betaFeedback

WorksPhysics and Astronomy

Turbulent velocity statistics in convective boundary layers

Abstract

Large-eddy simulations (LES) of the convective atmospheric boundary layer (CBL) are used to investigate velocity variance statistics in turbulent sheared convection. The thirteen simulations span conditions from weak convection with longitudinal rolls to free penetrative convection with no mean wind. In the presence of thermal rolls, the near-surface streamwise variance profile exhibits an approximately logarithmic decay with a slope proportional to the friction velocity, u∗u_*, and an added constant proportional to the convective velocity, w⋆w_\star. After the structural transition to convective cells, both the streamwise and spanwise variances follow the empirical stability and height dependencies of Wilson (Boundary-Layer Meteorol., 2008, vol. 129, pp. 353–369). The near-surface vertical (wall-normal) velocity variance is well described by Monin–Obukhov similarity with improved agreement using local-in-height scaling. Within the mixed layer, the peak vertical variance initially decreases relative to w⋆w_\star as shear diminishes, before increasing and shifting downward as the flow approaches free convection. In contrast, the total turbulent energy decreases relative to w⋆w_\star across the parameter space. The spanwise velocity variance represents the dominant turbulent component under weak convection and exceeds the streamwise variance throughout most of the convective regime. These statistical trends are attributed to the role of pressure, evaluated through variance budget contributions. Large-scale buoyancy production in the mixed layer is transported downward by pressure and redistributed preferentially to the spanwise component, coinciding with the presence of longitudinal roll motions. The weakening of this pressure-transport pathway for convective cells coincides with greater retention of vertical variance within the mixed layer.

Introduction

Sheared convective boundary layers represent a fundamental class of fluid flows with widespread relevance across engineering and the natural sciences. These environments are characterised by the complex interaction between wall-bounded shear turbulence and positive surface buoyancy ([52, 37 Salesky et al.] 2017; Bou-Zeid et al. [10]; Chávez-Medina et al. 2025, with applications ranging from predicting heat transfer in industrial thermal management systems (Blass et al. [9]; Wang et al. [81]) to understanding land-atmosphere interactions (Albertson & Parlange [2]; Stoll et al. [74]). The convective atmospheric boundary layer (CBL) serves as a prominent and widely studied example of this physical system within the atmospheric sciences, controlling the vertical transport of momentum, energy, moisture, and trace chemicals between the Earth’s surface and the free troposphere (Brutsaert [12]; Stull [75]; Garratt [20]; Wyngaard [86]; Dewani et al. 2023).

A notable characteristic of these environments is the large-scale structural organisation of the turbulent motions. As the relative influence of buoyancy and shear changes, the flow undergoes a transition from highly-organised longitudinal rolls aligned with the mean wind to buoyancy-driven polygonal cells (Sykes & Henn [77]; Khanna & Brasseur [37]; Weckwerth et al. [83]; Salesky et al. [70]). The topological shift from rolls to cells occurs for all sheared convection configurations, including wall-sheared Rayleigh–Bénard convection and various internal flows (Pandey et al. [58]; Blass et al. [9]; Wang et al. [82]). This universality suggests that these structures and their dynamical influence have broad relevance to fundamental fluid mechanics across applications. Despite the well-documented visual evolution of these large-scale structures, the turbulent second-order statistics associated with these structural transitions are less well understood and agreed upon (Salesky et al. [70]; Zhou et al. [92]). Further, the exact physical mechanisms by which turbulent kinetic energy (TKE) is produced, redistributed among the velocity components, and dissipated as the flow reorganises remain an active area of research (Ding et al. [18]; Bou-Zeid et al. [10]).

Within the surface layer of the CBL, variance statistics and profiles are traditionally evaluated within the framework of Monin–Obukhov similarity theory (MOST), which does not explicitly account for the large-scale structural organisation. MOST postulates that these near-surface turbulent statistics follow a universal function of the surface friction velocity u∗u_* and a local stability parameter ζ=z/L\zeta= z/L (Foken [19]), where zz indicates the height (or wall-normal distance) above the surface and LL is the Obukhov length:

L=−u∗3θ0κgQ.(1)L = -\frac{u_*^3\theta_0}{\kappa g Q}. \tag*{(1)}

Here, κ≈0.4\kappa\approx0.4 is the von Kármán constant, gg is gravitational acceleration, θ0\theta_0 is the reference potential temperature, QQ is the surface kinematic heat flux, and LL is negative for unstable convection with positive QQ. The global importance of shear versus buoyancy across the entire boundary layer is typically represented by the bulk stability parameter −zi/L-z_i/L, where ziz_i is the overall CBL depth. In this framework, −zi/L→0-z_i/L \to0 corresponds to a neutrally-stratified boundary layer with no net buoyancy, while −zi/L→∞-z_i/L \to\infty corresponds to the absolute free-convection limit where mean wind shear is entirely absent. The conventional equation forms derived from MOST to model the normalised variances of the fluctuating streamwise (u′u'), spanwise (v′v'), and vertical (wall-normal, w′w') velocity components are typically empirical power-law functions of z/Lz/L or zi/Lz_i/L. While historically foundational, this framework remains strictly constrained to the aforementioned surface layer (i.e. logarithmic region), comprising the lowest 10% of the CBL in the high-Reynolds-number limit (Garratt [20]; Yaglom [89]). Accurately mapping the behaviour of the turbulence statistics across the parameter space of sheared convection therefore requires consideration of the neutral and free-convection limits, in addition to the constraints imposed by the outer layer farther from the surface.

Deardorff [16] used dimensional analysis to predict the scaling of turbulence in buoyancy-dominated conditions. The scales u∗u_\ast and zz associated with wall-bounded shear become irrelevant both in the global limit of free convection and far from the surface in the convective mixed layer. Instead, large-scale turbulent behaviour is controlled by the surface kinematic heat flux and the CBL depth, which combine to form the convective velocity scale w∗=(gQzi/θ0)1/3w_\ast= (gQz_i/\theta_0)^{1/3} (Deardorff [16]). The balance of the shear and buoyancy velocity scales is directly related to the bulk stability parameter as u∗/w∗=κ1/3(−zi/L)−1/3u_\ast/w_\ast= \kappa^{1/3}(-z_i/L)^{-1/3}, providing a direct connection between −zi/L-z_i/L and the transition in scaling of turbulent velocities in these flows.

Conversely, in the neutral limit −zi/L→0-z_i/L \to0, turbulence is governed by the aforementioned u∗u_\ast and zz scales. Variance statistics are well-predicted by the attached eddy hypothesis in which eddies have size confined by the distance from the wall (Townsend [80]). The vertical velocity variance is approximately constant in the near-surface logarithmic region, while the horizontal variances exhibit a logarithmic decay with height:

⟨u′2⟩u∗2=Bu−Aulog⁡(z/zi).(2)\frac{\langle u'^2\rangle}{u_\ast^2} = B_u - A_u \log(z/z_i). \tag*{(2)}

Here, AuA_u and BuB_u are constant coefficients and angle brackets indicate ensemble averaging. Extensive experimental and numerical evidence from wall-bounded shear flows has since supported these neutral-limit predictions (see, e.g., [48, 4, 90, 67]), establishing baseline structural profiles for purely shear-driven turbulence. As seen in Table 1, the constant value for the vertical variance is in the range Bw≈1.45B_w \approx1.45 to 1.9 ([59, 61, 26]). Values for the constants AuA_u and BuB_u in (2) are given in Table 2, noting BuB_u is dependent on the flow configuration. Both variances are known to have corrections for finite Reynolds numbers ([62, 63, 26] (Bernardini et al. 2014)), but only asymptotically high Reynolds number CBLs are considered here following the traditional assumption for atmospheric flows.

ConditionFormulationEmpirical constantsSources
Neutral surface (log) layer⟨w2⟩/u∗2=Bw\langle w^{2}\rangle/u_{*}^{2}=B_{w}Bw≈1.45B_{w}\approx1.45–1.91.9Townsend [80]; Panofsky & Dutton [59]; Perry & Li [61]; Kunkel & Marusic [40]; Heisel et al. [26]
Convective surface layer⟨w2⟩/u∗2=Bw12(1−Aw1ζ)2/3\langle w^{2}\rangle/u_{*}^{2}=B_{w1}^{2}(1-A_{w1}\zeta)^{2/3}Aw1≈3.0A_{w1}\approx3.0–9.59.5<br>Bw1≈0.8B_{w1}\approx0.8–1.31.3Panofsky et al. [60]; Wilson [85]; Maronga & Reuder [47]; Kaimal & Finnigan [34]; Khanna & Brasseur [36]
Local free-convection surface layer (−ζ≫1-\zeta\gg1)⟨w2⟩/u∗2=Aw22(−ζ)2/3\langle w^{2}\rangle/u_{*}^{2}=A_{w2}^{2}(-\zeta)^{2/3}Aw2≈1.9A_{w2}\approx1.9Priestley [66]; Wyngaard et al. [88]
Convective mixed layer⟨w2⟩/w∗2=Bw3(z/zi)2/3(1−Aw3 z/zi)p\langle w^{2}\rangle/w_{*}^{2}=B_{w3}(z/z_{i})^{2/3}(1-A_{w3}\,z/z_{i})^{p}Aw3≈0.8A_{w3}\approx0.8–1.01.0<br>Bw3≈1.17B_{w3}\approx1.17–1.81.8<br>p∈{2/3,2}p\in\{2/3,2\}Lenschow et al. [45]; Sorbjan [72]; Zhou et al. [92]; Dewani et al. (2023)

Table 1. Common formulations for predicting the vertical velocity variance ⟨w2⟩\langle w^{2}\rangle across different conditions within convective boundary layers (CBLs), noting some formulas are derived from definitions for the standard deviation.

ConditionFormulationEmpirical constantsSources
Neutral surface (log) layer⟨u′2⟩/u∗2=Bu−Aulog⁡(z/zi)\langle u'^2\rangle/u_*^2 = B_u - A_u \log(z/z_i)Au≈0.91A_u \approx0.91–1.331.33, BuB_u flow dependentTownsend [80]; Perry et al. [62]; Marusic et al. [49]; Bernardini et al. (2014); Banerjee et al. [7]; Baars & Marusic [4]; Huang & Katul [31]
Convective surface layer⟨u′2⟩/u∗2=[12+0.5(−zi/L)]2/3\langle u'^2\rangle/u_*^2 = [12 + 0.5(-z_i/L)]^{2/3}—Panofsky et al. [60]
Convective surface layer⟨u′2⟩u∗2=[4+0.6(−ziL)2/3]×[1−(zzi)0.38]\frac{\langle u'^2\rangle}{u_*^2} = \left[4 + 0.6\left(\frac{-z_i}{L}\right)^{2/3}\right]\times\left[1-\left(\frac{z}{z_i}\right)^{0.38}\right]—Wilson [85]
Convective surface layer⟨u′2⟩/u∗2=Bu−Aulog⁡(z/zi)\langle u'^2\rangle/u_*^2 = B_u - A_u \log(z/z_i)Au=f(ζ)A_u = f(\zeta), Bu=f(ζ)B_u = f(\zeta)Banerjee et al. [7]
Convective surface layer⟨u′2⟩/u∗2\langle u'^2\rangle/u_*^2Au2=f(−zi/L)A_{u_2} = f(-z_i/L), Bu2=f(−zi/L)B_{u_2} = f(-z_i/L)Yang et al. [90]
Convective surface layerAu22[U∞/u∗−U/u∗]+Bu2A_{u_2}^2[U_\infty/u_* - U/u_*] + B_{u_2}
Convective mixed layer⟨u′2⟩/w∗2=f(z/zi,−zi/L)≈Bu3\langle u'^2\rangle/w_*^2 = f(z/z_i,-z_i/L) \approx B_{u_3}Bu3≈0.35B_{u_3} \approx0.35Caughey & Palmer [13]; Khanna & Brasseur [36]; Khanna & Brasseur [37]

Table 2. Same as Table 1, except the formulations are for the streamwise velocity variance ⟨u′2⟩\langle u'^2\rangle. Here, U∞U_\infty denotes the outer or reference mean streamwise velocity.

Historical predictions for the vertical velocity variance in the CBL are consistent with the constraints discussed above. The most common functional form shown in Table 1 for the convective surface layer spans from the constant neutral baseline to convective scaling, and traces the transition between them via MOST. In MOST, the convective velocity in the free-convection limit is a variation of w∗w_\ast defined using zz rather than ziz_i due to the definition of ζ\zeta. This limit is known as local free convection owing to the dependence on the local-in-height stability ζ\zeta as opposed to global convection based on zi/Lz_i/L ([88]). Observations of ⟨w′2⟩\langle w'^2\rangle across a diverse range of simulations and atmospheric field measurements are well described by this functional form (e.g., [35, 60, 36, 85, 47]). For the bulk of the CBL above the surface layer, universal functions for ⟨w′2⟩\langle w'^2\rangle have been proposed using the global convective velocity w∗w_\ast and the relative height z/ziz/z_i (Lenschow et al. [45]). Similar to the surface layer, measurements and simulations have been shown to fit this functional form with some variability in the coefficients and exponents (e.g., [13, 72, 44, 92] (Dewani et al. 2023)). Other formulations and conditions for ⟨w′2⟩\langle w'^2\rangle have been proposed in the literature – such as those incorporating shear-buoyancy superposition ([93, 94]), local buoyancy-shear scaling ([56, 71]), or dependence on anisotropy ([73]) – but these are not evaluated in the present analysis, aside from local scaling in the surface layer.

Unlike the vertical variance, early empirical models for the streamwise variance ⟨u2⟩\langle u^{2}\rangle in the CBL were inconsistent with the neutral limit in (2). The first predictions for near-neutral conditions considered a constant ⟨u2⟩/u∗2≈Bu2\langle u^{2}\rangle/u_{*}^{2}\approx B_{u2} with no height dependence (Lumley & Panofsky [46]). Panofsky et al. [60] later added a zi/Lz_{i}/L correction to Bu2B_{u2} to account for a wider range of convection. This formulation follows the assumption that horizontal velocity fluctuations in the surface layer are largely governed by large-scale, buoyancy-driven convective structures that extend throughout the mixed layer, rendering the boundary-layer depth ziz_{i} the relevant scaling parameter rather than ζ\zeta (MOST). More recently, Wilson [85] incorporated a power-law height dependence with the Panofsky et al. [60] relation to better fit field observations. Further efforts to address the limitations of MOST in parameterising ⟨u2⟩\langle u^{2}\rangle include introducing a horizontal length scale for multipoint Monin–Obukhov similarity (Tong & Nguyen [79]) and predicting variances based on anisotropy of the Reynolds-stress tensor (Stiperski & Calaf [73]). However, these frameworks do not explicitly account for the logarithmic profile and presence of attached eddies in the neutral limit.

Banerjee et al. [7] first attempted to reconcile ⟨u2⟩\langle u^{2}\rangle in convective conditions with (2) using a spectral budget method, yielding the same logarithmic equation but with the coefficients depending on local stability ζ\zeta. In an alternate approach, Yang et al. [90] relaxed Townsend’s [80] assumption of an attached eddy population to structural modifications to the attached eddies under weak to moderate convection, but have received limited corroboration across a wide range of conditions.

Further aloft in the convective mixed layer where mechanical shear is negligible, ⟨u′2⟩\langle u'^2\rangle is characterised by the convective velocity scale w∗w_*. While ⟨u′2⟩\langle u'^2\rangle exhibits systematic height and stability dependence in the lower mixed layer due to the interaction of near-surface shear and convective eddies ([36, 37]), it has historically been approximated in the outer mixed layer as a vertically uniform profile ([60, 13]). The approximately constant variance with respect to w∗w_* reflects the dominant role of non-local, boundary-layer-spanning convective structures that carry outer-layer velocity scales down to the surface ([94]).

For the spanwise (cross-wind) variance, ⟨v′2⟩\langle v'^2\rangle, the logarithmic spatial decay evaluated in (2) is consistent with predictions and observations under neutral conditions ([80, 41, 95, 5]). Within the convective surface layer, proposed formulations for ⟨v′2⟩\langle v'^2\rangle generally follow the same empirical corrections applied to the streamwise component: the variance is evaluated as a constant relative to u∗u_* in the neutral limit of MOST ([54]), a −zi/L-z_i/L stability correction accounts for buoyancy effects in the surface layer ([87, 60, 28]), and the variance is considered constant relative to w∗w_* in the mixed layer ([13, 94, 10]). Despite its integral role in characterising the three-dimensional (3D) structures of large-scale rolls and cells, the behaviour of ⟨v′2⟩\langle v'^2\rangle across the full range of convective conditions remains less investigated than its streamwise and vertical counterparts.

In this study, simulations across the parameter space of −zi/L-z_i/L are used to further investigate the variance formulations in Table 1 and Table 2, and to evaluate the scaling of turbulent statistics with u∗u_* and w∗w_*. Particular attention is paid to the spanwise velocity, which has historically received less attention within sheared convection literature. The objective is to reconcile the statistical observations with the structural transition between thermal rolls and cells. This is achieved by assessing contributions to the variance budget equations, including scale-wise spectral budgets. The budget analysis reveals the pathways of energy as it is produced and distributed across heights and velocity components, highlighting the important role of pressure in this process. The remainder of the work is organised as follows: § 2 outlines the methodology and numerical framework. § 3 and § 4 present the structural evolution of the variance profiles and the accompanying budget analyses. Finally, § 5 summarises the results and findings.

Methodology

Large-eddy simulations

A suite of thirteen high-resolution large-eddy simulations (LES) is used to investigate, in a systematic and quantitative way, the structural and statistical evolution of turbulent variances as convective instability increases from a shear-dominated boundary layer to the absolute limit of penetrative free convection. The first six cases in weaker convection were previously published [25], and this study expands the dataset to simulate the remainder of the parameter space. The numerical discretisation and boundary conditions remain consistent to ensure statistical fidelity.

The LES flow solver and numerical methods used in this study are described by Heisel & Chamecki [25] and are based on a modified version of the code developed by Albertson and co-authors [1, 2, 65]. The solver utilises a staggered vertical grid for the wall-normal velocity component to optimise vertical differentiation via second-order-accurate finite differencing. Horizontal derivatives are computed spectrally with full de-aliasing using the 3/2 zero-padding rule. The flow is advanced in time using a second-order-accurate Adams–Bashforth integration scheme. Boundary conditions are periodic in the horizontal directions, with zero penetration enforced at the bottom of the computational domain; the upper boundary employs a stress-free condition. Fluctuations in the top 25% of the domain are systematically relaxed using a Rayleigh damping layer to achieve the upper boundary conditions and mitigate gravity wave reflections [57].

The estimation of subgrid-scale (SGS) stresses uses a Lagrangian averaged scale-dependent (LASD) dynamic model to define the SGS eddy viscosity [11]. The SGS heat flux is closed using the same dynamic formulation alongside a constant subgrid turbulent Prandtl number, Prsgs=0.4Pr_{\mathrm{sgs}} = 0.4, with a comprehensive technical description of the numerical framework detailed elsewhere [39, 70]. The surface stress in this wall-modelled LES framework is estimated locally in space and time using MOST [51]. To mitigate log-law mismatch and wall-model overshoot, the wall model evaluates surface stresses using velocity values at 0.05zi0.05z_i instead of the first grid point [25].

The simulation parameters and resulting bulk characteristics are summarised in Table 3. Following Heisel & Chamecki [25], all simulations employ an aerodynamic surface roughness length of z0=0.1z_0 = 0.1 m, an initial reference potential temperature of θ0=300\theta_0 = 300 K, an imposed stably-stratified capping inversion with 0.003 K m−1^{-1} temperature gradient, and a constant Coriolis frequency of f=1×10−4f = 1 \times10^{-4} s−1^{-1}. The latter two forcings are widely utilised to represent realistic atmospheric conditions (see, e.g., Sullivan & Patton 2011; [70, 68] The imposed geostrophic wind velocity, UgU_g, decreases gradually from the maximum value Ug=15.0U_g = 15.0 m s−1^{-1} that characterises the initial cases dominated by shear to the limit of penetrative free convection (Ug=0.0U_g = 0.0 m s−1^{-1}). Penetrative convection in this context indicates the presence of an overlying stable capping inversion layer that bounds the convective mixed layer but can be penetrated by buoyant plumes. The imposed surface heat flux, QQ, increases from 0.10 K m s−1^{-1} to 0.24 K m s−1^{-1} across the initial shear-dominated simulations, and is subsequently held constant at 0.24 K m s−1^{-1} for the moderately to highly convective cases as UgU_g decreases. This results in a range −zi/L-z_i/L covering more than four orders of magnitude from −zi/L=2.6-z_i/L = 2.6 up to effectively infinity for the zero-mean-wind case. The finite reported value of 5.7×1045.7 \times10^4 for the free convection case (c13) in Table 3 results from a small residual mean stress with u⋆=0.02u_\star= 0.02 m s−1^{-1} arising from imperfect statistical convergence of the momentum flux.

CaseUgU_g (m s−1^{-1})QQ (K m s−1^{-1})Domain Size Lx×Ly×LzL_x \times L_y \times L_z (km)Grid Size NN (points)Turnover Times (zi/w⋆z_i/w_\star)Turnover Times (zi/u⋆z_i/u_\star)−zi/L-z_i/L
c1150.1012 × 12 × 21024 × 1024 × 5125.22.82.6
c2150.1712 × 12 × 21024 × 1024 × 5126.02.74.3
c3150.2412 × 12 × 21024 × 1024 × 5126.72.76.0
c4120.2412 × 12 × 21024 × 1024 × 5126.62.39.6
c590.2412 × 12 × 21024 × 1024 × 5128.82.518
c660.2412 × 12 × 21024 × 1024 × 5126.61.440
c750.245.76 × 5.76 × 2480 × 480 × 500203.955
c840.245.76 × 5.76 × 2480 × 480 × 500233.984
c930.245.76 × 5.76 × 2480 × 480 × 500273.8140
c1020.245.76 × 5.76 × 2480 × 480 × 500313.5260
c1110.245.76 × 5.76 × 2480 × 480 × 500342.7810
c120.50.245.76 × 5.76 × 2480 × 480 × 500382.12400
c1300.245.76 × 5.76 × 2480 × 480 × 500380.757 000

Table 3. Summary of the six previous and seven new LES cases, including the fixed geostrophic wind speed UgU_g, fixed surface heat flux QQ, duration relative to both convective (zi/w⋆z_i/w_\star) and shear (zi/u⋆z_i/u_\star) time scales, and the resulting bulk stability parameter −zi/L-z_i/L.

In the cases with weaker convection (c1–c6), a large area of 12 × 12 km is simulated in the horizontal and 2 km in the vertical. This horizontal domain corresponds to approximately 10zi10z_i to 12zi12z_i, consistent with previous LES studies ([70, 68, 15, 22]), which is required to avoid any artificial constriction of the highly elongated turbulent roll structures that scale with ziz_i.

Following the structural shift from rolls to cellular convection, the large-scale structures are less elongated in the horizontal and an O(10zi)O(10z_i) domain length is no longer required. Accordingly, the grid dimensions were adjusted for the strongly convective regime (c7–c13) to maintain computational efficiency. While the horizontal physical domain was reduced from 12 km to 5.76 km (∼5zi\sim5z_i), the spatial resolution was preserved by adjusting the grid-point allocation from 1024 × 1024 × 512 to 480 × 480 × 500. The minor adjustment in vertical grid points was required for parallel partitioning of the new simulations on the utilised computing infrastructure. The vertical grid spacing for all cases is Δz/zi≲0.004\Delta z/z_i \lesssim0.004. Case c6 was repeated using both domains to ensure the results are not affected by the smaller domain size. Appendix A shows that variance statistics closely match between the two domains.

All cases used an iterative spin-up procedure to mitigate inertial oscillations during the simulation [25]. The post-spin-up duration of each simulation varied from 60 physical minutes for case c1 to 5.5 hours for case c13. The goal was to capture sufficient turbulent events related to both buoyancy (whose relevant timescale is zi/w⋆z_i/w_\star) and shear (zi/u⋆z_i/u_\star timescale). This was assessed by the post-spin-up turnover times in Table 3 and the statistical convergence of the heat and momentum flux profiles. The long durations under strong convection were required to sufficiently capture the relatively infrequent, prolonged shear-related events and accurately evaluate their diminishing strength.

Figure 1(a) shows the mean wind magnitude for each LES case. In this and all later results, statistical moments are averaged across horizontal position and time. The profiles indicate the presence of a well-developed mixed layer, which is a common feature of the CBL [75]. Within this layer, buoyancy-driven thermal plumes efficiently redistribute momentum and potential temperature, producing nearly uniform mean profiles. Above this, there is an entrainment layer between the CBL and the capping inversion where the wind magnitude reaches UgU_g.

Vertical profiles of the mean wind properties for each LES case

Figure 1. Vertical profiles of the mean wind properties for each LES case in table 3 plotted as a function of normalised height ⟨z⟩/zi\langle z\rangle/z_i: (a) magnitude; (b) direction α\alpha relative to the LES grid. In all subsequent results, the xx–yy coordinate system is aligned with the near-surface wind direction, as shown in the inset of (b), which shows instantaneous w′w' in a horizontal plane at z/zi=0.1z/z_i = 0.1 for the weakest convective case c1.

In a rotating atmosphere governed by Coriolis forces, the mean wind vector typically changes direction with height through the planetary boundary layer [78, 23]. Throughout most of the CBL, this veering is mitigated by vertical mixing, as seen in the Figure 1(b) profiles of wind direction α\alpha. To account for the mismatch between the wind direction within the CBL and the imposed UgU_g aligned with the LES grid, the horizontal velocity components and their statistics are rotated according to the surface wind direction in this study. Because the wind direction is relatively constant throughout the surface and mixed layers (i.e. ±5∘\pm5^\circ for z/zi≲0.85z/z_i \lesssim0.85), a single fixed rotation by the mean surface direction at z/zi≤0.05z/z_i \le0.05 is applied across the full vertical domain. This fixed rotation provides a consistent streamwise–spanwise (x−yx-y) coordinate system that preserves the geometric orientation of boundary-layer-spanning structures across height while providing a physically meaningful decomposition of the turbulent fluctuations aligned with the near-surface mean flow and shear [84]. The rotation aligns the coordinate system with the orientation of the dominant turbulent structures, as illustrated by the elongated rolls in the Figure 1(b) inset.

The inset fluctuating field is shown for an additional five cases in Figure 2 using the rotated coordinate system. The instantaneous horizontal cross-sections at z=0.1ziz = 0.1z_i illustrate the change in coherent turbulence structure with increasing instability: the weakly convective near-surface flow is populated by longitudinal convective rolls that organise the flow into stable, counter-rotating circulations [43]. As instability increases, these large-scale structures gradually break down into a mixed roll–cellular organisation comprising predominantly buoyancy-driven cellular plume structures. Previous studies report that cells become dominant around −zi/L≈20-z_i/L \approx20–25 [23, 83, 70] based on the visual end of roll and cell coexistence [23], the loss of linear organisation [83], and a geometric roll factor [70]. Here, the absence of longitudinal coherence associated with rolls occurs for −zi/L≈50-z_i/L \approx50 between Figure 2(c) and Figure 2(d) based on visual observation and the same roll factor. This higher threshold is further supported by later quantitative results. The later transition relative to −zi/L-z_i/L may be due to a number of methodological factors: the impact of inertial oscillations and wall stress overshoot on u∗u_\ast and LL [25], which have both been mitigated here; the substantial difference in spatial resolution; and different grid aspect ratios. Considering the uncertainty from these factors, the present value −zi/L≈50-z_i/L \approx50 is used only to differentiate trends before and after the transition, and no conclusions are made regarding a universal −zi/L-z_i/L value for the transition.

Instantaneous fluctuations of normalised vertical velocity $w'/w_\star$ in the horizontal plane

Figure 2. Instantaneous fluctuations of normalised vertical velocity w′/w⋆w'/w_\star in the horizontal plane at z/zi=0.1z/z_i = 0.1 across six cases to visualise the transition from thermal rolls to cellular plumes: (a) c1, (b) c5, (c) c6, (d) c7, (e) c8, (f) c13. In each case the plane is rotated to align with the surface wind direction and confined to the same area relative to ziz_i.

Temporal averaging and normalisation

For convective conditions with a stable capping inversion, the CBL slowly grows as the inversion is entrained into the boundary layer. This growth is significant over the duration of the highly convective cases. For example, the capping inversion depth, ziz_i, grows by up to 65% in c12. In the setup of each case, the initial capping inversion was positioned to avoid growth of the CBL into the damping layer by the end of the simulation. An important additional consideration is that the key scaling parameters, ziz_i and w⋆w_\star, evolve as the simulation progresses.

Figure 3 demonstrates the influence of the growing CBL on time-averaged statistics using case c12. The coloured lines represent independent profiles averaged within ten one-minute intervals spaced throughout the simulation time tt, while the solid black curve denotes the overall time-averaged profile. The growth over time of the CBL is evident in Figure 3(a,Figure 3c), noting the capping inversion depth ziz_i is defined as the height of the minimum in the kinematic heat flux profile. This growth leads to a smoothing of the time-averaged profile, particularly near the inversion. In contrast, applying local-in-time scaling using zi(t)z_i(t) and w⋆(t)w_\star(t) produces substantially improved profile collapse in Figure 3(b,Figure 3d), with the individual time intervals remaining closely aligned with the overall

Profiles of total heat flux and vertical velocity variance

Figure 3. Profiles of total heat flux ⟨w′θ′⟩\langle w'\theta'\rangle (top row) and vertical velocity variance ⟨w′2⟩\langle w'^2\rangle (bottom row) at different time intervals during case c12, including the time-averaged profile (black curve). In (a,c), statistics are normalised by the overall time-averaged CBL depth ⟨zi⟩\langle z_i\rangle and convective velocity ⟨w⋆⟩\langle w_\star\rangle. In (b,d), local-in-time scaling is applied before temporal averaging using the instantaneous parameters zi(t)z_i(t) and w⋆(t)w_\star(t). In this and later figures, the star (⋆\star) superscript denotes local-in-time normalised statistics, e.g. z⋆=⟨z/zi(t)⟩z^\star=\langle z/z_i(t)\rangle and w′2⋆=⟨w′2/w⋆2(t)⟩w'^{2\star}=\langle w'^{2}/w_\star^2(t)\rangle.

mean. The resulting collapse indicates that local-in-time normalisation accounts for the evolving convective scales and avoids contamination of statistics due to non-stationary CBL growth. Accordingly, all statistics in this study – including for Figure 1 and Figure 2 – are normalised using local-in-time parameters before averaging across time. Here, the star (⋆\star) superscript denotes normalisation using the instantaneous convective scales. The normalised vertical coordinate is defined as z⋆=⟨z/zi(t)⟩z^\star= \langle z/z_i(t) \rangle, and the normalised velocity variances are w′2⋆=⟨w′2/w⋆2(t)⟩w^{\prime2\star} = \langle w^{\prime2}/w_\star^2(t) \rangle, where time-averaging indicated by brackets occurs after normalisation. Owing to the CBL growth, the z⋆z^\star points represented by the LES grid change over time. For computing statistics as a function of z⋆z^\star, the local-in-time statistics were first interpolated onto a common z/ziz/z_i grid before averaging the statistics across time.

Subgrid-scale variance statistics

The SGS model of the LES solver does not directly estimate the SGS variance (i.e. normal stresses), and the corresponding SGS normal stresses are instead grouped with the pressure term [1]. Because the SGS energy can be substantial near the surface where characteristic turbulent length scales are restricted, the SGS contribution to the variance statistics is estimated in post-processing for all cases to ensure the results are independent of the grid resolution.

The SGS variance estimation extrapolates the energy to subgrid scales using the Leonard stress and assuming a −5/3-5/3 inertial subrange [38, 69]. The SGS variances for uu and vv are transformed using the same coordinate rotation. described above for the resolved statistics. The SGS contributions are then added to the resolved quantities prior to normalisation and time averaging.

The SGS estimation relies on the grid resolution being within the inertial subrange of scales exhibiting a −5/3-5/3 power law in the spectrum. To evaluate this criterion, SGS variances were estimated using a range of grid and test filter sizes to compute the Leonard stress, where consistency across filters indicates the assumption is valid and the estimate is reliable. Variance statistics are only presented for points above a cutoff height where the filter comparisons converge, and near-surface points with discrepancies across filter sizes are excluded. Further details of the SGS estimate and the determination of the cutoff height are provided in Appendix B.

Resolved pressure statistics

As noted above, kinematic pressure statistics from the LES solver include the SGS kinetic energy. In comparison with literature [18], it was found that some statistics derived from this pressure such as the variance compare well with the true pressure, while statistics dependent on smaller scales such as the pressure–strain correlation show substantial discrepancies. Correcting the pressure using estimates of the instantaneous SGS energy is not reliable because the SGS estimates only work well for the average contribution [38].

Instead, the resolved pressure field is determined offline via the Poisson equation derived from the Navier–Stokes equations [53] (Yi et al. 2025), excluding the SGS stress term. Previous studies demonstrated that the SGS stress contribution to the resolved pressure field is small to negligible for these high-Reynolds-number turbulent statistics [3, 50, 24]. The boundary conditions for the pressure solver are zero gradient at the surface and zero pressure at the upper boundary [24]. For each case, the resolved pressure was computed from 60 instantaneous flow volumes evenly spaced across the simulation duration. The resolved pressure trends align with previous literature, and corresponding pressure statistics closely match with indirect estimates from the residual of the variance budgets described below.

Variance budget equations

The subsequent analysis evaluates the variance budget equations to quantify the production, transport, and redistribution of turbulent energy across the CBL under varying convection. These equations and their notation are introduced here to facilitate this assessment. Upon Reynolds decomposition and averaging of the Navier–Stokes equations, the variance budget equations are written as [75, 20]:

∂⟨u′2⟩∂t=−2⟨u′w′⟩∂U∂z−∂⟨u′2w′⟩∂z+2f⟨u′v′⟩+2⟨p′∂u′∂x⟩−εu(3)\frac{\partial\langle u'^{2}\rangle}{\partial t} = -2\langle u'w'\rangle\frac{\partial U}{\partial z} - \frac{\partial\langle u'^{2}w'\rangle}{\partial z} + 2f\langle u'v'\rangle + 2\left\langle p'\frac{\partial u'}{\partial x}\right\rangle -\varepsilon_u \tag*{(3)}
∂⟨v′2⟩∂t=−2⟨v′w′⟩∂V∂z−∂⟨v′2w′⟩∂z−2f⟨u′v′⟩+2⟨p′∂v′∂y⟩−εv(4)\frac{\partial\langle v'^{2}\rangle}{\partial t} = -2\langle v'w'\rangle\frac{\partial V}{\partial z} - \frac{\partial\langle v'^{2}w'\rangle}{\partial z} - 2f\langle u'v'\rangle + 2\left\langle p'\frac{\partial v'}{\partial y}\right\rangle -\varepsilon_v \tag*{(4)}
∂⟨w′2⟩∂t=2gθ0⟨w′θ′⟩−∂⟨w′3⟩∂z−2∂⟨w′p′⟩∂z+2⟨p′∂w′∂z⟩−εw(5)\frac{\partial\langle w'^{2}\rangle}{\partial t} = 2\frac{g}{\theta_0}\langle w'\theta'\rangle - \frac{\partial\langle w'^{3}\rangle}{\partial z} - 2\frac{\partial\langle w'p'\rangle}{\partial z} + 2\left\langle p'\frac{\partial w'}{\partial z}\right\rangle -\varepsilon_w \tag*{(5)}

Here, (⋅)′(\cdot)' denotes the fluctuating component of a given variable, U=⟨u⟩U = \langle u\rangle and V=⟨v⟩V = \langle v\rangle denote the mean streamwise and spanwise velocities, respectively, and pp is the resolved pressure computed offline as described above. The vertical velocity variance incorporates buoyancy effects through the Boussinesq approximation, consistent with the LES solver.

The factors analysed herein are the production of energy by shear (Pu\mathcal{P}_u and Pv\mathcal{P}_v), buoyancy (Bw\mathcal{B}_w), vertical transport by turbulence (Ti\mathcal{T}_i) and pressure (Πw\Pi_w), pressure–strain correlation or redistribution by pressure (Ri\mathcal{R}_i), and SGS dissipation of turbulent energy (εi\varepsilon_i). In the LES solver, the total TKE dissipation rate ε=εu+εv+εw=−⟨τijsgsS~ij⟩\varepsilon= \varepsilon_u + \varepsilon_v + \varepsilon_w = -\langle\tau_{ij}^{\mathrm{sgs}}\widetilde{S}_{ij}\rangle is determined directly from the LASD Smagorinsky model, where τijsgs\tau_{ij}^{\mathrm{sgs}} and S~ij\widetilde{S}_{ij} are the deviatoric SGS stress tensor and resolved strain-rate tensor respectively. In later results, dissipation is presented as the total ε\varepsilon.

Besides the dissipation, each term is computed from the resolved LES variables with the exceptions noted here. The computed production terms are based on the total fluxes including the SGS contribution. Similarly, the transport T\mathcal{T} includes the correlation between the resolved velocity and SGS stresses [3, 50]. The redistribution R\mathcal{R} acts to transfer energy between velocity components and for incompressible flow has a zero-trace pressure–strain tensor (Ru+Rv+Rw=0\mathcal{R}_u+\mathcal{R}_v+\mathcal{R}_w=0), such that any energy removed from one variance is injected into the other two [64]. The storage term on the left-hand side of (2.1)–(2.3) is small but non-zero in the present simulations owing to the slow growth of the CBL. The redistribution of energy between u′u' and v′v' by Coriolis (±2f⟨u′v′⟩\pm2f\langle u'v'\rangle) is also negligible relative to the other terms and is not shown in the later results.

Variance statistics

Convective similarity

The variance statistics are first evaluated relative to the convective velocity w⋆w_\star to identify overall trends including adherence to traditional free convection scaling. Figure 4 presents profiles of the total velocity variances and TKE across the thirteen LES cases, where each profile includes the resolved and estimated SGS component as described in § 2.3. A comparison of Figure 4(a) and Figure 4(b) reveals that u′u' is the dominant velocity component near the surface under weak convection for case c1. However, v′v' consistently exceeds u′u' throughout most of the CBL under moderately to strongly convective conditions, with v′>u′v' > u' most evident for cases c4–c10 spanning the transition from rolls to cells. The spanwise variance also increases more than u′u' in the upper half of the CBL, corresponding to the top of the counter-rotating thermal rolls. As buoyancy becomes increasingly dominant, both horizontal velocity variances decrease across all heights to the same value of 0.18 for case c13, which is approximately half the Bu3=0.35B_u^3 = 0.35 value in Table 2. Both variances continuously decrease without collapsing onto a common curve, indicating that a universal profile with mixed-layer scaling w⋆w_\star is only approached asymptotically for the horizontal velocities under strong convection.

Profiles of the total resolved plus SGS turbulent energy

Figure 4. Profiles of the total (resolved + SGS) turbulent energy: (a) streamwise variance u′2⋆u^{\prime2\star}, (b) spanwise variance v′2⋆v^{\prime2\star}, (c) vertical velocity variance w′2⋆w^{\prime2\star}, and (d) turbulent kinetic energy (TKE⋆\mathrm{TKE}^{\star}). The dashed grey curve in (c) shows the empirical parametrisation of Lenschow et al. [45] from table 1.

In contrast, the normalised vertical velocity in Figure 4(c) exhibits limited variability across the parameter space and is in generally good agreement with the empirical mixed-layer parametrisation proposed by Lenschow et al. [45] throughout much of the mixed layer, particularly approaching free convection. The largest deviations occur in the upper half of the CBL where capping inversion and entrainment effects may be relevant. Despite the better collapse, there is a systematic trend in which the variance peak initially decreases under weakly convective conditions before increasing with instability, accompanied by a downward shift in its vertical location. The initial decrease in the w′2⋆w'^{2\star} peak occurs while shear-generated turbulence is still dynamically important, whereas the subsequent increase begins only after shear production has become relatively weak. A similar qualitative trend has been observed in previous simulations [92], although its dynamical origin has not been explicitly examined.

The corresponding normalised total turbulent kinetic energy TKE⋆=12(u′2⋆+v′2⋆+w′2⋆)\mathrm{TKE}^\star= \frac{1}{2}(u'^{2\star} + v'^{2\star} + w'^{2\star}) shown in Figure 4(d) provides a useful measure of the combined response of the three velocity components to increasing convection. Following an initial decrease under weak convection, TKE⋆\mathrm{TKE}^\star becomes approximately invariant near the bottom of the mixed layer, around z⋆≈0.4z^\star\approx0.4, beginning with cases c6–c7. This behaviour coincides with the transition to predominantly cellular structures at −zi/L≈50-z_i/L \approx50 discussed previously. It is also the point at which the mixed-layer TKE first becomes comparable to, and locally exceeds, the near-surface value. In these conditions, shear production at the surface has already been substantially reduced and buoyant plumes within the mixed layer become the dominant energy-containing motions. The energy TKE⋆\mathrm{TKE}^\star continues to decrease in the lower and upper limits of the CBL until free convection is reached. This decrease is primarily associated with the weakening of the horizontal velocity fluctuations in Figure 4(a,Figure 4b) as the roll and cell structures progressively lose their horizontal circulations at the top and bottom of the CBL. The invariance in the mixed layer indicates that the reduction in u′u' and v′v' are offset by the increased w′w' at these heights.

Taken together, the results in Figure 4 indicate that mixed-layer scaling with w⋆w_\star captures the leading-order magnitude of the variances and TKE under strongly convective conditions, but does not completely describe the evolution of the profiles with changing convection. The remaining variability is attributed to the changing organisation of the largest turbulent structures and its impact on the distribution of turbulent energy, motivating the later examination of variance budget contributions.

The non-monotonic w′2⋆w^{\prime2\star} peak and higher spanwise variance v′>u′v^\prime> u^\prime in Figure 4 can be further assessed by summing the variances across the full CBL depth using integration. Depth integration removes the detailed vertical structure while retaining the total contribution of each velocity component to the boundary-layer turbulent energy. The integral is evaluated from the lowest valid grid level above the surface up to the top of the domain to include any residual turbulence in the entrainment layer above ziz_i. Figure 5 shows the depth-integrated variances and TKE, where the integral notation in the legend is shorthand for ∫0∞u′2⋆ dz⋆\int_{0}^{\infty} u^{\prime2\star}\,\mathrm{d}z^\star.

Variance and TKE statistics integrated across the CBL depth

Figure 5. Variance and TKE statistics integrated across the CBL depth and shown as a function of global instability −zi/L-z_i/L. The dashed line at −zi/L=50-z_i/L=50 indicates the approximate point where cells become dominant over rolls.

The integrated vertical variance ∫w′2⋆\int w^{\prime2\star} exhibits the same trends as the peak in Figure 4: while the value varies weakly across four orders of magnitude in −zi/L-z_i/L, there is an initial decrease followed by a gradual increase approaching the free-convection limit. Figure 5 reveals that the minimum coincides with the end of the transition from rolls to cells (dashed line), indicating a connection between the change in large-scale structure and the evolving magnitude of w′2⋆w^{\prime2\star} which is further explored in § 4.

The comparison of horizontal velocities in Figure 4 is more apparent from the integrated statistics in Figure 5, where ∫v′2⋆\int v^{\prime2\star} exceeds ∫u′2⋆\int u^{\prime2\star} from approximately case c2 onward. The difference decreases progressively until a common value is reached under near-free convection conditions. Further, v′v^\prime is the most energetic velocity component across the CBL for cases c2–c4, despite the absence of turbulence production along v′v^\prime.

The integrated TKE decreases substantially across the longitudinal roll regime as shear-generated turbulence diminishes. The TKE then decreases weakly in the cell regime, corresponding to the reduction of u′u^\prime and v′v^\prime at the CBL top and bottom observed in Figure 4. These depth-integrated results therefore reinforce the observations from the Figure 4 profiles: the transition to cellular convection changes how turbulent energy is distributed among the velocity components. The nature of these changes is evaluated further in § 4 using the variance budgets.

Surface layer similarity

Section 3.1 presented statistics across the CBL with respect to the convective scales. Here, the focus is on the transition between u∗u_\ast and w⋆w_\star scaling within the surface (log) layer. The surface-layer variances are first examined using MOST in Figure 6, where the competing effects of surface shear and buoyancy are expressed through u∗u_\ast and the Monin–Obukhov stability parameter ζ\zeta. The height range shown for each case extends from the SGS cutoff height defined in Appendix B to an upper bound of 0.1zi0.1z_i.

Comparison of surface layer variances with Monin–Obukhov similarity

Figure 6. Comparison of surface layer variances with Monin–Obukhov similarity: (a,b,c) streamwise, spanwise, and vertical velocities, respectively, normalised by u∗u_\ast and shown as a function of −z/⟨L⟩-z/\langle L\rangle based on the surface Obukhov length in (1); (d) same as (c), but with u∗(z)u_\ast(z) and the local Obukhov length Λ\Lambda based on local-in-height momentum and heat fluxes. Empirical MOST formulations from Table 1 and Table 2 are included for comparison ([60, 85]). The Panofsky et al. [60] points in (a,b) are for 0.1zi0.1z_i, which is the highest position included for each profile.

From case to case, the streamwise and spanwise velocity variances in Figure 6(a) and Figure 6(b) increase with convection when normalised by u∗u_\ast. Both velocities align with a 2/3 power law across cases under strong convection, consistent with the w⋆2∼u∗2(−zi/L)2/3w_\star^2 \sim u_\ast^2(-z_i/L)^{2/3} scaling. The v′v' profiles in Figure 6(b) exhibit this power law even for weak convection, indicating that the near-surface spanwise variance in this regime is dominated by convective-scaled roll motions that span the CBL [36]. The Panofsky et al. [60] relation aligns well with the w⋆w_\star-scaled regime in strong convection, but does not accurately capture the transition from u∗u_\ast in weak convection.

As expected, similarity in Figure 6(a,Figure 6b) using MOST is prevented by the height dependence within each profile. The Wilson [85] empirical fit captures the height dependence reasonably well for cases c7–c13 in the presence of cells, but underestimates the height dependence for cases c1–c6 with rolls. Thus, despite qualitatively capturing the trends for strong convection, an alternative formulation to MOST and the empirical fits is required to account for horizontal variance statistics in weak convection dominated by longitudinal rolls.

In contrast, the vertical velocity variance in Figure 6(c) yields a reasonable collapse of the profiles across the parameter space, indicating that MOST captures the transition in ⟨w′2⟩\langle w^{\prime2}\rangle from shear to buoyancy-dominated scaling. The profiles are also qualitatively consistent with the empirical fits reported in the literature for the convective surface layer and mixed layer, including the approach to free convection [60, 85, 88].

Although the adherence of the vertical velocity variance to MOST is well established, uncertainty remains in the precise formulation and fitted constants, particularly under weak convection. For example, Högström [27] found that the formulation of Panofsky et al. [60] overestimated normalised vertical velocity fluctuations by approximately 25% relative to reanalysed field measurements. Additionally, both previous LES and the present profiles exhibit lower variance values than the literature fits under weak and moderate convection [47], with the weak-convection values falling below Bw≈1.45B_w \approx1.45–1.91.9 predicted for neutral boundary layers (see, e.g., Table 1). Owing to uncertainty in the total (resolved + SGS) variances, no physical conclusions are drawn here regarding these lower values.

The collapse of ⟨w′2⟩\langle w^{\prime2}\rangle improves further when local scaling is used [56], as shown in Figure 6(d). The local scaling replaces the surface fluxes with local-in-height fluxes through the local friction velocity, u∗(z)=[⟨u′w′⟩(z)2+⟨v′w′⟩(z)2]1/4u_\ast(z) = [\langle u^\prime w^\prime\rangle(z)^2+\langle v^\prime w^\prime\rangle(z)^2]^{1/4}, and the local Obukhov length, Λ(z)=−u∗(z)3θ0/[κg⟨w′θ′⟩(z)]\Lambda(z) = -u_\ast(z)^3\theta_0/[\kappa g\langle w^\prime\theta^\prime\rangle(z)], noting the magnitude of the ⟨v′w′⟩\langle v^\prime w^\prime\rangle contribution to u∗(z)u_\ast(z) remains small (≤4%\leq4\%) compared to ⟨u′w′⟩\langle u^\prime w^\prime\rangle in the rotated coordinate system. The improved collapse obtained using −z/⟨Λ⟩-z/\langle\Lambda\rangle indicates that the height-dependent local momentum and heat fluxes provide a more complete similarity of ⟨w′2⟩\langle w^{\prime2}\rangle, consistent with findings for neutral boundary layers [26].

The results in Figure 6 therefore show that MOST captures the qualitative scaling trends across varying zi/Lz_i/L for all three velocity variance components, but complete similarity is precluded by underlying height dependencies. The collapse of profiles in Figure 6(d) is due to w′w^\prime being height-invariant in a neutral surface layer, such that the near-surface height dependence in Figure 4(c) is solely due to buoyancy effects that can be parameterised by −z/L-z/L. In the next section, the logarithmic profile in (2) is used to separate the convection trends from the height dependence observed for the streamwise variance in Figure 6(a).

Horizontal variances

Banerjee et al. [7] predicted a logarithmic decay for the streamwise variance in weak convection, consistent with the attached eddy framework. Figure 7(a) shows that the LES u′2u^{\prime2} profiles follow this logarithmic shape for cases c1–c6 with longitudinal rolls. Fits to (2) (dotted lines) are evaluated over the same case-dependent height range as Figure 6, extending from the SGS cutoff height defined in Appendix B to an upper bound of 0.1zi0.1z_i.

Streamwise velocity variance profiles and fitted logarithmic parameters

Figure 7. The streamwise velocity variance for weak convection fitted to ⟨u′2/u∗2⟩=Bu−Aulog⁡(z/0.4zi)\langle u^{\prime2}/u_*^{2}\rangle= B_u - A_u \log(z/0.4z_i) within the surface layer. (a,b) Profiles relative to u∗u_* and w∗w_*, respectively, where the dotted lines represent surface layer fits. The inset in (b) evaluates the surface layer fit for the free-convection limit (case c13) to demonstrate the deviation from a strictly logarithmic profile. (c,d) The fitted logarithmic slope AuA_u and intercept BuB_u, respectively, as functions of global instability. The grey markers in (c,d) indicate the cases deviating from the logarithmic variance profile.

A distinct logarithmic profile is observed within the surface layer for these cases, with limited change in the slope across cases but varying offset BuB_u corresponding to a shift along the abscissa. The u′u^\prime profiles diverge from a strictly logarithmic decay beginning with case c7 and thereafter more closely resemble the shape of the v′v^\prime profiles and the Wilson [85] height dependence. The inset in Figure 7(b) evaluates the surface layer fit for the free-convection limit (case c13), visually demonstrating the departure from the logarithmic formulation under stronger convection.

The logarithmic fits use 0.4zi0.4z_i as the reference height rather than ziz_i. This choice reflects the characteristic depth over which shear turbulence remains significant, noting the mixed layer is characterised by zero mean shear and constant variance (Figure 4). The bottom of the mixed layer occurs near 0.4zi0.4z_i ([25]), although the height at which the mean shear approaches zero varies between approximately 0.3zi0.3z_i and 0.55zi0.55z_i among the cases. Using a representative value of 0.4zi0.4z_i therefore provides a common reference for the shear-dominated region while preserving a simple two-parameter logarithmic formulation.

Figure 7(b) shows that the fitted profiles converge towards a common variance level near this height when expressed using w⋆w_\star, providing additional support for its interpretation as the relevant reference level for shear-generated turbulence in the CBL.

The fitted values of AuA_u in Figure 7(c) indicate that in the presence of rolls, AuA_u remains approximately constant to first-order accuracy. The constant value indicates that mechanical shear and wall-attached structures remain the dominant organising process for the u′(z)u^\prime(z) height dependence. Once the stability parameter exceeds −zi/L≈50-z_i/L \approx50, the slope departs sharply from this first-order approximation to Au∼(−zi/L)2/3A_u \sim(-z_i/L)^{2/3}, coinciding with a shift from u∗u_\ast to w⋆w_\star in governing the shape of u′(z)u^\prime(z).

The fitted slopes Au≈1.36A_u \approx1.36–1.81.8 in the roll regime are somewhat larger than values reported for canonical neutral boundary layers ([49, 6, 67]). One possible explanation is the finite horizontal extent of the present LES domain, which limits the longest resolved turbulent wavelengths. Such a restriction preferentially removes very-large-scale variance, with the effect becoming more important at increasing height as the characteristic eddy size grows, and may consequently produce a steeper apparent logarithmic decay. Figure 7(c) should therefore be interpreted primarily in terms of the qualitative trends with convection rather than a precise comparison with canonical AuA_u values.

The fitted BuB_u values in Figure 7(d) follow Bu≈0.2(−zi/L)2/3=0.2w⋆2/u⋆2B_u \approx0.2(-z_i/L)^{2/3} = 0.2w_\star^2/u_\star^2, indicating that the variance offset is predominantly governed by the convective velocity. The constant offset with respect to w⋆w_\star ensures the logarithmic profile approaches a finite value independent of u⋆u_\star in the mixed layer where shear is negligible. The confinement of the buoyancy correction to the coefficient BuB_u in the roll regime indicates that the impact of convection on near-surface u′u^\prime is non-local and approximately independent of height.

These results are qualitatively consistent with the extended attached-eddy formulation of Banerjee et al. [7], in which the logarithmic slope AuA_u remains approximately constant relative to u⋆u_\star under weak convection while the intercept BuB_u varies with stability. In the present LES, however, the approximately constant AuA_u persists to stronger convection up to −zi/L≈50-z_i/L \approx50 and connects this range with the transition from rolls to cells. The transition point in AuA_u corresponds to a local stability of −ζ≈5-\zeta\approx5 at z⋆=0.1z^\star= 0.1, compared to the predicted threshold of −ζ<0.5-\zeta< 0.5 and z⋆<0.02z^\star< 0.02 in Banerjee et al. [7]. Additionally, the fitted intercept is seen here to depend on the global convection and −zi/L-z_i/L rather than the local −ζ-\zeta, suggesting that the outer mixed-layer dynamics exert an important influence on the surface-layer horizontal variance. The different BuB_u dependence may also help to explain observed quantitative discrepancies between the Banerjee et al. [7] predictions and other convective simulations ([90]).

Fitting the ⟨u′2⟩\langle u^{\prime2}\rangle profiles to the mean deficit formulation in Table 2 yields similar values of Au2A_{u2} and Bu2B_{u2} as Yang et al. [90]. However, the profiles adhere more closely to a logarithmic decay than to the shape of the mean deficit, where the former leads to higher goodness-of-fit values. This difference is physically relevant because the logarithmic variance in Figure 7(a) and linear relation between ⟨u′2⟩\langle u^{\prime2}\rangle and U(z)U(z) in the mean-deficit formula implies the mean velocity profile is also logarithmic, but U(z)U(z) departs substantially from the log law in the CBL, which corresponds to a height-dependent non-dimensional gradient ϕm\phi_m (see, e.g., Heisel & Chamecki 2024). While the mean deficit model operates well as a first-order approximation and yields consistent fitted constants, it requires further adjustments to account for the difference in the height dependence of ⟨u′2⟩\langle u^{\prime2}\rangle and UU.

Under strong convection, the streamwise variance departs from the logarithmic profile evaluated in Figure 7 and appears to align more closely with the height dependence proposed by Wilson [85] shown in Figure 6(a). To assess this trend, Figure 8 compares the streamwise and spanwise variance profiles with ⟨ui′2/u⋆2⟩=Ci[1−(z⋆)0.38]\langle u_i^{\prime2}/u_\star^2\rangle= C_i[1 - (z^\star)^{0.38}] representing the height dependence from the Wilson [85] formulation in Table 2. The surface layer spanwise variance in Figure 8(b) is well described by the fitted height dependence across all cases, whereas the streamwise variance in Figure 8(a) only follows this dependence for cases c6 onward representing strong convection. The poor alignment with cases c1–c5 is more readily apparent from Figure 6(a). The fact that case c6 approximately follows both the logarithmic decay in Figure 7(a) and the power law in Figure 8(a) indicates a relatively smooth transition between the two dependencies. The consistency of the spanwise profile shape across cases and the transition in streamwise variance from (2) to the same dependence as v′v' coincides with the evolving large-scale structure: rolls predominantly impact v′v' while u′u' retains its shear-driven dependencies, and the more geometrically isotropic cells lead to greater similarity in u′u' and v′v'. It is expected that the spanwise variance also becomes logarithmic under near-neutral conditions in the absence of rolls, but this cannot be assessed from the present cases.

Horizontal velocity variances for strong convection

Figure 8. The horizontal velocity variances for strong convection fitted to ⟨ui′2/u∗2⟩=Ci[1−(z⋆)0.38]\langle u_i^{\prime2}/u_*^2\rangle= C_i[1 - (z^\star)^{0.38}] [85] within the surface layer. (a,b) Profiles of the normalised streamwise and spanwise velocity variances, respectively, where the dotted lines represent surface layer fits. (c,d) Fitted coefficients CuC_u and CvC_v, respectively, as functions of global instability. In (c,d), dashed lines represent empirical fits and grey markers identify the weak-convection streamwise cases whose profiles follow a logarithmic trend.

The fitted coefficients in Figure 8(c,Figure 8d) quantify how the variance levels change with global instability. The approximate invariance of cases c1–c5 in Figure 8(c) is consistent with the u∗u_* scaling of AuA_u in Figure 7(c). The dashed curves in Figure 8(c,Figure 8d) represent fits to the same empirical global instability dependence as Wilson [85] in Table 2. The fits lead to the horizontal variance profiles

⟨u′2⟩u∗2=[0.5+0.82(−⟨zi⟩L)2/3][1−(z∗)0.38],(6)\frac{\langle u^{\prime2}\rangle}{u_*^2} = \left[0.5 + 0.82\left(-\frac{\langle z_i\rangle}{L}\right)^{2/3}\right]\left[1-\left(z^*\right)^{0.38}\right], \tag*{(6)}
⟨v′2⟩u∗2=[3.5+0.89(−⟨zi⟩L)2/3][1−(z∗)0.38],(7)\frac{\langle v^{\prime2}\rangle}{u_*^2} = \left[3.5 + 0.89\left(-\frac{\langle z_i\rangle}{L}\right)^{2/3}\right]\left[1-\left(z^*\right)^{0.38}\right], \tag*{(7)}

where the exponents are imposed to match Wilson [85] but the coefficients differ. In particular, the 2/3 exponent in (6) and (7) is consistent with the scaling w∗2/u∗2∝(−zi/L)2/3w_*^2/u_*^2 \propto(-z_i/L)^{2/3}. The fitted trends are therefore consistent with the transition from surface shear to convective scaling with increasing buoyancy. The difference in offset values between CuC_u and CvC_v is required to account for the trends observed in Figure 4, where the spanwise variance is greater in weak convection before the horizontal variances converge to the same value in free convection. A matching coefficient for the −zi/L-z_i/L term in CuC_u and CvC_v is expected to ensure this convergence, noting that the difference in the fitted values is within their combined 95% statistical uncertainty ranges (0.82±0.060.82 \pm0.06 for u′u' and 0.89±0.060.89 \pm0.06 for v′v').

Variance budget statistics

The preceding analyses identified statistical changes in the velocity variances from weak to free convection. The enhanced v′v' and the non-monotonic evolution of w′w' are suggestive of underlying trends in the production, transport, and redistribution of turbulent energy across the roll-to-cell transition. Therefore, the variance budget equations introduced in § 2.5 are used here to assess the energy transfer pathways with changing convection.

Pressure redistribution and return to isotropy

The pressure redistribution Ri\mathcal{R}_i is the only mechanism in (3)–(5) for energy exchange across the velocity components. The redistribution profiles are shown in Figure 9(a–Figure 9c), with the star superscript indicating local-in-time normalisation by w∗3(t)/zi(t)w_*^3(t)/z_i(t) (see, e.g., § 2.2). In the surface layer, shear production of u′u' is redistributed to both v′v' and w′w' for the weakest convection case c1. For the remaining cases within the roll regime, v′v' receives net energy from both u′u' and w′w'. Energy distribution is from w′w' to u′u' and v′v' in the surface layer under stronger convection and for all cases within the mixed layer.

Profiles of pressure redistribution for streamwise, spanwise, and vertical velocities

Figure 9. Profiles of the pressure-redistribution for the streamwise (Ru\mathcal{R}_u, left), spanwise (Rv\mathcal{R}_v, middle), and vertical (Rw\mathcal{R}_w, right) velocities. Rows correspond to decomposition of the pressure field: (a–c) total redistribution Rtotal\mathcal{R}_{\mathrm{total}}; (d–f) redistribution Rbuoyancy\mathcal{R}_{\mathrm{buoyancy}} by pressure due to buoyancy; (g–i) redistribution Radvection\mathcal{R}_{\mathrm{advection}} by pressure due to nonlinear advection.

The pressure field can be decomposed into contributions from individual terms in the Navier–Stokes equations by solving variations of the pressure Poisson equation based on each term ([53, 24, 18] (Yi et al. 2025)). Figure 9(d–Figure 9f) shows Ri,buoyancy\mathcal{R}_{i,\mathrm{buoyancy}} due to pressure fluctuations associated with buoyancy arising from the gravity force term and the Boussinesq approximation. Similarly, Figure 9(g–Figure 9i) shows Ri,advection\mathcal{R}_{i,\mathrm{advection}} due to the pressure field from the nonlinear advection term. Advection can be further decomposed into shear and turbulent-turbulent components [24], but the turbulent-turbulent contribution was found to be dominant with no unique convection trends revealed by the additional decomposition. Pressure statistics due to other terms such as Coriolis are negligibly small and not shown here.

The buoyancy contribution to the vertical component, Rw,buoyancy⋆\mathcal{R}^{\star}_{w,\mathrm{buoyancy}} in Figure 9(f), transfers buoyancy-generated energy away from w′w' toward the horizontal velocity components for all heights and cases. Its profiles collapse across the strongly convective cases when normalised by w∗w_*, with all profiles exhibiting a similar vertical distribution. The minimum value in the mixed layer and local peaks at the top and bottom of the CBL correspond to the large-scale rolls and cells, in which the vertical motions in the mid-CBL connect to horizontal displacements at the boundaries. The horizontal redistributions, Ru,buoyancy⋆\mathcal{R}^{\star}_{u,\mathrm{buoyancy}} and Rv,buoyancy⋆\mathcal{R}^{\star}_{v,\mathrm{buoyancy}} in Figure 9(d,Figure 9e), have the same trend as w′w^\prime but with positive Ri\mathcal{R}_i. The strong near-surface peaks in Figure 9(d–Figure 9f) reflect the kinematic wall-blocking pressure response enforced by the wall boundary condition ([21, 32] (Chávez-Medina et al. 2025)). Under weak convection, most of the buoyancy-related redistribution at the top and bottom of the CBL goes to Rv,buoyancy⋆\mathcal{R}^{\star}_{v,\mathrm{buoyancy}} rather than Ru,buoyancy⋆\mathcal{R}^{\star}_{u,\mathrm{buoyancy}},

consistent with the circulations of the longitudinal rolls. The difference between the two progressively decreases across the rest of the parameter space until the redistribution to u′u' and v′v' is equal for cells under free convection.

The advection contribution exhibits substantially greater variation with height and convection. The redistribution of shear-generated turbulence from Ru,advection∗R^{\ast}_{u,\mathrm{advection}} in Figure 9(g) to Rv,advection∗R^{\ast}_{v,\mathrm{advection}} and Rw,advection∗R^{\ast}_{w,\mathrm{advection}} in Figure 9(h,Figure 9i) decreases with diminishing shear, and the profiles become approximately invariant for cases c7–c13. The non-zero profiles in these cases indicate an indirect contribution of buoyancy through the turbulent-turbulent advection term, such that the trends in Figure 9(g–Figure 9i) do not solely reflect shear effects.

To further evaluate the trends with varying convection, Figure 10 shows the depth-integrated redistribution terms in the same manner as the variances in Figure 5. Also included are the production and dissipation budget terms in Figure 10(a). The integrated statistics contain uncertainty associated with the estimation of the SGS contributions and with the absence of information between the wall and the first grid point in the wall-modelled LES. Accordingly, the interpretation of Figure 10 emphasises the trends across the parameter space rather than the absolute magnitudes of the integrated quantities.

Variance budget statistics integrated across the CBL depth

Figure 10. Variance budget statistics integrated across the CBL depth. (a) Production of turbulence by shear (∫Pu∗\int P_u^{\ast} and ∫Pv∗\int P_v^{\ast}) and buoyancy (∫Bw∗\int B_w^{\ast}) as well as dissipation (∫ε∗\int\varepsilon^{\ast}) as functions of global instability −zi/L-z_i/L. (b) Redistribution of ∫Ri∗\int R_i^{\ast} across velocity components by pressure–strain correlations versus −zi/L-z_i/L; (c) Redistribution relative to the total rate of production versus the fraction of production due to buoyancy. The vertical dashed line across all panels denotes the approximate end of the transition from roll- to cell-dominated convection at −zi/L≈50-z_i/L \approx50. The coloured dashed lines in (c) represent the (9) return-to-isotropy predictions for RiR_i leading to isotropic dissipation.

The productions in Figure 10(a) show a clear transition in the dominant source of turbulent energy. The integrated streamwise shear production, ∫Pu⋆\int\mathcal{P}_{u}^{\star}, decreases rapidly with increasing instability and becomes small relative to buoyancy by the end of the transition from rolls to cells. The spanwise shear production ∫Pv⋆\int\mathcal{P}_{v}^{\star} remains close to zero throughout the parameter range, with the small values under weak convection corresponding to production in the entrainment layer where there is strong wind shear and veering as seen in Figure 1. The integrated buoyancy production, ∫Bw⋆\int\mathcal{B}_{w}^{\star}, remains relatively constant across cases. By definition, the normalised production Bwzi/w⋆3\mathcal{B}_{w} z_i/w_{\star}^{3} reduces to the heat flux profile ⟨w′θ′⟩/Q\langle w'\theta'\rangle/Q, see, e.g., Figure 3(b), which is invariant with convection aside from small changes in the negative flux at ziz_i. Buoyancy production becomes the dominant source of TKE as shear diminishes, with more than 90% of the total integrated production due to buoyancy at −zi/L=50-z_i/L = 50. The integrated dissipation decreases with the decreasing shear and approximately balances the buoyancy production under the strongly convective cases, consistent with the quasi-steady state of the flow.

The evolution of the pressure-redistribution terms shown in Figure 10(b) is consistent with the trends observed in the Figure 9 profiles. In the weakly convective roll regime, there is a net transfer of energy away from the streamwise component ∫Ru⋆\int\mathcal{R}_{u}^{\star} to the spanwise component ∫Rv⋆\int\mathcal{R}_{v}^{\star} and to a lesser extent the vertical component ∫Rw⋆\int\mathcal{R}_{w}^{\star}. Energy redistribution away from w′w' progressively increases with convection until reaching an approximate constant in the cell regime. Given the contribution of buoyancy to Rw⋆\mathcal{R}_{w}^{\star} is fixed relative to the convective scales (Figure 9(f)), this trend is due to a reduction in redistribution from u′u' to w′w' as surface shear diminishes to negligible values (Figure 9(g,Figure 9i) and Figure 10(a)). The horizontal redistributions ∫Ru⋆\int\mathcal{R}_{u}^{\star} and ∫Rv⋆\int\mathcal{R}_{v}^{\star} converge to similar values in the same manner as Figure 9(d,Figure 9e), corresponding to the progressive loss of horizontal alignment in the roll structures as they transition toward isotropic cellular plumes.

In addition to its connection to the large-scale structure of the CBL, pressure redistribution is often considered for its role in isotropisation of the flow (Bou-Zeid et al. [10]), noting a common feature of pressure closure models is a return-to-isotropy component (Pope [64]). In the present context, isotropy refers to the dissipation tensor and its components in ((2.1)–(2.3)). The vertical transport terms in these budget equations are zero upon integration across the CBL depth, given that transport only acts to distribute existing energy across heights. The Coriolis contribution is negligible and the storage term associated with the slow growth of the CBL is small relative to the remaining terms, such that there is an approximate balance between production, redistribution, and dissipation in the integrated budgets:

0≈∫Pu+∫Ru−∫Eu,0≈∫Pv+∫Rv−∫Ev,0≈∫Bw+∫Rw−∫Ew.(8)\begin{aligned} 0 &\approx\int\mathcal{P}_{u} + \int\mathcal{R}_{u} - \int\mathcal{E}_{u}, \\ 0 &\approx\int\mathcal{P}_{v} + \int\mathcal{R}_{v} - \int\mathcal{E}_{v}, \\ 0 &\approx\int\mathcal{B}_{w} + \int\mathcal{R}_{w} - \int\mathcal{E}_{w}. \tag*{(8)} \end{aligned}

Same as before, the integral notation is shorthand for the integral across depth, e.g. ∫0∞Pu dz\int_0^\infty\mathcal{P}_u\,\mathrm{d}z. Dividing through by the total production and assuming the dissipation is isotropic and in balance with the total production leads to the following return-to-isotropy prediction for the redistribution:

∫Ru∫Pu+∫Pv+∫Bw≈13−∫Pu∫Pu+∫Pv+∫Bw,∫Rv∫Pu+∫Pv+∫Bw≈13−∫Pv∫Pu+∫Pv+∫Bw,∫Rw∫Pu+∫Pv+∫Bw≈13−∫Bw∫Pu+∫Pv+∫Bw.(9)\begin{aligned} \frac{\int\mathcal{R}_u}{\int\mathcal{P}_u+\int\mathcal{P}_v+\int\mathcal{B}_w} &\approx\frac{1}{3}-\frac{\int\mathcal{P}_u}{\int\mathcal{P}_u+\int\mathcal{P}_v+\int\mathcal{B}_w},\\ \frac{\int\mathcal{R}_v}{\int\mathcal{P}_u+\int\mathcal{P}_v+\int\mathcal{B}_w} &\approx\frac{1}{3}-\frac{\int\mathcal{P}_v}{\int\mathcal{P}_u+\int\mathcal{P}_v+\int\mathcal{B}_w},\\ \frac{\int\mathcal{R}_w}{\int\mathcal{P}_u+\int\mathcal{P}_v+\int\mathcal{B}_w} &\approx\frac{1}{3}-\frac{\int\mathcal{B}_w}{\int\mathcal{P}_u+\int\mathcal{P}_v+\int\mathcal{B}_w}. \tag*{(9)} \end{aligned}

These equations match Bou-Zeid et al. (2018) but with Bw\mathcal{B}_w included in the normalisation. The integrated LES redistributions are compared with these predictions in Figure 10(c) as the fraction of energy production due to buoyancy varies from 0 to 1, noting that the present work does not include fractions less than 0.2 which encompasses the near-neutral regime.

The redistributions in Figure 10(c) align relatively well with the (9) return-to-isotropy predictions, including the approach to +1/3 for each horizontal component and −2/3 for the vertical component as buoyancy becomes the sole source of TKE under free convection. The modest discrepancies from the prediction may in part be due to missing SGS contributions and statistics below the first grid point, which is particularly relevant for the surface production and redistribution under weak convection. Accordingly, the primary conclusion from Figure 10(c) is that the bulk redistribution of energy across the CBL is consistent with the isotropisation of ε\varepsilon to first-order accuracy.

Because ∫Rv\int\mathcal{R}_v does not exceed the return-to-isotropy prediction, the redistribution magnitude in Figure 10 cannot by itself explain how v′v' is the largest turbulent component under certain convective conditions as seen in Figure 5. In the following section, the spectrum of the variance budget is used to determine whether scale-wise trends can elucidate how the global isotropisation of dissipation leads to the observed variance statistics.

Spectral variance budgets and scale-dependent energy transfer

The variance budgets in (2.1)–(2.3) are transformed into the wavenumber domain to identify the characteristic horizontal scales and heights at which production, pressure transport, and pressure–strain redistribution occur (Lee & Moser 2019; Huang et al. (2026)). An additional scale-to-scale energy transfer term arises in the wavenumber domain [64]. This new term is not discussed here, except to note that the net scale-to-scale transfer is predominantly downscale across the sampled heights and convection states in the present simulations. The spectral analysis is used here to identify the dominant scales and pathways of variance production, transport, and redistribution.

The co-spectrum of each budget term was evaluated from two-dimensional (2D) Fourier transforms of the horizontal fields from 60 instantaneous 3D flow volumes distributed throughout each simulation. As an example, the spectral pressure-transport term Π^w∗\widehat{\Pi}_w^* is estimated from the co-spectrum of w′w' and p′p', with the vertical derivative applied to the resulting co-spectrum. The other spectral budget terms were obtained analogously from the velocity, pressure, and temperature co-spectra. Here, the hat notation (⋅)^\widehat{(\cdot)} denotes the spectral density of a given variable. The 2D spectral densities across wavenumbers kxk_x and kyk_y were translated to 1D functions by integrating the spectra azimuthally along fixed values of the horizontal wavenumber magnitude k=(kx2+ky2)1/2k = (k_x^2 + k_y^2)^{1/2}. Same as the previous results, the star (⋆\star) superscript denotes local-in-time normalisation by w⋆∗(t)w_\star^*(t) for the spectra and zi(t)z_i(t) for the wavenumber. The spectral densities were interpolated to common values of both height z/ziz/z_i and wavenumber magnitude kzikz_i prior to averaging across flow volumes.

Colour contours of the premultiplied variance budget spectral densities are shown in Figure 11 to visualise the scale-dependent pathways of turbulent energy production and redistribution. The changes are gradual within the cell regime, such that only the limits of this regime (c7 and c13) are shown here. Under weak convection within the roll regime, the spectral budgets identify two distinct pathways transferring energy to the spanwise velocity. Near the surface, streamwise variance produced by mean shear P^u\widehat{\mathcal{P}}_u across a range of scales is locally redistributed through pressure–strain interactions R^u\widehat{\mathcal{R}}_u, with a preferential contribution to the spanwise component R^v\widehat{\mathcal{R}}_v. Separately, buoyancy production B^w\widehat{\mathcal{B}}_w in the mixed layer generates vertical turbulence at moderately larger scales, which is subsequently transported downward by pressure Π^w\widehat{\Pi}_w and redistributed near the surface through R^w\widehat{\mathcal{R}}_w to R^v\widehat{\mathcal{R}}_v.

Premultiplied colour spectra of variance budget terms

Figure 11. Premultiplied colour spectra of variance budget terms as functions of the normalised wavenumber k∗=kzik^* = k z_i and height z∗z^*. Rows correspond to cases c1–c7 spanning the transition from rolls to cells and free convection case c13. Columns show the streamwise shear production, P^u∗\hat{P}_u^*, vertical buoyancy production, B^w∗\hat{B}_w^*, vertical pressure transport, Π^w∗\hat{\Pi}_w^*, and the pressure–strain redistribution terms R^u∗\hat{R}_u^*, R^v∗\hat{R}_v^*, and R^w∗\hat{R}_w^*.

This vertical transport and displacement from w′w' to v′v' corresponds to the large-scale roll motions, indicating the thermal rolls are mediated by pressure. Large eddies approaching the wall experience kinematic wall-blocking (Glendening [21]; Hunt & Morrison [32]), which suppresses wall-normal motion and requires a compensating horizontal response to satisfy continuity (Hunt & Morrison [32]; Nguyen et al. [55]; Jayaraman & Brasseur [33]). This process occurs through the wall-reflection pressure, which increases the vertical pressure gradient near the surface and redistributes part of the vertical turbulent energy into horizontal motions (Pope [64]; Ding et al. [18]; Chávez-Medina et al. (2025)). This redistribution from the vertical component is preferentially directed toward v′v' due to the alignment of the thermal rolls.

As the large-scale flow organisation restructures from rolls to cells, the spectral energy pathways also change. The surface shear production progressively diminishes across cases c1–c6, leading to negligible redistribution from R^u\widehat{\mathcal{R}}_u to R^v\widehat{\mathcal{R}}_v by −zi/L≈40-z_i/L \approx40–55. At the same time, the largest scales of buoyancy production B^w\widehat{\mathcal{B}}_w occur at moderately higher wavenumbers under stronger convection. This shift towards smaller scales is attributed to both a loss of streamwise elongation of the structures and a concurrent reduction in the characteristic lateral spacing of the cellular plumes relative to the rolls (Kaimal et al. [35]; Khanna & Brasseur [37]; Salesky & Anderson [68]). Pressure–strain redistribution R^w\widehat{\mathcal{R}}_w to the two horizontal components becomes increasingly equal with increasing convection and R^u\widehat{\mathcal{R}}_u and R^v\widehat{\mathcal{R}}_v appear to be the same under free convection. This convergence is consistent with Figure 10(b) and the loss of directional preference in the structures as shear reduces.

The spectra in Figure 11 further reveal that the large-scale, pressure-mediated coupling between the mixed layer and the surface weakens with increasing convection. Both the pressure transport Π^w\widehat{\Pi}_w connecting these layers and the near-surface redistribution R^w\widehat{\mathcal{R}}_w decrease with increasing convection across the parameter space. The results indicate that thermal rolls more efficiently transfer large-scale energy across heights and velocity components compared to cells. At the same time, redistribution from w′w' to the horizontal velocities increases across a range of smaller scales at all heights throughout the CBL, leading to a preservation of the net redistribution seen in Figure 9(c).

The changing magnitude of the buoyancy-driven energy pathways observed in Figure 11 are further assessed in Figure 12 which shows the scale-wise budget terms for all cases at two representative heights to evaluate the changes in low-wavenumber statistics. Figure 12(a–Figure 12e)

Spectra of the variance budget terms for near-surface and mixed-layer regions

Figure 12. Spectra of the variance budget terms as functions of the normalised wavenumber, k∗k^*, for the near-surface (z∗≈0.05z^* \approx0.05; a–e) and mixed-layer (z∗≈0.50z^* \approx0.50; f–j) regions. The rows show vertical buoyancy production, B^w∗\widehat{B}_w^*, vertical pressure transport, Π^w∗\widehat{\Pi}_w^*, and pressure–strain redistribution for the streamwise, spanwise, and vertical velocity variances, R^u∗\widehat{R}_u^*, R^v∗\widehat{R}_v^*, and R^w∗\widehat{R}_w^*, respectively.

shows the near-surface spectra (z∗≈0.05z^* \approx0.05), while Figure 12(f–Figure 12j) shows the corresponding mixed-layer spectra (z∗≈0.50z^* \approx0.50). The spike at k∗≈3k^* \approx3–44 for case c1 in Figure 12(i,Figure 12j) corresponds to a negative R^v∗\widehat{R}_v^* region in the upper half of the CBL seen in Figure 11, which is attributed to top-down effects related to entrainment layer production which is outside the scope of the present analysis.

Net buoyancy production is largest at the surface in Figure 12(a), but near-surface production is relatively broadband. There is a significantly higher narrowband concentration of large-scale B^w∗\widehat{B}_w^* in the mixed layer (Figure 12(f)) despite overall lower net production farther from the surface. For the weak convection cases, a significant portion of this large-scale production is transported by pressure from the mixed layer (Figure 12(g)) towards the surface (Figure 12(b)). Almost all of this energy is transferred to the horizontal velocity components (Figure 12(c,Figure 12d)) with preferential redistribution to v′v', noting Π^w⋆\widehat{\Pi}^{\star}_{w} in Figure 12(b) and R^w⋆\widehat{R}^{\star}_{w} in Figure 12(e) have approximately equal magnitudes in the low-wavenumber range. The redistribution of shear-generated turbulence from u′u' to v′v' and w′w' (Figure 12(d,Figure 12e)) is also apparent at smaller scales (k⋆≳10k^{\star} \gtrsim10) under weak convection. While this combined redistribution of both shear- and buoyancy-generated turbulence to v′v' does not exceed isotropy predictions, the redistribution mediated by pressure and the roll motions originates in the mixed layer and occurs at the largest scales in the flow.

Consistent with Figure 11, the concentrated large-scale production, transport, and redistribution peaks are progressively lost in Figure 12 as the rolls (c1) evolve into cells (c7 onward) with increasing convection. At the same time, the weakening peaks shift toward higher normalised wavenumber from k⋆≈2k^{\star} \approx2 to k⋆≈4k^{\star} \approx4–55 in Figure 12(b–Figure 12f). In the mixed layer, the weaker pressure transport toward the surface for cases c7–c13 in Figure 12(g) is compensated by increased pressure–strain redistribution in Figure 12(j). However, this increase is specifically at smaller scales (k⋆≳5k^{\star} \gtrsim5), and there is a deficit in R^w⋆\widehat{R}^{\star}_{w} relative to the roll cases at the scale of the production peaks (k⋆≈4k^{\star} \approx4–55 in Figure 12(f)). The deficit indicates that more large-scale buoyancy-generated energy is retained locally within w′w' with subsequent downscale transfer and redistribution at smaller scales, thus providing a possible explanation for the increase in the w′w' peak across the cell regime in Figure 4(c).

The downward displacement of the peak toward z⋆≈0.3z^{\star} \approx0.3 may be related to a coinciding shift in the symmetry of the pressure transport; the height with low-wavenumber Π^w⋆=0\widehat{\Pi}^{\star}_{w}=0 in Figure 11 moves downward with increasing convection in a manner consistent with the w′w' peak. However, this interpretation remains speculative and requires further investigation.

Figure 13 provides the 1D velocity variance spectral densities for the same heights as Figure 12 to facilitate discussion on how the energy pathways identified from Figure 11 and Figure 12 result in the variance trends observed in § 3. The 1D velocity spectra are denoted by u^⋆\widehat{u}^{\star}, v^⋆\widehat{v}^{\star}, and w^⋆\widehat{w}^{\star} for the streamwise, spanwise, and vertical components, respectively, where the star (⋆\star) superscript denotes local-in-time normalisation by w⋆2(t)zi(t)w_{\star}^{2}(t)z_{i}(t).

Velocity spectra as functions of the normalised wavenumber for near-surface and mixed-layer regions

Figure 13. Velocity spectra as functions of the normalised wavenumber, k⋆k^\star, for the near-surface (z⋆≈0.05z^\star\approx0.05; aa–cc) and mixed-layer (z⋆≈0.50z^\star\approx0.50; dd–ff) regions. The rows show the streamwise u^⋆\hat{u}^\star, spanwise v^⋆\hat{v}^\star, and vertical w^⋆\hat{w}^\star velocity spectra, respectively. The same ordinate range is plotted for each component at a given height, except for (cc) due to its substantially smaller magnitude.

Near the surface, the roll-dominated cases exhibit a distinct low-wavenumber peak in both horizontal components in Figure 13(a,Figure 13b) at k⋆≈2k^{\star} \approx2, the same wavenumber as the large-scale buoyancy production, pressure transport and redistribution peaks in Figure 12. The spanwise v^⋆\widehat{v}^{\star} low-wavenumber peak exceeds the streamwise u^⋆\widehat{u}^{\star} peak across the roll regime in cases c1–c6, consistent with the preferential redistribution toward v′v' identified from R^w⋆\widehat{R}^{\star}_{w} in Figure 12. While the overall redistribution ∫R\int\mathcal{R} approximately aligns with return-to-isotropy (Figure 10(c)), the energy redistributed to v′v' by roll motions is predominantly at the largest scales (Figure 12(d)), whereas dissipation ε\varepsilon is exclusively within subgrid scales in the present LES. This pathway allows for v′v' to accumulate more large-scale energy than the other velocity components, leading to larger variance statistics.

In the comparison of horizontal variances, the greater large-scale energy in v^⋆\widehat{v}^{\star} is partially offset by higher u^⋆\widehat{u}^{\star} at intermediate and smaller scales in cases c1–c6. While some of this shear-generated energy is redistributed from u′u' to v′v' (Figure 9(h) and Figure 12(d)), the greater small-scale energy in u^⋆\widehat{u}^{\star} reduces the difference between components in Figure 5. In contrast, the vertical velocity spectrum w^⋆\widehat{w}^{\star} near the surface in Figure 13(c) has smaller magnitudes centred at higher wavenumbers, consistent with the redistribution of large-scale buoyancy-generated energy into the horizontal components (Figure 12(e)).

The velocity spectra under stronger convection are also consistent with the previous figures: the spectral peaks shift to moderately higher wavenumbers as the large-scale structures lose their streamwise elongation and contract in lateral scale. Further, the horizontal velocity spectra become increasingly similar to each other at both heights, consistent with the increasingly symmetric pressure–strain redistribution between u′u' and v′v' seen in Figure 12. The increase in the w′w' peak with convection is less evident in Figure 13(f), although a modest increasing trend is apparent in the wavenumber range k⋆=5k^\star= 5–1010. One consideration is that the downward shift in the peak leads to a maximum variance at different heights across cases, such that the spectra at a fixed height may partially obscure spectral trends associated with the peak.

Summary

The present study assesses turbulent velocity variances in the CBL using thirteen wall-modelled LES cases summarised in Table 3. Each case exhibits a pronounced mixed layer (Figure 1), such that near-neutral conditions are not considered here. The LES spans the remainder of the sheared convection parameter space, including six cases within the transition from large-scale longitudinal rolls to polygonal cells ending around −zi/L≈50-z_i/L \approx50, followed by seven cases within the cell regime as the flow asymptotically approaches penetrative free convection (Figure 2).

This structural transition is accompanied by systematic changes in the velocity variances apparent from both the profiles (Figure 4) and depth-integrated statistics (Figure 5). The horizontal variances and TKE progressively decrease in magnitude relative to w⋆w_\star. The spanwise (cross-wind) variance exceeds the streamwise component for almost all of the cases and is the most energetic velocity component for weak-to-moderate convection (i.e. −zi/L≈4-z_i/L \approx4–1010). The vertical (wall-normal) variance yields better similarity with w⋆w_\star compared to the horizontal components (Figure 4(c)) and agrees well with existing relations [45], but the magnitude of the w′w' peak exhibits a weak non-monotonic trend with convection and its position moves downward approaching free convection.

Within the surface layer, the horizontal variances are poorly captured by MOST due to a height dependence separate from the local buoyancy effects represented by −z/L-z/L (Figure 6(a,Figure 6b)). In contrast, the vertical variance is well described by MOST (Figure 6(c)), particularly with local-in-height scaling (Figure 6(d)), demonstrating that the vertical velocity responds more directly to the local balance of shear and buoyancy than do the horizontal velocity components.

Under weaker convection within the surface layer, corresponding to cases c1–c6 and the presence of longitudinal rolls, the streamwise velocity variance retains the logarithmic dependence in (1.2) applicable to neutral conditions (Figure 7). The fitted logarithmic slope AuA_u remains approximately constant with respect to u∗u_\ast consistent with previous predictions for this regime [7], while the fitted offset BuB_u is proportional to w⋆w_\star. The latter enforces the mixed layer boundary condition above the surface layer, for which w⋆w_\star is the leading-order scaling parameter. For cells under strong convection, both horizontal variances more closely follow the empirical formulation of Wilson [85] which includes zi/Lz_i/L and z/ziz/z_i dependencies (Figure 8). These results indicate that the height dependence of the streamwise variance remains predominantly organised by mechanical shear in the presence of coherent rolls which have spanwise-aligned circulations (Figure 13(b)), whereas the horizontal variances become increasingly similar to each other as the preferential orientation of the rolls is lost (Figure 9(d,Figure 9e)).

The trends in the variance statistics are accompanied by analysis of the variance budget contributions in (2.1)–(2.3). Streamwise shear production decreases rapidly with increasing convection and supplies less than 10% of the total production across the CBL by the conclusion of the roll-to-cell transition (Figure 10(a)). As energy production changes, the overall pressure–strain redistribution adjusts to promote isotropic dissipation (i.e. return to isotropy) to first-order accuracy (Figure 10(c)).

Within the thermal roll regime represented by Figure 14(a), the largest-scale energy is produced by buoyancy in the mixed layer before being transported to the surface and redistributed preferentially to v′v' by pressure. In the transition to cells represented by Figure 14(b), the large-scale pressure transport connecting the surface and mixed layers progressively reduces, and the redistribution between the horizontal components becomes more even (figures 11 and 12). As this pathway weakens, the redistribution R\mathcal{R} becomes more active at smaller scales across all heights (Figure 11). These trends allow for increased local accumulation of large-scale w′w^\prime in the mixed layer, helping to explain the increasing vertical variance peak under strong convection. Further, there is a shift in the largest scales towards moderately higher wavenumbers as the rolls lose their streamwise coherence and transition to cells of smaller horizontal extent. The continuous decrease in TKE across the cases is attributed to this reduction in the largest scales as the cells contract and become geometrically isotropic. In the context of Figure 14, the roll-to-cell transition can therefore be interpreted as a reorganisation of the buoyancy-driven motions from large-scale, pressure-mediated rolls connecting the surface and mixed layers to more local, smaller-scale cells with increasingly symmetric horizontal redistribution.

Schematic summary of dominant pathways for turbulent energy production, transport, and redistribution in the CBL

Figure 14. Schematic summary of the dominant pathways for turbulent energy production, transport, and redistribution in the CBL: for the (a) roll-dominated and (b) cell-dominated regimes. Blue and orange arrows denote predominantly larger- and smaller-scale pathways, respectively. Arrow thickness denotes the relative strength or preferential nature of the corresponding transfer.

Acknowledgements. The authors acknowledge the high-performance computing resources provided by the Sydney Informatics Hub, a Core Research Facility of the University of Sydney, and the National Computational Infrastructure (NCI Australia), which is supported by the Australian Government.

Funding. This research received no specific grant from any funding agency, commercial or not-for-profit sectors.

Declaration of interests. The authors report no conflict of interest.

Appendix A

Computational domain sensitivity

Figure 15 presents a domain-size sensitivity test comparing variance statistics for the moderate-convection case c6 (−zi/L=40-z_i/L = 40) simulated on its original computational domain (Lx=Ly=12L_x = L_y = 12 km ≈10.9zi\approx10.9z_i) and the smaller domain (Lx=Ly=5.76L_x = L_y = 5.76 km ≈5.2zi\approx5.2z_i) used for the more convective cases in Table 3. The simulation conditions are identical aside from the domain size. The normalised spanwise (b) and vertical (c) velocity variance profiles exhibit close agreement across the entire boundary layer depth, with the mixed-layer peak in w′2⋆w^{\prime2\star} increasing by less than 4% in the smaller domain. There is a modest decrease in streamwise variance (a), but the decrease is small relative to the differences across the convective cases. This quantitative alignment demonstrates that the energy is sufficiently contained within the smaller computational domain for the more convective cases when the longitudinal rolls have fragmented, and the domain size does not impact the conclusions of the study.

Comparison of moderate-convection case c6 statistics for larger and smaller horizontal domains

Figure 15. Comparison of moderate-convection case c6 statistics using the larger (solid lines) and smaller (dashed lines) horizontal domains from Table 3, where the legend indicates the domain length LxL_x relative to ziz_i. Normalised streamwise, spanwise, and vertical velocity variance profiles are shown in (a), (b), and (c), respectively. The shaded grey regions denote the envelope of variance statistics across all cases in Figure 4.

Appendix B. SGS variance sensitivity

The LES grid filter is assumed to lie within the inertial subrange when the SGS variance is estimated, so that the unresolved energy can be extrapolated from the resolved scales using a k−5/3k^{-5/3} spectral slope [38, 69, 29]. The extrapolation is based on the Leonard stress, which represents the energy difference between the primary grid filter and a coarser test filter.

This assumption becomes increasingly questionable close to the wall because the characteristic turbulent length scales are constrained by the boundary. The integral scale can then approach the grid scale, leaving insufficient separation between the energy-containing range and the nominal inertial subrange. To quantify this uncertainty, the SGS variance was estimated using two primary grid-filter widths, Δ~=Δx\widetilde{\Delta} = \Delta x and Δ~=1.5Δx\widetilde{\Delta} = 1.5\Delta x, where Δx\Delta x is the horizontal grid spacing. A test filter twice as coarse was applied in each case.

Figure 16 illustrates the sensitivity of the estimated total velocity variances to the filter scale used to determine the SGS contribution. A third filter using Δ~=2Δx\widetilde{\Delta} = 2\Delta x is shown in the figure to illustrate the strong scale dependence close to the wall, but is not used in determining the cutoff.

Resolved and total velocity variances for cases c1, c4, and c12 at several filter scales

Figure 16. Resolved variances (solid lines) in cases c1, c4, and c12 compared with total variances using a range of filter scales to estimate the SGS contribution – Δ~=Δx\tilde{\Delta} = \Delta x (dashed lines), Δ~=1.5Δx\tilde{\Delta} = 1.5\Delta x (dotted lines), and Δ~=2Δx\tilde{\Delta} = 2\Delta x (dash-dotted lines) – for the normalised (a) streamwise, (b) spanwise, and (c) vertical velocity components. The near-wall region below the horizontal cutoffs is excluded from the main analysis.

The SGS estimate is considered reliable where the resulting total variance is insensitive to the choice of grid filter. In the present analysis, the cutoff height is therefore defined as the lowest height at which the total variances obtained with Δx\Delta x and 1.5Δx1.5\Delta x differ by less than 1%. Only data above this height are retained for the surface-layer fits and the variance analyses presented in the main text. This criterion ensures that the reported trends are based on the region where Δx\Delta x is well within the inertial subrange and the SGS variance extrapolation is accurate.

References

  1. [1]ALBERTSON, J. D. 1996 Large eddy simulation of land-atmosphere interaction. PhD thesis, University of California, Davis.
  2. [2]ALBERTSON, J. D. & PARLANGE, M. B. 1999 Surface length scales and shear stress: Implications for land-atmosphere interaction over complex terrain. Water Resour. Res. 35, 2121–2132.DOI
  3. [3]ANDRÉN, A., BROWN, A. R., MASON, P. J., GRAF, J., SCHUMANN, U., MOENG, C.-H. & NIEUWSTADT, F. T. M. 1994 Large-eddy simulation of a neutrally stratified boundary layer: A comparison of four computer codes. Q. J. R. Meteorol. Soc. 120, 1457–1484.
  4. [4]BAARS, W. J. & MARUSIC, I. 2020 Data-driven decomposition of the streamwise turbulence kinetic energy in boundary layers. Part 2: Integrated energy and A₁. J. Fluid Mech. 882, A26.
  5. [5]BAIDYA, R., PHILIP, J., HUTCHINS, N., MONTY, J. P. & MARUSIC, I. 2021 Spanwise velocity statistics in high-Reynolds-number turbulent boundary layers. J. Fluid Mech. 913, A35.DOI
  6. [6]BANERJEE, T. & KATUL, G. G. 2013 Logarithmic scaling in the longitudinal velocity variance explained by a spectral budget. Phys. Fluids 25 (12), 125106.DOI
  7. [7]BANERJEE, T., KATUL, G. G., SALESKY, S. T. & CHAMECKI, M. 2015 Revisiting the formulations for the longitudinal velocity variance in the unstable atmospheric surface layer. Q. J. R. Meteorol. Soc. 141, 1699–1711.DOI
  8. [9]BLASS, A., ZHU, X., VERZICCO, R., LOHSE, D. & STEVENS, R. J. A. M. 2020 Flow organization and heat transfer in turbulent wall sheared thermal convection. J. Fluid Mech. 897, A22.DOI
  9. [10]BOU-ZEID, E., GAO, X., ANSORGE, C. & KATUL, G. G. 2018 On the role of return to isotropy in wall-bounded turbulent flows with buoyancy. J. Fluid Mech. 856, 61–78.DOI
  10. [11]BOU-ZEID, E., MENEVEAU, C. & PARLANGE, M. B. 2005 A scale-dependent Lagrangian dynamic model for large eddy simulation of complex turbulent flows. Phys. Fluids 17, 025105.DOI
  11. [12]BRUTSAERT, W. 1982 Evaporation into the Atmosphere: Theory, History, and Applications. Springer.
  12. [13]CAUGHEY, S. J. & PALMER, S. G. 1979 Some aspects of turbulence structure through the depth of the convective boundary layer. Q. J. R. Meteorol. Soc. 105, 811–827.DOI
  13. [15]DE ROODE, S. R., SIEBESMA, A. P., JANSSON, F. & JANSSENS, M. 2022 Dependency of mesoscale organization on grid anisotropy in large-eddy simulations of convective boundary layers at Gray Zone resolutions. J. Adv. Model. Earth Syst. 14, e2022MS003095.DOI
  14. [16]DEARDORFF, J. W. 1970 Convective velocity and temperature scales for the unstable planetary boundary layer and for Rayleigh convection. J. Atmos. Sci. 27, 1211–1213.doi.org/10.1175/1520-0469(1970)027<1211:cvatsf>2.0.co;2
  15. [18]DING, M., NGUYEN, K. X., LIU, S., OTTE, M. J. & TONG, C. 2018 Investigation of the pressure–strain–rate correlation and pressure fluctuations in convective and near-neutral atmospheric surface layers. J. Fluid Mech. 854, 88–120.
  16. [19]Foken, T. 2006 50 years of the Monin–Obukhov similarity theory. Boundary-Layer Meteorol. 119, 431–447.DOI
  17. [20]Garratt, J. R. 1992 The Atmospheric Boundary Layer. Cambridge University Press.
  18. [21]Glendening, J. W. 2000 Budgets of lineal and nonlinear turbulent kinetic energy under strong shear conditions. J. Atmos. Sci. 57, 2297–2318.doi.org/10.1175/1520-0469(2000)057<2297:bolant>2.0.co;2
  19. [22]Greene, B. R., Otterstatter, L. M. & Salesky, S. T. 2025 How representative are uncrewed aircraft system measurements of the convective boundary layer? Geophys. Res. Lett. 52, e2024GL114291.DOI
  20. [23]Grossman, R. L. 1982 An analysis of vertical velocity spectra obtained in the BOMEX fair-weather, trade-wind boundary layer. Boundary-Layer Meteorol. 23, 323–357.DOI
  21. [24]Heinze, R., Mironov, D. & Raasch, S. 2016 Analysis of pressure-strain and pressure gradient-scalar covariances in cloud-topped boundary layers: A large-eddy simulation study. J. Adv. Model. Earth Syst. 8, 3–30.DOI
  22. [25]Heisel, M. & Chamecki, M. 2024 On the departure from Monin–Obukhov surface similarity and transition to the convective mixed layer. Boundary-Layer Meteorol. 190, 28.arxiv.org/abs/2308.16405
  23. [26]Heisel, M., Deshpande, R. & Katul, G. G. 2026 On the wall-normal velocity variance in canonical wall-bounded turbulence. J. Fluid Mech. 1033, A30.DOI
  24. [27]Högström, U. 1990 Analysis of turbulence structure in the surface layer with a modified similarity formulation for near neutral conditions. J. Atmos. Sci. 47, 1949–1972.doi.org/10.1175/1520-0469(1990)047<1949:aotsit>2.0.co;2
  25. [28]Højstrup, J. 1982 Velocity spectra in the unstable planetary boundary layer. J. Atmos. Sci. 39, 2239–2248.doi.org/10.1175/1520-0469(1982)039<2239:vsitup>2.0.co;2
  26. [29]Hossen, M. K., Mulayath Variyath, A. & Alam, J. M. 2021 Statistical analysis of dynamic subgrid modeling approaches in large eddy simulation. Aerospace 8 (12), 375.DOI
  27. [31]Huang, K. Y. & Katul, G. G. 2022 Profiles of high-order moments of longitudinal velocity explained by the random sweeping decorrelation hypothesis. Phys. Rev. Fluids 7, 044603.DOI
  28. [32]Hunt, J. C. R. & Morrison, J. F. 2000 Eddy structure in turbulent boundary layers. Eur. J. Mech. B/Fluids 19, 673–694.DOI
  29. [33]Jayaraman, B. & Brasseur, J. G. 2021 Transition in atmospheric boundary layer turbulence structure from neutral to convective, and large-scale rolls. J. Fluid Mech. 913, A42.
  30. [34]Kaimal, J. C. & Finnigan, J. J. 1994 Atmospheric Boundary Layer Flows: Their Structure and Measurement. Oxford University Press.
  31. [35]Kaimal, J. C., Wyngaard, J. C., Haugen, D. A., Coté, O. R., Izumi, Y., Caughey, S. J. & Readings, C. J. 1976 Turbulence structure in the convective boundary layer. J. Atmos. Sci. 33, 2152–2169.doi.org/10.1175/1520-0469(1976)033<2152:tsitcb>2.0.co;2
  32. [36]Khanna, S. & Brasseur, J. G. 1997 Analysis of Monin–Obukhov similarity from large-eddy simulation. J. Fluid Mech. 345, 251–286.DOI
  33. [37]Khanna, S. & Brasseur, J. G. 1998 Three-dimensional buoyancy- and shear-induced local structure of the atmospheric boundary layer. J. Atmos. Sci. 55, 710–743.doi.org/10.1175/1520-0469(1998)055<0710:tdbasi>2.0.co;2
  34. [38]Knaepen, B., Deplinqué, O. & Carati, D. 2002 Subgrid-scale energy and pseudo pressure in large-eddy simulation. Phys. Fluids 14, 4235–4241.DOI
  35. [39]Kumar, V., Kleissl, J., Meneveau, C. & Parlange, M. B. 2006 Large-eddy simulation of a diurnal cycle of the atmospheric boundary layer: Atmospheric stability and scaling issues. Water Resour. Res. 42, W06D09.DOI
  36. [40]Kunkel, G. J. & Mariusic, I. 2006 Study of the near-wall-turbulent region of the high-Reynolds-number boundary layer using an atmospheric flow. J. Fluid Mech. 548, 375–402.DOI
  37. [41]Lee, M. & Moser, R. D. 2015 Direct numerical simulation of turbulent channel flow up to Reτ ≈ 5200. J. Fluid Mech. 774, 395–415.DOI
  38. [43]LeMone, M. A. 1976 Modulation of turbulence energy by longitudinal rolls in an unstable planetary boundary layer. J. Atmos. Sci. 33, 1308–1320.doi.org/10.1175/1520-0469(1976)033<1308:motebl>2.0.co;2
  39. [44]Lenschow, D. H., Lothon, M., Mayor, S. D., Sullivan, P. P. & Canut, G. 2012 A comparison of higher-order vertical velocity moments in the convective boundary layer from lidar with in situ measurements and large-eddy simulation. Boundary-Layer Meteorol. 143 (1), 107–123.DOI
  40. [45]Lenschow, D. H., Wyngaard, J. C. & Pennell, W. T. 1980 Mean-field and second-moment budgets in a baroclinic, convective boundary layer. J. Atmos. Sci. 37, 1313–1326.doi.org/10.1175/1520-0469(1980)037<1313:mfasmb>2.0.co;2
  41. [46]Lumley, J. L. & Panofsky, H. A. 1964 The Structure of Atmospheric Turbulence, Interscience Monographs and Texts in Physics and Astronomy, vol. 12. New York: Interscience Publishers.
  42. [47]Maronga, B. & Reuder, J. 2017 On the formulation and universality of Monin–Obukhov similarity functions for mean gradients and standard deviations in the unstable surface layer: Results from surface-layer-resolving large-eddy simulations. J. Atmos. Sci. 74, 989–1010.
  43. [48]MARUSIC, I. & MONTY, J. P. 2019 Attached eddy model of wall turbulence. Annu. Rev. Fluid Mech. 51, 49–74.DOI
  44. [49]MARUSIC, I., MONTY, J. P., HULTMARK, M. & SMITS, A. J. 2013 On the logarithmic region in wall turbulence. J. Fluid Mech. 716, R3.DOI
  45. [50]MIRONOV, D. V., GRYANIK, V. M., MOENG, C.-H., OLBERS, D. J. & WARNCKE, T. H. 2000 Vertical turbulence structure and second-moment budgets in convection with rotation: A large-eddy simulation study. Q. J. R. Meteorol. Soc. 126 (563), 477–515.DOI
  46. [51]MOENG, C.-H. 1984 A large-eddy-simulation model for the study of planetary boundary-layer turbulence. J. Atmos. Sci. 41, 2052–2062.doi.org/10.1175/1520-0469(1984)041<2052:alesmf>2.0.co;2
  47. [52]MOENG, C.-H. & SULLIVAN, P. P. 1994 A comparison of shear- and buoyancy-driven planetary boundary layer flows. J. Atmos. Sci. 51, 999–1022.doi.org/10.1175/1520-0469(1994)051<0999:acosab>2.0.co;2
  48. [53]MOENG, C.-H. & WYNGAARD, J. C. 1986 An analysis of closures for pressure-scalar covariances in the convective boundary layer. J. Atmos. Sci. 43, 2499–2513.doi.org/10.1175/1520-0469(1986)043<2499:aaocfp>2.0.co;2
  49. [54]MORAES, O. L. L., ACEVEDO, O. C., DEGRAZIA, G. A., ANFOSSI, D., DA SILVA, R. & ANABOR, V. 2005 Surface layer turbulence parameters over a complex terrain. Atmos. Environ. 39, 3103–3112.DOI
  50. [55]NGUYEN, K. X., HORST, T. W., ONCLEY, S. P. & TONG, C. 2013 Measurements of the budgets of the subgrid-scale stress and temperature flux in a convective atmospheric surface layer. J. Fluid Mech. 729, 388–422.
  51. [56]NIEUWSTADT, F. T. M. 1984 The turbulent structure of the stable, nocturnal boundary layer. J. Atmos. Sci. 41, 2202–2216.doi.org/10.1175/1520-0469(1984)041<2202:ttsots>2.0.co;2
  52. [57]NIEUWSTADT, F. T. M., MASON, P. J., MOENG, C.-H. & SCHUMANN, U. 1993 Large-eddy simulation of the convective boundary layer: A comparison of four computer codes. In Turbulent Shear Flows 8 (ed. F. Durst, R. Friedrich, B. E. Launder, F. W. Schmidt, U. Schumann & J. H. Whitelaw), pp. 343–367. Berlin, Heidelberg: Springer.
  53. [58]PANDEY, A., SCHEEL, J. D. & SCHUMACHER, J. 2018 Turbulent superstructures in Rayleigh–Bénard convection. Nat. Commun. 9, 2118.DOI
  54. [59]PANOFSKY, H. A. & DUTTON, J. A. 1984 Atmospheric Turbulence: Models and Methods for Engineering Applications. Wiley.
  55. [60]PANOFSKY, H. A., TENNEKES, H., LENSCHOW, D. H. & WYNGAARD, J. C. 1977 The characteristics of turbulent velocity components in the surface layer under convective conditions. Boundary-Layer Meteorol. 11, 355–361.DOI
  56. [61]PERRY, A. E. & LI, J. D. 1990 Experimental support for the attached-eddy hypothesis in zero-pressure-gradient turbulent boundary layers. J. Fluid Mech. 218, 405–438.DOI
  57. [62]PERRY, A. E., LIM, K. L. & HENBEST, S. M. 1987 An experimental study of the turbulence structure in smooth- and rough-wall boundary layers. J. Fluid Mech. 177, 437–466.DOI
  58. [63]PIROZZOLI, S. 2024 On the streamwise velocity variance in the near-wall region of turbulent flows. J. Fluid Mech. 989, A5.DOI
  59. [64]POPE, S. B. 2000 Turbulent Flows. Cambridge University Press.
  60. [65]PORTÉ-AGEL, F., MENÉVEAU, C. & PARLANGE, M. B. 2000 A scale-dependent dynamic model for large-eddy simulation: application to a neutral atmospheric boundary layer. J. Fluid Mech. 415, 261–284.
  61. [66]PRIESTLEY, C. H. B. 1954 Convection from a large horizontal surface. Aust. J. Phys. 7, 176–201.DOI
  62. [67]QIN, Y., KATUL, G. G., LIU, H. & LI, D. 2025 Asymptotic coefficients of the attached-eddy model derived from an adiabatic atmosphere. J. Fluid Mech. 1011, A29.arxiv.org/abs/2411.02756
  63. [68]SALESKY, S. T. & ANDERSON, W. 2018 Buoyancy effects on large-scale motions in convective atmospheric boundary layers: implications for modulation of near-wall processes. J. Fluid Mech. 856, 135–168.DOI
  64. [69]SALESKY, S. T. & CHAMECKI, M. 2012 A similarity model of subfilter-scale energy for large-eddy simulations of the atmospheric boundary layer. Boundary-Layer Meteorol. 145, 69–91.
  65. [70]SALESKY, S. T., CHAMECKI, M. & BOU-ZEID, E. 2017 On the nature of the transition between roll and cellular organization in the convective boundary layer. Boundary-Layer Meteorol. 163, 41–68.DOI
  66. [71]SORBJAN, Z. 1986 On similarity in the atmospheric boundary layer. Boundary-Layer Meteorol. 34, 377–397.DOI
  67. [72]SORBJAN, Z. 1988 Local similarity in the convective boundary layer (CBL). Boundary-Layer Meteorol. 45, 237–250.DOI
  68. [73]STIPERSKI, I. & CALAF, M. 2023 Generalizing Monin–Obukhov similarity theory (1954) for complex atmospheric turbulence. Phys. Rev. Lett. 130, 124001.DOI
  69. [74]STOLL, R., GIBBS, J. A., SALESKY, S. T., ANDERSON, W. & CALAF, M. 2020 Large-eddy simulation of the atmospheric boundary layer. Boundary-Layer Meteorol. 177, 541–581.DOI
  70. [75]STULL, R. B. 1988 An Introduction to Boundary Layer Meteorology. Springer Netherlands.
  71. [77]Sykes, R. I. & Henn, D. S. 1989 Large-eddy simulation of turbulent sheared convection. J. Atmos. Sci. 46, 1106–1118.doi.org/10.1175/1520-0469(1989)046<1106:lesots>2.0.co;2
  72. [78]Tennekes, H. & Lumley, J. L. 1972 A First Course in Turbulence. MIT Press.
  73. [79]Tong, C. & Nguyen, K. X. 2015 Multipoint Monin–Obukhov similarity and its application to turbulence spectra in the convective atmospheric surface layer. J. Atmos. Sci. 72, 4337–4348.DOI
  74. [80]Townsend, A. A. 1976 The Structure of Turbulent Shear Flow. Cambridge University Press.
  75. [81]Wang, B., Wang, Q., Wan, Z.-H. & Sun, D.-J. 2021 The influence of aspect ratio on flow states in the buoyancy-driven turbulence with free slip boundaries. Int. J. Heat Mass Transfer 178, 121639.DOI
  76. [82]Wang, Q., Chong, K. L., Stevens, R. J. A. M., Verzicco, R. & Lohse, D. 2020 From zonal flow to convection rolls in Rayleigh–Bénard convection with free-slip plates. J. Fluid Mech. 905, A21.arxiv.org/abs/2005.02084
  77. [83]Weckwerth, T. M., Horst, T. W. & Wilson, J. W. 1999 An observational study of the evolution of horizontal convective rolls. Mon. Wea. Rev. 127, 2160–2179.doi.org/10.1175/1520-0493(1999)127<2160:aosote>2.0.co;2
  78. [84]Wilczak, J. M., Oncley, S. P. & Stage, S. A. 2001 Sonic anemometer tilt correction algorithms. Boundary-Layer Meteorol. 99, 127–150.DOI
  79. [85]Wilson, J. D. 2008 Monin–Obukhov functions for standard deviations of velocity. Boundary-Layer Meteorol. 129, 353–369.DOI
  80. [86]Wyngaard, J. C. 2010 Turbulence in the Atmosphere. Cambridge University Press.
  81. [87]Wyngaard, J. C. & Côté, O. R. 1974 The evolution of a convective planetary boundary layer — a higher-order-closure model study. Boundary-Layer Meteorol. 7, 289–308.DOI
  82. [88]Wyngaard, J. C., Côté, O. R. & Izumi, Y. 1971 Local free convection, similarity, and the budgets of shear stress and heat flux. J. Atmos. Sci. 28, 1171–1182.doi.org/10.1175/1520-0469(1971)028<1171:lfcsat>2.0.co;2
  83. [89]Yaglom, A. M. 1993 Similarity laws for wall turbulent flows: their limitations and generalizations. In New Approaches and Concepts in Turbulence (ed. Th. Dracos & A. Tsinober), pp. 7–27. Basel: Birkhäuser.DOI
  84. [90]Yang, X. I. A., Pirozzoli, S. & Abkar, M. 2020 Scaling of velocity fluctuations in statistically unstable boundary-layer flows. J. Fluid Mech. 886, A3.DOI
  85. [92]Zhou, B., Sun, S., Sun, J. & Zhu, K. 2019 The universality of the normalized vertical velocity variance in contrast to the horizontal velocity variance in the convective boundary layer. J. Atmos. Sci. 76, 1437–1456.DOI
  86. [93]Zilitinkevich, S. 1994 A generalized scaling for convective shear flows. Boundary-Layer Meteorol. 70, 51–78.DOI
  87. [94]Zilitinkevich, S. S., Hunt, J. C. R., Esau, I. N., Grachev, A. A., Lalas, D. P., Akylas, E., Tombrou, M., Fairall, C. W., Fernando, H. J. S., Baklanov, A. A. & Joffre, S. M. 2006 The influence of large convective eddies on the surface-layer turbulence. Q. J. R. Meteorol. Soc. 132, 1423–1456.
  88. [95]Zimmerman, S., Philip, J., Monty, J., Talamelli, A., Marusic, I., Ganapathisubramani, B., Hearst, R. J., Bellani, G., Baidya, R., Samie, M., Zheng, X., Dogan, E., Mascotelli, L. & Klewicki, J. 2019 A comparative study of the velocity and vorticity structure in pipes and boundary layers at friction Reynolds numbers up to 10⁴. J. Fluid Mech. 869, 182–213.

Paper details

vlnc:W_9secs7ca
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 10

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents