Hostname: page-component-7b7567c8c-w7t8q Total loading time: 0 Render date: 2026-11-18T06:48:45.926Z Has data issue: false hasContentIssue false

Robust relation of streamwise velocity autocorrelation in atmospheric surface layers based on an autoregressive moving average model

Published online by Cambridge University Press:  21 February 2024

Fei-Chi Zhang
Affiliation:
Department of Mechanics and Engineering Science at College of Engineering, and State Key Laboratory for Turbulence and Complex Systems, Peking University, Beijing 100871, PR China
Jin-Han Xie*
Affiliation:
Department of Mechanics and Engineering Science at College of Engineering, and State Key Laboratory for Turbulence and Complex Systems, Peking University, Beijing 100871, PR ChinaJoint Laboratory of Marine Hydrodynamics and Ocean Engineering, Laoshan Laboratory, Shandong 266237, PR China
Song Xi Chen*
Affiliation:
School of Mathematical Sciences, Guanghua School of Management and Center for Statistical Science, Peking University, Beijing 100871, PR China
Xiaojing Zheng*
Affiliation:
Department of Mechanics and Engineering Science at College of Engineering, and State Key Laboratory for Turbulence and Complex Systems, Peking University, Beijing 100871, PR ChinaCenter for Particle-Laden Turbulence, Lanzhou University, Lanzhou 730000, PR China
*
†Email addresses for correspondence: jinhanxie@pku.edu.cn, csx@gsm.pku.edu.cn, xjzheng@lzu.edu.cn
†Email addresses for correspondence: jinhanxie@pku.edu.cn, csx@gsm.pku.edu.cn, xjzheng@lzu.edu.cn
†Email addresses for correspondence: jinhanxie@pku.edu.cn, csx@gsm.pku.edu.cn, xjzheng@lzu.edu.cn

Abstract

We construct an autoregressive moving average (ARMA) model consisting of the history and random effects for the streamwise velocity fluctuation in boundary-layer turbulence. The distance to the wall and the boundary-layer thickness determine the time step and the order of the ARMA model, respectively. Based on the autocorrelation's analytical expression of the ARMA model, we obtain a global analytical expression for the second-order structure function, which asymptotically captures the inertial, dynamic and large-scale ranges. Specifically, the exponential autocorrelation of the ARMA model arises from the autoregressive coefficients and is modified to logarithmic behaviour by the moving-average coefficients. The asymptotic expressions enable us to determine model coefficients by existing parameters, such as the Kolmogorov and the Townsend–Perry constants. A consequent double-log expression for the characteristic length scale is derived and is justified by direct numerical simulation data with $Re_\tau \approx 5200$ and field-measured neutral atmospheric surface layer data with $Re_\tau \sim O(10^6)$ from the Qingtu Lake Observation Array site. This relation is robust because it applies to $Re_\tau$ from $O(10^4)$ to $O(10^6)$, and even when the statistics of natural ASL deviate from those of canonical boundary-layer turbulence, e.g. in the case of imbalance in energy production and dissipation, and when the Townsend–Perry constant deviates from traditional values.

Information

Type
JFM Papers
Copyright
© The Author(s), 2024. Published by Cambridge University Press
Figure 0

Figure 1.A schematic diagram of the two points in space used for autocorrelation and second-order structure function.

Figure 1

Figure 2.An illustration of the three ranges of the second-order structure function in log-normal coordinates. Here, $\eta _1$ and $\eta _2$ are the dimensionless lower and upper ends of the dynamic range.

Figure 2

Figure 3.Variation of $T_{\eta _1}$ and $T_{\eta _2}$ with $z/\delta$ according to (2.21a) and (2.21b), where $M_2=3.62$, $B_1=1.25$, $A_1=1.73$ and $B_2=2.5$.

Figure 3

Figure 4.Variations of (a) $f_1$ and (b) $f_2$ with $\mu =B_2/(2\langle u^{+2} \rangle )$ for $\xi _2=2/3$.

Figure 4

Figure 5.Profiles of statistics for selected intervals at ${Re}_\tau = O(10^6)$. (a) Log-linear plot of the mean streamwise velocity. The open symbols are current QLOA data and the dashed line corresponds to the log-law with $\kappa = 0.41$ (Pope 2000). (b) Logarithmic decay of streamwise turbulent intensity. The blue symbols is the ASL result of Hutchins et al. (2012) at ${Re}_\tau = O(10^6)$. The dashed and dotted lines correspond to (2.17) with $B_1=0.55$ and $0.80$, respectively. (c) Dimensionless shear Reynolds stress. The blue symbols are ASL data from Hutchins et al. (2012) and the dashed line is the similarity formulation from Chauhan (2007).

Figure 5

Figure 6.Diagrammatic flow of applying the ARMA model to boundary-layer turbulence.

Figure 6

Figure 7.(a) Raw fluctuating velocity $u$; (b) reproduced fluctuating velocity $\tilde u$; (c) comparison of autocorrelations of ASL data and ARMA model in normal-normal coordinates; and (d) comparison of the global expression, second-order structure functions of ASL data and ARMA model in log-normal coordinates, where $u_\tau =0.37$ m s$^{-1}$, ${Re}_\tau =3.78\times 10^6$ and $z^+=2.27\times 10^4$.

Figure 7

Figure 8.Fitted model parameters $\beta _{i}$, $i=1,2,\ldots,q$, with ${Re}_\tau =3.78\times 10^6$.

Figure 8

Figure 9.(a) Power spectra $\varPhi _{uu}$ and (b) premultiplied spectra $k_x\varPhi _{uu}$ obtained from ASL data and ARMA(1, $q$) model, respectively. Here ${Re}_\tau =3.78\times 10^6$ and $z^+=7.55\times 10^5$.

Figure 9

Figure 10.(a) Comparison of $\langle \Delta u^{+2} \rangle$ obtained from ASL data and global asymptotic expression at ${Re}_\tau =3.78\times 10^6$. The intermediate scales of theoretical curves are as follows: Curve 1 – $\eta _1 = 1.06$, $\eta _2 = 27.71$; Curve 2 – $\eta _1 = 0.87$, $\eta _2 = 23.12$; Curve 3 – $\eta _1 = 0.73$, $\eta _2 = 15.42$. The black curves correspond to expressions for inertial and dynamic ranges presented by de Silva et al. (2015). (b) Comparison of $\eta _2$ obtained from ASL data and the expression (2.18) with $A_1/B_2=1.15$, $A_2/B_2=1/\xi _2=1.50$, $B_1/B_2=0.41$ and $B_1=0.80$.

Figure 10

Figure 11.(a) Comparison of $\langle \Delta u^{+2} \rangle$ obtained from DNS and experiment data and global asymptotic expression. Here, $z^+$ locates in the logarithmic region. The triangle symbols correspond to the DNS data at ${Re}_\tau \approx 5200$ (Lee & Moser 2015), and the circle and square symbols correspond to the experiment data ${Re}_\tau = 19\,000$ (Hutchins et al.2009; de Silva et al.2015). (b) Comparison of $\eta _2$ obtained from DNS data and the expression (2.15b) with $A_1=1.72$ and $B_1=1.50$.

Figure 11

Figure 12.Behaviour of covariance of streamwise velocity $u$ and vertical velocity $w$ in the (a) dynamic range and (b) large-scale range.

Figure 12

Figure 13.(a) Dynamic and large-scale ranges of second-order structure function with new scale $\eta _2 z$ at different high friction Reynolds numbers ${Re}_\tau$. (b) Exponential scaling of the second-order structure function for scales larger than $\eta _2 z$.

Figure 13

Figure 14.(a) Characteristic length scales summarized by Liu et al. (2017) and the data comes from Lee & Sung (2011), Volino, Schultz & Flack (2007), Tutkun et al. (2009), Hutchins & Marusic (2007), Hutchins et al. (2012), Liu et al. (2017) and Önder & Meyers (2018). Equation (2.22) provides the prediction of parameters $A_1/B_2$ and $B_1/B_2$. (b) Some ASL data that deviate from the $L_x$ summarized by Liu et al. (2017) can also be characterized by (2.22). The height-averaged ratio of energy production and dissipation rate $P/\varepsilon$ for the time intervals corresponding to the circle, triangle and square symbols are 1.84, 1.47 and 1.37, respectively.

Figure 14

Figure 15.Comparison of integral length scale $\mathcal {L}$ and approximate expression (2.29a). The autocorrelation is integrated up to the first 0.01 crossing. The parameter $\mu$ in $f_1$ is obtained from ASL data and $\xi _2=2/3$.

Figure 15

Figure 16.Errors in the optimization of $\eta _1$.

Figure 16

Table 1.Uncertainty ranges of $M_2$, $\xi _2$ and $\eta _1$ in figure 10(a).