Computational Fluid Dynamical Modelling of Concentration Fluctuations in an Idealized Urban Area

Turbulent mixing induces variability in concentration that is important in many applications, such as reactive plumes, risk assessments or odour impact analyses (when the effects can have time scales on the order of a second). In urban canopies, the variability may be modified by the presence of buildings. Our purpose is to study concentration fluctuation variance in built-up areas using an Eulerian approach. We performed numerical simulations with the computational fluid dynamics model Mercure_Saturne, which is a three-dimensional model adapted to atmospheric flow and pollutant dispersion. We use a k − ϵ turbulence closure and predict the concentration variance with a transport equation model. The model performance is evaluated with the near-full scale experiment MUST (Mock Urban Setting Test), a field experiment conducted in Utah’s West Desert Test Center. The modelled root-mean-square of the concentration fluctuations is compared to measurements for 20 of the MUST trials. The model shows good agreement with the measurements, with the fraction of predictions within a factor of two of observations of 60.1%, with better results for horizontal lines of detectors than for the detectors on vertical masts (with fractions of predictions within a factor of two of observations of respectively 66.4% and 52.6%). The influence of different parameters on the fluctuation variance is also studied and we show the importance of taking into account the stability of the stratification when modelling the turbulent kinetic energy.     

Air-Parcel Residence Times Within Forest Canopies

We present a theoretical model, based on a simple model of turbulent diffusion and first-order chemical kinetics, to determine air-parcel residence times and the out-of-canopy export of reactive gases emitted within forest canopies under neutral conditions. Theoretical predictions of the air-parcel residence time are compared to values derived from large-eddy simulation for a range of canopy architectures and turbulence levels under neutral stratification. Median air-parcel residence times range from a few sec in the upper canopy to approximately 30 min near the ground and the distribution of residence times is skewed towards longer times in the lower canopy. While the predicted probability density functions from the theoretical model and large-eddy simulation are in good agreement with each other, the theoretical model requires only information on canopy height and eddy diffusivities inside the canopy. The eddy-diffusivity model developed additionally requires the friction velocity at canopy top and a parametrized profile of the standard deviation of vertical velocity. The theoretical model of air-parcel residence times is extended to include first-order chemical reactions over a range of of Damköhler numbers (Da) characteristic of plant-emitted hydrocarbons. The resulting out-of-canopy export fractions range from near 1 for \(Da =10^{-3}\) to less than 0.3 at \(Da = 10\). These results highlight the necessity for dense and tall forests to include the impacts of air-parcel residence times when calculating the out-of-canopy export fraction for reactive trace gases.     

Large-eddy Simulation of Turbulent Flow across a Forest Edge. Part II: Momentum and Turbulent Kinetic Energy Budgets

Momentum and turbulent kinetic energy (TKE) budgets across a forest edge have been investigated using large-eddy simulation (LES). Edge effects are observed in the rapid variation of a number of budget terms across this vegetation transition. The enhanced drag force at the forest edge is largely balanced by the pressure gradient force and by streamwise advection of upstream momentum, while vertical turbulent diffusion is relatively insignificant. For variance and TKE budgets, the most important processes at the forest edge are production due to the convergence (or divergence) of the mean flow, streamwise advection, pressure diffusion and enhanced dissipation by canopy drag. Turbulent diffusion, pressure redistribution and vertical shear production, which are characteristic processes in homogeneous canopy flow, are less important at the forest transition. We demonstrate that, in the equilibrated canopy flow, a substantial amount of TKE produced in the streamwise direction by the vertical shear of the mean flow is redistributed in the vertical direction by pressure fluctuations. This redistribution process occurs in the upper canopy layers. Part of the TKE in the vertical velocity component is transferred by turbulent and pressure diffusion to the lower canopy levels, where pressure redistribution takes place again and feeds TKE back to the streamwise direction. In this TKE cycle, the primary source terms are vertical shear production for streamwise velocity variance and pressure redistribution for vertical velocity variance. The evolution of these primary source terms downwind of the forest edge largely controls the adjustment rates of velocity variances.     

Turbulent Pressure and Velocity Perturbations Induced by Gentle Hills Covered with Sparse and Dense Canopies

How the spatial perturbations of the first and second moments of the velocity and pressure fields differ for flow over a train of gentle hills covered by either sparse or dense vegetation is explored using large-eddy simulation (LES). Two simulations are investigated where the canopy is composed of uniformly arrayed rods each with a height that is comparable to the hill height. In the first simulation, the rod density is chosen so that much of the momentum is absorbed within the canopy volume yet the canopy is not dense enough to induce separation on the lee side of the hill. In the second simulation, the rod density is large enough to induce recirculation inside the canopy on the lee side of the hill. For this separating flow case, zones of intense shear stress originating near the canopy-atmosphere interface persist all the way up to the middle layer, ‘contaminating’ much of the middle and outer layers with shear stress gradients. The implications of these persistent shear-stress gradients on rapid distortion theory and phase relationships between higher order velocity statistics and hill-induced mean velocity perturbations (Δu) are discussed. Within the inner layer, these intense shear zones improve predictions of the spatial perturbation by K-theory, especially for the phase relationships between the shear stress (~ ?Δu/?z) and the velocity variances, where z is the height. For the upper canopy layers, wake production increases with increasing leaf area density resulting in a vertical velocity variance more in phase with Δu than with ?Δu/?z. However, background turbulence and inactive eddies may have dampened this effect for the longitudinal velocity variance. The increase in leaf area density does not significantly affect the phase relationship between mean surface pressure and topography for the two simulations, though the LES results here confirm earlier findings that the minimum mean pressure shifts downstream from the hill crest. The increase in leaf area density and associated flow separation simply stretches this difference further downstream. This shift increases the pressure drag, the dominant term in the overall drag on the hill surface, by some 15%. With regards to the normalized pressure variance, increasing leaf area density increases ${\sigma_p/u_{*}^{2}}$ near the canopy top, where u _* is the longitudinally averaged friction velocity at the canopy top and σ _p is the standard deviation of the pressure fluctuations. This increase is shown to be consistent with a primitive scaling argument on the leading term describing the mean-flow turbulent interaction. This scaling argument also predicts the spatial variations in σ _p above the canopy reasonably well for both simulations, but not inside the canopy.     

Similarity Scaling Of Surface-Released Smoke Plumes

Concentration fluctuation data from surface-layer released smokeplumes have been investigated with the purpose of finding suitable scaling parametersfor the corresponding two-particle, relative diffusion process.Dispersion properties have been measured at downwind ranges between 0.1 and 1 km from a continuous, neutrally buoyant ground level source. A combinationof SF₆ and chemical smoke (aerosols) was used as tracer. Instantaneous crosswind concentration profiles of high temporal (up to 55 Hz) and spatialresolution (down to 0.375 m) were obtained from aerosol-backscatter Lidar detectionin combination with simultaneous gas chromatograph (SF₆) reference measurements. The database includes detailed crosswind concentration fluctuation measurements. Each experiment, typically of 1/2-hour duration, containsplume mean and variance concentration profiles, intermittency profiles andexceedence and duration statistics. The diffusion experiments were accompanied by detailed in-situ micrometeorological mean and turbulence measurements. In this paper, a new distance-neighbour function for surface-released smoke plumes is proposed, accompanied by experimental evidence in its support. The new distance-neighbour function is found to scale with the surface-layer friction velocity,and not with the inertial subrange dissipation rate, over the range of distance-neighbour separations considered.     

Experiments on scalar dispersion within a model plant canopy part I: The turbulence structure

This is the first of a series of three papers describing experiments on the dispersion of trace heat from elevated line and plane sources within a model plant canopy in a wind tunnel. Here we consider the wind field and turbulence structure. The model canopy consisted of bluff elements 60 mm high and 10 mm wide in a diamond array with frontal area index 0.23; streamwise and vertical velocity components were measured with a special three-hot-wire anemometer designed for optimum performance in flows of high turbulence intensity. We found that:

1. The momentum flux due to spatial correlations between time-averaged streamwise and vertical velocity components (the dispersive flux) was negligible, at heights near and above the top of the canopy.
2. In the turbulent energy budget, turbulent transport was a major loss (of about one-third of local production) near the top of the canopy, and was the principal gain mechanism lower down. Wake production was greater than shear production throughout the canopy. Pressure transport just above the canopy, inferred by difference, appeared to be a gain in approximate balance with the turbulent transport loss.
3. In the shear stress budget, wake production was negligible. The role of turbulent transport was equivalent to that in the turbulent energy budget, though smaller.
4. Velocity spectra above and within the canopy showed the dominance of large eddies occupying much of the boundary layer and moving downstream with a height-independent convection velocity. Within the canopy, much of the vertical but relatively little of the streamwise variance occurred at frequencies characteristic of wake turbulence.
5. Quadrant analysis of the shear stress showed only a slight excess of sweeps over ejections near the top of the canopy, in contrast with previous studies. This is a result of improved measurement techniques; it suggests some reappraisal of inferences previously drawn from quadrant analysis.

     

Dispersion in sheared Gaussian homogeneous turbulence

By integrating the Fokker-Planck equation corresponding to a Lagrangian stochastic trajectory model, which is consitent with the selection criterion of Thomson (1987), an analytical solution is given for the joint probability density functionp(x_i, u_i, t) for the position (x _i) and velocity (u _i) at timet of a neutral particle released into linearly-sheared, homogeneous turbulence. The solution is compared with dispersion experiments conforming to the restrictions of the model and with a shortrange experiment performed in highly inhomogeneous turbulence within and above a model crop canopy. When the turbulence intensity, wind shear and covariance are strong, the present solution is better than simpler solutions (Taylor, 1921; Durbin, 1983) and as good as any numerical Lagrangian stochastic model yet reported.     

Turbulent transport of scalar quantities within and above a paddy field

An experiment was conducted to study turbulent transport processes of scalar quantities within and above a rice plant canopy. A sonic anemometer-thermometer and a Lyman- humidiometer were used to measure the turbulent fluxes of sensible and latent heat and related turbulence statistics within a paddy field. The sensible and latent heat fluxes measured at two heights within and above the plant canopy showed that the upper layer of this plant canopy was an active source region and that the source strength of sensible and latent heat depended on the solar radiation and physiology of rice plants. Analysis of joint probability distributions of w and T and of w and q within this plant canopy showed that downdrafts were remarkably efficient for upward transport of sensible and latent heat in the daytime. The vertical fluxes of temperature and humidity variance were also divergent from the upper layer of plant canopies. The power spectra of temperature and humidity within the plant canopy decreased rapidly in the high frequency range, compared with the - 2/3 law relationship of nS(n) vs log n observed above plant canopies.     

Effect of Vertical Wind Shear on Concentration Fluctuation Statistics in a Point Source Plume

Measurements of concentration fluctuation intensity, intermittency factor, and integral time scale were made in a water channel for a plume dispersing in a well-developed, rough surface, neutrally stable, boundary layer, and in grid-generated turbulence with no mean velocity shear. The water-channel simulations apply to full-scale atmospheric plumes with very short averaging times, on the order of 1–4 min, because plume meandering was suppressed by the water-channel side walls. High spatial and temporal resolution vertical and crosswind profiles of fluctuations in the plume were obtained using a linescan camera laser-induced dye tracer fluorescence technique. A semi-empirical algebraic mean velocity shear history model was developed to predict these concentration statistics. This shear history concentration fluctuation model requires only a minimal set of parameters to be known: atmospheric stability, surface roughness, vertical velocity profile, and vertical and crosswind plume spreads. The universal shear history parameter used was the mean velocity shear normalized by surface friction velocity, plume travel time, and local mean wind speed. The reference height at which this non-dimensional shear history was calculated was important, because both the source and the receptor positions influence the history of particles passing through the receptor position.     

Dispersion in the Wake of a Rectangular Building: Validation of Two Reynolds-Averaged Navier–Stokes Modelling Approaches

When modelling the turbulent dispersion of a passive tracer using Reynolds-averaged Navier–Stokes (RANS) simulations, two different approaches can be used. The first consists of solving a transport equation for a scalar, where the governing parameters are the mean velocity field and the turbulent diffusion coefficient, given by the ratio of the turbulent viscosity and the turbulent Schmidt number Sc _t . The second approach uses a Lagrangian particle tracking algorithm, where the governing parameters are the mean velocity and the fluctuating velocity field, which is determined from the turbulence kinetic energy and the Lagrangian time T _L . A comparison between the two approaches and wind-tunnel data for the dispersion in the wake of a rectangular building immersed in a neutral atmospheric boundary layer (ABL) is presented. Particular attention was paid to the influence of turbulence model parameters on the flow and concentration field. In addition, an approach to estimate Sc _t and T _L based on the calculated flow field is proposed. The results show that applying modified turbulence model constants to enable correct modelling of the ABL improves the prediction for the velocity and concentration fields when the modification is restricted to the region for which it was derived. The difference between simulated and measured concentrations is smaller than 25% or the uncertainty of the data on 76% of the points when solving the transport equation for a scalar with the proposed formulation for Sc _t , and on 69% of the points when using the Lagrangian particle tracking with the proposed formulation for T _L .     

Application of a Bivariate Gamma Distribution for a Chemically Reacting Plume in the Atmosphere

The joint concentration probability density function of two reactive chemical species is modelled using a bivariate Gamma distribution coupled with a three-dimensional fluctuating plume model able to simulate the diffusion and mixing of turbulent plumes. A wind-tunnel experiment (Brown and Bilger, J Fluid Mech 312:373–407, 1996), carried out in homogeneous unbounded turbulence, in which nitrogen oxide is released from a point source in an ozone doped background and the chemical reactions take place in non-equilibrium conditions, is considered as a test case. The model is based on a stochastic Langevin equation reproducing the barycentre position distribution through a proper low-pass filter for the turbulence length scales. While the meandering large-scale motion of the plume is directly simulated, the internal mixing relative to the centroid is reproduced using a bivariate Gamma density function. The effect of turbulence on the chemical reaction (segregation), which in this case has not yet attained equilibrium, is directly evaluated through the covariance of the tracer concentration fields. The computed mean concentrations and the O₃–NO concentration covariance are also compared with those obtained by the Alessandrini and Ferrero Lagrangian single particle model (Alessandrini and Ferrero, Physica A 388:1375–1387, 2009) that entails an ad hoc parametrization for the segregation coefficient.     

Mechanisms Controlling Turbulence Development Across A Forest Edge

In this paper we discuss the development of turbulence back from the transition fromopen moorland to a forest. Data from a field study and a wind-tunnel experiment arepresented. These show that the variance in the streamwise velocity begins to adjust tothe new surface between 2 to 4 tree heights downwind of the transition. This is soonerthan either the vertical velocity variance or the shear stress, both of which begin to adjust in a zone 3 to 5 tree heights downwind of the edge. Key terms in the prognostic equations for streamwise and vertical velocity variance are evaluated in order to explain these differences. The flow distortion caused by the forest edge, which extends to 4 tree heights downwind of the forest edge, is shown to be crucial in the delayed turbulence development. Initially the shear production term, which is the dominant source for the streamwise velocity variance, is counteracted by a sink in the vertical advection term. After the flow levels out the pressure redistribution (return-to-isotropy) term becomes the main sink of streamwisevelocity variance and feeds energy into the vertical velocity component. Therefore, thedevelopment of the vertical velocity variance and shear stress cannot begin until afterdevelopment of an increase in the streamwise velocity variance. Results are comparedwith other experiments, including the flow across shelterbelts, and large-eddy simulations of forest flow.     

Large-Eddy Simulation of Coherent Turbulence Structures Associated with Scalar Ramps Over Plant Canopies

Large-eddy simulations were performed of a neutrally-stratified turbulent flow within and above an ideal, horizontally- and vertically-homogeneous plant canopy. Three simulations were performed for shear-driven flows in small and large computational domains, and a pressure-driven flow in a small domain, to enable the nature of canopy turbulence unaffected by external conditions to be captured. The simulations reproduced quite realistic canopy turbulence characteristics, including typical ramp structures appearing in time traces of the scalar concentration near the canopy top. Then, the spatial structure of the organised turbulence that caused the scalar ramps was examined using conditional sampling of three-dimensional instantaneous fields, triggered by the occurrence of ramp structures. A wavelet transform was used for the detection of ramp structures in the time traces. The ensemble-averaged results illustrate that the scalar ramps are associated with the microfrontal structure in the scalar, the ejection-sweep structure in the streamwise and vertical velocities, a laterally divergent flow just around the ramp-detection point, and a positive, vertically-coherent pressure perturbation. These vertical structures were consistent with previous measurements made in fields or wind tunnels. However, the most striking feature is that the horizontal slice of the same structure revealed a streamwise-elongated region of high-speed streamwise velocity impacting on another elongated region of low-speed velocity. These elongated structures resemble the so-called streak structures that are commonly observed in near-wall shear layers. Since elongated structures of essentially similar spatial scales were observed in all of the runs, these streak structures appear to be inherent in near-canopy turbulence. Presumably, strong wind shear formed just above the canopy is involved in their formation. By synthesis of the ensemble-averaged and instantaneous results, the following processes were inferred for the development of scalar microfronts and their associated flow structures: (1) a distinct scalar microfront develops where a coherent downdraft associated with a high-speed streak penetrates into the region of a low-speed streak; (2) a stagnation in flow between two streaks of different velocities builds up a vertically-coherent high-pressure region there; (3) the pressure gradients around the high-pressure region work to reduce the longitudinal variations in streamwise velocity and to enhance the laterally-divergent flow and lifted updrafts downstream of the microfront; (4) as the coherent mother downdraft impinges on the canopy, canopy-scale eddies are formed near the canopy top in a similar manner as observed in conventional mixing-layer turbulence.     

论边界层中的大气扩散PDF模式

基于大气扩散Ｋ理论，用作为风速脉动均方差和拉氏时间尺度函数的湍流交换系数，得到了直接利用风速脉动几率密度而不用扩散参数的大气扩散ＰＤＦ模式。分别研究了对流边界层上升气流区与下降区垂直速度的统计特征，求得双正态ＰＤＦ模式。在给定ＣＢＬ自身参数如对流特征速度ｗ＊，顶高ｈｉ和源高度上的平均风速时，该模式计算出的无量纲浓度分布与室内外测试结果一致。     

CFD simulation of airflow over a regular array of cubes. Part II: analysis of spatial average properties

In the first part of this study, results of a computational fluid dynamics simulation over an array of cubes have been validated against a set of wind-tunnel measurements. In Part II, such numerical results are used to investigate spatially-averaged properties of the flow and passive tracer dispersion that are of interest for high resolution urban mesoscale modelling (e.g. non resolved obstacle approaches). The results show that vertical profiles of mean horizontal wind are linear within the canopy and logarithmic above. The drag coefficient, derived from the numerical results using the classical formula for the drag force, is height dependent (it decreases with height). However, a modification of the formula is proposed (accounting for subgrid velocity scales) that makes the drag coefficient constant with height. Results also show that the dispersive fluxes are similar in magnitude to the turbulent fluxes, and that they play a very important role within the canopy. Vertical profiles of turbulent length scales (to be used in k–l closure schemes, where k is the turbulent kinetic energy and l a turbulent length scale) are also derived. Finally the distribution of the values around the mean over the reference volumes are analysed for wind and tracer concentrations.     

Intermittency of turbulence within open canopies

Eddy covariance data have been analyzed to examine intermittency and clustering properties of turbulence within open canopies. Intermittency consists of two aspects: one is related to amplitude variation and the other to clustering. Using the telegraph approximation (TA), the clustering properties have been separated from amplitude effects. Intermittency of canopy turbulence has been explored via clustering exponent, probability density distribution of inter-pulse period of TA, intermittency exponent and structure kurtosis. Intermittency and clustering properties of turbulence within open canopies show similar features to those within dense canopy but some differences are also noted. Unlike within a dense canopy, temperature does not show larger clustering than velocity, which seems to be due to a different thermal structure of the sub-canopy and larger vertical scale of canopy eddy within open canopies. Within the crown region, the inter-pulse probability distribution of TA does not show the ‘double regime’ which was observed within the crown of a dense canopy, indicating less influence of near-field source on canopy turbulence within open canopies. For TA series of the flow variables, intermittency exponent is higher for temperature than for two velocity components within open canopies, which are opposite within a dense canopy. When comparing intermittency for flow variables and their TA series, it is shown that amplitude variation mitigates intermittency for both velocity components and temperature although amplitude variations play a much larger role in velocity intermittency than in temperature counterpart. Kurtosis analysis demonstrates that structure kurtosis is higher at large scales in stable conditions than in unstable conditions, indicating the existence of global intermittency due to stable stratification. The intermittency features of canopy turbulence within open canopies have been discussed in comparison with those within a dense canopy.     

Particle Image Velocimetry Measurements of Turbulent Flow Within Outdoor and Indoor Urban Scale Models and Flushing Motions in Urban Canopy Layers

We investigate the spatial characteristics of urban-like canopy flow by applying particle image velocimetry (PIV) to atmospheric turbulence. The study site was a Comprehensive Outdoor Scale MOdel (COSMO) experiment for urban climate in Japan. The PIV system captured the two-dimensional flow field within the canopy layer continuously for an hour with a sampling frequency of 30 Hz, thereby providing reliable outdoor turbulence statistics. PIV measurements in a wind-tunnel facility using similar roughness geometry, but with a lower sampling frequency of 4 Hz, were also done for comparison. The turbulent momentum flux from COSMO, and the wind tunnel showed similar values and distributions when scaled using friction velocity. Some different characteristics between outdoor and indoor flow fields were mainly caused by the larger fluctuations in wind direction for the atmospheric turbulence. The focus of the analysis is on a variety of instantaneous turbulent flow structures. One remarkable flow structure is termed ‘flushing’, that is, a large-scale upward motion prevailing across the whole vertical cross-section of a building gap. This is observed intermittently, whereby tracer particles are flushed vertically out from the canopy layer. Flushing phenomena are also observed in the wind tunnel where there is neither thermal stratification nor outer-layer turbulence. It is suggested that flushing phenomena are correlated with the passing of large-scale low-momentum regions above the canopy.     

Mass Transfer Velocity and Momentum Vertical Exchange in Simulated Deep Street Canyons

A box model to simulate mass transfer inside deep street canyons and with atmospheric flow above is introduced and discussed. Two ideal deep street canyons with aspect ratios of 3 and 5 (the aspect ratio being the ratio between building height and street width H/W) are considered. This range of aspect ratios, found in many densely populated historical centres in Mediterranean cities as well as in other cities around the world, potentially creates high air pollutant concentration levels. Our model is based on a combination of analytical solutions and computation fluid dynamics (CFD) simulations using carbon monoxide (CO) as a tracer pollutant. The analytical part of the model is based on mass transfer velocity concepts while CFD simulations are used both for a preliminary validation of the physical hypothesis underlying the model (steady-state simulations) and to evaluate the concentration pattern with time (transient or wash-out simulations). Wash-out simulation curves were fitted by model curves, and mass transfer velocities were evaluated through a best-fitting procedure. Upon introducing into the model the contribution of traffic-produced turbulence, the modelled CO concentration levels became comparable with those obtained in real-world monitoring campaigns. The mass transfer rate between the canyon and the above atmosphere was then expressed in terms of an overall mass transfer velocity, which directly allows the evaluation of the mass transfer rate between the bottom volume of the canyon (pedestrian level) with the above atmosphere. Overall mass transfer velocities are reported as a function of the operating conditions studied (H/W = 3–5 and wind speeds = 2–8 ms⁻¹). Finally, a simple expression is reported for determining pollutant concentrations at the pedestrian level based on the overall mass transfer velocity defined.     

An Analytical one-Dimensional Second-Order Closure Model of Turbulence Statistics and the Lagrangian Time Scale Within and Above Plant Canopies of Arbitrary Structure

An analytical one-dimensional second-order closure model is developed to describe the within canopy velocity variances, turbulent intensities, dissipation rates, Lagrangian time scale and Lagrangian far field diffusivities for vegetation canopies of arbitrary structure and density. The model incorporates and extends the model of momentum transfer developed by Massman (1997) and the model of within canopy velocity variances developed by Weil (unpublished) from the second-order closure model of Wilson and Shaw (1977). Model predictions of within and above canopy velocity variances, turbulent intensities, dissipation rates and the Lagrangian time scale are in reasonable agreement with previously measured or estimated values for these parameters. The present model suggests that the Lagrangian time scale and the far field diffusivity could be strongly dependent upon foliage structure and density through the foliage effects on the velocity variances. A simple formulation for the Lagrangian time scale at canopy height is derived from model results. Taken as a whole, the present model may provide a relatively simple way to incorporate turbulence parameters into models of soil/canopy/atmosphere mass transfer.     

Coherent eddies and turbulence in vegetation canopies: The mixing-layer analogy

This paper argues that the active turbulence and coherent motions near the top of a vegetation canopy are patterned on a plane mixing layer, because of instabilities associated with the characteristic strong inflection in the mean velocity profile. Mixing-layer turbulence, formed around the inflectional mean velocity profile which develops between two coflowing streams of different velocities, differs in several ways from turbulence in a surface layer. Through these differences, the mixing-layer analogy provides an explanation for many of the observed distinctive features of canopy turbulence. These include: (a) ratios between components of the Reynolds stress tensor; (b) the ratio K _H/K _M of the eddy diffusivities for heat and momentum; (c) the relative roles of ejections and sweeps; (d) the behaviour of the turbulent energy balance, particularly the major role of turbulent transport; and (e) the behaviour of the turbulent length scales of the active coherent motions (the dominant eddies responsible for vertical transfer near the top of the canopy). It is predicted that these length scales are controlled by the shear length scale % MathType!MTEF!2!1!+-% feaafeart1ev1aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn% hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr% 4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq-Jc9% vqaqpepm0xbba9pwe9Q8fs0-yqaqpepae9pg0FirpepeKkFr0xfr-x% fr-xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaGaamitamaaBa% aaleaacaWGtbaabeaakiabg2da9iaadwfacaGGOaGaamiAaiaacMca% caGGVaGabmyvayaafaGaaiikaiaadIgacaGGPaaaaa!3FD0!\[L_S = U(h)/U'(h)\] (where h is canopy height, U(z) is mean velocity as a function of height z, and % MathType!MTEF!2!1!+-% feaafeart1ev1aaatCvAUfeBSjuyZL2yd9gzLbvyNv2CaerbuLwBLn% hiov2DGi1BTfMBaeXatLxBI9gBaerbd9wDYLwzYbItLDharqqtubsr% 4rNCHbGeaGqiVu0Je9sqqrpepC0xbbL8F4rqqrFfpeea0xe9Lq-Jc9% vqaqpepm0xbba9pwe9Q8fs0-yqaqpepae9pg0FirpepeKkFr0xfr-x% fr-xb9adbaqaaeGaciGaaiaabeqaamaabaabaaGcbaGabmyvayaafa% Gaeyypa0JaaeizaiaadwfacaGGVaGaaeizaiaadQhaaaa!3C32!\[U' = {\rm{d}}U/{\rm{d}}z\]). In particular, the streamwise spacing of the dominant canopy eddies is _x = mL _s, with m = 8.1. These predictions are tested against many sets of field and wind-tunnel data. We propose a picture of canopy turbulence in which eddies associated with inflectional instabilities are modulated by larger-scale, inactive turbulence, which is quasi-horizontal on the scale of the canopy.