physics0202047/art.tex
1: \documentclass[12pt]{article}
2: %\textheight=26cm
3: \usepackage{epsfig,amsmath,amsfonts}
4: \usepackage[figuresright]{rotating}
5: \newcommand{\bm}[1]{\mbox{\boldmath $#1$}}
6: %%%%\renewcommand{\baselinestretch}{2}  %%double spaced
7: %\pagestyle{empty}
8: \begin{document}
9: \title{Statistics of temperature fluctuations in a buoyancy dominated
10: boundary layer flow simulated by a Large-eddy simulation model}
11: \author{Marta Antonelli$^1$, Andrea Mazzino$^{2,1}$ and Umberto Rizza$^{2}$\\
12: \small{$^1$ INFM--Department of Physics, University of Genova, I--16146
13: Genova, Italy}\\
14: \small{$^2$ ISAC/CNR - Sezione di Lecce - Strada provinciale 
15: Lecce-Monteroni km 1.2 73100 Lecce}}
16: \date{\today}
17: \maketitle
18: \begin{abstract}
19: Temperature fluctuations in an  atmospheric convective boundary layer
20: are investigated by means of Large Eddy Simulations (LES).
21: A novel statistical characterization for both weak temperature fluctuations
22: and strong  temperature fluctuations has been found. Despite the
23: nontriviality of the dynamics of temperature fluctuations, our data
24: support the idea that the most relevant statistical properties can be
25: captured solely in terms of two scaling exponents, 
26: characterizing the entire mixed layer.
27: Such exponents control asymptotic (i.e. core and tails) 
28: rescaling properties of the probability density functions of 
29: equal-time
30: temperature differences, $\Delta_r \theta$,  between points separated 
31: by a distance ${\bm r}$. A link between statistical properties of
32: large temperature fluctuations and geometrical properties of the
33: set hosting such fluctuations is also provided.
34: Finally, a possible application of our new findings to the problem 
35: of subgrid-scale parameterizations for the temperature field
36: in a convective boundary layer is discussed.  
37: \end{abstract}
38: 
39: %\noindent PACS number(s)\,: 47.52.+j, 05.45.Ac
40: \vspace{4mm}
41: 
42: \section{Introduction}
43: Temperature in an atmospheric boundary layer (ABL) is typically convected by
44: a velocity background, ${\bm v}$, and diffuses by the action of
45: molecular motion and/or small-scale turbulent eddies. The basic
46: equation governing such a process is the well-known
47: advection-diffusion equation (Pielke, 1984) for the (potential)
48: temperature, $\theta$,
49: \begin{equation}
50: \partial_t\theta + v_{\alpha} \partial_{\alpha}\theta=D_0\partial^2\theta +
51: S_{\theta}\;\;\;,
52: \label{FP}
53: \end{equation}
54: where $S_{\theta}$ represents the sources and sinks of heat, 
55: eventually present within the domain, ${\bm v}$ is the 
56: velocity field advecting the
57: temperature, and $D_0$ may represent either the diffusion
58: coefficients
59: or, alternatively, an eddy diffusion coefficient if one intends to
60: focus on the large scale behavior (i.e.~large eddies) of $\theta$
61: and thus needs to parameterize, 
62: in some way, small scale temperature dynamics. Repeated indexes are summed.\\
63: Here, we use the following short notations:
64: $\partial_t\equiv \partial\bullet/\partial t$;  
65: $\partial_i\equiv \partial\bullet/\partial x_i$, $i=1,\cdots ,3$;
66: %${\bm v}\cdot {\bm \partial}\equiv \sum_{i=1}^3 v_i\partial_i\bullet$; 
67: $\partial^2\equiv\sum_{i=1}^3\partial_i\partial_i\bullet$.
68: 
69: In many situations, temperature dynamics, driven by the
70: velocity field via the advection term in Eq.~(\ref{FP}), does not
71: react back on the velocity field. This is the case, for instance,
72: in a neutrally stratified boundary layer (occurring, e.g., 
73: under windy conditions with a complete cloud cover) where buoyancy forces are
74: negligible compared
75: to the other terms in the Navier-Stokes (NS) equations ruling the velocity
76: field dynamics. In this case, temperature behaves as a passive scalar 
77: as a good approximation. \\
78: As a matter of fact, during the diurnal cycle  
79: neutral stratification is rarely observed in the ABL (Garratt, 1999).
80: More frequently, one observes stable conditions (occurring, e.g., at night in
81: response to surface cooling by longwave emission to space) 
82: or unstable, convective, ones (occurring, e.g., when strong surface heating
83: produces thermal instability or convection).
84: The role of temperature is active in both cases, that amounts to say
85: that temperature drives the velocity field dynamics through the buoyancy
86: contribution (usually modeled by means of the well-known Boussinesq
87: coupling). The latter contribution is now the leading one in the NS equations.
88: 
89: The main feature characterizing both active and passive scalar
90: dynamics is the presence of strong fluctuations in the temperature
91: field. Such fluctuations affect the whole range of scales involved in
92: the temperature dynamics, from the largest scales of motion
93: to the smallest ones where diffusive effects become important.
94: This huge excitation of degrees of freedom,
95: gives meaning to the term ``scalar
96: turbulence'' recently used to denote the dynamics
97: of temperature fluctuations (Shraiman and Siggia, 2000).
98: 
99: The first unpleasant consequence of persistent fluctuations is the failure
100: of  any attempt to construct dimensional theories for the statistics 
101: of temperature fluctuations (Frisch, 1995), quantitatively defined as 
102: temperature differences $\Delta_r\theta$ between points separated by $r$:
103: $\Delta_r\theta\equiv\theta({\bm r},t)-\theta({\bm 0},t)$. 
104: The common strategy
105: for dimensional approaches consists to define {\it typical}
106: length/time scales 
107: and {\it typical} amplitude for the fluctuations of
108: the unknown fields (e.g., $\Delta_r \theta$ and $\Delta_r{\bm v}$)
109: and then try to balance the various terms in the basic equations
110: (e.g.~Eq.~(\ref{FP}) coupled to the NS equations)
111: to deduce predictions for $\Delta_r \theta$ and 
112: $\Delta_r{\bm v}$ as a function of the separation ${\bm r}$.\\
113: This is, for instance, the essence of the first dimensional
114: theory for scalar turbulence due to Kolmogorov, Obukhov and Corrsin
115: (1949) and Bolgiano (1959). As a result of these theories,
116: probability density functions (pdfs), $P(\Delta_r\theta)$, 
117: of temperature differences, $\Delta_r \theta$, obey the following simple 
118: rescaling: $P(\Delta_r\theta)=r^{-\alpha}\tilde{P}(\Delta_r\theta/r^{\alpha})$,
119: where $\tilde{P}$ is a function of the sole ratio 
120: $\Delta_r\theta/r^{\alpha}$.\\
121: Such property immediately implies that, dimensionally, one has
122: $\Delta_r\theta\sim r^{\alpha}$ and, for the $p$-th moment of $\Delta_r\theta$:
123: $\langle(\Delta_r \theta)^p\rangle\sim r^{\zeta_p}$, with $\zeta_p$
124: (known as scaling exponents) linear function of $p$:  $\zeta_p=\alpha p$\\
125: The linearity of $\zeta_p$ vs $p$ reflects the fact that only one parameter,
126: $\alpha$, is necessary to explain most of the statistical properties of
127: $\theta$.
128: One has, in other words, `single-scale fluctuations';
129: that amounts to say that it is irrelevant which part of the
130: probability density function of temperature differences is sampled to
131: define a typical fluctuation.
132: 
133: Rather than the above simple scenario predicted by dimensional
134: theories, turbulent systems show an infinite hierarchy  of
135: `independent' fluctuations (Frisch, 1995), that amounts to say the
136: strong dependence on the order $p$ considered to define the 
137: typical fluctuation. More quantitatively, turbulent systems exhibit 
138: a nonlinear behavior of $\zeta_p$ vs $p$\footnote{Such curve must be
139: concave and not decreasing,
140: as it follows from general inequalities of probability theory
141: (see, e.g.,~ Frisch (1995)).} where the infinite set of exponents, 
142: $\zeta_p$, select different classes of fluctuations. The departure of the
143: actual curve of $\zeta_p$ vs $p$ from the linear (dimensional)
144: prediction is named ``intermittency'' or ``anomalous scaling''
145: (Frisch, 1995). Intermittency is probably the most representative
146: property characterizing a turbulent system.
147: 
148: Our aim here is to provide a statistical characterization of
149: temperature fluctuations in a convective boundary layer dominated by
150: well-organized plumes, simulated by
151: a large eddy simulation model (Moeng, 1984).\\
152: We shall focus, in particular, on the statistical characterization 
153: of two different classes of fluctuations: weak temperature
154: fluctuations, mainly occurring in the inner plume regions and strong
155: temperature fluctuations, associated to the plume interfaces (see Fig.~1). 
156: As we shall show, weak fluctuations are associated to linear behavior
157: of scaling exponents $\zeta_p$ vs $p$ for small $p$'s, 
158: while strong fluctuations
159: (captured by large $p$'s) cause the so-called intermittency saturations, i.e.
160: the curve $\zeta_p$ vs $p$ tends a to a constant value, $\zeta_{\infty}$,  
161: for $p$ large enough. The saturation exponent $\zeta_{\infty}$  is
162: simply connected (see Sec.~\ref{geom}) 
163: to the fractal dimension of the set hosting 
164: large temperature excursions: $\zeta_{\infty}=d-D_F$, where $d$ 
165: and  $D_F$ are the  usual dimension of the space and the fractal 
166: dimension of the large temperature fluctuation set, respectively.
167: \\
168: Despite the complexity of temperature fluctuation dynamics in a 
169: convective boundary layer (CBL), 
170: reflected in the strong intermittency of the system,
171: only two exponents are necessary to capture most of the statistics
172: of temperature fluctuations.  \\
173: It is worth emphasizing that the same statistical 
174: characterization has been recently found 
175: for two dimensional idealized models of both passive 
176: (see Frisch et al (1999), Celani et al
177: (2000) and Celani et al (2001a)) 
178: and active scalar turbulence (see Celani et al (2001b))  
179: simulated by means of both direct numerical
180: simulations and Lagrangian methods (see, e.g., Frisch et al (1998)).
181: This points toward the generality of our new findings
182: within the context of scalar transport.
183: 
184: 
185: 
186: \section{Statistical tools}
187: \label{tools}
188: The aim of this section is to provide a quick summary of
189: the statistical tools we have exploited to characterize temperature 
190: fluctuations in a CBL. \\
191: The basic and well-known indicator is the probability density function,
192: $P(\Delta_{{\bm r}; {\bm x}}\theta)$, 
193: of temperature differences, $\Delta_{{\bm r}; {\bm x}}\theta$, 
194: over a scale $r$, 
195: defined as:
196: \begin{equation}
197: \Delta_{{\bm r}; {\bm x}}\theta \equiv \theta({\bm x}+
198: {\bm r},t)-\theta({\bm x},t)\;\;\; .
199: \label{SF} 
200: \end{equation}
201: The pdf $P(\Delta_{{\bm r}; {\bm x}}\theta)$ 
202: will depend other than on ${\bm r}$
203: also on ${\bm x}$ if the system is not homogeneous. In our CBL,
204: we have homogeneity along $x$-$y$ planes, but not
205: along the vertical direction $z$. We thus shall have a dependence on the 
206: vertical coordinate $z$.
207: Moreover, in the analysis of Sec.~\ref{risu}, separations ${\bm r}$ will be
208: taken along $x$-$y$ planes in the direction forming an angle
209: of $\pi /4$ with the geostrophic wind direction. We shall thus denote
210: our pdf simply as $P(\Delta_{r; z}\theta)$. The choice for the
211: direction $\pi /4$ has been done in order to
212: reduce the contamination of the
213: scaling exponents by anisotropic effects.
214: More details on this `magic' angle can be found in 
215: Celani et al. (2001a).
216: 
217: By definition, weak fluctuations (i.e. small values of $|\theta({\bm x}+
218: {\bm r},t)-\theta({\bm x},t)|$ with respect to a typical fluctuation
219: defined  as 
220: $\sigma^{(z)}\equiv [\langle\theta^2\rangle- \langle\theta\rangle^2
221: ]^{1/2}$) 
222: are associated to the pdf core while, on the contrary, large
223: fluctuations
224: (i.e. $|\theta({\bm x}+
225: {\bm r},t)-\theta({\bm x},t)| \gg \sigma^{(z)}$) are associated to the 
226: pdf tails.\\
227: The above considerations can be easily paraphrased 
228: in terms of the moments, 
229: $S_p({\bm r};z)\equiv \langle(\Delta_{r;z} \theta)^p\rangle$,
230: known as structure functions.
231: Large fluctuations are captured by large $p$'s, while weak fluctuations
232: by small $p$'s.
233: 
234: Let us now  introduce two possible behaviors for the pdf 
235: $P(\Delta_{r; z}\theta)$. As we shall show in the sequel,
236: such behaviors will characterize, within the entire mixed layer,
237: the statistical properties of
238: weak temperature fluctuations and strong temperature fluctuations, 
239: respectively.\\
240: 
241: \noindent{\bf Self-similar behavior}
242: 
243: In terms of probability density functions of
244: $\Delta_{r;z}\theta$,
245: such a  behavior is defined by the rescaling property:
246: \begin{equation}
247: P(\Delta_{r;z}\theta)=r^{-\alpha^{(z)}}\tilde{P}
248: \left (\frac{\Delta_{r;z}\theta}{r^{\alpha^{(z)}}}\right )\;\;\; .
249: \label{pdf-resc1}
250: \end{equation}
251: It can be immediately verified from the definition of moments:
252: \begin{equation}
253: S_p({\bm r};z)=\int_{-\infty}^{+\infty} P(\Delta_{r;z}\theta)
254: (\Delta_{r;z}\theta)^p d(\Delta_{r;z}\theta)
255: \label{mom}
256: \end{equation}
257: that (\ref{pdf-resc1}) is equivalent to the following behavior 
258: for the structure functions, $S_p$:
259: \begin{equation}
260: S_p({\bm r};z)\sim r^{\zeta_{p}^{(z)}}\ \ \mbox{with}\ \ \ 
261: \zeta_{p}^{(z)}=\alpha^{(z)} p, ~~~ \alpha^{(z)}>0\;\;\; ,
262: \label{lin} 
263: \end{equation}
264: that is a linear behavior, with the factor $\alpha^{(z)}$ in general depending
265: on the elevation $z$ within the mixed layer.
266: 
267: It is worth noticing that (\ref{pdf-resc1}) and (\ref{lin})
268: does not necessarily imply a Gaussian shape for
269: $P(\Delta_{r;z}\theta)$. On the contrary, if $P(\Delta_{r;z}\theta)$
270: is Gaussian then (\ref{lin}) (and thus (\ref{pdf-resc1})) are
271: immediately satisfied as it follows from the well-known property of 
272: Gaussian statistics (see, e.g., Frisch (1995)): 
273: $S_{2p}({\bm r};z)=(2p-1)!!\;[S_2({\bm r};z)]^p$,
274: together with the assumption that $S_2({\bm r};z)\sim r^{2\alpha^{(z)}}$.\\
275: 
276: \noindent{\bf Intermittency saturation}
277: (i.e.~the strongest violation of dimensional predictions):
278: 
279: In terms of $P(\Delta_{r;z}\theta)$, intermittency saturation
280: is defined as 
281: \begin{equation}
282: P(\Delta_{r;z}\theta)=\frac{r^{\zeta_{\infty}^{(z)}}}{\sigma^{(z)}}
283: Q\left (\frac{\Delta_{r;z}\theta}{\sigma^{(z)}}\right )
284: ~~~\mbox{for}~~~|\Delta_{r;z}\theta|>\lambda\sigma^{(z)}~~~(\lambda >1)\;\;\; ,
285: \label{resca}
286: \end{equation}
287: where $Q$ is some function (not determined a {\it priori}) which 
288: does not depend on the separation $r$. \\
289: In terms of cumulated probabilities, i.e.~the sum (integral)
290: of the pdfs over the large temperature fluctuations (i.e.~for
291: $|\Delta_{r;z}\theta| > \lambda \sigma^{(z)}$, with $\lambda>1$),
292: defined as:
293: \begin{equation}
294: Prob[|\Delta_{r;z}\theta|>\lambda\sigma^{(z)}]
295: \equiv \int_{-\infty}^{-\lambda\sigma^{(z)}}
296: P(\Delta_{r;z}\theta) d(\Delta_{r;z}\theta)+
297: \int_{\lambda\sigma^{(z)}}^{+\infty}
298: P(\Delta_{r;z}\theta) d(\Delta_{r;z}\theta),
299: %\sim r^{\zeta_{\infty}^{(z)}}
300: \label{cum}
301: \end{equation}
302: saturation is equivalent to the following power law behavior, holding 
303: for different values of $\lambda>1$:
304: \begin{equation}
305: Prob[|\Delta_{r;z}\theta|>\lambda\sigma^{(z)}]
306: \sim r^{\zeta_{\infty}^{(z)}}\;\;\; .
307: \label{pawer}
308: \end{equation}
309: The scaling exponents, $\zeta_{\infty}^{(z)}$, can be thus easily extracted
310: by measuring the slope of 
311: $\log \left\{Prob[|\Delta_{r;z}\theta|>\lambda\sigma^{(z)}]\right\}$ vs
312: $\log r$.
313: 
314: Finally, it is worth observing that,
315: in terms of structure functions intermittency saturations means:
316: \begin{equation}
317: S_p({\bm r};z)\sim r^{\zeta_{p}^{(z)}}\ \ \mbox{with}\ \ \ 
318: \zeta_{p}^{(z)}=\zeta_{\infty}^{(z)}, ~~~
319: \mbox{for}~~~p>p_{crit}\;\;\; ,
320: \label{sat} 
321: \end{equation}
322: as one can easily verify from the definitions of moments (\ref{mom})
323: and from (\ref{resca}).\\
324: The scaling exponents $\zeta_p^{(z)}$ thus tend to 
325: a constant value $\zeta_{\infty}^{(z)}$
326: for orders, $p$'s large enough. Such behavior justifies the
327: word `saturation' to denote the laws (\ref{resca}) and (\ref{pawer}).
328:  
329: 
330: \section{The Large-Eddy simulation model}
331: 
332: In order to gather statistical informations on the turbulent structure
333: of a CBL, we used the LES code described in Moeng (1984)
334: and Sullivan et al (1994). Such model 
335: has been widely used and tested to investigate fundamental 
336: problems in the framework of boundary layers
337: (see, e.g., Moeng and Wyngaard (1989), Moeng et al. (1992), 
338: Andr\'en and Moeng (1993), Moeng and Sullivan (1994), among the
339: others).\\ For this reason we confine ourselves only on 
340: general aspects of the LES strategy. Details can be found
341: in the aforementioned references.
342: 
343: The key point of the LES strategy is that the large scale motion 
344: (i.e. motion associated to the large turbulent eddies) is explicitly solved 
345: while the smallest scales (typically in the inertial range of scales)
346: are described in a statistical consistent way
347: (i.e. parameterized in terms of the resolved, large scale, 
348: velocity and temperature fields).
349: This is done by filtering the governing equations for velocity and
350: potential temperature
351: by means of a filter operator. Applied, e.g., to the  potential
352: temperature field $\theta$, the filter
353: is defined as the convolution:
354: \begin{equation}
355: \overline{\theta}({\bm x})=\int \theta({\bm x}')G({\bm x}-{\bm x}')d{\bm x}'
356: \end{equation}
357: where $\overline{\theta}$ is the filtered variable and $G({\bf x})$ is a
358: tridimensional filter function. The field $\theta$ can be
359: thus decomposed as 
360: \begin{eqnarray}
361: \label{decom}
362: \theta=\overline{\theta}+{\theta}' .
363: \end{eqnarray}
364: Applying the filter operator both to the Navier--Stokes equations 
365: and to the equation for the potential temperature, and exploiting 
366: the decomposition (\ref{decom}) (and the analogous for the velocity field)
367: in the advection terms one 
368: obtains the corresponding filtered equations. For the sake of brevity,
369: we report the sole filtered equation for the potential temperature:
370: \begin{eqnarray}
371: \label{filtNS2} 
372: \partial_t\overline{\theta}=-\overline{\overline{v}_{\alpha}\partial_{\alpha}
373: \overline{\theta}}-\partial_{\alpha}\tau^{(\theta)}_{\alpha}
374: \end{eqnarray}
375: where $\tau^{(\theta)}_{\alpha}$ are the subgrid turbulence fluxes of
376: virtual temperature (in short SGS fluxes). They are related to the
377: resolved-scale field
378: as 
379: \begin{equation}
380: \tau^{(\theta)}_{\alpha}=-K_{H}\partial_{\alpha}\overline{\theta}
381: \end{equation}
382: $K_{H}$ being the SGS eddy coefficient for heat. 
383: A similar expression holds for the subgrid turbulence fluxes of
384: momentum (see Moeng, 1984) that are defined in terms of the 
385: SGS eddy coefficient for momentum ($K_{M}$).\\
386: The above two eddy coefficients are related to the velocity scale
387: $\overline{e'}^{1/2}$, $\overline{e'}$ being the SGS turbulence energy the 
388: equation of which is solved in this model, and to the length scale
389: $l\equiv (\Delta x\Delta y\Delta z )^{1/3}$ (valid for the convective
390: cases) $\Delta x$, $\Delta y$, and $ \Delta z $ being the grid mesh
391: spacing in $x$, $y$ and $z$. Namely:
392: \begin{equation}
393: K_{M}=0.1\; l\; \overline{e'}^{1/2}
394: \end{equation}
395: \begin{equation}
396: K_{H}=3 K_{M}.  
397: \end{equation}
398:  
399: \section{The simulated  convective experiment}
400: \label{simula}
401: 
402: In the present first study, our attention has been focused on the
403: Simulation B (hereafter referred to as Sim B) by Moeng and Sullivan
404: (1994). Sim B is a buoyancy-dominated flow with a 
405: relatively small shear effect, where vigorous thermals (see again Fig.~1) 
406: set up due to buoyancy force.\\
407: The sole difference of our simulation
408: with respect to the Moeng and Sullivan's simulation 
409: is the increased spatial resolution, here of $128^3$ grid points.\\
410: A preliminary sensitivity test at the lower resolution 
411: $96^3$ (as in  Moeng and Sullivan (1994)) did not show
412: significant differences in the results we are going to present.
413: Sensitivity tests at higher resolutions are still in progress
414: and seem to confirm our preceding conclusion.
415: 
416: Our choice for a convective boundary layer lies on the fact that, 
417: in such regimes,
418: dependence of resolved fields
419: on SGS parameterization should be very weak, and thus LES strategy
420: appears completely justified.
421: Indeed, in  convective regimes,  SGS motion acts as net 
422: energy sinks that drain energy from
423: the resolved motion. This is another way to say that energy blows
424: from large scales of motion toward the smallest scales and 
425: the cumulative (statistical) effect of the latter scales can be 
426: successfully captured
427: by means of simple eddy-diffusivity/viscosity SGS models. 
428: Uncertainties
429: eventually present at the smallest scales directly affected by 
430: SGS parameterizations (that are not the concern of the analysis
431: we are going to show)
432: do not propagate upward but are promptly diffuse
433: (and thus dissipated)  owing to the action of the aforementioned 
434: eddy-diffusivity/viscosity character of SGS motion. 
435: Genuine inertial range dynamics can thus develop and, as we shall see,
436: the typical features
437: characterizing an inertial range of scales (e.g., rescaling properties
438: of statistical objects) to appear. 
439: 
440: The following parameters characterize
441: the Sim B. Geostrophic wind, $U_g=10\;m/s$;
442: friction velocity, $u_*= 0.56\; m/s$; convective
443: velocity, $w_*= 2.02\; m/s$;
444: PBL height, $z_i= 1030\;m$; large-eddy
445: turnover time, $\tau_*= 510\;s$; stability parameter, $z_i/L=-18 $ ($L$
446: being the Monin--Obukov length);
447: potential temperature flux at the surface, $Q_*=0.24\;m K/s$.\\
448: Moreover, the numerical domain size in the $x$, $y$ and $z$ directions
449: are $L_x=L_y=5\;km$ and $L_z=2\;km$, respectively; the time step for the
450: numerical integration is about $1\;s$. 
451: For details on  the simulated experiment, readers can
452: refer to Moeng and Sullivan (1994).
453: 
454: To perform our statistical analysis, we first reached the
455: quasi-steady state. It took, as in Moeng and Sullivan (1994), 
456: about six large-eddy turnover times, $\tau_*$.
457: After that time, a new simulation has been made for about $37 \tau_*$ 
458: and the simulated potential temperature field saved at $0.5 \tau_*$
459: intervals for the analysis. Our data set was thus formed by 74
460: (almost independent)  potential temperature snapshots.
461: 
462: Each simulation hour required about 24 computer hours on an
463: Alpha-XP1000  workstation. 
464: 
465: \section{Results and discussions}
466: \label{risu}
467: \subsection{Statistics of large temperature fluctuations}
468: Let us start our statistical analysis from the large temperature
469: fluctuations. These are controlled by the pdf tails of temperature
470: differences, $\Delta_{r;z}\theta$, and, as we are going to show, they
471: are compatible with
472: the laws (\ref{resca}) and (\ref{pawer}), that means
473: intermittency saturation. \\
474: To show that, it is enough to see whether or not there exist a
475: positive number, $\zeta_{\infty}^{(z)}$, such that the quantities
476: $\sigma^{(z)} P(\Delta_{r;z}\theta) r^{-\zeta_{\infty}^{(z)}}$ 
477: collapse on the
478: same curve, $Q$, for different values of the separation $r$. Indeed, as
479: showed in Sec.~\ref{tools}, in the presence of saturation
480: the function $Q$, appearing in (\ref{resca}),
481: does not depend on $r$. \\
482: The validity of (\ref{resca}) can be seen in Fig.~2, where the
483: behavior of $P(\Delta_{r;z}\theta)$ for $z/z_i=0.3$ and two 
484: values of $r$ are shown, 
485: $z$ and $z_i$ being the elevation above the bottom boundary and
486: the mixed layer height, respectively. In the graph (a), 
487: $P(\Delta_{r;z}\theta)$ is reported without any $r$-dependent 
488: rescaling; in (b) 
489: we show $\sigma^{(z)}P(\Delta_{r;z}\theta) r^{-\zeta_{\infty}^{(z)}}$ for
490: $\zeta_{\infty}^{(z)}\sim 0.6$. The data collapse occurring on the 
491: tails of the curves of graph (b) is the footprint of intermittency 
492: saturation.\\ 
493: In Figs.~3 and 4 we show the analogous of Fig.~2 but for 
494: $z/z_i=0.45 $ and $z/z_i=0.6 $. Also in these cases, the exponent
495: giving the data collapse is $\zeta_{\infty}^{(z)}\sim 0.6$.
496: Similar behaviors have been observed for all $z$'s within the mixed layer.\\
497: As a conclusion, from the evidences of Figs.~2, 3 and 4, it turns out that
498: the saturation exponent $\zeta_{\infty}^{(z)}$ does not depend on $z$
499: within the mixed layer. It is thus a property of the entire mixed
500: layer and, for this reason, it will be simply denoted by
501: $\zeta_{\infty}$.
502: 
503: Let us now corroborate the scenario of intermittency saturation
504: by looking at the cumulated probability (\ref{cum}). \\
505: For the saturation to occur,  such probability 
506: has to behave as a power
507: law with exponent  $\zeta_{\infty}$ (see (\ref{pawer})).  
508: Such behavior is indeed observed and showed 
509: in Fig.~4(a) (for $z/z_i=0.3$ and $\lambda = 5$ and $5.5$)
510: and in Fig.~4(b) (for $z/z_i=0.6$ and $\lambda = 5$ and $5.5$).
511: The continuous lines have the slope $\zeta_{\infty}\sim 0.6$ as measured from
512: Figs.~2 and 3. The fact that there exist a region of scales, $r$, where
513: that slope is parallel to the slope of the cumulated probabilities
514: means, again, intermittency saturation with a unique
515: (i.e.~characterizing the whole mixed layer) exponent.\\
516: It is worth noting 
517: that figures similar to Figs.~5(a) and 5(b) have been obtained
518: also for smaller values of $\lambda$, e.g., $\lambda=2.5$ and
519: $\lambda=3.5$. Population of strong events 
520: being decreasing as $\lambda$ increases, the above independence on
521: $\lambda$  points for the robustness of our statistics.
522: 
523: In Fig.~6(a) we report, for $z/z_i=0.45$, the behaviors of the 
524: sixth and eight-order structure functions of temperature
525: differences 
526: vs the separation between points (squares). Stright lines have the
527: slope $\zeta_{\infty}=0.6$.  This is a further, direct, evidence of
528: intermittency saturation. To investigate the statistical convergence
529: of our sixth and eight-order  moments, we reported in  Fig.~6(b) 
530: the bulk contribution to such moments: 
531: $(\Delta_r \theta)^p P(\Delta_r \theta)$,
532: with $p=6$ and $p=8$,
533: $r/L\sim 7\times 10^{-2}$ in the inertial range of scales. 
534: For comparison $P(\Delta_r \theta)$
535: is also shown. Note that the maximum contribution
536: to the moments six and eight comes  from fluctuations $\Delta_r \theta$
537: in the region where $P(\Delta_r \theta)$ (i.e., $p=0$) is well
538: resolved, i.e., our statistics appears reliable up to the order eight. 
539: 
540: 
541: Once $\zeta_{\infty}$ is known, we evaluated from (\ref{resca})
542: the unknown function $Q$. Such function is
543: shown in Fig.~7 for two different values of $z$ within the mixed layer. 
544: Differences among the two curves are evident, signaling that $Q$
545: contains a dependence on the elevation $z$. Such dependence can be
546: associated to the relatively small shear  present in our
547: convective simulation (see Sec.~\ref{simula}) .\\
548: Further simulations spanning intermediate ABL (i.e.~where both shear
549: and buoyancy are important) have however to be performed in order
550: to confirm the above last conclusion. 
551: 
552: It is worth stressing that the scales $r$ at which we observe 
553: scaling behaviors are always larger than $\sim 8$ grid-points
554: (i.e.~sufficiently far from the scales directly affected
555: by SGS parameterizations). 
556: Our attention being focused in a region sufficiently 
557: far from boundaries, this is another point in favor for the
558: possible SGS independence of our results. 
559: 
560: 
561: \subsubsection{A link between geometry and statistics}
562: \label{geom}
563: Let us now conclude this section with a geometric point of view 
564: for the intermittency saturation (see also Celani et al, 2001a).
565: As we shall see, the saturation exponent $\zeta_{\infty}$ is related
566: to the fractal dimension, $D_F$, of the set hosting the strong
567: temperature fluctuations. \\
568: To do that, let us schematize in a very rough way our strong 
569: (i.e.~larger than some $\sigma^{(z)}$)
570: temperature fluctuations in the form of quasi-discontinuities 
571: (i.e.~step functions). Each quasi-discontinuity 
572: will define a point (we are performing the analysis on 
573: planes at constant $z$) of given coordinate  
574: in our two-dimensional plane. The ensemble of all
575: points defines the set, $S$, hosting strong temperature fluctuations.
576: Roughly, $S$ is formed by the intersection of our two-dimensional
577: plane with the plume interfaces across which
578: strong temperature jumps occur.\\
579: A useful indicator to characterize geometrically our set, $S$, is the 
580: fractal dimension, $D_F$, (see, e.g., Frisch (1995) for a presentation 
581: oriented toward turbulence problems). We briefly recall the standard
582: way to define $D_F$.\\
583: \begin{itemize}
584: \item Take boxes of side, $r$, and cover the whole plane at fixed $z$.
585: Denote with $N_{tot}$ the total number of those boxes;
586: \item Define the function $N(r)$ as the number of boxes containing
587: at least one point of $S$;
588: \item For $r$ sufficiently small, one expects power law behavior for
589: $N(r)$ in the form: $N(r)\sim r^{-D_F}$, which defines the fractal
590: dimension of $S$.  
591: \end{itemize}
592: Given the fractal dimension, $D_F$, it is now easy to compute
593: the probability, $Prob[|\Delta_{r;z}\theta|>\lambda\sigma^{(z)}]$,
594: of having strong (i.e.~larger than some $\sigma^{(z)}$)
595: temperature jumps within a certain
596: distance $r$. Indeed, by definition, we have:
597: \begin{equation}
598: Prob[|\Delta_{r;z}\theta|>\lambda\sigma^{(z)}]
599: \equiv\frac{\mbox{favorable cases}}{\mbox{possible cases}}=
600: \frac{N(r)}{N_{tot}}\sim \frac{r^{-D_F}}{r^{-2}}=r^{2-D_F}\;\;\; .
601: \label{frac}
602: \end{equation}
603: From (\ref{pawer}) and (\ref{sat}) the identification
604: $\zeta_{\infty}=2-D_{F}$
605: immediately follows. Notice that if one does not restrict the
606: attention on the sole planes at constant $z$, but focuses  on the whole
607: three dimensional space, the above relation becomes
608: $\zeta_{\infty}=3-D_{F}'$ where $D_{F}'$ is the fractal dimension
609: of the new set $S$.
610: 
611: 
612: 
613: \subsection{Statistics of weak temperature fluctuations}
614: Let us now pass to investigate the statistics of well-mixed regions of
615: the temperature field, corresponding to the inner parts of plumes that
616: are likely to be present in our CBL (see again Fig.~1).\\
617: In these regions, fluctuations turn out to be very gentle and,
618: as an immediate consequence, statistics is expected to be controlled
619: in terms of single-scale fluctuations (see the Introduction).
620: The best candidate to characterize, from a statistical point of view,
621: weak fluctuations is thus the rescaling form given by (\ref{pdf-resc1}). \\
622: To investigate whether or not our data are compatible with such
623: rescaling, it is enough to verify
624: whether there exist a number, $\alpha^{(z)}$, ({\it a priori} dependent
625: on the elevation $z$ within the mixed layer) such that, 
626: looking at $r^{\alpha^{(z)}}P(\Delta_{r;z}\theta)$ vs
627: $\Delta_{r;z}\theta/r^{\alpha^{(z)}}$ for different values of $r$, all
628: curves collapse one on the other for each value of $z$.\\
629: Our data support this behavior for the pdf cores (as expected, the rescaling
630: (\ref{pdf-resc1}) holds solely for weak fluctuations), as it can be 
631: observed in Fig.~2(c) (for $z/z_i=0.3$), in Fig.~3(c) 
632: (for $z/z_i=0.45$) and Fig.~4(c)  (for $z/z_i=0.6$). 
633: In all cases, the values of $\alpha^{(z)}$
634: is $\sim 0.2$, that means that $\alpha^{(z)}$ does not depend on $z$.
635: As the exponent $\zeta_{\infty}$, $\alpha\equiv\alpha^{(z)}$ thus 
636: characterizes the entire mixed layer.
637: 
638: 
639: \section{Conclusions and discussions}
640: We have characterized, from a statistical point of view, both large and weak 
641: temperature fluctuations of a convective boundary layer simulated by a 
642: large eddy simulation model. \\
643: The main results of our study can be summarized as follows.
644: \begin{itemize}
645: \item Large temperature fluctuations, occurring across plume interfaces,
646: turn out to be strongly intermittent. This is the cause of the observed 
647: break down of mean field theories {\'a} la Kolmogorov, predicting a linear 
648: behavior of the scaling exponents, $\zeta_p$, of  
649: the structure functions of temperature differences, vs the order $p$. 
650: We found, on the contrary,
651: a pdf rescaling which corresponds to 
652: a nonlinear shape of $\zeta_p$ vs $p$, with 
653: $\zeta_p\to\zeta_{\infty}\equiv const$ for $p$ large enough. This behavior
654: is named {\it intermittency saturation}, i.e.~the strongest violation of
655: dimensional predictions.\\
656: Hence, the concept of `typical fluctuation'
657: does not make sense: it is necessary to specify which part of the pdf of 
658: temperature differences is sampled for the definition of `typical fluctuation'.
659: 
660: \item Weak temperature fluctuations, characterizing the inner plume region
661: where temperature  is extremely well-mixed,  
662: have a self-similar character. This amounts to say 
663: that, despite the fact that many scales are excited in the well-mixed regions,
664: the concept of `typical fluctuation' here makes sense. In this case
665: a simple rescaling characterizes the pdf core, which corresponds to
666: a linear behavior of the curve $\zeta_p$ vs $p$ for small $p$'s.
667: The slope of the straight line $\zeta_p$ vs $p$ 
668: is $\alpha\sim 0.2$.
669: 
670: \item Exponents $\alpha$ and $\zeta_{\infty}$ appear to be independent on 
671: the elevation within the mixed layer. They are thus an intrinsic property of 
672: the entire mixed layer.
673: 
674: \item Statistics and geometry turn out to be intimately related.
675: A simple relationship holds indeed between $\zeta_{\infty}$ and the 
676: fractal dimension, $D_F$, of the set hosting the large temperature 
677: fluctuations:
678: $\zeta_{\infty}=d-D_F$ where $d$ is the usual dimension of the physical 
679: space.\\
680: As for $\zeta_{\infty}$, $D_F$ appears an intrinsic, i.e.~$z$-independent,
681: property of the entire mixed layer.
682: \end{itemize}
683: 
684: It is worth stressing that the present scenario holds also for
685: idealized, two-dimensional, models of scalar turbulence both passive
686: (see Frisch et al (1999), Celani et al
687: (2000) and Celani et al (2001a)) 
688: and active (see Celani et al (2001b)),  
689: simulated by means of direct numerical simulations.
690: This fact naturally points toward the  possible generality of the 
691: present statistical characterization for the entire class of scalar 
692: transport problems.
693: 
694: Finally, let us discuss a possible application of our results
695: within  boundary layer physics, and, more specifically, in the
696: LES approach. As well known, one of the most challenging
697: problem in the  LES strategy is to find a proper way to describe
698: the dynamical effect of small-scale unresolved motion on the
699: resolved large scale dynamics.
700: Recently, new approaches have emerged as alternatives to the eddy
701: viscosity and similarity models (see, e.g., Meneveau and Katz, 2000).
702: They construct the small-scale unresolved part of a total
703: field (e.g., the velocity field) by extrapolating properties
704: of the (resolved) coarse-grained field. A specific
705: form for the subgrid scale field is thus postulated 
706: exploiting scale-invariance (i.e.~inertial range scaling behavior)
707: of the coarse-grained field.
708: Note that, standard approaches postulate the form of the stress
709: tensor rather than the structure of the unresolved field.\\
710: The mathematical tool which permits to perform such an interpolation
711: is known as ``fractal interpolation'' (see Meneveau and Katz
712: (2000) and references therein) where the free parameter of the
713: theory is the fractal dimension of the field.\\
714: Up to now, many efforts have been devoted to exploit such
715: strategy for the velocity fields. The latter exhibit however 
716: a  multifractal structure (roughly speaking,
717:  an infinite set of fractal dimensions characterizes the whole field)
718: and thus the fractal dimension parameter is a sort of `mean field'
719: description. \\
720: The suggestion arising from
721: our results is that the same technique exploited for the velocity field
722: appears  successfully applicable
723: in the convective case to the temperature field as well. Indeed,
724: our results support a fractal structure of the temperature
725: field. The situation seems to be even better than that 
726: for the velocity field. As we have stressed in the preceding
727: sections, solely two exponents characterize most of the statistical
728: properties of temperature fluctuations. We thus propose to schematize
729: temperature fluctuations as a bi-fractal object (i.e., the simplest
730: multifractal object) described by our two exponents $\alpha$ and 
731: $\zeta_{\infty}$, and to generalize ``fractal interpolation'' to 
732: this case.\\
733: To do that, it seems necessary to investigate how the exponents
734: we found for the analyzed experiment change by varying, e.g., the
735: weight of buoyancy with respect to shear. We are currently working
736: on this point and, as far as we remain on convective experiments,
737: it seems that only small variations appear. \\
738: It should also be  interesting to investigate what happens to the
739: above scenario in stable stratified boundary layers. In that
740: cases LES approach appears more delicate than in a CBL.   
741: Observations in field and/or in a wind tunnel 
742: should  be thus necessary to investigate the problem.
743:  
744: 
745: 
746: \section*{References}
747: 
748: \noindent  Andr\'en, A., and C.-H. Moeng, 1993: 
749: Single-point closure in a neutrally 
750: stratified boundary layer. {\it J. Atmos. Sci.}, {\bf 50}, 3366-3379. \\
751: 
752: \noindent  Bolgiano, R., 1959: Turbulent spectra in a stably 
753: stratified atmosphere.  {\it J. Geophys. Res.}, {\bf 64}, 2226-2229.\\
754: 
755: \noindent  Celani, A., A. Lanotte, A. Mazzino, and M. Vergassola, 2000:
756: Universality and saturation of intermittency
757: in passive scalar turbulence.
758: {\it Phys. Rev. Lett.}, {\bf 84},  2385-2388.\\
759: 
760: \noindent  Celani, A., A. Lanotte, A. Mazzino, and M. Vergassola, 2001a:
761: Fronts in passive scalar turbulence.
762: {\it Phys. Fluids}, {\bf 13}, 1768-1783.\\
763: 
764: \noindent  Celani, A., A. Mazzino, and M. Vergassola, 2001b:
765: Thermal plume turbulence.
766: {\it Phys. Fluids}, {\bf 13}, 2133-2135.\\
767: 
768: \noindent Frisch, U., 1995: Turbulence. {\it The legacy of A.N. 
769: Kolmogorov.} Cambridge University  Press, 296 pp.\\
770: 
771: \noindent Frisch, U., A. Mazzino, and M. Vergassola, 1998:
772: Intermittency in passive scalar advection.
773: {\it Phys. Rev. Lett.}, {\bf 80}, 5532-5537.\\
774: 
775: \noindent Frisch, U., A. Mazzino, and M. Vergassola, 1999:
776: Lagrangian dynamics and high-order moments
777: intermittency in passive scalar advection.
778: {\it Phys. Chem. Earth}, {\bf 24}, 945-951.\\
779: 
780: \noindent Garratt, J.R., 1999: {\it The atmospheric boundary layer.} 
781: Cambridge University Press, 316 pp.\\
782: 
783: \noindent Meneveau, C. and J. Katz, 2000: 
784: Scale-Invariance and turbulence models for Large-eddy simulation,
785: {\it Annu. Rev. Fluid Mech.}, {\bf 32}, 1-32.\\
786: 
787: \noindent Moeng, C.-H., 1984: A large-eddy-simulation model for the study of 
788: planetary boundary-layer turbulence, {\it J. Atmos. Sci.}, {\bf 41}, 2052-2062. \\
789: 
790: \noindent Moeng, C.-H. and J.C. Wyngaard, 1989: Evaluation of turbulent transport and dissipation closures in second-order modeling. 
791: {\it J. Atmos. Sci.}, {\bf 46}, 2311-2330. \\
792: 
793: \noindent Moeng C.-H., and P.P. Sullivan, 1994: A comparison of shear and 
794: buoyancy driven Planetary Boundary Layer flows.
795: {\it J. Atmos. Sci.}, {\bf 51}, 999-1021.\\
796: 
797: \noindent Obukhov, A., 1949: Structure of the temperature field in turbulence. 
798: {\it Izv. Akad. Nauk. SSSR. Ser. Geogr.}, {\bf 13}, 55-69.\\
799: 
800: \noindent Pielke, R.A., 1984: Mesoscale Meteorological Modeling.
801: Academic Press, 612 pp.\\
802: 
803: \noindent Shraiman B.I, and E.D. Siggia, 2000: Scalar turbulence. {\it
804: Nature},
805: {\bf 405}, 639-646.\\
806: 
807: \noindent Sullivan, P.P., J.C.~McWilliams, and C.-H.~Moeng, 1994:
808: A subgrid-scale model for large-eddy simulation of planetary boundary layer
809: flows. {\it Bound. Layer Meteorol.}, {\bf 71}, 247-276.\\
810: 
811: 
812: \newpage
813: 
814: \vskip 0.2cm
815: \noindent {\bf Acknowledgments}\\
816: We are particularly grateful to Chin-Hoh Moeng  and Peter Sullivan,
817: for providing us with their LES code as well as
818: many useful comments, discussions and suggestions. 
819: Helpful discussions and suggestions by
820: A.~Celani, R.~Festa, C.F.~Ratto and M.~Vergassola are also acknowledged.
821: This work has been partially supported by the INFM project GEPAIGG01
822: and Cofin 2001, prot. 2001023848.
823: Simulations have been performed at CINECA (INFM parallel computing initiative).
824: 
825: 
826: \newpage
827: 
828: \section*{List of Figures}
829: 
830: \begin{enumerate}
831: 
832: \item 
833: A typical snapshot of the potential temperature field $\theta$,
834: in the quasi-steady state of a convective boundary layer
835: simulated by a Large Eddy Simulation with resolution $128^3$.
836: Above: vertical cross-section restricted to the mixed
837: layer; below: horizontal  cross-section inside the  mixed
838: layer. Colors are coded
839: according to the intensity of the field: white corresponds to large
840: temperature, black to small ones. Plumes and well-mixed regions are
841: clearly detectable.
842: 
843: \item 
844: The pdf's $P(\Delta_{r;z} \theta)$, for two values of $r$
845: inside the inertial range of scales (dotted lines: $r/L=0.22$; 
846: continuous line: $r/L=0.11$, $L$ being the side of the (squared)
847: simulation domain) and $z/z_i=0.3$, $z_i$ being the elevation
848: of the mixed layer top.
849: (a): pdf's are shown without
850: any $r$-dependent rescaling; (b) the pdf is multiplied by the factor
851: $\sigma^{(z)} r^{-\zeta_\infty}$ with $\zeta_{\infty}\sim 0.6$:
852: the collapse of the curves indicate the asymptotic behavior
853: $P(\Delta_{r;z} \theta)\sim r^{\zeta_\infty}$ for large $\Delta_{r;z} \theta$,
854: that means saturation of temperature scaling exponents (see,
855:  and (\protect\ref{resca}), (\protect\ref{pawer}) and (\protect\ref{sat}));
856: (c) pdf's are multiplied by the factor $r^{\alpha^{(z)}}$
857: while $\Delta_{r;z} \theta$ by $r^{-\alpha^{(z)}}$:
858: the collapse of pdf cores indicates the validity of (\protect\ref{pdf-resc1})
859: that is equivalent to the
860: linear behavior of low-order  temperature scaling exponents 
861: (see (\protect\ref{mom})).
862: 
863: \item 
864: As in Fig.~2 but for $z/z_i=0.45$.
865: 
866: \item 
867: As in Fig.~2 but for $z/z_i=0.6$.
868: 
869: \item 
870: The cumulated probabilities 
871: $Prob[|\Delta_{r;z}\theta|>\lambda\sigma^{(z)}]$
872: for two values of $\lambda$
873: are shown for (a): $z/z_i=0.3$ and (b): $z/z_i=0.6$. 
874: The slope of these curves (continuous line) are compatible with the 
875: exponent $\zeta_{\infty}\sim 0.6$. The error bar on this slope is
876: of the order of  $0.1$, evaluated by means of the local scaling exponents (on
877: half-decade ratios) as customary in turbulence data analysis.
878: 
879: 
880: \item 
881: (a) Sixth and eight-order structure functions of temperature
882: differences 
883: vs the separation between points (squares). Stright lines have the
884: slope $\zeta_{\infty}=0.6$.  (b) The bulk contribution to the 
885: moments $p=6$ and $p=8$, $(\Delta_r \theta)^p P(\Delta_r \theta)$,
886: with $r/L\sim 1.1\times 10^{-1}$ in the inertial range of scales. 
887: For comparison $P(\Delta_r \theta)$
888: (i.e. for $p=0$) is also shown. Note that the maximum contribution
889: to the moments six and eight comes  from fluctuations $\Delta_r \theta$
890: in the region where $P(\Delta_r \theta)$ is well resolved. This proves
891: the reliability of our statistics to compute moments up to the order
892: eight.
893: 
894: \item 
895: The function $Q$ defined in (\protect\ref{resca}) is shown for two
896: different values of $z$ within the mixed layer: $z/z_i=0.3$
897: (dotted line) and $z/z_i=0.6$ (continuous line).
898: Differences in the shape of these two curves, reveal that $Q$ contains
899: a dependence on the elevation, $z$,  within the mixed layer.
900: 
901: 
902: \end{enumerate}
903: 
904: \newpage
905: \pagestyle{empty}
906: 
907: %
908: %
909: %
910: \begin{figure}
911: \vspace{-0cm}
912: \begin{center}
913: \includegraphics[scale=0.5,angle=0]{fig1a.eps}
914: \end{center}
915: \vspace{2cm}
916: \begin{center}
917: \hspace{2cm}\includegraphics[scale=0.5,angle=0]{fig1b.eps}
918: \end{center}
919: \caption{}
920: \end{figure}
921: %
922: %
923: %
924: \begin{figure}
925: \vspace{-0cm}
926: \begin{center}
927: \includegraphics[scale=0.43,angle=-90]{fig2a.eps}
928: \end{center}
929: \begin{center}
930: \includegraphics[scale=0.43,angle=-90]{fig2b.eps}
931: \end{center}
932: \begin{center}
933: \includegraphics[scale=0.43,angle=-90]{fig2c.eps}
934: \end{center}
935: \caption{}
936: \end{figure}
937: %
938: %
939: %
940: \begin{figure}
941: \vspace{-0cm}
942: \begin{center}\includegraphics[scale=0.43,angle=-90]{fig3a.eps}\end{center}
943: \begin{center}\includegraphics[scale=0.43,angle=-90]{fig3b.eps}\end{center}
944: \begin{center}\includegraphics[scale=0.43,angle=-90]{fig3c.eps}\end{center}
945: \caption{}
946: \end{figure}
947: %
948: %
949: 
950: 
951: %
952: \begin{figure}
953: \vspace{-0cm}
954: \begin{center}
955: \includegraphics[scale=0.43,angle=-90]{fig4a.eps}
956: \end{center}
957: \begin{center}
958: \includegraphics[scale=0.43,angle=-90]{fig4b.eps}
959: \end{center}
960: \begin{center}
961: \includegraphics[scale=0.43,angle=-90]{fig4c.eps}
962: \end{center}
963: \caption{}
964: \end{figure}
965: %
966: %
967: %
968: \begin{figure}
969: \vspace{-0cm}
970: \begin{center}
971: \includegraphics[scale=0.45,angle=-90]{fig5a.eps}
972: \end{center}
973: \begin{center}
974: \includegraphics[scale=0.45,angle=-90]{fig5b.eps}
975: \end{center}
976: \caption{}
977: \end{figure}
978: %
979: %
980: %
981: 
982: \begin{figure}
983: \vspace{-0cm}
984: \begin{center}
985: \includegraphics[scale=0.45,angle=-90]{fig6a.eps}
986: \end{center}
987: \begin{center}
988: \includegraphics[scale=0.45,angle=-90]{fig6b.eps}
989: \end{center}
990: \caption{}
991: \end{figure}
992: %
993: 
994: %
995: \begin{figure}
996: \vspace{-0cm}
997: \begin{center}
998: \includegraphics[scale=0.5,angle=-90]{fig7.eps}
999: \end{center}
1000: \caption{}
1001: \end{figure}
1002: %
1003: %
1004: 
1005: 
1006: %
1007: 
1008: 
1009: 
1010: 
1011: \end{document}
1012: 
1013: