HIGH TIME-RESOLVED COMPARISONS
FOR IN-DEPTH PROBING OF CMAQ
FINE-PARTICLE AND GAS PREDICTIONS
Robin L. Dennis, Shawn J. Roselle, Rob Gilliam and Jeff Arnold*
EP A/600/A-04/094
1.	INTRODUCTION
In this paper, two major sources of bias in the Community Multi-scale Air Quality
Model (CMAQ), one physical and one chemical process, are examined. The examination
is conducted with hourly gas and particle data for the inorganic system of sulfate, total
ammonia, also called NHX, (gaseous ammonia, NH3 plus aerosol ammonium, NH4+) and
total nitrate (gaseous nitric acid, HN03 plus aerosol nitrate, N03~) and with hourly gas
and particle data for inert or conservative species. The physical source of bias stems
from the meteorological inputs related to mixing, in particular the behavior of the
simulated mixed layer in the evening. The chemical source of bias stems from the
nighttime heterogeneous production of HN03 from N205. The analyses are carried out
for a summer and a winter period to examine the seasonal dependence of the biases.
2.	GENERAL MODEL AND DATA DESCRIPTION
2.1 CMAQ
CMAQ is an Eulerian model that simulates input of precursor emissions and
atmospheric transport, transformation, and deposition of photochemical oxidants (ozone),
particulate matter, airborne toxics and acidic and nutrient species (Byun and Ching,
1999). The 2004 release version of CMAQ was used for these simulations. The
meteorological fields were derived from MM5, the Fifth-Generation Pennsylvania State
University/National Center for Atmospheric Research Mesoscale Model (Grell et al.,
1994) with data assimilation and use of the Pleim-Xiu land-surface model (PX) option
(Pleim and Xiu, 1995). The modeling domain covered the contiguous U.S. with a 36-km
horizontal grid dimension. A 24-layer vertical layer structure was used that reached to
*Robin Dennis, Shawn Roselle, Rob Gilliam®, Jeff Arnold", NERL, U.S. Environmental Protection Agency,
RTP NC, 27711, USA. # On Assignment from University Corporation for Atmospheric Research, CO 80303,
USA. ®On assignment from Air Resources Laboratory, National Oceanic and Atmospheric Administration,
RTP, NC 27711, USA.

-------
2
R.L. DENNIS ETAL.
the top of the free troposphere. The simulations were performed with the SAPRC99 gas-
phase chemical mechanism, with the U.S. EPA 2001 National Emissions Inventory and
biogenic emissions from BEIS 3.12. Two periods corresponding to EPA Supersite
Program intensives were simulated: July 2001 and January 2002.
2.2 Observational Data
The data used are high time resolution gas and particle data from July 2001 and
January 2002 taken at two supersites in the EPA Supersite Program: Jefferson Street,
Atlanta, a Southeastern Aerosol Research and Characterization Study (SEARCH) site
(web address in references) and Schenley Park, Pittsburgh (Wittig et al., 2004; Takahama
et al., 2004). In addition, companion (SEARCH) sites in the Southeastern US. were used.
3. RESULTS
3.1. Bias Stemming from Physical Process
Previous comparisons of CMAQ predictions of conservative species against data
taken in 1995 in Nashville, TN (as part of the Southern Oxidant Study (SOS)) indicated
that there was a systematic nighttime over-prediction of conservative species in the
model. Comparison of simulations of mixing heights from MM5 (using the PX land-
surface model option) and from radar profilers around Nashville indicated: (1) that the
mixed layer heights were in good agreement during the mid-day and (2) that the mixed
layer in MM5 was collapsing too soon in the late afternoon. It was hypothesized that the
premature collapse of the boundary layer was contributing significantly to the nighttime
over-predictions.
Comparisons against aircraft spirals over surface sites also indicated that during the
day the atmosphere is well mixed and the surface concentrations are representative of the
overall column concentration levels. Because pollutants in CMAQ are very well mixed
in the vertical and the mixed layer heights appears to be in reasonable agreement when
using the PX option with MM5, we expect the mid-day predicted and measured
concentrations to be in good agreement when daytime emissions are reasonably correct.
If there is a premature collapse of the mixed layer, we expect a very rapid rise in
predicted surface air concentrations leading to an over-prediction or positive bias in the
late afternoon across many inert (e.g., CO and elemental carbon (ECO) or quasi-inert
species (e.g., NOY and NHX) that are emitted late in the day and at night. The key here is
the rate of increase from mid-day levels. We expect little to no effect for species such as
sulfate whose gas-phase formation shuts down as the sun goes down.
We start with EC as the example inert tracer. The average diurnal pattern of EC
predictions and measurements at the Atlanta supersite location at Jefferson Street is
shown in Figure la for July 2001 (summer) and lb for January 2002 (winter). To better
illustrate the relative afternoon rate of increase, the scales are adjusted so that the mid-
day levels for model and measurements match. Indeed, we find a very rapid rise in
surface air concentrations and a rapid increase in bias in the late afternoon. The bias is
most pronounced in summer and least in winter. We will explain why later in this section.

-------
TIME-RESOLVED COMPARISONS OF CMAQ FINE-PARTICLE AND GAS PREDICTIONS 3
The Atlanta diurnal patterns for CO and NOY (not shown) are very similar in each season
to that in Figure 1 for EC. There appear to be nighttime sources of CO, EC and NOx,
consistent with diesel emissions, close to this site that are important in the winter and
affect the nighttime comparisons. Diesel emissions are implicated because the EC to NOY
relationship of the nighttime emissions is the same as for the AM and PM drive peaks.
These sources are either not in the emissions inventory or not handled well by the
chemical transport model. Figure 2 shows the diurnal pattern of EC comparisons in
Pittsburgh for summer to show the similarity in another urban area (without the nighttime
local source issue). Figures 3 shows the Pittsburgh diurnal patterns for S042" for summer,
indicating that, as expected, sulfate concentrations are not seriously affected, because the
S042" gas-phase production shuts down as the OH levels go to near zero in the evening.
The winter diurnal comparison is very similar but with a slight sulfate over-prediction.
Atlanta-JST July 2001 Diurnal EC Comparison
8	12	16
Hour (EST)
—JST AethEC - ¦- - CMAQ 04Rdease EC
Atlanta-JST January 2002 Diurnal EC Comparison
8	12	16
Hour (EST)
- JST AethEC - ¦- - CMAQ 04Release EC I
Figure 1. Comparison of the observed Athelometer Black Carbon and predicted Elemental Carbon hourly
diurnal pattern based on a monthly average of each hour at Atlanta for (a) July 2001 and (b) January 2002.
Pittsburgh July 2001 Diurnal EC Comparison
12	16
Hour (EST)
- SCHPK EC - - CMAQ 04Release EC I
Pittsburgh July 2001 Diurnal Sulfate Comparison
Hour (EST)
- SCHPK SQ4 - ¦- - CMAQ 04Release SQ4 I
Figure 2. Comparison of diurnal pattern of EC at Figure 3. Comparison of diurnal pattern of sulfate
Pittsburgh for July 2001	at Pittsburgh for July 2001
Figure 4 shows the monthly-averaged diurnal temperature bias for MM5 for July
2001 for Atlanta, Pittsburgh and the continental area of the Eastern U.S. excluding
Florida. Given the regular diurnal pattern in the summer of the rise and collapse of the

-------
4	R.L. DENNIS ETAL.
boundary layer and the subsequent decrease in wind speed, we expect the temperature
bias to provide a good indication of what is happening in the boundary layer. For Atlanta
and Pittsburgh, the temperature bias goes positive at mid-day and then steadily falls to
negative values shortly after 1600 EST and continues declining until 1900 EST and then
reverses direction. We associate this rapid over-cooling in the MM5 with the premature
collapse of the mixed layer. Interestingly, the pattern of cold bias maximum in the
morning and afternoon is seen to be a systematic feature across the entire Eastern US. So
we expect this issue of evening over-prediction to be present across the entire model
domain. We expect Pittsburgh and Atlanta, with their larger swings in temperature bias,
to be representative of urban areas; thus, we expect this issue to have to strongest effect
in urban areas. In winter there is a flat, smooth temperature bias with only a monotonic
rise towards no bias during the day and then a monotonic fall again to a constant evening
bias level. In winter there is more competition between winds (mechanical turbulence)
and density stratification and the atmosphere is less stationary, hence, the atmosphere
does not become stable as often (or reaches that state much later in the evening). Thus,
we expect less of an impact on the mixed layer and, hence, the impact on surface
concentrations is expected to be less in winter than in summer.
MM5 Temperature Bias - July 2001
1.5
1
o
o
o
0.5
U)
o
Q
0
(/)
.2
CO
o
3
ro
o
Q.
E
o
l-
¦0.5
•1
¦1.5
¦2
Hour (EST)
— - Atlanta - - Pittsburgh —*— Eastern US
Figure 4. Plot of the monthly averaged hourly diurnal pattern of MM5 temperature bias for July 2001 at
Atlanta, Pittsburgh and across the Eastern U.S.
Figure 5a presents the July 2001 diurnal pattern of NHX in Pittsburgh, where there is
a strong diurnal swing in the CMAQ predictions with under-prediction during the day
and over-prediction at night, whereas there is little to no diurnal variation in the
measurements. Figure 5b shows that the agreement between the 24-hour average
predictions and observations looks fairly good; however, the daily averages have covered
up the fact that there are compensating errors involved, possibly raising questions about
emissions. The 24-hour averages also cover up the fact that the diurnally varying NHX
errors will significantly affect the partitioning of total-nitrate to aerosol nitrate, creating a
companion bias for particulate nitrate. The exact same behavior of compensating errors

-------
TIME-RESOLVED COMPARISONS OF CMAQ FINE-PARTICLE AND GAS PREDICTIONS
5
was seen with the August 1999 Atlanta supersite comparisons for EC, raising questions
about the accuracy of the EC emissions - questions that would be missed when only
looking at daily averaged concentrations. Figure 5c shows that there are also larger
diurnal swings in the January 2002 predictions of NHX for Pittsburgh compared to the
data, with peaks in the morning and afternoon. Although not as dramatic as the summer
situation, these diurnal swings will affect the total-nitrate partitioning as well.
Pittsburgh July 2001 Diurnal NHx Comparison
4.0		
		/	r
l:^. ^
1«		/	
0	4	8	12	16	20
Hour (EST)
SCHPK Total NH3 - ¦- - CMAQ 04Release NHx I
Pittsburgh January2002 Diurnal NHx Comparison
0	4	8	12	16	20
Hour (EST)
SCHPK Total NH3 — — - CMAQ 04Release NHx
Pittsburgh July 2001 Daily NHx Comparison
6.0
5.0
.0
3.0
2.0
1.0
O.i
i.O
6/30/2001
Day (24-Hr Ave: EST)
SCHPK Total NH3 — - CMAQ 04Release NHx
Figure 5. Comparison of the observed and predicted NHx at Pittsburgh for (a) monthly averaged hourly diurnal
pattern in July 2001, (b) daily 24-hour average in July 2001, and (c) monthly averaged hourly diurnal pattern in
January 2002.
3.2. Bias Stemming from Chemical Process
Previous comparisons of simulation predictions from the CMAQ 2002 public release
version showed a very large over-prediction of fine particle nitrate (less than 2.5 microns
in size), to the point of being unacceptable. Comparisons against hourly data for January
2002 at the special sites of Atlanta and Pittsburgh showed huge over-predictions of nitric
acid and/or total nitrate, especially at night. CMAQ includes the nighttime heterogeneous
production of HN03 from N205 on wetted particles. The over-predictions peaked at
night, suggesting an issue with this nighttime heterogeneous production of nitric acid.
The heterogeneous reaction probabilities being used in the 2002 version of CMAQ
were based on Dentener and Cruzen (1993). Recent estimates of the N205 hydrolysis
reaction probability are however two to three orders of magnitude smaller than those

-------
6	R.L. DENNIS ETAL.
suggested by Dentener and Cruzen (1993). The parameterization in CMAQ was updated
to reflect the latest research (Riemer et al., 2003, based on experiments reported in
Mentel et al, 1999), which suggested the reaction probability is much smaller, in the
range of 0.02 and would be inhibited by the presence of nitrate on the aerosols, reducing
it to 0.002 when nitrate is a dominant component of the mixed aerosol. In addition, the
gas-phase reaction of the nitrate radical with water that produces N205 has been turned
off in CMAQ. The argument is that this reaction is highly uncertain and could be a
chamber wall artifact that belongs in the chamber wall model and not in the chemical
mechanism. The SAPRC mechanism developer did not object to this decision, noting
there was a significant degree of uncertainty about the gas-phase reaction (Carter, 2003).
A test of these new literature values for the heterogeneous reaction probabilities
showed a dramatic improvement in the predictions of CMAQ for nitric acid and aerosol
nitrate, although CMAQ potentially is still over-predicting nitric acid at night. Sensitivity
tests with CMAQ were also conducted to further explore the degree of over-prediction in
which the nighttime production of HN03 from N205 was turned off completely.
Figure 6 shows the monthly-averaged diurnal concentration patterns for CMAQ-
predicted HN03 at Atlanta for (a) summer and (b) winter for the base CMAQ that
includes the heterogeneous production of HN03 compared to CMAQ with the
heterogeneous pathway completely turned off. Both model versions are compared to
monthly-averaged diurnal measurements. The differences at Pittsburgh for winter (not
shown) are very similar for the two CMAQ model versions compared to measurements.
Atlanta-JST July 2001 Diurnal HN03 Comparison
Hour (EST)
—JST HN03 — - CMAQ 04Release Gamma - -a ¦ • CMAQ Zero Gamma
Atlanta-JST January 2002 Diurnal HN03 Comparison
4	8	12	16
Hour(EST)
—JST HNQ3 — - CMAQ 04Release Gamma--*- - CMAQ Zero Gamma
Figure 6. Comparison of the observed and predicted (base CMAQ and CMAQ with heterogeneous chemistry
turned off) monthly averaged hourly diurnal pattern of HN03 at Atlanta for (a) July 2001 and (b) January 2002.
The comparisons of the base case and sensitivity study case to the measurements
show that eliminating altogether the nighttime heterogeneous production of nitric acid in
CMAQ brings its predictions much more in line with the nighttime levels of nitric acid at
both special sites, although now the predicted HN03 levels can be below the
measurements. The comparisons of these CMAQ sensitivity runs show that the nighttime
chemistry is most important to the overall HN03 budget in winter. Figure 6(b) also shows
that there can be a noticeable over-prediction of nitric acid occurring in daylight hours in
the winter. That is, daytime the photochemical mechanisms for ozone production (either
CB4 or SAPRC99) are also contributing to the HN03 or total-nitrate over-prediction
during the winter.

-------
TIME-RESOLVED COMPARISONS OF CMAQ FINE-PARTICLE AND GAS PREDICTIONS 7
These high time resolution evaluations indicate that nighttime conversion of N205 to
HN03 is the source of the majority of HN03 in winter and daytime photochemistry is the
source of a majority of HN03 in summer. These time-resolved resolution evaluations also
indicate that the heterogeneous reaction probability is still too high; thus, the "nitrate
problem" has been greatly ameliorated, but not eliminated. There are suggestions in the
laboratory research community and in recent ambient observations that a variety of
factors exist that further inhibit these nighttime reactions, including the presence of
organic aerosols or mixtures that include organics, but none are published and available
for use by the CMAQ developers at this time. In addition, a wintertime issue with the
photochemical production of HN03 has been suggested.
4. SUMMARY
There is a systematic over-prediction in the late afternoons for species that are either
emitted or produced at the surface in the late afternoon and early evening. The most
consistent explanation at this time is that this over-prediction is due to a premature
collapse of the boundary layer in the model. For example, the late afternoon and early
morning cold bias of MM5 is consistent with this explanation. The over-prediction
appears to be relatively larger in urban areas and smaller in the rural areas, lending
further support to the hypothesis that an important source of the problem could be the
inability of the meteorological models to adequately account for the urban heat island.
The degree of over-prediction also appears to be larger in summer than in winter.
Nighttime over-predictions can create compensating errors for some of the
pollutants, such as EC in Atlanta (masking errors), and create systematic biases for
others, such as CO, NOY, and summer NHX in Atlanta (giving an incorrect sense of
error). These biases also can amplify the tendency of CMAQ to over-predict fine-particle
nitrate. This analysis shows that comparisons against 24-hour averages are unable to
discern whether the model is getting the right answer for the right reason or for
compensating wrong reason and that these comparisons form a necessary but not
sufficient component of model evaluation.
The diurnal analyses together with a variety of sensitivity analyses show that the
influence of the nighttime heterogeneous reactions on overall HN03 production is much
larger in winter than in summer. In summer, the daytime, photochemical production of
HN03 is dominant. The analysis also indicates that there can be five sources of error
affecting the levels and diurnal pattern of HN03 concentrations: (1) error in the nighttime
heterogeneous production, (2) error in the daytime photochemical production, (3) error in
the NOx emissions, (4) error in the pbl height and mixing, and (5) error in the NH3
emissions (affecting the partitioning of total-nitrate). High time-resolution analyses of
several species or species combinations besides HN03, including NOY, EC, NHX, NH3,
S042", and total-nitrate, are needed to sort out the possible sources or error and to check
for compensating errors.
A serious chemical problem, relating to the heterogeneous production of HN03 at
night in CMAQ, was, to an acceptable degree, fixed in the 2003 public release version of
CMAQ with the help of these high time-resolution evaluations. They indicate, however,
that the nighttime production of nitric acid in the 2003 and 2004 versions of CMAQ most

-------
8	R.L. DENNIS ETAL.
likely is still too high across much of the U.S., leading to a systematic over-prediction of
total nitrate and, hence, particulate nitrate. This is corroborated by comparisons against
the CASTNet data (Eder and Yu, 2004). In a relative sense the degree of over-prediction
of particulate nitrate is larger in the winter, when it is more easily formed due to the
lower temperatures, than in the summer. Interestingly, the two sources of bias (physical
and chemical process) have roughly opposite seasonal dependencies. Thus, the seasonal
balancing of surface NHX and total-nitrate concentrations will vary across the seasons. In
certain areas of the country CMAQ might be expected to predict ammonia limitation
more often than it should in the colder months. Thus, CMAQ might be expected to over-
emphasize the nitrate replacement (in absolute concentration terms) that can potentially
offset part of the reduction in sulfate that will accompany reductions of S02 emissions.
The hypothesis that this bias is affecting CMAQ's predicted changes of fine particles
associated with emissions reductions is now being tested with further high time-
resolution data coupled with model sensitivity analyses. We are continuing to investigate
further both sources of bias with the intent of improving the predictions of CMAQ.
5. REFERENCES
Byun, D.W. and J.K.S. Ching, Eds., Science algorithms of the EPA Models-3 Community Multi-scale Air
Quality (CMAQ) modeling system, EPA Report No. EPA/600/R-99/030, Office of Research and
Development, U.S. Environmental Protection Agency, Research Triangle Park, NC, 1999.
Carter, W., 2003, personal communication.
Dentener F. J. and Crutzen P.J. 1993. Reaction of N205 on the tropospheric aerosols: impact on the global
distribution of NOX, 03, and OH. J. Geophys. Res., 98, 7149-7163.
Eder, B. and Yu, S, A Performance Evaluation of the 2004 Release of Models-3 CMAQ. Proceedings of the 27th
NATO CCMS ITM, 25-29 October 204, Banff Center, Canada, 2004.
Grell, G. A., Dudhia, J., Stauffer, D.R., 1994. A description of the Fifth-Generation Penn State/NCAR
Mesoscale Model (MM5), Report NCAR/TN-389+STR, 138 pp., National Center for Atmospheric
Research, Boulder, Colorado.
Mentel T.F, Sohn M. and Wahner A. Nitrate effect in the heterogeneous hydrolysis of dinitrogen pentoxide on
aqueous aerosols. Phys. Chem. Chem. Phys., 1, 5451-5457, 1999.
Pleim, J.E. and A. Xiu. Development and testing of a surface flux and planetary boundary layer model for
application in mesoscale models, J. Applied Meteor., 34, 16-32, 1995.
RiemerN., Vogel H., Vogel B., Schell B., Ackermann I., Kessler, C. and Hass H. Impact of the heterogeneous
hydrolysis of N205 on chemistry and nitrate aerosol formation in the lower troposphere under photosmog
conditions. J. Geophys. Res., Vol. 108, No. D4, 4144, doi:10.1029/202JD002436, 2003.
SEARCH. Data can be accessed at http://www.atmospheric-research.com
Takahama, S., B. Wittig, D.V. Vayenas, C.I. Davidson, and S.N. Pandis, Modeling the diurnal variation of
nitrate during the Pittsburgh Air Quality Study, J. Geophys. Res., 2004 (in press).
Wittig, B., S. Takahama, A. Khlystov, S.N. Pandis, S. Hering, B. Kirby, and C. Davidson, Semi-continuous
PM2.5 inorganic composition measurements during the Pittsburgh Air Quality Study, Atmos. Environ., 38,
3201-3213,2004.
DISCLAIMER
The research presented here was performed under the Memorandum of Understanding between the U.S.
Environmental Protection Agency (EPA) and the U.S. Department of Commerce's National Oceanic and
Atmospheric Administration (NOAA) under agreement number DW13921548. Although it has been reviewed
by EPA and NOAA and approved for publication, it does not necessarily reflect their policies or views.

-------