Evaluation of GEOS-5 Sulfur Dioxide Simulations during the Frostburg, MD 2010 Field Campaign

Sulfur dioxide (SO2) is a major atmospheric pollutant with a strong anthropogenic component mostly produced by the combustion of fossil fuel and other industrial activities. As a precursor of sulfate aerosols that affect climate, air quality, and human health, this gas needs to be 5 monitored on a global scale. Global climate and chemistry models including aerosol processes along with their radiative effects are important tools for climate and air quality research. Validation of these models against in-situ and satellite measurements is essential to ascertain the credibility of 10 these models and to guide model improvements. In this study the Goddard Chemistry, Aerosol, Radiation, and Transport (GOCART) module running on-line inside the Goddard Earth Observing System version 5 (GEOS-5) model is used to simulate aerosol and SO2 concentrations. Data taken in 15 November 2010 over Frostburg, Maryland during an SO2 field campaign involving ground instrumentation and aircraft are used to evaluate GEOS-5 simulated SO2 concentrations. Preliminary data analysis indicated the model overestimated surface SO2 concentration, which motivated the examination 20 of mixing processes in the model and the specification of SO2 anthropogenic emission rates. As a result of this analysis, a revision of anthropogenic emission inventories in GEOS-5 was implemented, and the vertical placement of SO2 sources was updated. Results show that these revisions improve the 25 model agreement with observations locally and in regions outside the area of this field campaign. In particular, we use the ground-based measurements collected by the United States Environmental Protection Agency (US EPA) for the year 2010 to evaluate the revised model simulations over 30

monitored on a global scale. Global climate and chemistry models including aerosol processes along with their radiative effects are important tools for climate and air quality research. Validation of these models against in-situ and satellite measurements is essential to ascertain the credibility of 10 these models and to guide model improvements. In this study the Goddard Chemistry, Aerosol, Radiation, and Transport (GOCART) module running on-line inside the Goddard Earth Observing System version 5 (GEOS-5) model is used to simulate aerosol and SO 2 concentrations. Data taken in 15 November 2010 over Frostburg, Maryland during an SO 2 field campaign involving ground instrumentation and aircraft are used to evaluate GEOS-5 simulated SO 2 concentrations. Preliminary data analysis indicated the model overestimated surface SO 2 concentration, which motivated the examination 20 of mixing processes in the model and the specification of SO 2 anthropogenic emission rates. As a result of this analysis, a revision of anthropogenic emission inventories in GEOS-5 was implemented, and the vertical placement of SO 2 sources was updated. Results show that these revisions improve the 25 model agreement with observations locally and in regions outside the area of this field campaign. In particular, we use the ground-based measurements collected by the United States Environmental Protection Agency (US EPA) for the year 2010 to evaluate the revised model simulations over 30 North America.

Introduction
Sulfur dioxide (SO 2 ) is a trace gas which poses significant health threats near the surface, with consequences on human 35 health (Ware et al., 1986;US EPA, 2011) and on the ecosystem acidification (Schwartz, 1989). With a mean lifetime of few days in the troposphere (Lee et al., 2011;He et al., 2012), emitted SO 2 is quickly oxidized to form sulfate aerosols. The resulting aerosols exert influences on the atmospheric 40 radiative balance and cloud microphysics (e.g., McFiggans et al., 2006). SO 2 is emitted into the atmosphere mainly from anthropogenic sources such as fossil fuel combustion and industrial facilities. In the US these emissions represent more than 90% of SO 2 released into the air (US EPA, 45 2011). Since the implementation of national environmental regulations (e.g. 1990 Clean Air Act Amendments in the United States), a significant decrease of these emissions has been observed over the past 30 years. To keep track of SO 2 emissions, this gas is monitored throughout the country by a 50 system of continuously sampling ground-based instruments, and also by episodic intensive field campaigns. These campaigns are particularly valuable because the instruments deployed on the ground and from aircraft give not only the opportunity to validate and improve the ability of space-based 55 instruments to monitor air pollutants, but also provide the opportunity to evaluate chemical transport models that simulate the SO 2 and sulfate lifecycle (Easter et al., 2004;Liu et al., 2005;Goto et al., 2011). The purpose of this paper is to take advantage of the data measured during the Frostburg field 60 campaign held in Maryland during November 2010 to evaluate the SO 2 simulated with the GEOS-5/GOCART model. We first describe in Section 2 the aerosol model and give a brief description of the SO 2 sources and the chemical processes considered within the model. In Section 3 we start 65 by validating the modeled SO 2 at the surface over the continental US using the data collected by EPA. In Section 4 we evaluate the GEOS-5 simulated SO 2 with measurement https://ntrs.nasa.gov/search.jsp?R=20140012045 2020-03-04T12:57:09+00:00Z 2 V. Buchard et al.: Evaluation of GEOS-5 SO 2 simulations during the Frostburg, MD 2010 field campaign. data taken during the campaign. Section 5 reports the conclusions.

2 Representation of Aerosols in the GEOS-5 Earth
Modeling System The Goddard Earth Observing System version 5 (GEOS-5) model, the latest version from the NASA Global Modeling and Assimilation Office (GMAO), is a weather and climate 75 capable model described by Rienecker et al. (2008). The GEOS-5 system includes atmospheric circulation and composition, oceanic and land components. By including an aerosol transport module based on the Goddard Chemistry Aerosol Radiation and Transport (GOCART) model (Chin et 80 al., 2002), GEOS-5 provides the capability of studying atmospheric composition and aerosol-chemistry-climate interaction (Colarco et al., 2010). In addition to providing reanalyses of traditional meteorological parameters (winds, pressure and temperature fields, Rienecker et al. (2008)), the inclusion 85 of aerosols provides the background information for GEOS-5 to produce reanalyses of aerosol fields using retrieved aerosol optical depth (AOD) from the space-based instrument Moderate Resolution Imaging Spectroradiometer (MODIS  (2000a). Sulfate aerosol is mostly formed from the oxidation of SO 2 . All simulations include emissions of dimethysulfide (DMS), SO 2 and sulfate and we use prescribed oxidant fields (hydroxyl radical (OH), nitrate radical (NO 3 ) and hydrogen peroxide (H 2 O 2 )) from a monthly 115 varying climatology produced from simulations in the NASA Global Modeling Initiative (GMI) model (Duncan et al., 2007;Strahan and Douglas, 2004). A small amount of SO 2 is produced by the oxidation of DMS, which is emitted naturally from marine phytoplankton. We use a monthly vary- ing climatology of oceanic DMS concentrations (Kettle et al., 1999), with emissions calculated using the surface windspeed dependent (Liss and Merlivat, 1986) parameterizations of air-ocean exchange processes. The main source of SO 2 is anthropogenic, mainly from fossil fuel combustion from 125 power plants and industrial activities (US EPA, 2011). Figure 1 maps the emissions of SO 2 released from coal fired power plants (in tons) over the US in 2007. In this study, two different data sets of anthropogenic emissions and two assumptions about the injection height are considered in our 130 simulations to assess the effect of the emissions on SO 2 surface concentration. At the time of the campaign, the annual anthropogenic emissions of SO 2 were taken from Streets et al. (2009). In the GEOS-5 control simulation, this emission was injected into the lowest model level. All simulated re-135 sults using this configuration are hereafter called the "Control Run" or CR.
Recently, a new Emission Database for Global Atmospheric Research (EDGAR) version v4.1 dataset (European Commission, 2010) became available at 0.5 • horizontal res-140 olution and has the advantage of providing the 2005 anthropogenic emissions of SO 2 by source categories. This new set of emissions allowed us to emit the non-energy emissions (from transportation, manufacturing industries, residential) into the lowest GEOS-5 layer and the energy emissions from 145 power plants at higher levels between 100 and 500 meters (between the 2 nd and 4 th model layers). The results are herein referred to as the "Revised Run" or RR. Figure 2 shows a comparison of the SO 2 anthropogenic emissions by source category: energy-source sector and non-150 energy-source sector, based on the EDGAR 2005 database as used in our revised simulation. Most SO 2 emissions are released from power plants, so it is important to consider the emission injection above 100 m due to the stack height and plume rise. We assume these emissions are constant through-155 out the year. Furthermore, other anthropogenic emissions include aircraft and ship traffic emissions from Mortlock et al. (1998) and Eyring et al. (2005) respectively. We assume 3% of the SO 2 anthropogenic emissions are directly emitted as sulfate. All the simulations include also biomass burning 160 emissions of SO 2 following the Quick Fire Emission Dataset (QFED) inventory and SO 2 emissions from continuously eruptive volcanoes that are based on data from the Global Volcanism Program database (Siebert et al., 2002) and Total Ozone Mapping Spectrometer (TOMS) and Ozone Mon-165 itoring Instrument (OMI)'s SO 2 retrievals (Carn et al., 2003;Krotkov et al., 2006) while emissions from explosive volcanoes follow the Aerocom inventories (Dentener et al., 2006). SO 2 is removed in the atmosphere by dry and wet deposition and oxidized to sulfate by chemical reaction. The main ox-170 idation pathways are in the aqueous phase by H 2 O 2 and in the gas phase by OH (Chin et al., 2000a). We save the model tracer fields every three hours during our simulation. Figure 3 shows results of the simulated SO 2 surface concentrations for January and July 2010. The highest SO 2 concentrations are 175 found over eastern Asia, Europe, and North America, which are major anthropogenic source regions. SO 2 concentrations are higher during the winter; this seasonal variation can be explained by the seasonal SO 2 oxidation rates, which are slower in winter than in the summer (Chin et al., 2000b). The 180 planetary boundary layer (PBL) dynamics is also responsible for this seasonal cycle of SO2 concentrations. Figure 4 shows an evaluation of the GEOS-5 simulation of the SO 2 lifetime in black by comparison with the analysis made by Lee et  the GEOS-5 SO 2 lifetime values are quite close or within the range defined by the uncertainty interval of in-situ measurements. The differences in the transport and in the emissions are among the possible reasons that may explain the discrep-200 ancy with the GEOS-Chem model. In addition the oxidant fields in GEOS-5 are not interactive and depend instead on fields from a different model from a different period.

Model comparison to EPA surface measurements
In this section we evaluate the modeled surface concentra-205 tions of SO 2 and sulfate over the US for the control and revised runs for the year 2010. For this study we used data collected by EPA, local and state control agencies which maintain air quality monitoring networks over the US available from the EPA Air Quality System (AQS) (US EPA, 2010). 3.1 Sulfur dioxide Figure 5 shows the SO 2 daily mean comparisons for the control run (top) and the revised run (middle). The "EPA" daily averages of SO 2 concentration were calculated using hourly concentrations collected from 102 sites obtained from the 215 EPA AQS. A kernel density estimation (KDE) (Silverman, 1986;Scott, 1992) was applied to approximate the joint probability density function (PDF) of observed and modeled SO 2 daily mean surface concentrations. Since SO 2 is usually lognormally distributed, the correlation coefficient (r), the Root 220 Mean Square of the differences (GEOS-5-EPA) (RMS), the standard deviation (STDV) and the mean differences are calculated for logarithmically transformed data (summarized in Table 1 as well as the parameters in the original units calculated using the equations described in Limpert et al. 225 (2001) (Appendix A)). For both plots, the scatter between modeled and observed daily means is significant with correlation coefficients, r=0.49 and r=0.42 for the control and revised run respectively. However, the agreement between the observed and modeled daily mean is better with the re-230 vised run, with lower values for the RMS and the mean difference. The STDV is almost the same for both the control and revised runs. One of the reasons for this discrepancy might be attributed to the change in absolute magnitude of the SO 2 emissions datasets used in the control and 235 revised runs, but we noticed only small differences between the two datasets. Another plausible explanation is the emission injection height considered in the model. The vertical placement of emissions in the revised run decreases the high bias between observations and simulations at the sur-240 face. The remaining bias between observations and revised model SO 2 simulations may be explained by the error of representativeness associated with the incompatibility between in-situ measurements and grid-box mean values predicted by the model. As an attempt to filter out the in-situ measure-245 ments that are very unrepresentative of the grid-box mean conditions, the bottom plot of Fig. 5 presents the results after a statistical quality control was performed with the adaptive buddy check of Dee et al. (2001).
For a given ob-   6. The first column is r, the STDV and the absolute value of the mean difference between the modeled (control run) and observed daily averaged SO2 surface concentrations for each SO2 EPA site in 2010. The second column is the change in r, STDV, and absolute value of the mean difference for the revised run relative to the control run. The third column is the same, but showing the difference between the revised run (with buddy check of Dee et al. (2001)) and the control run. The color coding in the second and third column is such that blue indicates improvement relative to the control run.
observations discrepancies and discarding those observations that cannot be corroborated by their neighbors. A brief summary of the algorithm is given in Appendix B. After removing observations that failed this adaptive buddy check (Fig. 5  -bottom Fig. 6 presents the change in the r (top), the STDV(middle) and the absolute value of the mean difference (bottom) between modeled and observed daily averaged surface SO 2 for the control run on the left, the revised run in the middle and after the buddy check on the right. While the cor-270 relation coefficient increased from values lower than 0.4-0.6 for the control run to values greater than 0.6 after the buddy check, we see that the STDV increased over New England and slightly decreased elsewhere for the revised run, the decrease is more significant after the buddy check. Concerning 275 the absolute value of mean difference, we notice a decrease more and more significant between the control, the revised run and after the buddy-check.  but for sulfate. The daily means are directly provided by the EPA AQS and are available every one, three or six days for a total of 250 sites. Figure 7 includes also a comparison with the sulfate simulated with the GEOS-5 aerosol assimilation system, assimilation of MODIS AOD in the re-285 vised version of the model has been performed. On average the modeled sulfate concentrations are higher than the observations, regardless of the model or data assimilation system used. The values of r, the RMS, STDV and the average differences are slightly different for the control, revised sim-290 ulations and the reanalysis (summarized in Table 2). This suggests that the SO 2 emissions injections as well as the assimilation of AOD observations into the model have a low impact on the daily mean sulfate comparisons. Like for the SO 2 study, the measurements have been quality controlled 295 using the buddy-check scheme (Fig. 7), permitting an increase r from 0.71 to 0.79, the RMS, the STDV and the mean difference have been divided by almost a factor 2. Coupled with the longer lifetime of SO 2 in Fig. 4 and 5 and, hence, too slow production of sulfate, our results suggest we may 300 strongly underestimating the losses of sulfate aerosol. When looking site by site (Fig. 8), while the values of r decrease with the revised simulations for some sites, the application of the buddy check lead generally to greater and significant correlation coefficient values; the STDV values have not really 305 changed between the control and revised runs but the values tend to decrease after the buddy check. Finally we see also an improvement in the absolute values of the mean differences after the revised and more importantly after the buddy check simulations.

310
4 Evaluation of SO 2 in the model: comparison with measurement data during the Frostburg campaign in Maryland In Section 4 we concentrate our evaluation of the model performance in a smaller region using data collected dur-

315
ing an air quality campaign in western Maryland in November 2010. The Frostburg campaign was a regional air quality campaign conducted by investigators from Washington State University (WSU), the University of Maryland (UMD) and the NASA Goddard Space Flight Center (GSFC) dur-320 ing two weeks in November 2010. The campaign took place in Western Maryland and provided direct measurements of SO 2 among other atmospheric constituents. The interest of this region is based on the abundance of SO 2 from the Ohio River Valley, surrounded by several power plants (Figure 9).

325
In this section, we make use of several data sets available during this campaign to evaluate the anthropogenic SO 2 concentration simulated by GEOS-5.

Surface analysis: comparisons at Piney Run Station
The observed and simulated monthly mean SO 2 at the sur-330 face at Piney Run station are shown in Figure 10. This site is located in a mountain valley close to Frostburg, and is an ideal location for SO 2 monitoring due to its close proximity to power plants stations, with the nearest one, Warrior Run, located south of Cumberland. Globally, the model captures 335 the observed month-to-month variability of SO 2 with a winter maximum for both the control run in red and the revised run in black, as stated in section 2, the oxidation rates and the PBL dynamics are responsible for this seasonal variation. In the control run (the red line in Figure 10), we see that

Column amount analysis: comparisons to a MF-DOAS instrument
Simulated SO 2 column amount is evaluated with measurements from the Multifunction Differential Optical Absorp-360 tion Spectroscopy (MFDOAS) instrument developed at WSU (Herman et al., 2009;Spinei et al., 2010), deployed on the roof of a building at Frostburg State University (FSU) for the campaign. This instrument measures the direct sun irradiance and scattered sunlight in spectral UV and visible wave-365 lengths 281 -498 nm at 0.83 nm spectral resolution recorded simultaneously with a CCD detector in the spectrograph focal plane. Analysis of the measured spectra is done using the DOAS technique which is based on the Beer-Lambert law which states that the relationship at a wavelength between 370 the intensity of the incident solar light and the transmitted one attenuated due to absorption and scattering by aerosols and molecules in the atmosphere (e.g., Platt, 1994;Plane and Smith, 1995). SO 2 column density is measured with an un-certainty less than 0.03 DU. A description of this instrument 375 as well as the DOAS technique can be found in Spinei et al. (2010). Figure 12 shows the comparison between the column density measured by the MFDOAS and simulated by GEOS-5 during daylight hours from 13:30 UTC until 21:00 UTC on November 08 and 09. We notice that changing from 380 one emission dataset to the other shows not much change on the total column amount between the two runs; it confirms the small changes in the absolute magnitude of the SO 2 emissions between the two datasets. Accounting for the uncertainty on the ground-based instrument, the comparison is 385 rather satisfying with both the control and revised run but we notice that the model does not reproduce the observed diurnal variations. Besides the lack of diurnal variation in the prescribed emissions, an explanation might be the spatial resolution of the model (∼25 km) and the offset pointing of 390 the MFDOAS instrument when looking at the sun.

Vertical analysis: comparisons to aircraft measurements
The GEOS-5 simulated vertical distribution of SO 2 is compared to aircraft measurements conducted on two different 395 days during the campaign. The flights were made on the UMD Cessna 402B aircraft, which was equipped with a modified pulse-fluorescence instrument to measure the in situ SO 2 concentration (Taubman et al., 2006). The aircraft flight path on November 8 is shown on Fig. 9. Important regional 400 power plants are marked by yellow circles in Figure 9, with   Figure 13 shows the simulated vertical profile of SO 2 for the control (left) and revised (middle) runs sampled along the air-415 craft flight path, as well as the comparisons of the modeled SO 2 concentration from the revised run only to the aircraft observations for both days. The dark black lines in Figure  13 show the modeled SO 2 extracted exactly at the aircraft position, while the blue shading shows the range of the mod- at the beginning and at the end of the flight on both days.

Conclusions
The Frostburg campaign that took place in Maryland in November 2010 was a good opportunity to evaluate the SO 2 simulated by the GEOS-5/GOCART system. By comparing 445 the modeled SO 2 against observed data, such as aircraft and ground-based measurements from a ground-based system in Frostburg, we have first diagnosed that the SO 2 concentrations was overestimated at the surface and adjusting the vertical placement of the SO 2 anthropogenic emissions inside 450 GEOS-5 improved the SO 2 surface concentrations without changing considerably the integrated total column amount. The improvement in our treatment of the SO 2 anthropogenic emissions was confirmed with the analysis performed over the US using the EPA ground-based measurements.

455
The comparisons of the vertical profile with aircraft data showed that despite the spatial coarse resolution of GEOS-5, most of the major features of the aircraft observations were reproduced by the model on November 8 because the weather was dynamic with turbulent mixing and strong winds. In 460 contrast the analysis on November 9 shows that during quiet days, GEOS-5 will have difficulty of detecting plumes, especially in the vicinity of point source. Concerning the GEOS-5 simulated sulfate, the comparisons with the EPA data show that the changes in the SO 2 emissions dataset and vertical 465 distribution did not affect much the simulation of the sulfate at the surface, the positive bias observed with the control run remains with the revised run. These comparisons suggests that there might have an underestimated loss of sulfate in the model. A full analysis of the chemical processes could not 470 be performed with the available data and there is a possibil-ity that part of this process could also explain part of the bias remaining in the SO 2 and sulfate comparisons.

475
The lognormal distribution A random variable X is lognormally distributed if Y = log(X) has a normal distribution. The meanX and the standard deviation s X of the normal variable are related to theȲ and s Y of the lognormal variable by (Limpert et al., 2001) : Adaptive Buddy Check

485
In the buddy-check algorithm of Dee et al. (2001), first a background check is performed where differences between the observed and modeled daily means are analyzed in order to identify a set of suspect observations, given a specified tolerance. An iterative buddy-check is then performed on 490 each suspect observation using the remaining reliable observations (called "buddies") within a specified radius to perform a refined acceptance test. The tolerance used for this buddy check is adaptive in the sense that current values of the observation minus model departures are used as a local 495 modulator of the innovation variances used in the threshold test. Notice that before applying the buddy check the observation-model departures must be unbiased by removing the mean value. Figure B1 shows the PDF of the points removed after the buddy check is performed for SO 2 . Although 500 in some cases GEOS-5 simulates lower SO 2 surface values than the ground-based measurements, the majority of points removed after the buddy check are due of an overestimation of the GEOS-5 simulations compared to EPA measurements. While misplacement of plumes by the model could account 505 for some large discrepancies that would be flagged by the buddy check, there is no reason to expect that these discrepancies would be of a given sign. Therefore, the positive bias of the removed observations may point to excessive emissions by GEOS-5 at specific locations.

510
Acknowledgements. The campaign participants want to acknowledge significant logistical support from Dr. J. Hoffman (dean of Sciences) and the operations staff at Frostburg State University. WSU acknowledges NASA grant NNX09AJ28G for instrument development and deployment. The authors would like to thank Lacey Brent, 515 flight scientist, for collecting the aircraft data.  (middle), the white line is the aircraft altitude, on the right, the red line is the observed SO2 concentration, the black line is the modeled SO2 concentration (revised run), and the blue shading shows the range of simulated SO2 for the surrounded grids. Fig. B1. Points removed after the adaptive buddy check of Dee et al., (2001) was performed on the model revised SO2 simulations.