United States
Environmental Protection
Agency
Environmental Sciences Research
Laboratory
Research Triangle Park NC 27711
EPA-600 4-79-023
April 1979
Research and Development
A Lagrangian
Approach to
Modeling Air
Pollutant Dispersion

Development and
Testing in the
Vicinity of a  Roadway

-------
                RESEARCH REPORTING SERIES

Research reports of the Off ice of Research and Development, U.S. Environmental
Protection Agency, have been grouped into nine series. These nine broad cate-
gories were established to facilitate further development and application of en-
vironmental technology.  Elimination  of traditional grouping was  consciously
planned to foster technology transfer and a maximum interface in related fields.
The nine series are:

      1.   Environmental  Health Effects Research
     2.   Environmental  Protection Technology
     3.   Ecological Research
     4.   Environmental  Monitoring
     5.   Socioeconomic Environmental Studies
     6.   Scientific and Technical  Assessment Reports (STAR)
     7.   Interagency  Energy-Environment Research and Development
     8.   "Special" Reports
     9.   Miscellaneous Reports

This report has been assigned to the ENVIRONMENTAL MONITORING series.
This series describes research conducted to develop new or improved methods
and  instrumentation  for the identification and quantification of environmental
pollutants at the lowest conceivably significant concentrations, tt also includes
studies to determine the ambient concentrations of pollutants in the environment
and/or the variance of pollutants as a function of time or meteorological factors.
This document is available to the public through the National Technical Informa-
tion Service, Springfield, Virginia  22161.

-------
                                              EPA-600/4-79-023
                                              April 1979
      A LAGRANGIAN APPROACH TO MODELING
          AIR POLLUTANT DISPERSION  .
           Development and Testing
        in the Vicinity of a Roadway
                     by

     R.  G.  Lamb, H.  Hogo, and L. E. Reid
         Systems Applications, Inc.
             950 Northgate Drive
        San Rafael,  California  94903
           Contract No.  68-02-2733
               Project Officer

               R.  E.  Eskridge
     Meteorology and Assessment Division
 Environmental  Sciences Research Laboratory
Research Triangle Park, North Carolina  27711
 ENVIRONMENTAL SCIENCES RESEARCH LABORATORY
     OFFICE OF RESEARCH AND DEVELOPMENT
    U.S. ENVIRONMENTAL PROTECTION AGENCY
RESEARCH TRIANGLE PARK, NORTH CAROLINA  27711

-------
                             DISCLAIMER


     This  report has been reviewed by the Environmental Science Research
Laboratory,  U.S. Environmental Protection Agency, and approved for publi-
cation.  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.

-------
                               ABSTMCT
     We have developed and demonstrated a microscale  roadway dispersion
model based on Lagrangian diffusion theory.   The  model  incorporates
similarity expressions for the mean wind and turbulence energy  in the
atmospheric boundary layer, through which the effects of  wind shear and
atmospheric stability are taken into account; a parameterization of vehicle
wake turbulence; and provisions for predicting the  turbulence quantities
required to simulate second-order chemical reactions.   Through  simple
modifications, the model can be structured to treat particle settling,
deposition, and resuspension, as well  as buoyancy of  the  effluent material.
Calm winds, winds parallel to the roadway, flows  around depressed or elevated
roadways, shallow mixed layers, and transient or  spatially  variable meteor-
ological conditions can all be explicitly taken into  account within the frame-
work of our modeling approach.  The model requires  minimum  computer core
storage and fairly small execution times.  It is  not  based  on finite dif-
ference equations, and therefore, it is free of the truncation  errors and
computational stability and convergence problems  that models based on those
methods encounter.

     We demonstrated the model by applying it to  52 of  the  30-minute experi-
mental periods reported in the GM sulfate study.  Of  the  1040 predicted
values of the mean concentration of an inert material (SFg), we found half
of them to be within ±30 percent of the measured  values.  The overall
correlation coefficient was 0.91.  The monitoring sites treated ranged in
distance from 2 to 100 meters from the edge of the  roadway. The computer
time (but not core storage) required by our model is  directly  proportional
to the distance between the farthest receptor and the road.  For the  studies
reported here,  the model  required on the average  20 seconds of CPU  time
on the CDC  7600 to simulate each of the  30-minute GM experiments.
                                    m

-------
     One finding from  our  tests  of  the model  is  that  vehicle  wake turbu-
lence is important  in  dispersing material  near the  roadway, especially
under stable atmospheric conditions.  The  great  improvement in  the accuracy
of the predicted concentrations  achieved by adding  just  a  simple para-
meterization of the wake indicates  that concentration levels  close to  the
road are relatively insensitive  to  the intensity of ambient turbulence.

     Further refinement and  testing of this model is  planned  to improve
the wake turbulence parameterization, to evaluate in  more  detail  the
accuracy of the method used  to simulate ambient  turbulence effects under
combined light wind and very unstable situations, to  incorporate the non-
linear chemical reaction simulation capability,  to  make  the simple modifica-
tions necessary to  simulate  fugitive dust  dispersion, and  to  optimize  the
computer code.
                                    IV

-------
                                CONTENTS

Abstract	  i i i
Illustrations	,	   vi
Tables	 viii
   I.  Introduction	    1
  II.  Formulation of the Model	    7
       A.  General Formulation 	    7
       B.  Determining the Function J and Its Time Integral 	   13
       C.  Application to Photochemical  Pollutants 	   18
 III.  Model Testing 	   21
  IV.  Meteorological  Inputs to  the Model 	   27
   V.  Results of Preliminary Model Validation Experiments 	   52
       A.  Simulations That Neglect Wake Effects 	   52
       B.  Simulations That Include a Simple Parameterization
           of Wake Effects 	   55
  VI.  Summary 	   68
Appendices
       A.  A Scheme for Simulating Particle Pair Motions
           in Turbulent Fluid 	   69
       B.  User's Guide to the Lagrangian Highway Model 	   91
References 	  104

-------
                               ILLUSTRATIONS

Number                                                                     page

   1    Schematic Illustration of the Relationship Between the Lagrangian
       and  Eulerian Approaches to Turbulent Diffusion 	   4

   2    Coordinate System Used in the Roadway Model To Compute Concentra-
       tions at an Arbitrary Receptor Site	   8

   3    Schematic Illustration of Some of the Parameters Required and
       the  Method Used TO Calculate p	15
                                   	V
   4    Example of the Dependence of J^ on the Number of Realizations K. .  .  17
                                        _           p
   5    Simple Particle Trajectories with u  = 3.0, u1  =1, and
       w7?  = 0.6  (MKS Units)	22

   6    The  Root-Mean-Square Particle Dispersion Component o  as a
       Function of Travel Time for Two Values of the Integral
       Time Scale T	23

   7    Comparison of Roadway Model Predicted Cross-Wind Integrated
       Concentration with Gaussian Plume Formula Predictions at a
       Distance of 100 m Downwind of a Point Source at 50 m Elevation  ...  24

   8    Comparison of Roadway Model Predicted Cross-Uind Integrated
       Concentrations with Gaussian Plume Formula Predictions at a
       Distance of 100 m Downwind of a Point Source at 5 m Elevation.  ...  25

   9    Coordinate System Used To Express Mean Wind and Turbulence
       Velocities	29

  10    Observed Versus Predicted Mean Wind Speed  	  36
                                          —x-1/2     -VI / 2
  11    Observed Versus Predicted Values of u'^    and wl<:    	39
                                               -^1/2          —?l/2
  12    Observed Versus Predicted Values of u' = u1^    and w1 = w1^    ...  42

  13    Observed Versus Predicted Values of u' and w'	44

  14    Plots of the  Ratio of the Wind Speed at 10.5 and 1.5
       Meters and  the Drag Coefficient u*/u(10.5) Given by
       Similarity Theory as Functions of 1/L	47

-------
Number                                                                 Page

  15   Comparison of the u* and 1/L Values of Table 1  with Those
       Derived Using the WSR Method	49

  16   Comparison of Observed SF5 Concentrations with Those
       Simulated by the Roadway Model  with Vehicle Wake Turbulence
       Neglected:  All Experiments Included 	   54

  17   Comparison of Observed SF5 Concentrations with Those
       Simulated by the Roadway Model  with Vehicle Wake  Turbulence
       Neglected:  Data Points for Day 295 (H < 1  m/sec,  Ri  > 0.25)
       Omitted	56

  18   Comparison of Observed SFs Concentrations with Those
       Predicted by the Roadway Model  with Wake Turbulence Effects
       Parameterized:  All Experiments (1100 Data Points) Included  .  .   59

  19   Comparison of Observed SF5 Concentrations with Those
       Predicted by the Roadway Model  with Wake Turbulence Effects
       Parameterized:  Results Displayed for Stably Stratified
       Flows Only (580 Data Points)	61

  20   Comparison of Observed SFs Concentrations with Those
       Predicted by the Roadway Model  with Wake Turbulence Effects
       Parameterized:  Results Displayed for Unstably Stratified
       Flows Only (520 Data Points)	62

  21   Comparison of Observed Peak SFg Concentrations for Each  of
       the 55 GM Experiments with the Values Predicted by the
       Model (Wake Effects Included)  	   64

  22   Distribution of Fractional Error 	   67
                             	    2 1/2
 A-l   Stceamwise Component U2 - £0)x   of the rms Separation  of
       Simulated Particle Motions Under the Conditions u - 3.0,
       u7"2 = 1,  and w7? = 0.6 for Two  Values of the Parameter aQ  ...   89
                                     VII

-------
                                   TABLES
Number                                                                    Page
   1   Estimated Values of u* and_L for the GM Experiment Periods
      (zn - 5 cm) and the Angle 6  of the Mean Wind Relative
      to the Road	   33
   2   Model Performance  Statistics 	   66
 B-l   Format for  Card 1   	   93
 B-2   Format for  Cards 3 and 4	   94
 B-3   Format for  the Monitor Site  Cards	   96
 B-4   Format for  the First Meteorological Data Card	   96
 B-5   Format for  the Second Meteorological Data Card	   98
 B-6   Format for  the Source Location Card	   98
 B-7   Format for  the Miscellaneous Data Cards	   99
 B-8   Sample Input  Deck  Used in the GM Simulations	100
                                      vm

-------
                           I   INTRODUCTION


     The General  Motors (GM)  Sulfate Dispersion Experiment  conducted  in
late 1975 and reported by Cadle et al.  in 1976 revived  interest  in  roadway
diffusion modeling.   The study not only raised a controversy  about  the
adequacy of existing roadway models, but also collected a unique set  of
turbulence and concentration data in a  roadway environment  under well-
monitored conditions.  The work described in this report is in part an
indirect outgrowth of the GM study:   Although the basic theory of the
model described here was developed several  years ago,  it was  implemented
and tested under contract to EPA as  part of the roadway model development
activity at EPA that the GM study encouraged.

     Prior to 1976,  the most widely used model for performing environmental
impact assessments of roadways was HIWAY, a model based on  the Gaussian
plume formula (Zimmerman and Thompson,  1975).  Comparisons  made  by  Chock
(1977) of the predictions of HIWAY and  concentration values measured  during
the GM study revealed that HIWAY seriously overpredicted concentrations,
especially during stable, light wind conditions.  The generally  poor  per-
formance of HIWAY was not surprising,  considering the host  of simplifying
assumptions inherent in the Gaussian plume formula on which it is based.
In addition to neglect of streamwise diffusion, an assumption that  is
invalid during light winds, other simplifying assumptions  include neglect
of vehicle wake turbulence, dismissal  of wind shear effects,  and use  of
interpolation to obtain values for the  dispersion parameters  a   and GZ
from some arbitrarily chosen initial value at the source and  the values
measured approximately 100 meters downwind.  Furthermore,  the plume
formula is not applicable under nonsteady or spatially  inhomogeneous
situations, it cannot describe dispersion that occurs around  depressed
or elevated roads, and it is limited to the treatment of inert pollutants.

-------
     Seeking to develop roadway models  free of the restrictions  inherent
in the Gaussian plume formula,  many investigators have turned  to the  dif-
fusion equation.   Among these are Danard (1972),  Ragland and Pierce  (1973),
Egan, Epstein,  and Keefe (1973), and Eskridge and Demerjian  (1977).
However, the diffusion equation is not  a rigorously sound basis  for multi-
point or line source models.   In deriving the K-theory diffusion equation
from basic principles, one must assume  that the smallest length  scale of
the spatial variation in the  mean concentration distribution is  much
larger than the Lagrangian integral length scale of the turbulence (see,
for example, Lamb, 1973).

     The effects of this assumption and the limitation it imposes can be
seen clearly in the special case of stationary, homogeneous  turbulence.
In this case, the eddy diffusivity K is a constant and the diffusion
equation can be solved analytically.  Its solution for a point source has
                                 2
the standard Gaussian form with a  = 2Kt, where t = x/u, x is  distance
downwind from the source,  and u is the  mean (uniform)  wind speed. This
expression is consistent with Taylor's  (1921) classical  result for diffu-
sion at large distances from a source,  but it is at odds with  the observed,
theoretically derivable relationship  o  = u^ t2 that applies when  t is
small.  Here, u"^ is the variance of the turbulent velocity  fluctuations.
Thus, the assumption of smooth concentration fields invoked  in the deriva-
tion of the diffusion equation implicitly filters out  the turbulence
processes that are important during the initial period of diffusion  from
                                                 2    2
a point source and that are responsible for the o  ^ t  relationship  cited
above.  By assuming that the diffusivity K is a function of  travel time  t,
viz., K = 1/2 u1^ t, one can make the diffusion equation compatible with
the observed behavior of point source dispersion.  But this  modification
can be used only if a single point or line source is present.  Otherwise,
no unique value of K(t) or K(x) exists.

     Teske and Lewellen (1978) have developed a roadway model  using a
second-order turbulence closure technique.  It is not clear  to what  extent
this approach circumvents the problem just cited.   Two  disadvantages of

-------
diffusion models based on second-order turbulence closure is  that they are
complicated and costly to operate.   Also, they suffer from several problems
that any model  based on finite difference methods incurs.  One of these is
that when the winds are very light  or are blowing parallel  to the road, the
pseudo-diffusion associated with many finite difference  forms of the  dif-
fusion equation can spread material over much larger areas  than ambient
and vehicle wake turbulence together are capable  of doing.   Finite dif-
ference schemes that minimize pseudo-diffusion do so at  increased costs
in computer time and storage requirements.  A second problem  with finite
difference models is that the number of grid points required, and hence
the computer time needed, increases with the square of the  width of the
region treated.  This occurs because the top of the model tr.ust be at
least several times the maximum value of o  for the area of interest,
and as we  pointed  out above, a  ^ x  initially.  As a result,  it  is
difficult to achieve the fine spatial resolution  needed  around the
roadway without making the core and machine time  requirement  of the model
excessive.  Also, it is awkward to  apply finite difference  methods to
areas having complex shapes, such as those of elevated and  depressed
roadways.  Finally, one must always bear in mind  that the finite dif-
ference equation is a model of the  differential equation it represents
and, unless care is taken, under some conditions  the finite difference
equation might possess solutions that are radically different from any of
those of the diffusion equation.

     All of the problems discussed  above can be circumvented by casting
the roadway model  in Lagrangian form.  Figure 1,  which is taken from Lamb
(1978a), illustrates the way in which this approach is related to the
Gaussian plume and  K-theory models  discussed  above.  The theoretical
aspects of the model are developed  in Chapter  II.   We discuss  below sev-
eral important features of the Lagrangian approach that make it ideally
suited to  simulating diffusion  and even  some chemical reactions  over
short distances  (up to several hundred meters) from point or line sources.

-------
FIGURE 1.   SCHEMATIC ILLUSTRATION OF THE RELATIONSHIP BETWEEN THE
           LAGRANGIAN AND EULERIAN APPROACHES TO TURBULENT DIFFUSION

-------
     First, Lagrangian models require no grid network;  as  few or as  many
receptor points as desired can be used,  and these can be arbitrarily dis-
tributed in any configuration over the area of interest.   The computer
core needs are therefore minimal.  Also, the absence of a  grid network
and of finite differencing schemes makes the computational  task of modeling
dispersion around elevated or depressed roadways (or even  that of modeling
transport and diffusion in complex terrain) relatively  simple.

     Secondly, the Lagrangian approach is essentially free of the restrict-
ing assumptions (mentioned earlier) that hinder the plume  and K-theory
models in certain instances.  The roadway model we develop here can
explicitly account for the following effects:  wind shear; particle  settling,
deposition, and resuspension; buoyancy of the effluent; calm winds or
flows parallel to the roadway; vehicle wake turbulence; atmospheric
stability; and time or space variability in the meteorological or source
emissions conditions.  The temporal  evolution of the dispersion from the
 2                         2
3  ^ t1" regime to that of o  ^ t described above is also properly duplicated.

     Thirdly, when used in conjunction with the closure approximation devel-
oped by Lamb and Shu  (1978),  the Lagrangian roadway model  can treat pollu-
tant species such as NO, NOp, 0~, and others, taking into  account the
effects on chemical reaction rates of turbulence-induced concentration
fluctuations.

     Finally, this particular modeling approach utilizes readily measurable
Eulerian turbulence statistics, primarily the variances of the velocity
fluctuations, rather than Lagrangian statistics like the a parameter used
in puff and plume models.  In this report, we show that available empirical-
theoretical expressions for the vertical profiles of mean wind and the
required  velocity variances are accurate enough for obtaining good concen-
tration predictions through this modeling technique using only the fric-
tion velocity  u*, the Monin-Obukov length L, and an estimate of the
Lagrangian  integral time  scale of the turbulence as input parameters.

-------
     The remainder of this report discusses the formulation of the model
(Chapter II), testing of the model  through comparisons of its prediction
with those of another model  (Chapter II), the meteorological  inputs to
the model, and the results of our preliminary model  validation experiments.
Appendix A describes a scheme for simulating particle pair motions in
turbulent fluid that is the heart of the roadway model.  Appendix B
describes the input variables required by the model.

-------
                   II    FORMULATION OF  THE MODEL


     This chapter discusses  the  theoretical formulation of the roadway
model and its application to photochemical  pollutants.

A.    GENERAL  FORMULATION

     For simplicity,  the  roadway model  is cast  in  a two-dimensional form.
The process by which  this is achieved  is  illustrated  in Figure 2.  For a
roadway link  of sufficient length (to  be established  later), oriented as
shown in Figure 2, the mean  concentration at  the receptor shown in the
figure is given by:
                  t  »
   = (  f  p(x,0,z,t|xs,y',zs,tl)S(y',tl) dy' dt'
(1)
where z  is the effective height  of  the  vehicle emissions; x  denotes the
location of the road in  the chosen frame (only a single lane of traffic
is considered here); S is the function describing  the distribution of
vehicle emissions; and p(r,t|rs,t')  is the  probability density that a
particle released at r1  at time t1 will  be  found at the point r at time t.
In Lamb (1978a), Eq. (1) is derived  from basic Lagrangian principles.  Note
that Eq. (1) is the generalized form of  the solution of the inhomogeneous,
ensemble-averaged mass continuity equation  of an inert substance, and p is
the Green's function of that equation.   For traffic of uniform speed v$,
we have:

                                  N
                      s(y'.t') = £ Qn6(y' -y0n - vnf)    ,         (2)

 where  Qn  is the emissions  rate (mass/time) of the n-th vehiclo;  N is the
 total  number of vehicles operating on the  link;  yQ  is a random  variable
 representing the position  at  t = 0 of the  n-th vehicle;  vp  =  vg  if the

-------
n-th vehicle is moving toward +y, vn = -v  otherwise;  and 6  is  the
delta function.
                                             RECEPTOR
                                             COORDINATES =  (x,0,z)
             H = mean wind  vector

             u = component  of V,, nonnal  to roadway

             v = component  of VH parallel  to  roadway

             0 = coordinate origin; y axis parallel to road
                 with receptor lying  in  x-z plane
      FIGURE 2.   COORDINATE SYSTEM USED IN THE ROADWAY MODEL TO COMPUTE
                  CONCENTRATIONS AT AN ARBITRARY RECEPTOR SITE
       Owing to  the  symmetry  of  the  roadway about the x-z plane, we can
  assume that the component of diffusion  parallel to the roadway is negli-
  gible, in  which case,


  p(x,0,z,t|x ,y',z ,t') = p(x,z,t|xs,zs,t')6[y'  +  (t  - t')v]    ,       (3)

-------
where v Is the component of the mean wind parallel  to the  road (see  Figure
2).  Substituting Eqs.  (2) and (3)  into Eq.  (1)  and evaluating the y'  inte-
gral , we obtain
 - Z Qny   P(x,z,t|xs,zs,t'h[t'(v  - vn)  -  tv  - y0n] df

                                                                      (4)

       Let a = (v - vn)t'  = Vnt'.   Then  Eq.  (4)  can be written in the  form:
N      _V_t
                                -  vt  -
      ,o    'I -    •**»„/"
                       r n
 = S  Qn /    p(x,z,t|xs,zs,^-
               n=l     *                    ^

where
                                       -^n)    ,      1f  0 0, or if
                                                      0 > vt + ynn > V t
                                                              ^un    n
                                                      and Vn < 0;
                    0    ,    otherwise.                              (5b)
 To arrive  at Eq.  (5a),  we made use of the following well-known property of
 the delta  function:

             /*b
            /  f(x)6(x  - xQ)dx =  f(x0)    ,    a < XQ <  b
           *a
       It is perhaps instructive to comment  briefly  on  the  physical meaning
  of Eq. (5).  The delta function in this  equation is the effective source
  strength function.  Keeping in mind that 6(5)  is nonzero  only when ? = 0
  and that yn  is a random sequence describing  vehicle  locations at t = 0,

-------
we see that the source function S in our roadway model is a random sequence
of instantaneous point sources distributed in time in a manner determined by
the combination of vehicle and wind speed components v  and v.  For example,
if v = 0, i.e., if the winds are either calm or blowing perpendicular to
the road, then the delta function in Eq. (5) is nonzero only when
t'vn + yQn - 0.  By definition of yQ , this occurs only at the instant
t1 when the n-th vehicle passes through the x-z plane of the receptor.
This obvious result occurs because, when v = 0 and the roadway is symmetric
about the receptor plane, only those pollutants released in that plane can,
in effect, influence that receptor.  As a second example, suppose that all
vehicles are stopped, v  = 0, but v i- 0.  In this case, the delta function
in Eq. (5) is nonzero, for a given t, when (t - tl)v = -yfi .  In other
words, emissions released at time t' from the n-th vehicle affect the
receptor at the tiir
the receptor plane.
receptor at the time t = t1  - (yn /v)  when those emissions  pass  through
      Consider the special case where V  = v - v  =0, which represents the
 situation where the speed and direction of the n-th vehicle are identical
 to those of the wind component parallel to the roadway.   In this case, the
  equation is best obtained from Eq. (4) rather than from the transformed
 Eq. (5).  We then have:
                 N                   J:
 <(c(x,0,z,t))= £ Qn6(v t  +  y   )  /   p(x,z,t|x ,zs,t') dtl    .      (6)
                n=l                 JQ

 This is simply the superposition of N two-dimensional plumes (recall that  we
 are neglecting diffusion  in the y-direction), each located at the  y-position
 of its  source at time  t,  i.e., at y = y~  + vnt.  The n-th plume passes
 through the receptor plane y  = 0 at time t = ~yQn/vn-  Tnis relationship
 reveals the obvious result  that, if the vehicles are northbound, i.e.,
 v  (= v) > 0, then only  those whose initial locations are south of the
 receptor plane, i.e.,  yn < 0, will affect concentrations at the monitoring
 site.
       In most problems of  interest, it is the time-averaged concentration
  near  the  roadway that is  of concern.  The averaging time T should be long

                                     10

-------
compared with  the  time  scales of mixing and transport from the road to the
farthest monitor,  but short, compared with  the  smallest  time  scale  of tem-
poral  changes in the vehicle flow rate and meteorology;  otherwise  the  time-
averaged concentration would depend on the length of the averaging interval
A time average is  denoted by an  overbar (  ) and defined  as:

                                      t+(T/2)
                                                                        (7)
on t is a consequence of the dependence of c  on  vehicle
flow rate and meteorology, which are functions of the time of day,  t.   From
here on, we drop the reference to t in 
-------
where n  is the flow rate (vehicles/time)  of vehicles of type m,  all  of
which have the same emission rate Q ,  speed v .  and location (x  ,z  )  in
                                   m         rn                 sm  sm
the receptor plane.  For example, automobiles in one lane might represent
the group designated by m = 1, trucks  in that lane might constitute m = 2,
autos in another lane might be represented by m = 3, and so on.  [To obtain
Eq. (9), note that the number N of vehicles that pass through the receptor
plane in a period T is N =X)fLi  ii T.]

Equation (9) is identical to that of the mean concentration produced by M
steady, continuous line sources of strengths n Q (|v - v |)~ .  The depen-
dence of the effective line source strength on the wind speed component v
is noteworthy because it has not been  taken into account in previous line
source roadway models.  The error produced by this omission is small  in
cases of light winds where |v |  » |v| ; but on congested roadways or in
situations where there is a strong component of the wind parallel to the
road, errors of 20 percent or larger can result in the  estimates if v
is neglected in the source strength calculation.

     Looking back over the steps that  led to Eq. (5), we find that this
equation becomes invalid if a significant fraction of the emissions that
affect the receptor arise from portions of the roadway that are not parallel
to the y-axis.  A condition on the length of roadway that must be straight
in the vicinity of the receptor to render Eq. (5) accurate can be derived as
follows.  Let
                                                Kc
                                                                     (.0,
where  (XD,ZD)  denotes the receptor coordinates,
         K   K

                              uc = ^(7+ax)    ,                  Ola)


                              wc=^(z+oz)    ,                  (lib)

and  (x,z) is the centroid of an instantaneous cloud of emissions and o  and
                                                                      /\
o  are  (roughly) the cloud dimensions.  [See, for example, Monin and Yaglom
                                    12

-------
(1971) for similarity theory expressions for x, z, o ,  and o .]  Under these
                                                    A       £-
definitions, T.  is a measure of the time required for material  released at
(x .  z) to reach the receptor (XD, ZD)  through the combined actions of
  S   S                          K   K
transport and diffusion.  Thus, the roadway in the vicinity of the receptor
must be straight for a distance L  that satisfies the condition

                             Ls » vTL    ,                          (12)

because otherwise significant quantities of material could reach the receptor
from points whose projection on the receptor plane are  different from the
assumed point (x , z ).  Subject to this constraint, Eq.  (8) can be used as
the basis of a roadway model in general  situations, and Eq. (9) can be used
in cases in which the vehicle and wind speed components are such that the
line source approximation is valid [see the discussion  preceding Eq. (9)].

B.   DETERMINING THE FUNCTION J AND ITS TIME INTEGRAL

     The kernel  p that enters in the definition of J [see Eq.  (5b)] is the
probability density that a material particle released at  the point Ug.O
at a  specified time, say t1, will be at the point (x,z) at the  later time  t.
In situations in which the meteorology is quasi-steady, as we have already
assumed, the turbulence can be considered to be stationary, in  which case
the kernel  p is  a function only of the travel  time T =  t  - t1  and not of t
and t1 jointly.   In this case, by virtue of the definition of p, the quantity
P(X,Z,T|X ,z ,0) dx dz is equal to the fraction of all  realizations of the
turbulence in which a particle released at (x ,z ) is located after a travel
time  T in the box of dimensions (dx.dz)  centered at (x,z).   That is,
           ry 7  iy  7  n\ - lim    1     V V(x,z.T|x ,z )     ,     (13a)
           ,X,Z,T x ,z ,u; -        .     ^  K        s  s
                   S  S      N->-«> N ax aZ i._i
where
                            ,  if a particle released at (xs,zs)
                               is in the box (dx,dz)  centered at
              I      ,    t      (x,z) after a travel  time T on
      *kvx,z,T|xs,zsJ  = <^      realization k;

                            ,  otherwise.

                                     13

-------
Combining Eqs. (13a) and (5b) and integrating with respect to T,  we
obtain:

       T                                  ]/
/                             lim    1    A
         J(x,z,T|x ,z ) dt = K™ i/ Hy H7 2^ At,(x,z|x ,z ;T)   ,    (14)
                  S  S       IV'33 i\ UA Ui. . _-,   K.      b  b
      u
where At,(  x.z! x ,z ;T) is the total time spent during the period T by a
        K
particle i
at (x,z).
particle released at (x ,z ) on realization k in the box (dx,dz) centered
     Equation (14) provides an operational procedure for implementing the
roadway model numerically:  If the random turbulence velocity fields could
be simulated by a Monte Carlo type of process, then particles could be
released at any desired locations (x ,z ) in the simulated flows, their
motions could be tracked, and Eq. (14) could be evaluated computationally.
The resulting values could then be used in either Eq. (8) or Eq. (9) to
compute mean concentrations at any desired locations.

     We can illustrate the method in more detail with the aid of Figure 3.
For the given roadway site and meteorological conditions, we specify the
profiles of the mean wind u"(z)--note  that only the component u normal  to
the roadway affects the kernel p, the profiles of the variances u'2(z) and
—P
w  (z) of the ambient turbulent velocity components in the receptor plane,
and the profile of the Lagrangian integral time scale r(z).  The functional
forms used for each of these profiles under various meteorological conditions
are listed later in Chapter IV.  To account for the effects of vehicle wake
turbulence, it is also necessary to specify the speed and drag coefficient
of the vehicles and certain other parameters descriptive of their wakes.  At
present, there are no accurate theoretical or empirical descriptions of the
wake turbulence generated by automobiles moving along a roadway.  Consequently,
only approximations of the wake effects can be incorporated into the disper-
sion model.  The method we have chosen is to express the variances of the
lateral and vertical components of the wake turbulence velocities as functions
                                   14

-------
of the energy dissipated by drag forces on the vehicle and distance from
the vehicle and then to estimate the integral  time scale T  .of the wake
turbulence.  The last quantity and the prescribed wake velocity variances
are then used to simulate particle dispersion  by the wake in the same
manner (described later) in which the velocity and time scale properties
of the ambient turbulence depicted in Figure 3 are used to describe dis-
persion by boundary layer turbulence.  The technique is described below.
                                                   RECEPTOR
                                                   SAMPLE VOLUMES
                                                    SIMULATED PARTICLE
                                                    TRAJECTORY
     FIGURE 3.  SCHEMATIC ILLUSTRATION OF SOME OF THE PARAMETERS REQUIRED
                AND THE METHOD USED TO CALCULATE p
     Let (x ,z ) be the release point at which we wish to evaluate the
           j  o
density function p.  A particle is released from this point at time T = 0.
In the atmospl
are given by:
In the atmosphere,  it acquires  velocity components  u (T)  and w (T)  that
                                                                        (15a)
        W (T)  =
                       = W[ZP(T).T] + W"(T)
(15b)
where (x , z )  denotes the projection of the particle's coordinates on the
receptor plane; tl is the mean wind speed component normal  to the roadway
                                     15

-------
(depicted in Figure 3); u1  and w1 are random velocities representing the
ambient turbulence; and u"  and w" are random velocities representing the
vehicle wake turbulence.  If there is a nonzero mean vertical wind component
w or if the particle is positively or negatively buoyant, Eq. (15b) must
be modified accordingly.

     In the numerical  simulation, the random velocity components u1 and w1
are generated at each time step by an algorithm, described in detail in
                                                  ?       ?
Appendix A, that utilizes the variance profiles u'  and w'  and the
Lagrangian time scale T of the ambient turbulence.  The random wake velo-
cities u" and w" are generated by the same process, except in this case the
energy and Lagrangian time scales of the wake turbulence are used.  (Further
details are given in Chapter V.)  Once values for all the random velocity
components have been specified, these are used together with the given mean
wind profile u in Eq.  (15)  to calculate the particle trajectory
[x (T),Z (T)] for 0 <. T <. T.  As mentioned earlier, the integrating time
period T must be large enough for any particle to pass just beyond the
location of the most remote monitor.

     At those sites at which the predicted particle trajectory passes
through the sample volume (dx,dz) of a monitor, the time  At, that  the
particle spends in that volume  is determined.  Here  the subscript  1 signifies
that this is the first realization of the turbulence processes.

     All of the steps outlined above are now repeated, starting with the
release of a new particle at (x  ,z ) and using  a  new set of random velocity
                               j  o
components u1, w1, u",  and  w"  that  are uncorrelated with the first set.   From
the new realization of the particle trajectory, we determine the residence
time   At2 in each receptor's sample volume.  By continuing this process K
times, we obtain the sequence of At.  values required in Eq. (14) to evaluate
the time integral of 0 that enters into the equation for  the mean concentra-
tion.  To determine how many realizations K are needed to render a finite
sum (over K realizations) an acceptable approximation of  the infinite sum
of Eq. (14), we use a method in which intermediate approximations of the J
integral are stored and analyzed for trends that reveal convergence.
                                     16

-------
      Specifically,  let

                                     K
                                        Atk
represent the estimate of the J integral given by K realizations of the
                                           _K
particle trajectories.  Figure 4 shows how 0  might vary with K.  The
estimate usually varies rapidly as K increases from zero, but it becomes
                                           _i/
increasingly steady as K becomes large and J  converges to the true value
                                                       	K
of the J integral.  As a measure of the variability of J  at realization
K, we use the standard deviation OJ(K) of the n + 1 values J*, 0K"AK, ...,
LTn  ,  where n is any integer greater than 2 (we use 5) and AK is some
sampling interval in K (we use AK = 100).  We consider that CT is an accept-
                                                                      i/
able approximation of the true value of the J integral when oj(K) <. AiT,
where X is some fraction, e.g., A = 0.01.  In our studies with n = 5 and
                                                                	I/
AK = 100, we have found that the convergence condition a,(K) <. \j  is
                                                        j
satisfied within K = 500 realizations at points near the core of a point
source plume, while K = 2500 may be necessary to satisfy the criteria at
                            	I/
the edge of the plume where J  is only one-tenth the centerline value or
less.  We have also found that a great reduction in the computer time
required by the model  can be achieved with virtually no loss of accuracy
by setting A = 0.5 rather than 0.01.  (We were surprised to find that by
using A = 1.0, the overall performance of the model simulations of the GM
experimental data deteriorated by only 15 percent, as measured by the
change in the rms error, while the computer time required dropped to 1/30
of that necessary to run the model using \ = 0.05.)
                  	K
      The value of J  derived  from  carrying  out  this  procedure  for each
 monitoring site is used in either Eq. (8) or Eq.  (9) for the J integral
 to obtain the mean concentration.   Chapter III discusses the agreement
 between the prediction of this model  and those of the Gaussian plume
 formula  for  the meteorological conditions  under which the latter is
 known  to apply.
                                   17

-------
                             K'-n-K
       FIGURE 4.  EXAMPLE OF THE DEPENDENCE OF T ON THE NUMBER OF
                 REALIZATIONS K.  The difference between 3* and the
                 time value of the time integral of J is estimated
                 from (n + 1) moving of U^ as described in the text.
     APPLICATION TO PHOTOCHEMICAL POLLUTANTS
     The roadway model  can be applied to photochemical  pollutants  by
using the closure technique developed in Lamb and Shu  (1978).   To  facili-
tate this operation,  we assume that pollutants are either pseudo-inert
or infinitely fast reacting.   This  division  is prompted by the  fact that,
in most cases of interest, T is of  the order of several minutes, which  is
a long time compared  with the reaction time  of species  such as  NO, 03,
and sulfates around a roadway, but  a short time compared with other
species such as hydrocarbons.  The  category  a given species falls  into
must be decided case  by case, but we anticipate that around roadways, NO,
03, and sulfates can  generally be treated as infinitely fast reacting.
In this case, the equations governing the mean concentrations of these
species are (from Lamb and Shu, 1978):
           cNO(r,t) = max JO ,  [cINO(r,t)  -
                   = X0 (r,t) - cINQ(r,t)  + cMn(r,t)
m^'~
(17)

(18)

(19)
                                   18

-------
Equation (19) assumes that sulfate is emitted by motor vehicles rather than
formed in the atmosphere through the oxidation of SC^.   Here XQ  is the
ambient ozone concentration,  assumed to be given; cfjNO is the mean
concentration of NO in the absence of any chemical  reactions [C\NQ can be
calculated direcely from Eq.  (8) or Eq. (9)]; and £NQ is a mixing parameter,
which we approximate by (see  Table 1  of Lamb and Shu, 1978):

                                 _2
                      r  (r t)  - CINO(!>t)                                 (20)
                      C^'I;    -—         '                             {  '
In Eqs. (17) through (20), r = (x,z) denotes the receptor location.

     In summary, Eq. (8) is the roadway model equation for the mean
concentrations of all pseudo-inert pollutants.  This same equation is
also used to compute quantities like CJNO that enter in Eqs. (17)
through (20).  The model equations for NO, 0.,, and sulfate, which we
consider to be rapid reactants, are Eqs. (17) and (19).  Similar equations
can be written for other rapid reactants.  The evaluation of Eqs. (17)
and (19) requires that the mixing parameter c, be determined.

     The Lagrangian theory from which Eq. (8) was derived (Lamb, 1978a)  also
provides an expression for c2 that can be used to evaluate ;.   We have:

 _            N   N   t t
 c (r,t)  =  Q  /]/J I  I  p«(x,z,t;x,z,t x  ,z  ,t';x  ,z  ,t")
    ~"         M^^MP ^^^^^> V  •   £
              n=l  m=l  0 0

           *[<$  t'(v - vsn)  - tv -  y0nj-6[t"(v -  vsm)  -  tv -  y0  1 dt1  dt"
                                                                         (21)

where  p« is  the joint  probability density for the displacements of a  pair
of particles.   Note  that p  is  a marginal density of P2  in that  the former
can be obtained by  integrating p2 with  respect  to the position  coordinates
of one particle over all space.  The method  described earlier for
                                  19

-------
determining the J integral can be used to obtain the integrals over p2 in
Eq. (21).  However, in this case we must release pairs of particles from
the source point (x ,z ) and determine the total times that both particles
                   ->  5
are simultaneously within the sample volumes of each receptor.  The algo-
rithm described in Appendix A that we use to generate the random velocity
components u1, w',  u",  and w" of single  particles can also be used to
simulate the velocities of pairs of particles where motions are correlated.
In this role, the algorithm requires (in addition to the turbulence velo-
city variance profiles and Lagrangian integral time scale) the energy
dissipation rate profile e(z) of the ambient turbulence and comparable
information for the vehicle wakes.   Some results of particle pair simula-
tions and mean square concentration prediction  are presented  in
Chapter  III,  but applications of the  roadway  model to photochemical pollu-
tants were  not  treated further  in  this  study.
                                    20

-------
                         Ill    MODEL  TESTING


     Since the model described in the previous chapter is somewhat new,
it is appropriate to compare its predictions with those of other diffusion
models in current use.  In particular, by comparing the predictions of our
model with concentration estimates obtained from the Gaussian plume formula
for situations in which this formula is known to apply (namely,  under
conditions with no wind shear, steady meteorology and source emission
rate, no chemical conversion or deposition, and so on), we can verify our
approach as being basically sound.  This chapter discusses these verifica-
tion efforts.

     We set the model up to simulate dispersion from a point source in a
uniform flow (no shear) of speed u = 3.0 m/sec over a flat surface.  The
                                                           —5"      2    2
variances of the ambient turbulence velocities were set to u   = 1 m /sec
    —5"         2    2
and w1' = 0.61 m /sec .  Figure 5 shows 20 simulated particle trajectories
from a source 50 m high for each of two values of the time scale T.
Note that as T •*• 0, the value representative of Brownian motion, the
trajectories acquire the erratic character associated with molecular scale
movement.  The root-mean-square (rms) particle dispersion component o
                                                                     X
corresponding to the cases presented in Figures 5(a) and 5(b) are plotted
                                                    112
in Figure 6.  It is evident that the a ^ t and a ^ t '   regimes  predicted  for
short and long travel  times t,  respectively,  by Taylor (1921)  are  reproduced
by the scheme we use to generate the turbulence velocities.  The pre-
dicted profiles of mean, cross-wind integrated concentration under the
 same conditions  such  as  those  shown in  Figures  5(a)  and  5(b)
are shown in Figure 7 along with the corresponding values given by the
Gaussian plume formula.  A similar comparison is given in Figure 8 for
a source height of 5 m.  The excellent agreement between the model and
                                    21

-------
 cO
 o.
 CSI
100
 2DO	

""t" 1 ^T  I I  I  I"
3DO
                                                          I 1  I  I I  I  I
4oa>
  o
  rvi
I §4-
                                                   o
                                                  4-o
                                   200

                                X (meters)
                           (a)   T = 3  Seconds
                                    zoo

                                 X (meters)
                           (b)  31 =  30 Seconds



         FIGURE 5.  SAMPLE PARTICLE  TRAJECTORIES WITH u" = 3.0,

                    u7? = 1, AND w7? = 0.6 (KKS UNITS)
                                   22

-------
  100
    9
    8
    7

    6
C
o
o
0)
00
 X W
3  »
   8
   7
   6
21 = 3 sec

     SLOPE  =  1/2
                                               1	I    I   I  I  I  I
                                               7     3    4   i  6  7  1
    7891         2     3    4S67«910
                                 t (seconds)
                  9 WO
      FIGURE 6.  THE ROOT-MEAN-SQUARE PARTICLE  DISPERSION  COMPONENT  a
                 AS A FUNCTION OF TRAVEL  TIME FOR TWO VALUES OF THE
                 INTEGRAL TIME  SCALE T
                                  23

-------
         90
         80
         70
         60
         50
         '40
         30
         20
 MODEL PREDICTION

 CALCULATED VALUES FROM
 GAUSSIAN PLUME FORMULA:

   h - SOra
   0 « 3ra/sec
   Q • 2.0
   oj • 12.0
   o, (predicted) • 11.5
   x • 100 m

 LAGRAN6IAN INTEGRAL
 TIM! SCALES:
                                        3 sec
                                        3 sec
                   0.1
                            0.2

                             <*>'
                                    0.3
                                             0.4
                                                      0.5
c«v
FIGURE 7.   COMPARISON OF  ROADWAY MODEL  PREDICTED CROSS-WIND
             INTEGRATED CONCENTRATION WITH GAUSSIAN  PLUME
             FORMULA PREDICTIONS  AT A DISTANCE  OF TOO m
             DOWNWIND OF A  POINT  SOURCE AT 50 m ELEVATION
                              24

-------
  30
ers
t>0
o
  10
                1
I
                               •  MODEL PREDICTION

                               *  CALCULATED VALUES  FROM
                                  GAUSSIAN PLUME  FORMULA:
                                      = 5 m
                                      = 3 m/sec
                                      = 2.0
                                      = r. = 3 sec
                                                   •X
t
I
1
         0.1   0.2   0.3   0.4   0.5   0.6   0.7   0.0   0.9   1.0
                                = c~6v
FIGURE 8.    COMPARISON OF ROADWAY MODEL PREDICTED CROSS-WIND INTEGRATED
            CONCENTRATIONS WITH GAUSSIAN PLUME FORMULA PREDICTIONS AT
            A DISTANCE OF 100 m DOWNWIND OF A POINT SOURCE AT 5  m ELEVATION
                                    25

-------
plume formula predictions is evidence of the validity of our diffusion
modeling approach inasmuch as the plume formula predictions  are correct
under the conditions used in this simulation.
                                    26

-------
              IV   METEOROLOGICAL  INPUTS TO  THE MODEL


     Unlike the  GM study in which extensive measurements  were made of  tur-
bulence quantities, most applied pollution dispersion studies must rely on
a limited set of available observations.   For this  reason,  rather than use
the GM turbulence data directly in the roadway model, we  utilize
theoretical-empirical  expressions for the quantities required that can be
evaluated given  only the local  values of  surface  roughness  z0, Obukhov
length L, and mean wind speed u,n at  10 m.  This  chapter  discusses these
expressions in detail  and compares the values they  yield  with those from
the GM wind and  turbulence data.

     In Chapter  II, we indicated that the model requires  profiles of the
                    	                                      	jy
mean wind component u(z) normal  to the roadway, the profile u  (z] of  the
same component of the turbulent velocjjty_  variance,  the  profile of the
vertical turbulent velocity variance  w'^(z), and  the Lagrangian integral
time scale T of  the turbulence.   The  last quantity  is difficult to measure
because it cannot be obtained directly from turbulence  velocity observa-
tions made at fixed points.  Draxler  (1976) has inferred  approximate values
for T from diffusion data collected at a  number of  sites.  His estimates
of T are of the  order of 100 sec for  the  horizontal velocity  component
under both stable and unstable conditions.  His suggested values for the
time scale of the vertical velocity fluctuations  are 60 sec under unstable
conditions, and 30 sec for stably stratified flows. We found that the
model predictions agreed better with  the  observations at  the most remote
sampling sites (where concentrations  are  most sensitive to T} if integral
time scale values of only one-half those  suggested  by Draxler were used
for the vertical velocity fluctuations.  Thus, in all of  our  simulations,
we used
                                    27

-------
                         = 100 sec

                             30 sec    ,     if    L   < 0   ;
                      lw
                             10 sec    ,     if    L   > 0   ,
where L is the Monin-Obukhov length.
                                  —    2        2
      To  formulate the profiles of u, u' , and w' , we must first express
 the  direction of the mean wind in the coordinate system shown in Figure 9.
 Recall that  in our model u and u' are wind components normal to the road-
 way,  regardless of the orientation of the road.  Keep in mind, too, that
 the  component 7 of the wind parallel to the highway enters only the source
 strength term [see Eq. (8) or Eq. (9)] and that the corresponding turbu-
 lence component v' does not enter at all.  This occurs because of our
 assumption of symmetry of the roadway with respect to the receptor plane.

      The expression for u(z) is the following:

                       u(z) = ^g(z,L,zQ)sin e    ,                 (22)

 where u* is  the friction velocity, k = 0.4 is the von Karman constant, L is
 the  Monin-Obukhov length, ZQ is the surface roughness, and e is the mean
 wind direction in the roadway coordinates shown in Figure 9.  The form of
 the  function g is determined from similarity theory and depends on the
 sign of L as follows  [fromflagland  (1973)]:

                  /z + zn\
 1/L  = 0:    g = £n (— - -^     ;                                      (23)


                                                                     (24)
                                    \       * - ]N*n + ])
                                    ) + £"-M)(*-1)   ;
L < 0:   g = 2 tan'  * - tan"  *J + In /.,. .  lW...   .  1J  ;        (25)
                                     28

-------
where
                       4* =  1  -  15
                                                (26a)
*o= M5r
                                                                      (26b)
                    ROADWAY-
                                                    RECEPTOR PLANE
                                                    (x-z) LOCATION
              FIGURE 9.  COORDINATE SYSTEM USED TO EXPRESS  MEAN WIND
                         AND TURBULENCE VELOCITIES
      Th.e expressions for 1/^(2) and w'2(z)  are  derived from the semi-
  empirical formulation of Binkowski  (1978).   He  gives  expressions for the
  rms  vertical velocity fluctuation o  and the fluctuating components o
  and  o  normal and parallel, respectively, to the  direction of the mean
  wind.  Translating these formulas into the  coordinate system of the road-
  way  shown in Figure 9, we obtain
  where
                                                                       (27)
u'2(z)  = [cos  2F]"1  • [
    oj +
                                               cos2e  -  o  +
                                                 (28)


                                                 (29a)
                                     29

-------
            2    2    2^2
           o  = Q   - J   • o
            U   ^     W    V
   ?r
= <£ 3.
    L
                     0 + 0.75
                                  m
1.8r
(29b)


(29c)
                 rm
                    -  L
   jo  -  i6,r1/4     ,

   (l + 5;   ,     ;> 0    ,

   0.4[0.4 + 0.6 exp(4?)j

   0.4(1.0 + 3.4?  -  0.25;2)
                                           o
                                                                     (29e)
                                                                     (29f)
                                                c  < o
                                                                     (29f)
                z + z,
                                                                     (29g)
and e is the angle of the  mean wind  direction  shown  in  Figure  9.
     Equations (22) through (26)  constitute the bases of the mean wind  esti-
mates used in the roadway model,  and Eqs.  (27)  through (29)  make up the tur-
bulence velocity variance equations employed.   Both sets of  equations require
as inputs the friction velocity u*, the Monin-Obukhov length L,  and the sur-
face roughness.  The remainder of this chapter  describes (_1_) our method of
checking the agreement between the values  of IT, u'2, and w'2 given by Eqs.
(22) through (29) and those measured during the GM study, and (2) the deri-
vation of an empirical formula from which  estimates of u* and L  can be
obtained using wind speed data alone.

     To obtain u* and L values for each of the  GM  experiment periods, we used
u^ and L as fitting parameters to arrive at least-squares fits of the mean
wind profiles given by Eqs. (22)  through (26) to the GM tower measurements.
In the process, we also obtained  a value for the surface roughness of the  GM
site.  The following procedure was used.
                                   30

-------
     First, we chose a value of ZQ in the range of values  that  is  reason-
able for the terrain surrounding the GM test facility,  i.e.,  0.01  <  zn < 0.1
meter.  Then, consideration of the wind data from Towers  1,  5,  and 6 only
(the other towers are so close to the roadway that they are  subject  to vehicle
wake influences) resulted in a total of nine observations  of  u\  three at
each of the levels of 1.5, 4.5, and 10.5 meters, to which  the least  squares
fitting routine could be applied.  Letting e  be the difference  between the
observed value of u" at station n (n = 1, 2, ..., 9) and the  value  calculated
from the formula u = u*g(zn,L,zQ)/k, we chose u* such that the mean  error,
T,  was  zero.   The following equations  were written:
                      "   . U*9n     .     n  =  1,  ...,  9     ;              (30)
                 en     n "  k
where
          u  = observation at station n,
          gn - g(zn.L,z0)

and
                  n
                     , = 0
or
 for  given ZQ and L.

      Thus,
                                                                      (3D
 A  computer  program was set up to solve the u* values for each experiment
 from the  initial guess of the surface roughness (ZQ).  Using the following
 equation, ICO Monin-Obukhov  lengths  were  generated:
                                    31

-------
                              L = exp(X)    ,                         (32)

where

                      X = 0, 0.1, 0.2, ..., 10.0

and

                          sic]n(L) = sign(Ri)

where R. is the Richardson number measured during each  experiment.   For  each
of the 100 values of L, we determined u* from  Eq.  (31)  and the  corresponding
theoretical value ~u at each of the nine observation sites  from  Eq.  (22), namely,

                       u{z) = ^fg(z,L,z0)     .                      (33)
                           ~~2
The mean square difference e (L) between the observed wind speeds and
those given by Eq. (32) for each of the 100 selected values of L was deter-
mined by:
                           9
                 2                             2
                c(L) -       [un - u(zn,z0,L)]       .               (34)
                          n~* i
Of the 100 values of L, the best estimate of the true value was taken to
                  ~2                                      ~~7
be that for which e  is a minimum.  The minimum values of e* for each of
the GM experiments were then summed to obtain a "global" error for the
chosen value of z~.  The entire process described above was repeated using
a different guess for z,.;  from the set of global  errors  obtained,  the
true value of the surface roughness ZQ was taken to be that corresponding
to the minimum global error.

     Using the technique outlined above, we found that a surface roughness
ZQ = 0.05 m gave the best fit of the observed wind data to the similarity
profiles.  The values of u* and L that we derived for this surface rough-
ness are listed in Table 1 for each of the GM experiment periods.
                                     32

-------
TABLE 1.   ESTIMATED VALUES OF u* AND L  FOR  THE
          GM EXPERIMENT PERIODS (ZQ = 5 cm)  AND
          THE ANGLE ¥ OF THE MEAN WIND  RELATIVE
          TO THE ROAD
                 Qbukhov Length
GM Experiment
275080
275083
275090
275093
276081
276084
276091
276094
279080
279084
279090
279093
281080
281083
281090
281093
283081
283085
283092
283095
286081
286084
286091
286094
290080
290083
290090
290093
293103
~~ (mf "
-2.7
-2.7
-2.7
-2.7
-270.4
-181.2
-99.4
-90.0
16.4
24.5
-121.5
-109.9
73.7
298.9
148.4
1.8
54.6
-20.1
-2.7
-66.7
110.0
73.7
2.2
-330.3
812.4
181.3
601.9
1096.6
-544.6
(m/sec)
0.20
0.29
0.37
0.41
0.20
0.20
0.27
0.30
0.09
0.09
0.13
0.18
0.17
0.11
0.08
0.19
0.11
0.10
0.15
0.13
0.21
0.23
0.17
0.25
0.19
0.18
0.19
0.22
0.19
(rad)
2.50
2.28
2.76
2.81
0.56
0.64
0.83
0.99
1.23
1.14
1.28
1.23
3.96
3.86
4.76
4.85
1.25
1.39
1.01
0.64
0.19
0.33
0.41
0.45
4.38
4.15
4.20
4.11
1.58
                        33

-------
TABLE 1  (Concluded)






      Obukhov Length
GM Experiment
293110
294080
294083
294090
294093
295080
295083
295090
295093
296080
296083
296090
296093
297080
297083
297090
300080
300083
300090
300093
302080
302083
302090
302093
303080
303083
303090
303093
(m)
-330.3
18.2
21.0
-200.3
-66.7
4.1
18.2
3.7
2.7
134.3
181.3
244.7
-134.3
99.5
90.0
200.3
36.6
-121.5
-200.3
-99.5
1.8
-30.0
-36.6
-30.0
1.5
-1.1
-1.1
-5.0
(m/sec)
0.19
0.14
0.14
0.10
0.13
0.01
0.04
0.03
0.02
0.23
0.25
0.28
0.26
0.20
0.19
0.25
o.in
0.19
0.21
0.21
0,19
0.23
0.28
0.27
0.11
0.11
0.14
0.21
(rad
1.57
0.87
0.86
0.63
0.77
4.65
4.25
4.15
3.70
0.04
0.04
0.16
0.19
0.06
0.07
6.24
0.28
0.43
0.37
0.41
2.84
2.97
3.06
3.15
2.47
2.97
3.34
3.54
          34

-------
Table 1  also includes the angle e of the mean wind.   Figures  10(a)  through
10(c) compare the observed wind speeds at each of the three  levels  of 1.5,
4.5, and 10.5 meters with the corresponding values obtained  from Eqs.  (22)
through (26) using the u* and L values shown in Table 1.   The observations
represent average wind speeds at each level in Towers 1,5,  and  6.   A vast
majority of the predicted speeds are within ±10 percent of the measured
values,  indicating that the similarity expressions incorporated  in  Eqs.  (22)
through (26) are indeed accurate representations of  the wind  speed  profiles
near the ground.

      Checking the accuracies of the turbulence velocity  variance expres-
 sions [Eqs. (27) through (29)] requires some caution because the measured
 values of these parameters are more sensitive to vehicle wake turbulence
 than are the mean wind components.   Accordingly, we first considered only
 situations with westerly winds, i.e., wind directions between 250° and 290°,
 and used only the turbulence data collected at Tower 1,  which is farthest
 west of the roadway.  The resulting observed and predicted  values  of
 u' = ~^Z *   and w1 = w'2 '   are displayed in Figures ll(a)  and  ll(b),
 respectively.  The observed and predicted values of w1 agree quite well--60
 percent of the predicted values  are within  ±20 percent  of the measurements;
 but the predicted values of u1 are consistently too low  by  about 35 percent.
 Almost all of the data points used in Figure 11 are for  cases of unstable
 stratification, and it is during these conditions that the  Kansas  and
 Minnesota  turbulence data, which Binkowski (1970)  used to develop  the
 formulas in Eq. (29), show the largest scatter in the o   and o   values.
 In fact, the scatter is sufficiently large that profiles of o  and a
 exist that would be compatible with both the Binkowski and  the  GM  data
 sets.  We  have not pursued this point.  In the model validation studies
 (discussed  in  Chapter V), we  simply multiplied all  the u1 predictions by
 1.5 to eliminate the discrepancy shown in Figure 11 (a).
                                    35

-------
        i.t
                    a.»          a.*
                        Observed
                                            4.*
                                                        I.t
         (a)  z = 10.5 meters  (zn = 0.05 meters)
FIGURE 10.  OBSERVED VERSUS  PREDICTED MEAN WIND SPEED
                          36

-------
    3.0  -
Cl
4->

U
•n
c
   i .0
                                             3 »


                                      Observed
                                                                      0.*
                      (b)  z = 4.5 meters  (z  = 0.05 meters)
                                FIGURE  10  (Continued)
                                        37

-------
o
4->
'_>
                                           3.*


                                      Observed
                    (c)  z = 1.5 meters  (ZQ  =  0.05 meters)
                             FIGURE 10  (Concluded)
                                     38

-------
 0.0
0.2
                       0.8
                                                           -0.8
  r.o
0.2
0.4        0.6
 Observations
0.8
                                                             0.0
                                —pl/2
                            (a)  u'2
                                                 •1/2
FIGURE  11.  OBSERVED VERSUS PREDICTED VALUED OF u'2    AND  w'2
           Only cases with wind  direction e in the range
           250 < ¥ < 290 are represented, and only data from
           Tower 1 are included.
                                               1/2
                            39

-------
0.0        0.1
                                        ±20% ERROR BOUNDS -
                       0.2         0.3
                         Observations
                         (b)   w
                               ,2
                                 1/2
                   FIGURE 11  (Concluded)
                           40

-------
     Our efforts to check the accuracies of the variance formulas  [Eqs.  (27)
through (29)] were extended by considering data from Tower  6,  the  east-most
tower, only in cases of wind directions e" in the range of 70°  < ¥  <  110°.
The predicted and observed values of u1 and w1  are  shown in Figures  12(a)
and 12(b), respectively.  The tendency for the  predicted u1  value  to be  too
low is again apparent.   One noteworthy set of points shown  in  both figures
is that for which the predicted values of u1 and w'  are near zero--u'  =  0.02,
w' = 0.01--but the observed values are over 10  times larger.  This particular
set of points corresponds to a case of extremely stable flow and light winds
(Day  295) with a vertical temperature gradient  of +1°C per  5 meters  and
u ^0.4 m/sec.  The large value of u1 measured  in this case is most  likely
due to meanders in the  local circulation induced by topography and inhomogen-
eities in surface roughness, and the large w' values observed  may  be due to  horv
zontal divergence in the circulation, to gravity waves, or  to  both.   In  any
event, the fluctuating  velocities observed are  not  the result  of the mech-
anical and buoyancy phenomena that generate the turbulence  described by
similarity theory.  Describing the dispersive properties of very stably
stratified flows is one of the largest unsolved problems of diffusion
modeling.

     Finally, we compared the predicted u'  and  w1  values with  those  mea-
sured at Towers 1, 5, and 6 when the wind direction  e" fell  in  either of
the sectors 340 < ? < 20 or 160 < ¥ < 200.   These comparisons  are  given  in
Figures 13(a) and 13(b).  The dashed line in Figure 13(a) is the best fit
line of the u' values shown in Figure ll(a). It is  apparent that  with
northerly or southerly winds the predicted u' values are not as low  as
they are for westerly flows.  The dependence on wind direction of  the dis-
crepancy between the observed and predicted values  of u' may be the  result
of the dependence of o  and a  on inversion height.   Panofsky  et al.  (1977)
have proposed the relationship:

                ou = ay = u*[12 - h/(2L)]1/3    ,    L < 0     ,       (35)
                                   41

-------
     0.0
      0.2
          0.6
0.8
    0.8
                                                      0.8
 0
 •5
 cu
 Q.
    O.B
    0.4
                                                      O.B
    0.2
                                                      0.2
                 0.2
J	I
.4        0.6
 Observations
                                        0.8
                                                                 0.0
FIGURE  12.
                             (a)  u' = u'2
                                         1/2
OBSERVED VERSUS PREDICTED VALUES OF u' = u'2
                                                      1/2
                           AND w' = w'2
                                                          1/2
           Only cases with wind direction e in the range 70 < e  < 110 are
           represented,  and  only data from Tower  6 are included.
                                 42

-------
0.0
0. t
            0.1
           0.2        0.3
             observations
                       1/2
                                                         -0.1
                                                          D.O
                     (b)   w1  = w'2
                  FIGURE 12  (Concluded)
                           43

-------
     0.0
              0.2
0.4
0.6
0.8
•*->
o
•^
-o
OJ

Q_
O.J
                 0.2
                          0.4        0.6

                           Observations


                             (a)  u'
                        0.8
                                                                  0.0
      FIGURE  13.  OBSERVED VERSUS PREDICTED VALUES OF u1 AND w1.  Only
                 data for cases with wind direction  in the sectors
                 34Q < Q < 20 and 160 < "5" < 200 are  represented, and
                 only measurements from Towers  1,  5, and 6 are included.
                               44

-------
  0.0
0.1
0.2
0.3
o.q
O.J
             0.1
           0.2        0.3
             Observations

               (b)  W
                      0.1
                                              0.0
                    FIGURE  13 (Concluded)
                            45

-------
where h is the mixed layer depth or inversion base height.   They have found
very good agreement between this equation and measured data.  Binkowski's
formulas [Eq. (29)] for o  and a  are independent of h.  Consequently, it
is quite likely that the dependence on wind direction of the accuracy of
Eq. (29) is a result of a systematic variation of h with wind direction at
the GM site.   Since inversion data  were  not  measured  during the GM  study,
we  cannot  test  this hypothesis now.

     On the whole,  the comparisons presented in Figures 11  through 13 indi-
cate that Eq. (27)  provides acceptably accurate estimates of w'2 '   but
         _ -I /O
that the u'2 '  values obtained through Eq. (29) are subject to some uncer-
tainty.  We believe that through proper modifications of Eq. (29) or by
replacing this formula altogether with Eq. (35), acceptably accurate esti-
mates of the horizontal turbulence variances can also be obtained.

     As a result of the good agreement between the if values predicted by
similarity  theory and the measurements (see Figure 10), we decided to
explore the possibility of deriving estimates of u* and L from wind data
alone.  According to Eq. (22), the ratio of the wind speeds measured at
two heights,  z,  and Zp,  is:

                              IT,   g(z,,L,zn)
                              _ l_ _     I __ U__
                              TI2 " g(z2,L,zQ) ~    '

For given values of z, , z^, and ZQ, the right side of Eq. (36)  is a
unique  function of  L,  at  least in the interval  -0.1  <. 1/L  <. 0.1.  Figure 14
shows  a plot of this function  for z,  =" 10.5, z? = 1.5, and z~ = 5 cm.  Also
shown  is  a  plot of  the drag coefficient:
                               c   = -u-*  =
                               CD
 obtained from Eqs.  (22)  through  (26).   Using the wind speed measure-
 ments made at 10.5  and 1.5  meters and  the two curves  plotted  in Figure 14,
                                    46

-------
LT)

O


I =3

  *
    0.10 -
    0.08-
    0.06 -
    0.04 -
     -0.10
-0.05
0.05
                               1/L  (per  meter)
                                                                  = 2.5
                                              -2.0
                                                   LO

                                                   O
                                                                  -1.5
                                                                    1.0
0.10
  FIGURE 14.   PLOTS OF THE RATIO  OF  THE  WIND  SPEEDS AT 10.5 AND 1.5 METERS
              AND THE DRAG COEFFICIENT u*/u(10.5) GIVEN BY SIMILARITY
              THEORY AS FUNCTIONS OF 1/L.  The surface roughness, ZQ, is 5 cm.

-------
we can obtain estimates of both u* and L.  For convenience, we refer
to this as the wind speed ratio (WSPx) method.  The technique can  be used
for any values of z,, z2, and ZQ by recomputing the Zu/ZL and u*/u.,  curves.
We should also point out that with this method, L = 10 is assumed if Lu/IL
exceeds the theoretical value at 1/L = 0.1,  and L = -10  is assumed if the
wind speed ratio is less than the theoretical value at 1/L = -0.1.   Figure
15 compares the u* and L values given by the WSR method for each  of the GM
experiments with the corresponding values listed in Table 1.  In  applying
the WSR method, we used data from only Tower 1 or Tower 6, depending on
which was upwind of the roadway, rather than averages  for  several towers.

     Looking first at the u* estimates, we find that the two sets of values
are in fair agreement:  Over 67 percent of the WSR estimates are  within ±20
percent of the values given in Table 1.  However, there is a distinct ten-
dency for the WSR method to overpredict u*,  a direct result of the method's
tendency to predict unstable stratification  {for which u*/lT,) is  large)  in
situations where the values in Table 1 indicate that stable stratification
existed.  Recall that in obtaining the L values of Table 1, the sign of  L
was taken from the Richardson number, which  was measured during each of  the
GM experiments.  In general, there is little agreement between the two sets
of 1/L values shown in Figure 15(b).  In fact, only 60 percent of the values
given by the WSR method agree in sign with those in Table  I—only
slightly better than the agreement that  could be obtained by making  random
guesses of the sign of L.

     A case-by-case examination of the meteorological conditions  existing
during those periods when the signs of the L values given by the  two
methods disagree revealed the following findings.  On 4 of the 23 occa-
sions for which the signs differ, the sign of the Richardson number mea-
sured at Tower 1 was opposite that recorded  at Tower 6 some 100 meters
away.  In 9 of the remaining 19 cases, the temperature stratification was
not uniform vertically at the tower from which the wind data used in the
WSR method were taken; the sign of the temperature difference between the
top two levels of the tower was opposite that between the bottom two levels.
The 4 cases in which the Richardson numbers  at Towers 1 and 6 had opposite
                                   48

-------
                                            +15%
   0.3
o
.c
-i->

-------
-0.1   -0.08 -0".06 -0'.04  -0'.02'"V
 .*  "0.02   0.04  0.06  0.08  OJO

        Observed  1/L  (Table 1)

  -0.02


  • 4

  -0.04 K
        3



V-0.06 -
4 -0.08
                                 + -0.1
                                         •o
                                         0)
                                         4->
                                         o
                                         •I—
                                         •o
                                         o

                                         CL
                   (b)   Monin-Obukhov Length, L
                      FIGURE 15 (Concluded)
                                50

-------
signs all  occurred on Day 290,  when the flow was stably stratified upwind
of the roadway during the observation period and unstable downwind.   This
suggests that the heat from large numbers of automobiles and  the release
of heat stored in the roadway itself can significantly alter  the turbulence
locally when the ambient flow is stably stratified.   Another  possible cause
of change in stability downwind of the road is the overturning  of the cold
air near the ground and the warmer air above by the vehicle wake turbulence.
Our finding that the vertical temperature structure near the  ground  was
not uniform in a number of instances in which the two methods of estimating
L give radically different results emphasizes the problem of  characteriz-
ing stability when the vertical temperature profile is nearly isothermal.
Chapter V quantitatively examines the magnitude of the errors in mean con-
centration estimates resulting from uncertainties in the values of u* and L,
                                   51

-------
                  V  RESULTS  OF PRELIMINARY  MODEL
                      VALIDATION  EXPERIMENTS
     This  chapter discusses the SFg concentrations  predicted by our model
for each of the GM experiment periods  and  compares  them with the values
actually observed.   The roadway used in the GM  study  is oriented north-south.
Instrumented towers were placed 30 and 2 meters west  of the roadway; one
tower was  sited in the roadway median; and additional  towers were located
at distances of 4,  15, 30,  50, and 100 meters east  of the roadway's eastern
edge.  Thirty-minute averaged SFg concentrations were measured at the 1.5,
4.5, and 10.5 meter levels  of all towers except those located 50 and 100
meters east of the road, where measurements were made at 1.5 meters only.   Thus,
a total  of 20 SFg measurements were made during each  experiment.  The reader
is referred to Cadle et al. (1976) for further  details on the GM monitoring
network  and study conditions.

A.   SIMULATIONS THAT NEGLECT WAKE EFFECTS

     In  previous chapters we detailed  all  aspects of  the roadway model
except one—the vehicle wake turbulence parameterization.  Most early
roadway  models ignored the  dispersive  effects of wake turbulence entirely,
but models developed within the last few years  have included some means
of approximating these effects.  Apparently, however, no  quantitative model-
ing studies have ever been  performed to establish the actual importance of
wake effects to accurate predictions.   We  performed such a study by
exercising our model under  the conditions  of the GM experiments, with a
wake turbulence parameterization omitted entirely.   Obtaining good  agree-
ment between the simulation results and the measured  data would  indicate
that, at least  insofar as  the mean  concentration of an inert material  (SFg)
is  concerned, wake effects are minor and can be neglected.
                                  52

-------
     In Figure 16, the SFg concentrations predicted by the model  with  a
wake parameterization omitted are compared with the GM data.   We  emphasize
that the only input data used in the model for these calculations were the
SFg emissions rates reported in Cadle et al.  (1976), the u*,  F, and L  values
given earlier in Table 1, and estimates of the time scales T   and T given
on page 28.  The only exceptions are the four experiments conducted on Day
295.  As we noted earlier in Chapter IV, similarity theory completely  failed
to duplicate conditions on this extremely calm, stable day.   The  bulk
Richardson's numbers measured during the four experiments conducted on
Day 295 ranged between 0.35 and 7.5, which are outside the limit  (viz, 0.2)
of applicability of the similarity expressions that we use in the model.
Consequently, in the simulations of the four experiments conducted on
Day 295, we forced the IT, a , and o  profiles to agree with the averaged
measured values at the 10.5 meter levels of the towers 30 meters  east  and
west of the road.  Our justification for altering the input data  in this
manner lies in the fact that we are concerned here with how well  the dif-
fusion model itself performs under various meteorological conditions.
Therefore, we are attempting in these validation and testing  exercises to
minimize errors in the input data as much as possible so that errors in
the predicted concentrations will reveal inadequacies in the  model itself.
We showed in the previous chapter that the similarity expressions that we
have incorporated in the diffusion model yield quite accurate mean wind
and turbulence profiles given only the local  values of the friction velocity
and Monin-Obukhov length.  However, under extremely stable conditions,
namely R. > 0.2, these expressions break down altogether and  the  profiles
they produce are completely unreliable.  Consequently, under  these con-
ditions one must rely on measured winds and turbulence parameters for
reliable estimates.  We have adhered to this procedure in our validation
studies.

     We find in Figure 16 that, with the exception of several of  the values
predicted for Day 295 (designated by arrows in the figure), the  performance
of the model with wake effects ignored is not particularly bad.   The rms
error of the 1100 predicted values is 0.80 ppb (the averaged  observed  value
                                    53

-------
                  12.5
en
                  ii.i
                   7.5
               on
               u
               
-------
is 0.71), the correlation coefficient p = 0.79,  and the regression  line  is
PRED = 0.07 +1.27 •  OBS.  One of the features that sets Day 295  apart from
all the others is extremely light winds:   IT--0.5 m/sec.  On all  other days,
wind speeds were typically higher than 1.5 m/sec.

     Considering the results shown in Figure 16 from this perspective, we
speculate that wake effects are apparently most pronounced when wind speeds
are less than 1 m/sec, at least when the vehicle speeds are 22 m/sec and
higher, as they were in the GM study.  This speculation is supported by
Figure 17, where we have replotted the results shown in Figure 16 with
the 80 values predicted for Day 295 deleted.  In this subsample of  1020
points, and with the wake parameterization scheme still omitted,  the rms
error is o = 0.53 ppb, the correlation coefficient p = 0.87, and  the regres-
sion line is PRED = 0.07 +1.16 • OBS.

B.   SIMULATIONS THAT INCLUDE A SIMPLE PARAMETERIZATION OF WAKE EFFECTS

     At least three quantities are required to model the wake properly:
the initial turbulent energy in the wake, the time scale of the energy
decay, and the Lagrangian integral time scale of the wake turbulence.
Because values are not currently available for any of these parameters,
it was necessary to formulate a wake  parameterization empirically.
Our initial efforts, which are quite primitive, were based on the follow-
ing assumptions.  First, for simplicity, the integral time scale of the
wake turbulence was assumed  to be zero.   (This  is  tantamount to assuming
that the  diffusive action of  the wake  is  akin to that of  Fickian diffusion.)
Second,  the  initial wake turbulence energy was assumed  to be proportional
to  the work  done by the drag  force on  the motor vehicle;  that is,
                                    55

-------
                      it
en
CTl
                   -Q

                   §:
                   oo

                   -a
                    0.25) OMITTED

-------
                                                                     (38)
where
               Cn = the vehicle drag coefficient,
                A = the cross-sectional  area of the vehicle,
                V = the vehicle speed relative to  the air,
            y ,y  = constants,
            2   2
          u" ,w"  = the horizontal  and vertical components,
                    respectively,  of the initial mean square
                    velocity fluctuations in the wake.

Third, to account for the decay of the wake energy, the wake  intensity
was assumed to be constant at the level  given by Eqs. (38)  and (39)  for a
period of T  seconds, after which the energy was assumed to drop abruptly
to zero and to remain there for all later travel  time.

     The foregoing description of the wake was incorporated into the
stochastic velocity generator used to produce the  simulated wake velocity
components u" and w" described in Chapter II [see  Eq. (15)].

     To avoid using y , y , and T  as devices for  "tuning"  the model, we
                     U   W       W
attempted to determine values for the first two parameters  from the wind
and turbulence measurements made at the sites closest to the roadway.
We chose
                             •y  = Y
                              u   rw
                                        2
and then using CD = 0.5, and A = (2.5 m) , values chosen somewhat arbi-
trarily, we selected a value for y that made Eqs. (38) and (39) yield
velocity variances close to those suggested by the wind and turbulence
                                    57

-------
measurements.  Through this process, we found 0.005 < y < 0.01,   Next,
to refine the estimate of y and to obtain a value for T , we exercised
the model with the wake parameterization included to simulate Experiment
279080.  (This period was chosen completely arbitrarily.)  We varied the
value of y within the range noted above, and we experimented with various
values of T  to determine which set of values produced concentration
           w
predictions in closest agreement with the measured data.  We were partic-
ularly concerned with reproducing the upstream diffusion that did not
occur in the simulation in which the wake was ignored.  The following
parameter values were selected from these exercises:
                        Tw - 3 sec    ,

                         Y = 0.007 sec1/2/m3/2

     Using these values in the wake parameterization along with the values
of C~ and A given earlier, we incorporated the wake simulation algorithm
into the model and then repeated the simulation of all of the GM experi-
ment periods.  The source emission rates and the values of u*, 9", L, T  , and
T employed  in these calculations were  identical to those used in the simu-
lations  illustrated in  Figure 16 in which the wake effects were ignored.
Figure 18 shows the results of the comparison of the new calculations {with
the wake simulation included) with the  observed data.

     Comparison of the  results presented in Figure 16 and 18 indicates that
the wake parameterization scheme, as simple as it is, has improved the
accuracy of  the model predictions considerably.  The correlation coefficient
of the new results is very high (p = 0.91), and 53 percent of the 1100 pre-
dicted concentrations are within ±30 percent of the measured values (as
illustrated  later in Figure 22).  The rms error a = 0.33 and the regres-
sion line is  PRED = 0.13 + 0.91 • OBS.  In the course of testing the wake
parameterization scheme, we discovered  that while the overall performance
of the model  was greatly improved with  the scheme included, there were a
number of simulation periods in which the model performed considerably
                                    58

-------
en
10
                                                       2              3

                                                      Observed SFg (ppb)
                       FIGURE 18.  COMPARISON OF OBSERVED SF6 CONCENTRATIONS WITH THOSE PREDICTED BY THE

                                   ROADWAY MODEL WITH WAKE TURBULENCE EFFECTS PARAMETERIZED:  ALL

                                   EXPERIMENTS  (1100 DATA POINTS)  INCLUDED

-------
better with the wake parameterization omitted.   On seeking an explanation
of this finding, we discovered that in all  cases in which the model  per-
formed better with the wake effects ignored,  the mean wind direction e"
was within about ±10° of the roadway, namely, (sin ¥ | < 0.2.  The number
of cases involved (13 of the 55 GM study periods), the high correlation
with wind direction, and the difference in the rms error of the model
realized between simulations made with and without the wake parameteriza-
tion were strong evidence that the phenomenon in question was not merely
a chance event, but rather the result of some physical process associated
with the vehicle wake.

     After examining the condition of the GM experiment more closely,  we
concluded, with some confidence,  that the process in question is the so-
called "slipstream" phenomenon.  The vehicles in the GM experiment were
spaced only a few vehicle lengths apart, so that each vehicle was within
the wake of the one ahead of it,  except when the mean wind component normal
to the roadway was strong enough to force each vehicle's wake out of the
path of the vehicle immediately behind.   When the vehicles are well  within
each other's wakes, the air speed relative to each vehicle is considerably
lower than that which a single vehicle would experience in the same meteo-
rological conditions.  Thus, with close vehicle spacing, the drag work done
by each vehicle, and hence the intensity of turbulence in the wake pro-
duced, are lower than that which would occur if each vehicle were not
within the wakes of others.

     To account for this phenomenon, we set the air velocity relative to
each vehicle to 1/10 of its nominal value when the wind angle e" is in the
range  |sin ¥| < 0.2.  The model results displayed in Figure 18 include
this modification.

     In order to analyze the model's capabilities further, we have plotted
in Figures 19 and 20 scatter diagrams of predicted versus observed concen-
trations for stable and unstable conditions separately.  These figures
reveal a tendency for the model to overpredict slightly in stably
                                    60

-------
                               Observed  SFg (ppb)


FIGURE 19.   COMPARISON OF OBSERVED SF6 CONCENTRATIONS  WITH  THOSE  PREDICTED  BY THE
            ROADWAY MODEL WITH WAKE TURBULENCE  EFFECTS PARAMETERIZED:   RESULTS
            DISPLAYED FOR STABLY STRATIFIED FLOWS  ONLY (580 DATA  POINTS)

-------
ro
                                                   I  I  I  |  I  I  I  I	1   I  I  I	1

                                                        l.S        l.t        8.5
                                                     Observed SF6 (ppb)
3.S
                      FIGURE 20.  COMPARISON OF OBSERVED SF6 CONCENTRATIONS WITH THOSE PREDICTED BY THE
                                  ROADWAY MODEL WITH WAKE TURBULENCE EFFECTS PARAMETERIZED:  RESULTS
                                  DISPLAYED FOR UNSTABLY STRATIFIED FLOWS ONLY (520 DATA POINTS)

-------
stratified flows and to underpredict during unstable situations.   These
tendencies are also apparent from the following  performance statistics:

     >  Stable cases only
        Number of data points N = 580
        Average predicted concentrations P = 0.85 ppb
        Average observed concentrations 0 = 0.76 ppb
        a = 0.38 ppb
        P = 0.91
        PRED = 0.13 + 0.95 • OBS.
     >  Unstable cases only
        N = 520
        P = 0.68 ppb
        0 = 0.65 ppb
        a = 0.28 ppb
        p = 0.91
        PRED = 0.14 + 0.83 • OBS.

     The fact that the intercept of the regression line for the cases
with the wake scheme employed is larger than that obtained with the scheme
omitted indicates that the wake parameterization is spreading material
over too large an area.  Another feature of the  model results is the lower
slope of the regression line achieved.  Contributing to this deficiency  is
the set of points predicted for Day 275.  For some unknown reason, the
highest observed values on one of the experiments of this day are consis-
tently higher than the predicted values by a factor of two.  The other
experiment shows a similar underestimation.

     A quantity of great concern in applied air  pollution studies is the
expected peak concentration.  Figure 21 displays a scatter plot of the
predicted versus observed peak values obtained in the 55 experiments.   The
sloping dashed lines in the figure delineate the ±25 percent error bounds.
Over 75 percent of the calculated peak values fall within these bounds.
                                    63

-------
                              PRED =  -
                              1.25 •  OBS
CL
 10
u_
CO


"So   3
 
 o
• r-
•o
 (!)

Q-


                                                      RED =
                                                     0.75 •  OB
         III!   'I 1  I  I    I  I I
                                                                1  1  1  »
                                                                           fill
     1.6
                 2.*
8.6         3.«         3.6         «.»

      Observed  Peak SFg (ppb)
4.S
6.»
       FIGURE 21.   COMPARISON OF OBSERVED PEAK SFe CONCENTRATION  FOR  EACH  OF THE
                   55 GM EXPERIMENTS WITH THE VALUES PREDICTED BY THE MODEL
                   (WAKE EFFECTS INCLUDED)

-------
     The performance statistics of the model  are summarized  in  Table  2,
and plots of the fractional  error distributions found for several  of  the
validation studies are given in Figure 22.   The basic model  requires  about
36,000 words of core storage and it used about 10 minutes of CPU  time on
the UNI VAC 1110 to simulate all 55 of the GM experiments. Shorter execu-
tion times can be achieved by relaxing the error bounds on the  predicted
concentrations (see Table 2).  The computer memory requirement  increases
by 3 words for each additional monitor site added, regardless of  the  dis-
tance of the monitor from the roadway.  The machine time required by  the
model increases very slightly as additional sampling sites are  added  to
a given area.  However, if the area itself is extended, the  machine time
requirement increases.  We have not established the quantitative  dependence
of the machine time requirement on the size of the sampling  domain.
                                    65

-------
TABLE 2.  MODEL PERFORMANCE STATISTICS
                         Error Distribution--
                            percentage of
                           predicted value
                            with error <.:
Regression Line--
PRED = a + B -DBS

1.
2.
3.
4.
5.
N =
P" =
0 =
a =
P
Conditions
All experiments,
wake "on"
Day 295 omitted,
wake "on"
Same as No. 1 but
with relaxed
error (execution
time - 3 min;
Case 1 = 10 min)
All experiments,
wake "off"
Day 295 omitted,
wake "off"
total data points.
Average predicted SF
Average observed SFg
rms error.
N P 0 o p 25% 50% 100% a
1100 0.76 0.71 0.33 0.91 47% 68% 83% 0.12
1020 0.74 0.70 0.30 0.92 48 70 85 0.11
1100 0.76 0.71 0.38 0.88 42 65 81 0.13
1100 0.96 0.71 0.80 0.79 36 57 80 0.07
1020 0.88 0.70 0.53 0.87 37 59 82 0.07
6 (Ppb).
•

e
0.91
0.90
0.90
1.27
1.16



correlation coefficient.

-------
     100
cu
C1 C
(D •!-
-l-> -C
C 4J
0> •..-
U 3
i-
OJ
ex
      80
CO
cu
  O

Si  60
u
•»- C
-a cu
a> >
40
      20
WAKE "ON"

WAKE "OFF"--DAY 295
EXCLUDED

WAKE "OFF"
                    20
                          40           60

                                Error (percent)
   80
100
120
  FIGURE 22.   DISTRIBUTION OF FRACTIONAL  ERROR.   The curve labeled  "wake  'on'"  includes
              wake parameterization;  "wake 'off'" simulations ignore wake  effects.

-------
                           VI   SUMMARY
     We have developed a model  of air pollutant transport,  dispersion,
and removal  processes based on  Lagrangian principles.   The  model  requires
as inputs estimates of the local  friction velocity,  the Monin-Obukhov
length, the  surface roughness,  and source emission  rates.   In  applications
to pollutant dispersal within 100 meters  of a roadway,  we showed  that the
model performs very well provided that the Richardson's number does  not
exceed approximately 0.25.   We  also found that

     >  Vehicle wake turbulence plays a dominant role  in
        dispersing material within 10 meters of roadways when
        the  wind speeds are less  than about 1 m/sec  and the
        vehicle speeds are about  20 m/sec or larger.
     >  The  dispersive effects  of wake turbulence are
        greatly reduced when vehicles are closely spaced and
        the  wind is parallel to the roadway.
     >  Except for situations in  which wind speeds  are  less
        than 1 m/sec, spatial and temporal variations  in
        material concentration  are dominated by ambient tur-
        bulence.  A correlation coefficient of 0.87  was obtained
        under these conditions  with wake  effects ignored
        entirely.   However, the rms error of the estimates  is
        greatly improved if wake  effects  are taken  into account.
                                   68

-------
                          APPENDIX A

             A SCHEME FOR SIMULATING  PARTICLE
             PAIR  MOTIONS IN TURBULENT FLUID
                        Robert G. Lamb*

              Meteorology and Assessment Division
          Environmental  Sciences Research Laboratory
                Environmental Protection Agency
         Research Triangle Park, North Carolina  27711

         Preprint.  Submitted for publication to the
           Journal of Computational Physics, 1978
* On assignment with the EPA  from  NOAA, U.S. Department of Commerce.
                              69

-------
                               APPENDIX A
                  A SCHEME FOR SIMULATING PARTICLE
                  PAIR MOTIONS IN TURBULENT FLUID
                               ABSTRACT

     A technique is  developed  and  tested for simulating the random
motion of particles  in  turbulent fluid.  The technique is applicable to
single particles or  to  pairs of particles whose motions are correlated.
The method is intended  for  use in  determining one- or two-particle
Lagrangian turbulence statistics from  instantaneous velocity fields
generated by numerical  turbulence  models.  The technique can also be
used as the basis of the Monte Carlo type of models of turbulent diffu-
sion.  It requires as inputs the mean,  variance, and Lagrangian inte-
gral time scale of the  turbulence, and  for particle pair simulations,
it additionally requires the energy dissipation rate of the turbulence.

1.   INTRODUCTION

     In numerical studies of material  dispersion in turbulent fluid,
only a portion of the spectrum of  fluid motions can be described in a
deterministic manner.  These motions comprise the largest scales of
fluid motion, and their combined effect is represented by the so-called
"mean" velocity in the  calculation process.  The remaining, small scales
of motion are regarded  as "turbulence"  and are treated in various ways.
The most common of these in air pollution modeling studies is the eddy
diffusivity hypothesis.  In this approach, the effects of small-scale
velocity fluctuations on the mean  material concentration c are param-
eterized by a semi-empirical eddy  diffusivity K in an equation of
the form:
                                   70

-------
                        || + y-vc = V-Kvc + S    ,                  (A-l)

where v is the "mean" wind vector referred to above and S is the source
strength function.

     There are numerous situations in which either accurate expressions
for K are unavailable or the eddy diffusivity hypothesis itself is
inconsistent with the observed character of material diffusion.  An
example of the former is the modeling of air pollution on scales of
1000 km or larger.  In this case, the "turbulence"  comprises eddy sizes
up to about 100 km, i.e., roughly the average separation of wind monitor-
ing stations.  An eddy diffusivity representative of this portion of the
velocity spectrum is not available.  An example of a situation in which
the eddy diffusivity hypothesis is inconsistent with the character of the
diffusion process is unstable boundary layers.  Deardorff and Willis
(1975) have shown that in free convection the vertical component K   of K,
which intrinsically is a positive quantity, would have to assume negative
values to describe the diffusion phenomena observed.

     The problems of K-theory can be circumvented by working with
Lagrangian diffusion theory.  The Lagrangian counterpart of Eq. (A-l)
for c at a specified point x and time t is expressed as follows:


             c(x,t) =  /YpU.tlxJt'Wxit1) dx1 dt1    .          (A-2)


     The difficulty with the Lagrangian approach is determining the
form of the kernel p(x,t|xjt')» which is the probability density that a
material particle released at x1 at time t1 will be found at x at time t.
Until recently, the only method of implementing Eq. (A-2) was to assume
a form for p.  The Gaussian puff and plume formulas are examples of
models derived from Eq. (A-2) under the assumption that p has the Gaussian
form (see Lamb and Seinfeld, 1973).
                                   71

-------
     Within the last few years,  numerical turbulence models have been
developed [e.g.,  Deardorff (1974)] that make it possible to derive
Lagrangian turbulence properties such as p computationally.  In these
models, which explicitly resolve large portions of the spectrum of turbu-
lent motions, particles can be released and tracked at will to obtain
the ensembles of particle trajectories required to derive Lagrangian
infornation.  Each realization of the ensemble of trajectories needed
to compute p is given by:


                 rN                                 i
      x(t;x0) =  I   uLxtt'^Qht1] + y'[x(t';x0),t']  dt'  + XQ
                JQ  '                                              (A-3)

where xn denotes the particle release point (at t = 0); u(x,t) is the
      ~u                                                ~ ~
resolvable portion of the Eulerian velocity at (x,t), given by the
turbulence model;  and u'(x,t) is a random velocity that represents the
combined effect of all fluctuations in the instantaneous velocity field
that the turbulence model cannot explicitly resolve.  In general, u'
represents all velocity variations smaller than the model's grid dimen-
sions.  This appendix describes a technique for generating stochastic
velocity fields u' suitable for use in calculating Lagrangian turbulence
properties from numerical turbulence data.  [For convenience, we shall
hereafter refer to u1 as the subgrid-scale (SGS) velocity.]

     Approximation of the kernel p can also be obtained using so-called
Monte Carlo techniques.  With these methods (see Chapter II of this report),
p  is estimated from an ensemble of particle trajectories created from
computer-generated random velocity fields whose statistics are made to
satisfy given Eulerian statistical properties of the flow, such as velocity
variance profiles, energy dissipation rate profiles, and so on.

     For generality, we developed a scheme for generating pairs of cor-
related velocity  functions u1,, u'2-  Such functions are required to
simulate the  trajectories x-,(t;x.|Q) and  Xottjx^n) of a pair of particles
                                   72

-------
simultaneously released at points x,g, X^Q located close enough together
that the motions of the particles are correlated for a time.  Particle
pair trajectory simulations are needed to determine the two particle
displacement probability density function P2(x,t;x;t'|x,Q,t,;x2Q,t2)
that a pair of particles released at the space-time points (x,n, t,),
                                                            - I U   I
(XPQ, tp) will be found at (x,t), (xjt')> respectively.  The function
p,, is required to calculate the mean square concentration c2(x,t), viz,

	            t  t
c2(x,t) =11 I I   P2(x,t;x,t|x;t1;xl,'t"}S(x;t')S(x',1t")  df dt" dx'  dx"
         JJJ0J0                                               "   (A-4)

     Shu, Lamb, and Seinfeld (1978) have used this equation in conjunc-
tion with p2 estimates, derived in the manner outlined above from Deardorff's
(1974) numerical simulation of the convective planetary boundary layer,  to
model nonlinear chemical reactions among the constituents of a point
source plume and species in the ambient air.  Lamb (1976) proposed the
concept of a macroturbulence through which historical  meteorological data
can be used to determine a kernel PO, which, when combined with Eq.  (A-4)
yields estimates of air pollution episode probabilities.

     Deardorff and Peskin (1970), Riley and Patterson (1974), Lamb,  Chen,
and Seinfeld (1975), and Lamb (19785) have computed single-particle dis-
placement statistics using Eq. (A-3) and velocity data u from numerical
turbulence models.  In all of these studies, the effect of the subgrid-
scale velocity u1 were found to be negligible, and they were generally
ignored.  While it is true that the effects of u' on the displacement
statistics of a single particle are small, if u1 constitutes only a
small fraction of the total energy spectrum of fluid motion, the subgrid-
scale velocity component u1 plays a dominant role in determining
the joint displacement statistics of a pair of particles initially
separated by a distance smaller than the dimensions of the grid on which
the model data u are available.  This is true even when u1 represents
only a small fraction of the total fluid kinetic energy.  For this reason,
the parameterization of the subgrid velocity field is essential in the
                                    73

-------
calculation of mean square concentration distributions.   Thus, the main
uses that we foresee for the two-particle scheme described here are in
modeling the mean square concentration by Lagrangian methods, i.e. by
using Eq. (A-4) in conjunction with Eulerian velocity data.  The scheme
will also be of value in implementing Eq. (A-2) when the available velocity
data are on such a coarse grid that the subgrid-scale portion of the
energy spectrum is significant; and it can be used as well in conjunction
with Eq. (A-2) as the basis of Monte Carlo diffusion models.

2.   DESIGN OF THE PARAMETERIZATION SCHEME

     The basic approach to developing an approximation for the subgrid-
scale particle velocities u, and u' is to formulate a rule for generating
random functions such that the statistical properties of those functions
are equivalent to those that the SGS velocities are known to possess.
                                                      7
For example, the mean square subgrid-scale velocity u'  is usually known.
Thus, the random functions chosen to represent u\ and u~ must satisfy
-^    —j   — *-                                ~i     -t.
    =     =   1
 n  = u,-,  = u1  .   The Lagrangian integral  time scale of the SGS velocities
and certain mean values of particle displacements that they induce may
also be known.   In this case,  the selected random functions u-j  and ui
must have statistical properties that are consistent with all  of these
known quantities.   We outline below the SGS velocity and displacement
statistics that are generally known and that serve as constraints on the
formulation of the parameterization scheme.  For simplicity, we treat
only the x-component of the SGS velocity and displacement and ignore cross
correlations among the three vector components.  Extensions of the scheme
to more general situations are straightforward.

a.   Constraints on the Subgrid-Scale Particle Displacements

     By definition the mean subgrid-scale velocity u' is zero locally;
and although it does not necessarily follow from this definition that
the mean SGS particle displacements are zero, nevertheless, we assume
that
                                    74

-------
                       = x]0    ,     x2(t) = X2Q    ,              (A-5)
where x, Q and X^Q denote the release points of particles 1 and 2.
This assumption will generate an error roughly proportional to the mag-
nitude of the shear in the deterministic flow speed component u at the
points X,Q, X~Q of release.  The mean square SGS particle displacement
and separations can be expressed in the following forms:
                         *(t) = o2(t) + x2Q    ,                  (A-6a)
                              = o2(t) + x2     ,                  (A-6b)
(t)]T= *2(t)  + (x1Q - x2Q)2    .           (A-7)
                    - x2
Implicit in Eq. (A-6) is the assumption that x,n and xon are sufficiently
                     p       n       ij        I U      f.\J
close together that o (t) = a0(t) = a (t) .  For the moment, we assume
      22
that a (t) and s. (t) are known functions.

     To specify the random displacements  x-, and x2 uniquely, we would
 have  to specify the entire infinite hierarchy of joint moments of
these variables.  In practical situations, at best only the  first-  and
second-order moments are known.  Thus, in developing the SGS parameter-
ization, it suffices to incorporate only  Eqs. (A-5) through (A-7)
as constraints.

b.   General Form of the Displacement Parameterization

     Let x  be the displacement of the n-th particle of a pair (n = 1, 2)
produced by our simulation of the turbulence velocity u-' .  The constraints
of Eqs. (A-5) through (A-7) on the statistics of x  can be satisfied
if xn is of the following form:
                                    75

-------
               xn - xnQ = as + bxn     >     n  =  1, 2     ;            (A-8)
where a and b are functions of travel  time and £  and  x   are  computer-
generated random numbers.   To satisfy  Eq.  (A-5),  we must have
                                 xn = 0    .                        (A-9)
It is also convenient to have 5  and xn statistically independent  and Xi
and X?  independent of each other.   In this case,

                      cxn =  exn = o = 7jx^    .                   (A-io)

To satisfy Eq. (A-6), we need
and to satisfy Eq. (A-7), we must have
2 -  x   - x2 = b22 + b22  = a2
           (x1 - x2)  - (x1Q - x2Q)   = bx  + bx  = a

 If we make
           = o

                                  2                              (A-13a)

                                      a2    ,                    (A-135)
then it follows that Eqs. (A-5) through (A-7) will be satisfied by
Eq. (A-8) if, in addition to Eqs. (A-9), (A-10), and (A-13)  we select
a and b such that
                                   76

-------
                          a2 = i  -h?/2o2 I    ,                    (A-14a)
                          b2 = 1  -  a2    .                        (A-14b)

c.   General  Form of the Velocity Parameterization

     The algorithm for generating u'  is  obtained by differentiating
Eq. (A-7):

                      u; = a| + ca  +  bxn +  X{f>    .                (A-15)

With a and b given by Eqs. (A-14a)  and (A-14b),  respectively,  and £;  and
x  satisfying conditions in Eqs.  (A-9),  (A-10),  and (A-13),  Eq.  (A-15)
constitutes the basis of our parameterization of the subgrid-scale 	
                                                            2        2
velocity field.  Next, we must specify the  forms of £,  x >  o , and £ .

     In view of conditions in Eq. (A-13), our choice of forms  for c  and
x  will be guided by our knowledge of a, which until now we have regarded
as a given function.  In actuality, we have very little information  about
the form of o.  We  know only that for small times,

                  a2(t) = u'2t2    ,     small t    ,              (A-16)

      —2
where u1  is the mean square subgrid-scale  velocity fluctuation  and  t is
travel time.  The behavior of a at large times is not known.  Lamb (1971)
showed that if the turbulence is stationary and the Lagrangian velocity
spectrum is a band-pass spectrum, similar to that which the Eulerian
spectrum of the SGS velocity is known to be, then in the limit of large
travel times the mean square particle displacement a approaches  a constant
value.  This is in marked contrast to the effect that a white  noise
                                         2
velocity spectrum would produce,  namely cr  ~. t in the limit of large travel
times.  Whether the SGS velocities  produce  a constant,  a linearly increas-
                                            2
ing, or an altogether different dispersion  o  in the limit  of  large  travel
                                    77

-------
times is not known, but we expect that the details of the variations in
the long-time regime are not crucial.   Even if particles are released near
a boundary where the SGS velocities constitute a significant fraction of
the total turbulence, by the time the  dispersion enters  the long-time
regime the particles will  have moved into a region of the fluid where
the resolvable scale, rather than the  SGS, motions dominate the particle
displacements.  Proceeding under this  assumption, we adopt expressions
for the velocities £ and xn that are consistent with Eqs. (A-16) and (A-9)
and with estimates of the Lagrangian integral time scale T of the SGS
velocity field.  The long term statistical behavior of £ and xn that
result from these specifications will  be regarded as best estimates of
o2  [cf. Eq. (A-13)].

     In the context of the Monte Carlo type of models of diffusion, the
previous comments do not apply because there u' represents the entire
spectrum of turbulent motion.  In this case Eq. (A-16) holds for small t
     2
and o  ~ t applies in the limii
replicates both these regimes.
     2
and o  ~ t applies in the limit t -> ».   The scheme described here
     Since the statistical properties of £ and xn are identical, for
simplicity we treat only 5 in the following discussion.   A suitable
expression for | is the Markov-1 form

                     c(tn) = txc
where t  = nAt, At is the time step used in the diffusion simulation,
a and 6 are constants, and p (t ) is a random variable with the following
properties:

                              = 0                                (A-18a)

                              =0    ,    n f m    ,             (A-18b)

                              =0    ,    n f m    .             (A-18c)
                                    78

-------
The first of these is necessary to satisfy Eq. (A-9).

     Through straightforward analyses, we find that with £ given by
Eq. (A-17), its autocorrelation is
and its mean square varies temporally as
                  ?(tn) = c^U^) + B2P2(tn)     .              (A-20)
Consider for the moment the case  of stationary,  homogeneous  turbulence:

                     C2(t)  = u'2(t) = Constant    ,                (A-21)

      —2
where u'  is the mean square SGS  velocity fluctuation.   To ensure  that
Eq. (A-21) is satisfied when Eq.  (A-17)  is used  to describe  particle
velocities in stationary,  homogeneous turbulence, we  must set  [see
Eq. (A-20)]
                           a2 + B2 =  1     ,                        (A-22)
                             p|(t)  = u'2                          (A-23)

and the initial  values of \ must be selected such  that

                                           .                      (A-24)
     The Lagrangian integral  time scale  T  of I,  which  is  defined  by
                                    79

-------
                         u'   m=0

acquires a particularly simple value in  stationary,  homogeneous  turbulence,
namely [see Eqs.  (A-19) and  (A-21)],
                     T, = At     am = 7--                       (A-26)
                             ^^      I  - a
                             m=0
Thus, in applications to stationary,  homogeneous  SGS  velocity fields  with
prescribed values of T  and u1  ,  the  parameterization scheme must satisfy
for given At:

                         a = 1  -  f--    ,                           (A-27)

                                   o  1/2
                         6 = (1  - a )       ,                      (A-28)

and p  and e(0) must satisfy Eqs. (A-23)  and (A-24).

     In many cases of applied interest, the SGS field is stationary but
not homogeneous.  In this case,
                    r(tn) f r(tm)    ,     n t m    .             (A-29)
     If we were to apply the parameterization scheme to problems of this
type, we could continue to use Eqs. (A-24), (A-27), and (A-28)  to specify
£(0), a, and g, but we would have to require that p  satisfy [see
Eq.  (A-20)]
                 -y
                                    80

-------
        2
where u1 (x ) denotes the mean square SGS velocity at the location x
of a particle at time t = mAt on any given realization.   Placing this
condition as a constraint on the random variable p  is undesirable because,
during the execution of the scheme on the computer, Eq.  (A-30) would
require a square root calculation, which is time consuming; and in
regions of highly inhomoyeneous SGS fields, Eq.  (A-30) might require
                       2
(for a given At) that p  < 0.  For these reasons, we adopt the following
approximation:
                                = u'2(x)
It can be shown that the error incurred through the use of this approxima-
tion in inhomogeneous turbulence is negligible if
                       At «
vu'2    II      .                   (A-32)
                                                   2         2/3
For example,  in the unstable surface layer where w1   ~ (Z/L)     [see
Panofsky et al. (1977)], the right side of this expression has a  value
of about 100 sec at z = 10 m when l = -25 m and u* =  0.1  m/sec.   In short
Eq. (A-17)  can be applied to inhomogeneous SGS velocity fields provided
that the time step At used in this equation satisfies Eq. (A-32)  where
u'2 denotes the mean square SGS velocity as a function of space.

     When the parameterization scheme is implemented, a random number
generator is used to produce values of p .  The mean  and variance of this
function are specified by Eqs. (A-18) and (A-31), respectively.   It
should be kept in mind that the right-hand side of Eq. (A-31) is a known,
local property of the SGS velocity field.

     A Gaussian random number generator can be used to produce P£ in the
Monte Carlo type of models applied to neutrally stratified flows  because
                                   81

-------
turbulent velocity fluctuations are known to be normally distributed in
this case (Batchelor, 1953).  The Gaussian generator should also be appro-
priate  in simulating subgrid-scale turbulence in both neutrally and unstably
stratified flows provided that the grid size is much smaller than the
scale of the energy-containing eddies.  However, the joint distribu-
tion of turbulence velocities at two sufficiently close points is not
normally distributed under any conditions (Batchelor, 1953).  Thus, the
use of jointly normal random number generators in the parameterization
scheme will  result in errors in the joint statistics of the particle
pair displacements, but this error is probably no larger than that
incurred from other assumptions in the scheme.

     Generating random numbers  with a Gaussian  distribution requires more
computer time than  that needed  to  generate uniformly distributed random
numbers.  Therefore,  it is  advantageous  to use  a uniform random number
to index a table of numbers  configured in such  a way that the numbers
retrieved from the  table have a Gaussian distribution with zero mean
and unit variance.   This technique requires less than half of the machine
time required to generate Gaussian numbers using conventional  algorithms.

     In conclusion, l(t ) is given by Eq. (A-17) with a and g
specified by Eqs. (A-27) and (A-28), respectively.  The random vari-
able p  has  a Gaussian distribution with zero mean and variance given
by Eq. (A-31).   If the time step At satisfies Eq. (A-32), Eq.  (A-17)
can be applied to inhomogeneous SGS velocity fields.

     The equivalence of the statistical  properties of \. and xn suggests
that an expression similar to Eq.  (A-17) should be suitable for producing
xp.  In particular, if
                                   82

-------
where a and B are given by Eqs.  (A-27)  and (A-28),  respectively,  and

                        pp(tm) = u'2(xm)    ,                      (A-34)

                         Ao) = So)  = u'2(x J     ,             (A-35)
                                              n0
then all of the conditions imposed on xn» namely Eqs. (A-9), (A-10),  and
(A-13b), are satisfied.  Note that T  = T  = T, where T is the Lagrangian
                                    X    s>
integral time scale of the SGS velocity.  The expressions for c and xn
follow immediately from those of £ and x :

                                    n-1
                                                                  (A'36)
                     c(tn) =
                                    m=0
                                     m-1
                    xn(tm) = xn(0) +xn(tk)At    .              (A-37)
These equations together with Eqs. (A-15), (A-17), and (A-33) constitute
the basis of our parameterization scheme.  All that remains to render
these equations operational is to express the functions a and b in terms
of  available quantities and  to  check whether the scheme  produces  the
proper  particle behavior near reflective boundaries.

d.   The  Velocity  Parameterization at  Reflective Boundaries

     The  Markov form of the velocity parameterization we have chosen imparts
a "memory"  of  approximate  length T to  the motion of each particle.  Conse-
quently,  a  particle that has a  positive velocity,  say, just prior to its
striking  a  reflective boundary, is more  likely to have a positive velocity
after impact than  a negative one; and  hence  it is more likely to strike
the boundary a second time than to move  away.  As a result, the particle
will linger at the boundary.  This phenomenon is manifested in  diffusion
 simulations by fictitiously  large  particle concentrations  at  all  reflec-
 tive  surfaces.
                                   83

-------
     Actually,  turbulent eddies that carry particles into a wall  are most
likely dissipated by viscous forces near the wall  and,  consequently, the
memory they impart to particles is lost.  We can approximate this effect
in the parameterization scheme by modifying the expression for £, l»
X , and xn as follows:
                                    if reflection
               _  = Pf(tJ      >   occurred during    ,            (A-38)
                     5
where p  and p  are Gaussian random numbers with zero means and variances
._m     ^  - - -  M            ^j~
P^ = p2 = u'2(x ), where u' (x_) denotes the mean square SGS velocity at
the particle's location at time t.  Equation (A-38) is used only after
reflection.  At all other times, Eqs. (A-17), (A-33), and (A-37) are
used.  Tests of this boundary modification have shown that it eliminates
the erroneous accumulation of particles at the boundaries.

e.   Specification of the Parameters a and b

     When two particles are separated by a distance that  is larger than
the  largest turbulent eddies, the random motions of those particles are
                                                     ~~2~    22     2
uncorrelated.  In  this case, it is easy to show that I  = cr, + 0o (= a
in the present situation).  At the other extreme where the particles are
released at virtually the same time  and locations,_they never separate
                                                    2
(under the action  of turbulence alone), and  thus,  a  remains  zero for all
time.  These  two limiting cases indicate that a, as given by Eq. (A-14a),
ranges in value from unity for  particles  whose  motions  are totally  cor-
related  to  zero  for those whose velocities are  statistically  independent.
These observations  indicate that  in  the limit of large travel times, as
the  mean particle  separation becomes large compared with  the  grid size,  the
function a approaches zero.  In order to determine  the initial  value of
a  and the manner  in which  it approaches its  final  value,  we need explicit
                  P    —»j-
expressions for oc and v-.

                                   84

-------
              2
     Like a, £  is a parameter about which we have little information.
Using Kolmogorov's similarity hypothesis,  Batchelor (1950)  showed that
if the initial separation £Q of two particles lies in the inertial  sub-
range, then for small times:
                                                          ,         (A-39)

where c-,  is a constant on the order of 1, G is the energy dissipation rate
of the turbulence, and t is travel  tine.   In  Deardorff's  (1974)  turbulence
model, A = (AxAyAz)    (where Ax, AU,  and AZ  are the grid cell dimensions)
lies in the inertial  subrange; so Eq.  (A-39)  should apply for all  ju  less
than A but large enough that viscous effects  are unimportant.

     Deardorff and Drake (1975) suggest the semi -empirical formula

                                                                   (MO)
           G = 1 + -, - = -        ,    zj.4r    ,          (A-41)
where z is elevation above ground and E is the S6S kinetic energy.  By
combining Eqs. (A-14a), (A-16), (A-39) , and (A-40) and by assuming iso-
tropy of the SGS velocity field, i.e., u'2 = 2/3E, we obtain

                                       p/-j/4n\2/3
                aQ = a(t0) = 1 - (0.7Gr J(Y)       .             (A-42)

For convenience, we  have  chosen c-| =  2/5.   It is  interesting that the
initial value of a is independent of E.  The indicated dependence of aQ
on £Q is consistent with the fact that aQ -»• 1  as ZQ -*- 0, but Eq. (A-42)
cannot be valid for  ju » A because it yields large negative values for
a.  Although it is possible that a could achieve a small negative value,
in our applications  of the parameterization scheme we assume that
0 i a i 1.
                                   85

-------
     In addition to deriving Eg.  (A-39),  Batchelor obtained  an  expression
for the temporal variation of fc2  during the intermediate  time period
      2/3  -1/3
t » IQ  6  '   with t small enough  that the particle separation  is small
compared with the scale of the energy-containing  eddies of the  turbulence.
We could utilize this information to  help describe the behavior of a(t)
between its initial and final values,  but this  approach would be  of very
limited use.  Because it is applicable only in  the range  £Q  «  £2    « A,
this additional information would be  of value only when JL, is several orders
of magnitude smaller than A.  Rather than piece together  an  expression
for a(t), we adopt a single, ad hoc  formulation that is consistent with
available information.

     Specifically, we note from Eq.  (A-42) that the slope of a(t) is  zero
                                 2/3  -1/2
at t = 0.  Also, we expect that A   e     must  be proportional  to the
characteristic  time scale of a(t).  This information and  the known initial
and final values of a(t) can all  be  taken into  account by the following
simple formulas:

                      a(t) = a  exp  (-A)    ,                     (A-43a)
                           •  /*    *'
                             ^0 T2(t')
A =  /  -?r—	    ,                    (A-43b)
         •(t1) dt1
                      T(t) = -yr^—     ,                          (A-43c)
                             EV2(t)
where E(t) is the SGS kinetic energy at the centroid of the particle pair
at time t, and  X is a constant.  Observational studies of mean square con-
centration fluctuations c? in turbulent pipe flows indicate that cVc2
is self-similar over a considerable region downstream of the source.
Lamb  (1977) speculates that this is a characteristic feature of dispersion
in decaying turbulence and has shown that it implies that £2/a2 = constant.
[To simulate this condition with our parameterization, we would need
a = constant, which in turn would require that the constant in Eq. (A-43c)
have  a very large value.  In a future study, we plan to investigate the
value of  x in some detail.
                                   86

-------
     If the scheme developed here were used in a  Monte  Carlo  model  of  dif-

fusion in the planetary boundary layer,  in place  of Eq.  (A-40)  one  would

need the c(z) profile appropriate for boundary layer turbulence (see Wyngaard

and Cote, 1971); and in place of u'2 = 2/3E,  one  might  use  the  u'2(z)
                                                      f
profiles given by Panofsky et al.  (1977).   This approach  is described  in

this report in Chapter IV.



f.   Recapitulation



     For convenience, we summarize below the complete parameterization

scheme for the two-particle subgrid-scale velocities ul and uj>.
                                          n = 1,  2   ;
                             if a reflection of either particle
                             occurred during t  ,  <. t < t ;
+ &p?(tm)
                                         otherwise.
                        if reflection of either particle  occurred
                        during t    < t <
              "n".'
      if a reflection of particle n
      occurred during t  -j  <. t < tm;
                                          otherwise.
                        if reflection of particle n occurred
                        during t  , <. t < t ;
                                   87

-------
Initial  values are c(0)  = PC(O),  c(0)  = 0,  *n(0)  = Pn(0),  and xn(0)  = 0.'
Here,  Pn(t)  (n = 1,  2) and pF(t)  are Gaussian random numbers with zero
                      — — .    ^
means  and mean square p|(t), and  so on equal  to the mean square subgrid-
                 o"     ^
scale  velocity u1^ at the location of the particle at time t.  The param-
eters  a and 6 are given by Eqs.  (A-27) and (A-28), a and a are given by
Eq. (A-43),  and b and 6 are specified by Eq.  (A-14b).

3.   RESULTS OF TEST CALCULATIONS

     In Figure A-l,  we show that  the scheme gives the correct temporal
variation in the rms separation of particle pairs when we set the param-
eter a(t') = an = constant [see Eq. (A-14a)].   In this case Eq. (A-14a)
      ~~?    9
gives  IL ^ a .  This relationship is clearly satisfied by the rms particle
separation values shown in Figure A-l  that were determined from the
simulation.   Also, by comparing the curves  given in Figure A-l for the
case aQ = 0, i.e., uncorrelated motion of the particle pairs, with the
a  profiles shown in Figure 6 (Chapter III),  one can see that the param-
 A
eterization scheme reproduces correctly the limiting relationship
                         lim &2(t) = 2o2(t)
                         t -*-"0
The reader is referred to Chapter III for results of single particle
displacement statistics and concentration fields simulated using the
technique developed here.

4.   CONCLUSION

     We have developed a method of generating pairs of random velocity
fields suitable for simulating particle pair motions in turbulent fluid.
The scheme can be used to simulate the effects of subgrid-scale turbulence
in situations in which one is interested in determining Lagrangian tur-
bulence statistics from instantaneous velocity fields obtained from a
numerical turbulence model; or the scheme can be used as the basis of a
                                   88

-------
CO
                        100
                         9
                         a
                         7
                         t
                   CM
                   r- __ X
                    CO O
                     «*
                      I
                         71*1
                                            3   4  S 6 7 8 9 K>
                                                                          3   4  5  6 7 S 9100
                         FIGURE A-l.  STREAMWISE COMPONENT  (a2  -  S-Q)l/2 OF THE rms SEPARATION OF
                                      SIMULATED PARTICLE  MOTIONS  UNDER THE CONDITIONS u = 3.0,
                                      u7^ =  1, AND v7^ -  0.6 (MKS UNITS) FOR TWO VALUES OF THE
                                      PARAMETER aQ

-------
Monte Carlo model  of turbulent dispersion,  in situations  in  which  one
attempts to determine Lagrangian turbulence statistics  from  Eulerian
statistics.  The inputs to the scheme include the profiles of mean wind
and variances of the turbulence velocities, the Lagrangian integral time
scale, and the energy dissipation rate of the turbulence.  With very simple
modifications, the scheme can be used to simulate buoyant particles,
particles with nonzero settling velocities, and particles that are absorbed
or resuspended from boundary surfaces.  We are using this scheme in all
of the roles described here.
                                   90

-------
                 APPENDIX B
USER'S GUIDE TO THE LAGRANGIAN HIGHWAY MODEL

-------
                              APPENDIX B
          USER'S  GUIDE  TO THE LA6RANGIAN HIGHWAY MODEL


     The theoretical  aspects  of  the highway model are discussed in the main
text of this document.   This appendix presents a short discussion  of the
use of the model.   Although the  computer program was coded using American
National Standards  Institute  (ANSI) FORTRAN IV, there are some machine-
dependent portions  that  the user must check to ensure that the program will
work properly on his  computer system.  These are discussed in Section 3.

1.   DESCRIPTION OF THE  INPUT CARDS

     Five sets of input  cards must be used to  run the computer program:

     >  Overall  control  cards
     >  Monitor site  cards
     >  Meteorological data cards
     >  Source location  card
     >  Miscellaneous data cards.

These sets are discussed in  turn below.

a.   Overall Control  Cards

     Four control cards  must  be  placed at  the  beginning of the input deck.
Card No. 1 contains the  number of simulations  to be performed.  Two param-
eters are entered on this card:   the first is  the number of the first
experiment (usually 1),  and the  second is  the  number of the last experi-
ment.  Thus the user wishing  to  perform  three  simulations must specify
1 and 3 in Columns 5 and 10 on Card 1.   The two variables are entered as
five-digit integer variables. The format  for  the Card 1 is given in
Table B-l.
                                     92

-------
                       TABLE  B-l.  FORMAT FOR CARD 1
           Variable    Column     Format    	Description
            NSTART      1-5         15      The  number of the
                                          first simulation
            NSTOP       6-10       15      The  number of the
                                          last simulation
     Card 2,  the second card in the input deck,  contains  the title of the
simulations.   The title may be up to 72  characters.

     Cards 3  and 4 in the input deck contain  21 variables (12 on the
first  and 9  on  the  second).   Each  variable and the format for that var-
iable  are described  in  Table  B-2.

b.   Monitor  Site Cards

     The calculated PI  and P2 kernels (and ultimately  the concentration
of pollutant  at each monitor site)  are stored for  each monitor  site.
Therefore a simulation is not performed  unless  there  is at least one monitor
site.  The maximum number of monitors is 100.   The format for the monitor
site cards is given in Table B-3.  One card is  entered for each monitor
site, and these cards are placed after Card 4.   Although  the program can
handle up  to  four lanes,  the  computer cost increases with the number of lanes.
If the lanes  are located  symmetrically  about the  center of the road-
way,  the user can create pseudo-stations to reduce the number of lanes  to
be treated.  (This procedure is discussed in  the main  text.)  The pseudo-
sites will monitor the PI and P2 calculations for  the  extra lanes.

c.   Meteorological  Data Cards

     Meteorological  data are read from two cards.   The first, Card 6A,
is described  in Table B-4.
                                    93

-------
ZMAX



STRTIM


ENDTIM
            TABLE  B-2.   FORMAT  FOR  CARDS 3 AND 4

                        (a)   Card 3
Variable    Column     Format
                          Description
LFILE
LGRID
LPRNT
NPLT
LMET
NDATA
NX
NZ
XMAX
1
2
3
4
5
6
11-15
16-20
21-25
11
11
.11
11
11
11
15
15
F5.0
Option to save data on an
external file
Option to print calculated
results on a grid
Option to print input data
Option to plot particle
trajectories
Option to print calculated
meteorological data
Flag for comparison of
calculated data and
observed data
Number of cells in the
x-direction
Number of cells in the
z-direction
Length of the region in
26-30



31-35


36-40
F5.0



F5.0


F5.0
the x-direction (in
meters)

Length of the region in
the z-direction (in
meters)

Starting time of the
simulation
Ending time oi
simulation
the
                              94

-------
                TABLE B-2  (Concluded)

                     (b)  Card 4
Variable    Column    Format
                           Description
NMON

NSAMP



NITER
  1-5

 6-10



11-15
15

15



15
NRLMX
TOL
SMPDIM
SEPR
MONSGS
MONWAK
16-20
21-30
31-40
41-50
54
55
15
F10.0
F10.0
F10.0
LI
LI
Number of monitors

Sampling number of conver-
gence test (recommended
value is 7)

Number of 100 iterations
used for sampling (recom-
mended value is 7)
Maximum number of itera-
tions for each simulation

Error tolerance at each
monitor site

Sampling dimensions of
each monitor

Separation between the two
paired particles

Option to activate print-
ing of particle statistics

Option to activate print-
ing of particle statistics
caused by the wake
turbulence
                              95

-------
  TABLE B-3.  FORMAT FOR THE MONITOR SITE CARDS  (CARDS  5}
Variable
NAMMON(I)

XMON(I)

ZMON(I)
Column    Format
                  Description
   1-4

  6-15

 16-25
   A4

F10.2

F10.2
Name of the i-th monitor (if
any
Location of the i-th monitor
in the x-direction
Location of the i-th monitor
in the z-direction
TABLE B-4.  FORMAT FOR THE FIRST METEOROLOGICAL  DATA CARD (CARD  6A)
Variable
Z0
ZI
ANEHMT
OBUKOV
USTAR
WNDSPD
WNDANG
Col umn
1-5
6-10
11-20
21-30
31-40
41-50
51-60
Format
F5.0
F5.0
F10.1
F10.1
F10.1
F10.1
F10.1
DTDZ
 61-70
F10.1
	Description	
Surface roughness
Inversion height
Anemometer height
Monin-Obukhov length
Friction velocity
Mean wind speed
Mean wind angle perpendicular
to the roadway
Vertical temperature gradient
                               96

-------
     The user may enter experimental  data to compare with model  predictions
by including Card 6B (see Table B-5)  and an external file called TAPE 10.
The meteorological parameters read from the external file will  replace
the meteorological parameters read on Card 6A.

d.   Source Location Card

     The location and description of the roadway is entered on Cards 7,
after the meteorological data.  The format for Cards 7  is  given  in  Table B-6.
All distances and heights on Cards 7  are entered  in units  of meters.

e.   Miscellaneous Data Cards

     Three cards are required for miscellaneous data, as described in
Table B-7.  The data entered on Card 7 control the simulation time.
Card 9 contains parameters that describe the wake turbulence profile.
Card 10 contains descriptions of the vehicles found on each lane.

2.   SAMPLE INPUT DECK

     A sample input deck is presented in Table B-8.  In this example,
the input data are those used in the simulations of the GM experiments.
Users may wish to use these data for their initial simulations.

3.   COMPUTER-DEPENDENT PARAMETERS AND SUBROUTINES

     The Highway  Program requires approximately 44,000 decimal words to
load on the CDC 7600 computer.  This number may vary among computer
systems, and some users may find the program too large for their system.
However, certain  routines that handle input-output options can be elim-
inated by making  them dummy routines.  Consider, for example, routine
SGSMON, which is  coded as follows:
                                   97

-------
TABLE B-5.   FORMAT FOR THE SECOND METEOROLOGICAL DATA CARD (CARD 6B)
     Variable
      Column    Format
                      Description
(TITLE(I), 1=12,14)
OBUKOV
USTAR
WNDSPD
WNDANG
DTDZ-
4-15
16-25
26-35
36-45
46-55
56-66
3A4
F10.2
F10.2
F10.2
F10.2
Ell. 2
Title of the experiment
Observed Monin-Obukhov
length
Observed friction
velocity
Observed wind speed
Observed mean wind angle
perpendicular to the
roadway
Observed vertical tem-
perature gradient
 Next  card:

 (STREN(I),I=1,4)


 Next  cards:

 (SF6(I),1-1,20)
        1-44    4E11.2
                Source strength from
                each lane of the roadway
         1-72     9F8.3    Observed concentrations
                          (up to nine on each card)
    TABLE B-6.   FORMAT  FOR THE SOURCE LOCATION CARD (CARD 7)
   Variable    Column    Format
                          Description
    NLANE


    LAN1LO



    DELTAL


    TRUCKH


    AUTOH
 1-10
11-20
  110
F10.1
21-30     F10.1


31-40     F10.1


41-50     F10.1
Number of lanes on the
roadway (maximum = 4)

Location of the first lane
(in meters from the
origin)
Distance between the cen-
ters of the lanes

Height of source emis-
sions from a truck

Height of source emissions
from an automobile
                              98

-------
      TABLE  B-7.   FORMAT  FOR  THE  MISCELLANEOUS  DATA  CARDS
                  (CARDS  8, 9,  AND 10)
                          (a)  Card 8
Variable
DELTAT
BETA
TIMSCL
A0

Variable
TIMWAK
A0WAKE
BETAWK
UWKCOF
WWKCOF

Variable
VEHWD(I)
VEHSPD(I)
Col limn
1-5
6-10
11-20
21-30

Column
1-5
6-10
11-20
21-30
31-40

Column
1-5
6-10
Format
F5.0
F5.0
F10.1
F10.1
(b)
Format
F5.0
F5.0
F10.0
F10.0
F10.0
(c)
Format
F5.0
F5.0
Description
Time step (in seconds)
Time decay multiplier of A0
Lagrangian integral time scale
Boundary layer initial par-
ticle separation coefficient
Card 9
Description
Lagrangian integral time scale
of the wake component
Boundary layer initial par-
ticle separation coefficient
due to the wake turbulence
Time decay multiplier of A0WAKE
Coefficient of the u-component
of the wake turbulence
Coefficient of the w-component
of the wake turbulence
Card 10*
Description
Width of the vehicle in the i-th
lane
Speed of the vehicle in the i-th
DRGCOF(I)    11-20     F10.1
lane

D!~ag coefficient of the vehicle
in the i-th lane
* One Card 10 is entered for each lane specified in Card 7.
                              99

-------
TABLE 8-8.   SAMPLE INPUT DECK USED  IN THE GM SIMULATIONS
      1   63
                          GM EXPERT MEM* WO.
  111111      103   30330.  90.1000.1100.
     69    7  100 3000     .03       3.0       0.0     FF
  11        104.4      9.38
  12        104.4      3.31
  13        104.4      0.31
  21        132.3      9.50
  22        132.3      3.63
  23        132.3      0.36
  31        147.1      9.38
  32        147.1      3.03
  33        147.1      0.36
  41        163.6      9.63
  42        163.6      3.61
  43        163.6      0.31
  51        174.8      9.30
  52,        174.8      3.48
  33        174.8      0.36
  61        189.8      9.60
  62        189.8      3.84
  63        189.8      0.58
  73        209.8      0.36
  83        259.8      0.36
  IIP       101.0      9.38
  12P       101.0      3.31
  13P       101.0      0.31
  21P       129.1      9.30
  22P       129.1      3.63
  23P       129.1      0.56
  3 IP       143.7      9.38
  32P       143.7      3.03
  33P       143.7      0.36
  41P       160.2      9.63
  42P       160.2      3.61
  43P       160.2      0.31
  31P       171.4      9.30
  32P       171.4      3.48
  33P       171.4      0.36
  6 IP       186.4      9.60
  62P       186.4      3.84
  63P       186.4      0.38
  73P       206.4      0.36
  83P       236.4      0.36
  41         83.8     9.38
  42         85.8     3.51
  43         83.8     0.31
  444       113.9     9.3
  434       113.9     3.6
  464       113.9     0.6
  32        136.0     9.5
  33        136.0     3.3
  33        156.0     0.3
  60        241.2     0.5
  44        82.4      9.58
  44        82.4      3.3
  43        82.4-      0.3
  64        110.3     9.3
  63        110.3     3.6
  66        110.3     0.3
  67        125.1     9.3
  68        123.1     3.5
  69        123.1     0.3
  70        141.6     9.3
                               100

-------
              TABLE B-8 (Concluded)
69
                CM EXPERIMENT JO.
111111
69
61
61
63
64
6-3
66
67
68
80
.03

0.8
0.0
2.3
100 30 330. 90
7 100 3000
141.
141.
132.
132.
132.
167.
167.
167.
237.
300.
1
1.0
0.8
22.2
6
6
8
8
8
8
8
3
8
10.3
136, 1
30.
1.0
.3
3.3
0.3
9.3
3.3
0.3
9.3
3.5
0.3
0.3



0

.08









-2.72
Id. 6
0.
.008

                           3.0       ».0     FT
                            .20       1.9      2.o2      .0028
                             1.         1.         .8
                         0.008
                         101

-------
          SUBROUTINE  SGSMON
          COMMON  .  .  .
          RETURN
          END

This routine can be made into a dummy by simply eliminating  the code
between the second card and the RETURN card:

         - SUBROUTINE SGSMON
          RETURN
          END

     If the user is interested only in the predicted results,  the follow-
ing routines can be made into dummy routines:

          METRRT   j
          WAKMON   /  should be kept for testing purposes
          SGSMON   )
          ERPRNT
          WRITGD
          GRID
          CONVT
          PLTRAJ
          FRAME
          AXES
          MNSTAT—used only if experimental data are entered

     An option  of the Highway Program is an off-line plot (CALCOMP) of
the first 20 realizations of the first 100 particle trajectories.  If
this option  is  used, standard CALCOMP routines must be supplied by the
user:
                                    102

-------
          NUMBER
          LINE
          PLOT
          SYMBOL
          PLOTS

If the user does not have the facility for off-line plotting,  he  must
create dummy routines for each of the above.

     The Highway Program requires the use of a random number generator  to
generate a number between 0 and 1.   This is done through the use  of the
system routine for the random number generator.  On the CDC computer the random
number generator function is called RANF(0).   This  function varies  between
computer systems.  Therefore, the user must rewrite the statement

                           RANUM(X)=RANF(X)*GMAX

found in subroutines SGSSET and ADVECT.  The card should be changed to

             RANUM(X)=(user's random number function)*GMAX
                                   103

-------
                              REFERENCES
Batchelor, G.  K.  (1953), The Theory of Homogeneous  Turbulence  (University
     Press, Cambridge, United Kingdom).

           (1950),  "The Application of the Similarity Theory of Turbulence
     to Atmospheric Diffusion,"  Quart.  J.  Roy.  Meteor.  Soc.,  Vol.  76,
     pp. 133-146.

Binkowski, F.  S.  (1978), "Consistency of a Semi-empirical  Theory  for Turbu-
     lence in the Atmospheric Surface Layer," Atmos.  Environ.,  in press.

Cadle, S. H., et al.  (1976), "Results of the GM Sulfate Dispersion
     Experiment,"  GMR-2107,  General  Motors Corporation,  Warren, Michigan..

Chock, D. P.  (1977),  "General Motors Sulfate Dispersion Experiment:
     Assessment of the EPA HIWAY Model," J. Air Poll.  Contr.  Assoc., Vol. 27,
     No. 1, pp. 39-45.

Danard, M. B. (1972), "Numerical Modeling of Carbon Monoxide  Concentrations
     Near Highways,"  J.  Appl. Meteor.,  Vol. 11, pp. 947-957.

Deardorff, J. W.  (1974), "Three-Dimensional Numerical  Study of  the Height
     and Mean Structure of a Heated Planetary Boundary Layer,"  Bound.  Layer
     Meteor., Vol. 7, pp. 81-106.

Deardorff, J. W., and M. A.  Drake (1975), "Boundary Layer  Data  from a  Numer-
     ical Integration in Three Dimensions," National  Center for Atmospheric
     Research, Boulder,  Colorado.

Deardorff, J. W., and R. L.  Peskin (1970), "Lagrangian Statistics from
     Numerically Integrated Turbulent Shear Flow," Phys. Fluids.  Vol.  13,
     pp. 584-595.

Deardorff, J. W., and G. E.  Willis (1975), "A Parameterization  of Diffusion
     into the Mixed Layer," J. Appl. Meteor., Vol. 14, pp. 1451-1458.

Draxler, R. R. (1976),  "Determination of Atmospheric Diffusion  Parameters,"
     Atmos. Environ.. Vol.  10, pp. 99-105.

Egan, B. A., A. Epstein, and M.  Keefe (1973), "Development of Procedures To
     Simulate Motor Vehicle  Pollution Levels," Department  of  Highways  and
     Traffic, Washington, D.C. (NTIS PB-233-938).
                                   104

-------
Eskridge, R. E., and K. L. Demerjian (1977), "A Highway Model  for the
     Advection, Diffusion and Chemical  Reaction of Pollutants  Released
     by Automobiles," Preprint Volume,  Joint Conference on Applications
     on Air Pollution Meteorology, American Meteorological  Society,
     29 November-2 December 1977, Salt  Lake City,  Utah.

Lamb, R. G. (1978a), "Mathematical Principles of Turbulent Diffusion Modeling,"
     Lectures presented at the International School  of Atmospheric Physics,
     13-27 February, 1978, Erice, Sicily, Environmental Protection Agency,
     Research Triangle Park, North Carolina, in press.

	 (1978b), "A Numerical  Simulation of Dispersion from an Elevated
     Point Source in the Convective Planetary Boundary Layer," Atmos.
     Environ.,  Vol.  12,  pp. 1297-1304.

	 (1977), "Continued Research  in Mesoscale Air Pollution Simulation
     Modeling--Vol. VI:  Further Studies in Modeling Microscale Phenomena,"
     EF77-143, Systems Applications, Incorporated, San Rafael, California.

	 (1976), "The Calculation of  Long-Term Atmospheric Pollutant
     Concentration Statistics Using the Concept of a Macro-Turbulence,"
     Preprint Volume, Third Symposium on Atmospheric Turbulence, Diffusion,
     and Air Quality, American Meteorological Society, Raleigh, North
     Carolina.

	 (1973), "Note on the Application of K-Theory to Diffusion
     Problems Involving Nonlinear Chemical Reactions," Atmos.  Environ.,
     Vol. 7, pp. 257-263.

	 (1971), "Numerical  Modeling  of Urban Air Pollution," Ph.D. Thesis,
     University of California, Los Angeles, California (available from
     University Microfilm, Ann Arbor, Michigan).

Lamb, R. G., and J. H. Seinfeld (1973), "Mathematical Modeling of Urban
     Air Pollution:  General Theory," Environ. Sci.  Techno!.,  Vol. 7,
     pp. 253-261.

Lamb, R. G., and VJ. R. Shu (1978), "A Model of Second-Order Chemical
     Reactions in Turbulent Fluid," Atmos. Environ.,  Vol.  12,  pp. 1685-1694.

Lamb, R. G., W. H. Chen, and J. H. Seinfeld (1975), "Numerico-Empirical
     Analyses of Atmospheric Diffusion  Theories," J. Atmos. Sci., Vol. 32,
     pp. 1794-1807.

Monin, A. S., and A. M. Yaglom (1971),  Statistical Fluid Mechanics
     (MIT Press, Cambridge, Massachusetts).

Morse, P. M., and H. Feshback (1953), Methods of Theoretical  Physics, p.  795
     (McGraw-Hill Book Company, New York, New York).

Panofsky, H.  A., et al. (1977), "The Characteristics of Turbulent Velocity
     Components in the Surface Layer Under Convective Conditions," Bound.
     Layer Meteor., Vol.  11, pp.  355-361.


                                 105

-------
Ragland, K.  W.  (1973),  "Multiple  Bay  Model  for  Dispersion  of Air  Pollutants
     from Area  Sources,"  Atmos.  Environ.,  Vol.  7,  pp.  1017-1032.

Ragland, K.  W., and 0.  J. Pierce  (1975),  "Boundary Layer Model  for  Air
     Pollutant  Concentrations Due to  Highway  Traffic,"  J.  Air  Poll.  Contr.
     Assoc., Vol.  25,  No. 1,  pp.  48-51.

Riley, J. 0., and G. S. Patterson, Jr.  (1974),  "Diffusion  Experiments with
     Numerically Integrated Isotropic Turbulence," Phys.  Fluids,  Vol. 17,
     pp. 292-297.

Shu, W. R.,  R.  G.  Lamb, and J. H. Seinfeld (1978), "A Model of Second-Order
     Chemical Reactions in Turbulent  Fluid—Part II:   Application to
     Atmospheric Plumes," Atmos.  Environ., Vol. 12,  pp. 1695-1706.

Taylor, G.  I. (1921),  "Diffusion by Continuous  Movements," Proc.  London
     Math.  Soc., Vol.  20, pp. 196-212.

Teske, M.(E., and W. S. Lewellen (1978),  "Prediction  of Estimated Highway
     Emissions by a Second-Order Closure  Model," Draft Report, Environ-
     mental  Protection Agency, Research Triangle Park, North  Carolina.

Wyngaard, J. C., and 0. R. Cote (1971), "The Budgets  of Turbulent Kinetic
     Energy and Temperature Variance  in the Atmospheric Surface Layer,"
     J. Atmos.  Sci., Vol. 28, pp. 190-201.

Zimmerman,  J. R., and R.  S. Thompson  (1975),  "User's  Guide for HIWAY, a
     Highway Air Pollution Model," EPA-650/4-74-008,  Environmental  Protection
     Agency, Research Triangle Park,  North Carolina.
                                 106

-------
                                  TECHNICAL REPORT DATA
                           (Please read Instructions on the reverse before completing}
  REPORT NO
   EPA-60Q/4- 79^023
             3. RECIPIENT'S ACCESSION>NO.
   ri_5 AND SUBTITLE
  A LAGRANGIAN APPROACH TO MODELING AIR POLLUTANT
  DISPERSION
 Development and Testing in the Vicinity of a Roadway
                                                          5. REPORT DATE
                                                             April  1979
             6. PERFORMING ORGANIZATION CODE
 '. AUTHORIS)
                                                          8. PERFORMING ORGANIZATION REPORT NO.
  R. G. Lamb,  H.'  Hogo,  L.  E.  Reid
9. PERFORMING ORGANIZATION NAME AND ADDRESS
  Systems Applications,  Inc.
  950 Northgate  Drive
  San Rafael,  CA  94903
             10. PROGRAM ELEMENT NO.

              1AA815B CA-033 (FY-78)
             11. CONTRACT/GRANT NO.
                                                            68-02-2733
12. SPONSORING AGENCY NAME AND ADDRESS
Environmental  Sciences Research Laboratories - RTP,  NC
Office of  Research  and Development
U. S. Environmental Protection Agency
Research Triangle Park. NC  27711	_.
             13. TYPE OF REPORT AND PERIOD COVERED
             Final  3/78 - 9/78	
             14. SPONSORING AGENCY CODE
              EPA/600/09
15. SUPPLEMENTARY NOTES
16. ABSTRACT
 A microscale  roadway dispersion model based on Lagrangian  diffusion theory has been
 developed and tested.   The model incorporates similarity expressions for the mean
 wind and turbulence  energy in the atmospheric boundary  layer,  through which the
 effects of wind  shear  and atmospheric stability are  taken  into account, and a
 parameterization of  vehicle wake turbulence.  Through simple modifications, the model
 can be structed  to  treat particle settling, deposition, and  resuspension, as well
 as buoyancy of the  effluent material.  Calm winds, winds parallel to the roadway,
 flows around  depressed or elevated roadways, shallow mixed layers,  and transient of
 spatially variable meteorological conditions can all be explicity taken into account
 within the framework of the modeling approach.

 The model was  tested by applying it to the 30-minute experimental periods
 reported in the  General Motors sulfate study.  Of the 1040 predicted values of the
 mean concentration of  an inert material (SF-), half  were found to be within +30
 percent of the measured values.   The overall correlation coefficient was 0.91.  The
 computer time  (but not core storage) required by the model is  directly proportional
 to the distance  between the farthest receptor and the road.  For the studies reported
 the model requires on  the average 20 seconds of CPU  time on  the CDC 7600 to
 simulate each of the 30-minute General Motors experiments.
17.
                                KEY WORDS AND DOCUMENT ANALYSIS
                  DESCRIPTORS
                                              b.lDENTIFIERS/OPEN ENDED TERMS
                          c.  COSATI Held/Group
      *Air pollution
      *Roads
      *Atmospheric diffusion
      *Mathematical models
       Lagrangian functions
       Development
       Tests
                                13B
                                04A
                                12A
                                14B
13. DISTRIBUTION STATEMENT
19. SECURITY CLASS (This Report/
           RELEASE TO PUBLIC
	'IED	
20. SECURITY CLASS (Thispage)
 UNCLASSIFIED
21. NO. OF PAGES
      115
                                                                        22. PRICE
EPA Form 2220-1 (9-73)
                                             107

-------
U.S. ENVIRONMENTAL PROTECTION AGENCY
     Office of Research and Development
    Environmental Research Information Center
           Cincinnati, Ohio 45268

           OFFICIAL BUSINESS
    PENALTY FOR PRIVATE USE. S3OO
  AN EQUAL OPPORTUNITY  EMPLOYER
        POSTAGE AND FEF.S PAID

US ENVIRONMENTAL PROTECTION AGENCY

                EPA-335
                                         If your address is incorrect, please change on the above label
                                         tear off- and return to the above address.
                                         If you do not desire to continue receiving these technical
                                         reports, CHECK Wffffd, tear off label, and return it to the
                                         above address.
                                                          EPA-600/4-79-023

-------