气溶胶,流体中速度场,速度矢量场检测方案(粒子图像测速)

智能文字提取功能测试中

Nonlinear Processes in Geophysics (2004) 11： 343-350SRef-ID： 1607-7946/npg/2004-11-343 A. Eidelman et al.： Turbulent Thermal Diffusion344 Correspondence to： T. Elperin Nonlinear Processesin Geophysics C European Geosciences Union 2004 Turbulent thermal diffusion of aerosols in geophysics and inlaboratory experiments A. Eidelman，T. Elperinl，N. Kleeorinl， A. Krein， I.Rogachevskiil， J.Buchholz， and G. Gruinefeld? The Pearlstone Center for Aeronautical Engineering Studies， Department of Mechanical Engineering， The Ben-GurionUniversity of the Negev， POB 653， Beer-Sheva 84105， Israel Faculty of Mechanical Engineering， RWTH Aachen University， Aachen， Germany Received： 10 November 2003 -Revised： 19 May 2004-Accepted： 16 June 2004 - Published： 27 July 2004 Abstract. We discuss a new phenomenon of turbulent ther-mal diffusion associated with turbulent transport of aerosolsin the atmosphere and in laboratory experiments. Theessence of this phenomenon is the appearance of a nondif-fusive mean flux of particles in the direction of the meanheat flux， which results in the formation of large-scale in-homogeneities in the spatial distribution of aerosols that ac-cumulate in regions of minimum mean temperature of thesurrounding fluid. This effect of turbulent thermal diffusionwas detected experimentally. In experiments turbulence wasgenerated by two oscillating grids in two directions of the im-posed vertical mean temperature gradient. We used ParticleImage Velocimetry to determine the turbulent velocity field，and an Image Processing Technique based on an analysis ofthe intensity of Mie scattering to determine the spatial dis-tribution of aerosols. Analysis of the intensity of laser lightMie scattering by aerosols showed that aerosols accumulatein the vicinity of the minimum mean temperature due to theeffect of turbulent thermal diffusion. Introduction Aerosols are a universal feature of the Earth’s atmosphere.They can significantly affect the heat balance and dynam-ics of the atmosphere， climate， atmospheric chemistry， radia-tive transport and precipitation formation (see， e.g. Twomey，1977； Seinfeld， 1986； Flagan and Seinfeld， 1988； Pruppacherand Klett 1997； Lohmann and Lesins， 2002； Kaufman et al.2002； Anderson et al. 2003； and references therein). For-mation of aerosol clouds is of fundamental significance inmany areas of environmental sciences， physics of the atmo-sphere and meteorology (see， e.g. Seinfeld， 1986； Flagan andSeinfeld， 1988； Paluch and Baumgardner， 1989； Shaw et al.1998； Shaw 2003； and references therein). It is well knownthat turbulence causes decay of inhomogeneities in spatial ( (elperin@menix.bgu.ac.il) ) distribution of aerosols due to turbulent diffusion (see， e.g.Csanady， 1980； McComb 1990； Stock1996)， whereas theopposite effect， the preferential concentration of aerosols inatmospheric turbulence is not well understood. Elperin et al. (1996，1997，1998，2000b， 2001) recentlyfound that in a turbulent fluid flow with a nonzero meantemperature gradient an additional mean flux of aerosols ap-pears in the direction opposite to that of the mean tempera-ture gradient， which is known as the phenomenon of turbu-lent thermal diffusion. This effect results in the formation oflarge-scale inhomogeneities in the spatial distribution of theaerosol particles. The phenomenon of turbulent thermal diffusion is impor-tant for understanding atmospheric phenomena (e.g. atmo-spheric aerosols， smog formation， etc.). There exists a corre-lation between the appearance of temperature inversions andthe aerosol layers (pollutants) in the vicinity of the tempera-ture inversions (see， e.g. Seinfeld， 1986； Flagan and Seinfeld，1988). Moreover， turbulent thermal diffusion can cause theformation of large-scale aerosol layers in the vicinity of tem-perature inversions in atmospheric turbulence (Elperin et al.2000a，2000c). The main goal of this paper is to describe the experimentaldetection of a new phenomenon of turbulent thermal diffu-sion. To accomplish this， we constructed an oscillating gridsturbulence generator (for details see Eidelman et al.， 2002).Recent studies by De Silva and Fernando (1994)； Srdic etal. (1996)； Shy et al. (1997) have demonstrated the feasibilityof generating nearly isotropic turbulence by two oscillatinggrids. Turbulent diffusion in oscillating grids turbulence wasinvestigated by Ott and Mann (2000). In order to study tur-bulent thermal diffusion we used Particle Image Velocimetryto characterize a turbulent velocity field， and an Image Pro-cessing Technique based on Mie scattering to determine thespatial distribution of the particles. Our experiments wereperformed in two directions of the vertical mean tempera-ture gradient： an upward mean temperature gradient (formedby a cold bottom and a hot top wall of the chamber) anda downward mean temperature gradient. We found that in a flow with a downward mean temperature gradient， parti-cles accumulate in the vicinity of the top wall of the chamber(whereby the mean fluid temperature is minimal)， and in aflow with an upward mean temperature gradient particles ac-cumulate in the vicinity of the bottom wall of the chamberdue to the effect of turbulent thermal diffusion. The paper is organized as follows. Section 2 discusses thephysics of turbulent thermal diffusion， and Sect. 3 describesthe experimental set-up for a laboratory study of this effect.The experimental results are presented in Sect. 4， and a de-tailed analysis of experimental detection of turbulent thermaldiffusion is performed in Sect. 5. Finally， conclusions aredrawn in Sect. 6. 2 Turbulent thermal diffusion Evolution of the number density n(t， r) of small particles ina turbulent flow is determined by where JM is the molecular flux of the particles and vp is thevelocity of the particles in the turbulent fluid velocity field.Averaging Eq.(1) over the statistics of the turbulent veloc-ity field yields the following equation for the mean numberdensity of particles N=(n)： where Dr=(t/3)(u) is the turbulent diffusion coefficient，T is the momentum relaxation time of the turbulent velocityfield，vp=V，+u， Vp=(vp) is the mean particle velocity， andthe effective velocity is Here J=-D(VN+kVT/T) is the mean molecular fluxof particles， D is the coefficient of molecular diffusion， k，is the thermal diffusion ratio， and T=(T) is the mean fluidtemperature. Equations (2) and (3) were previously derivedby different methods (see Elperin et al. 1996， 1997， 1998，2000b，2001；Pandya and Mashayek 2002). For noninertial particles advected by a turbulent fluidflow， particle velocity Vp coincides with fluid velocity v，and V.v~-(v.V)p/p~(v.V)T/T， where p and T are thedensity and temperature of the fluid.Thus， the effec-tive velocity (4) for noninertial particles can be given byVerr=-D，(VT)/T， which takes into account the equationof state for an ideal gas but neglects small gradients of themean fluid pressure. For inertial particles， velocity vp depends on the velocityof the surrounding fluid v. Velocity Vp can be determinedby the equation of motion for a particle. Solving the equa-tion of motion for a small solid particle with pp>p yields：vp=v-tpdv/dt+O(t，) (see Maxey， 1987)， where tp is the Stokes time and pp is the material density of the particles. Inthat case and the effective velocity of the inertial particles can be givenby Verr=-D，(1+x)(VT)/T， where P is the fluid pressure.Coefficient k depends on particle inertia (mp/mu)， the pa-rameters ofturbulence (Reynolds number) and the mean fluidtemperature (Elperin et al.， 1996，1997，1998，2000b，2001).Here mp is the particle mass and mp is the mass of moleculesof the surrounding fluid. The turbulent flux of particles (3)can be rewritten as where k=(1+k)N. The first term in the RHS of Eq. (6)describes turbulent thermal diffusion， while the second termin the turbulent flux of particles (6) describes turbulent diffu-sion. Parameter k， can be interpreted as the turbulent ther-mal diffusion ratio， and Drk， is the coefficient of turbulentthermal diffusion. Turbulent thermal diffusion causes the for-mation of a large-scale pattern wherein the initial spatial dis-tribution of particles in a turbulent fluid flow evolves intoa large-scale inhomogeneous distribution， i.e. particles ac-cumulate in the vicinity of the minimum mean temperatureof the surrounding fluid (Elperin et al.， 1996， 1997， 1998，2000b，2001). The mechanism responsible for the occurrence of turbu-lent thermal diffusion for particles with pp>p can be de-scribed as follows. Inertia causes particles inside the tur-bulent eddies to drift out to the boundary regions betweenthe eddies (i.e. regions with low vorticity or high strainrate and maximum of fluid pressure). Thus， particles ac-cumulate in regions with maximum pressure of the turbu-lent fluid. For simplicity， let us consider a pure inertial ef-fect， i.e. we assume that V.v=0. This inertial effect resultsin V·vpot，AP0. On the other hand， Eq. (1) for largePeclet numbers yields V.vpo-dn/dt. The latter formulaimplies that dn/dto-tpAP， i.e. inertial particles accumu-late(dn/dt>0) in regions with maximum pressure of the tur-bulent fluid (where △P<0). Similarly， there is an outflow ofparticles from regions with minimum pressure of fluid. Inhomogeneous and isotropic turbulence without large-scaleexternal gradients of temperature， a drift from regions withincreased or decreased concentration of particles by a turbu-lent flow of fluid is equiprobable in all directions， and pres-sure and temperature of the surrounding fluid do not correlatewith the turbulent velocity field. Thus only turbulent diffu-sion of particles takes place. In a turbulent fluid flow with a mean temperature gradient，the mean heat flux (u@) is not zero， i.e. the fluctuations offluid temperature Q=T-T and the velocity of the fluid cor-relate. Fluctuations of temperature cause fluctuations of fluidpressure. These fluctuations result in fluctuations of the num-ber density of particles. Indeed， an increase in pressure of the Fig.1. The experimental set-up. surrounding fluid is accompanied by an accumulation of par-ticles. Therefore， the direction of the mean flux of particlescoincides with that of the heat flux， (vpn)x(u@)x-VT， i.e.the mean flux of particles is directed to the area with mini-mum mean temperature， and the particles accumulate in thatregion. 3 Experimental set-up In this Section we investigate experimentally the effect of tur-bulent thermal diffusion. The experiments were conductedin an oscillating grids turbulence generator in air flow (seeFig. 1). The test section consisted of a rectangular cham-ber of dimensions 29×29×58 cm (see Fig.2). Pairs of ver-tically oriented grids with bars arranged in a square arraywere attached to the right and left horizontal rods.s. Bothgrids were driven independently with speed-controlled mo-tors.S.TThe grids were positioned at a distance of two-gridmeshes from the chamber walls parallel to them. A two-gridsystem can oscillate at a controllable frequency up to 20 Hz.The grid stroke was adjusted within a range of 1 to 10 cm. A vertical mean temperature gradient in the turbulent flowwas formed by attaching two aluminium heat exchangers tothe bottom and top walls of the test section (see Fig. 2). Theexperiments were performed in two directions of the meantemperature gradient： an upward mean temperature gradi-ent (a cold bottom and a hot top wall of the chamber) and adownward mean temperature gradient (a heated bottom anda cold top wall of the chamber). In order to improve theheat transfer in the boundary layers at the walls we used aheat exchanger with rectangular pins 3×3×15mm. This al-lowed us to support a mean temperature gradient in the coreof the flow up to 200 K/m in the downward direction and up Fig.2. The test section of the oscillating grids turbulence generator. to 110 K/m in the upward direction at a mean temperatureof about 300 K. The temperature was measured with a high-frequency response thermocouple. The velocity field was measured using a Particle Im-age Velocimetry (PIV)， see Raffel et al. (1998). Atal PIV system with LaVision Flow Master III was used.A double-pulsed light sheet was provided by a Nd-YAGlaser source (Continuum Surelite 2×170mJ). The light sheetoptics includes spherical and cylindrical Galilei telescopeswith tuneable divergence and adjustable focus length.Weused a progressive-scan 12 bit digital CCD camera (pixelsize 6.7 umx6.7 umeach) with a dual-frame-technique forcross-correlation processing of captured images. A pro-grammable Timing Unit (PC interface card) generated se-quences of pulses to control the laser， camera and data ac-quisition rate. The software package DaVis 6 was appliedto control all hardware components and for 32 bit imageacquisition and visualization. This software package con-tains PIV software for calculating the flow fields using cross-correlation analysis. Velocity maps and their characteris-tics， e.g. statistics and PDF， were analyzed with this pack-age plus an additional developed software package. An in-cense smoke with sub-micron particles (with pp/p~10) asa tracer was used for the PIV measurements. Smoke wasproduced by high temperature sublimation of solid incenseparticles. Analysis of smoke particles using a microscope(Nikon， Epiphot with an amplification of 560) and a PM-300portable laser particulate analyzer showed that these parti-cles have a spherical shape and that their mean diameter is0.7 um. Mean and r.m.sS.velocities， two-point correlation func-tions and an integral scale of turbulence from the measuredvelocity fields were determined. A series of 100 pairs ofimages acquired with a frequency of 4 Hz were stored forcalculating the velocity maps and for ensemble and spa-tial averaging of turbulence characteristics.. The center ofthe measurement region coincides with the center of thechamber. We measured the velocity for flow areas from60x60mm²up to 212×212 mm² with a spatial resolution of1024×1024 pixels. This size of the probed area corresponds Fig. 3. Vertical temperature profiles at frequency of grid oscilla-tions f=10.5 Hz in turbulent flows with an upward mean tempera-ture gradient (filled squares) and with a downward mean tempera-ture gradient (unfilled triangles). Here Z is a dimensionless verticalcoordinate measured in units of the height of the chamber， and Z=0at the bottom of the chamber. to a spatial resolution from 58 um/pixel up to 207 um/pixel.These regions were analyzed with interrogation windows of32×32 and 16×16pixels. A velocity vector was determinedin every interrogation window， allowing us to construct a ve-locity map comprising 32×32 or 64×64 vectors. The ve-locity maps were determined in two planes. In the one-gridexperiments the plane was parallel to the grid， while in thetwo-grid experiments the plane was normal to the grids. Themean and r.m.s. velocities for each point of the velocity map(1024 points) were determined by averaging over 100 inde-pendent maps， and then over 1024 points.. The two-pointcorrelation functions of the velocity field were determinedfor each point of the central part of the velocity map (16x16vectors) by averaging over 100 independent velocity maps，and then over 256 points. An integral scale L of turbulencewas determined from the two-point correlation functions ofthe velocity field. These measurements were repeated fordifferent distances from the grid， and for various tempera-ture gradients， Reynolds numbers and particle mass load-ings. The PIV measurements performed in the oscillatinggrids turbulence generator confirmed earlier results (Thomp-son and Turner 1975； Hopfinger and Toly 1976； Kit et al.，1997) for the dependence of varies characteristics of the tur-bulent velocity field on the parameters of the oscillating gridsturbulence generator. To determine the spatial distribution of the particles， weused an Image Processing Technique based on Mie scatter-ing. Theadvantages ofthis method have been demonstrated Fig. 4. Ratios E|E of normalized average distributions of the in-tensity of scattered light versus the normalized vertical coordinate Zin turbulent flows with an upward mean temperature gradient (filledsquares) and with a downward mean temperature gradient (unfilledtriangles). Here Z=0 is at the bottom of the chamber. The fre-quency of grids oscillations is f=10.5 Hz. by Guibert et al. (2001). The light radiation energy fluxscattered by small particles is E，xE w(ndp/入；a；n)， whereEond/4 is the energy flux incident at the particle， dp is theparticle diameter， A is the wavelength， a is the index of re-fraction and is the scattering function. In the general case， is given by Mie equations. For wavelengths A， which aresmaller than the particle size， tends to be independent ofdp and A. The scattered light energy flux incident on the CCDcamera probe is proportional to the particle number densityn，i.e. E，xEn(nd/4). Mie scattering does not change from temperature effectssince it depends on the permittivity of particles， the particlesize and the laser light wavelength. The effect of the temper-ature on these characteristics is negligibly small. Note that ineach experiment before taking the measurements， we let thesystem run for some time (up to 30 min after smoke injectioninto the flow with a steady mean temperature profile) in orderto attain a stationary state. We found that the probability density function of the par-ticle size measured with the PM300 particulate analyzer wasto the most extent independent of the location in the flow forincense particle size of 0.5-1 um. Note that since the num-ber density of the particles is small (about 1 mm apart)， itcan be assumed that a change in particle number density willnot affect their size distribution. Therefore， the ratio of thescattered radiation fluxes at two points and at the image mea-sured with the CCD camera remains equal to the ratio of theparticle mean number densities at these two locations. Fig. 5. Normalized particle number density Nr=N/No versus nor-malized temperature difference Tr=(T-To)/To in a turbulent flowwith an upward mean temperature gradient. The frequency of gridoscillations is f=10.5 Hz. Experimental detection of turbulent thermal diffu-sion The turbulent flow parameters in the oscillating grids tur-bulence generator are： r.m.s.velocity√(u²)=4-14cm/sdepending on the frequency of the grid oscillations， inte-gral scale of turbulence L=1.6-2.3 cm， and the Kolmogorovlength scale n=380-600 um. Other parameters are given inSect. 5 (Tables 1-2). These flows involve a wide range ofscales of turbulent motions. Interestingly， a flow with a widerange of spatial scales is already formed at comparatively lowfrequencies of grid oscillations. We found a weak mean flowin the form of two large toroidal structures parallel and adja-cent to the grids. The interaction of these structures results ina symmetric mean flow that is sensitive to the parameters ofthe grid adjustments. We particularly studied the parametersthat affect a mean flow such as the grid distance to the wallsof the chamber and partitions， and the angles between thegrid planes and the axes of their oscillations. Varying theseparameters allowed us to expand the central region with ahomogeneous turbulence. We found that the measured r.m.s.velocity was several times higher than the mean velocity inthe core of the flow. The temperature measurements were taken in the oscillat-ing grids turbulence generator. The temperature differencebetween the heat exchangers varied within a range of 25 to50 K. The mean temperature vertical profiles at a frequencyof grids oscillations f=10.5 Hz in turbulent flows with down-ward and upward mean temperature gradients are shown inFig. 3. Here Z is a dimensionless vertical coordinate mea- Fig. 6. Normalized particle number density Nr=N/No versus nor-malized temperature difference Tr=(T-To)/To in a turbulent flowwith a downward mean temperature gradient. The frequency of gridoscillations is f=10.4 Hz. sured in units of the height of the chamber， and Z=0 at thebottom of the chamber. Hereafter， we use the following sys-tem ofcoordinates： Z is the vertical axis， the Y-axis is per-pendicular to the grids and the XZ-plane is parallel to thegrids plane. Measurements performed using different concentrationsof incense smoke showed that in an isothermal flow the distri-bution of the average scattered light intensity over a verticalcoordinate is independent of the mean particle number den-sity. In order to characterize the spatial distribution of parti-cle number density NoE /E in a non-isothermal flow， thedistribution of the scattered light intensity E for the isother-mal case was used to normalize the scattered light intensityE obtained in a non-isothermal flow under the same condi-tions. The scattered light intensities E and E in each ex-periment were normalized by corresponding scattered lightintensities averaged over the vertical coordinate. The ratiosE/E of the normalized average distributions of the inten-sity of scattered light as a function of the normalized verti-cal coordinate Z in turbulent flows with a downward meantemperature gradient and with an upward mean temperaturegradient are shown in Fig.4. The figure demonstrates thatparticles are redistributed in a turbulent flow with a meantemperature gradient. Particles accumulate in regions of min-imum mean temperature (in the lower part of the chamber inturbulent flows with an upward mean temperature gradient)，and there is an outflow of particles from the upper part of thechamber where the mean temperature is larger. On the otherhand， in a flow with a downward mean temperature gradient(a hot bottom and a cold top wall of the test section) particles Table 1. Parameters of turbulence and the turbulent thermal dif-fusion coefficient for a turbulent flow with an upward mean tem-perature gradient. Here Re=L√(u2)/v， L is the integral scale ofturbulence， and v is the kinematic viscosity. accumulate in the vicinity of the top wall of the chamber， i.e.in the vicinity where the mean fluid temperature is minimum. To determine the turbulent thermal diffusion ratio， we plot-ted the normalized particle number density Nr=N/No ver-sus the normalized temperature difference Tr=(T-To)/To，where To is the reference t10/0，emperature and No=N(T=To).The function N，(Tr) is shown in Fig. 5 for a turbulent flowwith an upward mean temperature gradient and in Fig. 6 for aturbulent flow with a downward mean temperature gradient. We probed the central 20×20 cm region in the chamber bydetermining the mean intensity of scattered light in 32×16interrogation windows with a size of 32×64 pixels. The ver-tical distribution of the intensity of the scattered light wasdetermined in 16 vertical strips， which are composed of 32interrogation windows.Variations of the obtained verticaldistributions between these strips were very small. Thereforein Figs. 4-6 we used spatial average over strips and the en-semble average over 100 images of the vertical distributionsof the intensity of scattered light. Figures 5-6 were plottedusing the mean temperature vertical profiles shown in Fig. 3.The normalized local mean temperatures [the relative tem-perature differences (T-To)/To] in Figs. 5-6 correspond tothe different points inside the probed region. In particular，in Fig. 5 the location of the point with reference temperatureTo is Z=0.135 (the lowest point of the probed region with amaximum N in turbulent flows with a downward mean tem-perature gradient)， and the point with a maximum normalizedtemperature difference is the highest point of the probed re-gion atZ=0.865. The size ofthe probed region did not affectour results. 5 Discussion The vertical profiles of the mean number density of parti-cles obtained in our experiments can be explained by con-sidering Eqs. (2) and (3). If one does not take into accountthe term NVeff in Eq. (2) for the mean number density ofparticles， the equation reads 0N/0t=D，AN， where we ne-glected a small mean fluid velocity V， and a small molecularmean flux of particles (that corresponds to the conditions ofthe experiment). The steady-state solution of this equationis N=const. However， our measurements demonstrate thatthe solution N=const is valid only for an isothermal turbu- Table 2. Parameters of turbulence and the turbulent thermal diffu-sion coefficient for a turbulent flow with a downward mean temper-ature gradient. f(Hz) V(u2) (cm/s) L (cm) Re 6.5 3.6 1.9 46 1.79 10.5 7.2 2.1 101 1.65 14.5 10.7 2.3 164 1.43 16.5 12.4 2.0 165 1.33 lent flow， i.e. when T=const. By taking into account theeffect of turbulent thermal diffusion， i.e. the term NVerr inEq. (2)， the steady-state solution of this equation for non-inertial particles is： VN/N=-VT/T. The latter equationyields N/No=1-(T-To)/To， where T-T01 for inertialparticles. As can be seen in Tables 1-2， parameter a slightlydecreases when the Reynolds number increases. There are other factors that can affect the spatial distribu-tion of particles. The contribution of the mean flow to thespatial distribution of particles is negligibly small. Indeed，the normalized distribution of the scattered light intensitymeasured in the different vertical strips in the regions wherethe mean flow velocity and the coefficient of turbulent dif-fusion vary strongly are practically identical (the differencebeing only about 1%). Due to the effect of turbulent thermaldiffusion， particles are redistributed in the vertical directionin the chamber. In turbulent flows with an upward mean tem-perature gradient particles accumulated in the lower part ofthe chamber， and in flows with a downward mean temper-ature gradient particles accumulated in the vicinity of thetop wall of the chamber， i.e. in regions with a minimummean temperature. The spatial-temporal evolution of the nor-malized number density of particles N， is governed by the conservation law of the total number of particles (see Eqs. 2and 3). Some fraction of particles sticks to the walls of thechamber， and the total number of particles without feedingfresh smoke slowly decreases. The characteristic time of thisdecrease is about 15 min. However， the spatial distribution ofthe normalized number density of particles does not changeover time. The number density of particles in our experiments wasof the order of 1010 particles per cubic meter. Therefore，the distance between particles is about 1 mm， and their colli-sion rate is negligibly small. Indeed，calculation of the parti-cle collision rate using the Saffman-Turner formula (Saffmanand Turner， 1956) confirmed this finding. Consequently， wedid not observe any coalescence of particles. The effect ofthe gravitational settling of small particles (0.5-1 um) is neg-ligibly small (the terminal fall velocity of these particles be-ing less than 0.01 cm/s). Thus， we may conclude that the twocompetitive mechanisms of particle transport， i.e. mixing byturbulent diffusion and accumulation of particles due to tur-bulent thermal diffusion， exist simultaneously and there is avery small effect of gravitational settling of the particles. In regard to the accuracy of the performed measurements，we found that uncertainties in optics adjustment and errorsin measuring a CCD image background value are consider-ably less than the observed effect of a 10% change of nor-malized intensity of the scattered light in the test sectionin the presence of an imposed mean temperature gradient.In the range of the tracer concentrations used in the exper-iments the particle-air suspension can be considered as anoptically thin medium， from which we can infer that the in-tensity of the scattered light is proportional to the particlenumber density. The total error in our measurements was de-termined by (8N/N+8T/T)/√g~0.3%(see Eq. 7)， where8T=0.1 K is the accuracy of the temperature measurements；8N/N=0.8% is the accuracy of the mean number densitymeasurements； and Q=8 is the number of experiments per-formed for each direction of the mean temperature gradientand for each value of the frequency of the grid oscillations.The total variation of the normalized particle number densitydue to the effect of turbulent thermal diffusion is more than10% (see Figs. 4-6). The relative error of the measurementsis less than 3% of the total variation of the particle numberdensity. Thus， the accuracy of these measurements is con-siderably higher than the magnitude of the observed effect.As such， our experiments confirm the existence of an effectof turbulent thermal diffusion， as predicted theoretically byElperin et al. (1996，1997). 6 Conclusions A new phenomenon of turbulent thermal diffusion has beendetected experimentally in turbulence generated by oscillat-ing grids with an imposed vertical mean temperature gradientin air flow. This phenomenon implies that there exists an ad-ditional mean flux of particles in the direction opposite to thatof the mean temperature gradient， that results in the forma- tion of large-scale inhomogeneities in the spatial distributionof particles. The particles accumulated in the vicinity of theminimum mean fluid temperature. In the experiments in twodirections of the vertical mean temperature gradient， it wasfound that in a flow with a downward mean temperature gra-dient particles accumulate in the vicinity of the top wall ofthe chamber. In a flow with an upward mean temperaturegradient particles accumulate in the vicinity of the bottomwall of the chamber. Turbulent thermal diffusion can explainthe large-scale aerosol layers that form inside atmospherictemperature inversions. Acknowledgements. We are indebted to F. Busse， H. J. S. Fer-nando， J. Katz， E. Kit， V. L’vov， J. Mann， S. Ott and A. Tsinoberfor illuminating discussions. We also thank A. Markovich for hisassistance in processing the experimental data.This work waspartially supported by the German-Israeli Project Cooperation(DIP) administrated by the Federal Ministry for Education andResearch (BMBF) and by the Israel Science Foundation governedby the Israeli Academy of Science. Edited by： R. Grimshaw Reviewed by： two referees References Anderson，L.， Charlson， R. J.，Schwartz， S. E.， Knutti， R.，Boucher，O.， Rodhe， H.， and Heintzenberg，J.： Climate forcing by aerosols-a hazy picture， Science， 300， 1103-1104， 2003. Csanady， G. T.： Turbulent Diffusion in the Environment， Reidel，Dordrecht， 1980. De Silva， I. P. D. and Fernando H.J. S.： Oscillating grids as a sourceof nearly isotropic turbulence， Phys. Fluids， 6， 2455-2464，1994. ( Eidelman， A .， E lperin， T.， Kapusta， A.， Kleeorin， N.， Krein， A.，and Rogachevskii， I. ： O s cillating grids tu r bulence gen e rator for turbulent transport studies， Nonlin. Proc. Geophys.， 9，201 - 205，2002. ) Elperin， T.， Kleeorin， N.， and Rogachevskii， I.： Turbulent thermaldiffision of small inertial particles，Phys. Rev. Lett.， 76，224-228，1996. ( Elperin， T.， Kleeorin， N.， and R ogachevskii， I . ： T urbulent barodif- fusion， turbulent t hermal d i ffusion and large-scale instability in gases， Phys. Rev. E， 55， 2713-2721，1 9 97. ) ( Elperin， T.， Kleeorin， N.， and Rogachevskii， I . ： F ormation of inho- mogeneities i n two-phase low-mach-number compressible tur b u-lent flows， Int. J. Multiphase Flow， 24， 1 163-1182，1 9 98. ) ( Elperin， T.， Kleeorin， N.， and Rogachevskii， I.： Mechanisms of fo r - mation of a erosol and gaseous inhomogeneities in the t u rbulentatmosphere， Atmosph. Res.， 5 3， 117- 1 29， 2000a. ) ( Elperin， T.， Kleeorin， N.， Rogachevskii， I . ， and S okoloff， D.： Pas-sive scalar transport in a random flow w i th a finite renewal time： mean-field equations， Phys. Rev.E， 61，2617-2625，2000b. ) ( Elperin， T.， K leeorin， N.， Rogachevskii， I. ， and S o koloff， D.： Tur-bulent transport o f a tmospheric aerosols and formation of l a rge- scale structures， Phys. Chem. Earth， A25， 7 9 7-803， 2000c. ) ( Elperin， T.， Kleeorin， N.， Rogachevskii，I.， and Sokoloff， D.： Me a n-field theory for a passive scalar advected b y a turbulent velocity field with a random renewal t ime， Phys. Rev.E.， 64， 026304(1- 9)，2001. ) Flagan， R. and Seinfeld， J. H.： Fundamentals of Air Pollution Engi-neering， Prentice Hall， Englewood Cliffs，1988. Guibert， P.， Durget， M.， and Murat， M.： Concentration fields ina confined two-gas mixture and engine in cylinder flow： lasertomography measurements by Mie scattering.， Exp. Fluids， 31，630-642，2001. Hopfinger， E. J. and Toly， J.-A.： Spatially decaying turbulence andits relation to mixing across density interfaces， J. Fluid Mech.，78，155-175，1976. ( Kaufman， Y . J . ， T anre， D .， a nd Boucher， O . ： A satellite v iew of.aerosols in the climate system， Nature， 419， 2 15-2 2 3，2002. ) ( Kit， E.， Strang， E. J. ， and Fernando， H. J. S.： Measurement of tur- bulence n ear shear-free density interfaces， J . Flui d Mech.， 334， 293-314， 1 997. ) ( Lohmann， U . and Lesins， G . ： Stronger constraints on the a nthro- pogenic inderect aerosol effect， Science ， 298 ， 1012-1015，2002 ) ( Maxey， M . R.： The gr a vitational settling of aerosol particles in ho-mogeneous turbulence a n d random flow field， J. F l uid M e ch.，174，441-465，1987. ) ( McComb， W.D.： Th e Physics of Fluid Turbulence， Clarendon， Ox-ford， 1990. ) ( Ott， S . and Mann， J.： An experimental inve s tigation of t h e relativediffusion of particle pairs in three-dimensional tur b ulent flow， J. Fluid Mech.， 422，207-234，2000. ) ( Pandya， R. V. R . and Mashayek， F.： Turbulent thermal diffusion and barodiffusion of passive scalar and dispersed phase of particles i n turbulent flows， P hys. Rev. Lett.， 8 8， 044501 (1-4) ， 2002. ) ( Paluch， I. R. and Baumgardner， D. G.： E ntrainment and fine-scalemixing in a continental convective cloud， J. Atmos. Sci.， 46，261-278， 1 989. ) Pruppacher， H. R. and Klett， J. D.： Microphysics of Clouds andPrecipitation， Kluwer Acad. Publ.，Dordrecht， 1997. Raffel， M.， Willert， C.， and Kompenhans， J.： Particle Image Ve-locimetry， Springer， 1998. Saffman， P. G. and Turner，J. S.： On the collision of drops in turbu-lent clouds，J. Fluid Mech.， 1， 16-30， 1956. Seinfeld， J. H.： Atmospheric Chemistry and Physics of Air Pollu-tion， John Wiley， New York， 1986. Shaw， R. A.， Reade， W. C.， Collins L. R.， and Verlinder J.： Prefer-ential concentration of cloud droplets by turbulence： effects onthe early evolution of cumulus cloud droplet spectra， J. Atmosph.Sci.， 55， 1965-1976， 1998. Shaw，R. A.： Particle-turbulence interactions in atmospheric clouds，Ann. Rev. Fluid Mech， 35，pp. 183-227，2003. Shy， S. S.， Tang， C. Y.， and Fann， S. Y.： A nearly isotropic turbu-lence generated by a pair of vibrating grids， Exp. Thermal FluidSci.， 14，251-262，1997. Srdic， A.， Fernando， H. J. S.， and Montenegro， L.： Generation ofnearly isotropic turbulence using two oscillating grids， Exper. inFluids， 20， 395-397， 1996. Stock， D.： Particle dispersion in flowing gases， ASME J. FluidsEngineering， 118，4-17，1996. Telford， J. W.： Clouds with turbulence： the role of entrainment，Atmos. Res.，40，261-282， 1996. Thompson， S. M. and Turner， J. S.： Mixing across an interface dueto turbulence generated by an oscillating grid， J. Fluid Mech.，67，3'549-368，1975. Twomey， S.： Atmospheric Aerosols， Elsevier， Amsterdam， 1977.