EVALUATION OF AN ANNUAL SIMULATION OF
OZONE AND FINE PARTICULATE MATTER OVER
THE CONTINENTAL UNITED STATES: WHICH
TEMPORAL FEATURES ARE CAPTURED?
C. Hogrefe, J.M. Jones, A. Gilliland, P.S. Porter, E. Gego, R. Gilliam, J.
Swall, J. Irwin, and ST. Rao*	EPA/600/A-04/075
1. INTRODUCTION
Motivated by growing concerns about the detrimental effects of fine particulate
matter (PM25) on human health, the U.S. Environmental Protection Agency (EPA)
recently promulgated a National Ambient Air Quality Standard (NAAQS) for PM2 5. The
PM2 5 standard includes a 24-hour limit (65 |J.g/m3 for the 98th percentile) and annual (15
)_Lg/in3) limit. Except for a few cases, the annual standard will be the primary concern for
attainment issues. Over the next several years, grid-based photochemical models such as
the Community Multiscale Air Quality (CMAQ) model (Byun and Ching, 1999) will be
used by regulatory agencies to design emission control strategies aimed at meeting and
maintaining the NAAQS for 03 and PM25. The evaluation of these models for a
simulation of current conditions is a necessary prerequisite for using them to simulate
future conditions. The evaluation presented in this study focuses on determining the
temporal patterns in all components of the modeling system (meteorology, emissions and
* C. Hogrefe and J.M. Jones, ASRC, University at Albany, Albany, NY, U.S.A. A.
Gilliland, R. Gilliam, J. Swall, J. Irwin, and S.T. Rao, NOAA Atmospheric Sciences
Modeling Division, On Assignment to the U.S. Environmental Protection Agency,
Research Triangle Park, NC, U.S.A. P.S. Porter, University of Idaho, Idaho Falls, ID,
U.S.A. E. Gego, University Corporation for Atmospheric Research, Idaho Falls, ID,
U.S.A.

-------
2
C. HOGREFE ETAL.
air quality) and comparing them against available observations. Furthermore, we briefly
investigated the weekday/weekend differences in the observed and predicted pollutant
concentrations and outlined steps for future research. Since anthropogenic emissions are
known to have a distinct weekly cycle, such analyses would help us in evaluating the
modeling system's ability to accurately reproduce the observed response to emission
changes.
2. MODELS AND DATABASE
Meteorological fields for the photochemical simulations were prepared by the MM5
model (Grell et al., 1994) version 3.6.1 over the continental United States at a horizontal
resolution of 36 km for the time period from January 1 - December 31, 2001 (McNally,
2003). The MM5 fields were then processed by the Meteorology-Chemistry Interface
Program (MCIP) version 2.2. Emissions were processed by the SMOKE processor
(Carolina Environmental Programs, 2003) which incorporated the MOBILE6 module
(U.S. EPA, 2003) for mobile source emissions and the BEIS3.12 model for biogenic
emissions (http://www.epa.gov/asmdnerl/biogen.html). The emission inventory was
based on the USEPA National Emissions Inventory for 2001, which relies on state
reported values. The seasonality of the ammonia emissions, an important consideration
for prediction of PM2 5, was estimated based on seasonal information from Gilliland et al.
(2003) and Pinder et al. (2004). These meteorological and emission fields were then
provided as input to two photochemical models, namely CMAQ (February 2004 version)
and REMSAD version 7.061 (ICF Consulting, 2002), both run at a resolution of 36 km
over the continental United States. Chemical boundary conditions for both models were
prepared from a global simulation with the GEOS-CHEM model (Bey et al., 2001).
This study utilizes a variety of observations from different networks. Observations
of surface temperature and wind speed were retrieved from the TDL data set maintained
by the Data Support Section at the National Center for Atmospheric Research (NCAR-
DSS). Hourly surface ozone observations, hourly PM25 concentrations measured by
tapered element oscillation microbalance (TEOM) monitors
(http://www.rpco.com/products/ambprod/ambl400/), and 24-hr average PM25
concentrations measured at monitors following the Federal Reference Method (FRM)
protocol were retrieved from EPA's Air Quality System (AQS) database
(http://www.epa.gov/air/data/aqsdb.html). Speciated PM25 measurements were obtained
from the Interagency Monitoring to Protect Visual Environments (IMPROVE) network,
the Clean Air Status and Trends Network (CASTNet), and Speciated Trends Network
(STN). Because of differences in measurement techniques and instrumentation, sampling
frequencies, and site location criteria, model performance was calculated on a species-by-
species and network-by-network basis. The analysis presented in this paper focuses on
the Eastern United States. Monitoring sites were only included in the analysis if at least
70% of the data were available.
Following the approach outlined in Rao et al. (1997), Hogrefe et al. (2000) and
Hogrefe et al. (2001), a spectral decomposition technique was applied to compare
temporal variations in observed and predicted time series. To this end, time series of
meteorological variables and pollutant concentrations were spectrally decomposed into
fluctuations occurring on the intraday (time period less than 12 hours), diurnal (12-48

-------
EVALUATION OF AN ANNUAL SIMULATION OF PARTICULATE MATTER
3
Table la. Correlations between different temporal components embedded in hourly time series of
observed and predicted temperature, wind speed, ozone and total PM2.5. Median values are
shown for each network/variable.

#Sites
Intra-day
Diurnal
Synoptic
Baseline
Temperature TDL/MM5
738
0.18
0.90
0.95
0.99
Wind Speed TDL/MM5
735
0.02
0.60
0.84
0.90
03 AQS/CMAQ
193
0.07
0.70
0.64
0.87
PM2.5 TEOM
CMAQ | REMSAD
67
0.01 | 0.03 | 0.25 | 0.25 | 0.70 | 0.63 | 0.04 | 0.10
Table lb. Correlations between different synoptic and baseline components embedded in time
series of observed and predicted PM2 5 from different networks. Median values are shown for each
network/variable.

#Sites
Synoptic
Baseline


CMAQ
REMSAD
CMAQ
REMSAD
PM2.5 FRM (daily)
938
0.68
0.65
0.60
0.51
PM2.5 STN (daily)
25
0.60
0.63
0.38
0.35
SO4 Improve (daily)
44
0.77
0.70
0.89
0.77
S04 CASTnet (weekly)
48
0.85
0.72
0.94
0.88
SO4 STN (daily)
23
0.72
0.70
0.85
0.74
NO3 Improve (daily)
44
0.46
0.54
0.88
0.78
N03 CASTnet (weekly)
48
0.51
0.46
0.89
0.83
NO, STN (daily)
23
0.39
0.42
0.83
0.66
NH4 CASTnet (weekly)
48
0.71
0.72
0.55
0.45
NH4 STN (daily)
23
0.63
0.66
0.52
0.37
EC STN (daily)
23
0.41
0.39
0.15
0.32
OC STN (daily)
22
0.48
0.55
0.24
0.28
Crustal STN (daily)
23
0.34
0.29
-0.35
-0.39
hours), synoptic (2-21 days) and baseline (greater than 21 days) time scales using the
Kolmogorov-Zurbenko (KZ) filter as described in Hogrefe et al. (2000). Note that the
intraday and diurnal components could only be estimated for variables measured hourly,
while the synoptic and baseline components could be estimated for variables measured
hourly, daily or weekly. All analyses presented in this paper were performed over the
entire annual cycle from January 1 - December 31, 2001, with the exception of the ozone
weekday/weekend analysis which was performed for June 1 - August 31, 2001.
3. RESULTS AND DISCUSSION
3.1. Correlations on different time scales
Correlations between different temporal components embedded in time series of the
observed and predicted variables were computed for temperature, wind speed and ozone
as well as total and speciated PM2 5 from the different measurement networks (Tables la-
b). The correlations were computed at each site for a given variable/network/model
combination, and Tables la-b list the median value of the correlation across all sites for a
given variable/network/model combination. For the meteorological variables
(temperature and wind speed), correlations increase with increasing time scale, i.e.

-------
4
C. HOGREFE ETAL.
correlations are lowest for the intra-day component (r < 0.2) and highest for the baseline
component (r > 0.9).
While correlations are relatively high for the diurnal component (r > 0.6), part of this
correlation is due to the inherent cyclical nature of this component, and correlations are
lower when the time series of the diurnal amplitudes are considered (not shown). It is not
surprising that the correlation is highest on the synoptic and baseline time scale since
MM5 model predictions were nudged towards analysis fields using 4-Dimensional Data
Assimilation techniques. For ozone, correlations follow a similar pattern as temperature
and wind speed, with correlations on the intraday time scale being less than 0.1 and
correlations on the baseline time scale being 0.87. The results presented here are
consistent with those presented in Hogrefe et al. (2001) who analyzed ozone from a
three-months summertime simulation over the Eastern United States.
Except for hourly measurements of total PM2 5 by TEOM instruments retrieved from
EPA's AQS, all PM2 5 measurements analyzed in this study are based on filter samples of
either 24-hr average or 7-day average concentrations. Consequently, the intra-day and
diurnal components could only be estimated for the comparison of CMAQ and
REMSAD model predictions with total PM2 5 measurements by TEOM instruments. It is
striking that the correlations between TEOM observations and model predictions are
poor on the diurnal and baseline components for both CMAQ and REMSAD. Figures 1 a
and b show the average observed and predicted diurnal cycles and the time series of the
observed and predicted baseline components averaged over all TEOM monitors and
corresponding model grid cells. Although the time of occurrence of maxima are
simulated well, there is a large difference in the amplitude of the diurnal forcing; both
models overestimate observations during nighttime and severely underestimate
observations during daytime hours. Plausible reasons for this discrepancy are
misrepresentations of the strength of vertical mixing in the model or the magnitude of
primary PM2 5 emissions from area and mobile sources. Figure lb illustrates that the low
correlations on the baseline time scale stem from the higher PM2 5 predictions by both
models during wintertime while TEOM measurements show a decrease. Part of this
decrease in TEOM measurements during wintertime is likely caused by the high
operating temperatures of most of the currently-deployed TEOM instruments (30°C and
50°C). Volatilization losses can occur when the sample is heated from ambient
temperature to the operating temperature, and such losses tend to be higher during colder
ambient temperatures (Allen et al., 1997). On the other hand, CMAQ and REMSAD
utilize MM5-simulated temperatures to calculate the partitioning between gas and
particle phase. In other words, there is an inherent difference between measurement
technique and modeling approach, and this difference exhibits seasonality, thereby
affecting the baseline comparisons. Support for this explanation comes from the higher
baseline correlations when CMAQ and REMSAD are compared against PM25 filter
observations from FRM monitors and the STN network. This highlights the importance
of conducting PM2 5 model evaluation on a network-by-network basis. In other words,
data from different air monitoring networks should not be combined into a single dataset
for the purpose of model evaluation.
Correlations between the synoptic and baseline components of sulfate measured by
the IMPROVE, CASTNet, and STN networks, and predicted by CMAQ and REMSAD
are consistently greater than 0.7, with baseline correlations exceeding 0.85 for CMAQ
and 0.74 for REMSAD. It is noteworthy that there is relatively little difference in model

-------
EVALUATION OF AN ANNUAL SIMULATION OF PARTICULATE MATTER
5
a)
-*-*¦ Observations
b-ej CMAQ
$	0- REMSAD
REMSAD
} oh ok o'e 0*8 l'o 1*2 1*4 lb 1^8 2*0 h.
Hour (Local Standard Time)	Month
Figure 1: a) Average observed and predicted diurnal cycles and b) time series of the observed and
predicted baseline components averaged over all TEOM monitors and corresponding model grid
cells for PM2.5.
performance across the different networks, a finding that is consistent with the regional-
scale nature of sulfate concentrations in the eastern United States that has also been
discussed by Gego et al. (2004). Furthermore, correlations for REMSAD are consistently
lower than those for CMAQ for this pollutant across all networks. For nitrate,
correlations on the baseline time scale are similar to those for sulfate, but correlations on
the synoptic time scale are lower. For the baseline, CMAQ correlations are consistently
higher than those for REMSAD.
In contrast to baseline correlations for sulfate and nitrate, correlations are relatively
low for ammonium. A likely contributor to these lower correlations is the seasonal
characterization of NH3 emissions. The seasonality for NH3 emissions is a well-known
uncertainty that is currently being investigated from both bottom-up inventory
development and from top-down estimation methods (Gilliland et al., 2003). An inverse
modeling study is underway using this 2001 annual simulation to consider how the
current seasonality estimates for NH3 emissions should be modified to improve model
predictions of ammonium aerosols and wet deposition.
Model predicted concentrations of elemental carbon (EC), organic carbon (OC), and
crustal material are strongly influenced by emissions of primary PM since there is no
secondary formation mechanism for EC and crustal material in CMAQ and REMSAD.
Consequently, the relatively weak correlations between the observed and predicted
Observations
CMAQ
-&¦¦¦¦&¦¦ REMSAD
Observations
CMAQ
REMSAD
Emissions
-220.
-200.
180. m
-160.
-140.»
O 1.2
-120.0
-------
6
C. HOGREFE ETAL.
baseline components for these species point to potential problems in the temporal
allocation of PM emissions during emission processing. To investigate this issue, we
constructed the baseline component of EC observations, CMAQ and REMSAD
predictions, and total PM25 emissions at several STN monitoring locations. Examples of
this analysis are shown in Figures 2a and b. The strong correlation between PM25
emissions and model-predicted EC concentrations is clearly visible at the Decatur, GA
monitor and, to a slightly lesser extent, at the Bronx, NY monitor. In both cases, the
relatively poor correlation between observations and model predictions seems to be
largely driven by temporal signature of the PM25 emissions. Therefore, in order to
improve model performance on longer time scales for primary species such as EC and
crustal material, it is necessary to improve the temporal characterization of primary PM2 5
emissions.
In summary, the results presented in Tables la-b illustrate that the models exhibit
greatest skills at capturing longer-term (seasonal) fluctuations for temperature, wind
speed, ozone, sulfate and nitrate. For total PM2 5, ammonium, EC, OC and crustal PM2 5,
correlations are highest for the synoptic time scale, implying problems with factors other
than meteorology in capturing the baseline fluctuations. For the variables for which
hourly measurements were available, correlations were insignificant on the intraday time
scale, suggesting that these models are not skillful in simulating the shorter-term
variations in pollutant levels.
3.2. Analysis of the weekday/weekend effect
Anthropogenic emissions of NOx and VOC are reduced on weekends due to reduced
traffic as well as industrial and commercial activities. The impact of these cyclical
reductions of precursors on ozone has been the subject of numerous studies (e.g.
Cleveland et al., 1974; Croes et al., 2003). Many studies found ozone increases on
weekends in urban core areas and laid out several hypotheses explaining this behavior,
including reduced 03 titration by NO and more efficient ozone production on weekends.
One of the main motivations for such studies is to infer the likely response of ozone
concentrations to emission control policies from these cyclical real-world emission
reductions. While there is no direct way to evaluate photochemical model responses to
hypothetical emission control scenarios, analysis of the weekday/weekend effect could
provide a tool to evaluate the modeling system's ability to accurately reproduce the
observed response to emission changes. While performing such analysis in sufficient
detail is beyond the scope of this paper and will be presented in future work, we briefly
outline some of the necessary steps in performing this analysis.
As a first step, it is necessary to establish that the weekday/weekend effect indeed
exists during the period of analysis in both observations and model predictions. To this
end, average weekly cycles need to be computed for each station, and the difference
between average weekday (Monday-Friday) and weekend (Saturday-Sunday)
concentrations needs to be determined. As an example, Figure 3 presents a scatter plot of
the average CMAQ predicted difference between weekend daily maximum 1-hr ozone
concentrations and weekday daily maximum 1-hr ozone concentrations versus the
corresponding difference computed from observations at the same location. This figure
illustrates that a weekday/weekend cycle of comparable magnitude is indeed present in
observed and CMAQ-predicted ozone concentrations during the summer of 2001.
-------
EVALUATION OF AN ANNUAL SIMULATION OF PARTICULATE MATTER
7
-1U.-0. U. D. 10. 10. £U. ao.		a 	o 	g 	1 n i o	A
Observations	' observations
Figure 3: a) Scatter plot of the average CMAQ predicted difference between weekend daily maximum
1-hr ozone concentrations and weekday daily maximum 1-hr ozone concentrations versus the
corresponding difference computed from observations at the same location, b) As in a), but for daily
maximum temperature.
Additional analyses are needed to determine the location of stations that show opposite
magnitudes of the weekday/weekend differences, to perform analysis on early morning
concentrations (when the difference in motor vehicle emissions between weekdays and
weekends is most pronounced), to restrict analysis to high ozone days only, and to
include analysis of ozone precursors as well. Furthermore, it is important to ascertain
whether the weekday versus weekend differences in ozone concentrations are caused by
emission fluctuations or might be largely explained by meteorological effects when
analysis is restricted to a single summer season only. Indeed, Figure 3b shows that for the
summer of 2001 there was a distinct weekday/weekend fluctuation in temperature at
most monitors in the eastern U.S. Most monitors show lower daily maximum
temperatures on weekends than on weekdays, and MM5 captures this behavior. Because
temperature both directly influences the rate of ozone formation and serves as a proxy for
other meteorological parameters conducive to ozone formation, this figure illustrates that
the existence of a weekday/weekend cycle in ozone concentrations for the summer of
2001 (Figure 3a) can not unequivocally be attributed to cyclical changes in precursor
emissions. In the absence of longer time records of observed and predicted ozone
concentrations, one might be able to apply statistical techniques to adjust observed and
predicted ozone concentrations for meteorological variability to isolate the effect of
emission reductions. Also, it is important to include an analysis of weekday/weekend
differences in the processed emission files in the overall evaluation of the modeling
system's ability to simulate this phenomenon.
4.	ACKNOWLEDGMENTS
The Department of Commerce partially funded the research described here under
contracts with Dr. E. Gego (EA133R-03-SE-0710), with the University of Idaho to Dr. P.
5.	Porter (EA133R-03-SE-0372), and with the State University of New York to Dr. C.
Hogrefe (EA133R-03-SE-0650).
5. DISCLAIMER
-------
8
C. HOGREFE ETAL.
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)
and under agreement number DW13921548. This paper has been reviewed in
accordance with the EPA's peer and administrative review policies and approved for
presentation and publication.
6. REFERENCES
Allen, G., C. Sioutas, P. Koutrakis, R. Reiss, F.W. Lurmann, and P.T. Roberts, 1997: "Evaluation of the TEOM
method for measurement of ambient particulate mass in urban areas, J. Air Waste Manage. Assoc., 47,
682-689.
Bey, I., D. J. Jacob, R. M. Yantosca, J. A. Logan, B. D. Field, A. M. Fiore, Q. Li, H. Y. Liu, L. J. Mickley, and
M. G. Schultz, 2001: Global modeling of tropospheric chemistry with assimilated meteorology: Model
description and evaluation, J. Geophys. Res., 106(D19), 23073-23096, 10.1029/2001JD000807
Byun, D.W. and Ching, J.K.S. (eds.), 1999. Science algorithms of the EPA Models-3 Community Multiscale
Air Quality Model (CMAQ) modeling system. EPA/600/R-99/030, U. S. Environmental Protection
Agency, Office of Research and Development, Washington, DC 20460.
Carolina Environmental Programs, 2003: Sparse Matrix Operator Kernel Emission (SMOKE) Modeling System,
University of Carolina, Carolina Environmental Programs, Research Triangle Park, NC.
Cleveland W.S., T.E. Graedel, B. Kleiner, and J.L. Warner, 1974: Sunday and workday variations in
photochemical air pollutants in New Jersey and New York. Science, 186, 1037-1038.
Croes, B.E., L.J. Dolislager, L. Larsen, and J.N. Pitts, 2003. Forum - the 03 "weekend effect" and NOx control
strategies - scientific and public health findings and their regulatory implications. Environ. Manag., July
2003,27-35.
Gego, E., P.S. Porter, J.S. Irwin, C. Hogrefe, and S.T. Rao, 2004: Assessing the comparability of ammonium,
nitrate and sulfate concentrations measured by three air quality monitoring networks, Pure Appl. Geoph.,
in press.
Gilliland A.B., R. L. Dennis, S. J. Roselle, and T. E. Pierce, 2003: Seasonal NH3 emission estimates for the
eastern United States based on ammonium wet concentrations and an inverse modeling method, J.
Geophys. Res., 108 (D15), 4477, doi:10.1029/2002JD003063
Grell, G. A., J. Dudhia, and D. Stauffer, 1994: A description of the fifth-generation Penn State/NCAR
Mesoscale Model (MM5). NCAR Technical Note, 138 pp., TN-398 + STR, National Center for
Atmospheric Research, Boulder, CO
Hogrefe, C., S. T. Rao, I. G. Zurbenko, and P. S. Porter, 2000: Interpreting the information in ozone
observations and model predictions relevant to regulatory policies in the eastern United States; Bull.
Amer. Meteor. Soc., 81, 2083 - 2106.
Hogrefe, C., S. T. Rao, P. Kasibhatla, W. Hao, G. Sistla, R. Mathur, and J. McHenry 2001: Evaluating the
performance of regional-scale photochemical modeling systems: Part II - ozone predictions. Atmos.
Environ, 35, 4175-4188
ICF Consulting, 2002: User's Guide to the Regional Modeling System for Aerosols and Deposition (REMSAD)
Version 7. Systems Applications International/ICF Consulting, San Rafael, CA 94903. 153 pages.
McNally, D., 2003: Annual application of MM5 for calendar year 2001, prepared for U.S. EPA by D. McNally,
Alpine Geophysics, Arvada, CO, March 2003, 179 pp.
Pinder R.W., R. Strader, C.I. Davidson and P.J. Adams, 2004: A temporally and spatially resolved ammonia
emission inventory for dairy cows in the United States, Atmos. Env., 38, 3747-3756.
Rao, S.T., I.G. Zurbenko, R. Neagu, P.S. Porter, J.-Y. Ku, and R.F. Henry, 1997: Space and time scales in
ambient ozone data. Bull. Amer. Meteor. Soc., 78, 2153-2166.
U.S. Environmental Protection Agency, 2003: User's guide to MOBILE6.1 and MOBILE6.2; EPA Office of Air
and Radiation, EPA420-R-03-010, Assessment and Standards Division, Office of Transportation and Air
Quality, U.S. Environmental Protection Agency. 262pp.
-------