drag, integrated across the wall-normal direction, for different values of $Ca$.

3.2. Canopy drag

Here we focus on the measurement of the canopy drag coefficient, $C_d$. First, in close analogy to Ghisalberti & Nepf (Reference Ghisalberti and Nepf2006), we define

(3.3)\begin{equation} C_d a (y)= \frac{\left.\dfrac{\partial \langle u' v' \rangle}{\partial y}\right\rvert_{y_{out}< y< H} - \dfrac{\partial \langle u' v' \rangle}{\partial y}(y)}{\dfrac{1}{2}\langle u \rangle^2 (y)}, \end{equation}

where $a$ is the frontal canopy area per unit volume, and the first term in the numerator denotes the mean vertical gradient of the Reynolds’ shear stress above the canopy tip, $y_{out}$. We thus compute the mean value of $C_d a$, denoted with angle brackets, between the inner and the outer inflection points for different values of $Ca$. This approach circumvents the complexity of the computation of $a$ (Ghisalberti & Nepf Reference Ghisalberti and Nepf2006), but yields a dimensional quantity that we thus make dimensionless upon multiplication with the channel height. The outcome is compared with the experimental measurements of Ghisalberti & Nepf (Reference Ghisalberti and Nepf2006) in figure 8(a), highlighting a good correspondence between the two for the most flexible cases considered in our investigation. The data appear to tend towards $\langle C_d a \rangle H \sim Ca^{-1/4}$ at large values of $Ca$.

Figure 8. Canopy drag measurements. We show (a) the mean value of $C_d a$ from our simulations, compared with that measured experimentally by Ghisalberti & Nepf (Reference Ghisalberti and Nepf2006). We also report (b) the value of the canopy drag, $C_d$, throughout our simulations.

In order to decouple the value of $C_d$ from $a$, we consider the integral form of the streamwise momentum balance

(3.4)\begin{equation} -H\frac{{\rm d} P}{{\rm d}x}=\tau_w + \int_0^{y_{out}}D_c(y)\,{{\rm d}y}, \end{equation}

where the streamwise pressure gradient ${{\rm d} P}/{{\rm d}x}$ equates the sum of the mean shear stress at the wall, $\tau _w$, and of the canopy drag, $D_c(y)$, integrated across the canopy height. Thus, $C_d$ can be computed through such balance, referring it to the frontal canopy area,

(3.5)\begin{equation} C_d= \frac{L_x L_z}{H L_z} \frac{\displaystyle\int_0^{y_{out}}D_c(y) \,{{\rm d}y}}{\dfrac{1}{2}\rho U_b^2} ={-} \frac{L_x}{H} \frac{H\dfrac{{\rm d} P}{{\rm d}x}+\tau_w}{\dfrac{1}{2}\rho U_b^2}. \end{equation}

The outcome of this approach in our simulations is reported in figure 8(b). The value of $C_d$ in the rigid case is compatible with what is reported in the literature for canopies with similar properties (Raupach & Thom Reference Raupach and Thom1981; Shimizu et al. Reference Shimizu, Tsujimoto, Nakagawa and Kitamura1992; Finnigan Reference Finnigan2000; Ghisalberti & Nepf Reference Ghisalberti and Nepf2006; Nepf Reference Nepf2012a), where $C_d \approx 0.5\unicode{x2013}1.5$, and quickly decreases reducing the rigidity of the filaments, reaching a plateau for $Ca \gtrapprox 50$. Overall, the drag measurements reported here are in good agreement with the experimental measurements and shed light on the trend of $C_d$ with $Ca$.

3.3. Energy spectra

The amount of kinetic energy retained by the turbulent fluctuations of the flow is quantified by the mean TKE, defined as $K(y)=0.5(\langle u' u' \rangle + \langle v' v' \rangle +\langle w' w' \rangle )$. Nevertheless, the space averaging operation cancels any information about the distribution of the kinetic energy across the different scales of motion. To this purpose, we therefore resort to the spatial spectrum of the TKE, $E(k_x,k_z;y)=\overline {\mathcal {F}_x(\mathcal {F}_z(0.5(u' u'+v' v'+w' w')))}$, where $\mathcal {F}$ denotes the Fourier transform operator across either of the homogeneous directions and the overbar indicates averaging in time. Integration along the spanwise wavenumbers $k_z=2{\rm \pi} i/L_z$ (with $i \in \mathbb {N};\ i = 1,\ldots, n_z$) yields the streamwise spectrum $E_x(k_x;y)$ as a function of the streamwise wavenumber $k_x$ and the wall-normal coordinate; with a similar procedure we also attain the spanwise spectrum $E_z(k_z;y)$. In figure 9(a,b) we inspect the profiles of $E_x$ and $E_z$, respectively, for different values of $Ca$ at the inner inflection point, $y=y_{in}$, at the outer inflection point, $y=y_{tip}$, and above the canopy, at $y=H/2$. We also report the typical $-5/3$ slope for comparison. Within the canopy, both spectra exhibit a non-monotonous behaviour characterised by a first peak at the wavenumber associated to the channel height and a second, sharper one, associated to the mean separation of the filaments. In between the two, the spectra approach a $-5/3$ slope when $Ca$ increases. The filaments absorb energy in the shear production range, close to the largest scales of motion, and re-inject it in the flow through their wakes and their waving motion, in close agreement to the spectral short cut process described by Finnigan (Reference Finnigan2000) and Olivieri et al. (Reference Olivieri, Brandt, Rosti and Mazzino2020). A variation in the $Ca$ is observed to affect both the spectral short cut and the amplitude of the spectra. In fact, the spectral short cut is more intense in the case of rigid filaments, where the spectra almost approach a plateau before their second peak, while it appears less accentuated in the cases at higher $Ca$, consistent with the observations of Olivieri, Cannon & Rosti (Reference Olivieri, Cannon and Rosti2022a). A more regular decay of the spectra is recovered moving up to the drag discontinuity at the canopy tip. In particular, while $E_x$ decreases monotonously from the largest scales of motion, $E_z$ peaks at $k_z\sim {O}(10)$: this behaviour, which is also found in the outer flow at $y=H/2$, appears compatible with the presence of large structures dominating the outer flow above the canopy (Monti et al. Reference Monti, Olivieri and Rosti2023). Conventional full spectra are observed in the outer flow, denoting the persistence of a fully developed turbulent state there. Interestingly, the amplitude of the spectra within the canopy appears to increase with $Ca$, while an opposite trend is observed outside. An increase in the flexibility of the filaments is in fact associated to more intense velocity fluctuations at the inner inflection point, located deep in the canopy and marginally affected from the outer flow turbulence, due to the increased motion of the filaments. At and above the canopy tip, instead, more flexible filaments are associated to the generation of a weaker shear layer with respect to the rigid case due to the lower drag discontinuity at their tip.

Figure 9. (a) Streamwise and (b) spanwise one-dimensional spectra of the TKE, integrated along the remaining homogeneous direction, for different values of $Ca$. The spectra are sampled (left) at the inner inflection point, (centre) at the canopy tip and (right) in the outer flow.

3.4. Lumley triangle

The DNS of turbulent canopy flows is a challenging task that might provide an excessive level of detail for most engineering applications, where an accurate model of the principal phenomena governing the fluid motion can instead prove sufficient. Turbulence models often rely on a detailed description of the Reynolds stress anisotropy tensor $\boldsymbol {b}$, which we therefore thoroughly characterise in this section resorting to the Lumley triangle formalism. As outlined by Pope (Reference Pope2000), the turbulent stress state can be described with only two scalar variables, $\xi$ and $\eta$, with the latter quantifying the anisotropy of the stress state. Adopting the index notation and implying Einstein's summation over repeated indices ($i,j \in \mathbb {N};\ i,j=1,2,3$) we can write

(3.6)\begin{equation} b_{ij}=\frac{\langle u'_i u'_j \rangle}{\langle u'_k u'_k \rangle}- \frac{1}{3}\delta_{ij}, \end{equation}

where $\delta _{ij}$ represents the second-order tensor identity. Therefore, $\xi$ and $\eta$ are defined as

(3.7a,b)\begin{equation} 6\eta^2=b_{ij}b_{ji},\quad 6\xi^3=b_{ij}b_{jk}b_{ki}. \end{equation}

The so-called Lumley triangle is attained delimiting in the $\xi$–$\eta$ plane the set of realisable states of the Reynolds stress tensor, hence, those associated to real and positive eigenvalues.

To set a starting reference, we first report in all panels of figure 10 the states attained at each wall-normal grid point of a turbulent open channel flow simulation carried out in the same set-up detailed in § 2 at $Re_b=5000$, without the filaments. The first grid point close to the wall is marked in blue, while the one at the centreline is coloured in grey. Turbulence at the wall is two component (2C) due to the no-penetration condition and the anisotropy peaks moving upward, at about $y u_\tau ^{in}/\nu \approx 7$. An axisymmetric state with $\xi >0$ is achieved towards the log layer (the Reynolds stress ellipsoid therefore resembles a prolate spheroid); such a condition is nevertheless lost approaching the centreline, in contrast to what is observed for a full channel, as the no-penetration condition forces again a 2C state, which this time is approached for $\xi <0$ (the Reynolds stress ellipsoid therefore resembles an oblate spheroid). We impute the difference in the sign of $\xi$ attained at the wall and at the free-slip surface to the effect of the no-slip condition. Next, we plot the trends of $\xi$ and $\eta$ for the different canopy cases: (a) refers to the flow within the canopy and the evolution of the variables is therefore truncated at the canopy tip, highlighting the relevant points with distinctive symbols; (b), instead, refers to the outer flow and reports the evolution of the variables from the canopy tip upward. Since any information concerning the wall-normal coordinate is lost in the triangles, we complement them with two additional panels each, showing the trends of $\xi$ and $\eta$ with $y$ scaled in inner (a) and outer (b) viscous units, as introduced in § 3.1. As a matter of consistency, for the open channel case, we set $u_\tau ^{in} = u_\tau ^{out} = \sqrt {\tau _w/\rho _f}$ and $y_{vo}=0$.

Figure 10. Characterisation of the turbulence state (a) inside and (b) outside the canopy for different values of $Ca$, according to the Lumley triangle formalism. Data for the reference open channel case are reported across the whole channel height in both panels and relevant points within the canopy are denoted with distinctive symbols. In the two main plots, each dot corresponds to a grid point along the wall-normal direction, with the first one close to the wall coloured in blue and the last one at the centreline coloured in grey; note how the first grid point of the open channel lays in the right half of the triangle, while those of all the canopy cases lay in the left one. The smaller plots at the bottom report the trends of $\xi$ and $\eta$ (a) inside and (b) outside the canopy, scaled in the viscous units of the inner and outer flow introduced in § 3.1, respectively.

Turbulence close to the bottom of the canopy lays in a 2C axisymmetric state with $\xi <0$ (opposite to the 2C state at $\xi >0$ found in the open channel), arguably, due to the shielding effect of the filaments on the wall and the consequent attenuation of sweep events ($u'>0,v'<0$) reaching the bottom layer. The anisotropy reduces moving upward and reaches a minimum between the inner inflection point and the virtual origin, where turbulence becomes almost isotropic for the most rigid canopy cases. At the same time, starting close to the inner inflection point, $\xi$ begins to grow more rapidly and shifts to positive values, leading to an axisymmetric state with $\xi >0$ characteristic of both the virtual origin and the outer inflection point, where $\eta$ has a local maximum. Moving out of the canopy, above the shear layer at the outer inflection point, the anisotropy is depleted and approaches an almost linear trend that is maintained far into the outer flow for all cases. For the most rigid canopies, $\xi$ settles on a constant value and the stress state is reminiscent of that attained in the log-law region of the open channel. Nevertheless, for the most flexible canopies, $\xi$ exhibits an irregular decrease and soon switches sign again. Finally, approaching the free-slip wall, a 2C axisymmetric state with $\xi <0$ is forced by the boundary condition; $\xi$ therefore undergoes a sudden fall, as $\eta$ sharply increases.

Our analysis supports a multi-layer approach to the modelling of canopy flow turbulence like in Poggi et al. (Reference Poggi, Porporato, Ridolfi, Albertson and Katul2004), based on the collection of different turbulence states along the wall-normal direction. Turbulence in the inner flow exhibits a peculiar structure close to the bottom wall, tending towards a more isotropic condition immediately above and further transitioning smoothly towards a state similar to that attained in the log-law region of turbulent wall flows. This picture is also supported by the visualisations of the instantaneous, local anisotropy of the flow reported in the Appendix. The profiles of $\xi$ and $\eta$ in figure 10 show a reasonable collapse in both the most rigid ($Ca=0,1,10$) and most flexible cases ($Ca=50,100$), once scaled in inner viscous units, with the case at $Ca=25$ laying in between the two. Turbulence immediately above the canopy is more isotropic than over the no-slip wall of the open channel ($\eta$ is smaller), consistent with the observations of Kuwata & Suga (Reference Kuwata and Suga2016) for an isotropic porous medium. Instead, the portion of the outer flow from the log-law region up to where the upper free-slip wall is felt appears analogous to a turbulent wall flow, thus suggesting outer similarity arguments. There, the anisotropy settles on an almost linear trend common to all $Ca$.

3.5. Inner/outer flow interaction

Multiple authors have proposed layered descriptions of canopy flows, identifying regions with specific dynamical properties while moving along the wall-normal direction. Despite the differences between the various approaches proposed (Belcher, Jerram & Hunt Reference Belcher, Jerram and Hunt2003; Poggi et al. Reference Poggi, Porporato, Ridolfi, Albertson and Katul2004; Okamoto & Nezu Reference Okamoto and Nezu2009) there is consensus in separating the flow in the canopy from that outside, and in isolating the thin shear layer region at the canopy tip. In the previous subsection we characterised the inner and the outer flow separately; here, instead, we aim at elucidating how they interact with each other. As extensively discussed in the literature, the flows over plant canopies (Finnigan Reference Finnigan2000) along with those above permeable (Breugem & Boersma Reference Breugem and Boersma2005) and elastic (Rosti & Brandt Reference Rosti and Brandt2017) walls are often characterised by the presence of large-scale vortical structures elongated in the spanwise direction (rollers), generated by a Kelvin–Helmholtz-like instability. We provide evidence of the presence of these structures in our set-up in the Appendix. These structures are likely to dominate the momentum exchange between the inner and the outer flow, but they are not the only process active close to the shear layer: also high/low-speed streaks (Monti et al. Reference Monti, Olivieri and Rosti2023) and quasi-streamwise vortices might affect the interaction. The flow dynamics close to the canopy tip is in fact a highly non-trivial phenomenon, which can not be exhaustively described only accounting for the presence of the rollers. Furthermore, as noted by Nicholas et al. (Reference Nicholas, Monti, Omidyeganeh and Pinelli2023), drag model-based simulations provide reasonable results in capturing the flow behaviour near the canopy, but they fall short in providing a detailed characterisation of the flow within the canopy layer and at the interface with the outer flow. A better understanding is therefore crucial for the development of more accurate models.

Thus, in figure 11 we observe the sweep ($u'>0,v'<0$) and ejection ($u'<0,v'>0$) events taking place at three locations distinctive of the three regions mentioned above: at $y=H/2$, at the canopy tip, corresponding to the outer inflection point $y_{out}$ and at the virtual origin $y_{vo}$. (The continuous evolution of sweeps and ejections along the wall-normal direction can be found in Poggi et al. (Reference Poggi, Porporato, Ridolfi, Albertson and Katul2004), for a rigid canopy.) We consider the rigid canopy case (a–c) as well as a flexible one ($Ca=100$, (d–f)). The most intense sweeps and ejections occur at the canopy tip (b,e), while they attenuate and become more coherent moving in the outer flow (a,d). Consistent with the observations of Gao et al. (Reference Gao, Shaw and Paw U1989), sweep and ejections are of about equal strength at approximately twice the canopy height. At the virtual origin (c,f), instead, only the most intense sweep events are able to penetrate and they occupy small, well-defined regions; ejections there are less frequent. Overall, the turbulent state is more strongly driven in the rigid canopy case, where it is characterised by more coherent events compared with the flexible one. Sweep events reach the virtual origin of the flexible canopy flow more frequently, but with diminished intensity.

Figure 11. Instantaneous sweep and ejection events in a rigid (a–c) and a flexible (d–f) canopy flow with $Ca=100$, at $y=H/2$ (a,d), at the canopy tip (b,e) and at the virtual origin (c,f). The flow is sampled on wall-parallel planes with the mean velocity aligned to the vertical direction, going from bottom to top. Regions where the events are occurring are delimited with black lines, while their magnitude is quantified as $|uw|/U_b^2$ and visualised with a linear colourmap ranging from white to orange (ejections) or violet (sweeps) in [0,0.4].

In order to provide a more quantitative description of the momentum transfer between the inner and outer flows, we compute the joint probability density function (J-PDF) of the streamwise and wall-normal velocity fluctuations at the canopy tip for different values of $Ca$. As visible in figure 12, the peaks always lay in the second quadrant ($u'<0,v'>0$) denoting the dominance of ejections over sweeps, as well documented in the literature (Gao et al. Reference Gao, Shaw and Paw U1989; Finnigan Reference Finnigan2000). The downward motion of high-speed fluid is mitigated by the presence of the canopy, which instead opposes little resistance to the uplift of low-speed fluid from its interior. Such observation is consistent with the experiment of Chowdhuri et al. (Reference Chowdhuri, Ghannam and Banerjee2022), who noted that long-lasting ejection events are more relevant than short-lived sweep events close to the canopy tip. Furthermore, the shape of the J-PDFs closely resembles that reported by Manes et al. (Reference Manes, Poggi and Ridolfi2011) in the case of a highly permeable porous medium. Notwithstanding this similarity, our J-PDFs differ from one another due to the effect of canopy flexibility. In the most rigid cases the J-PDFs elongate in the fourth quadrant ($u'>0,v'<0$), remarking the occurrence of rare and intense sweep events that are attenuated increasing the flexibility. In the most flexible cases, in fact, the filaments are deflected by the mean flow and shield the inner region of the canopy from downward interactions. Furthermore, at $Ca=0$ and $Ca=1$, the J-PDFs exhibit a secondary peak in the third quadrant ($u'<0,v'<0$) associated with the deflection of the streamlines around the filament tip. As seen in Monti et al. (Reference Monti, Omidyeganeh, Eckhardt and Pinelli2020) and in figure 16 of Nicholas et al. (Reference Nicholas, Monti, Omidyeganeh and Pinelli2023), immediately after the tip the flow decelerates and plunges.

Figure 12. Isolines of the J-PDF (normalised to a unitary integral over the domain) associated to the streamwise and wall-normal velocity fluctuations at the mean position of the canopy tip, for different values of $Ca$. Levels are evenly distributed between $0.4$ and $6$ with $0.4$ increments, while the locations of the peaks are denoted by black dots.

Our analysis appears to support a scenario where turbulence inside the canopy is sustained by intense events induced by the outer flow. Nevertheless, while the sweeps are hindered by the presence of the filaments, the canopy opposes little resistance to ejections. The filaments, in fact, regulate the vertical exchange of streamwise momentum allowing for the transpiration of the canopy and shielding the inner flow from all but the most intense downward interactions (likely associated to the secondary instability of the rollers, Nepf Reference Nepf2012a). Such intense sweeps induce a significant local deflection of the filaments and a large-amplitude flapping motion. Turbulence inside the canopy is therefore driven by frequent ejections and rare intense sweep events (Nepf Reference Nepf2012a), along with the flapping motion of the filaments in the flexible cases.

3.6. Filament density variation

In order to investigate the effects of a variation in the inertial properties of the filaments, we consider two additional values of the density ratio $\{1.0+1.46\times 10^{-2}, 1.0+1.46\times 10^{-1}\}$ starting from the case at $Ca=25$ and $\rho _s/\rho _f=1.0+1.46\times 10^{-3}$.

The first macroscopic effect of an increase in the density of the filaments is their reduced compliance to the flow and the consequent upshift of the canopy tip. A taller canopy is associated to a higher blockage effect: the streamwise velocity profiles close to the bottom wall are therefore depleted (figure 13a), while higher velocities are attained far above the canopy tip due to the imposition of a constant flow rate. The positions of the inner inflection point and of the virtual origin remain essentially unchanged (inset of figure 13a), while the upward shift of the canopy tip is associated to an increase in the mean streamwise velocity there. Moving to the shear stress balance reported in figure 13(b), we observe how the peak in the viscous shear stress becomes more definite in the case of denser filaments, denoting the presence of a sharp shear layer above their tip. Nevertheless, their limited flapping motion yields less intense turbulent fluctuations there. Equivalently, the shear layer is blurred for lower density ratios and the filaments undergo a more pronounced motion, responsible for the higher Reynolds shear stress. As the most significant effects of a variation in the filament density are appreciated at the canopy tip, we are motivated to better investigate the events taking place there observing the J-PDFs of the streamwise and wall-normal velocity fluctuations reported in figure 14. While a clear peak is always observed in correspondence of ejection events, sweep events penetrating the canopy become more frequent for higher density ratios due to the diminished shielding of the inner flow exerted by the filaments. The J-PDFs therefore elongate in the fourth quadrant, as the ejection peak becomes less pronounced. A deviation towards negative values of $u'$ close to $v'=0$ can also be appreciated, due to the augmented inertia of the filaments and the consequently increased intensity of impact interactions with high-speed fluid.

Figure 13. (a) Mean profiles of the streamwise velocity for different values of $\rho _s/\rho _f$ at $Ca=25$ and associated relevant points (inset). In (b) the total shear stress (black line) for the same cases, normalised by the wall shear stress, is given by the sum of the turbulent shear stress (continuous lines), the viscous shear stress (dash-dotted lines) and the canopy drag (dashed lines).

Figure 14. Isolines of the $u'/v'$ J-PDF at the canopy tip for different values of $\rho _s/\rho _f$ at $Ca=25$. Levels are distributed between $0.4$ and $6$ with $0.4$ increments, while peaks are denoted by black dots.

For the dynamics of the fluid considered so far, an increase in the density of the filaments yields effects similar to those of an increase in their rigidity from one of the most flexible cases. Such effects can be appreciated comparing the case with lowest density ratio (belonging to the bulk of our investigation) to the others. Nevertheless, the differences between the two cases at higher $\rho _s/\rho _f$ are marginal and denote the attainment of a saturation for $\rho _s/\rho _f-1={O}(10^{-1})$, from which most of our investigation remains far from.

3.7. The effect of the filaments’ flexibility

In a flexible canopy the motion of the filaments is tightly coupled to the large-scale coherent fluctuations of the turbulent flow (Monti et al. Reference Monti, Olivieri and Rosti2023); in turn, the flow is affected by the flapping motion of the filaments along with their wakes at significantly smaller scales, due to the spectral short cut mechanism highlighted in § 3.3. To assess the consequences of this complex fluid–structure interaction, we inhibit it, by ‘freezing’ the flexible canopy in one of its instantaneous configurations, and we compare the flow attained above such a peculiar rigid canopy to the flow developed above its flexible counterpart. We repeat the comparison for all the non-zero values of $Ca$ in this study. The independence of our observations from the specific canopy configuration considered is assessed comparing the pressure gradient needed to impose the same flow rate above two different frozen canopies originated from the same simulation at $Ca=100$: no significant variation is observed between the two, thus confirming the adequacy of our numerical test. In particular, the canopy proves large enough to accommodate multiple configurations of the filaments in spite of their large-scale coherent motion and, therefore, retains almost constant space-averaged characteristics across time.

The first macroscopic effect of freezing the flexible canopy in its instantaneous configuration, despite the increased average velocity difference between the filaments and the fluid, is a reduction of its total drag (which nevertheless remains higher then in an open channel at $Re_b=5000$ without the canopy). As visible in figure 15, the driving pressure gradient $\mathrm {d}P/\mathrm {d}\kern 0.06em x$ is reduced with respect to all the flexible canopy cases, reported for reference as thinner vertical bars highlighted with a red hatched fill. The viscous shear contribution remains small and constant, while the canopy drag is slightly depleted. Nevertheless, the most significant reduction is undergone by the turbulent shear, which is therefore almost completely responsible for the drag reduction effect. Indeed, the depletion of the driving pressure gradient is accompanied by a reduction of all the components in the Reynolds stress tensor across the whole half-channel height. Such an effect is shown for the shearing contribution only in figure 16, where data from the frozen canopy cases are denoted with continuous lines and data from the flexible cases are shown with dashed lines. No significant variation can be appreciated in the location of the maxima of $\langle u' v' \rangle$ between the flexible and frozen canopy cases, suggesting that the position of the shear layer above the canopy remains essentially unchanged. To clarify how the nature of the turbulent fluctuations relates to the (now prevented) motion of the filaments, we once again observe the J-PDFs of the streamwise and wall-normal velocity fluctuations at the canopy tip. In figure 17 the same levels are shown for both the frozen and flexible canopy cases, respectively denoted with red dashed lines and black continuous lines, for different values of $Ca$. The deviation from the flexible cases is understandably enhanced increasing the initial $Ca$, as we compare canopies of increasing flexibility to rigid ones. For the frozen cases, intense events of the wall-normal velocity become less relevant in both the positive and negative quadrants, leading to sharper peaks in the second quadrant ($u'>0, v'<0$). Intense sweep and ejection events are therefore diminished: the frozen canopy behaves as a less permeable porous medium compared with the flexible one, thus justifying its decreased drag (Manes et al. Reference Manes, Pokrajac, McEwan and Nikora2009).

Figure 15. Different contributions to the shear stress balance, integrated across the wall-normal direction, for the frozen canopy cases at different initial values of $Ca$. Results for the corresponding flexible cases are reported, for reference, as thinner bars with a red hatched fill.

Figure 16. Shearing component of the Reynolds stress tensor for different values of $Ca$. Data from the frozen canopy cases are denoted with continuous lines and data from the corresponding flexible cases are shown with dashed lines.

Figure 17. Isolines of the $u'/v'$ J-PDF at the tip of the frozen canopies for different initial values of $Ca$. Levels are distributed between $0.4$ and $6$ with $0.8$ increments, while peaks are denoted by black dots. The same isolines from the corresponding flexible cases are reported, for reference, as dashed red lines.

The drag reduction observed freezing a flexible canopy in its instantaneous configuration can therefore be imputed to an overall reduction of the turbulent activity with respect to the flexible case. We identify two phenomena contributing to such an effect: first, upon freezing the filaments, no more energy is pumped into the small-scale turbulent fluctuations of the flow by their flapping motion. This is consistent with the maximum drag reduction being attained in the intermediate regime ($Ca=25\sim 50$), where the large-amplitude flapping of the filaments is most regular. As further discussed in the following section, in fact, at intermediate values of $Ca$ the natural response of the filaments resonates with a characteristic frequency of the turbulent forcing. Second, the frozen canopies are more effective then the flexible ones at preventing intense vertical interactions between the inner and outer flows, thus yielding a weakly driven turbulent state within the canopy and depleting the turbulent fluctuations close to the shear layer.

4.1. Characterisation of motion

Multiple quantities can be observed in order to characterise the motion of the flexible filaments constituting the canopy. Here, we focus on the spanwise velocity of the Lagrangian points at the tip of the filaments: we prefer the measurement of a velocity to that of a displacement since it is more directly related to the turbulent fluctuations of the flow, and we favour the spanwise direction as it is the one less affected by the constraints of the problem. The streamwise and vertical dynamics of the filaments are the most influenced by the constraints, i.e. the inextensibility and the presence of the wall. We therefore collect the Lagrangian velocity signal in the spanwise direction at the tip of all the filaments throughout all the values of $Ca$ considered in this study, and highlight its features both in the time and frequency domains. For the purpose of visualisation, we also choose one canopy filament per value of $Ca$ and show in figure 18(a) the velocity signal over ten bulk time units extracted after the attainment of a fully developed flow state. Here we also discuss the results of the simulation at $Ca=500$ to encompass the case in which the filaments are fully compliant to the flow.

Figure 18. (a) Lagrangian velocity of the filament tips in the spanwise direction. We report the signal from one selected filament per value of $Ca$ over ten bulk time units and (b) the energy spectrum of the signal, ensemble averaged over all the filaments at a given $Ca$. The colour scale is the same between the two panels.

We immediately observe that the amplitude of the fluctuations increases with the flexibility of the filaments up to $Ca=25\sim 50$, where it appears to saturate. Furthermore, the small-amplitude fluctuations of the most rigid case ($Ca=1$) are characterised by a nearly sinusoidal shape, modulated in amplitude by slower dynamics. Such a behaviour contrasts with the motion of the most flexible filaments ($Ca=100\sim 500$), which instead exhibit fluctuations with a less definite shape and period. In these cases it is also possible to appreciate abrupt and fast small-scale fluctuations on top of the slower dynamics: those arise from the collisions of the filament with the wall, and indeed are not observed for the lowest values of $Ca$ as the filament never gets sufficiently deflected. In between these two behaviours ($Ca=50$), the large-amplitude fluctuations are almost sinusoidal in shape and exhibit a more definite period; they therefore appear compatible with a resonance between the natural structural response of the filaments (responsible for the nearly sinusoidal oscillations of the most rigid cases) and the turbulent fluctuations of the flow (dominating the motion of the filaments in the most flexible cases).

To confirm the picture above, we investigate the energy spectrum of the signal, $E_z^{tip}=\frac {1}{2}\overline {\overline {\mathcal {F}(w_{tip})\mathcal {F}^*(w_{tip})}}$, where $\mathcal {F}$ is the temporal Fourier's transform operator, $*$ is the conjugate of an arbitrary complex number and the double overbar denotes ensemble averaging on all the filaments of each case. After computing $E_z^{tip}$ over a sufficiently long-time span cleaned of any initial transient, we report it in figure 18(b) for the different values of $Ca$ considered. We also mark with vertical dashed lines the natural frequencies associated to the first structural mode of the different filaments, $f_{nat}$, computed as described in § 2. For the lowest values of $Ca$ ($1$ to $50$), the spectrum peaks in correspondence of the natural frequency (and higher harmonics), thus confirming that the structural natural response dominates the dynamics of the most rigid filaments. The value of $f_{nat}$ decreases with $\gamma$, hence, it gets shifted towards the left for increasing values of $Ca$. The peak of the spectrum also obeys this shift in frequency up to $Ca=50$, while beyond that its position does not to change significantly with the $Ca$. Such behaviour suggests that the most flexible filaments adjust to a motion no more dictated by their structural properties, but presumably compliant to the turbulent fluctuations of the flow. The transition between the two distinct regimes of motion occurs at $Ca\approx 50$, and thus, we justify the large amplitude nearly sinusoidal oscillations observed in that case with the resonance between the first structural mode of the filaments and a characteristic frequency of the turbulent forcing to which they are exposed.

The identification of two distinct regimes of motion is not unexpected: previous investigations addressing the motion of flexible fibres in homogeneous isotropic turbulence (Rosti et al. Reference Rosti, Banaei, Brandt and Mazzino2018a,  Reference Rosti, Olivieri, Banaei, Brandt and Mazzino2020; Olivieri, Mazzino & Rosti Reference Olivieri, Mazzino and Rosti2022b) highlighted the existence of a structure-dominated and a turbulence-dominated regime, partially similar to what was found when studying vortex induced vibrations (Bearman Reference Bearman1984). We recently characterised numerically the motion of a flexible fibre clamped in wall turbulence over a wide range of structural parameters (Foggi Rota et al. Reference Foggi Rota, Koseki, Agrawal, Olivieri and Rosti2024), also observing the emergence of the two regimes discussed above. Recently, Fu et al. (Reference Fu, He, Huang, Dey and Fang2023) experimentally investigated the individual motion of an isolated flexible plant model and how that changes when multiple plants are arranged in a canopy, while Monti et al. (Reference Monti, Olivieri and Rosti2023) studied the motion of the filaments in a flexible canopy, mostly to characterise their collective dynamics (honami/monami). To reconcile the different insights offered by all these works, in the following section we further describe the flapping state of the individual filaments and better contextualise our observations.

4.2. The flapping state

To better investigate the different dynamical regimes of the filaments, we compute their dominant flapping frequency along the spanwise direction, $f^{flap}_z$, i.e. the peak location of the spectra reported in figure 18(b), and plot it against the natural frequency associated to the corresponding value of $Ca$, i.e. with reference to § 2, $f_{nat}\approx ({3.516}/{d h^2})\sqrt {{\rho _f d h^3 U_b^2}/{2 Ca \rho _s {\rm \pi}^3}}\approx 0.45 ({U_b}/{h})\sqrt {({\rho _f}/{\rho _s})({h}/{d})({1}/{Ca})}$. Since for the highest values of $Ca$ the position of the spectral peak does not appear to undergo any significant change with the structural parameters of the filaments, we denote it as $f_{turb}$ (i.e. the characteristic flapping frequency of the filaments in the turbulence-dominated regime) and employ it to adimensionalise both $f^{flap}_z$ and $f_{turb}$. The outcome is reported in figure 19(a), where flapping frequencies close to the natural one lay on the dashed-dotted line with unitary slope, $f^{flap}_z/f_{turb}=f_{nat}/f_{turb}$, while those close to the chosen value of $f_{turb}$ approach the dashed horizontal asymptote. We thus confirm that the most rigid filaments exhibit their natural response, while the most flexible ones deviate from it and approach $f^{flap}_z\approx f_{turb} = 0.5 U_b/H$. Such frequency, independent from the structural characteristics of the filaments, is consistent with the outcome of our previous investigation (Foggi Rota et al. Reference Foggi Rota, Koseki, Agrawal, Olivieri and Rosti2024) focused on the flapping states of an isolated flexible fibre clamped in wall turbulence. Also in that case indeed, we identified a turbulence-dominated regime for the highest values of $Ca$ where the fibre is compliant to the flow and its motion is characterised by a slow flapping with a period comparable to the bulk time scale of the flow. Furthermore, despite the absence of a physical justification for $f^{flap}_z\approx 0.5 U_b/H$, we observe that its occurrence in both the motion of an isolated flexible fibre in a full-channel flow and in the motion of the canopy filaments in the half-channel considered here supports the reproducibility and generality of the result.

Figure 19. Flapping states of the filaments for different values of $Ca$, hence $f_{nat}/f_{turb}$. We first show (a) the dominant frequency of oscillation along the spanwise direction, $f^{flap}_z$, extracted as the peak location of the spectra in figure 18. Squares denote data from the cases at different density ratios. (b) We also report the trends of the elastic ($\varGamma _s$) and kinetic ($K_s$) energy of the filaments per unit length, averaged over time and over all the filaments. The error bars are computed carrying out the same measurement on the time signals truncated to their first half, and fall within the markers size for most points.

In order to assess the effect of an increase in the structural density of the filaments on their dynamics, we also report in the map of figure 19(a), the results of the cases at ${\rho _s}/{\rho _f}=\{1+1.46\times 10^{-2},1+1.46\times 10^{-1}\}$, respectively associated to a higher and a lower value of the natural frequency. Differently from what was found in our previous investigation of a single filament (Foggi Rota et al. Reference Foggi Rota, Koseki, Agrawal, Olivieri and Rosti2024), here an increase in ${\rho _s}/{\rho _f}$ prevents the filaments from flapping at $f_{turb}$ and they thus exhibit their natural response. This difference originates from the fact that denser filaments yield a stronger obstruction of the mean flow in the case of a canopy (as elucidated in § 3.6), and consequently get less deflected and maintain an almost upright configuration, swaying at $f_{nat}$. This effect is lost in the case of an isolated fibre, as the back reaction on the mean flow is negligible; in this isolated case the fibre is less shielded and consequently gets more deflected by the flow, swaying with it.

Interestingly, also the experimental investigation of Fu et al. (Reference Fu, He, Huang, Dey and Fang2023) addresses the dynamical response of a synthetic plant submerged in a turbulent flow, comparing the isolated case to the one in which the plant is part of a vegetation patch (i.e. a canopy). There, a simplified plant model constituted by a series of five wooden buoyant pellets connected by a fine rope is exposed to a turbulent channel flow at different Reynolds numbers, thus focusing on a different scenario from that explored here. The synthetic plant used in the experiments, in fact, appears hinged to the bottom wall rather than clamped, and it does not have homogeneous structural properties like our filaments. Consequently, while the motion of our filaments is dominated by a single dynamics even (and particularly) in the highest range of $f^{flap}_z$, the plant of Fu et al. (Reference Fu, He, Huang, Dey and Fang2023) alternates phases of swaying at a lower frequency synchronous mode to phases at a higher frequency asynchronous one, both of them laying far above the range of $f^{flap}_z$ we considered here.

The effect of the clamp on the dynamics of the filaments is further understood observing the variation of their elastic energy with $Ca$, while their motion is better described in terms of kinetic energy. To those ends, we therefore consider the weak form of (2.3) and (2.4), integrated over a time span $\tilde {t}$ and over the length of the filaments,

(4.1)\begin{equation} \int_{\tilde{t}} \int_h \left[ {\rm \Delta} \tilde{\rho} \frac{\partial \boldsymbol{X}}{\partial t} \boldsymbol{\cdot} \frac{\partial \boldsymbol{X}}{\partial t} + \gamma \frac{\partial^2 \boldsymbol{X}}{\partial s^2} \boldsymbol{\cdot} \frac{\partial^2 \boldsymbol{X}}{\partial s^2} -T + \boldsymbol{F} \boldsymbol{\cdot} \boldsymbol{X} \right] \mathrm{d}s \,\mathrm{d}t = 0, \end{equation}

which can be interpreted as a balance equation for the structural energy under the assumption of a beam configuration $\boldsymbol {X}$ compatible with the constraints. In particular,

(4.2)\begin{equation} \varGamma_s=\frac{1}{n_x n_z}\frac{1}{\tilde{t}}\frac{1}{h}\sum_{n_x n_z} \int_{\tilde{t}} \int_h \gamma \frac{\partial^2 \boldsymbol{X}}{\partial s^2} \boldsymbol{\cdot} \frac{\partial^2 \boldsymbol{X}}{\partial s^2} \mathrm{d}s \,\mathrm{d}t \end{equation}

represents the structural elastic energy per unit length, averaged over time and over all the filaments, while

(4.3)\begin{equation} K_s=\frac{1}{n_x n_z}\frac{1}{\tilde{t}}\frac{1}{h}\sum_{n_x n_z} \int_{\tilde{t}} \int_h {\rm \Delta} \tilde{\rho} \frac{\partial \boldsymbol{X}}{\partial t} \boldsymbol{\cdot} \frac{\partial \boldsymbol{X}}{\partial t} \mathrm{d}s \,\mathrm{d}t \end{equation}

represents the structural kinetic energy per unit length, likewise averaged. We report the trends of $\varGamma _s$ and $K_s$ with $f_{nat}/f_{turb}$, and hence, with the $Ca$, in figure 19(b). While the magnitude of $\varGamma _s$ is comparable to that measured by Rosti et al. (Reference Rosti, Banaei, Brandt and Mazzino2018a), the position of its peak contrasts with their findings. Here, in fact, $\varGamma _s$ peaks well within the regime dominated by the natural response of the filaments, confirming what can also be observed in the case of an isolated clamped fibre. We therefore deduce that the clamp to the wall, constraining the shape of the filaments, hinders their deflection and shifts the maximum of $\varGamma _s$ towards higher values of $\gamma$. The kinetic energy, instead, follows a trend consistent with the amplitude of the fluctuations in figure 18(a), increasing with the $Ca$ in the most rigid cases and reaching a plateau close to the resonance between $f_{nat}$ and $f_{turb}$.

Characterising the dominant dynamics of our canopy filaments, we have been able to compare our results with those attained considering a single fibre clamped in wall turbulence (Foggi Rota et al. Reference Foggi Rota, Koseki, Agrawal, Olivieri and Rosti2024) and plants with a different geometry (Fu et al. Reference Fu, He, Huang, Dey and Fang2023). We have also isolated the role played by the sheltering effect of the canopy upon increasing the density of the filaments, and by the model adopted to describe them. Nevertheless, we emphasise that the present investigation is unable to capture any sub-dominant dynamics of the filaments, such as the contribution of turbulence to their motion in the regime dominated by the natural response. Regardless of the value of $Ca$, in fact, the coherent motion of the filaments (honami/monami) investigated by Monti et al. (Reference Monti, Olivieri and Rosti2023) remains dictated by the turbulent structures in the flow.