United States                EPA-600/8-81 -009
               Environmental Protection           April 1981
               Agency
vvEPA        Research  and
               Development
               Guideline for
               Fluid Modeling of
               Atmospheric Diffusion
               Prepared for

               Office of Air Quality
               Planning and Standards
               Prepared  by

               Environmental Sciences Research
               Laboratory
               Research Triangle Park NC 27711

-------
                                             600/8-81-009
                                             April 1981
       GUIDELINE FOR FLUID MODELING
         OF ATMOSPHERIC DIFFUSION
                    by
             William H.  Snyder

    Meteorology and Assessment Division
 Environmental  Sciences  Research Laboratory
    U.S.  Environmental  Protection Agency
     Research Triangle  Park,  NC  27711
ENVIRONMENTAL SCIENCES RESEARCH LABORATORY
    OFFICE OF RESEARCH AND DEVELOPMENT
   U.S. ENVIRONMENTAL PROTECTION AGENCY
    RESEARCH TRIANGLE PARK, NC  27711
      us  Environmental Protection Agency.


      Slo^Sor
       ChLgo, Illinois  60604

-------
                                 DISCLAIMER

This report has been reviewed by the Environmental Sciences Research
Laboratory, U.S. Environmental Protection Agency, and approved for
publication.  Approval does not signify that the contents necessarily reflect
the views and policies of the U.S.  Environmental Protection Agency, nor does
mention of trade names or commercial products constitute endorsement or
recommendation for use.
     The author, William H. Snyder, is a physical  scientist in the Meteorology

and Assessment Division, Environmental Sciences Research Laboratory, U.S.

Environmental Protection Agency, Research Triangle Park, North Carolina.

He is on assignment from the National Oceanic and Atmospheric Administration,

U.S. Department of Commerce.
                                      ii
               U,S. Environmental Protection Agency

-------
                                PREFACE

     The U.S. Environmental  Protection Agency is charged by Congress with
establishing and enforcing air pollution control standards to protect the
public health and welfare.  To accomplish its mission, it is essential  to
be able to describe and predict the transport and diffusion of pollutants
in the atmosphere.  Present mathematical models are not yet adequate for
calculating concentrations of contaminants when the plume is affected by
obstructions such as hills and buildings.  Field programs to obtain adequate
data are very expensive and time consuming.  Small scale models immersed
in the flow of wind tunnels and water channels, i.e., fluid models, can fre-
quently be of use in simulating atmospheric transport and diffusion in a
timely and relatively inexpensive manner.
     It is the aim of this guideline to point out the capabilities and
limitations of fluid modeling and to recommend standards to be followed in
the conduct of such studies.  The guideline is intended to be of use both
to scientists and engineers involved in operating fluid modeling facilities
and to air pollution control officials in evaluating the quality and cred-
ibility of the reports resulting from such studies.
     The fundamental principles of fluid modeling are well-established, but
when decisions must be made concerning a particular model study, the fundamen-
tal principles frequently do not provide specific guidance.  There is a need
for basic and systematic modeling studies to provide more specific guidance.
This guideline will be periodically revised as more specific experience is
gained, new techniques are developed, and old ones refined.

-------
                                 TABLE OF CONTENTS

                                                                          Page
PREFACE	     iii
LIST OF FIGURES AND  TABLES	     vii
NOMENCLATURE	       x
ACKNOWLEDGEMENTS	     x1v

     1.  INTRODUCTION	       1

     2.  FUNDAMENTAL PRINCIPLES 	       3
         2.1  The Equations of Motion 	       3
         2.2  The Dimension!ess Parameters	       7
              2.2.1   The Rossby Number	       8
              2.2.2   The Reynolds Number	      14
                     2.2.2.1  The Laminar Flow Analogy	      14
                     2.2.2.2  Reynolds Number Independence	      17
                     2.2.2.3  Dissipation Scaling 	      27
              2.2.3   The Peclet Number and the Reynolds-Schmidt Product      29
              2.2.4   The Froude Number	      32
         2.3  Boundary Conditions 	      35
              2.3.1   General	      35
              2.3.2   Jensen's Criterion and Fully Rough Flow	      37
              2.3.3   Other Boundary Conditions	      39
         2.4  Summary and Recommendations 	      40

    3.  PRACTICAL APPLICATIONS   	       42
         3.1  Plume  Rise and Diffusion	      42
              3.1.1   Near-Field Plume Behavior	      44
              3.1.2   Summary and Recommendations on Modeling Near-Field
                     Plumes	      63
                     3.1.2.1  The Stack Downwash Problem	      63
                     3.1.2.2  The Near-Field Non-downwashed Plume
                              Problem	      64
              3.1.3   Far-Field Plume Behavior 	      65
                     3.1.3.1  Ignoring the Minimum Reynolds Number.  .  .      65
                     3.1.3.2  Raising the Stack Height	      67
                     3.1.3.3  Distorting the Stack Diameter 	      67
              3.1.4   Summary and Recommendations on Modeling Far-
                     Field Plumes	      70
         3.2  The Atmospheric Boundary Layer	      72
              3.2.1   Characteristics of the Atmospheric Boundary
                     Layer	      74
                     3.2.1.1  The Adiabatic Boundary Layer	      75
                     3.2.1.2  Summary of the Adiabatic Boundary
                              Layer Structure 	      87
                     3.2.1.3  The Dlabatic Boundary Layer 	      89
                     3.2.1.4  Summary of the Dlabatic Boundary
                              Layer Structure 	     112

-------
          3.2.2  Simulating the Ad1abat1c Boundary Layer	     114
          3.2.3  Simulating the D1abat1c Boundary Layer 	     127
          3.2.4  Summary on Simulating the Atmospheric Boundary
                 Layer	     129
     3.3  Flow Around Buildings 	     132
          3,3.1  Discussion	     132
          3.3.2  Recommendations	     140
     3.4  Flow Over HUly Terrain	     141
          3.4.1  Neutral Flow	     141
          3.4.2  Stratified Flow	     144
          3.4.3  Recommendations	     147
     3.5  Relating Measurements to the Field	     148
     3.6  Averaging Times and Sampling Rates in the Laboratory.  .  .     151

4.  THE HARDWARE	      156
     4.1  General Requirements	     156
          4.1.1  The Speed Range and Scale Reductions  	     156
          4.1.2  Test Section Dimensions	     159
     4.2  A1r versus Water	     161
          4.2.1  Visual  Observations	     161
          4.2.2  Quantitative Measurements	     163
          4.2.3  Producing Stratification 	     164
          4.2.4  Examples	     166
          4.2.5  Summary	     168

5.  CONCLUDING REMARKS  	     169

6.  REFERENCES	     172
                                 V1

-------
                                LIST OF FIGURES
NUMBER                               TITLE                              PAGE
   1            Schematic of diffusion in the  Ekman  Layer	9
   2            Turbulent jets  illustrating  Reynolds number
                  independence	18
   3            Shadowgraphs of the jets shown in  Figure 2	18
   4            Filter function in Equation  2.12	20
   5            Spectrum of wind speed at 100  m	21
   6            Form of turbulence spectrum	22
   7            Change of spectrum with Reynolds number	23
   8            Plume downwash  in the wake of  a stack	44
   9            Variation of plume rise with Reynolds number	59
  10            Laminar plume caused by low  Reynolds number effluent	66
  11            Effects of wind shear on the flow  round a building	72
  12            The depth of the adiabatic boundary  layer according
                  to the geostrophic drag law  compared with other
                  schemes	77
  13            Typical wind profiles over uniform terrain in neutral
                  flow	82
  14            Variation of power law index,  turbulence intensity,
                  and Reynolds  stress with roughness length in the
                  adiabatic boundary layer	82
  15            Shear stress distributions measured  at various down-
                  wind positions in a wind tunnel  boundary layer	83
  16            Variation of longitudinal  turbulence intensity with
                  height under  adiabatic conditions	85
  17            Variation of Integral length scale with height and
                  roughness length	86

-------
NUMBER                             TITLE                              PAGE
  18          Empirical curves for spectra and cospectrum for
                neutral conditions	87
  19          Typical  nonadiabatic boundary layer depths from
                the geostrophic drag relations	96
  20          Variation of friction velocity with stability from
                the geostrophic drag relations	96
  21          Theoretical  variation of the power-law exponent as
                a function of z  and L for z equal  to 100 m  	98
  22          Variation of the power-law exponent,  averaged over
                layer from 10 to 100m, as a function of surface
                roughness  and Pasquill stability class	99
  23          Typical  surface layer velocity profiles under
                nonadiabatic conditions	102
  24          Typical  temperature profiles in the surface layer	102
  25          The relationship between R1 and z/L	105
  26          Variation of »  and «  with z/L in the surface layer	105
                            W      6
  27          Variation of *  with z/L in the surface layer	107
  28          Variation of *y with z/L	107
  29          Universal spectral shape	108
  30          Location of spectral peak for u,v,w and e plotted
                against z/L	108
  31          Upstream view of a long wind tunnel	116
  32          Vortex generators and roughness in a short wind tunnel..117
  33          Schematic representation of the counter-jet technique...118
  34          Development of boundary layer in a long wind tunnel	119
  35          Development of mean velocity profiles along the smooth
                floor of a long tunnel	121
  36          Thickness parameters for boundary layer of Figure 35	121
                                  viil

-------
NUMBER                            TITLE                              PAGE
  37          Spectrum of the longitudinal component of velocity	123
  38          Velocity profiles above crest of triangular ridge
                i ndi cati ng effect of bl ockage	139
  39          Contour map of three-dimensional  hill showing
                inappropriate choice of area to be modeled	146
  40          Averaging time requirements for wind tunnel
                measurements of turbulent energy	153
  41          Example of limits of wind tunnel  simulation of buoyant
                effluent dispersal in the atmospheric boundary layer..158
                             LIST OF TABLES

NUMBER                            TITLE                              PAGE
   1           Typical Parameters for Modeling Plume Downwash	  54
   2           Techniques used for Simulation of Buoyant Plumes
                 at Various Fluid Modeling Facilities	  62
   3           Values of Surface Roughness Length for Various
                 Types of Surfaces	  80
   4           Typical Values for the Various Stability Parameters	  93
   5           Dimensionless Length Scales as Functions of Ri	109
                                   ix

-------
                              NOMENCLATURE
A      constant or area [L2]
B      constant
c      constant
C      constant or concentration [M/L3]
c,,u,    cospectrum of Reynolds stress [L2/T'4]
 uw
d      zero-plane displacement [L]
D      stack diameter [L]
E      spectrum function [L3/T2]
f      nondimensional frequency or arbitrary function
f      Coriolis parameter (^ 10" /sec in mid-latitudes)
f      nondimensioral frequency corresponding to spectral peak
F      source buoyancy flux = g(D2/4)(Ap/pJ [L1*/!3]
                                          c
F.      Lagrangian spectrum function [T]
Fm     source momentum flux = (ps/Pa)(D2/4)(W$) [L"/T2]
Fr     Froude number
g      acceleration due to gravity [L/T2]
G      geostrophic wind speed [L/T]
h      hill, building or obstacle height [L]
h      roughness element height [L]
H      stack or building height [L]
I      turbulence integral scale [L]
            *  *
k      von Karman constant
K      eddy viscosity or diffusivity [L2/T]
£n     buoyancy length scale [L]
am     momentum length scale [L]
L      characteristic length scale or Monin-Obukhov length [L]

-------
Lu     integral scale of longitudinal velocity in a-direction [L]
Lw     integral scale of vertical velocity in a-direction [L]
  a
n      frequency [T  ]
N      Brunt-Vaisala frequency  =  ((g/p)(dp/dz))1/2 [T"1]
p      pressure [K/LT2] or power-law index
Pe     Peclet number
Q      pollutant emission flow  rate [M/T]
Re     Reynolds number
Ri     Richardson number
Rio    bulk Richardson number
Ri,.    flux Richardson number
Ro     Rossby number
S      spectrum function [L2/T] or scale reduction factor
Sc     Schmidt number
t      time [T]
T      averaging time [T], time of travel from source [T], or temperature [T]
u      fluctuating velocity in x-direction (streamwise) [L/T]
u*     friction velocity [L/T]
U      mean wind speed [L/T]
U.     instantaneous flow velocity in i-direction [L/T]
v      fluctuating velocity in y-direction (cross-streamwise) [L/T]
w      fluctuating velocity in z-direction (vertical) [L/T]
W      effluent speed [L/T]
x      Cartesian coordinate (streamwise) [L]
Xj     coordinate in i-direction [L]
y      Cartesian coordinate (cross-streamwise) or particle displacement [L]
                                     xi

-------
z      Cartesian coordinate (vertical) [L]
z      roughness length [L]
a      molecular mass diffusivity [L2/T]
B      blockage ratio (area)
6      boundary layer depth [L]
6..    Kronecker's delta
 ' J
6P     deviation of pressure from that in neutral atmosphere [M/LT2]
<5T     deviation of temperature from that in neutral atmosphere [T]
Ah     plume rise [L]
Ap      density difference [M/L3]
E      dissipatition [L2/T3] roughness element height [L], or fractional error
eijk   alternating tensor
n      Kolmogoroff microscale [L]
e      potential temperature [T]
K      thermal diffusivity (air:0.21 cm2/s; water:0:0014 cm2/s)
A      resistance coefficient for pipe flow
A      wavelength corresponding to spectral peak [L]
v      stability parameter
v      kinematic viscosity (air:0.15 cm2/s; water:0.01 cm2/s)
£      time separation [T]
p      density (air: 1.3 g/1 ; water:! g/cm3) or autocorrelation function
o      standard deviation
T      fluctuating temperature (deviation from mean) [T]
u      Kolmogoroff velocity [L/T]
*h     nondimensional potential temperature gradient
*u     nondimensional horizontal (u plus v) turbulence intensity
                                    xii

-------
$      nondimensional wind speed gradient
$      nondimensional longitudinal turbulence intensity
$      nondimensional lateral turbulence intensity
$      nondimensional vertical turbulence intensity
 w
*.     nondimensional intensity of temperature fluctuations
 o
x      nondimensional concentration
uj      earth's rotation rate  [T  ]
Subscripts and Special Symbols
( )„   ambient value
   a
( )    equilibrium value
( )-   field value
( )    geostrophic value
( )L   Lagrangian value
( )    model value
( LY  maximum value
   IHA
( )    value of quantity in neutral atmosphere, except as noted
( )    prototype value
( )    reference quantity
( )    stack value
( )    value of quantity in x-direct1on
   ^
( )    value of quantity in y-direction
( )2   value of quantity in z-direction
( )at   free stream value
( )'   nondimensional quantity
IT    average value
( )    vector quantity
                                     xiti

-------
                              ACKNOWLEDGEMENTS

     The author wishes to acknowledge the many helpful  discussions with his
colleagues and, in particular, with Dr.  Ian Castro,  University of Surrey,  Dr.
Rex Britter and Dr. Julian Hunt, University of Cambridge,  Dr.  Igor Nekrasov,
University of Moscow, and Dr.  Gary Briggs, National  Oceanic and Atmospheric
Administration, all of whom spent extended periods of time at  the Fluid Model-
ing Facility.  Mr. Alan Huber's persistent prodding  saw this work to fruition.
Mr. Roger Thompson very willingly engaged in a large number of discussions and
expositions over the course of several  years.  Mr. Robert  Lawson kept the  lab-
oratory running smoothly, allowing me to concentrate on the manuscript. Many
people provided comments on the drafts  dated March or June 1979; especially
thorough or thought-provoking reviews were provided  by Dr. Alan Robins, March-
wood Engineering Laboratories, Central  Electricity Generating  Board, Dr. Robert
Meroney, Colorado State University, Dr.  Gary Ludwig  and Dr. George Skinner,
Calspan Corporation, Dr. James Halitsky, Croton-on-Hudson, NY, Dr. John
Wyngaard, National Center for Atmospheric Research,  and Dr. Frank Binkowski,
National Oceanic and Atmospheric Administration.  Mr. Mike Shipman, Northrop
Services, constructed many of the graphs as well as a library  program that
made the list of references a breeze.  Miss Laurie Lamb, Miss  Tammy Bass and
Ms. Carolyn H. Coleman painstakingly typed various versions of the drafts  and
final manuscript.  Finally, my wife Hazel and children JB and  Jennifer endured
through it all.  To each of these, I express my sincere thanks.
                                     xi v

-------
                               1.  INTRODUCTION

     Most mathematical models of turbulent diffusion in the lower atmospheric
layer tend to ignore the fundamental fluid-dynamical processes involved in
the dispersion of materials.  This results from the fact that the memory size
of the latest computer is far too small to keep track of the large number of
"eddies" in a turbulent flow.  Corrsin (1961a), in speculating on the future
role of large computing machines in following the consequences of the Navier-
Stokes equations under random initial conditions, estimated a required memory
size of ~1013 bits, then asked if "the foregoing estimate is enough to suggest
the use of analog instead of digital computation; in particular, how about an
analog consisting of a tank of water?" (emphasis added).  In spite of the tre-
mendous advances in computer memories in the past two decades, Corrsin's remark
1s still appropriate.
     Fluid models of various aspects of atmospheric motion have been described
in the literature many times.  The necessity of studying the dispersion of at-
mospheric pollutants, especially in urban areas, has further directed thoughts
of meteorologists towards fluid modeling.
     Many factors affect the dispersion of pollutants in the atmosphere; ther-
mal effects, the topography, the rotation of the Earth, etc.  Fluid modeling
studies are desirable mostly because essential  variables can be controlled at
will, and the time and expense are greatly reduced from that required in full-
scale studies.  It is not usual, however, for all the factors influencing

-------
atmospheric dispersion to be included in a model.   Normally,  the similarity
criteria are conflicting in some sense;  it may be  necessary to model  one phys-
ical process at the expense of not being able to model  another.
     For correct modeling, certain nondimensional  parameters  in the prototype
must be duplicated in the model.  Almost invariably, duplication of these non-
dimensional parameters is impractical or impossible.  Hence,  a decision must
be made as to which parameters are dominant.   The  less  important ones must be
ignored.  This decision will generally depend upon the  scale  in which the in-
vestigator is interested.  For example,  when  studying the upper air flow above
a city, the waffle-like topography may be treated  as surface  roughness.  The
heat island effect may be modeled by using a  heated plate.   If the city is
large enough, Coriolis forces may be important.   If, however, the interest is
in dispersion in the immediate vicinity of buildings, the topography cannot be
treated as surface roughness.  The heat-island effect would require a detailed
distribution of heat sources, and Coriolis forces  could be ignored because the
aerodynamic effects of the flow around the buildings would dominate.
     Chapter 2 reviews the fundamental principles  for fluid modeling relevant
to air pollution meteorology and evaluates the usefulness of  such models from
both scientific and engineering viewpoints.  Because many detailed decisions
must be made during the design and execution  of each model  study, and because
the fundamental principles frequently do not  provide enough guidance, discus-
sions of the details of the most common types of modeling problems are provided
in Chapter 3.  Air and water are most commonly used as  media  for the simulation
of atmospheric motions.  The potentials of both of these fluids are reviewed in
Chapter 4.

-------
                         2.  FUNDAMENTAL PRINCIPLES

     A discussion of the fundamental principles for fluid modeling of atmos-
pheric phenomena is presented here.  The dynamics of the flow in the fluid
model must accurately simulate those in the field.  Similarity criteria are
derived through analysis of the equations of motion.  This analysis shows that
various nondimensional parameters must be matched between the model and field
flows.  The significance of each of these parameters is discussed in detail.
Additionally, effects in the field upstream of the modeled area must be ac-
counted for in the fluid model by developing appropriate boundary conditions.
These are discussed at the end of the chapter.

2.1  THE EQUATIONS OF MOTION
     The equations of motion are the starting point for the similarity analysis.
With the earth as a reference frame rotating at an angular velocity n, the fluid
motion is described by the following equations (Lumley and Panofsky, 1964):
     Conservation of Momentum
                                                       ^L         (2.1)
                   «   CXj             Q0 CXf   TO       CX»C.Vk
     Continuity
                   ^ = 0                                             (2.2)
                   C.Vj
     Energy
                  cdT  cST      c2dT                                  ln „»
                  — + — Vt = K^^ (i =1,2,3)                       (2.3)
                   ct   cxj      CA-.C.Y,
where the x., axis is taken vertically upward, U. is instantaneous velocity,
6P and 6T are deviations of pressure and temperature from those of a neutral
atmosphere, pQ and TQ are density and temperature of a neutral atmosphere
(functions of height), v is kinematic viscosity, K is thermal diffusivity,
eiik 1S ^e alternatin9 tensor (if any two of the indices i,j,k are equal,
                                       3

-------
the component is 0; if i, j, and k are all unequal and are in cyclic order,
the component is +1; if not in cyclic order, the component is -1), 6^. is Kro-
neker's delta (fi^. = 1 if the two indices are equal and 0 if unequal), and the
summation convention is used here (whenever a suffix is repeated in a term,
it is to be given all  possible values and the terms are to be added for all).
     Equation 2.1 shows that the vector sum of the forces per unit mass acting
on a parcel of fluid must balance the acceleration of that parcel.  The first
term represents the unsteady acceleration of the fluid element.   The second
represents the advective acceleration.  The remainder are, respectively, the
Coriolis force, the pressure gradient force, the buoyancy force, and the fric-
tional force per unit  mass.
     Equation 2.2 is,  of course, the continuity equation, which  expresses the
conservation of mass in an incompressible fluid.  Equation 2.3 expresses the
conservation of thermal energy;  the time rate of change of thermal energy
(first term) equals the convection (or advection) of energy by the flow (sec-
ond term) plus the conduction of energy (third term).
     The assumptions made in deriving the above equations are:
     (1)  The atmosphere is  composed of a perfect gas of constant
          composition,
     (2)  the deviations of  pressure, temperature, and density are
          small compared with the neutral (adiabatic) values,
     (3)  the density  is independent of the fluctuating pressure
          (small Mach  number),
     (4)  variations of v and <  are negligible,
     (5)  the generation of  heat through viscous stresses is
          negligible,  and
     (6)  there are no sources of any kind.

     The second step in the  similarity analysis is to nondimensionalize the

-------
equations of motion through the use of appropriate reference quantities.
Reference quantities assumed to be supplied through the boundary conditions
are:  L, length; UR, velocity; pR, density; 6TR, temperature deviation; and
nR, angular velocity.  The dimensionless variables are
         ,  *i            v,
        x = -        U = —
           L         '   UR
        <'=^t       Q>,«>
           L            QR
             .
           o,
Using these definitions to nondimensionalize Eqs. 2.1 to 2.3 yields
                                                    2rr'
               8U,'   2    ,  ,     1 ddP'   1    ,     1  d2Ul
                                                                   (2.5)
and
       ddT     ddT   1 d26T
where RO=UP/LQD is the Rossby number,
          K   R
                    0}    is the densimetric Froude number,
      ResURL/v is the Reynolds number,
and   PeHURL/K is the Peclet number.
    Concerning the philosophy of modeling, Eqs. 2.4 to 2.6 with appropriate
boundary conditions completely determine the flow.  The question of uniqueness

-------
of solutions of the Navier-Stokes equations is avoided here (see Lumley, 1970).
These five equations contain five unknowns, U.J, Ui, 111, 6P1 and 6T1, so that,
in principle, their solutions may be determined.  Any two flows within the
same general category (i.e., insofar as they are governed by the above
equations) will be similar if and only if they are described by identical
solutions to the given set of Eqs. 2.4 to 2.6.  Solutions to this set of
equations will be identical if and only if the coefficients Ro, Fr, Re, and
Pe and the nondimensional boundary conditions are identical (Birkhoff, 1950;
Batchelor, 1953a).  The final statement, then, as it applies to laboratory
modeling of atmospheric motions becomes:  any atmospheric flow which can be
described by Eqs. 2.4 to 2.6 may be modeled by any other flow which can
also be described by the same set of equations, provided that the Rossby,
Froude, Reynolds, and Peclet numbers are identical, and provided that the
nondimensional boundary conditions (to be discussed later) are identical.
Equations 2.4 to 2.6 apply to both laminar and turbulent flows.  It is not
necessary to determine a priori whether the flow is laminar or turbulent.
     Thus far, only the requirements for similarity of flow patterns have
been discussed.  The dispersion of a pollutant within the system will now
be considered.  The contaminant is assumed to be completely passive in the
sense that it is without effect on the governing equation and undergoes no
transformations in the fluid.
     Prediction of the dispersion of this contaminant in space and time is
desired.  Since a passive contaminant is specified, its dynamic behavior must
be the same as that of the air; hence, the similarity criteria describing
its dynamic behavior have already been specified.  One additional parameter,
however, remains to be specified.  This is obtained through nondimensionali-
zation of the molecular diffusion equation:
                                      6

-------
where x represents the instantaneous concentration and a is the molecular
mass diffusivity.  Nondimensionalization of this equation requires
x1 = X/XR, and yields

       W+ ' ~dx'., ~ ReSc dx-dx'i   ,                                     (2.8)

where Sc = v/a is the Schmidt number.
     To summarize, Equations 2.4 to 2.6 and 2.8 form a complete set of equa-
tions which govern the dispersion of a dynamically passive contaminant in
the atmosphere and in a model.   If and only if the nondimensional coefficients
in these equations and the boundary conditions are identical, the dispersion
of the contaminant in a model will be identical to that in the atmosphere.

2.2  THE DIMENSIONLESS PARAMETERS
     It is generally impossible to simultaneously match all of the dimensionless
parameters when the ratio of the length scales (i.e., prototype to model) is
greater than about 10.  As an example, consider the Reynolds and Froude
numbers:            UL            UR
                Re =	,   Fr =  ,       .
If one models in a wind tunnel, the values of v and g are identical to values
in the atmosphere, and T  is roughly the same.  Thus, decreasing the length
scale by 10 requires an increase of 10 1n the velocity to satisfy the Reynolds
number criterion.  Considering the Froude number, L decreased by 10 and U~

-------
increased by 10 implies that 6TR must be increased by a factor of 1000, which
is, of course, highly impractical.  In general, a length scale reduction much
greater than 10 is desired.
     A factor of 15 in the Reynolds number may be gained by modeling with
water as the fluid medium, but then the Peclet number and Reynolds number
criteria cannot be satisfied simultaneously.   The Peclet number can be writ-
ten as the product of the Reynolds number and the Prandtl number.   Even if
the Reynolds number can be matched, the Prandtl number cannot, because it is
a fluid property and differs by a factor of 10 between air and water.   The
Prandtl number is not, however, a critical  parameter (see later discussion).
     Many examples of this type can be shown.  All  modelers recognize  that
rigorous modeling with significant reduction  in scale is impossible.   Under
certain circumstances, however, some of the criteria may be relaxed.   In the
first example, if the atmospheric flow were of neutral stability,  the  Froude
number would be infinite.   This is easily accomplished by making the model
flow isothermal.   (The vertical dimension of  a typical wind tunnel  is  small
enough that the temperature differences between isothermal  and strictly neu-
tral conditions is extremely small.  Hence, "neutral" and "isothermal" are
used interchangeably when  referring to wind tunnel  flows.)   Hence,  both Rey-
nolds number and  Froude number criteria may be satisfied simultaneously.
     It is instructive now to examine the nondimensional parameters in de-
tail.

2.2.1  The Rossby Number,  UR/LnR
     The Rossby number represents the ratio of advective or local  accelera-
        P
tlons (UR/L) to Corlolis accelerations (proportional to URnR).  Local  accel-

-------
erations may result from unsteadiness  or non-uniformities in the velocity
field.  Coriolis accelerations, of  course,  result from the fact that the earth
rotates.  The importance of the Rossby number criterion for modeling of atmos-
pheric diffusion is described as  follows.
     In the planetary boundary layer,  or "Ekman"  layer, which extends from
the Earth's surface to a height of  one to two kilometers, the combined effects
of the Coriolis acceleration, the pressure gradient, and surface friction cause
the wind vector to change direction or spiral with increasing height from the
surface.  The geostrophic wind is parallel  to the isobars, whereas the surface
wind blows to the left across the isobars,  typically at an angle of 20° to 40°.
The maximum rate of change of wind  direction with height occurs at the surface.
     Imagine a cloud of material  released at ground level in an Ekman layer.
Its transport and dispersion are  illustrated in Figure 1.  The surface wind is
directly into the paper.  The crosswind velocity profile is as shown.  The ini-
                                       CROSSWIND VELOCITY
                                       (SURFACE WIND INTO PAGE)\
                  I. INITIAL CLOUD          r- 4. DIFFUSION (NO
                   -2. INITIAL DIFFUSION      \  FURTHER TILTING)
                     \3. CROSSWIND TILTING
                       (NO DIFFUSION)
           Figure 1.  Schematic of diffusion  in  the Ekman layer.

tial cloud (Step 1), being  small,  is  transported mainly by the surface wind.
Its size increases mainly by  turbulent dispersion (Step 2).  At this point,
the upper levels of the  cloud will be advected in a different direction from
that of the surface wind.   Conceptually,  the  tilting of the cloud is imagined
to occur independently of diffusion,  whereas, in reality, tilting and diffu-
                                       9

-------
sion occur simultaneously.  The cloud is tilted by the crosswind (Step 3), and
the simultaneous turbulent diffusion (Step 4) increases the width of the cloud
at ground level over what it would have been by diffusion alone.  The center of
gravity of a slice of the cloud at ground level will  not follow the surface wind.
     The Rossby number describes the relative importance of the Coriolis ac-
celerations when compared with advective, or local accelerations.  If the
Rossby number is large, Coriolis accelerations are small, so that enhanced
dispersion due to directional  wind shear may be ignored.  Equivalently, a
near infinite Rossby number is automatically matched  in a model.
     To date, all modelers have assumed a large Rossby number and discarded
terms involving it in the equations of motion, or, equivalently, ignored it as
a criterion for modeling of diffusion.   (Howroyd and  Slawson, 1975, and others
have simulated the Ekman spiral using annular wind tunnels and rotating tanks,
but diffusion in such boundary layers has not been attempted.  Such facilities
are not yet practical for the  types of studies discussed here.)  Cermak et al.
(1966) and Hidy (1967) made the rather broad statement that, provided the typ-
ical  length in the horizontal  plane is  less  than 150km, the Rossby number can
generally be eliminated from the requirements for similarity.  McVehil  et al.
(1967) ignored the Rossby number when modeling atmospheric motions on the scale
of one kilometer in the vertical and several tens of  kilometers in the  horizon-
tal.   Ukeguchi et al. (1967) claimed that the cutoff  was 40 to 50km.  Mery (1969)
claimed that the Coriolis force may be neglected if the characteristic  length
is less than 15km.  The present discussion shows that the cutoff point  is on the
order of 5km for modeling diffusion under appropriate atmospheric conditions,
i.e., neutral or stable conditions in relatively flat terrain.
     The criterion is based on a length scale rather  than on the Rossby num-
ber itself because the angular rotation of the Earth, n , is a constant
                                      w

-------
(Ro = UD/Lfi ), and the characteristic velocity of the atmospheric flow does
       K   0


not vary by more than an order of magnitude, so that the characteristic length



is primarily responsible for determining the Rossby number.



     Several papers have examined the effect of crosswind shear on dispersion.



Pasquill  (1962) measured horizontal spread both in the longitudinal and cross-



wind directions for medium-range dispersion.  His data, however, were insuf-



ficient to allow firm conclusions to be drawn about the relative importance



of turbulence and shear in promoting horizontal spread.



     Corrsin (1953) showed that a <*t3'   in a uniform shear flow (a  is stream-
                                 X                                A


wise puff width), by considering Lagrangian particle motions.  Saffman (1962)

                                       *  *

applied the concentration-moment (von Karman integral) method to the classical



diffusion equation (he did not consider turning of wind with height, although



similar considerations are involved).  For a semi-infinite flow, ground-level



source  (puff), and linear velocity profile, he also found o «t ' .  For a com-
                                                           A

                                                     3/2
pletely unbounded flow, Smith (1965) showed that a «t   , using statistical
                                                  /\


techniques.



     Since the contribution to the spread from the turbulence alone is

    1 /o

o «t '   , it is clear that the shear effect will eventually dominate the dis-
 J\


persion process.  These solutions are valid only for constant diffusivity and



large times; they do not provide any indication of the early development.



Hence, they are of no help in determining at what distance the shear effect



becomes dominant.



     Tyldesley and Wellington (1965) used a numerical scheme and an analog



computer to study the effect of crosswind shear on dispersion.  They used an


                                                   1°
8-step Ekman spiral except with the surface wind 22*-  from the geostrophic



wind.  They claimed that the 3/2 power law does not apply because the cross-
                                      11

-------
wind shear is not constant with height.   Their estimates  indicate  that the



wind shear become? dominant around 4 to  6 km from the source.   For much larg-



er distances, the turbulence will  again  dominate because  the  shear goes to



zero for large heights.



     Hogstrom (1964) and Smith (1965) used statistical  approaches  to the



crosswind shear problem. Homogeneous turbulence, a mean wind  that  was constant



with height, and a crosswind component that varied linearly with height were



assumed.  They obtained  expressions valid for all  times of travel, but did



not estimate times or distances at which the shear would  dominate.



     Csanady (1969) attempted to confirm analytically the numerical  results



of Tyldesley and Wallington.  Because of analytical  difficulties,  he confined



his investigation to a slice at ground level of a cloud released from a source



at ground level.  He found that, indeed, the centroid of  the  slice at ground



level did not follow the surface wind.  By the time the cloud occupied 1/3  of



the Ekman layer (~600m), its distance from a line parallel to the  surface



wind was of order 10 km.  The contribution to the spread  from the  turbulence



and from the shear were  found to be equal at one kilometer from the source.



He estimated that, in the actual atmosphere, the shear effect would overtake



the turbulence effect at 3 to 4 km from  the source.   He showed that for small


           1 /2
times, a «t '  (i.e., turbulent diffusion dominates).  For intermediate times,


    3/2                                                                  1/2
a «t    (i.e., shear-induced diffusion dominates).  For large times, a at '
 •J                                                                    J


again, because the cloud height is the same as the Ekman  layer depth, and the



flow is, in effect, bounded (a  is cross-streamwise puff  width).   Maul (1978)



extended Csanady's approach and derived  expressions to include effects of



source height: and finite depth of mixing.



     Only a few diffusion experiments have been reported  that have been spe-



cifically designed to examine the relative effects of turbulence and shear.


                                      12

-------
 PasquHl  (1969)  reexarcinec!  tv/o  independent  studies  that  contain  information
 of interest in this  connection.   He  looked  at  Hogstrom's (1964)  data on the
 behavior  of smoke puffs  released  from an  elevated source under neutral and
 stable atmospheric conditions.  The  data  for crosswind spread show  the onset
 of a more rapid  than linear growth at 2.5 km.  These data are somewhat mis-
 leading because  they indicate the total puff width  rather than the  width at a
 given level.   Hence, they indicate bodily distortion (tilting) of the puff,
 but do not show  directly enhanced spread  at a  given level.   In accounting for
 this, Pasquill concluded that enhanced spread  at a  given level as a result of
 shear becomes  important  around  5  km  from  the source.  In analyzing  the Hanford
 data of Fuquay,  et al  (1964)  (continuous  ground release  of a tracer), Pasquill
 concluded that the effect on  spread  at ground  level under stable atmospheric
 conditions appeared  to have set in significantly at about 12.8 km.  He sum-
 marized:
      "...a bodily crosswind distortion of the  plume from a point source
      (either  elevated or on the ground) sets in between  2 and 3  km.  How-
      ever, the form  of the  crosswind growth curves  suggests  strongly that
      the  communication of the distortion  to the spread at a  given level
      was  not  of  practical importance below  about 5  km in the case of the
      elevated source and about  12 km in the case of the  ground level source.^
      Thereafter, the implication  is  that  the shear  contribution  is  dominant."
     Limited field measurements  by Brown and Michael (1974)  suggested that
directional wind shear was the dominant factor  in plume dispersion at great
distances  downwind (^50 km)  during stable  conditions.  More  extensive experi-
mental measurements by Howroyd and Slawson (1979) showed  qualitative and
quantitative agreement with the  analysis of  Pasquill (1969).
     The Rossby criterion should be  considered, therefore, if diffusion model-
ing is desired in a prototype with a  length  scale greater than about 5  km,
under neutral  or stable atmospheric  conditions, in relatively flat terrain.
One encouraging note  is that Harris'  (1968)  high wind  (neutral stability)
results show no systematic variation  in wind direction with  height over flat
                                     13

-------
terrain up to z*200m.

2.2.2  The Reynolds Number. UrL/v
     The physical significance of the Reynolds number becomes  apparent  by  not-
ing that it measures the ratio of inertial forces  (UJ^/L) to  viscous  or  fric-
                    2
tional forces (vUR/L ) in the equations of motion.   It  imposes  very  strong lim-
itations on rigorous simulation; it is the most abused  criterion  in  models of
atmospheric flows.  The scale reductions commonly  used  result  in  model  Reynolds
numbers three to four orders of magnitude smaller  than  found in the  atmosphere.
The viscous forces are thus relatively more important in the model than they
are in the prototype.  If strict adherence to the  Reynolds number criterion
were required, no atmospheric flows could be modeled.
     Various arguments have been presented which attempt to  justify  the use of
smaller Reynolds numbers in a model (i.e., to justify the neglect of the Rey-
nolds number criterion).  These arguments may be divided into  three  general
categories; the laminar flow analogy, Reynolds number independence,  and dissi-
pation scaling.  Each of these is discussed below.

2.2.2.1  The Laminar Flow Analogy
     Abe (1941) was the first to introduce this concept.  If the  instantane-
ous velocity, temperature, and pressure in Eq. 2.1 are written as the sum  of
mean and fluctuating parts (U.. = IT- + u..), and the equation  is then  averaged,
the following equation is obtained (after minor manipulation):

            80t .fjdU:             1 ddP   g  —      S2Ut   d^Tj
            ——h Uj —	\- 2eijklljUk =	1	oTo3i + v	 .            /o n \
            at     dXj            QO oXf  T0        dXjdXj   dXj              \£- -y I

An eddy viscosity is defined to relate the Reynolds  stress to  the mean  veloci-
ty, -(u.jU.)=K(3U../9x.).  The nondimensional equation is then
                                     14

-------
        du;    ,du:   i    _        i ddp1   i  —      i  d2u;    i  a2t//
where ReK = URL/K is called a "turbulent" Reynolds number.  Now if K/v is of
        3
order 10 , the term containing the turbulent Reynolds number is much larger
than the term containing the ordinary Reynolds number.  If the nondimensional
equation for laminar flow were now written, it would appear identical  to Equa-
tion 2.10, with the exception that the term containing the turbulent Reynolds
number would be absent.  Assuming that the prototype flow is turbulent and
                                                                    3
that the model flow is laminar, the scale ratio is of the order 1:10 , and UR
is the same order of magnitude in model and prototype, then,
       (Re)model = (ReKJprototype.
Hence, similarity may be established by modeling a turbulent prototype flow
by a laminar model flow when the scale ratio is on the order of 1:10 , all
else being equal in a nondimensional sense.
     This scheme is fundamentally incorrect for the same reasons that K-theo-
ries are fundamentally incorrect.  Eddy sizes scale with distance from the
ground, with the size of the obstacle, or generally, with the scale of sub-
stantial variation in the mean flow.  Turbulent diffusion is a flow property,
not a fluid property.  The laminar flow analogy assumes unrealistically that
eddy sizes are very small compared with the scale of variation of the proper-
ty being diffused.  Perhaps under very restrictive conditions, when there ex-
ists a small upper bound to the sizes of atmospheric eddies (i.e., extremely
stable conditions), there may be some realistic modeling possibilities, but
                                      15

-------
the chances of that being the case without any doubt are small.
     Perhaps qualitative use of this technique is the answer to  those (non-
diffusive) problems where the spread of a contaminant is controlled primarily
by advective transport (mean flow).  Two previous experimental  studies give
guidance here.  Abe (1941) attempted to model  the flow around Mt.  Fuji, Japan,
at a scale ratio of 1:50,000 using this analogy.   Cermak et al.  (1966) claimed
that "the model flow patterns obtained were not even qualitatively close to
that (sic) observed in actual field tests" (the original paper was not avail-
able for verification).  Cermak and Peterka (1966) made a second study of the
wind field over Point Arguello, California (a  peninsula jutting  into the
Pacific Ocean).  Cermak et al.  (1966) stated that:
     Comparison of the surface flow directions and smoke traces  for
     neutral and inversion flows established an excellent agreement
     in wind flow patterns over the Point Arguello area for flows  ap-
     proaching from the northwest.
After careful study of the figures presented,  the present author is not con-
vinced of the validity of this statement.  Large scatter of concentration lev-
els in the field data prevented firm conclusions concerning diffusion charac-
teristics of the two flow fields.  Rather surprisingly, a logarithmic plot of
concentration versus downwind distance showed  that rates of decrease of con-
centration with distance were grossly similar  in model and prototype.  In view
of the dissimilarity in surface flow patterns, this agreement is regarded as
fortuitous.
     Since kinematic viscosity is a fluid property, it is not an adjustable
parameter.  Turbulent eddy viscosity varies strongly with height,  stability,
and direction.  This severely limits the use of this laminar-turbulent analo-
gy for fluid modeling.  Mathematical modeling  techniques are superior to fluid
modeling techniques in the sense that K is a controllable variable in the math-
                                      16

-------
ematlcal model (e.g., a function of height, stability and direction).  The
quantity K is not a controllable variable in this sense in the fluid model.

2.2.2.2  Reynolds Number Independence
     This approach is based on the hypothesis that in the absence of thermal
and Coriolis effects and for a specified flow system, whose boundary conditions
are expressed nondimensionally in terms of a characteristic length L and ve-
locity UR, the turbulent flow structure is similar at all sufficiently high
Reynolds numbers (Townsend, 1956).  Most nondimensional mean-value functions
depend only upon nondimensional space and time variables and not upon the Reyn-
olds number, provided it is large enough.  There are two exceptions:  (1) those
functions which are concerned with the very small-scale structure of the turbu-
lence (i.e., those responsible for the viscous dissipation of energy), and
(2) the flow very close to the boundary (the no-slip condition is a viscous
constraint).  The viscosity has very little effect on the main structure of
the turbulence in the interior of the flow; its major effect is limited to
setting the size of the small eddies which convert mechanical energy to heat.
One way to avoid the effects of viscosity at the boundaries is to roughen the
surface of the model (see discussion of Boundary Conditions and Section 3.3.1).
     This hypothesis of Reynolds number independence was put forth by Town-
send (1956).  He called it Reynolds number similarity.   There now exists a
large amount of experimental evidence supporting this principle.   Townsend
stated it simply: "geometrically similar flows are similar at all  sufficient-
ly high Reynolds numbers."  This is an extremely fortunate phenomenon from the
standpoint of modeling.  The gross structure of the turbulence is  similar over
a very wide range of Reynolds numbers.  This concept is used by nearly all mod-
                                       17

-------
elers.  It is graphically illustrated in Figures 2  and 3.   The  two  jets  shown

in each figure are identical  in every way except for the viscosity  of the  flu-

ids, and therefore the Reynolds numbers, which differ by a  factor of 50.
Figure 2.  Turbulent jets showing that the Reynolds number does  not much
           affect the large scale structure, so long as  it is  sufficiently
           large that the jet is indeed turbulent.   The  upper  jet has  a
           Reynolds number 50 times that of the lower.   (Reproduced with
           permission from Illustrated Experiments  in Fluid Mechanics,
           National Committee for Fluid Mechanics Films, copyright 1972
           Education Development Center, Inc., Newton, MA).

Figure 3.  Shadowgraphs of the jets shown in Fig. 2.  Note how much finer
           grained is the structure in the high Reynolds number jet than
           that in the low Reynolds number jet.  (Reproduced with per-
           mission from Illustrated Experiments in Fluid Mechanics, National
           Committee for Fluid Mechanics Films, copyright 1972, Education
           Development Center, Inc. Newton, MA).
                                      18

-------
     To obtain a better idea of the contributions to dispersion from the vari-
ous scales of turbulence, it is convenient to examine the spectral  form of the
Taylor (1921) diffusion equation.  Taylor's expression for the mean-square
fluid particle displacement in a stationary homogeneous turbulence is given by
                                  T t
                              = 2? J J
                                  0 0
(2.11)
where v2" is the variance of particle velocities, pU)=v(t)v(t+c)/vz(t) is the
Lagrangian autocorrelation of particle velocities with time separation £, and
                                              *     *
T is the time of travel from the source.  Kampe de Feriet (1939) and Batchelor
(1949) applied the Fourier-transform between the autocorrelation and the cor-
responding Lagrangian spectrum function
                        FL(n)  E f p(t)  cos  (2irnt)  dt,
                               o
to obtain
                                                  ..                    (2.12)
where n is the frequency.  The squared term under the integral is the filter
function illustrated in Figure 4; it is very small when n>l/T and virtually
unity when n<0.1/T.  For very small travel times, the filter function is
virtually unity, so that all scales of turbulence contribute to the dispersion
with the same weight that they contribute to the total energy.  For larger
travel times, larger scales of turbulence progressively dominate the dispersion
process.  This means that eddies with diameter smaller than about one-tenth
the plume width do not significantly affect the spread of the plume; only
those eddies with diameter about the same size as the width of the plume and
larger substantially increase its width.
                                      19

-------
               0.01/T
0.1/T               1/T
    FREQUENCY, n (Hz)
10/T
    Figure 4.  Filter function in Eq. 2.12:  As the  travel  time  T  in-
               creases, the contribution of small scale  (high  frequency)
               motions to dispersion diminishes rapidly.
     It may be helpful here to examine typical energy  spectra  so  that  (1)
"weather" and "turbulence" may be defined as separate  and distinct  entities,
and (2)  the influence of Reynolds number upon the  shape of  the spectrum will
be more easily understood.  Figure 5 shows a spectrum
                  Su(n) - 4
 u(t)  u(t+t') cos 2imt' dt1
of wind speed near the ground from a study by Van der Hoven  (1957).   It  is
evident that wind effects can be separated roughly  into two  scales of motion:
                                      20

-------
large scale (low frequency) motions lasting longer  than  a  few  hours,  and  small
scale (high frequency) motions that last considerably  less  than  an  hour.   The
large scale motions are due to diurnal fluctuations, pressure  systems,  passage
                                      1	1	1	1	1    I    I
                                      SPECTRAL GAP 	*	TURBULENCE
        Cycles/hr
         Hours
                                             100
                                             0.01
1000
0.001
     Figure 5.
Spectrum of wind speed at 100m.  (Reprinted with permission
from J. Meteorology, American Meteorological Society, van
der Hoven, 1957.)
of frontal systems, seasonal and annual changes, etc., and are generally
called weather.  The small scale motions are associated with roughness
elements, topographical features, and differential surface heating  in the
boundary layer and are called turbulence.  The spectral gap (low energy
region) separating weather from turbulence is a very fortunate occurrence,
both from an analytical viewpoint and from a fluid modeling viewpoint.
Because of this gap, it is possible to consider these regions independent-
ly and to execute proper mathematical operations to determine the statistical
properties of the two regions.  It is the smaller scales of motion, the
turbulence, which are simulated in a fluid modeling facility.  Steady state
averages of fluctuating quantities in the model atmosphere correspond to
approximately one-hour time periods in the real atmosphere (during which
the mean wind is steady in speed and direction).  From results of model ex-
                                      21

-------
periments conducted at different mean wind speeds and directions,  the  low  fre-
quency contribution can be constructed analytically from distribution  charts
of wind speed and direction (wind roses).
     Figure 6 is a definition sketch of a turbulent energy spectrum from Wyn-
gaard (1973), which will be helpful in understanding future discussions.   (It
is not intended here to present a detailed discussion of turbulent energy
spectra.  Only those features of direct interest will be covered.  The ardent
student should consult Batchelor, 1953a, Hinze, 1975, or Tennekes  and  Lumley,
1972.)  In Figure 6, we have used wave number K instead of frequency n so  that
 ENERGY
 CONTAINING
-RANGE
                                  INERTIAL
                                 - SUBRANGE -
                                                I
                                                I
                                                I DISSIPATION!
                                                  RANGE j
                                                         I
                                     in
     Figure 6.
Form of turbulence spectrum.  (Reprinted with
permission from Workshop on_ Micrometeorology,
American Meteorological Society, Wyngaard, 1973.)
we will be more inclined to think in terms of length scales.  The relation-
ship is K=2irn/Lf.  (The spectrum function S  in Figure 5 is one-dimensional,
whereas that of Figure 6, E, is three-dimensional.  The differences need not
concern us here.)  Note that an integral scale I and a microscale n are de-
fined.  The integral scale may be thought of as the characteristic size of the
                                      22

-------
energy-containing eddies and Is located at the peak of the three-dimensional
spectrum.  The microscale may be thought of as characteristic of the smallest
eddies in a turbulent flow.  They are the ones primarily responsible for the
dissipation (e) of turbulent kinetic energy.  The ratio of the integral scale
to the microscale, then, is a measure of the width of the spectrum or the
range of eddy sizes in the turbulence.
     A pertinent question at this point is:  how does the spectrum of turbu-
lence in a simulated atmospheric boundary layer in, say, a wind tunnel compare
with that in the real atmospheric boundary layer?
       1nE(K)
                          RE
                       INCREASING
                   I/I                   1/T?           Inx
     Figure 7.  Change of spectrum with Reynolds number.

Figure 7 shows how the spectrum shape 1s affected by changing the Reynolds
                                       23

-------
number.  For a given class of turbulent flow, a decrease of the Reynolds num-
ber decreases the range of the high-frequency end of the spectrum,  whereas  the
size of the energy containing eddies changes only very slowly with  Reynolds
number.
     To be somewhat more quantitative, it is useful  to examine the  trends ob-
served experimentally and theoretically in grid-generated turbulence.   A good
measure of the width of a turbulent energy spectrum is the ratio of the inte-
gral scale, I, to the Kolmogoroff micrbscale, n.   It may be shown,  through  ar-
guments presented by Corrsin (1963), that this ratio is
                                 ^Re3/4,                            (2.13)
where Re is the grid Reynolds number (based upon  upstream velocity, U,  and
mesh size, M).  Ideally, both I and n would be reduced in the same  proportion
in a model (i.e., the geometrical scale ratio), so that the width (number of
"decades") of the spectra would be identical in model  and prototype.   It is
clear, however, that this would require identical Reynolds numbers  in  model
and prototype.  Eq. 2.13 may be used to estimate  the comparative widths of
model (m) and prototype (p) spectra:

3/4
                                                3/4
                                                                       (2.14)
assuming that the ratio of flow speed to viscosity is roughly the same in mod-
el and prototype.  (L is a characteristic length in the flow, for example, the
mesh size or the height of a building.)  Hence, at a scale ratio of 1:1000, a
seven-decade-wide atmospheric spectrum is "modeled" by a 4i-decade-wide labora-
tory spectrum.  This appears to be a drastic reduction in spectral  width, but
observations of grid-generated flows show that only the high-frequency end of the
                                      24

-------
spectrum is cut out,  so that this  reduction  in  spectral width  has  insignificant
effects.  It is found empirically  that I/M at a fixed  distance x/M downstream
from a grid is nearly independent  of Reynolds number (Corrsin, 1963).  Similarly,
it may be expected in other flow geometries  that I/L at corresponding geometrical
locations will be roughly independent of Reynolds number,  i.e.,

                   'i~-'t.                                             <2-15>

Indeed, the integral  scale is found to be roughly half the size  of the charac-
teristic length and independent of Reynolds  number in  a wide  variety of  class-
es of flows.
     Combining Eqs. 2.13 and 2.15  yields
                            3/4  /r \l/4
                  „   T /T  \3/4   /r \
                  n'~ MM   ~[-?i
                  ^~~T\ j  I   ~\T  I
                  fm  Jm \Lp/     \LmJ
In summary, integral  scales reduce with the  first  power  of  the geometrical
scale ratio (as desired), whereas Kolmogoroff microscales reduce with only the
one-fourth power of the geometrical  scale ratio.   As  we  have  seen  in our pre-
vious discussion, the largest eddies contribute to the spread of a plume and
the ones smaller than the plume width have little  dispersive  effect; hence, the
mismatch of Reynolds  number between the model and  the prototype is insignifi-
cant.
     A practical example here will make the  point  clear.  The Kolmogoroff mi-
croscale in the atmosphere is about one millimeter between  1  and 100 m  above
ground (Lumley and Panofsky, 1964).   As indicated  by  Eq. 2.14, at  a scale ra-
                                     25

-------
 tio of  1:500, the spectral width in a model would be approximately two orders
 of magnitude smaller than desired.  The Kolmogoroff microscale in the model
 would be about 100 times larger than required by rigorous similarity.  This
 would correspond to a Kolmogoroff microscale of 10 cm in the atmosphere.  It
 is difficult to imagine a practical atmospheric diffusion problem where eddies
 smaller than 10 cm would contribute significantly to the spread of a contami-
 nant.
     It should be noted that the arguments beginning with Eq. 2.13 have been
 concerned with Eulerian scales and spectra, whereas they would best have been
 posed in terms of Lagrangian coordinates, as required in Eq. 2.12.  However,
 it is reasonable to assume that if the Eulerian spectra and scales of model
 and prototype are similar, then so will  be the Lagrangian spectra and scales;
more appropriately, if Eulerian spectra  differ in certain respects between
model and prototype, then the Lagrangian spectra will be dissimilar in like
 fashion.  Hence, conclusions drawn in the Eulerian framework are expected to
 be valid in the Lagrangian framework (to at least within the same order of
magnitude).
     Concerning puff (relative) diffusion, the problem is somewhat different.
Corrsin (1961b) has argued that the principal  contribution to two-particle
relative diffusion comes from eddies of  roughly the same size as the particle
pair separation.  This contrasts with single particle diffusion, where the
principal contribution comes from eddies of the same size and larger.  Hence,
puff diffusion in a model  will  depend somewhat more strongly upon the model
Reynolds number.  Intuitively,  it appears that reasonable results would be ob-
tained if the Kolmogoroff microscale were small  compared with the initial  puff
width.
                                      26

-------
     The discussion of the Reynolds number criterion in the modeling litera-
ture normally centers around sharp-edged geometry where it is usually stated
that the mean flow patterns will  not be much affected by changing the Reynolds
number.  While this is true, it does not make full use of the concept.   Most
mean-value functions, including those describing the main turbulence structure,
will be nearly independent of Reynolds number, providing it is sufficiently
high; the only exceptions are those two discussed at the beginning of this
section.
     The question is: how high must the Reynolds number be to be high enough?
A precise answer would depend upon the geometrical shape of the boundaries,
the roughness of the model surface, the accuracy desired, the type of informa-
tion desired from the model, and possibly other effects (e.g., those charac-
terized by the Rossby and Froude numbers).  The answer to this question is rea-
sonably well known for simple flow classes such as jets and cylinder wakes,
but is largely unknown for models of atmospheric motions.  Specific recommen-
dations on minimum Reynolds numbers to be achieved in various classes of flows
will be made in Section 3, Practical Applications.

2.2.2.3  Dissipation Scaling
     A hypothesis on the similarity of the detailed turbulence structure of
model and prototype flows was proposed by Nemoto (1968).  Again, a basic as-
sumption is that thermal and Coriolis effects are negligible.  He reasoned
that mean flow patterns of both the model and prototype would be similar if
the turbulent structure of the two flows were geometrically similar.  Two as-
sumptions were made:
     (1)  the turbulence of both model and prototype flows was 'locally
          isotropic1 (Kolmogoroff, 1941),
                                      27

-------
      (2) the Kolmogoroff velocity, u, and microscale, n» characterize the
         turbulence at each point in the flow.
     Assumption 1 is satisfied if the Reynolds number is very large (Hinze,
1975).  Using assumption 2, Nemoto reasoned that the turbulence structures of
model and prototype flows would be similar when
                 Lp                                                   (2.17)
and
            o.  Ua
             r   "f
                                                                      (2.18)
where the subscripts m and p refer to model and prototype, respectively.
From the definitions of n and u, the following equation may be established:
              _»*_
                *                                                     (2.19)
From Eqs. 2.17, 2.18 and 2.19, it may be deduced that:
             URm  AmY/3 /LmY/3
             UR"  ^£p'   ^Lp'                                            (2.20)
Eq. 2.20 is the similarity criterion proposed by Nemoto.   He has also shown
how the above relationship may be obtained from a special  nondimensionaliza-
tion of the turbulent energy equation.
     It is agreed that the turbulence structures of model  and prototype flows
would be similar if Eqs. 2.17 and 2.18 could be satisfied.  However,
it is impossible to satisfy Eq.  2.17 using typical  length  scale reductions
(i.e., 1:300 to 1:1000).  As mentioned previously,  the  Kolmogoroff microscale
in the lower layer of the atmosphere is about 1mm.   Typical  values of
n in laboratory flows are 0.075mn (Wyngaard, 1967), and 0.5mm (Snyder and
                                      28

-------
Lumley, 1971).  Measurements 1n very high Reynolds number laboratory flows
(Klstler and Vrebalovich, 1966) show that the smallest Kolmogoroff microscale
which can reasonably be generated in the laboratory is 0.05 mm.   Thus,  the
largest ratio of microscales is of order 20, which is far short  of the
typical ratio required (i.e., 300 to 1000).
     Generally speaking, the satisfaction of Eq. 2.17 requires identical  Reyn-
olds numbers of the model and prototype flows, as shown in the discussion of
Reynolds number independence.  This is also easily shown by the  usual nondi-
mensionalization of the turbulent energy equation (Cermak et al., 1966).

2.2.3  The Peclet Number and the Reynolds-Schmidt Product
     The Peclet number is most easily discussed by writing it as
                 URL   URL v
                        V  K
where Pr is the Prandtl number.
     The Reynolds-Schmidt product may be written as
                    v  a
Both of these dimenslonless parameters have the same form (i.e., the product
of a Reynolds number and a ratio of molecular transport coefficients).   Both
the Prandtl and Schmidt numbers are fluid properties and not flow properties.
The Prandtl number is the ratio of the momentum diffusivity (kinematic  vis-
cosity) to the thermal diffusivity.  The Schmidt number is the ratio of the
momentum diffusivity to mass diffusivity.
     For air, the Prandtl number does not vary strongly with temperature.
When air is used as the medium for modeling, the Prandtl number is nearly the
same 1n model and prototype, and, 1f the Reynolds numbers were the same, the
                                      29

-------
Peclet number criterion would be nearly satisfied.  However, if water is used
as the medium for modeling, the Prandtl number at ordinary room temperatures
is a factor of about 10 larger than it is in air, and it varies rather great-
ly with temperature.  Thus, from the standpoint of rigorous similarity, it
does not appear that water would be a suitable medium in which to do model
studies.
     The Schmidt number for most gases in air is about one.  Thus, the Schmidt
number for an effluent plume (which contains only minor fractions of gases
other than air) diffusing in the atmosphere is about one.   If air is used as
the medium for modeling, the Schmidt number (for nearly any foreign gas intro-
duced) will be nearly the same in model and prototype.  If, at the same time,
the Reynolds number were the same, the Reynolds-Schmidt product criterion
would be nearly satisfied.
     When water is used as the medium for modeling,  salt water or alcohol  are
typically used to simulate the buoyancy of a plume.   The Schmidt number for
sodium chloride or alcohol in water is approximately 800.   Thus, it appears
that strict similarity using water as the modeling medium would be difficult
to obtain.
     The basic problem, however, in matching of the  Peclet number or Reynolds-
Schmidt product, is not in the Prandtl or Schmidt numbers, but rather in the
Reynolds number.  Arguments similar to those constructed for Reynolds number
independence may be used to justify the neglect of the Peclet number and Reyn-
olds-Schmidt product as modeling criteria.  The term on the right-hand side
of Eq.  2.6 represents the molecular diffusion of heat.  The term on the right-
hand side of Eq. 2.8 represents the molecular diffusion of mass.  In this
connection, both heat and mass are regarded as passive scalar contaminants.
                                     30

-------
If the flow is of a sufficiently high Reynolds number, then the main struc-
ture of the turbulence will be almost totally responsible for the transport
of the contaminant (heat or mass).  Molecular diffusion will contribute very
little to the bulk contaminant transfer; its main function is to smooth out
the very small-scale discontinuities of concentration or temperature (i.e.,
it acts as a low-pass filter on the concentration or temperature fluctuations).
Indeed, arguments of this nature have been used to postulate the form of con-
centration or temperature spectra at large wave numbers (see Corrsin, 1964;
Pao, 1965).  The main effect of the diffusivities is confined to setting the
high wave-number cutoff of the temperature or concentration spectrum.  Since
turbulent diffusion strongly dominates molecular diffusion in turbulent air
flows, especially at high Reynolds numbers, and since molecular diffusion is
even less important for Prandtl or Schmidt numbers larger than unity, the
effect of not matching the Prandtl or Schmidt numbers of the prototype in the
model is unimportant.  Generally speaking, the Peclet number and Reynolds-
Schmidt product may be neglected as modeling criteria if the flow exhibits
Reynolds number independence.   Both air and water are suitable media for mod-
eling, from this standpoint.  Further discussions of air versus water are giv-
en in Chapter 4.
     A cautionary note is advised here.  The foregoing comments were primarily
intended to apply to diffusion of a plume from a smokestack, and the implication
was that if the Reynolds number was large enough, then so would be the Peclet
number and Reynolds-Schmidt product.  However, experiments involving molecular
diffusion over smooth surfaces, such as heat transfer from model  buildings
(Meroney, 1978) and water evaporation from a simulated lake (Cermak and Koloseus,
                                      31

-------
1954), suggest that Peclet numbers and Reynolds-Schmidt products need to be
quite large for similarity, i.e., 500,000 or greater.  This is evidently due to
the existence of laminar sublayers beneath the turbulent boundary layers in the
models, where molecular diffusion may have been the controlling parameter.   It
is conceivable that this problem could have been overcome by roughening the
model surface (see further discussion in Section 3.3.1), although this is not
a straightforward solution because the roughness would, of course, increase the
surface area for heat and mass transfer, and possibly provide undesireable
insulation.

2.2.4  The Froude Number, Un/(gl-STn/T.)1/2
       ~~~~ — ~—~— ~       "   K      K  0
     The square of the Froude number represents the ratio of inertia! forces
to buoyancy forces.  A large value of the Froude number implies that buoyancy
forces are small compared to inertial forces.  Thus, thermal effects become
important as the Froude number approaches unity.  Batchelor (1953b) has shown
how this parameter is related to the Richardson number.  In the absence of a
clearly defined length in the atmospheric boundary layer, it is convenient to
         2
replace U^/L by a representative velocity gradient and  6TR/L by a representa-
tive temperature gradient.  Substitution of these gradients into the expres-
sion for the Froude number yields,
                            g  L2 5TR   g (ddTldz)R
Thus, the Froude number may be regarded as the inverse square root of a Rich-
ardson number.  It is also related to the Monin-Obukhov (1954) length.  Al-
though any of these parameters may be used as similarity criteria, the Froude
number is used here because it appeared naturally through the non-dimensional i
zation of Eq. 2.1.  Batchelor  (1953b) has also discussed the conditions under
                                      32

-------
which this parameter is the sole governing criterion for dynamical similarity
of motions of a perfect gas atmosphere.
     A different interpretation of the Froude number is quite useful in
considering stably stratified flow over hilly terrain.  Suppose the flow approach-
ing an isolated hill has a uniform velocity profile and a linear density gradient.
The appropriate form of the Froude number is then
                           Fr2 = PU2 /(ghAp),
where the characteristic length L has been replaced by the height h of the
hill and the density difference Ap is that between the base and top of the
hill.  The square of the Froude number represents the ratio of kinetic
to potential energy, i.e., it represents the ratio of the kinetic energy in the
approach flow to the potential energy required to raise a fluid element from the
base to the top of the hill.  It is clear that if the Froude number is much less
than unity (very strong stratification), there is insufficient kinetic energy in
the approach flow to raise fluid from the base to the top of the hill.  With a
two-dimensional hill perpendicular to the wind direction, this would result
in upstream blocking of the flow below the hill  top (Long, 1972).   For a
three-dimensional  hill, the fluid, rather than being blocked, can go round the
hill (Hunt and Snyder, 1980).   Hunt et al. (1978) and Snyder et al.  (1979) have
shown in more quantitative terms how the streamline patterns (hence, plume
trajectories) change drastically with changing Froude number.  It is thus
evident that the Froude number is an essential parameter to be matched when
modeling stably stratified flow over hilly terrain (see further discussion
in Section 3.4).
     The Froude number is not, by itself, a difficult parameter to duplicate
in a fluid model.   It is likely to be the most important individual  parameter
to be matched when the model  is to simulate atmospheric diffusion.  When the
                                     33

-------
modeling medium is air, provisions must be made for heating or cooling of the
air stream to obtain the temperature stratification.  However, in order to match
small Froude numbers of the prototype in a model  with a typical  scale reduction
(i.e., 1:300 to 1:1000), and in order to maintain reasonable temperatures
(i.e., maximum temperature difference of 200°C) in the model, it is necessary
to decrease the mean flow speed.  To match the Reynolds number between the
model and the prototype requires that the mean flow speed be increased.  This
conflict is resolved by matching the Froude number while insuring that a
Reynolds-number-independent flow is established.   This is not always possible.
     When considering water as the medium for modeling, it is necessary to
define the Froude number in terms of density, rather than temperature.  The
common method of producing stable density stratification in water is by
producing thin layers of various concentrations of salt in water.  In view
of the very small mass diffusivity of salt in water, an undisturbed stable
mass of salt water will remain that way for several weeks before the density
gradient is changed substantially by molecular diffusion.  Maximum density
differences are limited (about 20% in the dimensionless density difference),
so that flow speeds must be reduced as was the case with air as the modeling
medium.  Recirculating systems using this technique have been impractical
because of resulting mixing within a pump.  However, Odell and Kovasznay
(1971) have designed a rotating disk pump that maintains the gradient;
this device may permit the use of recirculating salt water systems, although
thus far it has only been used in very small channels.  An interesting
technique is reported by Homma  (1969) wherein fresh and saline water are
mixed to produce stable density gradients at the entrance of a once-through
open water channel.  This technique offers the possiblity of providing the
proper boundary conditions of turbulent flow (see next section), which is
quite difficult in a still tank wtth a towed model.
                                      34

-------
     Hydraulic engineers  and others  (e.g.,  Halitsky,  1979)  frequently  define


the Froude number as  U/(gL)^2  or  U2/gL;  at first  glance,  it would  appear  they


have omitted the dimensionless  density  difference  Ap/p.  However, the  Froude


number was originally introduced in  the field  of hydraulics and was  used to


describe the behavior of  the free  surface of rivers and  channels, where the


density difference was that across the  free surface,  i.e.,



                       Ap/p = (pwater -pair)/pwater * K
               P
The parameter U /gL thus  has a  very  meaningful  interpretation  in the field


of hydraulics; it does include  a density term,  albeit one  that degenerates


to unity.  However, density differences in  the  vast majority of atmospheric

                                                                    2
simulation problems are much less  than  unity,  so that the  parameter U  /gL


does not have a meaningful  interpretation in and of itself;  it has  a  mean-


ingful interpretation only  when combined with  the  density  term.  Further


support of this view comes  from an examination  of  the nondimensionalization


of the equations of motion  (e.g.,  see Eq. 2.4  or Halitsky,  1979); the  para-


meter U /gL appears only  in combination with the density term.





2.3  BOUNDARY CONDITIONS



2.3.1  General
     A statement was made earlier that it was not necessary to determine


a priori whether the flow was laminar or turbulent in order to apply Eqs.  2.4


to 2.6 to the determination of the similarity parameters.   It is certainly


necessary to determine whether or not the flow is turbulent in order to


specify the boundary conditions.  It is assumed here that  the atmospheric


flow is always turbulent.  Furthermore, it was stated that the model flow


would be identical to the prototype flow if, among other things, the




                                      35

-------
non-dimensional boundary conditions were identical.
     Batchelor (1953b) points out:
       Regarding the boundary conditions, we do not know enough about
       the differential equations concerned to be able to say with
       certainty what conditions must be specified to make the problem
       determinate, but it is a plausible inference from physical
       experience that u'  and p1 must be given as functions of t1  at
       all spatial  boundaries and ui p1, p1 must be given as functions
       of x1 at an  initial value o7 t1; it seems certain that such a set
       of Foundary  conditions is sufficient, although in some circum-
       stances the  conditions may well be over-sufficient.

(The prime here indicates  a nondimensional quantity and the underline signifies
a vector.)  In the  problem considered in the present paper, it is  clear that

-------
     Nearly all modelers have considered the specification of boundary condi-
tions from a different viewpoint, that Is, through the spectrum of turbulence
1n the approach flow.  Armltt and Coum'han (1968) have given qualitative argu-
ments which suggest that, for the study of plume dispersal, not only must the
turbulence intensity components be properly modeled, but also the spectrum of
each component 1s required, particularly the low-frequency end of the spec-
trum.  This idea 1s in agreement with the previous discussion on the contribu-
tion of the various scales of turbulence to the dispersal of contaminants.
Some control of the turbulence spectra 1n the approach flow is possible (see
Section 3.2).

2.3.2  Jensen's Criterion and Fully Rough Flow
     The specification of the velocity on solid boundaries is simple; it is
zero, and all of its moments are zero.  Hence, geometrical similarity of model
and prototype 1s required.   This raises another question; how much detail is
necessary?  From the standpoint of rigorous similarity, of course, every de-
tail of the prototype must be duplicated in the model.   However, in view of
the fact that the Reynolds number will not be duplicated, the fine detail is
unnecessary.  Jensen (1958) has suggested that, 1f the roughness length of
the prototype, ZQ, may be determined (or at least estimated), then it should
be scaled according to
                              zom
(the roughness length is a fictitious length scale characterizing flow over a
rough surface.  For uniformly distributed sand grains of size e,  the roughness
length Is typically e/30.)

                                     37

-------
This equation is known as Jensen's criterion, and has been widely used.   It
implies that elements or details smaller than e will  have very little effect
on the overall flow; hence they need not be matched in the model.  Details a-
bout the same size as e need to be matched only approximately (for example,
randomly distributed grains of sand).  It may be necessary, with large reduc-
tions in scale, to abandon Jensen's criterion, as discussed below.
     The flow of fluid close to a smooth boundary is  not Reynolds number indepen-
dent.  The no-slip condition at the surface is a viscous constraint.   A vis-
cous sublayer exists immediately adjacent to the wall where viscous stresses
dominate.  If the surface is roughened such that the  irregularities are larger
than the thickness of the viscous sub-layer which would have existed on a
smooth surface under otherwise identical flow conditions, viscous stresses be-
come negligible.  The irregularities then behave Hke bluff bodies whose re-
sistance is predominantly form drag, I.e., the resistance is due to the pres-
sure difference across the obstacle rather than to viscous stresses.   Such a
rough surface is said to be aerodynamically rough; the flow over an aerodynam-
ical ly rough surface is Reynolds-number independent.   The criterion which in-
sures that the flow is aerodynamically rough is u*zQ/v>2.5 (Sutton, 1949),
where u^ is the friction velocity.
     This is extremely fortunate from a modeling standpoint, because atmos-
pheric flows are almost always aerodynamically rough (Sutton, 1949).  If model
flow conditions are chosen such that u*z^/v>2.5, one can be certain that the
                                        o  —
boundary layers are turbulent, so that such things as separation  'bubbles' and
wakes behind obstacles and transition, separation, and reattachment of bounda-
ry layers on topographical surfaces will change very little with Reynolds num-
ber.  The critical roughness Reynolds number, then,  is that at which the bound-
ary layer on the model becomes qualitatively comparable to that on the proto-
                                      38

-------
type.
     For large reductions in scale, the simultaneous satisfaction of Jensen's
criterion and the critical Reynolds number may not be possible.  The critical
Reynolds number criterion is undoubtedly the more important of the two crite-
ria, because it controls the quality of the flow.  Over-roughening of the mod-
el surfaces, thereby ignoring the Jensen criterion, will merely limit the
resolution of the flow over the model (details about same size as z  will not
be capable of being resolved), but, since the Reynolds criterion is met, the
over-all flow patterns will most likely be matched.

2.3.3  Other Boundary Conditions
     Specification of the detailed temperature distributions at the solid
boundaries is rarely discussed from the modeling standpoint.  In current prac-
tice, the solid boundary is maintained at constant temperature.  It is plausi-
ble that the amount of detail in the temperature distribution should be deter-
mined on the same basis as the amount of detail in the geometrical  boundaries.
This has never been done, although basic attempts have been made by Chaudhry
and Cermak (1971).  Similar considerations apply to the specification of the
boundary conditions of the density distributions when the modeling medium is
salt water.
     Specification of boundary conditions on concentration distributions is,
in principle, easy.  In practice, the difficulty would depend on the type of
problem to be studied.   For example, if the problem were to determine the ef-
fect of a single source, the boundary conditions could be x'=0 initially every-
where and x'=constant at the location of the source for all time thereafter
     Very little is known about the boundary specification of the pressure, p'.
Normally, the mean pressure gradient in a wind tunnel is adjusted to zero.
                                     39

-------
Fluctuating pressures  are,  of course,  not controllable parameters  in  a  model.
     In current practice, the upstream boundary  conditions  on  velocity  and
temperature are specified to be reasonably similar to some  theoretical  formula
such as the logarithmic velocity distribution.   Cermak et al.  (1966)  argue  that
boundary layers (kinetic and thermal)  grown naturally over  long  lengths of
rough ground must be inherently similar to those in the atmosphere.   Others
(Ludwig and Sundaram,  1969; Armitt and Counihan, 1968; Mery,  1969) use  artifi-
cial techniques for generating thick boundary layers over short  distances.  Mery
(1969) and Ogawa et al. (1980) have attempted to model both velocity  and temp-
erature profiles using artificial techniques. Any of the present  techniques
for boundary-layer generation appears  to be suitable; all of them  come  reason-
ably close to matching the first two moments of  the velocity/temperature distri-
butions.  Boundary-layer heights should correspond with the geometric scale
ratio, but, as various studies have shown, if the depth of  the boundary layer
is large compared with the model height, there is some scope for selecting
boundary-layer heights for convenience.  Practical goals and techniques for
simulating the atmospheric boundary layer are discussed in  Section 3.2.

2.4  SUMMARY AND RECOMMENDATIONS
     Similarity criteria for modeling atmospheric flows in  air and water have
been derived.  Rigorous similarity requires that five nondimensional  parameters
plus a set of nondimensional boundary conditions must be matched in both model
and prototype.  It has been determined that the  Rossby number should be con-
sidered when modeling prototype flows with a_ length scale greater than about
5_ km. under neutral or_ stable atmospheric conditions, i_n_ relatively flat
terrain.   It is concluded that more work needs to be done to determine under
what conditions the prototype length scale may be extended while still  ignoring
the Rossby number criterion.  It is recommended that study be continued on
                                      40

-------
methods for simulating Coriolis forces in a model.
     The concept of Reynolds number independence  has  been  found  to  be  extreme-
ly useful and powerful.  Heuristic arguments have been  given  through the use
of this concept that rt is_ not necessary to_ match Reynolds  number,  Peclet
number or_ Reynolds-Schmidt product between model  and  prototype,  provided the
model Reynolds number is sufficiently large.  Current practice indicates that
sufficiently large Reynolds numbers are attainable at least for  sharp-edged
geometrical structures in ordinary meteorological wind  tunnels.   More  work
needs to be done to determine if sufficiently high Reynolds numbers may  be
obtained in the laboratory for the simulation of  flow over  more  streamlined
surfaces.  The Froude number jhs^ the^ most important single parameter describing
the prototype flow which must be duplicated in the model.   The specification
of boundary conditions was found to be nebulous both  in terms of how many
variables are necessary and sufficient and also in terms of the  type of
statistical information required (i.e., is the specification of  only a few
lower order moments of each variable sufficient?)  Geometrical similarity
(nondistorted models) ij^ required from the specification of_ zero velocity
ajb the solid boundaries.  It was decided that details cif the prototype p_f_ size
smaller than the roughness length need not be_ reproduced j_n_ the  model.   Objects
about the same size as_ the roughness length need  not  be_ reproduced jm^  geometrical
form but a.n_ equivalent roughness nuisj^ be_ established.  Over-roughening may be_
required to_ satisfy the roughness Reynolds number criterion.
     Boundary conditions in the fluid model are set by  simulating the  atmos-
pheric boundary layer.  Practical goals and techniques for  simulating  the
atmospheric boundary layer are summarized in Section  3.2.4.
                                     41

-------
                         3.  PRACTICAL APPLICATIONS

     The fundamental  principles of fluid modeling have been discussed in the
previous chapter.  When it comes to the details of a particular model study,
however, many decisions must be made, and the fundamental  principles frequen-
tly do not provide enough guidance.  It is the aim of this chapter to cover
the most common types of problems encountered by a modeler when designing a
particular model study, and to provide rational guidelines where possible or
to cite common practice where there is no rationale.
     The following sections discuss in detail  the special  problems encountered
1n modeling plume rise, the atmospheric boundary layer, flow around buildings,
and flow over complex terrain.  Each of these  sections is  summarized with a
set of recommendations.
3.1 PLUME RISE AND DIFFUSION
    Numerous investigators have studied the rise of plumes from model  stacks.
Many different kinds of facilities, including wind tunnels, water tanks, tow-
ing tanks, water channels, and even the calm stably-stratified environment of
an ice-skating rink, have been used.  The water tanks and the ice-skating rink
have been used to study the behavior of plumes in calm environments, both

                                      42

-------
stratified and unstratified.  The wind tunnels, water channels and towing tanks
have been used to study the behavior of plumes issuing from stacks into cross-
winds.  The effluent has ranged from pure jets to strongly buoyant plumes.
The crosswinds have ranged from neutrally stratified with uniform velocity
profiles to simulated atmospheric boundary layers (stable and unstable strati-
fication with, for example, logarithmic velocity profiles).
    The historical development of modeling techniques concerning plume rise
is analogous to the historical development of theoretical formulas for the
prediction of plume rise, I.e., the effluent buoyancy was thought to be neg-
ligible in comparison with its momentum.  Sherlock and Stalker (1940) appear
to have done the first wind tunnel study relating to plume behavior.  Specif-
ically, their experiments established the rule that the effluent speed must
exceed one-and-one-half times the wind speed in order to avoid downwash in the
lee of the stack.  This "one-and-one-half-times" rule is still widely applied
today.  They worked with a 1:300 scale model, but chose to use essentially
identical model and full-scale values of wind speed and effluent temperature.
One of their conclusions was "...the temperature of the stack gas is relativ-
ely unimportant as a means of controlling the downwash...".  At the present
time, it is still not clear exactly what effect buoyancy does have on the one-
and-one-half-times rule, but is evident that the buoyancy was not properly
scaled 1n the Sherlock and Stalker experiments.  Their experiments, even with
the very hot (400°F) effluent, were highly momentum-dominated plumes (effec-
tively jets), and corrections were made only for the change of momentum due
to change in temperature (density).   More recent work (cf. Huber et al., 1979)
Indicates that the buoyancy per se of the lighter effluent is ineffectual  in
preventing downwash; instead, the decrease 1n density alone contributes to
                                     43

-------
downwash because it reduces the effluent momentum.
     Numerous model studies have been conducted since those of Sherlock and
Stalker, but very few of the results have been compared with atmospheric data
or even with other model results.   Much worse, there has not been a  uniform
application of similarity criteria.  Each investigator appears to apply a dif-
ferent set of rules which ensure that his experiment models the rise of a
plume in the atmosphere.  It is evident after only  a little study that some of
these rules are conflicting and that all of them cannot be correct.   Indeed,
this is currently a highly controversial topic that has generated much discus-
sion, correspondence and further experimental work  over the past 3 to 4 years.
Several groups of experimenters claim to have done  "the definitive tests", but
as yet, a consensus on modeling buoyant plumes has  not emerged.

3.1.1  Near-Field Plume Behavior
     Let us consider the simplest conceivable problem of a plume downwashing
in the lee of a stack because the effluent contains insufficient momentum to
overcome the low pressure suction due to the crosswind (See Figure 8).
             Figure 8.  Plume downwash in the wake of a stack.
                                      44

-------
We will suppose that the stack walls are thin  and smooth,  the  stack  is  tall
and the effluent has the properties of air at  the same temperature as the
surroundings, i.e., D /D.=l, H /D»l, p /p =1, Ap=0 and Fr=°°.   Further,  since
                     01     s        s  a
we are concerned only -with the local flow field near the top of the  stack, we
have purposely omitted shear in the crosswind  (9u/8z«U/D)  as  well as strat-
ification and turbulence in the approach flow.  (Indeed, turbulence  in  the
approach flow may well affect the flow characteristics in  the  wake of a  cyl-
inder  (Goldstein, 1965, p. 430), and for a proper model  study, both  the
intensity and scale of the turbulence in the approach flow near the  top  of
the stack should be matched, but for purposes  of the present discussion, this
effect is ignored.)  To model this problem, we must match  only two parameters,
the ratio of effluent speed to wind speed and  the Reynolds  number:
                            WS/U,  WsD/v.
As discussed in Section 2.2.2.2, provided the  Reynolds number  is larger  than
some critical value, its precise magnitude is  irrelevant.   There are really,
however, three Reynolds numbers in this problem, corresponding to three  dif-
ferent classes of flow: one for the flow inside the stack  W D/v, one for the
jet issuing from the stack and entraining ambient air (also WsD/v),  and  one
for the flow around the outside of the stack UD/v.   The critical Reynolds
numbers may differ because the classes of the  flows differ. The critical Re
for pipe flow assuredly differs from that of a two-dimensional cylinder  wake.
There appears, however, to be considerable disagreement concerning the  partic-
ular values of these "critical" Reynolds numbers.
                                     45

-------
     Let us consider first the flow inside the stack and ascertain how we
would obtain similar velocity profiles at the stack exit.  Because full scale
stacks result in very large Reynolds numbers (a typical  small  valve might be
10 ), and if we assume the flow is fully developed (our H /D»l  assumption),
we can predict the shape of the velocity profile very closely  (i.e., from
Schlichting, 1968, Fig. 20.2).  By reducing the Reynolds number  in our model,
it may be seen that the shape of the velocity profile will  change only mar-
ginally; even though the resistance coefficient \ changes from 0.012 at Re=106
to 0.04 at Re=4X103 (ibid, Fig. 20.1), the power law exponent  a  in the velocity
profile U/Umx=(y/R)a changes from 0.12 to 0.17.  (From another viewpoint, the
ratio of the mean to maximum velocity changes from 0.85  at  Re=106 to 0.79 at
       o
Re=4X10 .)  In view of other possible uncertainties in the  problem, such as
knowing the real shape of the velocity profiles Inside the  stack (i.e., do
we have the 25 to 40 diameters required for full  development of  the velocity
profile), these small  changes with Reynolds number are regarded  as insignif-
icant.  Hence, we conclude that similar exit velocity profiles will  be obtained
at model Reynolds numbers of 4000.  Indeed, the lower limit would appear to be
2300, i.e., the value  required for the maintenance of turbulent  flow in a pipe.
    Note that, if our  full scale stack had had rough internal  walls, we could
have come even closer  to matching the full scale velocity profile by roughen-
ing the inside of our  model stack.  For example,  for a value of  R/k =30.6 (R
is the pipe radius, k$ is the Nikuradse equivalent sand  grain  roughness), the
resistance coefficient \ would vary from 0.045 at Re=106 to 0.040 at Re=4X103
(ibid, Fig. 20.18) and the shape of the velocity profiles would  have been es-
sentially identical.
                                      46

-------
     Let us consider second the flow immediately outside the stack.   Ricou

and Spaulding (1961) have shown that the entrainment rate of momentum-domin-

ated jets in calm surroundings is essentially constant for Reynolds  numbers

in excess of 25,000.  Only minor variations were observed between 15,000 and

25,000.  Substantial variations were observed below 10,000; the entrainment

rate was increased by more than 20%.  Hence, if minor errors are acceptable,

this critical Reynolds number is 15,000.

     Let us consider third the ambient flow around the outside of the stack.

It is useful here to consider the changes that occur in the flow pattern

around a circular cylinder as the  ,-ynolds number is increased (for  additional)

details, see Goldstein, 1965).  At very low Re (<1), the streamline  patterns

are symmetrical  fore and aft of the cylinder.  As Re is increased (^10), two

symmetrical standing vortices are formed at the back; they grow in size and

are stretched farther and farther downstream until at Re^lOO, they break down

and are shed alternately at regular intervals from the sides of the  cylinder.
                                                           O      C
This type of flow persists over a very wide range of Re (10 •
cylinder:  one where the Karman vortex street (shedding vortices) is formed

                                      47

-------
(lower) and one where the boundary layer becomes turbulent (upper).   At still
larger Reynolds numbers, the transition to turbulence occurs  earlier in the
boundary layer, so that the separation point moves  forward on the  cylinder,
and the drag coefficient increases gradually.
     The drag coefficient in this instance is quite important,  because  it  is
a reflection of the pressure distribution on the lee side  of  the stack  and
will have fairly strong influence on the downwash of the plume  (i.e., we can-
not ignore large variations in the drag coefficient as we  did in the case  of
the resistance coefficient for pipe flow).  If the  full  scale Re is  less than
about 105, then it is only necessary that the model  Re exceed the  lower cri-
tical Re of about 400, because the drag coefficient is essentially constant.
For a rectangular stack, because of the sharp corners forcing separation,  the
lower critical Re would be lower still.
     Modeling downwash around a full  scale stack where Re^lO5,  however,  is
much more difficult.   It may be accomplished by ensuring that the  model  Re
exceeds the upper critical Re of 10  which would require a large and/or fair-
ly high speed wind tunnel (i.e., a stack diameter of 10cm  and wind speed of
15m/s).  It is also possible to simulate a higher Re using the  common wind
tunnel practice of tripping the boundary layer, either through  use of a  trip
wire or by roughing the surface of the cylinder, thereby forcing the boundary
layer to become turbulent (Goldstein, 1965, Figs. 162 &  163).   This  technique,
however, would gain at most a factor of two in Reynolds  number; as the  size
of the roughness elements or wire diameters increases, the sharp drop in drag
coefficient occurs at lower Re, but the magnitude of the drop is progressively
diminished; at Re £ 30,000, the roughness and wires  are  completely ineffective.
For a rectangular stack, because of the sharp corners forcing separation,  the
upper critical Re would be much higher, possibly non-existent.
                                     48

-------
     First-hand experience with this problem was gained by Tunstall  and
Batham (Robins, private communication).   They followed on from Armitt's (1968)
work with cooling towers and investigated the feasibility of surface roughen-
ing to reproduce very high Reynolds number flow around model  chimneys in sim-
ulated boundary layers.  Their aim was to reproduce field measurements of the
pressure field on the Fawley Power Station chimney (200m x 20m, Re ^ 3x10 ).
This was achieved, but for scale reductions between 250 and 500, it was found
necessary to roughen the model with sand of about 0.5 mm dia.  and to operate
                          c
the tunnel so that Re ^ 10 .
     It should be noted that if we were  not concerned with entrainment of
effluent into the wake, i.e., if we wanted to model a non-downwashed plume,
then the cylinder Re would be relatively unimportant in any event, so that
we need be concerned only with the critical Re (15000) for the effluent as
it exits the stack under ordinary circumstances.  This simplified problem
would be relatively easy to model even in a fairly small wind  tunnel (say,
0.5m square test section) with moderate  wind speed (^20m/s) and a small stack
(1 cm dia.).  We will shortly see, however, that if the effluent is  buoyant,
the problem becomes much more complicated.  It will not be so  easy to obtain
such a large Reynolds number, and we must look harder to determine if the
15000 value for the critical Reynolds number can be reduced.   We will return
to our discussion of critical Reynolds numbers later in this  section.
     Notice that, provided the Reynolds  number exceeds 15,000, there is only
one parameter of importance, WS/U.  Since full scale stacks and effluent
speeds (even fairly small stacks and low speeds) result in huge Reynolds
numbers, the full scale flow is Reynolds number independent.   This implies
that the size and shape of the wake behind the stack and the amount  of
downwash depend on only one parameter, W /U, and not on wind speed per se.
                                      49

-------
Similarly, provided the Reynolds  number is  large  enough,  the  flow  structure
in a model in a wind tunnel  is  similar to that  of the  prototype  and  is  indep-
endent of wind speed per se.   (This  discussion  may appear obvious  and there-
fore trivial, but in demonstrating wind tunnel  experiments to novices,  or
even frequently to accomplished experimentalists, the  inevitable question  is:
to what full scale wind speed does this flow correspond?   The correct answer,
i.e., all wind speeds above the barest minimum, invokes  puzzled  glances and
disbelief 1)
     Now let us complicate the problem, one step at a  time, to see what addit-
ional issues arise in the modeling.   Suppose the effluent is  of  high tempera-
ture, so that its density is, say, half the ambient air  density.  In a  wind
tunnel, it is usually easier and more practical to use a  lighter gas to simu-
late this high temperature field effluent than  it is to  heat  the model  effluent.
Similarly, in a water channel or tank, it is usually easier to use salt water
or alcohol than to heat or cool the  model effluent. This low density manifests
itself in two opposite ways:   first, at a fixed effluent  speed,  the  effluent
momentum flux is reduced, tending to make the plume more  easily  bent-over,
thus promoting downwash, and second, the buoyancy of the  effluent  is increased,
tending to inhibit the downwash.   It is not clear which  is the more  important
effect.  Overcamp and Hoult (1971) showed rather convincingly that the  effect
of the increased buoyancy was to inhibit downwash of cooling  tower plumes,
                                    1 /2
where the Froude number W /(gDAp/p ) '   ranged  from 0.2  to 2.  Huber et al.
                         s        a
(1979), however, observed enhanced downwash as  the effluent density  was
                                      so

-------
decreased.   The Froude numbers  in  their experiments were  greater  than  4, which
is more typical of power plant  plumes.   It  would  appear that  the  crossover
point where the effect of the lower density switches  from inhibiting downwash
to enhancing it occurs at a Froude number around  3; however,  it must also be
a function of the effluent momentum to  crosswind  momentum ratio.
     Most investigators would agree that the following set of parameters to
be matched for this more complex problem are sufficient (although,  perhaps,
not all are necessary):
                        ws   PS          w
                         5>    •*   Do  	-	,- „
                        _,_,Re, -       -r/2                  (3J)
     Since products of similarity parameters  are themselves  similarity
parameters, the following set is fully equivalent to that above:
    _
5" > ~
    p
                                               .
                                       (gDAP/pa)
                                                                    (3.2)
     The first parameter expresses the ratio of effluent momentum flux to cross-
wind momentum flux, and must be matched if the initial  bending or the rise due
to the initial momentum of the plume is important.   The last parameter is the
Froude number, which expresses the ratio of inertia! to buoyancy forces in the
effluent.  (Note the different interpretation of Fr here as opposed to its
characterizing the stratification in the approach flow  as it was introduced
in Section 2.2.4.)
     A questionable parameter is the density ratio Ps/Pa per se.  As mention-
ed previously, the density difference manifests itself  through its effects on
                                     51

-------
the effluent momentum and the effluent buoyancy, which are expressed in the
first and last parameters of Eq.  3.2.   Of course, it is perfectly acceptable
to match the density ratio between model  and prototype, but if it is not an
essential parameter, then the full capabilities of modeling facilities will
not be realized.  It is frequently advantageous to exaggerate the density
differences in the model in order to achieve low Froude numbers.   Ricou and
Spalding (1961) showed very convincingly  that the rate of entrainment dm/dx
in a highly momentum-dominated jet (no crosswind) obeys the relation
                        1  dm _ r
                        ~~  j,, ~~ \f
                        P
                         a
                           dx
                                      1/2
WSD ,                     (3.3)
where C is a constant.  Hence the entrainment rate is a direct function of
the density ratio.  For small density differences, the entrainment rate is
not much affected.  However, if helium (S.G.  = 0.14) were used as the buoyant
effluent from a stack in a wind tunnel, maintaining geometric similarity and
the ratio of effluent to wind speed WS/U, to  simulate a full  scale effluent
with specific gravity of 0.7, then the entrainment rate would be halved.  (We
will refer to this phenomenon as an "impedance mismatch", analogous to its
usage in acoustic engineering as the ratio of pressure to volume displacement
in a sound-transmitting medium.)  Hence, its  rise due to initial momentum
would not be correctly modeled.  Considerations such as these are evidently
what lead Hoult (1973) to state that the density difference Ap/p  must not
                                                                a
exceed 0.4.  But such a statement cannot be made unequivocally.  If the den-
sity difference in the field were 0.5, as might be the case for a gas turbine
exhaust, then it would certainly be desirable to exceed 0.4 in the model.
                                      52

-------
More importantly, as shown by Eq.  3.3, it is  certainly possible to exaggerate
the density difference to 0.8 by using essentially pure helium as  the  model
effluent and still  maintain the same entrainment rate by increasing the ef-
fluent flow rate W .  But Eq. 3.3 of Ricou and Spalding applies only beyond  a
few diameters beyond the stack exit (say >10D).   In order to avoid an  "imped-
ance mismatch" in our downwash problem, where we are concerned with the flow
behavior right at the top of the stack, it is necessary to match the density
ratio.  Beyond a few diameters, it is only essential to match the momentum
flux ratio.
     A major point of disagreement among investigators concerns the definition
of the Froude number.  One group of investigators define the Froude number
with the effluent density as the reference density, Frg; another group defines
it with the ambient density (at stack top) as the reference density, Frfl.
Yet, nowhere is any reason given for the particular choice.  It might  appear
at first glance that the choice is completely arbitrary.  However, given that
two plumes have the same Froude number based on effluent density,  it is not
necessarily so that they have the same Froude number based on ambient  density.
Consider the example (Table 1) of a typical power plant plume being modeled
at a scale of 1:400 using helium as the buoyant effluent in a wind tunnel.
     The Froude numbers differ by a factor of the square root of the density
                         1 /2
ratios, i.e., Fr =(p /p) ' Fr , so that unless PC/P. is the same in model
                a   a  s      s                  bo
and prototype, the choice of the definition of the Froude number is not arbi-
trary.  Yet, almost all investigators exaggerate the density differences in
the model in order to obtain large enough buoyancy in the plumes (low Froude
numbers).  They do not match P../P,, as required by Eq. 3.2, so that it is not
                              s  a
possible to match both Froude numbers simultaneously.
                                      53

-------
      TABLE 1:  TYPICAL PARAMETERS FOR MODELING PLUME DOWNWASH.
Parameter
Ws
g
D
pa(V
PS(TS)
Fra
Frs
Re
Prototype value
20m/ s
9.8m/s2
10m
1.2g/l(20°C)
0.83g/l(150°C)
3.6
3.0
13xl06
Model value
1.67m/s
9.8m/s2
2.5cm
1.2g/l(20°C)
0.17g/l(20°C)
3.6
1.37
360
     A third group of modelers define the Froude number using the wind speed



at stack top rather than the effluent speed.   We will  designate this Froude



number as Fr;; or FrjJ.  Note that if Wg/U is matched (Eq.  3.1),  then matching



Fra is equivalent to matching Fr  and matching Fr  is  equivalent to matching
  a                             as

           22                             U
Fr : if p,W_/p U  is matched, then matching Fr  is equivalent to matching
  s      s s  a                               a


Fr  and matching Fr  is equivalent to matching (p_/pj '  Fr . which is not
  s                s                             s  a       s


the same as Fr.
              a


     This is a particularly vexing problem because most plume-rise theories



are founded on the assumption of small density differences, so that all



Froude numbers are essentially equivalent. There are  only a few places  in



the literature providing guidance on which Froude number is the most approp-



riate.  Hoult et al. (1977) stated that two complete,  independent wind tunnel



tests were run, one using the ambient density and the  second using effluent



density as the reference.  The tests involved the modeling of gas turbine



exhausts, which generally involve large effluent velocities and high effluent



temperatures.  They claim (unfortunately, without presenting any data to

                                      54

-------
support their claim) that the better choice is ambient density and that the
error between model and field observations (presumably, of far-field plume
rise) when using effluent density was about 90$, which was nearly 10 times
the error incurred using ambient density.  It is important to note that using
ambient density as a reference corresponds to using effluent temperature as
a reference, i.e.,
                               VTa   pa"ps ,                       (3.4)
                                 Ts  =   pa
due to the perfect gas law at constant pressure, pT=const.
     A physical interpretation of the difference between the two definitions
is suggested in a footnote by Briggs (1972) (his comments applied specifically
to alternative definitions of buoyancy flux, but are equally applicable here):
"the difference — amounts to different approximations for the effective den-
sity (inertia per unit volume) of the fluid being driven by the buoyant force:
       (1) that the effective density is approximately constant=p , which is
       reasonable very close to the stack, say within a few stack diameters
       downwind;
       (2) that the effective density is approximately constant=p , which is a
       better approximation at all larger distances."
     It is apparent, then, that in our stack downwash problem, we should use
the effluent density as the reference density in matching of Froude numbers.
It is also apparent that, if we were attempting to model far field plume rise,
we should use the ambient density as the reference density, in agreement with
Hoult et al. (1977).  Fr  is used as a scaling parameter by Isyumov et al.
(1976), Cermak (1971, 1975), Melbourne (1968) and Barrett (1973).
     A second point of disagreement among modelers has to do with whether the
relevant parameter is the ratio of effluent speed to wind speed HS/U or the
                                      55

-------
                                                              2      2
ratio of effluent momentum flux to crosswlnd momentum flux  peW_/(p,U ).   This
                                                            5  b    a
may also be thought of as a ratio of dynamic pressures.   Again,  if PP/P,  is
                                                                    S   a
matched between model  and prototype, the choice is  arbitrary.  As  discussed
above, however, most modelers drop the requirement  of matching the density
ratio, and the choice is no longer arbitrary.  Sherlock  and Stalker  (1940)
found that the behavior of the plume depended upon  the ratio of  the  momenta
and that the ratio of the two speeds was a close approximation,  provided  that
both velocities were reduced to equivalent velocities at a  common  temperature.
Their one-and-one-half-times rule was thus based on the  momentum ratio, a
fact not appreciated by most authors who quote the  rule. The  recommendation
made here, then, is that the relevant parameter is  the momentum  ratio, and
not the speed ratio per se.
     A third problem is that in water tanks or channels, a  heavy salt solution
is commonly used to simulate a buoyant effluent by  inverting the stack and
exhausting the salt solution into lighter fresh water.  The same principle
could conceivably be used in a wind tunnel by using a heavier-than-air gas
such as freon with an inverted stack.  There is a subtle question here that
has not been fully answered.  In the field, a lighter effluent entrains
heavier air, whereas in the water tank, a heavier effluent  entrains  a lighter
ambient fluid.   It is conceivable that the entrainment mechanisms could be
significantly altered due to this interchange of heavier and lighter fluids.
Eq. 3.3 indicates that a heavy fluid issuing from an inverted stack can be
used to simulate a lighter fluid from an upright stack, if the effluent
speed W  is appropriately reduced.  As argued previously, there  may be a
subtle effect on the entrainment very near the stack, but this disappears
quickly as the density difference is rapidly diluted.  The total rise of the
plume is not highly sensitive to the entrainment parameter; the forced plumes
                                      56

-------
of Hoult and Weil  (1972) and Lin et al.  (1974),  both  using  salt water effluent,



appear to simulate field results quite well.   It is clear that the  problem  is



not yet completely answered and requires  detailed systematic  study.   A tenta-



tive conclusion, in view of other inherent inaccuracies  in  modeling at small



scales, is that this subtle difference may be overlooked.



     It was shown in the early part of this section that a  critical  Reynolds



number of 15,000 was not difficult to achieve per se.  However, the introdu-



tion of buoyancy makes this Re  much more difficult to attain.  In  order to
                              \f


match Froude numbers, it was essential to introduce helium  as the effluent



(which, incidentally, has a kinematic viscosity  approximately 8 times that



of air) and to reduce the effluent speed  by a factor  of  12  (see Table 1),



so that the effluent Reynolds number was  only 360.  Thus, we  must determine



whether a lower critical Reynolds number  can  be  justified.



     The data of Ricou and Spalding (1961), which suggested Re =15,000,  was
                                                             \f


applicable to momentum-dominated jets in  calm surroundings.   Most investi-



gators would agree that for a bent-over plume, the critical value may be



substantially lower, of the order of 2300, i.e., a value that is well-estab-



lished for the maintenance of turbulent flow  in  a pipe.   This is equivalent



to saying that the plume behavior is independent of Reynolds  number provided



that the effluent flow is fully turbulent at  the stack exit.  Lin et al.



(1974) have taken this one step further.   They tripped the  flow to  ensure



that the effluent was fully turbulent at  the  stack exit  at  a  Re of  530 by



placing an orifice with opening D/2 inside the stack  and located 3D from the



exit.  Their data for (1) the terminal rise of a buoyant plume in a calm and



stably-stratified environment and (2) the trajectory  of  a buoyant plume  in  a



stably-stratified crosswind compared reasonably  well  with other laboratory



and field data.  Hoot, et al. (1973) and  Nakai and Shikata  (1977) have used


                                      57

-------
similar techniques.   Liu and Lin (1975),  however,  indicated  that  the  place-
ment of the orifice  relative to the top of the stack  was  critical  at  a  stack
Re of 290.  If the distance was smaller than  required,  the effluent flow was
governed by the orifice diameter and not  the  stack diameter; if larger, the
flow tripped by the  orifice would laminarize  before it  reached the stack exit.
     Isyumov and Tanaka (1979) reported on the influence  of  the shape of the
velocity distribution on subsequent plume behavior.  The  normally parabolic
(laminar) velocity profile of the effluent stream  was "flattened"  using a
somewhat larger diameter stack with a short contraction to the required in-
side diameter at the top of the stack. Experimental  data presented showed
the effects of this  improvement in the shape  of the velocity profile  to be
small.  The Reynolds numbers, however, were approximately 30, so  that the
"flattened" profile  was assuredly laminar.  In line with  our previous dis-
cussion (see also discussion in Section 3.1.3.1) requiring a turbulent  flow
at the exit, this technique is not recommended.
     Wilson (private communication) reported  that  the shape  of the stack exit
velocity profile as  well as its turbulence level was  important in determing
near-stack effects.   He inserted a perforated plug in the stack three dia-
meters from the exit to help produce a flat velocity  profile at the exit with
a turbulance intensity of about 20%.  His experiments were conducted  in a
simulated neutral atmospheric boundary layer.
     Briggs and Snyder (1980) tested the  rise of jets and buoyant plumes in
a calm, stably stratified salt water tank.  Data relevant to establishing
critical Reynolds numbers are presented in Figure  9,  where it may be  seen
that both the maximum (overshoot) heights and equilibrium (final  heights
are Reynolds number independent for large enough Reynolds numbers. For jets,
                                      58

-------
         0
         100.0
         10.0
     1000.0               10000.0
           REYNOLDS NUMBER

(a)  Neutrally buoyant plumes
                            T
                                                  Ahn
                                                  (F/N3)1/4
                                            Aheq
                                         Ah
                                                1
     100.0                 1000.0

           REYNOLDS NUMBER
                                          100000.0
                                                                  10000.0
                            (b)   Buoyant  plumes
Figure 9.   Variation of  plume rise with Reynolds number (Ah  =max1mum
                                                               niA
height reached by plume, Ah  =equ111br1um plume height, Resource
momentum flux, F=source  buoyancy flux,  N=Brunt-Va1sala frequency).
                                      53

-------
the critical Reynolds number was approximately 2000; for buoyant plumes, it
was approximately 200.  Below these critical values, the plumes were laminar
at the stack exit, with resulting rises too high (the upward trend of the
curves at low Reynolds numbers).  Limited attempts at tripping the flow in-
side the stacks yielded unpredictable results: sometimes the trips were
ineffective in establishing turbulent flow; in other cases, the trips were
overly effective: they enhanced entrainment to such a degree that the re-
sultant plume rise was too small (the, lower bound of the shaded areas in
Figure 9a).
     Experiments with buoyant plumes in neutrally stable crosswinds, conducted
by Hoult and Weil (1972), show: (1) at a Reynolds number greater than 300, the
plume appears to be fully turbulent everywhere; at lower Reynolds numbers,
the plume becomes turbulent only some distance downstream of the exit (there
was apparently no tripping of the flow inside the stack), (2) ignoring scatter
in the data, no dependence of far field plume trajectory on Reynolds number
was observed for Reynolds numbers between 28 and 2800,  (3)  the vertical  plume
width was substantially reduced close to the stack exit (within 10 stack dia-
meters downwind) for Reynolds numbers below 300.
     It is difficult to reconcile the results of the various sets of experi-
ments.  There are numerous possible reasons why the critical Reynolds numbers
are different.   The effluent flows differ (momentum-dominated versus buoyancy-
dominated), the stratification of the ambient fluids differ (neutral  versus
stable stratification), and in some cases, the jets issued  into calm environ-
ments, whereas  in others, the plumes were bent over by  a crosswind.   The two-
order-of-magnitude difference in critical Reynolds number,  however,  is
difficult to explain.  It is evident that a basic systematic study needs to
                                     60

-------
be undertaken to establish that Reynolds number (perhaps  different ones  for
different sets of conditions) above which the rise and spread  of model plumes
is independent of Reynolds number.
     Equation 3.1 provides 4 basic  parameters that, most  investigators would
agree, provide for a complete or "exact" simulation of buoyant plume rise,
provided that a minimum Reynolds number requirement is met.  However,  if the
density difference is exaggerated,  a "Pandora's box" is opened, allowing a
multitude of different interpretations and different sets of similarity  par-
rreters.  Table 2 provides a list of these sets of parameters as used (or at
least suggested) in various laboratories.  Even though each  of these sets is
unique, there are many more possible sets.
     Some of these techniques have  been "proven" through  comparison with field
data, albeit on very limited bases, and others have been  "proven" or "dispro-
ven" by comparison with "exact" simulations, i.e., matching  the 4 parameters
of Eq. 3.1, although it is not always clear that minimum  Re  requirements have
been met.  Notice that the CALSPAN  technique does not match  any Froude numbers;
they have combined the 4 basic parameters of Eq. 3.1, in  a different fashion
to match buoyancy and length scales as used in Briggs (1969) plume rise  eq-
uations (this will be discussed further in Section 3.1.3).  In our near-field
problem, the flow at the mouth of the stack is important, so that exaggeration
of the stack diameter is unacceptable.  Of the remaining  techniques, only 3
have been "proven" (Robins, 1980; Isyumov and Tanaka, 1979;  Hoult et al., 1977;
Wilson, private communication), but these tests have consisted primarily of
far field comparisons, i.e., maximum ground-level concentrations or plume rise
at x-lOH .  The one near-field comparison of the CALSPAN  technique "disproved"
its validity (Isyumov and Tanaka, 1979), but these results are questioned as
                                      61

-------
           TABLE  2:   TECHNIQUES USED  FOR SIMULATION  OF  BUOYANT  PLUMES

                         AT VARIOUS FLUID MODELING  FACILITIES.
SCALING
PARAMETER
Density ratio
WP,
a
Fra
Ws/(gDAp/p.)}1/2
^
lV(9DAp/P )]/2
d
Frs
ws/(gD/iP/Ps)1/2
^
U/(gDAp/ps)'/2
Velocity ratio
ws/u
Momentum ratio
pswf/(P/)
Momentum ratio
for distorted stack
"sD2wf/(PaH2U2)
Mass flow rate
P/Ws/(paHs2U)
Volumetric flow rate
D2Ws/(H2U)
Buoyancy length
VHs=gDipWsD/(4U3paHs)
Buoyancy length
^/Hs=gDAPWsD/(4U3psHs)
Momentum length
VHs=psWsD/<4Pau2V
Geometric scale
D/HS
Proven
Disproven
"EXACT"
SIMULATION
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
X
,
APPROXIMATE SIMULATIONS
UWO1.. CSU2/MIT3 CALSPAtl4 MONASH5 BRISTOL6 CERL7 CERL8 CERL9 CERL10
(DMA)" (DVA) (BMA) (EMS) (DMS) (DVS) (BMD)
	 	 	 1
10 no no no no no no no yes
no yes no no no no no no no
yes yes no yes yes no no no yes
yes no no yes yes no no yes no
no no no no no no yes yes yes
no yes no no no no no yes no
yes no yes yes yes yes yes no no
yes no yes no no yes yes no yes
no no no yes no no no no no
no yes no no yes no no yes no
no yes yes no no no no no yes
no no no no no yes no yes yes
yes no yes no no yes yes no no
yes yes yes no no yes yes yes no
X XXX X
XX XX X X X XX
1.  University of Western Ontario, Isyumov et al., 1976.  2.   Colorado  State University, Cermak,  1971, 1975.
3.  Massachusetts Institute of Technology, Hoult et al., 1977.  4.  Calspan Corp., Skinner and Ludwig, 1978.
5.  Monash University, Melbourne, 1968  (interpreted from Isyumov and Tanaka, 1980).  6.  Bristol  University,
Barrett, 1973 (interpreted from Isyumov and Tanaka, 1980).  7-10.  Central Electricity Research Laboratory,
Robins, 1980.  11.  Three-letter codes  in parentheses  are notations of  Robins, 1980.
                                                 62

-------
indicated in our previous remarks; they are questioned for other reasons  by
Ludwig (private communication).   Because of these uncertainties, none of
the approximate techniques can be recommended for modeling the near-field
rise of buoyant plumes.   The various  techniques  will  be discussed further in
Section 3.1.3.

3.1.2  Summary and Recommendations on Modeling Near-Field Plumes
3.1.2.1  The Stack Downwash Problem
     In summary, to model the problem where the  effluent is downwashed into
the wake of a cylindical  stack,  it is recommended that:
     1.  If the full-scale Reynolds number based on the wind speed and out-
         side diameter of the stack exceeds 10, it will be necessary to
                                       5
         exceed a Reynolds number of 10  in the  model also.  Some small
         relaxation of this requirement (at most, a factor of 2) may be pos-
         sible through the use of roughness or trip wires to force a turbulent
         boundary layer on the cylinder surface.
     2.  If the full-scale Reynolds number (as in 1)  is less that 10 , it will
         be necessary that the model  Re exceed 400.
                                                   1 /p
     3.  The parameters W_/U, p/pa and Wc/(gD/Ap ) '   must be matched between
                         5     S  cl      S       o
         model and prototype.  It is  expected that if Fr >4, the density  dif-
                                                        a
         ference tends to enhance downwash, so that a mismatch of Froude
         numbers would lead to conservative results;  if Fr <4, the density
                                                          a
         difference will  tend to inhibit downwash, so that matching all para-
         meters is important.
         Since there is only one possible reference length in this problem,
the stack diameter, it determines the scale ratio, and geometric scaling
is implicit, i.e., all lengths should be referenced to the stack diameter.
                                      63

-------
Other lengths that may be important are the stack wall  thickness  and  the
stack height.  Obviously, these lengths should be scaled with  the stack
diameter.

3.1.2.2  The Near-Field, Non-Downwashed Plume Problem
     To model the problem where the plume is expected to disperse in  the  pres-
ence of aerodynamic effects of buildings, etc. (but not in the wake of the
stack itself), it is recommended that:
     1.  The effluent Reynolds number be as large as possible.
         (a)  Fix the effluent Reynolds number to be as large  as  possible,
              preferably greater than 15,000.
         (b)  If it is necessary to reduce the effluent Reynolds  number below
              2300, it may be necessary to trip the flow to ensure a  fully
              turbulent exhaust.
         (c)  If it is desired to reduce the effluent Reynolds number below
              300, it will be necessary to do some experimentation to deter-
              mine under what conditions the plume will simulate  the  behavior
              of a plume in the field.
     2.  The set of parameters to be matched (equal in model and  prototype) is:
                        WS/U, ps/pa, Ws/(gDAP/ps)1/2.
     This "exact" simulation will generally limit the scale reduction to less
than 400.  Should the scale reduction exceed 400 or other techniques  be des-
ireable, they should be  "proven" by detailed comparisons with "exact"
techniques.
     Again,  the stack diameter determines the geometric scale ratio,  and all
other  lengths, such as building height, should be reduced by the same fraction.
                                      64

-------
3.1.3  Far-Field^ Plume Behavior
     The previous section reviewed criteria to be met for modeling plumes
close to the top of the stack (say, less than a few stack heights downwind).
To model plumes farther downwind of the stack, it is obvious that even greater
reductions in scale are required (larger geographical areas to be modeled),
and "exact" scaling will be difficult if not impossible to satisfy.  The
question to be answered in this section, then, is whether further compromises
can be made without making the results unduly suspect.
     As an example, suppose we wish to model a power plant in complex terrain,
where the scale reduction factor is 1:5000.  Typical conditions from the plant
operations record might be T =540°K, T =300°K, W =25m/s, U=10m/s, D =10m.  Sup-
                            s         a         s                  s
pose we try to match the "exact" conditions from the previous section using
pure methane as the model effluent (P./P =0.56) in a wind tunnel.  The model
                                     S  a
stack diameter would be 2mm.  The Froude number of the full scale effluent is
8, which implies a model effluent speed of 0.35m/s.  These conditions yield a
stack Reynolds number of
                           w n
                      D«    s    35cm/s x 0.2cm   ..
                      K6 - —             x       tH
                        s   v       O.lGcnT/s
considerably below the value recommended in the previous section.
     There are several directions available at this point.  Notice that the
problem was not created because of the matching of density or velocity ratios,
although the velocity requirement may later cause problems in obtaining a
minimum Reynolds number based on the roughness of the underlying terrain (see
Section 3.2).  The problem was caused because of the matching of Froude num-
bers.  This problem has been attacked in a variety of ways.

3.1.3.1.  Ignoring the Minimum Reynolds Number
     Ludwig and Skinner (1976) ignored the minimum Reynolds number require-
                                      65

-------
ment; thus, their plumes were laminar in the immediate vicinity of the stack
(see Figure 10).  Discussion in their report admitted that the rise of an
initially laminar plume would exceed that of an initially turbulent plume
because the turbulent one mixes more rapidly with the ambient air.  They felt
however, that this was not a serious limitation in their model because their
initial plume rise was quite small  before atmospheric turbulence began to
dominate the mixing process.  It is evident from Figure 10, however, that the
scale of the atmospheric turbulence is considerably larger than the initial
plume diameter, so that the plume trajectory is highly contorted, but little
real mixing of effluent with ambient air occurs for many stack heights down-
stream.  If the effluent plume were turbulent, it would be diluted very rapidly
(within a few stack diameters) by ambient air.  The resulting plume rise could
be substantially different in the two cases, depending on the precise effluent
parameters.  Ludwig and Skinner did not feel that tripping of the flow within
the stacks was possible because the stack diameters ranged from 0.25 to 1.3mm
and there were 49 separate stacks in the model.  Liu and Lin (1975), however,
were able to use a sapphire nozzle  of 0.18mm dia. to trip the flow in their
one stack.  As mentioned in Section 3.1.1, the size and placement of the ori-
fice is evidently critical and will require special experimentation.
                                                             ••   -As
         Figure 10.  Laminar plume caused by low Reynolds number effluent
                     (from Ludwig and Skinner, 1976).
                                      66

-------
3.1.3.2  Raising the Stack Height
       Facy (1971) ignored the plume buoyancy, per se, but instead extended the
stack and bent-over the top such that the effluent was emitted at the same
elevation as that calculated from plume rise formulas.  This technique has the
advantage that flow Reynolds numbers can be made as large as desirable.  The
disadvantages, however, are obvious.  Since the plume rise is a function of
wind speed, there is a contribution to vertical dispersion due to both long-
itudinal and vertical fluctuations in the wind speed that cannot be simulated
via this method.  Also, the physical stack height in the model must be changed
to simulate different wind speeds.  But the most serious limitation is that
the complex trajectory of the plume, which may be the most useful information
obtained from the model, cannot be obtained using this method.  It is frequently
desired to determine whether a plume goes over the top, is diverted around, or
impacts on the surface of a hill.  If the plume is emitted into a different
mean streamline, its resulting trajectory could be entirely different.  Adding
momentum to the effluent to obtain the same rise as for a buoyant plume is ob-
jectionable for similar reasons.   This technique, however, might be acceptable
under some circumstances; for example, if the problem were to determine concen-
trations on an isolated hill far downwind of the source (beyond the point of
maximum rise), then it might be acceptable to inject the plume at its terminal
height.  Another problem is that one must presume to know the rise.  This may be
acceptable for stable flows, but is an unsettled matter otherwise.
3.1.3.3  Distorting the Stack Diameter
       Briggs1 (1969, 1975) equation for the trajectory of a plume with both
initial momentum and buoyancy, valid only for distances considerably smaller
than that to the point of maximum rise, is rewritten here in a different form:
                                      67

-------
Ah
*T
4HS
                                                4p/Hs
                                                                        (3.5)
                                                                        (3.6)
where 3-j and 82  are entrainment coefficients  (3^1/3+U/Wg  and  B2=0.6),
1  is a momentum length scale and K is  a buoyancy  length scale,  defined  by
Briggs (1975) as
                                        P*   LJ
                                     -      w.
                                                                        (3.7)
                                           IT
                                     -,2 W.
and
                                                          (3.8)
A physical interpretation of these length scales  is  that they represent  the
initial radius of curvature of the plume due to momentum and buoyancy respec-
tively.  In our example problem, 1 =9,3m and lp=2.72m.   It is evident from
examination of Eq. 3.6 that close to the stack, the  first term on  the right
hand side of this equation will dominate and far  from the stack, the second
term will dominate.  That is, close to the stack, the initial momentum will  be
important, whereas, ultimately, the buoyancy will dominate.
     Hoult (1973) suggested that we ignore the initial  momentum, provided only
that we avoid stack downwash, and take as our requirement
                            «W.-Wp  •                      (3-9)
where subscripts m and p refer to model and prototype, respectively.  He fur-
ther suggested that Eq. 3.9 could be met by exaggerating the density difference
and/or by reducing the effluent speed Ws-  Lin and Liu (1976) started at
                                      68

-------
essentially the same point, but suggested additionally that Eq.  3.9  could  be
met by exaggerating the stack diameter.   They performed an experimental  run
on a complex terrain model at a scale of 1:10000 using a stack diameter
exaggeration factor of 2.  Photographs show that the plume was turbulent at
the stack exit, even though the effluent Re was only 68, but no experiments
were made to validate the use of this method.
     It is conceivable from inspection of Eq. 3.5 that, by clever manipula-
tion, we could vary any_ and_ al_l_ of the parameters p$, pa, W$,  D, or  U in such
a fashion that the coefficients would not be changed, and, therefore,  that the
plume trajectory would be unchanged, i.e., it would not be necessary to  ignore
the momentum term—we could include it too.  This is equivalent to reducing
the momentum and buoyancy lengths by the geometric scale reduction factor.
Obviously, however, if we change D, we will also change the plume width  at
the stack exit (=D).  This violates our previous requirement of geometric
similarity; it may or may not have serious consequences, dependent upon  the
amount of the exaggeration and on the particular flow field.   It is  not  en-
tirely clear what extraneous effects may be introduced by the  manipulation of
the other variables.
     Basically, the coefficients of the x/H$ terms in Eqs. 3.5 and 3.6 are
products of similarity parameters, i.e.,

                                        1/2
       and
B.I

                                           >au'
                                                3/2
                               3/2
                                                                       (3.10)
                                                                       (3.11)
                                      69

-------
 If we insist on geometric similarity, then Eq. 3.10 is identical to our
 previous momentum matching requirement, and Eq. 3.11 is a product of our
 previous Froude number, momentum and density ratio matching requirements.
 There is no reason, a priori, to favor one or the other; the choice will
 require experimental verification.
     In a rather unusual derivation, Skinner and Ludwig (1978) have arrived at
 scaling laws that are essentially equivalent to matching the ratios of momen-
 tum and buoyancy length scales to the stack height, i.e., Eqs. 3.10 and 3.11.
 They also conducted some experimental work showing that "enhanced" scaling
 (exaggerated density differences) produced the same results as "restricted"
 scaling (matched density differences), with both tests done in the wind tunnel.
 Further, they have pointed out the possibility of exaggerating the stack dia-
 meter, but they have not conducted tests to verify this.
     A review of the literature shows that a wide variety of approximate
 techniques have been proposed and used (see Table 2).   The only technique
 that has been independently tested and "proven" in different laboratories is
 the CALSPAN technique, and it is recommended for that reason.
 3.1.4  Summary and Recommendations on Modeling Far-Field Plumes
     Most likely more important than the decision on matching the momentum
 and buoyancy length scales versus diameter exaggeration versus matching of
 the Froude number and the momentum ratio are the effects of the approach flow,
 i.e., the stratification and ambient turbulence to which the plume is sub-
 jected, as well as the effects of downwash and flow diversion or channeling
 caused by buildings and terrain.  These will be discussed further in later
 sections.  For the present, the recommendation is to avoid a nonturbulent
effluent flow and to avoid raising the stack,  either physically or through

                                     70

-------
the addition of momentum.  Instead, the most advantageous of the methods
discussed in Section 3.1.3.3 should be used.

     Thus, to model the far-field rise of a buoyant plume from a stack,  1t
1s recommended that the modeler:
     1.  Insure a fully turbulent effluent flow and
     2.  Either (in order of decreasing "correctness"):
         a)  match lm/Hs and IB/HS,
             1)  following geometric similarity or
            11)  exaggerating the stack diameter,  but avoiding stack downwash,
      or b)  match IB/HS,
             1)  following geometric similarity or
            11)  exaggerating the stack diameter,  but avoiding stack downwash.
     Obviously, if the stack diameter is exaggerated, other lengths are  to be
referenced to the stack height and not the stack diameter.   It is Implicit
above that the simulated atmospheric boundary layer is matched and that  geo-
metric similarity is followed everywhere, with the possible exception of
exaggerating the stack diameter as noted.  Notice  that an exaggeration in
stack diameter will generally be accompanied by a  reduction in the momentum
ratio.  It must be remembered that the momentum ratio should not be reduced
to the point where the plume is downwashed in the  wake of the stack.
     It is obvious that there are many unresolved  problems  concerning the
modeling of plume rise, in spite of nearly 40 years of such modeling. Be-
cause of the lack of basic, systematic studies on  these  fundamental problems,
the above recommendations are tentatively proposed and are  subject to change
pending future developments.
                                     71

-------
3.2  THE ATMOSPHERIC BOUNDARY LAYER
     In early wind tunnel studies of flow around  buildings  (Strom et al.,  1957),
complex terrain (Strom and Halitsky, 1953), and urban  areas  (Kalinske et al.,
1945), care was taken to insure that the approach flow was  uniform and of  low
turbulence across the wind tunnel test  section.   Jensen  (1958)  was the first to
suggest that the simulation of the atmospheric boundary  layer was important; he
was also the first to produce a simulated atmospheric  boundary  layer approach
flow by matching the ratio of roughness length to building  height between  model
and prototype.  Strong variations in surface  pressure  coefficient were observed
along with variations in cavity size and shape downwind  from a  building with
different depths of boundary layers in  wind tunnels  by Jensen and Franck (1963)
and Halitsky (1968).  Tan-atichat and Nagib (1974) and Castro and Robins (1975)
have shown that the nature, strength, and locations  of vortices in the flow
pattern around buildings differs markedly with and without  a thick boundary
layer approach flow.  Wind shear and the presence of the ground produce a  down-
ward flow on the front face of a building, a  reverse flow and an increase  in
speed upwind, and high winds near the sides as sketched  in  Figure 11 (Hunt,
1975).  It is now generally agreed that a thick boundary layer  is essential if
similar concentration fields are to be  observed downwind of a model.
                         UPWIND
                      VELOCITY PROFILE
                                  SEPARATED
                                FLOW ON ROOF
                        MEAN VELOCITY IN
                        REVERSE DIRECTION
                                        INCREASE IN SPEED
                                          NEAR SIDES
 Figure 11.
Effects of wind shear on the flow round a building.   (Reprinted
with permission from Models and Systems in Architecture  and
Building, Construction Press, Ltd., Hunt, 1975.)
                          72

-------
     Further, not just any thick boundary layer will do.  It must simulate
the atmospheric boundary layer structure, including as a minimum, the mean
velocity profile and the intensity and spectral distribution of the turbulence.
That the simulation of the spectrum is essential is evidenced in a report by
Dean (1977).  He attempted to duplicate the results of Snyder and Lawson (1976)
at the somewhat smaller scale of 1:500 (compared with 1:300).  The boundary
layer depth was scaled properly and the mean velocity and turbulence profiles
were reasonable facsimilies of those of Snyder and Lawson (S&L).  However,
when measuring concentration profiles in the boundary layer downwind from an
isolated stack, he found a vertical plume width over 3 times that of S&L and
a maximum concentration l/9th as large, which was characteristic of Pasquill
diffusion category A, highly unstable.  A removal of the vortex generators,
leaving the roughness strips intact, did not change the velocity or turbulence
intensity profiles appreciably, but produced a marked change in the energy
spectra, which in turn brought the plume width and maximum concentration to
within a few percent of those of S&L, more nearly characteristic of category D,
neutral stability.
     If the atmospheric boundary layer is to be simulated in a wind tunnel or water
channel, it is necessary to decide at some point just what characteristics can
and should be matched.  If adequate data are available describing the atmos-
pheric boundary layer structure for the specific site to be modeled, it is,
of course, more appropriate to use these data.  But, generally, sufficient data
are not available, so that some model must be chosen.  For example, if we want
to simulate the dispersion of pollutants from a stack in the atmospheric
boundary layer, we need first to answer the question of what the approach flow
should look like.   What is a typical atmospheric boundary layer depth?  What
are appropriate parameters that describe atmospheric stability and  what are
                                     73

-------
typical  values for these parameters?  How do the turbulence spectra vary with
stability, height above ground,  etc.?  It is necessary to establish a goal  to
be met,  say, a simplified analytical description, of the flow in the atmos-
pheric boundary layer.   The first part of this section (3.2.1)  suggests  some
goals which the modeler should attempt to achieve.  What we attempt to
describe is the barotropic planetary boundary layer under steady-state and
horizontally homogeneous conditions.  The second part (Section  3.2.2) reviews
the most promising techniques that have been tried for generating thick, neutral
boundary layers simulating the atmospheric boundary layer.  The third part
(Section 3.2.3) reviews methods  which appear promising for simulating strati-
fied boundary layers.  Finally,  an attempt is made (Section 3.2.4) to summarize
the previous sections and to establish guidelines for modeling  of the atmospheric
boundary layer.  Because the discussion of Sections 3.2.1.1 and 3.2.1.3 go  into
considerable detail, the disinclined reader may wish to skip to the summaries
(Section 3.2.1.2 and 3.2.1.4) for the essentials.

3.2.1  Characteristics  of the Atmospheric Boundary Layer
     The atmospheric boundary layer, alternately referred to as the Ekman
layer, the friction layer, or the planetary boundary layer, is  concerned with
that portion of the atmosphere where the aerodynamic friction due to the motion
of the air relative to the earth's surface is of prime importance.  Above the
boundary layer, the air motion is geostrophic, reflecting a balance between
the horizontal pressure gradient and the Coriolis force, and the velocity
obtained there is the gradient velocity.  The depth of the boundary layer is
highly variable, although it is typically between 1/2 and 2 km under neutral
conditions.  The overall boundary layer may be divided into at least 2 sub-
layers, principally the surface layer, also misnamed the constant stress
                                      74

-------
layer , and a transition region above,  wherein  the  shear stress  diminishes  from
the nearly constant value in the surface layer  to a near-zero  value  at  the  grad-
ient height.  The surface layer is  herein defined to be  the lower 10 to 20% of
the planetary boundary layer.   Generally, the mean  velocity profile  in  the
surface layer is described by a logarithmic law, although deviations can be
large.  Above the surface layer, there  are numerous analytical expressions
for describing the velocity profile.   If the entire boundary layer is to be des-
cribed by one expression, it is common  engineering  practice to use a power  law.
     Since Coriolis forces cannot be  modeled in an  ordinary wind tunnel or
water channel, modeling efforts should  be restricted to  those  classes of prob-
lems where Coriolis forces are unimportant.  As discussed in Section 2.2.1,
Coriolis forces may be important under  neutral  or stably stratified  conditions
in relatively flat terrain when the length of the model  exceeds  approximately
5 km.  It appears that, if the length scale of  the  field situation to be mod-
eled is less than 5 km, Coriolis forces may be  ignored.   Likewise, if the
terrain is rugged, so that the flow is  highly dominated  by local (advective)
forces, Coriolis forces may be ignored.  This restricted class of flows limits
the usefulness of fluid modeling facilities, but there still exists  a very
large range of problems in which it is  not at all unreasonable to ignore these
effects.  Orgill et al. (1971), for example, suggested that diffusion over
complex terrain could be reasonably simulated over  distances of  50 km.

3.2.1.1  The Adiabatic Boundary Layer
     Picking a depth for the adiabatic  (neutrally stable) boundary layer is no
simple task.  After an extensive review of the  literature on adiabatic boundary
layers, Counihan (1975) concluded that  the boundary layer depth  is 600m,
1.  Strictly speaking, a constant stress layer exists only in a boundary layer
with zero pressure gradient, which is seldom the case in the atmosphere.
                                      75

-------
practically independent of wind speed and surface roughness.   Davenport's
(1963) scheme, previously accepted as the "standard" for wind tunnel  studies
of wind forces on buildings, specified the depth as a function of the rough-
ness length z  only, varying from 6=300 m at zQ=0.03 m to 6=600 m at zQ=3  m.
Another popular scheme which is claimed to fit observations quite well is:
                             6  =cu*/fc ,                                 (3.12)
where 6 is the boundary layer depth,  u* is  the  friction  velocity  (=/TQ/P),
and f  is  the Coriolis  parameter Ztosine,  where  u is  the  earth's rotation
                                                    -4   -1
rate and e is the latitude.   In mid-latitudes,  fca10  sec  .   Typical  values
for c range from 0.2 (Hanna, 1969)  to 0.3 (Tennekes, 1973b).   It  is common to
use the geostrophic drag law, which relates the "drag coeffient"  u*/G  to the
surface Rossby number G/f z :
m rr-A + ln u:
    CO           *
                                           k2G2    R2
                                           	7T ~  0
1/2
                 (3.13a)
where G is the geostrophic wind speed (with components  U   and  V  ),  k  is
     ^  ^
von Karman's constant (0.4), and A and B are "constants"  which differ consider-
ably from one author to the next.  From Blackadar and Tennekes (1968), A  is
about 1.7 and B about 4.7.  For the sake of completeness, we also write the
expression for the angle a between the surface stress and the  geostrophic
wi nd :
                               sin a =      .                         (3.13b)
These three schemes for specifying the boundary layer depth are compared in
Figure 12, where it may be seen that they yield drastically different results.
In view of the uncertainties involved and also because Counihan's  (1975)
literature review showed np_ measurements of depths in excess of 600m, if
specific measurements to the contrary are unavailable, the boundary layer
                                      76

-------
should be assumed to be approximately 600m  1n  depth.   (In  modeling,  the
depth of the boundary layer is usually  large compared  with the  model  height,
so that the precise depth chosen is not usually  critical;  within limits,
there is room for choice.)  This value,  as  well  as  recommendations  that
follow, are not in any sense to be taken as absolute.   They are recommended
in the sense that in the absence of other data,  these  values are not unreas-
onable to use as a model.  Also, many of Counihan's  (1975) as opposed to
Davenport's (1963) conclusions are repeated here because they are represen-
tative of a wider range of data and they are more thorough in the sense that
more kinds of statistics are covered.
2500

2250

2000

1750

1500

1250

1000

 750

 500

 250
                                20
                       11 il
                            DAVENPORT (1963)
                            . . ..nl     i
                                                                   i i i 11 i
         .001
                  .01
.1
                                       z0.m
10
     Figure 12.
           The depth of the adlabatic boundary layer according to
           the geostrophic drag law compared with other schemes.
                                      77

-------
The depth of the surface layer, in which the mean velocity profile follows
a logarithmic law, and from which the roughness length may be defined, is
generally stated to be 10 to 20% of the boundary layer depth.  Counihan (1975)
suggests a value of 100m as a reasonable average depth for the surface layer.
The roughness length ZQ may be derived from the mean velocity profiles in the
range 1.5h 
-------
elements!   He suggested that the cause may be due to the more vigorous  turbu
lence scouring the buildings, with the air stream "penetrating"  more deeply
between buildings, thereby increasing both the inter-building wind speed and
the depth  of the building contributing effectively to the drag.
     The zero plane displacement d may generally be neglected for terrain
types where the roughness length is less than about 0.2m.  It is suggested
by Simiu and Scanlan (1978) that reasonable values of d in cities may be
estimated using the formula
                                                                   (3J5)
                                                        ^  ^
where ff is the general roof-top level  and k is the von Karman constant (0.4).
     The mean velocity profile throughout the entire depth of the boundary
layer is adequately represented by a power law:
                                 U/Uw = (z/6)p ,                  (3.16)
where U  is the mean velocity at the top of the boundary layer of depth 6 and
                                      79

-------
TABLE 3:  VALUES OF SURFACE ROUGHNESS LENGTH FOP VARIOUS TYPES OF SURFACES'



Type of Surface
Sand
Sea Surface
Snow surface
Mown Grass (-U3.01 m)
Low grass, steppe
Fallow field
High grass
Palrretto
Pine forest (Mean height of trees: 15 m;
one tree per 1C m2; 2^=12 m)
Outskirts of towns, suburbs
Centers of towns
Centers of large cities
(a) From Simiu and Scanlan (1978).
(b) Wind speed at 10 m above surface = 1.
(c) Wind speed at 10 m above surface > 15
(d) These values are exceptionally small;
7.
0
(cm)
0.01 - 0.1
O.C003b - 0.5C
0.1 - 0.6
0.1 - 1
1 - 4
2 - 3
4 - 1C
10 - 30

90 - 100
20 - 40d
35 - 45d
60 - 80d

5 m/sec.
m/sec.
see text.
                                     80

-------
p is the power law index.   This form is popular in engineering practice  and  is
highly useful  frorr. a practical  viewpoint.   Davenport (1963)  claims  that  the
overall reliability of the power law is at least as good as  much more sophis-
ticated expressions and it is recommended  here for that reason.   It was  shown
by Davenport to work quite well for high geostrophic winds.   It should also
work well for light winds  as long as the atmosphere is neutral.   Even for light
winds in the atmosphere, Reynolds numbers  are very large. The problem is not
that the power law will not work for light winds, but that,  especially under
light winds, the atmosphere is seldom neutral.  Figure 13 shows typical  mean
wind profiles and Figure 14 shows the variation of p with the roughness  length
z  (from Counihan, 1975).   The power law index varies from about 0.1 in  excep-
tionally smooth terrain such as ice to about 0.35 in very rough terrain  such
as built-up urban areas.
     As shown by Counihan, the turbulence intensity at a 30  m elevation  follows
the same (empirical) formula as the power law index; their numerical values
as functions of roughness length are identical.
           p = (/uVO")30m = 0.24 + 0.096 log1Qz0 + 0.016(log1()z0)2 ,    (3.17)
where z  is to be specified in meters.  The scatter in the Reynolds stress
                                       _  o
measurements was considerable, and -100uw/lr could have been represented by
the identical formula (3.17), but Counihan felt that would underestimate the
stress in moderately rough terrain.  Hence, he proposed, for the surface
layer
             -ITvr/uf = uJ/U^ = 2.75xlO"3 + 6xlO"4 log1Qz0 ,          (3.18)
which is also shown in Figure 14.  Counihan does not suggest how uw varies with
                                       81

-------
45§!-
3M
          A
  Figure 13:  Typical wind profiles over uniform  terrain  ir  neutral  flew.
   .4


   .35


   3
  I lll(
X
=2.75xir3 + fair4 ut,,1
    Jtl
 Figure 14:  Variation of power law  irdex,  turbulence intensity, and
             Reynolds stress with  roughness length ir the adiabatic
             boundary layer  (fror  Counihar, 197E).
                                   82

-------
height; he irplies that Eq. 3.18 gives its ''constant' value in the "constant"

stress layer.  A convenient approximation is a linear decrease with height from

its surface value to zero at 2=6.
                             - uw(z) =
                                              (3.19)
Thus, at heights less than C.lf, the stress is within 10? of its surface value

(see Figure 15).
                        u
                        1.1
!\
l°n
i
>— \
\ \A
I \
I \ "
1 |
U00=1l-/i
= STATION Ifa
- STATION 13
: STATION It
• STATION 19
• STATION 22
i : STATION 24
\
i
—i



                        14
I-
\
                                          Oo£  A
                                              ^.n
                                                          4
                                    i.5
                    1J
                 1.5
     Figure 15.  Shear stress distributions measured at various dottmrind posi-
                 tions in a wind tunnel boundary layer (neutral flow). Adapted
                 fror, Zoric and Sandborn (1972). (Courtesy of Boundary-Layer
                 Meteorology, D. Reidel Publishing Co.)
                                       83

-------
     These figures may be used  for model  design  purposes  in a number of ways.
In a general  type of study,  such as diffusion  over  an  urban area, they can be
used directly to determine appropriate  values  for z and  p.  Or, in a specific
study, once U  is chosen, one has only  to determine zn (by measurement or
                                          ft
estimation) and, since p,"\uVll,  and  -uw/lr  are  principally  functions of z  ,
to obtain them from Figure 14.   If it is  desired to match  the wind  speed U-, at
a particular height z-j, say,  at the top of a stack, Eq.  3.16 may  be rewritten
as
and the free stream wind speed (gradient wind) may  be  determined.
     The variation of the longitudinal  turbulence intensity  in  the  surface
layer is given by
                       V?/U"= p ln(30/zQ)/ln(z/z0)  .                  (3.20)
(This formula is slightly different from that of Counihan, but  it is consistent
with his data and other formulas.  Counihan's formulation did not match  Eq.  3.17
for the turbulence intensity at z=30 m for low values  of z0  and was somewhat
ambiguous in the range 0.1 m
-------
                    1.0



                    0.9



                    0.8



                    0.7



                    0.6




                    °-5


                    0.4



                    0.3



                    0.2



                    0.1
                            0.001
                          0.05  0.10   0.15  0.20  0.25  0.30  0.35  0.40
     Figure 16.  Variation  of longitudinal  turbulence intensity with height

                 under  adiabatic conditions.



decreased with increase of  surface roughness and increased with height up to



200-300 m.  Above this  level,  Lu  was independent of surface roughness and
                                 n


decreased with height.   A summary showing the variation of Lu  with elevation
                                                              A


and roughness length  is given  in Figure 17.   For other integral length scales,



Counihan has concluded:
     UJ  - 0.3~0.4 Lu¥
       y             x


     Lu  = 0.5~0.6 Lu
10 m
-------
                  1000
                  '«
                5
                i
                  10
  TERRAIN   PRE POST
   TYPE    -M   '40


   J~ 4    O   •

-  v'lsfl.O	••
                            0.4
             (r • MTERMITTENCY    ••
              • FACTOR)         \,
                                             LENGTH SCALE - U. <•)
                        FOR: 10 < Km) < 240. UM • C (t)l/B

     Figure  17.   Variation of integral length scale with  height and roughness
                  length.   (Reprinted with permission  from Atmos.  Envir., v. 9,
                  Counihan, Copyright 1975, Pergamon Press, Ltd.)



     nSu(n)/u* =  105f/(l+33f)5/3 ,                                 (3.23a)


     nSv(n)/u* =  17f/(H-9.5f)5/3 ,                                 (3.23b)


     nSw(n)/u* =  2f/(l+5.3f5/3) ,                                  (3.23c)


and -nC^nJ/u* = 14f/(l+9.6f)2'4 ,                                (3.23d)


where f=nz/D" is a nondimensional  frequency (see Figure  18 for plots).

     It may  be seen  that  these spectral functions are dependent on z  insofar

as u* and U  are functions of ZQ.   These expressions may be used to estimate

integral scales (e.g.,  Kaimal, 1973), but scales thus derived are not  consist-

ent with those in Figure  17  (scales derived from Counihan's  suggested  spectral

                                      86

-------
         10
     CM*
     ,3

     ^  .1
         .01
        .001
                                                               T	T
             a=u
          .001
.01
.1
10
100
                                        f = nz/U
     Figure 18.  Empirical curves for spectra and cospectrum for neutral
                 conditions (from Kaimal et al., 1972).
forms are even less consistent with Figure 17 -- such is our knowledge of the
neutral atmospheric boundary layer!).  This is very unfortunate, because, as
was pointed out 1n the previous section, the larger scales of the turbulence
are highly important in simulating diffusion.
3.2.1.2  Summary of the Adiabatic Boundary Layer Structure
     If specific site data are available giving adequate information on the
structure of the adiabatic boundary layer, it is, of course, most desirable
to use those data as the target to simulate in the wind tunnel.  If not, as
is usually the case, it is recommended that the following model be used.
For the sake of conciseness, justifications for the particular choices are
                                      87

-------
omitted here.  The interested reader may consult the previous  section  and
the references given there.   Listed here are the main features of the
steady state adiabatic boundary layer over horizontally homogeneous  terrain
(uniform roughness).
          1.  The depth 6 of the boundary layer is  600m, independent of
              surface roughness and wind speed.
          2.  The mean velocity profile is logarithmic in the  surface  layer,
              which is 100m deep.
          3.  The roughness length ZQ and the friction velocity u* may be
              derived from the mean velocity profile in the range l.S
              where h  is the general height of the roughness elements, and
              d is the displacement height (neglected for z <0.2m and given
              by Eq. 3.15 for z >0.2m).  Typical values for ZQ are given in
              Table 3.
          4.  The mean velocity profile through the entire depth of the
              boundary layer is represented by a power law U/U^ = (z/6)P.
              The power law index p is a function of ZQ alone and may be
              obtained from Figure 14 or Eq. 3.17.  It varies from 0.1 over
              smooth ice to 0.35 in built-up urban areas.
          5.  The Reynolds stress in the surface layer may be calculated as
              a function of z  from Eq. 3.18.  Its vertical variation may
              be approximated as a linear decrease with height from its
              surface value (Eq. 3.18) to zero at z=600m (Eq. 3.19).
                                        88

-------
     6.  The variation of the local  longitudinal  turbulence intensity
         with z  and elevation is shown in Figure 16.   The vertical
         and lateral turbulence intensities are approximately half
         and three-quarters, respectively, of the longitudinal  turbu-
         lence intensity.
     7.  The variation of the longitudinal integral  length scale with
         z  and elevation is shown in Figure 17.   Other integral scales
         may be obtained from Eqs. 3.22.
     8.  Spectral shapes are given by Eqs. 3.23 and  shown in Figure  18.

3.2.1.3  The Diabatic Boundary Layer
     In many ways, our knowledge of diabatic boundary  layers, at least in  the
surface layers, is more extensive than that of adiabatic boundary layers.
This is so because diabatic boundary layers are far  more common and  because  the
change in the surface heat flux is generally slow enough that the surface
layer turbulence is able to track it, i.e., the boundary layer is stationary
long enough that reasonably stable averages are more readily obtainable
(Wyngaard, 1975).  The depth of the  boundary layer is  highly dependent upon
the stratification.  During the day  over land, the effective top of  the  bound-
ary layer may usually be defined as  the inversion height, i.e., a layer  with
stable density stratification exists at some height  that is typically in the
range of 0.5 to 2 km.  On a cloudless night with  light winds, the ground
cooling generates a strongly stably  stratified layer very close to the surface
that suppresses the turbulence; the  effective boundary layer, then,  may  be
very shallow indeed, as low as a few tens of meters  or even meters (Businger
and Arya, 1974; Caughey et al., 1979).
                                     89

-------
     It 1s convenient at this point to discuss various paramenters that charac-
terize the stratification.  (This discussion closely follows that of Businger,
1973.)  The dlabatlc surface layer differs from the neutral  one,  of course*
because of the presence of the heat flux that creates the stratification that
very markedly affects the turbulence structure.  This 1s clearly  seen by exam-
ination of the turbulent energy budget equation (see, for example, Busch, 1973),
where a very Important production term appears that 1s proportional to the heat
flux.  Another production term is, of course, the mechanical term due to wind
shear.  Richardson (1920) introduced a stability parameter that represented the
ratio, hence, the relative Importance of these two production terms:
                              R1 .
                                   6
where e 1s mean potential temperature (e=T+yz, where T 1s actual  temperature
and y 1s the adlabatic decrease of temperature with height).   This parameter
Is known as the gradient Richardson number or, simply, Richardson number.
In deriving this stability parameter, it was assumed that the eddy transfer
coefficients for heat and momentum were equal (K^K^).  Since this assumption
1s not quite valid, it is better to leave the flux terms in the form in which
they appear 1n the energy equation, instead of assuming the fluxes are pro-
portional to the gradients of the mean quantities.  Hence, we have a flux
Richardson number
                            Ri   = S.  :Z-£L - •                   (3.24b)
                              T   T  uw  3U/3Z

where T represents temperature fluctuations.
                                      90

-------
  This parameter is  rather difficult to determine  because of the covariance
terms, whereas the determination of Ri  involved only the measurement of mean
temperature and mean velocity separately as  functions of height.
     If we differentiate the expression for  the logarithmic velocity profile
(Eq. 3.14), we obtain 3U/3z=u*/kz.   Substituting this expression into the flux
Richardson number yields a dimensionless height
                               _z.      c[  wT kz
                                L ="  T     4    '
                                      T  "*
                  where         L = - ••-	                         (3.24c)
                                      9 k wT
is the Monin-Obukhov (M-0) length.   This length is a very useful stability
parameter.  It contains only constants and fluxes  that are approximately con-
stant throughout the surface layer (also called the constant flux layer,
analogous to the constant stress layer in the neutral boundary layer).   L there-
fore is a characteristic height that determines the structure of the surface
layer.  It has been found that many features of the turbulence in the surface
layer depend solely upon the dimensionless height  z/L.  Such dependence is
referred to as M-0 similarity, which we will return to later in this section.
     Another stability parameter is the Ekman-layer equivalent of z/L,  i.e.,
it governs diabatic scaling in the entire boundary layer, much as z/L governs
diabatic scaling in the surface layer (Tennekes, 1973).
                                       ku
                                                                   (3.24d)
                                      91

-------
     A final stability parameter is the Froude number
                                                                       (3.24e)
which was discussed in Section 2.2.4.   The Froude number might  characterize  the
stratification in the surface layer or that of the entire boundary  layer,
depending upon the height H chosen for specifying the  velocity  and  the  upper
level for the temperature difference.   More common in  the meteorological liter-
ature  is the inverse square of this Froude number, which is  called the bulk
Richardson number
                                        "HU
To give the reader a "feel" for the magnitudes  of these  various stability
parameters, we have listed typical  values  1n  Table 4.  These  values are not to
be taken as definitions or as absolute 1n  any sense.   Particular values depend
on the height chosen for specification of  the wind speed and  temperature and
there is not, in any event, a one-to-one correspondence  between the parameters.
     With these stability parameters in hand, we  will  be able to specify many
of the features of the diabatlc boundary layer  (albeit one  that is steady and
horizontally homogeneous).  Let us  first quantify our  rather  qualitative des-
cription earlier in this section of the boundary  layer depth.
     According to Hanna (1969), the formula
                                      0.75 11
                                 6  =
                                    tf 4H
where Ae/Az 1s the average vertical  gradient  of potential temperature through
the boundary layer, agrees well  with observed boundary  layer thicknesses.
                                     92

-------
      TABLE 4:  TYPICAL VALUES FOR THE VARIOUS STABILITY PARAMETERS.
Qualitative
description
Highly
unstable
Unstable
Slightly
unstable
Neutral

Slightly
stable
Stable
Highly
stable
(a)
(b)


Pasquill-Gifford
category
A
B
C
D
E
F
G
L,m z/L
-5
-10
-20 -0
00
100 0
20 0
10
-2
-1
.5
0
.1
.5
1
Rif
-5
-2
-1
0
0.07
0.14
0.17

Ri
n -3
-2 -0.03
-1 -0.02
-0

0.
0.
0.
The assumed height of the anemometer and
the lower thermometer: 2m. A roughness
assumed in the calculations.
The
friction velocity
listed
is
that
Fr
P «..*
- -4000
- -120
.5 -0.01 -
0
07 0.
14 0
17 0
upper
length
value used
0
004
.05
.17
00
16
7
5
thermometer
of 0.01 m
in
-60
0
12
40
100
was 10 m
was also
sb
3
3
3
3
3
2
1
*
calculating y.
This equation can be used most of the time,  because  the atmosphere  is  usually
stable, on average, throughout the boundary  layer.   This equation implies  that
the bulk Richardson number g6A6/(TU2) equals 0.56 for all  stable boundary
layers, i.e., the boundary layer adjusts itself until this criterion  is met.
     Arya (1977), to the contrary, claims that observations indicate  that  the
bulk Richardson number increases with increasing stratification and may
approach a constant (critical) value only under extremely stable conditions.
He, Businger and Arya (1974), Wyngaard (1975), Brost and Wyngaard  (1978) and
others using widely differing theoretical approaches all arrive at  the simple
form for the height where the stress is some specified small  fraction of the
surface value:
                          6/L - ayl/2   or   6 = a(Lu*/f J1/2,         (3.25b)
                                     93

-------
where a is a constant and y*=u*/fcL is a stability parameter related to
Eq. 3.24d through v=ky*.  The constant a, however, is highly dependent upon the
value chosen for the stress criterion.  For a 1% stress criterion, Businger and
Arya (1974) find a=0.72, whereas for 5%, a*0.4.  The latter value is supported
by the second order closure model of Brost and Wyngaard (1978) over a wide
range of cooling rates.  Comparisons with Wangara data (Arya, 1977) show very
large scatter, and that a^l would be a much better fit.
     To estimate u*, the geostrophic drag relation (Eq. 3.13) is used, where
the "constants" A and B are functions of the stability parameters.  In a
critical review, Arya (1977) has suggested

                        A = ln(6/L) - 0.96(6/L) + 2.5                 (3.26a)
and                          B = 1.156/L + 1.1 ,                      (3.26b)

where <5 is determined from Eq. 3.25b with a»l.  (This is obviously an itera-
tive procedure in that Eqs. 3.13, 3.25, and 3.26 all  involve u*, which we are
attempting to determine.)
       Finally, an interpolation formula suggested by Deardorff (1972), i.e.,
                          « - f ]  +   fc   + 1  I"1                      /, 97 \
                          6 " [3UT   O5U7   ZjTj   >                     (3<27)
in which ZT is the height of the tropopause, does  not suffer from "blowing up"
under neutral  conditions (where L-*») near the equator (where fc-*0)»
6+0.25u*/fc  under neutral  conditions in mid-latitudes (i.e., Eq. 3.12) and
6+30L under very stable conditions and/or in low latitudes.   Eq. 3.27 yields
results comparable to Eq. 3.25b in mid-latitudes (see Figure 19).
       The unstable boundary layer is almost always capped by an inversion at
some elevation.  It is now generally agreed that the  height of this boundary
                                      94

-------
layer is determined by the height of the base of the inversion, i.e., 6=z.
(Deardorff, 1972*. Wyngaard et al . , 1974).  The height of the inversion base
varies from day to day, but its diurnal trend is quite similar.  Kaimal et al .
(1976) describe their observations in the Minnesota experiments as follows:
     "Between sunrise and local noon (1300 CDT) z. grew rapidly in response
     to the steadily increasing heat flux (Qft).  the growth of z. slowed down
     between 1300 and 1600 CDT as Q  reached0 its maximum value.  But as Q
     decreased through the late afternoon, z. began to level off to a neaPly
     constant value which it maintained even after Q  turned negative."
     Even though this convective boundary layer depth changes rather rapidly
with time, there is justification for treating its midday structure as if it
were in steady state, or at least in a condition of moving equilibrium or
quasi-steady state (Kaimal et al., 1976).  To predict the height of this
boundary layer, Deardorff (1974) and Arya (1977) recommended a rate equation.
For purposes of fluid modeling, it is sufficient to pick typical values,
i.e., 6sl to 2 km, as the typical maximum height for the inversion base is
1 to 2 km.  Once a boundary layer height is chosen, we can estimate u* from
the geostrophic drag relation, Eq.  3.13, where the parameters A and B are
functions of the stability parameters 6/L and f  6/u* (Arya, 1977).
                          A = ln(-6/L) + ln(fc«/uj + 1.5                (3.28a)
                   B = Mfa/uJ'  + 1.8(f<5/u*) exp(0.26/L) .              (3.28b)
Figure 19 shows predicted boundary layer depths from Eqs.  3.25b and 3.27.   It
may be seen why the neutral boundary layer depth is so difficult to determine;
only slight departures from neutrality effect drastic changes in its depth.
Figure 20 shows how the friction velocity u* varies with stability as predicted
                                      95

-------
  1500
  1350
  1200
  1050
   900
   750
   600
   450
   300 -
   150 -
    -0.2   -0.15    -0.1
.05     .1     .15     .2
                                  •0.05      0
                                       1/L, m'1
Figure 19.   Typical nonadiabatic boundary layer depths from the  geostrophic
             drag relations  (G=10m/s,  z =0.01m,  zT=10km, f =10  /s, a=l).
                                        O          1C
*
   -0.2   -0.15    -0.1
           .15     .2
                                 •0.05     0     .05
                                       1/L. nr1
Figure 20.   Variation of  friction velocity  with stability  from the
             geostrophic drag relations  (Eqs.  3.13, 3.26, 3.28, G=10m/s,
             zQ=0.01m, zT=10km).
                                       96

-------
from the geostrophic drag relations using Eqs. 3.26 and 3.28.
     Regarding the mean wind profile under non-neutral conditions, DeMarrais
(1959) has measured the power law index p and has drawn the following general
conclusions:
     "During the day, when superadiabatic conditions and neutral lapse rates
     prevail, the values of p vary from 0.1 to 0.3.  This variation is princi-
     pally in proportion to the roughness of the terrain.  At night, when
     stable, isothermal, and inversion conditions exist, the value of p
     generally varies from 0.2 to 0.8; this variation is proportional to the
     degree of stability and the roughness of the underlying terrain."
     Panofsky et al. (1960) have used a formula due to Ellison (1957) to derive
a theoretical relationship for p as a function of z  and 1/L (L is the M-0
length).  Irwin (1978) has, analogously to Panofsky et al., used results of
Nickerson and Smiley (1975) to establish a theoretical relationship between p,
z  and 1/L.  The results, shown in Figure 21, support DeMarrais1 (1959) con-
clusions reasonably well.
     Air pollution meteorologists frequently use Pasquill stability classes
(or similar groupings) to categorize atmospheric diffusion.  Golder (1972) has
related the qualitative Pasquill classes to more definitive measures of sta-
bility through analysis of a large number of observations at 5 sites.  Irwin
(1978) has taken Golder's results relating Pasquill classes to the
Monin-Obukhov length and roughness length and overlaid them as shown in
Figure 21.  Irwin (1979) has further plotted the variation of p with z
where the Pasquill stability class is a parameter (see Figure 22).  It may
be seen from Figures 21 and 22 that the shape of the wind profile is much more
strongly dependent on stability than on the roughness length under stable con-
                                      97

-------
                     0.001
                             106     0.02     -OJ2     -0.06
                                      STABILITY LENGTH. 1/L (m'1)
                                                        •0.10
                                                               •0.14
     Figure 21,  Theoretical variation of  the  power-law exponent as a function
                 of z  and L for z equal to  100m.   The dashed curves overplot-
                 ted are the limits defined  by Colder (1972)  of the Pasquill
                 stability classes as adapted  by Turner (1964).  (from Irwin,
                 1978).
ditions.  It is relatively insensitive to  stability but more dependent upon
roughness under unstable conditions.  Comparisons  (and tailoring) of Irwin's
results with field data of DeMarrais  (1959), Touma (1977) and Izumi (1971)
agreed well and explained reported differences in  exponent values.  His theo-
retical predictions compare very well with Counihan's (1975) results for
neutral conditions, i.e., stability class  D (see  Figure 22).
     In the above discussions, we have largely ignored the influence of the
earth's rotation because this feature, in  general, cannot be simulated realis-

-------
                   0.6
                   ;0.4
                  I
                  oc
                  Ul
                  o
                  UJ
                   0.2
zt =  10.0
z2 = 100.0
0 =   2.0
                    0.01                    0.10
                              SURFACE ROUGHNESS LENGTH, z., meters
                                        1.0
     Figure 22.  Variation of the power-law exponent  p,  averaged  over layer
                 from 10m to 100m, as a function of surface  roughness and
                 Pasquill stability class.  Dashed curve is  result  suggested
                 by Counihan (1975) for adiabatic conditions which  should
                 agree with stability class D.  (Reprinted with permission
                 from Atmos. Envir., v.13, Irwin, Copyright  1979, Pergamon
                 Press, Ltd.)
tically in laboratory facilities in any event.  On the other hand,  many fea-
tures of the surface layer can be well simulated.  Panofsky  (1974)  has
suggested that we further subdivide the Ekman layer (overall boundary layer)
into a tower layer, i.e., below 150m or so in neutral or unstable conditions.
The surface layer proper extends to approximately 30m, but many of  the relat-
ionships developed for the surface layer may be extended to  the tower layer;
whereas the earth's rotation may be important in the  tower layer  (it  was  not
in the surface layer), the turning of the wind can be ignored.  In  stable air,
this subdivision is useless because significant turning  may  start at  much
                                      99

-------
lower heights.  In the discussion to follow, then,  we will  discuss 1n detail
various properties of the surface layer that may possibly be extended to the
tower layer in neutral and unstable conditions.
     It is customary in discussing surface layer profiles and fluxes to define
nondimensional vertical gradients of wind speed  and potential temperature as
where e*=-wt/u*.  It has been found that these nondimensional  gradients are
functions of z/L only (M-0 similarity).  In unstable air,
                                             z/L<0                      (3.30)
fits surface observations quite well (Panofsky, 1974).   The expression can be
Integrated to obtain the mean velocity profile (Paulson, 1970):

          U/u* = (l/k)(ln(z/z0) - 2 ln[Jd+l/fm)] - ln(J(l+l/«J;))
                          + 2 tan'1 (I/O - it/2]                       (3.31)
                                       ni       j •

This formulation is consistent in that under neutral conditions L-*«, «m-»-l,
and Eq. 3.31 reduces to the familiar log law.  Panofsky (1974) showed that
« •O-lSz/L)"1^3 fit various data sets better than Eq.  3.30 for large values
of  |z/L|.
     In stable air
                            *m = 1+Bz/L  , z/L>0                        (3.32)

integrates to the familiar log-11near wind profile:
                                      wo

-------
                       U/u* = (l/k)(ln Z/ZQ + 5z/L).                 (3.33)
Figure 23 shows typical velocity profiles as predicted by Eqs. 3.31 and 3.33.
     The behavior of the nondimensional temperature gradient 4^ is somewhat
controversial.  For simplicity we will here list the forms given by Panofsky
(1974)
                                         , for z/L<0 ,              (3.34a)

and                       *h = l+5z/L , for z/L>0 ,                  (3.34b)
These expressions integrate to
              (e-e )/e* = In (Z/ZQ) - 21n{(l+l/*h/2>, for z/L<0     (3.35a)
                   (e-e0)/e* = In (Z/ZQ) + 5z/L, for z/L>0,         (3.35b)
where e  is the extrapolated temperature for z=z  (not necessarily the actual
surface temperature).  Typical temperature profiles  as predicted by Eqs. 3.35
are shown in Figure 24.  It is useful in interpreting Figure 24 to note that
e* and (e-e ) change sign simultaneously, so that the slopes of the curves
are always positive.  It is also interesting to note that the limit as L-*« is
the same as the limit as L-»~«>, i.e., a logarithmic temperature distribution,
which is not the same as adiabatic, where the potential temperature would be
uniform with height.  This is an anomaly 1n the mathematics, because both
numerator and denominator of the left hand sides of Eqs. 3.35 approach zero
simultaneously as the surface temperature approaches the fluid temperature.
     Another useful relationship is that between the gradient Richardson
number and z/L:
                              Ri = (z/L) (*h/»2)                    (3.36a)
                                     101

-------
           10000
           1000 -
Figure 23.  Typical surface  layer velocity profiles under nonadiabatic
            conditions  (from Eqs. 3.31  and 3.33 with z  = 0.01m).
          10000
           1000 -
            10
                                         8     10     12     14
                                     (0 - 00) /0»

Figure 24.  Typical  temperature profiles in the surface  layer  (from
            Eqs. 3.35 with  ZQ = 0.01m).
                                      702

-------
or                        Ri  = z/L ,  for z/L<0,                  (3.36b)





                         Ri  =          ' for Z/L>0              (3.36c)
This relationship is shown in Figure 25.





     The mean squares of the fluctuations of various  turbulence quantities



are also found to obey similarity theory.  The variance of the vertical



velocity fluctuations o= jvr follows Monin-Obukhov scaling,  so that
                       W




                                                                 (3.37a)
where »  is a universal  function.   According to Panofsky (1974)
       W




                   (VI. 25 for z/L>-0.3(including all  z/L>0)/

             *w -            1/3                             .    (3.37b)

              w    h.9(-z/L)l/<3 for
The variance of temperature also follows M-0 similarity,  so that





                                                                 (3.38a)
                                     703

-------
where *  is also a universal  function which, according to Panofsky (1974)
       W


is given by



                                        "° for z/L<-0.1
                                                                    (3.38b)

                            .1.8 for z/L>0





*  and *  are shown in Figure 26.
 W      V


     The variances of the horizontal velocity components a  and o  do not appear



at present to follow any discernable pattern and do not obey M-0 similarity



(but see later discussion in this section of the convective boundary layer).



This is evidently due to the low-frequency contributions to these variances



that are possibly due to large scale terrain features or circulation systems



of large horizontal extent, unaccounted for in M-0 theory.  Part of the problem



may also be due to the difficulty in separating fluctuations from means in the



original time series, i.e., the "spectral gap" may not be so clearly defined.



     Nevertheless, various authors have attempted to force their observations



to fit M-0 scaling
                      ou = u,»u(z/L) , ov = u**v(z/L) .                 (3.39)





The "constants" $..(0) and * (0) for neutral stratification vary from 1.5 to 3



with "mean" values of 2.5 and 1.9, respectively.  Observations of the variation



of a  and o  with height often show little attenuation,  but there are notable



exceptions where slow and rapid decreases have been observed even in unstable



air (see Panofsky, 1974).
                                      104

-------
      -0.25
         -1    -0.8   -0.6   -0.4   -0.2    0    .2    .4     .6    .8    .10
       -0.5 -
      -0.75 -
Figure  25.   The relationship between Ri and  z/L (Eqs. 3.36).
         4     -3      -2     -1
3     4
Figure 26.   Variation  of *., and *   with z/L in  the surface  layer
                            W      v

             (Eqs. 3.37 and 3.38).
                                  105

-------
     Binkowski (1979) has derived expressions for $u and $y that fit through
the middle of the very wide scatter of the Kansas and Minnesota data.  (The
"wide scatter", however, most likely results from plotting in incorrect sim-
ilarity coordinates, not from scatter in the usual sense; see later discussion
in this section on the convective boundary layer.)  Due to the complexity of
the formulas, they are not repeated here, but are shown in Figures 27 and 28.
     Spectral forms are taken from Kaimal et al. (1972) and Kaimal (1973).
In stable air, spectral shapes of all velocity and temperature fluctuations
were found to be M-0 similar, and to have universal forms when appropriate-
ly normalized. Hence,
                          nS (n)     0.16 f/f
                          -Z-	P     ,                   (3.40)
                             2     l+0.16(f/fn)5/0).  As an engineering approximation, vertical
velocity and temperature fluctuations may be assumed to fit the universal
form (Eq. 3.40).  This universal spectral shape is shown  in Figure 29, and
the variation of  the peak  frequency with z/L is shown  in  Figure 30.
     The integral scales are difficult to evaluate directly from  the spectra;
                                     706

-------
                                                I     '
                                           x = KANSAS DATA
                                           + = MINNESOTA DATA
                                    I
                        •2
0
z/L
Figure 27.  Variation of »u with z/L  in the  surface  layer  (Reprinted
            with permission from Atmos. Envir.  v.  13,  Binkowski,  Copy-
            right, 1979, Pergamon Press, Ltd.).
                                          i	,	•	
                                          x = KANSAS DATA
                                          + = MINNESOTA DATA
            3 -
            2 ~
            1  -
Figure 28.  Variation of *y with z/L (Reprinted with permission from
            Atmos. Envir., v. 13, Binkowski, Copyright, 1979, Pergamon
            Press, Ltd.).
                               707

-------
          0.01
                                      10
                                       100
1000
     Figure 29.  Universal  spectral  shape (Eq. 3.40)
Figure 30.
        -25  -20  -15  -10  -O5   0  +05  +10   +15  +2.0
                          *'L
Location of spectral peak  for u,  v, w and e plotted against
z/L.  Curves shown are  fitted by  eye (from Kaimal et al., 1972;
Reprinted with permission  from the Quarterly Journal of the
Royal Meteorological Society).
                                      108

-------
a length scale that can be obtained directly is x^, the wavelength corres-
ponding to the peak in the logarithmic spectrum nS(n).  Using Taylor's
hypothesis,
                                               •                       <3-42'
where n  and f  are the cyclic and reduced frequencies at the spectral  peaks.
       mm
This length scale is used extensively (as opposed to the integral  scale) in
the interpretation of atmospheric spectra.  Kaimal (1973) has derived a simple
expression relating these two length scales under stable conditions
                    Lax = 0.041z/f0a = O.lGz/f^ = Xm(«)/2ir .           (3.43)

His findings for the variation of these length scales with Richardson number
are listed in Table 5.

   TABLE 5.  DIMENSIONLESS LENGTH SCALES AS FUNCTIONS OF Ri(Q.05
-------
with other empirical relationships.   No expressions are available for the
variation of A  in unstable conditions, but values for X (w)  and X (e) may be
deduced from Figure 30.
     Little is known about the variation of X  with roughness.   Wamser and
 •*
Muller (1977) noted that their data  showed a decrease in X (w)  with increasing
roughness under neutral and convective conditions, but could  not draw any con-
clusions for stable conditions.  They also noted that there was no systematic
dependence of ^(u) on roughness.  Higher order statistics such as cospectra
and structure parameters are beyond  the scope of this review.  The interested
reader is referred to Wyngaard and Cote (1972), Wyngaard et al. (1971).
     Above the surface layer, the turning of wind with height generally becomes
highly important and is not amenable to simulation in the usual  laboratory
facility.  But one case, in fact one that is fairly typical  of daytime convec-
tive conditions, deserves mention.  Kaimal et al. (1976) describe the structure
of this  "mixed layer" as obtained from their extensive measurements in Minne-
sota.  The surface layer is as described above, but is confined to the height
range z<|L|.  Immediately above the surface layer, they describe a "free con-
vection" layer, where the surface shear stress is no longer important, but the
height z continues to be important.  The upper level for this free convection
layer is approximately O.lz., where z. is the height of the base of the lowest
inversion, and is also a good measure of the boundary layer depth (typically
1 to 2 km).  The remaining 9/10 of the boundary layer, then, is the "mixed
layer" where the mean wind is essentially uniform and the wind direction
changes  little with height.  In the "worst case" run, the wind direction
varied by only 15° between the surface and the top of the boundary layer;
it was typically only a few degrees.
                                      110

-------
     It is conceivable that the entire depth of this  convective boundary
layer could be simulated in a laboratory facility,  albeit at very low Reynolds
number.  Deardorff and Willis (1974)  have done the  limiting case of pure con-
vection (no wind) and Schon et al.  (1974) and Rey et  al.  (1979) have done an
unstable boundary layer, but without  a capping inversion.   That the two ap-
proaches can be merged appears promising.
     For details of the boundary layer structure (variances, scaling, spectra,
etc.), the reader is referred to the  papers by Kaimal et  al. (1976), Kaimal
(1978), and Panofsky et al. (1977).   The latter authors show, for example, by
using observations from several data  sets over uniform surfaces, that $u and
*  depend not upon z/L, but instead upon z^/L.  Also, there were no signifi-
cant differences between the lateral  and longitudinal components.  Their ex-
pression fitted to the horizontal velocity data is
                     *H = (12-0.5z./L)1/3 , -400z.,)
and a transition region (z
-------
ity, where z^ was the sole-governing length scale, applied.  Interpolation
formulas for the transition region were derived.  Further, it was shown how
these surface layer spectra (including w) evolve with height into their mixed
layer forms.  As the empirical expressions are complicated and of somewhat
limited applicability, the interested reader is referred to the original  paper.

3.2.1.4  Summary of the Diabatic Boundary Layer Structure
     Listed here are the main features of the steady-state diabatic boundary
layer over horizontally homogeneous terrain.   Again,  if specific site data
are available giving, for example, typical strongly stable characteristics of
the boundary layer, it is, of course, most desirable  to use those data as  a
target to simulate.
        1.   The depth of the stable boundary layer may be estimated from
            Eq.  3.27, where the friction velocity u*  is obtained from the
            geostrophic drag law (Eq. 3.13),  and the  "constants" A and B
            are determined from Eqs.  3.26 (an iterative procedure).   It is
            typically 100m deep.   The unstable boundary layer undergoes a
            diurnal  trend with a  typical  maximum depth between 1  and 2 km.
        2.   Once the boundary layer depth is  chosen,  the friction velocity is
            obtained from the geostrophic drag relation (Eq.  3.13),  where  the
            "constants" A and B are obtained  from Eqs.  3.26 for stable
            conditions and Eqs.  3.28 for unstable conditions  (again,  an
            iterative procedure).   Typically,  u^O.OSU^ in unstable  conditions and
            u*=0.02Uoo in stable conditions.
        3.   The  power law exponent p characterizing the shape of the mean
            velocity profile over the depth of the boundary layer may be
            obtained from Figure  21  or 22.   In unstable conditions,  it is
                                      112

-------
     dependent primarily on the roughness length and essentially Indep-
     endent of the degree of Instability, varying 1n the range of 0.1  to
     0.2.  Under stable conditions, it 1s highly dependent upon the degree
     of stability and essentially independent of the surface roughness,
     varying in the range of 0.2 to 0.8.
4.   In neutral and unstable conditions, the surface layer properties
     may be extended to a depth of approximately 150m.   In stable con-
     ditions, the surface layer may be even thinner than 10 to 20 m in
     depth.  The Monln-Obukhov length L is currently the most popular
     stability parameter because most of the surface layer properties
     can be described solely in terms of the dimenslonless height z/L
     (M-0 similarity theory).  Given L and u*, we can predict the shapes
     of the mean velocity profile (Eqs. 3.30, 3.31 and 3.33), the mean
     temperature profile (Eqs. 3.34 and 3.35), the variance of vertical
     velocities (Eqs. 3.37), the variance of temperature (Eqs. 3.38),
     and to a rough approximation, the variances of the lateral and
     longitudinal velocities (Eqs. 3.39 and Figs. 27 and 28).  We can
     also predict spetral shapes (Eq. 3.40) and scales (Eq. 3.41 through
     3.45, Fig. 29 and Table 5).
 5.  Little 1s known of the boundary layer characteristics above the
     surface layer except that generally the turning of the wind
     with height 1s Important.  Flow above the surface layer is thus
     not usually amenable to simulation 1n a laboratory facility.
     One special case, however, is the convective boundary layer.  It
     appears that this entire boundary layer could be simulated in a
     laboratory facility as the change 1n wind direction with height
                                     113

-------
     is typically only a few degrees over its typically 1 km
     depth.  For additional details, the reader is referred to the
     original papers.
We have seen in our review of the atmospheric boundary layer that it is ever
changing, it is governed by a large number of parameters, and that its space-
time characteristics are difficult to determine.  Even the specification of one
of the "simplest" characteristics, its depth, is a horrendous problem.  We
have attempted to assimilate the results of the most recent theories, but
they continue to develop and are rapidly modified as new experimental results
                                                       .*  *
become available.  Even the classical "universal" von Karman constant is
questioned (Tennekes and Lumley, 1972).  There are few generic boundary
layers to emulate or to compare with our wind tunnel simulations.   Neverthe-
less, we have classified typical types and have described the salient
characteristics of those classical types as they are known at the present
time.
3.2.2  Simulating the Adi abatic Boundary Layer
     In the previous sections, we have established at least the main character-
istics of the adiabatic and nonadiabatic atmospheric boundary layer.  In this
section, we will examine several techniques commonly used to simulate the
neutral atmospheric boundary layer and note, where possible, how successful
these techniques have been.  Generally, such techniques have been applied
only in wind tunnels although, in principle, they could also be used in water
tunnels and towing tanks.
                                      114

-------
     The techniques can be broadly divided into three categories:
     1.  Long tunnels, in which a thick boundary layer develops naturally
         over a rough floor (Figure 31).  The length of the test section
         of such a tunnel is typically 30 m.
     2.  Short tunnels with passive devices, in which the boundary layer is
         generated by a fence, screens or grids of non-uniform spacing,
         spires or vortex generators, i.e., stationary devices that retard
         the mean flow close to the floor and induce vorticity and turbu-
         lence into the boundary layer (Figure 32).  In order to maintain a
         non-developing boundary layer, it is essential to "match" the
         generators with the roughness elements.
     3.  Short tunnels with active devices, in which the boundary layer is
         generated by jets directed at angles to the main flow stream at
         the entrance to the test section (Figure 33).  Again, "matched"
         roughness must line the floor of the test section to obtain a
         non-developing boundary layer.
     Subcategories might include tunnels equipped with machine-driven shutters
or flaps or possibly a program-driven variable speed fan.  In short tunnels
with active devices, it is claimed to be possible, within limits, to vary the
turbulence structure independently of the mean velocity profile, but it is
not clear that boundary layers with different properties can be in
equilibrium with the same surface roughness.
     Initially, the long tunnels were touted as superior to short ones
with devices for artificially thickening the boundary layers because in them
the boundary layers were developed "naturally" over rough ground.  The long
                                      115

-------
          Figure 31.   Upstream view of a  long  wind  tunnel  (Courtesy
                      of the Boundary Layer Wind  Tunnel  Laboratory,
                      University of Western Ontario).
tunnel advocates felt that the grids, jets and vortex generators  introduced

extraneous turbulence scales and the turbulence dissipated and its structure
                                                                          /
changed with downstream distance (Cermak and Arya,  1970).   The short tunnel

enthusiasts, on the other hand, pointed to the developing  boundary layer and

to the secondary flows caused by the growing sidewall boundary layers as not
                                     116

-------
           Figure 32.  Vortex generators and roughness in a short wind
                       tunnel (Courtesy of Marchwood Engineering Labora-
                       tory, Central  Electricity Generating Board, England)
representing steady and horizontally homogeneous atmospheric boundary
layers (Nagib et al., 1974).  Recently, however, drag-producing elements
have been used in the long tunnels as well, and many techniques for generating
                                     117

-------
            FREE-STREAM
            ENTERING
            TEST SECTION


AM

ON



	 ^
	 .


-- _.

r~~d tz -'-""
	 j ,•-'
~^
	 ^
_.-:

„- _ .,

r_7









^
.
-—

- - -*


. V





•j-
1



	 .

	 	 m

— .
	 /





^ f
I
,8
i


^ 	 .

	 _.

	 •
	 H
         T  T
                     '/xV/V*>^v
                     »^A     \
        //////,
    RECIRCULATING
•x\ FLOW REGION
/      VIEW
THICKNESS OF
GENERATED
BOUNDARY LAYER
                                 ».  • .   '  ' '  '  ...  . •  •
                                 I   •  -...•'  .I''..
Figure  33.   Schematic  representation of the counter-jet  technique
             (Reprinted with permission  from the American Institute
             of Aeronautics and Astronautics, AIAA Paper  No.  74-638,
             Figure 1,  Nagib et al., 1974).
                    DEPTH OF BOUNDARY LAYER OVER CARPET (z0~0.03 cm)

  DEPTH OVER RECTANGULAR BLOCKS (2.5 TO 10 cm HIGH; zn ~0.3 cm)
       BELL MOUTH-*) (4-DISTANCE FROM LEADING EDGE OF ROUGNESS (m) —»

Figure 34.   Development of boundary layer  in  a long wind tunnel  (Adapted
             from Davenport and Isyumov,  1968, Proc. Int. Res.  Seminar on
             Wind Effects on  Buildings and  Structures, Univ.  of Toronto
             Press).
                                     118

-------
thick boundary layers in short lengths of test section have been
developed.  There is no reason, in principle, why a fully developed layer
with unchanging turbulence properties cannot be achieved in short tunnels.
An experimentalist must only be clever (or lucky!) enough to determine the
proper size, number and arrangement of grids, vortex generators, jets,
roughness, etc. to obtain such.  Also, some development length is required;
current practice indicates that an equilibrium boundary layer may be estab-
lished in 5 to 10 boundary layer heights, a substantial improvement over
the long tunnels.  It is not the aim here to favor one system over another,
but instead to stress the necessity of adequately measuring the boundary
layer, however generated, to be sure that it is laterally homogeneous and
non-developing (if that is what is desired) and that is meets the target
flow characteristics (which are not, in all cases, of course, those of the
atmospheric boundary layer).
     Examples of the first category, long tunnels, are the Micro-Meteorologi-
cal Wind Tunnel at the Colorado State University (CSU) (Cermak, 1958; Plate
and Cermak, 1963) and the Boundary Layer Wind Tunnel at the University of
Western Ontario (UWO) (Davenport and Isyumov, 1968).  Figure 34 illustrates
the development of the boundary layer in a long tunnel and how the depth of
the boundary layer depends upon the roughness.   Because of the growth of
the boundary layers, these  tunnels generally have adjustable ceilings to
control the axial pressure distribution, and the ceiling is adjusted to give
a zero pressure gradient along the length of the test section.
     Sandborn and Marshall (1965)  were the first to show that the turbulence
in the boundary layer of the CSU long tunnel  exhibited characteristics of the
Kolmogoroff local isotropy predictions, i.e., a large separation between the
                                     119

-------
Integral scale and the microscale (see Section 2.2.2.2 and Figure 6) which is
characteristic of large Reynolds number turbulence, and is responsible for the
-5/3 power in the spectral equations (3.23).   Their measurements were made
over a coarse gravel floor approximately 20 m from the test section
entrance with a free stream wind speed of approximately 10 m/sec and boundary
layer thickness of about 60cm .   Whereas this feature is necessary for
simulation of wind forces on buildings (see Simiu and Scanlan, 1978), it
is regarded as relatively unimportant' (but certainly not harmful) for
diffusion studies (See Section 2.2.2.2).  The results do indicate, however,
that the flow Reynolds number may be reduced somewhat without reducing the
total turbulent energy or shifting the location of the peak in the energy
spectrum significantly (See Figure 7).
    Zoric and Sandborn (1972) have shown that profiles of mean velocity
nondimensionalized by  boundary layer depth are similar beyond 6 m from the
entrance (CSU tunnel).  Figure 35 shows that they are approximated quite
well by a l/7th power law.  Figure 36, however, shows that the boundary layer
grows nearly linearly and still  quite rapidly with downstream distance beyond
about 10 m.  Zoric (1968) obtained results similar to those of Figures 35
and 36 for freestream velocities between 18 and 30 m/s.  Turbulence
profiles were strikingly similar in shape to those suggested by Counihan
(1975) for very small roughness (see Figure 16).  The boundary layers
were developed over the smooth wind tunnel floor.  Vertical turbulence
1.  In fact, under these flow conditions, the turbulent Reynolds number,
based on eddy velocity and eddy size, may be estimated to be at most 2000.
Tennekes and Lumley (1972) suggest a bare minimum value for an inertia!
subrange (local isotropy) to exist is 4000.  Even though a substantial
spectral region with a -5/3 slope was measured, it is doubtful that local
isotropy existed, i.e., the existence of the -5/3 slope is not a critical test
of local isotropy.
                                      720

-------
                 0.4
                 0.2
- i..

0.2
                                  0.4
0.6
0.8
                                     z/5

— fto
SYMBOL
o

A

D
V
0
•
A

POWER LAW
STATION
3. Km)

6.1

9.1
12
15
18
21
24
-j
t
1
1
i
i
•j

"1
-1
J
!
Figure 35.   Development of mean velocity profiles along the smooth
             floor  of a  long tunnel  (from Zoric and Sandborn, 1972;
             Reproduced  by courtesy of Boundary-Layer Meteorology,
             D.  Reidel Publishing Co.).
       m   cm
     1.0 r
     0.8 -
     0.6 -
     0.4 -
     0.2 -
8
6

*• 4
K>
2
n
I

1 1 1 1 ! 1 ' '
o DISPLACEMENT THICKNESS, 5* (cm) ^^°^
0 MOMENTUM THICKNESS, 6 (cm) ^^ -x-
A BOUNDARY LAYER THICKNESS, 6 (meters) o*^^ ^^^***X^

' ^^^^^"^
	 .0-*>AA-^
-^
^ 	 D 	 D 	 D 	 ° 	 o 	 a 	 c— I
i i i i i i
) 5 10 15 20 25 30
x (meters)


H
1.5
1 0


Figure 36.  Thickness parameters  for  boundary layer of Figure 35
            (from Zoric and Sandborn,  1972;  Reproduced by courtesy of
            Boundary-Layer Meteorology,  D. Reidel  Publishing Co.).
                                    121

-------
intensities were about 50% of the longitudinal  intensities  close  to  the
ground in accordance with Eq. 3.21,  but were typically 70%  over the
upper 90% of the boundary layer depth.
     Surprisingly little has been published along the lines of neutral
boundary layer turbulence characteristics over  a rough floor that would
show, for example, that the flow was laterally  homogeneous  or how it
would compare as a small scale model of the neutral  atmospheric boundary
layer.  Evidently, detailed basic and systematic studies of the
turbulence structure in neutral boundary layers have not been made in the
long tunnels.
     Additionally, the above measurements were  made at wind tunnel speeds
much in excess of those allowable for modeling  buoyant plumes.  Isyumov,
et al. (1976), for example, suggest typical tunnel wind speeds of 0.5 to
0.7 m/s.  They do present one spectrum, reproduced here as  Figure 37, that
shows the rapid decrease of energy at frequencies in excess of the location
of the spectral peak.  Also, a relatively large amount of energy  at lower
frequencies  is rather surprising in that significant energy in this part
of the spectrum is not generally produced in wind tunnels.   Measurements
of the spectrum of lateral velocities would ascertain whether this energy
is,  in fact, due to turbulence or whether it is "pseudo-turbulence", i.e.,
one-dimensional fluctuations caused, for example, by low frequency
oscillations in fan speed.
                                      722

-------
                         Height=400 ft.
                                 0 Estimated spectral
                                  points from wind
                                  tunnel measurements
                                — Davenport's spectrum
                                  for neutral stability
               .01     .02
           Figure  37.   Spectrum of the longitudinal  component of velocity
                       (from Isyumov et al.,  1976; Courtesy of Air Pollution
                       Control Association).
     Low speed boundary layer development characteristics  in  short
tunnels with passive  devices are much better documented.   The most
popular of the passive  devices is the barrier/vortex generator/roughness
system developed by Counihan (1969).  It has been adopted  with minor
variations at numerous  laboratories.  Castro et al. (1975) have made
                                       123

-------
extensive measurements of a 2m deep boundary layer developed using
Counihan's system, primarily for study of dispersion of chimney
emissions in neutral flow.   They found that the turbulence in the
boundary layer reached a near-equilibrium state in approximately 7%
generator heights (boundary layer heights) downstream.  They were able
to draw this conclusion because they had measured the various terms
in the turbulent energy budget, unfortunately an all too rare measurement.
Their conclusions were that the boundary layer thus developed:
    1.  Had characteristics similar to those of a suburban (or
        somewhat rougher) layer of 600 m depth and roughness length
        of 1.3 m at a scale ratio of 1:300, and that departures
        from equilibrium were unimportant beyond about 5 boundary
        layer heights.  The lower 10 to 20% of the boundary layer
        could be used beyond about 3^6.
    2.  Was Reynolds number independent for free stream wind speeds
        in the range of 0.7 to 13 m/s.
    3.  Appeared to be unaffected by the proximity of the wind tunnel
        ceiling to the top of the boundary layer.1  This conclusion
        was drawn by comparing intermittency distributions with those
        of "natural" wind tunnel boundary layers.
    Moreover, Robins (1978), showed that dispersion in the above wind tunnel
boundary layer as well as that in a simulated rural boundary layer was a
reasonable model of the full scale process, i.e., it produced concentration
patterns approximating Pasquill category C-D atmospheric flows (slightly
unstable to neutral), which is normal for such a large roughness length.
1.  The present author notes that in some unpublished work, boundary layers
developed using Counihan's system were found to be very much dependent upon
the proximity of the ceiling, i.e., when the ceiling was several times the
height of the vortex generators, even the mean velocity profiles differed
drastically from his.  The ceiling thus appears essential; indeed, it is an
integral part of the simulation system.
                                      724

-------
     Somewhat less-well-documented techniques include the "spires"  of Templin
(1969) (also quite popular), the "fence" of Ludwig et al. (1971),  the "coffee"
cups" of Cook (1973).   Earlier methods employing graded blockage,  grids  of
rods or slats, etc., have been largely superseded in the Western Hemisphere,
but are still quite popular in Japan (Sato et al., 1974; Ogawa,  1976, 1981).
Hunt and Fernholz (1975) provided a list of wind tunnels (largely  European)
used for atmospheric boundary layer simulations and included characteristics
of the wind tunnels and relevant measurements of such boundary layers, so that
some comparisons of the different techniques may be made.

     Short test sections with active devices are also numerous.  Schon and
Mery (1971) injected air perpendicular to the flow through a porous plate at
the entrance to the test section.  This system may be thought of as a fluid-
mechanical fence, where the fence "height" is adjusted by varying  the
strength of injected air, but it has the additional potential  for  injecting
gas of different density in order to quickly establish a non-neutral  density
profile.  They have shown that this technique can produce a boundary layer
twice as thick as the  "natural" one over a smooth floor and that its charac-
teristics are essentially similar.  However, this system appears to require
a rather long development length compared with Counihan's system (Hunt
and Fernholz, 1975).  Also, because of the smooth floor, turbulence
intensities were somewhat lower than those in even a mildly rough  field
surface.  Mery et al.  (1974) have shown that this technique produced dis-
persion patterns similar to the Brookhaven experiments (Smith and  Singer,
1955), but only after  "adjusting" their o 's by a factor of 2 to account
for an equivalent wind tunnel  averaging time (converted to full scale) of
                                     725

-------
3 minutes compared with atmospheric averaging timesof 1 hour.  This
adjustment technique, however, appears somewhat arbitrary.  The small
values of the a  's in the wind tunnel were, in the present author's
               */
opinion, most likely due to the small turbulence intensities as well
as to the lack of large scale lateral fluctuations in velocity.

    Nagib et al. (1974) have added some flexibility to the Schon et al.
technique, in that the injected air is input through a line of holes in a
pipe perpendicular to the flow on the floor at the entrance to the test
section.  The pipe may be rotated (See Figure 33) to change the jet
injection angle and the jet speed may be varied; these, of course, change
the boundary layer characteristics.  The "counter-jet" technique, it is
claimed, avoids the objectionable introduction of extraneous turbulence
scales as from vortex generators or grids, but this claim is contested by
Cook (1978).   Nagib et al.  (1974) and Tan-atichat et al.  (1974) show that
this technique produces reasonable boundary layers with adequate lateral
uniformity and that equilibrium is achieved in approximately 4 boundary
layer heights.  Neither turbulence scales nor diffusive characteristics
of this boundary layer has  been measured, however.
    Other techniques in this third category include the multiple-jet
systems of Teunissen (1975)  and the "turbulence box" of Nee et al. (1974),
but neither of these systems appears to have been developed beyond the
initial stages.
                                      126

-------
3.2.3  Simulating the Diabatic Boundary Layer

     Only a few facilities exist for simulating the  diabatic boundary layer.
The oldest and best-known is the Micrometeorological  Wind Tunnel  at the Col-
orado State University (Cermak, 1958; Plate and Cermak,  1963).   It  has nominal
test section dimensions of 1.8 X 1.8 X 27 m and an adjustable ceiling for elim-
inating the pressure gradient due to growth of the boundary layers.   Speeds  in
the test section may be varied from 0 to 37 m/s.   A  12 m length  of  floor can
be heated or cooled and a heat exchanger in the return leg maintains ambient
air temperature equilibrium, permitting temperature  differences  between the
cold floor and hot air of about 65°C and between the hot floor and  cold air  of
about 105°C at "moderate" wind speeds.  At a speed of about 6 m/s,  a boundary
layer thickness between 70 and 120 cm can be obtained as the roughness is
varied (Cermak and Arya, 1970).

     Arya and Plate (1969) have described many characteristics of the stable
boundary layer generated in this wind tunnel and have shown that the surface
layer characteristics are in excellent agreement with field data when scaling
is done according to Monln-Obukhov similarity theory. Their data ranged from
neutral to moderately stable (0 <. z/L <.0.3) in the  lowest 15% of the boundary
layer, which was about 70 cm deep.   To obtain this range of stabilities, the
temperature difference between the cold floor and free stream air was maintain-
ed at 40°C while the wind speed was varied from 3 to 9 m/s.  Measurements
included distributions of mean velocity, temperature, turbulence intensities,
shear stress, heat flux, and temperature fluctuations.  Arya (1975) has pre-
sented additional measurements 1n this stable boundary layer. Thus far, all
measurements in stratified boundary layers in the CSU tunnel have been with
a smooth floor.
                                      127

-------
     The Fluid Mechanics Laboratory at the Ecole Centrale de Lyon has made
extensive measurements of an unstable wind tunnel boundary layer and compared
its properties with the atmospheric boundary layer (Schon et al., 1974;  Mery
et al., 1974; Schon and Mery, 1978).   Flow speeds were typically 2 to 4  m/s
while the floor temperature was maintained 50°C above ambient.   In general,
comparisons with the Kansas data (Businger et al., 1971;  Haugen  et al.,
were quite satisfactory, but, again,  these measurements were made over the
smooth wind tunnel floor; longitudinal turbulence intensities exhibited  a
slight Reynolds number dependence, and the lack of energy in the high fre-
quency portions of the spectra were quite evident, but as noted  earlier,
this effect is expected to be unimportant in terms of simulating diffusion.
The most unstable flow in which diffusion was studied was characterized  by
a Monin-Obukhov length of -1 m, which, when scaled to the atmosphere, cor-
responds to -500 to -1000 m, and is indeed only very slightly unstable (See
Table 4).  Attempts by Rey (1977) and Rey et al. (1979) in adding a rough
floor to this unstable boundary layer showed substantial  changes in the
boundary layer structure with roughness.
     A stratified and closed return wind tunnel has recently been construct-
ed at the Japanese Environmental Agency (Ogawa, et al., 1980).   It has test
section dimensions of 3m x 2m x 24m and an adjustable ceiling for pressure
gradient control.  Speeds may be varied from 0.1 to 11 m/s.  Velocity- and
temperature-profile carts at the test section entrance establish the initial
velocity and temperature profiles.  Sections of the floor (3 meter lengths)
can be independently heated or cooled to maintain stable or unstable boundary
layers, to generate an elevated inversion, or to simulate land-sea breezes.
                                      128

-------
3.2.4  Summary on Simulating the Atmospheric Boundary Layer
     The point of the previous two sections (3.2.2 and 3.2.3)  was to cite
evidence of our ability to simulate at least the main features of the lower
portion of the atmospheric boundary layer in wind tunnels.   No attempt was
made to include in detail  all of the various techniques that have been used,
as that is beyond the scope of this guideline.   The point is only to show
that it can and indeed has been done through various schemes.
     Adequate simulations  of the neutral  atmospheric boundary layer have
been obtained using short tunnels with either active or passive devices
and long tunnels.  Strengths and weaknesses of the three types, as far as
their ability to produce adequate boundary layers is concerned, appear to
be evenly balanced.  Hence, no one technique or system is recommended over
any other.  Due to the large number of permutations and combinations and to
the possible large changes in boundary layer structure with seemingly small
changes in configuration,  however, 1t is  imperative that, whatever technique
is used, the boundary layer characteristics are adequately documented.
     Simulations of diabatic boundary layers have been accomplished using
wind tunnels with heated and cooled floors, but present technology allows
only small deviations from neutrality, i.e., mildly stable or mildly unstable.
Also, because normal roughness elements on the floor would reduce heat trans-
fer even further, most measurements to date have been made over smooth wind
tunnel floors.  Recent exceptions indicate experiments over cooled, rough-
ened terrain made from formed aluminium foil by Peterson and Cermak (1979)
and unstable boundary layers developed by Rey et al. (1979).  As we have seen
in Figure 22, the inability to use a rough surface could be important, at
least for unstable flows and large roughness lengths.
                                      729

-------
     Adequate documentation of the boundary layer characteristics  should
include, as a minimum:
     1.   Several  vertical  profiles of mean  velocity,  turbulence  intensity
         (3 components),  and Reynolds stress  throughout  the  region  of
         interest to establish that the  boundary  layer is  non-develop-
         ing (or  at least very slowly developing),  and is  similar  to the
         target atmospheric boundary layer  (zQ,d,u*).  In  a  stratified  bound-
         ary layer, of  course, profiles  of  temperature and temperature
         fluctuations should also  be included and the stability  parameters
         should be matched.
     2.   Lateral  surveys  of mean velocity and turbulence structure  at
         various  elevations  to ascertain the  two-dimensionality  of  the
         boundary layer.
     3.   Spectral  measurements of  the turbulence  to determine that  the
         integral  scales  and the shape of the spectra are  appropriate.
     4.   Dispersive characteristics  of the  boundary layer  (in the absence
         of a model)  to determine  that the  concentration patterns are
         reasonably similar  to those expected in  the target  atmosphere,
         e.g., the appropriate Pasquill category.
                                      130

-------
Perhaps the most critical test of the boundary layer is the measurement of
its dispersive characteristics to determine whether appropriate concentration
patterns result.  This point cannot be over-emphasized.  Wind tunnels are
generally extremely difficult to operate at low wind speeds (
-------
3.3  FLOW AROUND BUILDINGS
     We discuss here guidelines to be followed in modeling flow around
buildings, e.g., in order to determine a necessary height for a stack on
a power plant to avoid downwash of the plume in the wake of the plant.
The class of problems covered includes single or small groups of buildings,
primarily isolated ones in a rural environment, i.e. scale reductions in the
range of 1:200 to 1:1000.  It is evident from preceding sections that the
building must be immersed in an appropriate boundary layer.  Geometrical
scaling implies that the ratio of the building height to boundary layer
height must be matched and, of course, that all length scales be reduced by
this same ratio.  A minimum building Reynolds number criterion must be met
as discussed in Section 2.2.2.2 and further elaborated here.   Finally, the
effluent plume behavior must be modeled as discussed in Section 3.1.
3.3.1  Discussion
     Geometrical scales that come to mind are stack height H   and diameter D,
                                                            5
building height H, boundary layer depth 6, roughness length ZQ, and, if
stratified, Monin-Obukhov length L.  There are, of course, many other length
scales and geometric scaling requires that all lengths be reduced by the
same ratio.  However, this brings up the question of how much detail is
required, i.e., is it necessary to include in the scale model a particular
protuberance, say, from the roof of the building?  The answer, of course,
depends upon the size and shape of the protuberance; it is a  question of
whether or not the obstacle has a separated wake.  Some guidance may be obtained
from Goldstein (1965), where it is stated that provided the size e of the
                                      732

-------
protuberance is such that e u*/v<_4, it will have little effect on the
flow in a turbulent boundary layer on a flat plate.   Hence, protuberances
smaller than e = 4v/u* need not be reproduced in the model.
     A closely related but more demanding problem is as follows.  A given
surface is aerodynamically smooth when the Reynolds  number is below
a certain value; it is rough when Re exceeds this value.  Hence, all
surfaces are rough when the Reynolds number is sufficiently large.  Because
field values of Reynolds numbers are almost always very large, we may
assume that surfaces of typical buildings are aerodynamically rough.
As we reduce sizes of buildings to fit into our wind tunnel, we also
reduce the Reynolds numbers, so that the surfaces become aerodynamically
smooth.  Hence, locations of separation points, the  drag coefficient,
and the general character of the flow along the model surface will be
affected.
     Again, from  Goldstein  (1965),  if Rec  = e  u*/v  >_ 100,  these flow
phenomena will be independent of Reynolds number. These results indicate
that small details need not be reproduced and, indeed, that model surfaces
should be roughened to the point that the critical Reynolds number is at
least 100.
     Crude estimates will suffice here and an example will  help to clarify
the procedure.  Suppose our model building has a height of 20 cm, and in
order to model the buoyancy in the plume, we have reduced the wind speed to
1 m/s.   The friction velocity is typically 0.05 U^,  so that size of
roughness elements with which to cover the surface of the model building
 S                       e « (100)  (0.15cm2/s)/5cm/s=3cm.
                                      733

-------
     This is, in general, an unacceptably large roughness size, as we should
probably also restrict e/H
-------
slight variations over the entire range of Reynolds numbers with neutrally
buoyant effluent and with an effluent-to-free-stream velocity ratio of unity.
However, the concentration at a point on the roof itself was found to vary
strongly with Reynolds numbers less than 11,000, but to be invariant with
Reynolds numbers between 11,000 and 94,000.  Thus, a critical Reynolds number
may be defined, which, with this type of geometry, appears to be 11,000.
     Golden's value for the critical Reynolds number for flow around cubes is
frequently cited in the fluid modeling literature on building downwash
problems.  Whereas Golden's value was established for a smooth surfaced
cube on a raised platform facing a uniform approach flow of very low turb-
ulence intensity, it is applied "across-the-board" to all shapes and orient-
ations of buildings, in all types of approach flow boundary layers, and without
regard to the building surface roughness, all of which will affect Re .   Also,
Golden's value was established primarily through the measurement of concentra-
tions at only one point on the roof of the cube, as opposed to measurements of,
say, the concentration fields in the wakes.  Far too much confidence seems
to have been placed in his result.  It is probably conservative as the
shear and high turbulence in an approach boundary layer as well as a rough
building surface are likely to reduce the critical Reynolds number.  Also,
as pointed out by Halitsky (1968) lower values are probably acceptable if
measurements are restricted to regions away from the building surface.  Hence,
a critical  Reynolds number of 11,000 is a useful and probably conservative
value for model  design purposes, but tests to establish Reynolds number
independence should be an integral part of any model study until such time
that firmer values are established.
                                      135

-------
     A study by Smith (1951)  may also be regarded as  a test of Reynolds  number
independence.  He investigated the sizes of the wakes created by both  model  and
prototype sharp-edged buildings.  He assumed that the flow was independent
of Reynolds number effects if the ratio of the length of the cavity region to
the building height was the same in both model and prototype.  In the  proto-
type tests, he found this ratio to be constant for Reynolds numbers (based
upon an appropriate characteristic length) between about 2x10  and 2x10  .   In
the model tests, this ratio was constant for various  block models over a range
of Re from 2x1O4 to 2x1O5.
     In more recent work, Castro and Robins (1977) investigated flow around a
surface mounted cube in uniform and turbulent upstream boundary-layer flows.
Reynolds number independence was observed in the uniform flow when Re >  30,000
and in the shear flow when Re > 4000.  These statements were based primarily
on measurements of mean surface pressures on the body, which are evidently
more sensitive than surface concentration measurements as made by Golden
(1961).  The trend toward much lower critical Reynolds numbers in turbulent
shear flows, however, is clear.  It should be noted that the simulated atmos-
pheric boundary layer of Castro and Robins was of a suburban character,  i.e.,
a simulated roughness length of 1.3m, so that it was, indeed, highly turbulent.
     Critical Reynolds numbers for other geometrical  shapes remain to be
determined.  A study by Halitsky et al. (1963) on a reactor shell (a hemis-
phere fitted on a vertical cylinder) indicated a critical Reynolds number
greater  than 79,000. The separation point, and, hence, the pressure
distribution for rounded buildings is affected by the Reynolds number.
Generally  speaking, the more streamlined is the object, the larger is the
critical Reynolds number.  It is quite  likely that with rough surfaces,
critical Reynolds numbers  for streamlined objects may be reduced substantially,
                                      136

-------
 and  systematic  studies need to be done in this area.
      Suppose  there  is another building a distance s upstream of our example
 power plant;  is  it  necessary to incorporate this building into our wind tunnel
 model?  Some  guidance is provided by Hunt (1974), who reviewed experimental
 results of  several  investigators and showed that the velocity deficit in the
 wakes of cubes  and  cylinders is given by
downwind of the separation bubble, where All   is the maximum mean velocity
                                           II IA
deficit created by the obstacle, h is the height of the obstacle, x is the
distance downstream of the obstacle, and A is a constant which is
dependent upon the building shape, orientation, boundary layer thickness,
and surface roughness.  Typically, A = 2.5, although it may vary from
that by a factor of 2.  Hence, if we require that the velocity be within
3% of its undisturbed value, then a cubical building as high as s/20
must be included upstream of our power plant.  This result, however,
is dependent upon the aspect ratio; a building with its width much greater
than its height, for example, would require inclusion if its height
were greater than s/100 (See Section 3.4).
     The ratio of the cross-sectional area of a model to that of the tunnel  is
referred to as "blockage", 3.  It is easy to show, through the principle of  mass
continuity, that the average speed-up S (increase in velocity) of the air stream
through the plane intersecting the model  is equal  to S = e/(l-e).  Of course,  in
the atmosphere, there are no sidewalls or roof to restrict divergence of the
flow around the model, so that the average speed-up is zero.  Wind tunnels
with adjustable ceilings can compensate to some extent by locally raising
                                      737

-------
the height of the ceiling above the model  itself (with gentle slopes upwind
and downwind of the model).   In fact, the  average speed-up can be reduced to
zero by raising the ceiling  such that the  additional  cross-sectional area
of the tunnel is exactly equal  to the cross-sectional  area of the model,  but
it is obvious that this is not a perfect "fix", as that would require local
expansion of the sidewalls at the same time.
     Some unpublished measurements by the  present author on the flow over a
two-dimensional ridge sheds  light on this  problem.  Measurements of velocity
profiles above the crest of  the ridge were made with a flat (unadjusted)
ceiling where the blockage caused by the ridge was 10%.  The ceiling
height was then adjusted until  longitudinal surveys of velocity at an elevation
5 hill heights above the tunnel floor showed a nonaccelerating flow.  Vertical
profiles of mean velocity were similar in  shape to those measured with the
flat ceiling, but the magnitude of the wind speed was lower by 10% everywhere
above the crest  (see Figure  38) with little change in the root-mean-square
values of the longitudinal or vertical fluctuating velocities (turbulence).
It is apparent  (but by no means proven) that blockage would reduce the
vertical width of a plume by approximately 10% as it traversed the ridge,
but, because its centerline would be 10% closer to the ridge crest, resulting
surface concentrations upstream of the crest would be essentially unchanged.
However, because the flow acceleration changes the pressure distribution around
the model, which will  in turn affect the location of the separation point,
the effects  downstream of the crest are not apparent.  Blockage  "corrections"
for conventional aeronautical wind tunnel  models  is a highly involved
engineering  science problem.   "Rules of thumb" indicate a  limit  of  5%
                                      138

-------
/DU
600
E
E
a.'
O
J] 450
X
o
CO
< 300
i—
X
o
IU
X

150


n
<• 	 T 1 I 1 | 1 1 1 U-t
_


A
—


n A
/
-
D A
L ' '
n A
i i
D A
1 1
n A
/ f
n A
n A
Q A
,,.11,,!,%,
.5   1   1.5   2  2.5   3   3.5


           MEAN VELOCITY, m/s
                                                           4.5
   Figure 38.  Velocity profiles above crest of  triangular ridge indicating
               effect of blockage  (A flat ceiling,  10$  blockage;  a  raised
               ceiling, nonaccelerating free-stream flow).
blockage in the ordinary wind tunnel and somewhat  higher,  perhaps  10%,  in

a tunnel with an adjustable ceiling.  Blockage effects  in  stratified facili-

ties are even less well understood and are potentially  much more  important.
                                      139

-------
3.3.2  Recommendations

     To model  the flow  and dispersion  around individual  or  small  groups  of
buildings, it  is recommended that:
     1.  The building be immersed in an  appropriate  boundary  layer,
         the main features of which  include  matching of  the ratios
         of roughness length, boundary layer depth and,  if  stratified,
         the Monin-Obukhov length to building height.
     2.  The effluent plume be modeled as  discussed  in Section  3.1.
     3.  Reynolds number independence  tests  be conducted as an  integral
         part  of the model study. For design purposes,  a minimum building
         Reynolds number UnH/v = 11,000  appears to be conservative.
     4.  The surface of the building be  covered with gravel of  size  e  such
         that  eu*/v=100.  If this results  in excessive roughness, i.e.,
         e/H > 1/30, compromises may be  made, but in no  case  should  eu*/v be
         less  than 20.
     5.  Another building or major obstruction upstream  of  the  test  building
         be included if its height exceeds l/20th of its distance from the
         test building.  This recommendation applies to  a roughly cubical
         obstacle.  An  obstruction whose crosswind dimension  is large
         compared to its height must be  included if  its  height  is greater
         than  1/100th of its distance  upstream (see  text).
     6.  Blockage caused by a model  be limited to 5% in  an  ordinary  wind
         tunnel and to  10% in a tunnel with  an adjustable ceiling.
                                        140

-------
3.4  FLOW OVER HILLY TERRAIN
     Guidelines for modeling neutral  flow over hilly terrain are essentially
similar to those for modeling that around buildings; hence, only a few of the
unique features of terrain will be discussed in this section.  Differences occur
primarily because terrain is generally much more streamlined than are buildings
and because the roughness is generally more patchy.   Whereas separation of the
flow from a building surface will almost always occur at a sharp corner, the
separation point for a hill with a rounded top may fluctuate in position with
time, and may occur on the downwind slope, or for a  hill with low or moderate
slope, may be absent entirely.  Also, the stratification in the approach flow
can drastically change the nature as  well as the location of the separation
and may enhance or eliminate separation entirely (see Hunt and Snyder, 1979).
     We will first discuss neutral flow, emphasizing the differences between
modeling the flow around hills and that around buildings.  Because stratified
flows are so different from neutral flows, they will be discussed in a
separate section.  The two sections are summarized with a set of recommendations.
3.4.1  Neutral Flow
     In the field, the ridge Reynolds number based on a ridge height of 75 m
and wind speed of only 3 m/s is 10 .   For these very large Reynolds numbers,
at least for a ridge with steep slopes, separation is certain to occur
near the apex, even for a ridge with  a smooth rounded top (see Scorer, 1968,
p. 113),  The Reynolds number for this model mountain ridge would lie
between 10  and 10 , much smaller than the full scale Re.  It is possible to
trip the flow at the apex (as done by Huber et al.,  1976) or to roughen  the
surface, so that the point of separation on the model will occur at its
apex and similarity of the two flow patterns will  be achieved.
                                     141

-------
Appropriate roughening of the surface, as outlined below, is the "safest" of
the two techniques, because proper placement of a trip requires foreknowledge
or possibly unwarranted presumptions of the location (or indeed existence) of
separation.
     Concerning the minimum size of protuberances that must be reproduced in
the model and also the size of the roughness elements that cover the model to
make the flow independent of the Reynolds number, we apply the same criterion
as established for buildings, namely e = 20 v/u* = 400 v/U^.  This may in some
instances conflict with Jensen's criterion that h/z  be matched between model
and prototype, but the minimum Reynolds number is regarded as more important.
     A common practice in constructing terrain models is to trace individual
contour lines from enlarged geographic maps onto plywood or styrofoam, then
to cut them out and stack them to form "stepped" terrain models.  Some
laboratories then fill in or smooth out these irregularities, while others use
rather large steps and do not smooth them.  One laboratory, in fact, proposed
to fill in and smooth the model, then to add randomly spaced blocks to simulate
surface roughness.  Application of the criterion in the previous paragraph
shows both the desired step size and the roughness element size.  It is not
necessary to fill-in the steps if the step size is chosen appropriately at the
beginning; the steps double, to some extent, as roughness elements, although
it is most likely better to densely cover the model surface with gravel of
about the same diameter as the step size.
     How much terrain is it necessary to include in the model upwind of a
power plant?  For a two-dimensional ridge, Counihan et al. (1974) have
shown that the maximum deficit of mean velocity in the wake, normalized by
                                     142

-------
the mean velocity at the height of the hill, decays  as
                            AUmx *  B
                            OW xTfi" '
where All   is the difference in mean velocity created by the hill,  h is the
        Ill/v
hill height, x is distance downstream of the hill, and B is a constant depen-
dent upon surface roughness and hill shape.  Typically, B=3.0.  Hence, if we
insist that the mean velocity be within 3% of its undisturbed value (i.e., its
value in the absence of the ridge), then all upwind ridges with heights as
large as x/100 should be included in the model.  In actual practice, one
should study the topographic maps of the area surrounding the plant, locate
prominent ridges upstream, then determine the height of each ridge  and its
distance from the plant.  If its height is greater than x/100, all  terrain
between the ridge and the plant should be included   If h
-------
the fonrulas do not indicate the fetch required for the development of the
boundary layer.
     The choice of a boundary layer depth for very rugged terrain is a difficult
task.  Our choice of 600m (Section 3.2.1.2) is obviously absurd if the heights
of the hills themselves are of the same magnitude.  Ore indication from the
literature is from Thompson (1978), who examined wind profiles obtained from
pilot balloons over complex terrain in southwestern Virginia.   The average
boundary layer depth, Thompson concluded, was about 800 m, or  4 hill heights
under moderate to high wind speed neutral conditions..   As mentioned in Sec-
tion 3.2.1.1, he also observed a logarithmic wind profile with z  of 35m.
3.4.2  Stratified Flow
       We have discussed in depth the stable boundary layer in Section 3.2.1.4.
It was shown that under stongly stable conditions, the  boundary layer is very
shallow, typically less than 100 m.  Frequently, pollutant sources discharge
their effluent at much higher elevations, i.e., above the stable boundary
layer, where the plume may be transported long distances with  little or no
dispersion (e.g., see cover photograph of AMS, 1979).   Further, results
of Godowitch et al. (1979) indicate that extremely shallow stable boundary
layers under quite deep surface-based temperature inversions are typical
at sunrise at a rural  site outside St. Louis, MO.  The  average depth of
the nocturnal inversion, for example, was 325 m (±90 m  standard  deviation).
The average temperature gradient was 1.4°C/100m.  Under these  conditions,  it
is evident that simulation of the stable boundary layer beneath an elevated
source is relatively unimportant.  Far more important is the simulation of
the stability above the boundary layer because, as shown by Lin et al. (1974),
                                     144

-------
Hunt and Snyder (1979) and Snyder et al.  (1979), the stability deter-
mines the essential structure of the flow, i.e., whether plumes
will impinge on the hill surface or travel over the hill top, the
size and location of hydraulic jumps, etc.
     Under strongly stable conditions, the flow is constrained to
move in essentially horizontal planes.  If a three-dimensional hill
is placed in the flow field, streamlines  below the hill top will
pass round the sides of the hill and not  over the top.   If a two-
dimensional hill is placed across the flow field, the fluid obviously
cannot pass round the sides and, because  it has insufficient kinetic
energy, it cannot pass over the top of the hill (see Section 2.2.4).
Thus, upstream blocking of the flow below the hill top will occur.
The point is that the modeler must be very careful in determining
the amount of terrain to duplicate in the model.  An example is
shown in Figure 39, where a portion of a  three-dimensional hill is
turned into a two-dimensional one by an inappropriate choice of
the area to be modeled.  Under strongly stable conditions, the com-
bination of the hill and tunnel sidewalls would result in upstream
blocking of the flow beneath the hill top, whereas, with a wider
tunnel or smaller scale model, the flow would be diverted around the
sides of the hill as would certainly occur in the atmosphere.  Similar
extensions of this type of reasoning apply to valleys and ridges angled
diagonally across the flow stream.  It is impossible to give firm and
                                     745

-------
  Figure  39.  Contour map of three-dimensional hill showing inappropriate
             choice of area to be modeled.
fixed rules for determining the appropriate area of terrain to model because
the flow field must be known a priori.  However, detailed study of topographic
maps of the area and common sense will avoid most pitfalls.
     As Scorer (1968) has pointed out, laboratory studies of stratified flows
                                      146

-------
tend to overemphasize the effect of stratification  in  the  approach  flow;  local
heating and cooling of hill  surfaces are equally, perhaps  more,  important.
Effects of anabatic and katabatic winds are not only local,  but  may have  large
effects on the flow structure by inhibiting or enhancing separation (Scorer,
1968, p. 113; Brighton, 1978).   Some attempts have  been made to  simulate  heat-
ing of terrain surfaces, but primarily to simulate  fumigation of elevated
plumes (Liu and Lin, 1976; Keroney et al., 1975; Ogawa et  al., 1975),  rather
than to model anabatic or katabatic winds (Petersen and Cermak,  1979).  Other
local heating and cooling problems include lake-shore breezes and urban heat
islands.  Little is known of the proper similarity  criteria  to be applied to
thermally-driven flows.  Any comparison between field and  model  experiments,
where such thermally generated winds are absent, must be made with  great
caution.

3.4.3  Recommendations
     Recommendations for modeling flow and dispersion over hilly terrain  in
neutral stability are essentially similar to those  for modeling  flow around
buildings.  It is recommended that:
     1.  The terrain be immersed in a simulated atmospheric  boundary layer,
         matching the ratios of roughness length and boundary layer depth to
         hill height.
     2.  The effluent plume be modeled as discussed in Section 3.1.
     3.  Reynolds number tests be conducted as an  integral part  of  the
         study.
                                     147

-------
     4.  The surface of the model be covered with gravel of size e such
         that eu*/v>20.  The step size in "stepped" terrain models should
         also be of order e.

     5.  An upstream ridge be included in the model if its height exceeds
         l/100th of its distance upstream from the test portion of the
         model.  A three-dimensional hill should be included if its height
         exceeds l/20th of its distance upwind.

     6.  Blockage be limited to 5% for an ordinary wind tunnel  and to 10%
         in a tunnel with an adjustable ceiling.

     Additionally, in modeling dispersion from elevated sources in strongly

stably stratified flow over hilly terrain, it is suggested that:

     7.  The simulation of the stable boundary layer, per se, is relatively
         unimportant.  More important is matching of the Froude number based
         on the hill height and the density difference between  the base and
         top of the hill.

     8.  Topographic maps of the area to be modeled should be studied care-
         fully to insure that an appropriate area is modeled (see text).

     Finally, laboratory models to simulate anabatic and katabatic winds

must be considered as exploratory in nature at the present time.
3.5  RELATING MEASUREMENTS TO THE FIELD

     Since buoyancy in a plume may be modeled using light gas as opposed to

temperature, the concentration measured in a model  facility may be related to

that in the field through the nondimensional concentration x " CUH2/Q, where

        C = mass concentration of pollutant (ML"3),

        U = wind speed (LT"1),

        H = characteristic length (L),

    and Q = pollutant emission rate (MT~ , e.g., grams of S02/second).

     The relation between field and model  concentrations beyond a few dia-

meters from the source is thus
                                       UU  •)  f\
                                   _   n_ c.  \jf
                                      148

-------
Note that both C and Q are measured in mass units.   More frequently, C and Q
are measured in volume units, in which case, they must be evaluated at ambient
(not stack) temperature, including Qf (Robins, 1975).   An example may help to
clarify the procedure.  Suppose we model  a buoyant  plume with a mixture of
helium, air and methane as indicated below:
              Property                 Field           Model
           Reference wind speed        10 m/s            1 ir/s
           Stack height                50 m             50 cm
           Stack diameter               5m              5 cm
           Effluent speed              20 m/s            2 m/s
           Pollutant emission rate    500 g/s(S02)        1 g/s(CH4)
           Effluent temperature       117°C             20°C
           Ambient temperature         20°C             20°C
           Effluent specific gravity   0.75             0.75
At some point downwind of the source in the wind tunnel, we might measure a
model concentration of 100 ppm (parts CH^ per million  parts air, by volume),
and the problem is to relate this to a field concentration value.  First, the
model volume concentration must be converted to a mass concentration (relevant
densities at 20°C are air: 1.20 g/£; CH4: 0.68 g/£; S02: 2.72 g/£):
                   100 £ CH.    , .  .      0.68 g  CH.    52.7 x 10"6g CHA
                = f	±\ ( ' *• air  \ I	iL •
                  V  C      / \1 On n a-i*^ V
um   v   f       I  M  ?f) n air'  v  1  9  CM    '         Q  air
      10  f  "'"'                      "
Hence, the field mass concentration will be:
                       52.7 x 10"6g CHA    ...   n ,._ 2  500 g S0?/s
                      /               4\ /im/s\ /u.om\  /
                       _        _
                  f "        g~avr        TOii/V  ~^Gm     1 g CH4/s
                    = 0.264 x 10"6g S02/g air.

Converting this mass concentration to volume concentration yields:
                                     J49

-------
                          0.264 x 10~6g S09   ,  9n    ,      1  £ S09
                    r  = t	       2% fl.20 g airx  /	2  }
                     f   v      g air      ' v 1  £ air'  V2.72 g S02'
                       = 0,116 x 10"6 £ S02/£ air = 0.116  ppm S02>

To summarize, the relation between model  and field concentrations in this
example is
                 1 ppm S02 .». 858 ppm CH.
            or   1 g S02/g air + 200 g CH4/g air.
Whereas it is tempting to bypass some of the above steps by using volume
emission rates, shortcuts are not recommended.  It is important to note that
all densities were specified and used at ambient  temperature,  i.e., 20°C.
     If model concentrations are to be related to field  concentrations very
close to the source and the density difference is exaggerated, then Eq. 3.48
is not strictly correct.  Considerations  of geometric similarity of model
and field plumes and perfect gas laws show that
                         ^m _ pp'pap ,
                         xp   pm/pam
where p1 is the local density (i.e., not reduced to ambient temperature).
This equation is not very useful  by itself because local  densities are not
generally measured.  To reduce it to useable form, however, will  depend
upon the precise means of simulating the buoyant plume and the concentration
measurement technique.  One such  correction scheme has been derived by
Meroney (1978) in connection with liquid-natural-gas spills.
     Regarding the comparison of  model  results with field results, it is
well-known that in the field the  averaging time has a definite effect on the
                                     /so

-------
measured concentrations.   This is not the case in model  tests.   (This
discussion is taken largely from Ludwig,  1974).   The model  results  correspond
to short-time-averaged field measurements, taken over not more  than 10 or 15
minutes in most cases.  Briefly, what is  involved here is the following.   The
energy spectrum of wind gusts in the atmosphere  generally shows a null, or near
null, in the frequency range of 1 to 3 cycles per hour (the "spectral  gap"
discussed in Section 2.2.2.2).  Thus, it  is possible to  separate the spectrum
into two parts and to deal with the phenomena associated with each  part
separately.  The high-frequency portion,  related to the  roughness of the
surface and the turbulence around obstructions is well-simulated in a  wind
tunnel.  The low-frequency portion, related to the meandering of the wind,
diurnal fluctuations, passage of weather  systems, etc.,  cannot  be simulated in
a wind tunnel.  However,  a correction for meandering of  the wind can be applied,
if desired, to derive longer term averages (Hino, 1968;  Isyumov, et al.,  1976).
Model averaging times, on the other hand, are chosen to  provide data that are
repeatable to within some specified accuracy, as discussed  later.  However, as
noted above, the data so  obtained will  correspond to field  data measured while
the wind direction is essentially steady, which  is generally not more  than 10
to 15 minutes.  Shorter term averages obtained from the  model can be related
to the short term fluctuations in the atmosphere, and instrumentation  is being
developed to accomplish this (Fackrell, 1978; 1979).

3.6  AVERAGING TIME AND SAMPLING RATES IN THE LABORATORY
     Because the flow is  turbulent, essentially  all  of the  quantities  we
attempt to measure will fluctuate in time.  Generally, we will  deal  with a
fluctuating electrical signal from a transducer, and it  is  not  the  precise
                                     151

-------
value at any particular instant of time that 1s  of Interest,  but  rather the
average values and the statistics  of the fluctuations.   It  is necessary at
some point to determine how long an averaging period is  required  to obtain
a stable average.  Frequently, it is convenient  to digitize an analog signal
(sample it and convert the analog voltage to digital form).  Sampling at too
high a rate is a waste of resources; sampling at too low a  rate may not allow
us to obtain the desired information and, in fact, may  lead to incorrect
answers .  Hence, it may be necessary to determine an appropriate  sampling
rate.
     To determine an appropriate averaging time  for measuring the mean of a
fluctuating quantity F(t) in a wind tunnel, it is useful to consider the
turbulence as a Gaussian process.   (Whereas turbulence  is not a Gaussian
process, experience has shown that this assumption leads to quite reasonable
estimates of the errors involved, except in extreme cases of signal inter-
                          n
mittency.)  The variance o  of the difference between the ensemble (true)
average and the average obtained by integration  over time T is (Lumley and
Panofsky, 1964):
                                 2    ~2~
                                a  = 2ri/T ,
      — x-                                                            _
where f  is the ensemble variance of F about its ensemble mean, f=F-F and I
is the integral scale of F.  The fractional error e, then is given by

                       e2 =
If, for example, it is desired to measure the turbulent energy u , it may
be shown, using the assumption that u has a Gaussian distribution, that
                            e2 = 4I/T.
     To obtain a conservative estimate of the error, it is convenient to
                                      752

-------
                   1000
                .5  100
                C9
                z

                3

                oc
                111

                <   10
                     1
                      1
                                           e=1%
3%
                                            10%
          6   7  8 9 10
                     2      345


                         VELOCITY, m/s


Figure 40.  Averaging time requirements for wind tunnel

                                              ~y
            measurements of turbulent energy, u .
estimate a maximum integral scale as 6/1^, where 6 is the boundary  layer


depth  and U^ is the free stream velocity; the required averaging time  is



                                 T = 46/U e2  .
                                         00


     This relationship is shown in Figure 40  for a typical boundary layer


depth of 1m.  For a wind tunnel speed of 4m/s and a desired accuracy of 10%,


a two-minute averaging time is required.  It  is readily observed that much


higher accuracies at such low wind speeds are impractical, as an accuracy of


1% would require an averaging time of over 3  hours.


     To estimate averaging times required for measuring other quantities


(besides turbulent energy), respective integral scales must be known and


numerical factors are generally larger; however, experience has shown that

                                      153

-------
T = 46/U e   is a reasonably good estimate of the averaging time required
        GO
for making all of the common measurements (mean velocity, turbulence intensity,
concentration, etc.).  Higher order moments require considerably longer averaging
times.
     To estimate an appropriate sampling rate, we begin by drawing from a
mathematical theorem (Miller, 1963):
               "If a signal f(t) extending from 0 to « contains no
               frequencies above W cycles per second, then it is
               completely determined by giving its ordinates at a
               sequence of points spaced 1/2W seconds apart."
Hence, in order not to lose information from our continuous signal through our
discrete sampling, it will be essential to determine the highest frequency
component in our signal and to sample at twice that rate.  The highest
frequency of any significance in the turbulence is the Kolmogoroff frequency,
f.=U/2irn (see Section 2.2.2.2).  Hence, assuming an excellent anemometer
(good frequency response and low electrical noise), all information about the
turbulent signal may be obtained by sampling at a rate of 2fd=U/irn.  At the
slow  flow speeds typical of fluid modeling studies (<10m/s), the Kolmogoroff
microscale  is not  likely to be much smaller than 0.5 mm.  Hence, a typical
sampling rate would  be approximately 2 kilohertz at a flow speed of 5m/s.
Of course,  if the  transducer or amplifier have slower frequency response, it
is pointless  to sample at  twice the Kolmogoroff frequency.  A flame ionization
detector, for example, has a time constant of approximately 0.5 sec., so that
a  sampling  rate in excess  of 4 hertz is not necessary.
         Strictly  speaking, the above  discussion on sampling rates applies
 "across-the-board" to  all  types of measurements; because  of aliasing, lower
sampling rates could conceivably yield incorrect results  (for more information,
                                     154

-------
see Lumley and Panofsky, 1964).  However, unless the signal is so periodic
as to be totally atypical of atmospheric or laboratory shear flows, such high
rates are necessary only for spectral (or autocorrelation) type measurements.
Generally, it is sufficient to ensure that
     1.  The averaging time is long enough, and
     2.  A sufficient number of samples is taken to reduce statistical  errors
         arising from finite sample size.
     For normally distributed velocity fluctuations, estimates of required
sample sizes for the determination of mean velocity and turbulence intensity
are (Bradbury and Castro, 1971; Mandel , 1964)
where n is the number of samples required to ensure that the estimates of
mean velocity and turbulence intensity are within ±AU and ±A(u2)   ,  respect-
ively, of the ensemble values,  t is a parameter called the Student's t
distribution function; it has values of 1.96 for a 95% probability of being
within the interval and 1.645 for a 90% probability.   As an example,  if the
turbulence intensity were 50%, then, to determine the mean velocity within 5%
of the ensemble mean velocity with 95% probability, it would be necessary to
average 384 samples.  The turbulence intensity determined with 384 samples
would be within 7% of the ensemble intensity with 95% probability. These
estimates do not, of course, include experimental errors.  They also  assume
that the samples are taken far enough apart as to be  independent or,  from
another viewpoint, that the total length of record meets the averaging time
requirements discussed previously.
                                      155

-------
                               4.   THE HARDWARE

4.1  GENERAL REQUIREMENTS
     Model experiments may be conducted using either air or water as  the
medium, although air is by far the most coronon.  The present section  con-
siders the size and performance characteristics required of the facility;
it is written primarily with a view toward wind tunnels, but, of course, the
principles are equally applicable  to water facilities.   The requirements of
the facility are defined by the scaling laws  that have  been covered in  prev-
ious chapters.  In Section 4.2, the advantages and disadvantages of air versus
water will be discussed in detail.

4.1.1  The Speed Range and Scale Reductions
     The roughness Reynolds number criterion  which insures  that the flow is
aerodynamically rough (Section 2.3.2) is
                                u*zQ/v>_2.5,                      (4.1)
which may be written as
                             U.6/V >.2.5(UaB/u*)(6/z0).              (4.2)
where 6 is the boundary layer depth, U^ is the free stream  tunnel  speed, and
the nondimensional parameters in parentheses  are to be  matched in model and
prototype.  As a typical example,  consider a  problem wherein it is required
to determine the height of a stack near a power plant that  is located in a
suburban area where zQ=lm and 6=600m.  From Figure 14,  u^/U^  0.05,  so  that
the free stream speed is
                                     756

-------
                             IL6   U fif
                             — = -fg1130,000,                   (4.3)
                       or    Uw >_0.00075S (m/s),                   (4.4)
where 6- is the atmospheric boundary layer depth and S is the geometric scale
reduction factor, typically 300
-------

-------
specifically applicable only to the particular example illustrated, the
general  features are universal, i.e., there are excluded regions of the
same general shapes for all  model  studies.   Similar graphs may also be
constructed to determine whether particular studies can be successfully
conducted in particular wind tunnels.  Other factors also enter into this
decision; these are discussed further in the next subsection.

4.1.2  Test Section Dimensions
     From the previous section, it is evident that boundary layer depths on
the order of 0.3 to 2m are required, so that the height of the test section
should be somewhat greater than 2m.  Following Robins (1975),  the length of
the test section is determined by four distances:
     L-,  - the length occupied by the boundary layer generation system
     Lp - the fetch required to establish a homogeneous, equilibrium boundary
          layer
     L^ - the upwind fetch of a topographical model
     L. - the distance downwind to and somewhat beyond the point of maximum
          concentration.
     For the Counihan (1969) system, L,=2.5<5 and Lp^S.Bfi.  The required upwind
fetch of a topographical model would depend upon the details of the specific
topography, but 20 stack heights (HS) may be considered typical (in fact, L2
and L3 may overlap in some cases).  The distance to the point  of maximum
concentration is typically 10 to 20(Hs+Ah), where Ah is the plume rise.
Hence, the required test section length is
                            L = 86 + 40Hs + 20Ah.
     For a 300m stack modeled at a scale ratio of 600:1, the required test
section length would be approximately 35m!
     The width of a plume at the point of maximum concentration downwind of
                                     759

-------
a tall chimney is typically 26.   Effects of the sidewalls  may extend into
the flow a distance of approximately 1  test section height.   Hence,  the
required width of the test section is about 46.
     It should now be evident that many factors must be considered in the
process of selecting a scale factor for a particular model  study or  in
deciding whether a particular study can be conducted in a  particular fac-
ility.  We can, however, give some fairly definite guidance.   Consider the
wind tunnel simulation of the dispersal of buoyant effluent from a tall
chimney in a neutrally stable atmospheric boundary layer.   From the  prev-
ious discussions, it is evident that scale reductions between 300 and 1000
are possible and, for a tall stack, a choice of 1000 would almost certainly
be made in order to accomodate the model within the test section. Thus,
a 600m atmospheric boundary layer implies a 0.6m model  boundary layer. The
fetch required for boundary layer development, that necessary for upstream
flow "conditioning", and that required to obtain the maximum ground  level
concentration within the test section imply a test section length of about
15m and, consequently, a height of 1m and a width of 3m; an operating speed
in the range of 0.7 to 3m/s is also implied, there being no point in model-
ing unrealistically high wind speeds.  A wind tunnel that  does not roughly
satisfy these requirements should not be used.
                                     160

-------
4.2  AIR VERSUS MATER
     The choice of air versus water as the fluid medium for modeling of
atmospheric flow and diffusion of pollutants will  depend on many different
factors:  the availability of the facility, economics, the type of problem to be
studied and the type of information to be obtained, to name a few.  The
kinematic viscosity of water at normal room temperature is a factor of 15 less
than that of air, so that, in principle, a factor of 15 in the Reynolds number
may be gained by modeling with water as the medium.  However, because water
is so much heavier than air, structural and pumping requirements dictate that
water facilities be much smaller and run at much lower flow speeds than wind
tunnels.  Thus, the full potential for obtaining larger Reynolds numbers using
water facilities is seldom realized.
     If it is essential to obtain very high Reynolds numbers, water has some
advantages.  Because of its incompresslbility, it may be run at high speeds
while maintaining low Mach numbers, which is not possible with air.  However,
a different problem appears with water at high speeds -- cavitation behind
obstacles.  This may be overcome by maintaining large pressures inside the
water tunnel, which then requires heavy steel construction, so that compromises
must again be made.

4.2.1  Visual Observations
     Smoke and helium filled soap bubbles (for which a generator is now
commercially available) are about the only visible tracers for use in air.   A
very much wider variety of tracer techniques is available for use in water,
making flow visualization much easier.  These include different colors and
densities of dye, hydrogen bubbles, potassium permanganate crystals,
shadowgraphs, and neutrally buoyant particles.  And because flow speeds are
generally low, it is easy to observe and photograph flow patterns in
                                    767

-------
water.  The comparable smoke in a wind tunnel  is difficult to regulate either
in concentration or specific gravity (oil  fog  smoke generators are notor-
iously cantankerous and they occasionally explode!).  Smoke is also difficult
to observe visually and photographically at flow speeds in excess of 1 to 2 m/s.
Titanium tetrachloride is relatively easy to use, but is corrosive, hazardous
to handle, and is not easily used as a stack effluent.
     The importance of flow visualization should not be underestimated.   Much
time and effort can be wasted searching for a  maximum ground level concentration
in complex terrain using a probe and some sort of analyzer, whereas visual
observations of smoke or dye would narrow the  area to be searched tremen-
dously.  Fixed rakes are frequently positioned downwind of a hill to sample the
vertical and lateral concentration profiles; but unless it is known a priori
about where the plume will be, the data collected will not be highly useful and
the experiment may have to be run again.  With flow visualization, it is obvious
at a glance, for example, whether a plume is going over or around a hill,
whereas extensive point-by-point measurements  would be required otherwise.
Flow visualization can also be of great help in the interpretation and
understanding of quantitative data.  Hot-wire  anemometry, in spite of its
increased sophistication and reliability in recent years, still cannot tell
us the direction of flow (there 1s a +_ 180° ambiguity) and reverse flows
commonly exist downwind of bluff obstacles.  Finally, some quantitative
results may also be obtained from flow visualization.  For example, Hunt and
Snyder (1980) used flow visualization to measure the displacement of stream-
lines by a hill, the surface streamline patterns, the increase of velocity
or speed-up over the top of the hill, for understanding lee waves, hydraulic
jumps, and separated flow regions downwind of the hill, and for extending
                                     162

-------
Drazin's (1961) theory to determine whether plumes  from upwind sources  would
pass over the top or impact on the surface and pass round the sides  of  the
hill in stratified flow.   Also, different colored dyes  emitted from  different
elevations on the hill surface showed oscillations  in the wake that  were
anticorrelated at different elevations; this kind of information would  have
been difficult to obtain  through other means.

4.2.2  Quantitative Measurements
     Quantitative measurements of flow speeds are more difficult in  water.
Wind tunnel techniques for these measurements have  been developed and advanced
to a level of high reliability and accuracy (Bradshaw, 1970).  For local
velocity measurements in  wind tunnels, numerous instruments are available:
pi tot tubes, hot-wire, hot-film, and pulsed-wire anemometers.  (The  pulsed-
wire anemometer is especially useful in low speed and reversing flows,  as it
is capable of detecting the direction of flow; it is also relatively insens-
itive to small fluctuations in temperature, and therefore would be useful in
stratified flows.)  Hot-film anemometers are used in water, but require much
travail to obtain reliable measurements.  At typical low flow speeds used in
water, pi tot tubes are not very useful.  Small propeller anemometers (Mem
dia.) have been developed for special studies in air and water, but are not
readily available.
     Highly accurate and reliable flame ionization  detectors are available
for quantitative measurements of pollutant concentrations downwind from a
source in a wind tunnel.   These instruments are presently the most popular
because they have a relatively fast response time  (~0.5s), their output is
linear with concentration over a very wide range (about 0.5 to 10,000 ppm
methane with proper adjustments), and they can be used with any hydrocarbon
                                     763

-------
gas, including methane, ethylene, and butane, which have specific gravities
of 0.5, 1 and 2, respectively.  Many other tracers and instruments have also
been used, including sulfur dioxide, carbon monoxide, temperature, smoke
(Motycka and Leutheusser, 1972), helium (Isyumov et a!., 1976), and radio-
active gases (Meroney, 1970), along with corresponding measuring devices.
Smoke, temperature, and helium techniques offer possibilities for the meas-
urement of concentration fluctuations, but are generally limited to the
measurement of small dilutions, i.e., 1:100, as compared with the 1:10,000
desired.  Fackrell (1978) has developed a "popper valve" to allow the flame
ionization detector to be used for the measurement of concentration fluctua-
tions.  Fackrell (1979) has also modified a flame ionization detector to
enable continuous measurements of concentration fluctuations at rates of up
to 350hz.  Another method suitable for measuring low concentration fluctua-
tions is laser/aerosol light scattering (Meroney and Yang,  1974).   When
properly designed, this instrument can detect individual particles within
the sampling volume.
     Salts in conjunction with conductivity meters, acids with pH meters,
temperature with thermistors, and dyes with colorimeters and fluorometers
have been used as tracers for quantitative measurements of  concentration in
water.  Except for temperature, these techniques offer a wide range in con-
centration detectability.  The conductivity probes and thermistors can be
quite fast-response devices.  They offer possibilities for  the measurement of
concentration fluctuations.

4.2.3  Producing Stratification
     There are two common methods for producing stratification in water.   The
most common method of producing stable density stratification in water is  by
                                     164

-------
slowly filling a tank through distribution tubes  on  the bottom with thin
layers of salt water, each layer increasing in specific gravity (Hunt et  al.
1978).  The heavier solutions flow under the lighter fluid above,  thus lifting
it.  In view of the very small  mass diffusivity of salt in water,  an undis-
turbed stable mass of salt water will  remain that way for weeks, even months,
before the density gradient is  substantially changed by molecular  diffusion.
Maximum dimensionless density differences are limited to about 20% using
common salt (NaCl).  Reclrculating systems using  this technique have been
impractical because of the mixing within the pump.  However,  Odell and
Kovasznay (1971) have designed a rotating disk pump  that maintains the
gradient; this device permits the use  of recirculating salt water  systems,  but,
thus far, has been used only for very  small channels (^lOcm depth).
     The second common method for producing stratification in water is by
heating and cooling.  Frenzen (1963) had produced both stable and  unstable
stratification in a towing tank using  this method.  Because of the large  heat
capacity of water, large amounts of energy are required for heating and cooling
to produce significant stratification, so that this  method is generally limited
to small tanks.  Deardorff and Willis  (1974) and  Liu and Lin  (1976) have  combined
heating and cooling (respectively) with stable salt  water stratification
to study inversion break-up phenomena.
     Air, with its low heat capacity,  is comparatively easy to stratify.
Provisions must be made, of course, for heating or cooling of the  floor of  the
test section and/or differential heating and/or cooling of the air entering
the test section and, if it is  a closed return tunnel, for cooling or heating
the return flow.  In order not to exceed reasonable  temperatures in the tunnel
(say 100°C), the maximum dimensionless density difference is  limited to about
35%.  The Micrometeorological Wind Tunnel at the  Colorado State University
                                      165

-------
(Cermak, 1975) has a test section 1.8m square and 27m long.   It has  heating
and cooling capabilities for maintaining the floor temperature between  1  and
200°C and the ambient air temperature between 5 and 95°C.  Calspan (Ludwig
and Skinner, 1976) used liquid nitrogen dripped onto aluminum plates upstream
of a model in their open-return wind tunnel  to produce stable stratification.
Dry ice has been used in a similar manner (Cermak et al.,  1970).   The problem
with the liquid nitrogen and dry ice is that a stable boundary layer is
created at the point of contact, but a growing mixed-layer (elevated inver-
sion) develops downstream because of the air contact with  the uncooled  tunnel
floor or model surface.

4.2.4  Examples
     Thus far in this chapter, we have discussed the comparative  advantages
of using air or water as the fluid medium for modeling studies.   There  are
no "hard and fast" rules for deciding which  type of facility is best for
a particular study.  Two example problems are given below, one of which
appears best suited for study in a wind tunnel and the other of which appears
best suited for study in a towing tank.  However, in principle and --
depending upon the ingenuity and perseverance of the investigator -- in
practice, similar information could be obtained from either  study in
either facility.
Problem 1:  We wish to determine the excess  ground level  concentrations
            caused by an insufficiently tall chimney next  to a power plant
            in essentially flat terrain.
Method of Solution:  A plume from a power plant is generally highly buoyant,
so that building downwash probably occurs only in high wind, hence neutral,
conditions.  The advantages of a wind tunnel over a water  channel here  are
                                     766

-------
obvious.  Thick, simulated neutral atmospheric boundary layers are easily
obtained in wind tunnels, whereas the development and testing of such in a
water channel would be a cumbersome task.  Measurement of the turbulent flow
structure in a water channel would be a very difficult task; accessability to
the model would be limited; instrumentation would be more expensive and less
reliable; concentration measurements would be more easily obtained in air
using hydrocarbons (probably methane in this case) and a flame ionization
detector; etc.  A large enough Reynolds number can probably be obtained in a
wind tunnel even though it is necessary to simulate the buoyant effluent.
Probably the only advantage to using a water channel in this case would be for
the ease of flow visualization, but smoke or helium-filled soap bubbles would
probably be adequate in a wind tunnel.
Problem 2:   We wish to determine the maximum ground level  concentration (glc)
            that may occur (once per year) on an isolated three-dimensional
            hill 200 m high downwind of a 100 m high stack.  Nocturnal surface-
            based inversions develop to 400 m depth with temperature gradients
            of 1.5°C/100 m and wind speeds of 2 m/s at the 200 m elevation.
Method of Solution;  The maximum glc will probably occur during the nocturnal
inversion.   The boundary layer will be below the plume and, hence, is probably
unimportant.  The most important parameter is the Froude number based on the
hill height and the density difference between the base and top of the hill:
            F = U/Nh = U/(ghA6/e1/2 = 2/(9.8x200x5/300)1/2 = 0.35.
(Notice that potential temperature instead of density has  been used in calcu-
lating the  Froude number.)  This problem is rather easily studied in a towing
tank of 1 m depth where the stratification is obtained using a continuous
gradient of salt water (s.g. = 1.0 at the top and 1.2 at the bottom, yielding
N=1.3 rad/s).  The required towing speed for a hill of height 0.2 m would be
U=FNh=9 cm/s, a reasonable towing speed for a water channel, yielding a Reynolds
number Uh/v=18,000.
                                     767

-------
     This type of flow has not yet (at least,  to the author's  knowledge)  been
obtained in a wind tunnel, but rough calculations will  easily  illustrate  the
difficulties.  The maximum temperature difference that  could be generated is on
the order of 100°C.  Let us suppose that the model hill  height is also 20cm
in the wind tunnel and that the 100°C temperature  difference  is imposed  over  a
40 cm depth, so that N=2.9.  The required tunnel speed  is thus 23 cm/s, which
is exceedingly difficult to maintain, control, and measure in  any wind tunnel,
especially when the temperature varies so drastically.   The Reynolds number
would be only 4600 (although it 1s most likely unimportant in  this case,  since
the flow will definitely not be turbulent).

4.2.5  Summary
     The ease and convenience of operating wind tunnels and associated measuring
equipment and the ability to adequately simulate the neutral atmospheric
boundary layer make the wind tunnel far superior to the water  tunnel for  small
scale studies where buoyancy is relatively unimportant.  However, the inability
of the wind tunnel to achieve adequate buoyancy or stratification and adequate
Reynolds numbers simultaneously make the towing tank indispensable for the
study of elevated plume dispersion in stably-stratified flow in complex terrain.
Somewhere in the middle, where the interest is in low-level dispersion in
mildly stratified flows, the two types of facilities have essentially equal
capabilities.
                                      168

-------
                             5.   CONCLUDING  REMARKS

      The problem with  simulating the  neutral boundary  layer  is that  the
 atmosphere  is  very  seldom  neutral.  There is "always"  an  inversion at some
 height with  surface heating  or cooling  from below.  Perhaps, occasionally, the
 atmosphere  is  truly neutral  for  a few minutes around sunrise or sunset, but such
 a  state  cannot be considered stationary because it lasts  only a few  tens of
 minutes  (Kaimal, et al., 1976) and, because the surface heat flux is changing
 so rapidly,  the  turbulence cannot track it  (Wyngaard,  1973).  Perhaps our only
 hope  is  cloudy,  high wind, conditions,  but  "cloudy" implies a temperature
 inversion (at  the base of the clouds),  so this cannot  be  truly neutral either.
      One might rightly ask at this point:   "Why bother with wind tunnel modeling?
 We can't simulate the rotational effects, and even if we  restrict ourselves to
 cases where  rotational effects are relatively unimportant, the type of flow that
 we have some chance of simulating well, the neutral  surface layer, hardly ever
 exists in the atmosphere."  Panofsky (1974)  rather summarily dismissed wind
 tunnel modeling because of our inability to simulate the turning of wind with
 height.  The answer is "fluid modeling  is heuristic."   We have the ability to
 control the flow and to independently adjust specific  parameters.   To paraphrase
 Corrsin (1961a), a wind tunnel  is, in effect,  an analog computer and, compared
with digital computers (numerical models),  it  has  the  advantages of "near-
 infinitesimal" resolution and "near-infinite"  memory.   The inability to achieve
 large Reynolds number turbulence limits the  size of  the dissipative eddies.   In
many ways, this situation is analogous to numerical  fluid-dynamic  models wherein
                                      169

-------
the small-scale turbulence is "parameterized."   Whereas  we  have  difficulty  in
simulating the large scale eddies, we are no worse  off than the  numerical
modelers and we need not make any "closure assumptions". Nor must we  deal
with an inviscid potential flow that cannot separate from any body, let alone
a sharp corner.  The point is that we need to understand the characteristics of
the flow we generate and to understand how changing those characteristics  changes
the result.  We must also recognize the limitations of our  facilities  and
interpret the results accordingly—with caution.
     There are two basic categories of fluid modeling studies:  (1)  The
"generic" study wherein idealized obstacles and terrain are used with idealized
flows in an attempt to obtain basic physical understanding  of the flow and
diffusion mechanisms, and (2) the engineering "case" study  wherein the miniature
scaled model of a specific building or hill is constructed  and a specific  decision
is to be made based upon the results of the tests,  i.e., the necessary stack
height or the siting of a plant.  Advances in the basic understanding obtained
from the generic studies will ultimately reduce the need for case studies, but
the present state-of-the-art falls far short of eliminating this need.
     There are many "doubting Thomases" concerning the applicability of fluid
modeling studies to the real atmosphere; yet those same "doubting Thomases" do
not hesitate to apply potential flow models with constant eddy diffusivities
in order to predict surface concentrations on hills under all types of stratified
flow conditions.  Frequently, they appear to be unaware that many of the under-
lying physical ideas and even many of the "constants" used directly in their
models were obtained from laboratory experiments.  A fluid modeling study, after
all, employs a real fluid, and  if a mathematical model  is to be applied to the
atmosphere, it should also be applicable to a fluid model, e.g., by eliminating
                                      770

-------
or adjusting that portion of the model dealing with rotational effects, by re-
ducing the Reynolds number, etc.  If a mathematical model cannot simulate the
results of an idealized laboratory experiment, how can it possibly be applic-
able to the atmosphere?  The point is that fluid models should be used to
bridge-the-gap between the mathematical model and its application to the field,
     A well-designed and carefully executed fluid modeling study will yield
valid and useful information - information that can be applied to real environ-
mental problems - - with just as much and generally more credibility than any
current mathematical models.
                                     777

-------
                                REFERENCES


Abe, M., 1941:  Mountain Clouds, Their Forms and Connected Air Currents,
Pt. II, Bull. Cent. Meteorol.  Obs.,  Tokyo, v.  7, no.  3.

American Meteorological  Society, 1979:  Fourth Symposium on Turbulence,
Diffusion and Air Pollution, Preprints voluire, Jan.  15-18, Reno,  NV,  676p.

Armitt, J., 1968:  The Effect of Surface Roughness  and Free Stream Turbulence
on the Flow around a Model  Cooling Tower at Critical  Reynolds  Numbers,  Proc.
Symp.  on Wind Effects on Bldgs.  and  Structures, Loughborough Univ.  of Tech.,
England, April.

Armitt, J.  and Counihan, J., 1968:   The Simulation  of the Atmospheric Boundary
Layer in a Wind Tunnel,  Atmos.  Envir., v.  2, p. 49-71.

Arya,  S.P.S., 1975:  Buoyancy Effects  in a Horizontal  Flat-Plate  Boundary
Layer, J. Fluid Mech., v. 68,  pt.  2, p. 321-43.

Arya,  S.P.S., 1977:  Suggested Revisions to Certain  Boundary Layer Param-
eterization Schemes Used in Atmospheric Circulation  Models, Mon.  Wea. Review,
v. 105, no. 2, p. 215-26.

Arya,  S.P.S. and Plate,  E.J.,  1969:   Modeling  of the Stably Stratified
Atmospheric Boundary Layer, J.  Atmos.  Sci., v. 26,  no. 4, p. 656-65.

Barrett, R.V., 1973:  Use of the Wind Tunnel to Investigate the Influence of
Topographical Features on Pollution  from a Tall Stack, Chimney Design Symp.,
Univ.  of Edinburgh, April.

Batchelor,  G.K., 1949:  Diffusion  in a Field of Homogeneous Turbulence,
pt. I, Eulerian Analysis, Aust.  J. Sci. Res.,  ser.  A,  v.  2, p.  437.

Batchelor,  G.K., 1953a:   The Theory  of Homogeneous  Turbulence,  Cambridge Univ.
Press, London, 197p.

Batchelor,  G.K.. 1953b:   The Conditions for Dynamical  Similarity  of Motions of
a Frictionless Perfect-Gas Atmosphere, Quart.  J. Roy.  Meteorol.  Soc., v. 79,
p. 229-35.

Binkowski,  F.S., 1979:  A Simple Semi-Empirical Theory for Turbulence in the
Atmospheric Surface Layer,  Atmos.  Envir.,  v. 13, no.  2,  p. 247-53.

Birkhoff, G., 1950:  Hydrodynamics,  Princeton  Univ.  Press, Princeton, NJ.

Blackadar,  A.K.  and Tennekes,  H.,  1968:  Asymptotic Similarity in Neutral
Barotropic Planetary Boundary Layers,  J. Atmos. Sci.,  v.  25, p.  1015-20.

                                     772

-------
Bradbury, L.J.S. and Castro, I.P.,  1971:   A Pulsed-wire  Technique  for  Velocity
Measurements in Highly Turbulent Flows.  J. Fluid Mech.,  v.  49,  pt.  4,  p.  657-91.

Bradshaw, P., 1970:   Experimental  Fluid  Mechanics,  2nd.  ed.,  Pergamon  Press,
NY, NY, 219p.

Briggs, G.A., 1969:   Plume Rise, Critical  Review Series,  U.S. Atomic Energy
Comm., TID-25075, Nat. Tech. Infor.  Sen/., Springfield,  VA, 81p.

Briggs, G.A., 1972:   Discussion  of Chimney Plumes In  Neutral  and Stable Sur-
roundings, Atmos. Envir., v. 6,  p.  507-10.

Briggs, G.A., 1975:   Plume Rise  Predictions,  ATDL No.  75/15,  Atmos. Turb. and
Diff. Lab., NOAA Envir. Res. Lab.,  Oak Ridge, TN, 53p.

Briggs, G.A. and Snyder, W.H.,  1980:  Laboratory Measurements of Plume
Rise in Calm, Stable Surroundings,  Envir.  Prot.  Agcy.  Rpt.  (to  be  published).

Brighton, P.W.M., 1978:  Strongly Stratified  Flow Past Three-Dimensional
Obstacles, Quart. J. Roy. Meteorol.  Soc.,  v.  104, p.  289-307.

Brest, R.A. and Wyngaard, J.C.,  1978:  A Model  Study  of  the Stably-Stratified
Planetary Boundary Layer, J. Atmos.  Sci.,  v.  35, p. 1427-40.

Brown, M. and Michael, P., 1974:  Measured Effect of  Shear  on Plume Dispersion,
Presented at AMS Symp. on  Atmos.  Diff.  and Air Poll., Santa  Barbara,  CA,
Sept. 9-13, p. 246-50.

Busch, N.E., 1973:  On the Mechanics  of  Atmospheric Turbulence, Workshop  on
Micrometeorology.  Am. Meteorol. Soc., Boston,  MA,  p.  67-100.

Businger, J.A., 1973:   Turbulent Transfer  in  the Atmospheric  Surface Layer,
Workshop on Micrometeorology, Am.  Meteorol. Soc., Boston, MA, p. 67-100.

Businger, J.A. and Arya, S.P.S., 1974:   Height of the  Mixed Layer  in the
Stably Stratified Planetary Boundary  Layer, Adv. in Geophys., Academic Press,
NY, NY, v. ISA, p. 73-92.

Businger, J.A., Wyngaard, J.C.,  Izumi, Y.  and Bradley, E.F.,  1971:  Flux-
Profile Relationships  in the Atmospheric Surface Layer,  J.  Atmos.  Sci.,
v. 28, no. 2, p. 181-9.

Castro, I.P., Jackson, N.A. and  Robins,  A.G., 1975:   The  Structure  and
Development of a 2 m Simulated  Suburban  Boundary Layer,  Cent. Elect. Gen. Bd.
Research Rpt. R/M/N800, Marchwood Engrg. Lab.,  Southampton, England.

Castro, I.P. and Robins, A.G.,  1975:  The  Effect of a  Thick Incident
Boundary Layer on the  Flow Around a Small  Surface Mounted Cube., Cent.
Elect. Gen. Bd. Research Rpt. R/M/N795,  Marchwood Engrg.  Lab.,  Southampton,
England, 38p.

Caughey, S.J., Wyngaard, J.C. and Kaimal,  J.C.,  1979:  Turbulence  in the  Evolv-
ing Stable Boundary Layer, J. Atmos.  Sci., v. 36, no.  6,  p. 1041-52.
                                     773

-------
Cermak, J.E., 1958:   Wind Tunnel  for the Study of Turbulence  in  the
Atmospheric Surface  Layer, Dept.  of Civil  Engr.,  Colo.  State  Univ. Rpt. no.
CER58JEC42, Ft.  Collins, CO,  40p.

Cermak, J.E., 1971:   Laboratory Simulation of the Atmospheric Boundary Layer,
AIAA J., v. 9, p.  1746-54.

Cermak, J.E. 1975:   Applications  of Fluid Mechanics  to  Wind Engineering—
A Freeman Scholar Lecture, J.  Fluids Engr., v.  97, ser.  1, no. 1, p. 9-38.

Cermak, J.E. and Arya,  S.P.S., 1970:  Problems  of Atmospheric Shear Flows
and Their Laboratory Simulation,  Bound.-Layer Meteorol.,  v. 1, p. 40-60.

Cermak, J.E. and Koloseus, H.J.,  1954:   Lake Hefner  Model Studies of Wind
Structure and Evaporation, Tech.  Rpt.,  Contract No.  N0bsr57053,  Pt. I, Nov.
1953, CER54JEC20 and Pt. II,  July,  1954, CER54JEC22, AD-105-493, Colo. State
Univ., Ft.  Collins,  CO.

Cermak, J.E., Orgill, M.M. and Grant,  L.O., 1970:  Laboratory Simulation of
Atmospheric Motion and  Transport  over  Complex Topography  as Related to Cloud
Seeding Operations,  Second Nat.  Ccnf.  on Wea.  Mod.,  Apr.  6-9,  Santa Barbara,
CA.

Cermak, J.E. and Peterka, J.,  1966:   Simulation of Wind Fields over Point
Arguello, Calif, by  Wind Tunnel  Flow over a Topographic Model, Rpt. to Navy
under Contract N123(61756)34361A(PMR),  Colo.  State Univ., Ft.  Collins, CO.

Cermak, J.E., Sandborn,  V./1.,  Plate, E.J., Binder, G.H.,  Chuang, H., Meroney,
R.N. and Ito, S.,  1966:   Simulation  of Atmospheric Motion by  Wind Tunnel Flows,
Fluid Dyn.  and Diff. Lab. Rpt. No.  CER66JEC-VAS-EJP-GJB-HC-RNM-S117, Colo.
State Univ., Ft. Collins, CO, May,  102p.

Chaudhry, F.H. and Cermak, J.E.,  1971:   Simulation of Flow and Diffusion over
an Urban Complex,  Paper Pres.  at  Conf.  on Air Poll.  Meteorol., Raleigh, NC,
Apr. 5-9, p. 126-31.

Cook, N.J., 1973:  On Simulating  the Lower Third  of  the Urban Adiabatic Bound-
ary Layer in a Wind  Tunnel, Atmos.  Envir., v.  7,  no. 7, p. 691-706.

Cook, N.J., 1978:  Wind Tunnel Simulation of the  Adiabatic Atmospheric Boundary
Layer by Roughness,  Barrier and Mixing-Device Methods,  J. Indus. Aero., v. 3,
p. 157-76.

Corrsin, S., 1953:   Remarks on Turbulent Heat Transfer, Proc.  Iowa Thermodyn.
Symp., State Univ. of Iowa, April  27-28, p. 5-28.

Corrsin, S., 1961a:   Turbulent Flow, Am. Scientist,  v.  49, no. 3, p. 300-25.

Corrsin, S., 1961b:   Theories of Turbulent Dispersion,  Mechanique de la
Turbulence, Marseille,  France, p.  27-52.

Corrsin, S., 1963:   Turbulence:  Experimental  Methods, Handbuch Der Physik,
Encyclopedia of Physics, edited by S.  Flugge/Freiburg,  v. VIII/2, p. 524-90.
                                     174

-------
Corrsin, S., 1964:  Further Generalization of Onsager's Cascade Model  for
Turbulent Spectra, Phys. Fluids, v.  7, no. 8, p.  1156-9.

Counihan, J., 1969:  An Improved Method of Simulating an Atmospheric Boundary
Layer in a Wind Tunnel, Atmos. Envir., v. 3, p.  197-214.

Counihan, J., 1975:  Adiabatic Atmospheric Boundary Layer:   A Review and
Analysis of Data from the Period 1880-1972, Atmos.  Envir.,  v. 9, no. 10,
p. 871-905.

Counihan, J., Hunt, J.C.R. and Jackson, P.S., 1974:  Wakes  Behind Two-
Dimensional Surface Obstacles in Turbulent Boundary Layers, J.  Fluid
Mech., v. 64, pt. 3, p. 529-63.

Csanady, G.T., 1969:  Diffusion in ar Ekman Layer,  J.  Atmos.  Sci.,  v.  26,  no
3, p. 414-26.

Davenport, A.G., 1963:   The Relationship of Wind  Structure  to Wind Loading,
Paper 2, Proc. Conf. on Wind Effects on Bldgs.  and  Structures,  Nat.  Phys.
Lab., June, HMSO, London, 1965, p. 54-102.

Davenport, A.G. and Isyumov, N., 1968:  The Application of  the  Boundary
Layer Wind Tunnel to the Prediction  of Wind Loading, Proc.  Int.  Res. Sem.
Wind Effects on Buildings and Structures, Ottawa, Can., v.  1, p.  201-30.

Dean, M., 1977:  Gas Pollution Discharge from Boiler Stacks (A  Wind Tunnel
Study), Dept. Arch. Sci., Univ. of Sydney, Australia,  135p.

Deardorff, J.W., 1972:   Numerical Investigation of  Neutral  and  Unstable
Planetary Boundary Layers, J. Atmos. Sci., v. 29, p. 91-115.

Deardorff, J.W., 1974:   Three-dimensional Numerical Study of the Height and
Mean Structure of a Heated Planetary Boundary Layer, Bound.-Layer Meteorol.,
v. 7, p. 81-106.

Deardorff, J.W. and Willis, G.E., 1974:  Computer and  Laboratory Modeling
of the Vertical Diffusion of Nonbuoyant Particles in the Mixed  Layer,  Adv.
Geophys., v. 18B, p. 187-200.

DeMarrais, G.A., 1959:   Wind-Speed Profiles at Brookhaven National  Laboratory,
J. Meteorol., v. 16, p. 181-190.

Drazin, P.G., 1961:  On the Steady Flow of a Fluid  of Variable  Density Past
an Obstacle, Tellus, v. XIII, no. 2, p. 239-51.

Duchene-Marullaz, P., 1976:  Full-Scale Measurements of Atmospheric Turbulence
in a Suburban Area, Proc.  4th Int. Conf.  on Wind  Effects  on Bldgs.  and Struc-
tures, Cambridge Univ.  Press, Cambridge,  England, p. 23-31.

Ellison, T.H., 1957:  Turbulent Transport of Heat and  Momentum  from an
Infinite Rough Plane, J. Fluid Mech., v.  2, p.  456-66.
                                     175

-------
Fackrell, J.E., 1978:  Measuring Turbulent Concentration Fluctuations  with
the Popper Valve, Cent. Elect. Gen.  Bd.  Rpt.  no.  R/M/N971,  Marchwood Engrg.
Lab., Southampton, England, 37p.

Fackrell, J.E. 1979:  A Flame lonisation Detector for Continuous  Monitoring
of Fluctuating Concentration, Cent.  Elect. Gen.  Bd.  Rpt. No.  R/M/N1056,  March-
wood Engrg. Lab., Southampton, England,  28p.

Facy, L., 1971:  Dispersion and Diffusion of  Effluents Emitted  by the
Conemaugh and Seward Power Plants in the Vicinity of Johnstown, PA,
Institute for Meteorol. Research and Dev., Magny-Les-Hameaux, France (orig.
in French - trans, for EPA - TR1654  - unpublished).

Frenzen, P., 1963:  A Laboratory Investigation of the Lagrangian  Autocor-
relation Function in a Stratified Fluid, Argonne  Nat. Lab.  Rpt. ANL-6794,
Argonne, IL, 168p.

Fuquay, J.J., Simpson, C.L. and Hinds, W.T.,  1964:   Prediction  of Envi-
ronmental Exposures from Sources near the Ground  Based on Hanford Experi-
mental Data, J. Appl. Meteorol., v.3, 761-70.

Godowitch, J.M., Ching, J.K.S. and Clark, J.F.,  1979:  Dissipation of  the
Nocturnal Inversion Layer at an Urban and Rural Site in St. Louis, MO» 4th
Symp. on Turb., Diff. and Air Poll., Jan.  15-18, Reno, NV, Am. Meterol. Soc.,
Boston, MA, p. 416-20.

Golden, J., 1961:  Scale Model Techniques, M.S. Thesis, College of Engr.,
Mew York Univ., MY, NY, 48p.

Colder, D.G., 1972:  Relations among Stability Parameters in  the  Surface
Layer, Bound.-Layer Meteorol., v. 3, no. 1, p.  47-58.

Goldstein, S., 1965:  Modern Developments in  Fluid Dynamics.  Dover,  NY,  NY,
702p.

Halitsky, J., 1968:  Gas Diffusion Near  Buildings, Meteorol.  and  Atomic
Energy, (ed. D.H. Slade), TID-24190, U.S. Atomic  Energy Comm.,  Sect.
5-5, p. 221-55.

Halitsky, J., 1979:  Wind Tunnel Modeling of  Heated  Stack Effluents  in
Disturbed Flows, Atmos. Envir., v. 13, no. 4,  p.  449-52.

Halitsky, J., Golden, J., Halpern, P. and Wu,  P., 1963: Wind Tunnel Tests
of Gas Diffusion from a Leak in the  Shell of  a Nuclear Power  Reactor and
from a Nearby Stack, Geophys. Sci. Lab.  Rpt.  63-2, Dept. of Meteorol.  and
Ocean., New York Univ., NY, NY.

Hanna, S.R., 1969:  The Thickness of the Planetary Boundary Layer, Atmos.
Envir., v. 3, no. 5, p. 519-36.

Hansen, A.C., Peterka, J.A. and Cermak,  J.E.,  1975:   Wind-Tunnel  Measurements
in the Wake of a Simple Structure in a Simulated  Atmospheric  Flow,  NASA
CR-2540, 45p.

                                     176

-------
Harris, P..I., 1968:  Measurement of Wind Structure at Heights up to 598 feet
above Ground Level, Symp. Wind Effects on Buildings and Structures, Dept.  of
Transport Technology, Loughborough Univ., England.

Haugen, D.A., Kaimal, J.C. and Bradley, E.F., 1971:  An Experimental  Study of
Reynolds Stress and Heat Flux in the Atmospheric Surface Layer,  Quart.  J.  Roy.
Meteorol. Soc., v. 97, p. 168-80.

Hidy, G.M., 1967:  Adventures in Atmospheric Simulation, Bull. Am.  Meteorol.
Soc., v. 48, no. 3, p. 143-61.

Hino, M., 1968:  Maximum Ground-Level Concentration and Sampling Time,  Atmos.
Envir., v. 2, p. 149-65.

Hinze, J.O., 1975:  Turbulence. 2nd ed., McGraw Hill, NY,  NY, 790p.


Hogstrom, A.S. and Hogstrom, U., 1978:  A Practical Method for Determining
Wind Frequency Distributions for the First 200 m from Routine Meteorological
Data, J. Appl. Meteorol., v. 17, p. 942-54.


Hogstrom, U., 1964:  An Experimental Study on Atmospheric  Diffusion,  Tellus,
v. 16, p. 205-51.

Homma, M., 1969:  Main Research Activities on Air Pollution,  Central  Res.
Inst. of Elec. Power Industry, Japan, unpublished.

Hoot, T.G., Meroney, R.N. and Peterka, J.A., 1973:  Wind Tunnel  Tests of
Negatively Buoyant Plumes, Envir.  Prot. Agcy. Rpt. No.  EPA-650/3-74-003,
Res. Tri. Pk., NC, 94p.

Hoult, D.P., 1973:  Simulation of Buoyant Pollutants in the Atmospheric
Boundary Layer, Flow Studies in Air and Water Pollution, Am.  Soc. Mech.
Engrg., NY, NY, p. 61-6.

Hoult, D.P., O'Dea, S.R., Touchtan, G.L. and Ketterer,  R.J.,  1977:   Turbulent
Plume in a Turbulent Cross Flow:  Comparison of Wind Tunnel Tests with  Field
Observations, 0. Air Poll. Cont. Assoc., v. 27, no. 1,  p.  56-60.

Hoult, D.P. and Weil, J.C., 1972:   Turbulent Plume in a Laminar  Cross Flow,
Atmos. Envir., v. 6, no. 8, p. 513-32.

Howroyd, G.C. and Slawson, P.R., 1975:  The Characteristics of a Laboratory
Produced Turbulent Ekman Layer, Bound.-Layer Meteorol., v.  8, p.  201-19.

Howroyd, G.C. and Slawson, P.R., 1979:  The Effects of Cross-Wind Shear on
Plumes Released from Elevated Sources, Proc. 4th Symp.  on  Turb.,  Diff.  and
Air Poll., Reno, NV, p. 56-61, Am. Meteorol. Soc., Boston,  MA.

Huber, A.H., Snyder, W.H., Thompson, R.S., and Lawson,  R.E. Jr.,  1976:
Stack Placement in the Lee of a Mountain Rtdge:  A Wind Tunnel Study, Envir.
Prot. Agcy. Rept. No. EPA-600/4-76-047, Res. Tri.  Pk.,  NC., 45p.

                                     777

-------
Huber, A.H., Snyder, W.H., Thompson, R.S.  and Lawson, R.E.  Jr.,  1980:   The
Effects of a Squat Building on Short Stack Effluents, Envir.  Prot.  Agcy.  Rpt.
(in press), Res. Tri. Pk., NC.

Hunt, J.C.R., 1974:  Wakes behind Buildings, Paper pres.  to Atmos.  Envir.
Comrr. of Aero. Res. Council, London, Oct.  23, ARC Rept.  35601, Atmos.  229,
28p.

Hunt, J.C.R., 1975:  Fundamental  Studies of Wind Flow Near  Buildings,  Models
and Systems in Arch, and Bldg., LUBFS Conf.  Proc. no. 2,  Construction  Press
Ltd., Lancaster, England.

Hunt, J.C.R. and Fernholz, H., 1975:  Wind-Tunnel Simulation  of  the Atmo-
spheric Boundary Layer:   A Report on Euromech 50, J.  Fluid  Mech.,  v. 70,
pt. 3, p. 543-59.
Stratified Flow over a Model Three-Dimensional  Hill,  J.  Fluid Mech., v. 96,
pt. 4, p. 671-704.

Hunt, J.C.R., Snyder, W.H., and Lawson, R.E. Jr., 1978:   Flow Structure and
Turbulent Diffusion around a Three-Dimensional  Hill:   Fluid Modeling Study on
Effects of Stratification:  Part  I:   Flow Structure,  Envir. Prot. Agcy. Rpt.
No. EPA 600/4-78-041, Res. Tri. Pk., NC, 84p.

Irwin, J.S., 1978:  Dispersion Estimate Suggestion No. 6, Unpublished  Note,
Envir. Appl. Branch, Meteorol. and Assess. Div., Envir.  Prot. Agcy., Res.
Tri. Pk, NC.

Irwin, J.S., 1979:  A Theoretical Variation  of the Wind  Profile  Power-Law
Exponent as a Function of Surface Roughness  and Stability,  Atmos. Envir.,  v.
13, no. 1, p. 191-4.

Isyumov, N., Jandali, T. and Davenport, A.G., 1976:   Model  Studies  and the
Prediction of Full Scale Levels of Stack Gas Concentration, J. Air  Poll.
Control Assoc., v. 26, no. 10, p. 954-64.

Isyumov, N. and Tanaka,  H., 1979:  Wind Tunnel  Modeling  of  Stack Gas Disper-
sion - Difficulties and Approximations, Preprints, Fifth  Int. Conf.  on Wind
Engr., July 8-14, Colo.  State Univ., Ft. Collins, CO, v.  II,  p.  VIII-3-1 to 15.

Izumi, Y., 1971:  Kansas 1968 Field Program  Data Report,  AFCRL-72-0041, Air
Force Cambridge Research Lab. (LYB), LG Hanscom Field, Bedford,  MA.

Jensen, M., 1958:  The Model-Law  for Phenomena in a Natural Wind,  Ingenioren,
Int. Ed., v. 2, no. 4.

Jensen, M. and Franck, N., 1963:   Model-Scale Tests in Turbulent Wind,
Pt. I:  Phenomena Dependent on the Wind Speed,  Danish Technical  Press,
Copenhagen, 97p.

Kaimal, J.C., 1973:  Turbulence Spectra, Length Scales and  Structure Para-
meters in the Stable Surface Layer, Boundary Layer Meteorol,  v.  4,  p.  289-
309.

                                    178

-------
Kaimal, J.C., 1978:   Horizontal  Velocity Spectra in an Unstable Surface  Layer,
J. Atmos. Sci., v.  35, p. 18-24.

Kaimal, J.C., Wyngaard, O.C., Haugen,  D.A.,  Cote, O.R., Izumi,  Y.,  Caughey,
S.J. and Readings,  C.J., 1976:  Turbulence Structure in the Convective
Boundary Layer, J.  Atmos. Sci.,  v.  33, no. 11,  p. 2152-69.

Kaimal, J.C., Wyngaard, J.C., Izumi,  Y.  and  Cote, O.R., 1972:   Spectral
Characteristics of Surface Layer Turbulence, Quart. J.  Roy. Meteorol.  Soc.,  v.
98, no. 417, p. 563-89.

Kalinske, A.A., Jensen, R.A.  and Schadt, C.F.,  1945:  Wind  Tunnel Studies  of
Gas Diffusion in a Typical Japanese District and Correlation of Wind Tunnel
Studies with Field Measurements  of  Gas Diffusion, OSRD, NDRC, Div.  10  Informal
Rpt. No. 10-3A-48- and 48a (available  from Lib.  of Congress).

Kampe de Feriet, J., 1939:  Les  Fonctions Aleatoires Stationnaires  et  la
Theorie Statistique de la Turbulence  Homogene,  Ann. Soc.  Sci. Brux., v.
59, p. 145.

Kistler, A.L. and Vrebalovich, T.,  1966:  Grid  Turbulence at Large  Reynolds
Numbers, J. Fluid Mech., v.  26,  pt. 1, p. 37-47.

Kolmogoroff, A.M.,  1941:  The Local Structure of Turbulence in  Incompres-
sible Viscous Fluid for very Large  Reynolds  Number, Compt.  Rend.  (Doklady),
Acad. Sci. U.S.S.R., v. 30,  p. 301-5.

Lin, J.T., Liu, H.T., Pao, Y.H., Lilly,  O.K., Israeli, M. and Orszag,  S.A.,
1974:  Laboratory and Numerical  Simulation of Plume Dispersion  in Stably
Stratified Flow over Complex Terrain,  Envir. Prot.  Agcy.  Rpt. No. EPA-650/4-
74-044, Res. Tri. PK., NC, 70p.

Liu, H.T. and Lin,  J.T., 1975:  Laboratory Simulation of Plume  Dispersion
from a Lead Smelter in Glover, Missouri, in  Neutral and Stable  Atmosphere,
Envir. Prot. Agcy.  Rpt. No.  EPA-450/3-75-066, Res.  Tri. Pk., NC,  63p.

Liu, H.T., and Lin, J.T., 1976:   Plume Dispersion in Stably Stratified Flows
over Complex Terrain:  Phase 2,  Envir. Prot. Agcy.  Rpt. No. EPA-600/4-76-022,
Res. Tri. Pk., NC,  66p.

Long, R.R. 1972:  Finite Amplitude  Disturbances  in the Flow of  Inviscid
Rotating and Stratified Fluids over Obstacles,  Ann. Rev.  Fluid  Mech.,
v. 4, p. 19-92.

Ludwig, G.R., 1974:   Atmospheric Simulation  Facility:  Technical  Supplement,
Calspan Corp., Buffalo, NY,  21p.

Ludwig, G.R. and Skinner, G.T.,  1976:   Wind  Tunnel  Modeling Study of the
Dispersion of Sulfur Dioxide in  Southern Allegheny County,  PA,  Envir.  Prot.
Agcy. Rpt. No. EPA 903/9-75-019, Res.  Tri. Pk.,  NC, 128p.

Ludwig, G.R. and Sundaram, T.R., 1969:  On the  Laboratory Simulation of
Small-Scale Atmospheric Turbulence, Cornell  Aero. Lab.  Rpt. No.  VC-2740-S-
1, Buffalo, NY, 63p.
                                     179

-------
Ludwig, G.R., Sundaram, T.R.  and Skinner,  G.T.,  1971:   Laboratory Modeling
of the Atmospheric Surface Layer with Emphasis on Diffusion,  Cornell  Aero.
Lab. Rpt. No. VC-2740-S-2, Buffalo,  NY,  59p.

Lumley, J.L., 1970:   Stochastic Tools in Turbulence. Academic Press,  NY,NY.

Lumley, J.L. and Panofosky, H.A., 1964:  The  Structure of Atmospheric
Turbulence. Interscience, NY, NY, 239p.

Mandel, J., 1964:  The Statistical Analysis of Experimental  Data, Interscience,
NY, NY.

Maul, P.R., 1978:  The Effect of the Turning  of  the Wind  Vector with  Height  on
the Ground Level Trajectory of a Diffusing Cloud, Atmos.  Envir, v.  12,  no. 5,
p. 1045-50.

McVehil, G.E., Ludwig, G.R. and Sundaram,  T.R.,  1967:   On the Feasibility
of Modeling Small Scale Atmospheric  Motions,  Cornell Aero.  Lab. Rpt.  ZB-
2328, Buffalo, NY, 178p.

Melbourne, W.H., 1968:  Wind Tunnel  Modeling  of  Buoyant Chimney Plumes, Inst.
of Engr. Aust., Proc. 3rd Australasian Conf.  on  Hyd. and  Fluid Mech., Sydney,
Paper No. 2631.

Meroney, R.N., 1970:   Gaseous Plume  Diffusion about  Isolated  Structures of
Simple Geometry, Rpt. No. CER69-70RNM29  (also AEC Rpt.  No.  COO-2053-1),
Colo. State Univ., Ft. Collins, CO.

Meroney, R.N., 1978:   Studying the Convective Heat Transfer from a Building
Model with Infrared Camera Techniques, ASME Winter Annual Mtg., Dec.  10-15,
San Francisco, CA, 28p.  (also CEP77-78RNM61, Colo. State Univ., Ft.  Collins,
CO).

Meroney, R.N., 1978:   Liquefied Natural  Gas:  Wind Tunnel Simulation  and
Instrumentation Assessments, Contract Rpt. No. SAN/W1364-01  to Dept.  of
Energy, Wash., DC.

Meroney, R.N., Cermak, J.E. and Yang, B.T., 1975: Modeling of Atmospheric
Transport and Fumigation at Shorelines,  Bound.-Layer Meteorol., v.  9, no. 1,
p. 69-90.

Mery, P., 1969:  Reproduction en Similitude de la Diffusion dans la Couche
Limite Atmospherique, la Houille Blanche,  no. 4, p.  327-43.

Mery, P., Schon, J.P., and Solal, J., 1974:  Comparison of Thermally  Neutral
and Unstable Shear Flows in the Wind Tunnel and  the Atmosphere, Adv.  Geophys.,
v. 18B, p. 273-87.

Miller, K.S., 1963:   Engineering Mathematics. Dover, NY,  NY,  417p.

Monin, A.S. and Obukhov, A.M., 1954:  Basic Laws of  Turbulent Mixing  in
the Ground Layer of the Atmosphere,  Geophys.  Inst. Acad.  Sci., U.S.S.R.,
v. 24, p. 163-87.

                                     750

-------
Motycka, J., and Leutheusser, H.J., 1972: Concentration Meter for Wind
Tunnel Studies of Gaseous Dispersion, Atmos. Envir., v. 6, no.  12, p.  911-6.

Nagib, H.M., Morkovin, M.V., Yung, J.T. and Tan-Atichat, J., 1974:  On
Modeling of Atmospheric Surface Layers by the Counter-Jet Technique, AIAA
Paper No. 74-638, AIAA 8th Aerodynamic Testing Conf., Bethesda, MD, Jul. 8-10,
9p.

Nakai, M. and Shikata, H., 1977:  Experimental Investigation on Non-buoyant
Exhaust Gas Rise, Rpt. No. 276048, Central Res. Inst. for Elec. Power Indus.,
Japan.

Nee, V.W., Szewczyk, A.A., Yang, K.T., Cheung, F.B. and Liu, V.K., 1974:
The Simulation of the Atmospheric Surface Layer in the Notre Dame Atmospheric
Wind Tunnel, Paper 74-213 pres. at 67th Ann. Mtg.  of APCA, Denver, June 9-13,
25p.

Nemoto, S., 1968:  Similarity between Natural Local Wind in the Atmosphere
and Model Wind in a Wind Tunnel, Papers in Meteorol. and Geophys., v.  19,
no. 2, p. 131-230.

Nickerson, E.C. and Smiley, V.E., 1975:  Surface Layer and Energy Budget
Parameter!'zations for Mesoscale Models, J. Appl. Meteorol., v.  14, p.  297-300.

Odell, G.M. and Kovasznay, L.S.G., 1971:  A New Type of Water Channel  with
Density Stratification, J. Fluid Mech., v. 50, pt. 3, p. 535-43.

Ogawa, Y., 1976:  A New Wind Tunnel for the Simulation of Atmospheric
Diffusion, Paper 76-17.07 pres. at 69th Ann. Mtg.  of the APCA,  June 27-July
1, Portland, OR.

Ogawa, Y., Diosey, P.G., Uehara, K. and Ueda, H.,  1981:  A Wind Tunnel  for
Studying the Effects of Thermal Stratification in  the Atmosphere, To appear
in Atmos. Envir.

Ogawa, Y., Griffiths, R.A. and Hoydysh, W.G., 1975:  A Wind Tunnel Study of Sea
Breeze Effects, Bound.-Layer Meteorol., v. 8, p. 141-61.

Overcamp, T.J. and Hoult, D.P., 1971:   Precipitation in the Wake of Cooling
Towers, Atmos. Envir., v. 5, no. 9, p. 751-66.

Panofsky, H.A., 1974:  The Atmospheric Boundary Layer Below 150 Meters,
Annual Review of Fluid Mech., v. 6, p~; 147-78.

Panofsky, H.A., Blackadar, A.K. and McVehil, G.E., 1960:  The Diabatic
Wind Profile, Quart. J. Roy. Meteorol. Soc., v. 86, p.  390-8.

Panofsky, H.A., Tennekes, H., Lenschow, D.H. and Wyngaard, J.C.,  1977:   The
Characteristics of Turbulent Velocity Components in the Surface Layer  under
Convective Conditions, Bound.-Layer Meteorol., v.  11, p. 355-61.

Pao, Y.H., 1965:  Structure of Turbulent Velocity  and Scalar Fields at
Large Wavenumbers, Phys. Fluids, v. 8, no. 6, p. 1063-75.

                                     181

-------
Pasquill, F., 1962:  Some Observed Properties of Medium-Scale Diffusion  in
the Atmosphere, Quart. J. Roy.  Meteorol.  Soc., v. 88,  p.  70.

Pasquill, F., 1969:  The Influence of the Turning of Wind with Height on
Crosswind Diffusion, Phil.  Trans.  Roy.  Soc.,  v.  265A,  p.  173-81.

Pasquill, F., 1971:  Wind Structure in  the Atmospheric Boundary Layer, Phil.
Trans. Roy.  Soc, v. 269A, p.  439-56.

Paulson, C.A., 1970:  The Mathematical  Representation  of  Wind Speed and
Temperature Profiles in the Unstable  Surface  Layer,  J.  Appl.  Meteorol.,  v.  9,
no. 6, p. 857-61.

Petersen, R. and Cermak, J.E.,  1979:   Wind-Tunnel Investigation of Plume Dis-
persion and Transport over Complex Terrain for Col strip Power Plant—Stable
Stratification, Fluid Dynamics  and Diffusion  Lab. Rpt.  No.  CER78-79RLP-JEC39,
Colo. State Univ., Ft. Collins, CO.

Plate, E.J.  and Cermak, J.E., 1963:  Micrometeorological  Wind Tunnel  Facility-
Description and Characteristics, Fluid  Dynamics  and  Diffusion Lab.  Rpt.  CER
63EJP-JEC9,  Colo.  State Univ.,  Ft. Collins, CO,  60p.

Rey, C., 1977: Effets du Nombre de Prandtl, de la Gravite et  de la  Rugosite
Sur les Spectres de Turbulence  Cinematique et Scalaire, Ph.D.  Dissertation,
TUniversite Claude Bernard,  Lyon, France, 327p.

Rey, C., Schon, J.P. and Mathieu,  J., 1979:  Buoyancy  Effects in  a  Wind
Tunnel Simulation of the Atmospheric  Boundary Layer, Phys.  Fluids,  v.  22,
no. 6, p. 1020-8.

Richardson,  L.F.,  1920:  The  Supply of  Energy from and to Atmospheric
Eddies, Proc. Roy. Soc. London, v. A97, p. 354-73.

Ricou, F.P.  and Spalding, D.B., 1961:  Measurements  of Entrainment  of A.xisym-
metrical Turbulent Jets, 0. Fluid Mech.,  v. 11,  p. 21-32.

Robins, A.G., 1975:  Wind Tunnel Modeling of  Plume Dispersal, Cent.  Elect.
Gen. Bd. Rpt. No.  236, Marchwood Engrg. Lab., Southampton,  England.

Robins, A.G., 1978:  Plume Dispersion from Ground Level Sources in  Simulated
Atmospheric Boundary Layers,  Atmos. Envir., v. 12, no.  5, p.  1033-44.

Saffman, P.G., 1962:  The Effect of Wind  Shear on Horizontal  Spread from an
Instantaneous Ground Source,  Quart. J.  Roy. Meteorol.  Soc., v. 88,  no. 378,
p. 382-93.

Sandborn, V.A. and Marshall,  R.D., 1965:   Local  Isotropy  in Wind  Tunnel
Turbulence,  Rpt. No. CER-65VAS-RDM 71,  Colo.  State Univ., Ft. Collins, CO,
25p.

Sato, H., Onda, Y. and Saito, T.,  1974:  Laboratory  Simulation of Atmospheric
Turbulence:   Generation of Arbitrary  Velocity Distributions and Model  Exper-
iment on Flow around Mt. Fuji,  Adv. Geophys., v.  18B,  p.  241-51.
                                     182

-------
Schlichting, H.t 1968:  Boundary-Layer Theory, McGraw-Hill, NY, NY, 744p.

Schon, J.P. and Mery, P., 1971: A Preliminary Study of the Simulation of
Neutral Atmospheric Boundary Layer Using Air Injection in a Wind Tunnel,
Atmos. Envir., v. 5, no. 5, p. 299-312.

Schon, J.P. and Mery, P., 1978:  Physical Simulation of Atmospheric Phe-
nomena in the Planetary Boundary Layer - Wind Tunnel Experiments, Pres.  at
Erice, Italy.

Schon, J.P., Rey, C., Mathieu, J. and Mery, P., 1974:  Experimental Study
of Unstable Stratified Boundary Layers:  Comparison with the Structure of
the Atmospheric Surface Layer, Paper 74-212, Proc.  67th Annual  Meeting of
the APCA, Denver, CO. June 9-13, 23p.

Scorer, R.S., 1968:  Air Pollution, Pergamon Press., NY,NY, 151p.

Sherlock, R.H. and Stalker, E.A., 1940:  The Control of Gases in the Wake
of Smoke Stacks, Mech. Engrg., v. 62, p. 455-8.

Simiu, E. and Scanlan, R.H., 1978:   Wind Effects on Structures, John Wiley
and Sons, NY, NY, 458p.

Skinner, G.T. and Ludwig, G.R., 1978:  Physical Modeling of Dispersion in
the Atmospheric Boundary Layer, Calspan Rpt. No. 201, Calspan Corp., Buffalo,
NY, 31p.

Smith, E.G., 1951:   The Feasibility of Using Models for Predetermining
Natural Ventilation, Res. Rep. No.  26, Tex. Engrg.  Exp.  Stn., Texas A &  M
College, College Station, TX.

Smith, F.B., 1965:   The Role of Wind Shear in Horizontal Diffusion of Ambient
Particles, Quart. J. Roy. Meteorol. Soc., v. 91, no. 389, p.  318-29.

Smith, M.E. and Singer, I.A.,  1955:  Sampling Periods in Air Pollution
Evaluations, Proc.  3rd Nat. Air Poll. Symp., April  18-20, Pasadena, CA,  p.  80-5.

Snyder, W.H., 1979:  The EPA Meteorological Wind Tunnel: Its Design, Const-
ruction, and Operating Characteristics, Envir. Prot. Agcy.  Rpt. No.
EPA-600/4-79-051, Res. Tri. Pk., NC, 78p.

Snyder, W.H., Britter, R.E. and Hunt, J.C.R., 1979:  A Fluid Modeling Study
of the Flow Structure and Plume Impingement on a Three-Dimensional Hill  in
Stably Stratified Flow, Proc.  Fifth Int. Conf. on Wind Engr., Pergamon Press,
NY, NY, v. 1, p. 319-29.

Snyder, W.H. and Lawson, R.E.  Jr.,  1976:  Determination  of a Necessary Height
for a Stack Close to a Building - A Wind Tunnel Study, Atmos. Envir., v. 10,
no. 9, p. 683-91.

Snyder, W.H. and Lumley, J.L., 1971:  Some Measurements  of Particle Velocity
Autocorrelation Functions in a Turbulent Flow, J. Fluid  Mech.,  v. 48, p. 41-
71.
                                     183

-------
Stewart, R.W., 1972:   Turbulence,  Illustrated  Experiments  1n  Fluid Mechanics,
Natl.  Comm.  for Fluid Mech.  Films,  MIT Press,  Cambridge, MA,  p.  82-8.

Strom, G.H., Hackman, M.  and Kaplin,  E.J.,  1957:   Atmospheric Dispersal  of
Industrial Stack Gases Determined  by  Concentration Measurements  in Scale Model
Wind Tunnel  Experiments,  J.  Air Poll.  Cont.  Assoc., v.  7,  no.  3, p.  198-204.

Strom, G.H.  and Halitsky, J., 1953:   Topographical Influences on Dispersal
of Stack Gases, Combustion,  v.  24,  no. 12,  p.  40-43.

Sutton, O.G., 1949:   Atmospheric Turbulence, Methuen,  London, 11Ip.

Tan-atichat, J. and  Nagib, H.N., 1974:  Wind Tunnel Simulation of Neutral
Atmospheric Surface  Layers by the  Counter-Jet  Technique, Paper 74-214  pres. at
67th Ann. Mtg. Air Poll.  Cont.  Assoc.., Denver, CO, June 9-13, 21p.

Taylor, G.I., 1921:   Diffusion  by  Continuous Movements, Proc.  London Math.
Soc., v. 20, p. 196-212.

Templin, R.J., 1969:   Interim Progress Note on Simulation  of  Earth's Surface
Winds by Artificially Thickened Wind  Tunnel  Boundary Layers,  Lab. Tech.  Rpt.
LA-22, Nat.  Res. Coun. of Canada,  Ottawa.

Tennekes, H., 1973:   Similarity Laws  and Scale Relations in Planetary  Bound-
ary Layers, Workshop in Micrometeorology,  Am.  Meteorol. Soc., Boston,  MA,
p. 177-216.

Tennekes, H. and Lumley,  J.L.,  1972:   A First  Course in Turbulence.  MIT
Press, Cambridge, MA, 300p.

Teunissen, H.W., 1975:  Simulation of the Planetary Boundary  Layer  in  a  Mul-
tiple Jet Wind Tunnel, Atmos. Envir., v. 9, no. 2, p.  145-75.

Thompson, R.S., 1978:  Note on the Aerodynamic Roughness Length  for Complex
Terrain, J. Appl. Meteorol., v. 17, no. 9, p.  1402-3.

Touma, J.S., 1977:  Dependence of  the Wind Profile Power Law  on  Stability
for Various Locations, J. Air Poll. Cont. Assoc., v. 27, no.  9,  p.  863-6.

Townsend, A.A., 1956:  The Structure of Turbulent Shear Flow. Cambridge
Univ. Press, Cambridge,~EFgland, 315p.

Turner, D.B.,  1964:   A Diffusion Model for an  Urban Area,  J.  Appl.  Meteorol.,
v. 3, p. 83-91.

Tyldesley, J.B. and Wellington, C.E., 1965:  The Effect of Wind  Shear and
Vertical Diffusion on Horizontal Dispersion, Quart. J. Roy. Meteorol.  Soc.,
v. 91, p. 158-74.

Ukeguchi, N.,  Sakata, H., Okamoto, H. and Ide, Y., 1967:   Study on  Stack
Gas Diffusion, Mitsubishi Tech. bul.  no. 52, p. 1-13.
                                     184

-------
Van der Hoven, I., 1957:  Power Spectrum of Horizontal Wind Speed in the
Frequency Range from 0.0007 to 900 Cycles Per Hour, J. Meteorol., v. 14,
p. 160-4.

Wamser, C. and Muller, H., 1977:  On the Spectral Scale of Wind Fluctuations
within and above the Surface Layer, Quart. J. Roy. Meteorol. Soc., v. 103,
p. 721-30.

Wyngaard, J.C., 1967:  An Experimental Investigation of the Small-Scale
Structure of Turbulence in a Curved Mixing Layer, Ph.D. Dissertation, Dept.
Mech. Engrg., Penn. State Univ., Univ. Pk., PA, 105p.

Wyngaard, J.C., 1973:  On Surface-Layer Turbulence, Workshop on Micrometeor-
ology, Am. Meteorol. Soc., Boston, MA, p. 101-49.

Wyngaard, J.C., 1975:  Modeling the Planetary Boundary Layer - Extension to
the Stable Case, Bound.-Layer Meteorol., v. 9, p. 441-60.

Wyngaard, J.C., Arya, S.P.S. and Cote, O.R., 1974:  Some Aspects of the
Structure of Convective Planetary Boundary Layers, J. Atmos. Sci., v. 31,
p. 747-54.

Wyngaard, J.C. and Cote, O.R., 1972:  Cospectral Similarity in the Atmospheric
Surface Layer, Quart. J. Roy. Meteorol. Soc., v. 98, no. 417, p. 590-603.

Wyngaard, J.C., Cote, O.R. and Izumi, Y., 1971:  Local Free Convection,
Siirilarity, and the Budgets of Shear Stress and Heat Flux, J. Atmos. Sci.,
v. 28, no. 7, p. 1171-82.

Yang, B.T. and Meroney, R.N., 1974:  A Portable Laser Light Scattering Probe
for Concentration Measurements, Rev. Sci. Insts., v. 35, no. 2, p. 210-5.

Zoric, D.L., 1968:  Approach of Turbulent Boundary Layer to Similarity, Ph.D.
Dissertation, Colo. State Univ., Ft. Collins, CO, 203p.

Zoric, D.L. and Sandborn, V.A., 1972:  Similarity of Large Reynolds Number
Boundary Layers, Bound.-Layer Meteorol., v. 2, p. 326-33.
                                     185             » US GOVERNMENT PRINTING OFFICE 1M1 757-064/0315

-------