United States
Environmental Protection
Agency  >
Research and Development
Atmospheric Research and      ^ ; :- •;#/.
Exposure Assessment Laboratory •*€!,  :\
Research Triangle Park. NC 27711 "% ^"^

EPA/600/S3-89/056 July 1989
 Project  Summary
 Evaluation  and Sensitivity
 Analyses  Results  of the
 MESOPUFF  II  Model with
 CAPTEX  Measurements
 James M. Godowitch
The MESOPUFF II regional Lagrangian
puff model has been evaluated  and
tested with the Cross-Appalachian
Tracer Experiment  (CAPTEX) data
base. The model was applied to the
six full-scale CAPTEX episodes in
order to Investigate its  ability to
transport and disperse  the  tracer
plume formed  from the  3-hour
release of an inert, non-depositing
perfluorocarbon  tracer gas from
either one of two  selected sites.
Model performance  was quanti-
tatively determined  from traditional
statistical measures of difference and
correlation between modeled  and
observed tracer  concentrations
paired in  time and location. Graphical
maps displaying  observed  and
modeled  plume patterns were also
employed to qualitatively  assess
spatial displacements, while analysis
of plume centroid positions provided
quantitative  information about  the
amount of separation and difference
in downwind distances between the
respective plumes with time for each
two day episode.
  Diagnostic test results  applying
optional  single level wind  fields
available  in the  model and certain
optional  dispersion methods  are
compared to results from the default
model runs, which  employed  a
mixed-layer averaged wind field  and
Gaussian dispersion parameters.
Transport time and location of impact
of the peak tracer concentration  and
its magnitude at the first sampling
arc were examined for the various
diagnostic test runs and compared
against measured results.
  Sensitivity test runs were also
performed that focused on selected
options and variations  in  key
parameters  in  the  model's  dry
deposition   and    chemical
transformation mechanisms in order
to assess their impact on 24-h mean
and peak sulfur dioxide and  sulfate
concentrations using emissions from
a realistic elevated point source.
  This  Project Summary was
developed by the EPA's Atmospheric
Research and Exposure Assessment
Laboratory,  Research Triangle Park,
NC, to announce key findings of the
research  project that  is  fully
documented in a separate report of
the same title (see Project  Report
ordering information at back).

Introduction
  The ability of  a regional-scale  air
quality model to  reproduce  spatial
pollutant  concentration  fields  or
deposition patterns on a short-term basis
is strongly dependent  upon  its
formulations for  simulating  the
atmospheric  transport and dispersion
processes. Thus, an evaluation of these
crucial model components against field
measurements is an important element in
establishing the credibility of any model.
  Experimental field studies with certain
tracer gases have provided valuable
concentration measurements  and
meteorological data on regional scales,

-------
which have allowed assessments of the
transport and  dispersion  methods  in
models.  One of the most recent  tracer
data sets  available for this  purpose is
from  the  Cross-Appalachian Tracer
Experiment (CAPTEX).  The  CAPTEX
data base has provided a challenging set
of episodes for  testing  regional-scale
transport and  dispersion models.  This
intensive field study  was  specifically
designed  and conducted with the
intention of acquiring accurate tracer data
over an extensive  surface sampling
network  and  concurrent  upper  air
meteorological measurements along with
ancillary measurements  on tower and
airborne platforms.
   The full report documents the results
of an evaluation and testing  effort of the
MESOPUFF II  model system  with  the
CAPTEX data base. MESOPUFF II is  a
second  generation  Lagrangian  puff
model, which was designed  to treat the
transport,  dispersion,  chemical
transformation for SOX and NOX, and the
removal  (dry and  wet)  processes
influencing pollutant  emissions from
elevated point and/or  area sources over
regional  scales  for  multiple-diurnal
cycles. Any non-reacting, non-depositing
gas may also be modeled by electing not
to simulate transformation or deposition
processes in  the model  runs.  The
MESOPUFF II  (Version 4.3) model was
executed  and evaluated  with the
measurements obtained from the six full-
scale experimental cases from CAPTEX.
   Quantitative  results  are provided by
traditional statistical  measures  of
difference between modeled  and
observed  mean  and  peak  tracer
concentrations in an  effort to assess
model performance.  However, the
analysis  tools were not  limited  to
statistical results  of concentration pairs.
Selected  graphical   maps   of
modeled/observed plume patterns were
also produced to  provide qualitative
evidence to assist in the interpretation of
model performance and to indicate where
improvements might be  needed. Plume
centroid positions were analyzed  to
obtain  quantitative measures  of  the
difference in  downwind distance and
separation  of the  observed and modeled
plumes as  a function of time.
   Diagnostic test runs were performed
with alternate  single  level  wind  fields
available  from  the  meteorological
processor and with  variations  in  the
dispersion  method provided in the model.
The results of the diagnostic model runs
are compared to  the default model runs
and to actual  values for the  CAPTEX
cases  to  investigate  differences in
transport and dispersion.
   Since  the  model  evaluation  was
limited to an assessment of the transport
and  dispersion components, a  select
group of model sensitivity runs was also
undertaken  to investigate the impact on
24-hour mean and peak  SOX concentra-
tions due to changes  in  key technical
parameters  or the selection of options in
the  dry  deposition  and  chemical
transformation methods. The model base
case run and sensitivity test runs were all
performed  using the  same meteor-
ological fields  from  a single CAPTEX
episode  and  the  emissions  from  a
realistic, large elevated point source.

Model Description
   The MESOPUFF II modeling system is
composed of separate computer codes to
process  meteorological data (READ56,
MESOPAC  II), to  compute pollutant
concentrations  (MESOPUFF II),  and to
perform  various  postprocessing
operations on  modeled  concentrations
from receptor  sites or  over  a  gridded
domain  (MESOFILE  II).  All of  these
model elements were  exercised in this
evaluation effort.
   MESOPAC II is the primary processor
program, which  generates the hourly
gridded fields of the horizontal  wind
components for  two  layers (default),
mixing height (Z,), surface friction velocity
(U«), convective  velocity  scale  (w«),
Obukhov  length (L), and  PGT  stability
class (A-F) from  National Weather
Service  (NWS)  hourly surface
observations, twice-daily  upper air data,
and  land use  categories.  Precipitation
measurements  are optional  and were not
used in this effort because dry conditions
generally  prevailed  during CAPTEX
episodes.
   The two  operational  (default)  wind
fields are:  a mixed-layer averaged wind
field to represent the flow between the
surface and current Z, and an upper layer
averaged wind field for the region from Z,
to the  700 mb height. There are  also
alternative single level wind fields, which
can be constructed if selected.
   The MESOPUFF II model  applies the
puff superposition technique to simulate a
continuous  pollutant  plume  from  either
point and/or area sources. Each puff is
horizontally symmetric with a Gaussian
distribution.  The  puff  release  and
sampling  rates must be specified by the
user in each application. The horizontal
(oy)  and   vertical  (oz)  dispersion
parameters  govern puff growth out to 100
km (default).  Values of  the  dispersion
parameters  are computed from power law
formulas  derived from curve fits to tl
Turner dispersion curves for the  differe
stability classes. Puff  dispersion follov
time dependent relationships at  great
downwind distances. No puff splitting
performed in the model.  The height of
puff center is automatically compared
Z, at each hour in  order to determine tl
appropriate layer-averaged  wind  field 1
transport.  Puff dispersion  above Z,
governed by E stability class.

Model Evaluation  Data
The  field  study   phase  of  CAPTE
consisted  of  individual  ground-lev
releases   of  a  perfluoromon
methylcyclohexane (C7H14)  tracer g
over a 3-hour period  on selected  da
during the period from  mid-September
late October 1983. The amount of trac
emitted was accurately controlled  and
constant release rate was assumed ov
the  release period.  The  desirab
attributes for  this tracer  include
extremely low background  level and
interference from  other sources. T
release sites were at  Dayton,   Ohio
Sudbury, Ontario.  There were  four fi
scale  experimental  releases  (#1-
conducted from  Dayton during afterno
periods, which  assured  strong  verti<
mixing  and  the  prevailing  flo\
transported the  tracer  plume across t
sampling region. Release #6 from Dayl
was  very  brief  and was not modeU
Releases #5 and #7 were conducted frc
Sudbury  under  northwesterly win
during two different nocturnal periods.
   The tracer was accurately measur
by  automatic  sequential  samplers
either 3-  or 6-hour intervals  at  87 si
over  an  extensive  surface  netwc
encompassing  the northeastern  Unit
States  and southeastern Canada. T
network of sites was  designed in ai
deployed at approximately  100  km  int
vals  extending from  300  km to  ab<
1100 km downwind of Dayton.
   The  tracer  measurements  we
provided  in units of femtoliters/liter (fl
with  an ambient background  of 3.4 fl
removed  from each concentration in
data  set.  With  a maximum  of
consecutive  samples  allowing  mi
measurements to  span up to 36  hoi
and  a sampling  strategy  designed
bring all  sites  on-line  simultaneou
along each  arc prior  to the expec1
arrival of the observed plume, the fi
experiments encompassed  about two
days of travel across the region in m
of the cases.
   The domains  for the meteorologi
processor and model grids were defir
to be the same size and  encompas;

-------
the CAPTEX  sampling network.  The
Tiodel domain  was defined by 30 E-W x
19 N-S grid squares for the Dayton cases
and it  was expanded  to  24 N-S  grid
squares for the Sudbury cases. The grid
spacing was set to 37 km  for all cases.
This value was partly selected due to the
resolution  of the  land  use data base.
Land use  types  needed for the model
domain  were obtained from an in-house
land use inventory, which contained the
fractional coverage of 12  different  land
use categories  at approximately twice the
resolution  of the model grid.  Since the
model allows only a single  land use  type
to  be specified for each grid square, the
land use category covering the  greatest
fraction  of the total area within each grid
square was selected.
   Upper air profiles from the  regular
National  Weather  Service  (NWS)
rawinsonde sites and  10  supplemental
locations  were obtained  at  6-hour
intervals  during  each experimental
period. However, profile data for  only the
00 and  12 GMT launches were  required
for model input. The upper air data  from
the six  NWS  upper air stations in  the
region  were  applied in  the   model
evaluation runs.
   Hourly surface observations from 25
regular  reporting NWS surface  stations
(model limit) distributed over the model
domain were also prepared for the model
runs for each CAPTEX  study  period.
Missing observations were  interpolated
with data from adjacent hours because
MESOPAC II requires continuous surface
measurements  of cloud cover,  ceiling
height,  wind  speed,  wind direction,
temperature,  pressure,  and  relative
humidity on an  hourly basis.


Model Evaluation Procedures
and Results
   The model  was  executed in order to
simulate the tracer release for the six full-
scale CAPTEX experiments. The model
simulated the neutrally  buoyant  ground-
level tracer releases  by emitting puffs
from a 1-m stack height with no plume
rise for  a  3-hour period with the actual
emission rate (g s-1) for each case. The
puff release rate and puff sampling rate
were both set to 4 h-1 as  tests revealed
negligible  concentration differences at
greater  rates for this  application.  The
chemical transformation and deposition
processes were  not  simulated.  Model
execution time was greatly reduced by
specifying that hourly concentrations be
calculated only at the grid  coordinates of
the surface sampling sites. The default
methods and features specified in  the
user's  guide  were  applied in  these
operational model runs.  The  modeled
results  were  averaged  to  obtain
concentrations at 3- or 6-hour intervals by
the MESOFILE II postprocessor program
in  order  to  correspond  with the  time
periods of the tracer measurements.
   Concentration  pairs with both values
exhibiting zero  were  excluded from
statistical analysis since neither observed
nor modeled  plume impacted  these
points.  This  eliminated  1088 out of the
1895 total pairs. Additionally, a relatively
few tracer concentrations of 1-2 fl  I-1
obtained  at  sites  separated from  the
observed  plume  pattern were screened
out of the data base  because they were
deemed to be anomalous.  The final data
set contained 734 modeled and observed
concentration pairs.
   Statistical  results from analysis  of
observed  (0) and  model  predicted  (P)
concentrations paired in time and  space
were computed for each experiment and
the overall  data  set. MESOPUFF  II
overpredicted  mean concentrations,
although  model  values were  within  a
factor of two  of the observed means in
experiments  #2,#4,#7, and for  the  full
data set. The  greatest  overprediction
occurred  in   CAPTEX  #3, which  also
exhibited  the most complicated vertical
wind   shear  pattern  among  these
episodes. The values of various statistics
for residuals (Oj - P,) were larger than the
observed mean  concentrations  and
correlations were  generally low.  These
results  from the statistical  measures are
similar to those  obtained  with  other
regional  model  evaluations with tracer
data sets. Spatial displacements between
the respective plumes were attributed for
the considerable  scatter  and  low
correlations in the statistical results. The
results  revealed  the  difficulty  that  the
regional   models  have  in  accurately
replicating the plume trajectory and
plume spread over long distances.
   Graphical  maps  of  modeled  and
observed  values over the study region
provided valuable  evidence  for
interpreting the amount of overlap of the
respective plumes. The  plume patterns
were  depicted  by  symbols  at the
locations of  sampling   sites  where
nonzero  observed  and  modeled
concentrations  occurred.  The  model
simulated the actual path  and pattern of
the tracer plume quite  well for the
relatively  strong,  steady  westerly  flow
situation  during CAPTEX  #4 when wind
direction  shear was small. Nevertheless,
differences between  the  modeled and
actual   transport  speeds  produced
variations in  the time of  impact of  the
modeled  plume,  which assisted in
explaining the  low correlation  for  this
experiment.  More  notable  spatial
differences were found  in the positions
and extents of the observed and modeled
plume patterns during CAPTEX #3 where
vertical direction shear was significant in
the mixed layer. Similar maps of  plume
patterns for individual event periods gave
useful qualitative  evidence  about  the
relative plume overlap during  the course
of each episode.  The maps of plume
patterns certainly verified the existence of
notable  plume  separations  particularly
during the late periods of each CAPTEX
case.
   A  valuable analysis  technique  that
emerged to assess spatial displacements
between  modeled  and observed  plume
patterns was  the determination of  plume
centroid positions.  The centroid location
of a plume was defined to be the density-
weighted  maximum  concentration
location. Therefore, quantitative measures
were computed  as  the  downwind
distance  difference and separation
distance between observed and modeled
plumes at each time period. Analyses of
these results  revealed that the observed
tracer was often found farther downwind
than the  modeled  results after  the  first
night  in each Dayton case. The  actual
plume was likely transported by  faster
winds aloft and was also  directed along a
different path than the modeled plume,
which was transported by slower  winds
derived over a shallow lower layer.


Diagnostic and Sensitivity Test
Results
   Model test runs with  different  wind
fields and  dispersion features  in  the
model provided  interesting  differences
from the  default methods applied  in the
operational  evaluation. This effort  was
undertaken  to investigate differences in
plume transport between the mixed-layer
averaged winds and optional single level
wind  fields  available  from   the
meteorological processor and to examine
variations in  peak concentrations  by
changes in the dispersion method  in the
model.
   Results at the  first arc at 300 km
revealed  that the mixed-layer averaged
wind  field  provided  a  better
representation of actual  plume transport
speed and direction than single-level
wind fields generated for the  surface or
850 mb height. The comparative results
indicated the modeled plumes  traveled
much slower  and  to  the left (counter-
clockwise) with a surface wind  field, while
850 mb winds transported the modeled

-------
plumes  more rapidly  and to the right
(clockwise) of the observed tracer plume.

   Modeled peak concentrations from the
default  model  runs  in  the  evaluation
overpredicted  observed peak values
whether unpaired in time or location.  For
the high-25 concentrations, the modeled
peak value of  1010 ±436 fl  I-1  can be
compared  to  the observed  peak  of
637±436  fl   1-1.   The   highest
concentrations were also found at the 300
km arc sites during the initial day of each
Dayton experiment. Overprediction  of the
peak values certainly was a strong reason
why the overall  mean  concentrations
values were also overpredicted for  these
cases. Examination of the meteorological
processor output fields  indicated that
stability  class  4 (neutral)  was  often
specified during  the afternoon hours of
the release periods due  to  the strong
winds in these  cases.  It  appeared that
vertical  dispersion  was underestimated
for the  neutral  stability cases with  the
current  formulation for the  Gaussian
dispersion coefficient at the  short-range
distances. In diagnostic test runs with the
uniform  vertical mixing  option, where
puffs are immediately distributed over the
entire depth of  the mixing layer, results
showed  modeled peak concentrations
were more comparable to the observed
peak values for  the Dayton cases.  While
observed plumes  had  become vertically
well-mixed, evidence  indicated  the
vertical  dispersion  coefficient in  the
model had not increased rapidly enough
under neutral conditions to disperse the
puffs through the  entire depth of  the
afternoon mixing  layer before reaching
the 300 km arc.  Results from other model
test runs indicated that a reduction  of the
cross-over  distance between distance-
dependent  and  the  time-dependent
dispersion schemes to 50 km and 10 km
generally produced even higher  peak
concentrations than with the default value
of 100 km for this application.
   The  purpose of the model sensitivity
runs was to investigate the impact on 24-
h  mean and peak SOX concentrations
from select variations in certain key
parameters in  the dry deposition and
chemical transformation  modules. The
emissions in  all model test runs were
continuous from a single elevated point
source.  The base case run included the
default  features for deposition and
chemical conversion of S02 to S04. Each
test case run involved the variation of a
single  parameter or  option.  All model
runs utilized  the same meteorological
fields from a CAPTEX episode  and the
simulation period was 48 hours.
   The   dry  deposition  method
incorporated into MESOPUFF II is based
on the deposition velocity concept, which
is  computed  from  the  sum  of
aerodynamic  and  surface  resistances.
The transformation  rate  of SO2  is
determined from a regression expression
which contains  the  dominant variables
controlling this process as derived from
analyses  of  photochemical  model
simulations.
   Results were obtained by computing
the 24-h plume  average concentration
and peak concentration from each model
run. The select group of model sensitivity
run cases included: no dry deposition, no
chemical transformation,  immediate
uniform vertical  mixing to Zj, changes to
the S02 or S04 surface resistance, and a
plus  or  minus  50%  variation   in
background  ozone  concentrations.
Generally, peak  S02 concentrations were
much less sensitive  than  peak sulfate
concentrations in the  model runs  when
either   deposition  or  chemical
transformation were not simulated. Peak
S04 values were affected  more by  the
selected  variations  in these model
components  because  high sulfate
concentrations occurred at much greater
distances  downwind that peak S02
values.  Tables of results  contain the
actual percentage differences from the
base case run for both 24-h periods.

Conclusions
   The results of the evaluation revealed
that the model overpredicted both mean
and  peak  concentrations.  The
overpredictions  were most  pronounced
for the Dayton releases  at the first two
sampling arcs.  Since  differences  in
horizontal plume spread were not evident
during  the  first  day of transport,  it was
concluded that the primary cause for the
model   overpredictions  was  an
underestimation of vertical plume growth
for neutral stability, which  was the
stability class most  often  specified during
the afternoon release periods. A different
dispersion formula for the  Gaussian
vertical dispersion  parameter for neutral
conditions fitted to the Pasquill  D1 curve
should be incorporated and tested within
the current framework of the model. This
revision  would  provide for more  rapid
vertical  plume spread with the default
Gaussian dispersion distance-dependent
scheme under neutral conditions.
   The statistical results also showed that
large scatter  and  low  correlations
occurred between modeled and observed
concentrations paired in  time and space
for both  mean  and peak values.  This
reflected the  inability  of this  model,  as
with  other similar models, to accurate
replicate the speed and/or direction of th
tracer  plume.  Thus,  the  larg
concentration differences where observe
and  modeled  plumes  did  not overla
were primarily attributed to trajector
errors.  However, the statistical analyse
did not provide sufficient information ft
an assessment of the causes for modi
errors in transport.
   Graphical displays and a technique I
determine  plume centroid  position;
which indicated  the  size of the spatii
plume  differences,  greatly  assisted i
characterizing  model trajectory  error;
Spatial displacements of varying amount
were evident  from graphical  maps <
modeled  and  observed plume pattern
with  time.  Plume centroid  position;
derived  as  the  concentration-weighte
locations from  non-zero values at 6-hoi
intervals,  were  used in quantifying  di
ferences in the  downwind distance  an
the  separation  distance  betwee
observed  and  modeled  plumes
Specifically, results from MESOPUFF
indicated that the observed plume move
faster and/or along a different path durin
the nocturnal  period and was general!
found further  downwind of the modele
plume  by the  second daytime period i
these episodes.  The  plume separatio
distance,  the  difference  in  centroi
locations between  the observed an
modeled  plumes, ranged from 100-30
km during  these cases at  downwin
distances of  500-1000 km.  The mos
rapid increase in plume separation als
occurred during the nocturnal periods.
   The  CAPTEX  tracer   emission
consisted of non-buoyant, ground-lev*
releases. For this source  type, pu
heights in MESOPUFF  II were  alway
less than the mixing height, which causei
modeled  plumes  to  be  continual!
transported  by the lower layer wind fiek
This  feature  is  believed to  be
shortcoming   of the  model  design
especially  during the  nocturnal perioi
when greater vertical shears in speed am
direction often prevail.  It  was conclude'
from analyses of the plume centroii
positions that the modeled plumes mus
have been  advected by slower  wind
derived over the shallow  mixed layer i
night, while observed plumes  had bee
transported  along a different trajectory b
faster winds at higher levels. In the cas<
of an elevated  source emitting a buoyar
plume  which rises to even higher  level;
the model  would have switched  thi
plume  transport at night to  the  uppe
layer wind field  when the mixing  heigh
dropped below  the  height  of the  pul
center. Based on  these  results, mode

-------
applications for multi-day simulations of
 round-level,  non-buoyant   point
jmissions are not recommended. The
model's treatment  for this  source type
could be improved if the  height  of the
puff center after release was redefined to
be one-half of its vertical dimension and
this revised puff center height reached a
limit  when vertical puff growth extended
up to the mixing height.
   Diagnostic  tests of optional features in
the model provided distinct differences in
plume transport and dispersion from the
default  methods.   The  mixed-layer
averaged wind field  (i.e.,  default lower
level wind field) was found to be superior
to  single level wind fields at the  surface
or 850 mb height in simulating the impact
time  and location of the  tracer plume at
the 300 km arc. Results suggested that it
should  remain as the  preferred wind field
for representing boundary layer transport
when applying this model. Model test run
results  with the  uniform vertical mixing
option indicated that peak concentrations
were  much  closer to observed values
than those from  the  default  Gaussian
dispersion parameter methods. With the
uniform vertical mixing method, puffs are
completely dispersed through the entire
extent of the boundary layer after release.
These  test results gave more evidence
that  modeled  plumes required greater
vertical mixing  during the  afternoon
release periods. In lieu  of the suggested
revision  to  the  Gaussian dispersion
scheme noted earlier, the selection of this
optional dispersion method appears to be
an attractive alternative since it produces
the desired vertical dispersion  when
neutral conditions  are prevalent.
   A limited group of model sensitivity
runs was also  performed in an effort  to
examine  the impact  on 24-h peak  and
plume  average S02 and  sulfate  con-
centrations from variations  in certain key
parameters and changes in methods in
the  dry  deposition  and  chemical
transformation  components of the model.
The model base case and test case runs
were  exercised  with  the  same
meteorological fields from  CAPTEX  #1
and  the  same  emission  rates from a
large,  elevated  point  source.  Results
showed that sulfate concentrations  were
more sensitive to the selected variation in
a parameter  or  method.  In particular,
sensitivity run results with differences in
the  surface  S04  resistance or  ±50%
differences  in  background  ozone
concentration  produced  negligible
variations  in  SO2  peak  and mean
concentrations; however,  the impact  on
sulfate concentrations was from  15-20%.
This finding is  particularly  relevant  to
regional  model applications since the
higher concentration levels of  sulfate
were found  at much greater downwind
distances than was S02.  It  follows that
S04 is influenced  to  a greater extent by
transformation and deposition processes.
A more  extensive evaluation  of  these
modules is advocated  with  suitable
experimental data.

-------