IMPORTANT COPYRIGHT INFORMATION

Transcription

1 IMPORTANT COPYRIGHT INFORMATION The following PDF article was originally published in the Journal of the Air & Waste Management Association and is fully protected under the copyright laws of the United States of America. The author of this article alone has been granted permission to copy and distribute this PDF. Additional uses of the PDF/article by the author(s) or recipients, including posting it on a Web site, are prohibited without the express consent of the Air & Waste Management Association. If you are interested in reusing, redistributing, or posting online all or parts of the enclosed article, please contact the offices of the Journal of the Air & Waste Management Association at Phone: , ext journal@awma.org Web: You may also contact the Copyright Clearance Center for all permissions related to the Journal of the Air & Waste Management Association: Copyright 2006 Air & Waste Management Association

2 TECHNICAL PAPER ISSN: J. Air & Waste Manage. Assoc. 59: DOI: / Copyright 2009 Air & Waste Management Association The Relation between Moderate Resolution Imaging Spectroradiometer (MODIS) Aerosol Optical Depth and PM 2.5 over the United States: A Geographical Comparison by U.S. Environmental Protection Agency Regions Hai Zhang and Raymond M. Hoff University of Maryland, Baltimore County/Joint Center for Earth Systems Technology, Baltimore, MD Jill A. Engel-Cox Battelle Memorial Institute, Arlington, VA ABSTRACT Aerosol optical depth (AOD) acquired from satellite measurements demonstrates good correlation with particulate matter with diameters less than 2.5 m (PM 2.5 ) in some regions of the United States and has been used for monitoring and nowcasting air quality over the United States. This work investigates the relation between Moderate Resolution Imaging Spectroradiometer (MODIS) AOD and PM 2.5 over the 10 U.S. Environmental Protection Agency (EPA)-defined geographic regions in the United States on the basis of a 2-yr ( ) match-up dataset of MO- DIS AOD and hourly PM 2.5 measurements. The AOD retrievals demonstrate a geographical and seasonal variation in their relation with PM 2.5. Good correlations are mostly observed over the eastern United States in summer and fall. The southeastern United States has the highest correlation coefficients at more than 0.6. The southwestern United States has the lowest correlation coefficient of approximately 0.2. The seasonal regression relations derived for each region are used to estimate the PM 2.5 from AOD retrievals, and it is shown that the estimation using this method is more accurate than that using a fixed ratio between PM 2.5 and AOD. Two versions of AOD from Terra (v4.0.1 and v5.2.6) are also compared in terms of the inversion methods and screening algorithms. The v5.2.6 AOD retrievals demonstrate better correlation with PM 2.5 than v4.0.1 retrievals, but they have much less coverage because of the differences in the cloud-screening algorithm. IMPLICATIONS Satellite AOD retrievals can provide decision-makers with air quality information containing larger spatial coverage than in situ measurements. The regression relation derived from this paper between the MODIS AOD and PM 2.5 can be used to estimate surface PM 2.5 with higher accuracy over the United States than using a simple fixed ratio between these variables. INTRODUCTION Particulate matter with aerodynamic diameter less than 2.5 m (PM 2.5 ) can cause serious health problems such as chronic and acute respiratory end points, including mortality. 1 PM 2.5 concentration is one of the air quality parameters monitored by the U.S. Environmental Protection Agency (EPA) Air Quality System 2 (AQS). Hourly PM 2.5 data are available at over 600 sites throughout North America, providing real-time air quality information to the public through the AIRNOW Web site. 3 One limitation of these ground-based PM 2.5 monitors is the spacing between stations and paucity of information over parts of the United States. Although there are many stations along the East and West Coast, the number of stations over the central United States is small. There are no stations over the ocean, which makes it hard to track and predict the transport of aerosols (e.g., the smoke from biomass burning from Central America and dust aerosols from the Saharan desert). In addition, people live between the monitors. Therefore, having a contiguous view of the United States is helpful for air quality analysts. Satellite measurements of aerosols are spatially much denser (although temporally more limited), and they can fill in the gaps between the PM 2.5 monitors. The parameter commonly measured from satellites is aerosol optical depth (AOD), which represents the integrated extinction from the amount of total column aerosol. The AOD-PM 2.5 relationship is a function of the aerosol mass concentration, mass extinction efficiency, hygroscopic growth factor, and effective scale height. 4 The authors discuss the relationship between AOD-PM 2.5 in considerable detail in the 2009 A&WMA Critical Review, as well as discussing prior work on this topic. 5 Previous research shows that there are good correlations between the AOD and PM 2.5 over the eastern United States, 6 8 suggesting the potential for the monitoring of particulate air quality through the use of satellites. Although this remains an active research field, the practicality of using AOD as a surrogate for PM 2.5 encourages its use at this time. Several different models have been proposed for the relation between AOD and PM 2.5 : simple linear regression relation 4 ; multivariate 1358 Journal of the Air & Waste Management Association Volume 59 November 2009

3 regression relation (linear and nonlinear), including meteorological parameters 9,10 ; or the relation with the aid of chemical transport models. 11 Furthermore, Liu et al. 12,13 explored a method to predict component concentration and size distribution using the fractional AOD from Multiangle Imaging Spectroradiometer (MISR). The literature has shown that the correlation between AOD and PM 2.5 is different over different geographical regions and different seasons. For example, it was found that the correlation coefficient can range from 0.14 to 0.6 over 5 cities located at different parts of the Earth, 14 and that the correlation coefficient is small during the months from December to March and large during other months over the southeastern United States. 15 Efforts have been made to improve the correlation coefficients, including removal of the hot spot in AOD retrievals, 16 removal of pixels with high cloud fractions, 14 the use of Light Detection and Ranging (LIDAR) measurements to determine the vertical profile and remove AOD from aerosols aloft, 17,18 the use of multivariate regression regressions and neural network approaches to include meteorological parameters in the AOD-PM 2.5 relation model, 10,19 etc. The relation between AOD and PM 2.5 can be used for monitoring air quality, verifying the air quality model forecast, and data assimilation. In a current nowcasting application (the Infusing satellite Data into Environmental Applications [IDEA] product), a single relationship for AOD-PM 2.5 was used for the whole nation. This paper will examine whether using the data at hand now from several years of providing that product allows researchers to refine the AOD-PM 2.5 relationship further and improve the predictions in IDEA. Many satellite instruments can derive AOD over land, including Total Ozone Mapping Spectrometer (TOMS), 20 Ozone Measurement Instrument (OMI), 21 Polarization and Directionality of the Earth s Reflectances (POLDER), 22 Along Track Scanning Radiometer (ATSR-2), 23 MISR, 24 Moderate Resolution Imaging Spectroradiometer (MO- DIS), 25,26 and Geostationary Operational Environmental Satellites (GOES) 27,28 imagers. Because of the difference in the sensor properties, the AOD retrievals from different sensors can be quite different. Even if from the same sensor, they can also vary greatly if they are derived from different retrieval algorithms. 29 The MODIS sensors provide global AOD on a twice-daily basis over land and ocean. They are located on Terra and Aqua polar-orbiting satellite platforms with daytime overpasses at 10:30 a.m. and 1:30 p.m. local time, respectively. Since launch, the aerosol products have experienced continuous updates from v3.0 to the most recent v ,31 Hutchison et al. 18 showed that the correlation between MODIS AOD and PM 2.5 demonstrates an improvement with the newer version over Texas regions but with less data coverage. This influence over the United States will be discussed later in this paper. The Three-Dimensional Air Quality System (3D- AQS) 32 is a multiagency project funded by the National Aeronautics and Space Administration (NASA) to integrate satellite aerosol data and ground PM 2.5 data into air quality decision support systems, which include EPA s AQS, the AirQuest database, the Remote Sensing Information Gateway (RSIG), and the National Oceanic and Atmospheric Administration (NOAA) National Environmental Satellite Data and Information Service (NESDIS) IDEA product, 7,33 and the U.S. Air Quality weblog. 34 One of the products arising from 3D-AQS is the addition of AOD data from MODIS colocated with EPA PM 2.5 monitors into EPA s AirQuest. Another product is an enhanced, operational NOAA NESDIS IDEA product, which includes products relating MODIS AOD, the GOES Aerosol and Smoke Product (GASP), and PM 2.5. The purpose of these products is to provide user satellite aerosol measurements as guidance for air quality monitoring, analysis, and nowcasting. In the current implementation of the IDEA product, a single AOD-PM 2.5 relationship is used for the entire nation. The literature above and the work below suggest that a refinement can be made to the IDEA product that uses regionally and seasonally adjusted relationships to provide better prediction of PM 2.5 from the measured AOD. This paper investigates the regional and seasonal variations of the relation between MODIS AOD from Terra and Aqua retrievals and surface PM 2.5 measurements over the United States. First, the MODIS aerosol algorithm is introduced. Then the dataset and methodology used in this work are described. Correlations and regression relations between MODIS AOD and PM 2.5 are derived for 10 EPA-defined geographical regions and for different seasons. The regression relations are then applied to derive PM 2.5 from the MODIS AOD retrievals. Finally, the two versions of the Terra MODIS AOD retrievals (i.e., v4.0.1 and v5.2.6) are compared. DATA COLLECTION AND METHODS This work uses 2 yr ( ) of MODIS AOD from Terra (v4.0.1 and v5.2.6), Aqua (v5.2.6), and surface hourly PM 2.5 data. The v4.0.1 MODIS AOD from Terra were acquired from the archive of the IDEA product at the University of Wisconsin, 35 and the v5.2.6 data were obtained from NASA s Level 1 and Atmosphere Archive and Distribution System (LAADS). 36 Hourly PM 2.5 concentration measurements at 521 stations over the United States (Figure 1) were obtained from the IDEA archive, which were accumulated daily from EPA s AIRNOW server. The match-up process between AOD and PM 2.5 used in the IDEA product was adopted. To find the AOD at the position of a PM 2.5 site, the MODIS AOD pixel that covers this site is located and then that value is used. The PM 2.5 concentration at the time of the passing of the satellite was obtained through the interpolation from the two closest measurements in time. For consistency with IDEA, other spatially, more extended (i.e., averaging or inverse weighting) techniques were not used. To find regional dependency of AOD and PM 2.5 relations, the United States were divided into 10 areas on the basis of the EPA geographic regions (Figure 1), which are grouped states geographically close to each other. Table 1 shows the states and the number of the hourly PM 2.5 sites within each region. The hourly PM 2.5 sites are not distributed evenly and they tend to be located with population centers (e.g., they are distributed sparsely in regions 7 and 8 and more densely in other regions). Volume 59 November 2009 Journal of the Air & Waste Management Association 1359

4 Figure 1. EPA hourly PM 2.5 sites and 10 geographical regions over the continental United States. The blue dots represent the hourly PM 2.5 stations, and the regions are plotted on different colors. RESULTS AND DISCUSSION Seasonal Variations The coincidence data between AOD retrievals and PM 2.5 were separated into different seasons and analyzed with the linear regression method. The correlation coefficients, slopes, and intercepts were calculated for each season. Figure 2 shows the seasonal variations of AOD and PM 2.5 slopes and correlations for January 2005 through November Seasons with less than 20 data points are not plotted. Strong seasonal variations of correlation coefficients are observed for regions 1 7 (the eastern and central United States). The correlations are usually good in summer and fall and poor in winter. In regions 1 3 (the northeast), the highest correlation occurs mostly in the summers, with correlation coefficients of approximately 0.6. In region 4 (the southeast), the correlation is high in the fall of 2005 and spring of 2006, consistent with the results of Gupta et al. 14 In region 5 (Great Lakes states), the correlation is highest in the fall of 2005, but it does not show much seasonal variation in In region 6 (southern states from New Mexico through Louisiana), the highest correlation occurs in fall in both years. In region 7 (the Wheatbelt states), the summer and fall of 2005 have very good correlation close to 0.8. In region 8 (the Rocky Mountain states), the highest correlations are observed in the spring of both years. In regions 9 and 10 (the west), the correlations do not have any seasonal patterns. For example, in region 10, the correlation is high in winter 2005, but it is very low in winter Comparing the 2 yr, the correlation is larger for the year 2005 than that for the year 2006 in regions 1 7 and region 9. In almost all of the seasons and regions, v5.2.6 AOD retrievals have higher correlation with PM 2.5 than v4.0.1; however, the slopes and intercepts of the linear regression are quite different. The seasonal averages of coincident AOD and PM 2.5 are shown in Figure 3. In regions 1 4, 6, and 7, the maximum PM 2.5 occurs in summer with similar magnitudes for both years. In region 5, the maximums are also in summer, but the magnitude of PM 2.5 is larger in 2005 than that in The average PM 2.5 in the three western regions (regions 8 10) do not vary much seasonally. Regions 8 and 10 have better air quality than the other regions, with average PM 2.5 values less than 10 g m 3 in most of the years. Except in the three western regions, average AODs follow the seasonal variations of PM 2.5 very well. Table 2 shows a summary of linear regression relationships calculated from all of the coincident data for each region, and regions are ranked by their correlation coefficients obtained from the v5.2.6 AOD. The southeast region (region 4) has the highest correlation coefficients at 0.63 and 0.67 for the v5.2.6 Terra and Aqua, respectively; it also has the largest number of coincidences. Region 9 in the southwest has the lowest correlation coefficient at approximately 0.2. Table 3 shows a summary of the slopes, intercepts, correlations, and number of coincidences between MODIS AOD and PM 2.5 for different seasons over the 2-yr period. The correlation coefficients and numbers of coincidences based on Terra and Aqua are generally similar over most of the seasons and regions. Most large differences in correlation coefficients occur in winter when the uncertainty of AOD retrievals is large and the numbers of coincidences are small. Because the MODIS aerosol retrieval algorithm is based on dark surface pixels in the near-infrared at 2.1 m, the difference in the surface reflectance in different regions and seasons can cause the difference in the accuracy of AOD retrievals. Over a high-reflectance surface, the retrieved AODs are less accurate. The eastern United States, which is covered with green vegetation, has low reflectance during the warm seasons, whereas the western United States is mostly rocky and desert, which has high reflectance. During the winter, early spring, and late fall, the vegetation in the east turns reddish and the surface reflectance increases. This is one reason that the western United States has lower correlation between AOD and PM 2.5 than the eastern United States and that the high correlations are usually observed in summer and fall. It also has been noted in previous papers 6,17,18 that AOD in the west arises more from smoke events, which may be transported more aloft than the high-aod pollution events in the east. High AOD-PM 2.5 correlations are usually found in areas with high variations of particulate matter. Because the intercepts in the regression relation are not zero, the PM 2.5 with low values cannot be detected Table 1. States and number of hourly PM 2.5 sites in each EPA region. EPA Region States Number of Hourly PM 2.5 Sites 1 ME, NH, VT, MA, RI, CT 32 2 NY, NJ 25 3 PA, DE, MD, DC, VA WV 28 4 KY, NC, TN, SC, GA, AL, MS, FL 93 5 MN, WI, MI, IL, IN, OH 80 6 NM, TX, OK, AR, LA 83 7 NE, IA, KS, MO 19 8 MT, ND, WY, SD, UT, CO 20 9 CA, NV, AZ WA, OR, ID Journal of the Air & Waste Management Association Volume 59 November 2009

5 Figure 2. Seasonal variations of MODIS AOD-PM 2.5 slopes (b, d, f, h, j, l, n, p, r, and t) and correlations (a, c, e, g, i, k, m, o, q, and s) ( ) by EPA regions. by AOD and small variations in PM 2.5 should not introduce corresponding AOD variations. Pollution events are mostly found over the eastern United States and cover a large area during warm seasons caused by smoke or anthropogenic air pollution. It can be seen that the average PM 2.5 has a maximum value over regions 1 7 during summer and a minimum value during winter. This may be another reason to explain the observed seasonal pattern over the east. The AOD-PM 2.5 relation is also a function of meteorological parameters, such as height of the mixing layer, temperature, wind, humidity, etc. PM 2.5 is usually well mixed in the planetary boundary layer (PBL), so PM 2.5 should be large for a small mixing layer height corresponding to the same AOD. Gupta et al. 14 found that AOD-PM 2.5 has a higher correlation if the mixing layer height is smaller. The slope between PM 2.5 and AOD should decrease with increasing mixing layer height. Larger wind speed can induce high mixing layer height, which can cause a reduction of the correlation coefficients and slope. The relative humidity (RH) can affect the AOD-PM 2.5 through changing the optical properties of the aerosols. The higher the RH, the larger the portion of light is scattered and hence the larger AOD. Therefore the slope should be smaller with larger RH and hydrated aerosol AODs. 5 The slopes and intercepts in Table 3 can be used as a guide for calculating PM 2.5 from MODIS AOD; that is, PM 2.5 can be estimated from AOD using the slopes and intercepts in the table. Because the parameters derived from the linear regression vary seasonally and geographically, this method is different from that used in IDEA, which applies a fixed ratio of 62:1 between PM 2.5 and AOD over all of the regions and seasons. To find the effectiveness of the regression relation on the estimation of PM 2.5, four cases from the year 2007 one from each season are examined. The summer case Volume 59 November 2009 Journal of the Air & Waste Management Association 1361

6 Figure 3. Seasonal averages of MODIS AOD (a, c, e, g, i, k, m, o, q, and s) and PM 2.5 (b, d, f, h, j, l, n, p, r, and t) ( ) by EPA regions (PM 2.5 is in g/m 3 ). Table 2. Summary of MODIS AOD-PM 2.5 relation ( ) over 10 regions for all seasons ranked by the correlation coefficients from v5.2.6 AOD. Rank EPA Region Terra v4.0.1 Terra v5.2.6 Aqua v5.2.6 N R Slope Intercept N R Slope Intercept N R Slope Intercept ,778 (16,856) 0.51 (0.61) 26.3 (36.3) 9.1 (9.1) 17,829 (16,856) 0.63 (0.64) 31.7 (34.8) 8.9 (8.6) 15, ,374 (2,040) 0.38 (0.55) 20.7 (34.7) 6.1 (4.8) 2,267 (2,040) 0.63 (0.57) 27.7 (29.5) 5.5 (5.3) 1, ,258 (3,808) 0.51 (0.60) 26.2 (34.2) 5.6 (5.7) 3,976 (3808) 0.60 (0.61) 28.9 (30.9) 6.0 (5.9) 3, ,917 (3,738) 0.47 (0.52) 30.3 (37.2) 8.8 (8.9) 3,995 (3,738) 0.57 (0.56) 33.5 (35.6) 8.6 (8.3) 2, ,809 (2606) 0.49 (0.57) 25.9 (32.8) 7.3 (7.1) 2,836 (2,606) 0.58 (0.59) 30.2 (32.2) 6.4 (6.2) 2, ,125 (8,779) 0.40 (0.46) 24.5 (31.0) 8.9 (9.0) 9,345 (8,779) 0.47 (0.47) 26.2 (28.3) 8.7 (8.4) 8, ,081 (8,747) 0.35 (0.47) 15.1 (27.1) 7.0 (6.7) 9,716 (8,747) 0.42 (0.43) 21.8 (23.7) 6.6 (6.5) 10, ,705 (9,375) 0.19 (0.30) 9.8 (15.5) 5.9 (5.6) 9,817 (9,375) 0.33 (0.34) 16.6 (16.3) 5.2 (5.1) 14, ,428 (2,241) 0.15 (0.25) 6.7 (11.4) 7.0 (6.6) 2,431 (2,241) 0.28 (0.30) 10.7 (11.6) 5.6 (5.5) 2, ,785 (7,096) 0.02 (0.22) 1.3 (15.0) 13.4 (12.3) 7,483 (7,096) 0.26 (0.27) 16.6 (17.1) 11.8 (11.7) 10, Notes: Numbers in parentheses are for the dataset in which Terra v4.0.1 and v5.2.6 have one-to-one correspondence Journal of the Air & Waste Management Association Volume 59 November 2009

7 Table 3. Summary of MODIS AOD-PM 2.5 relation by season ( ). Terra v4.0.1 Terra v5.2.6 Aqua v5.2.6 EPA Region Season N R Slope Standard Error Intercept Standard Error N R Slope Standard Error Intercept Standard Error N R Slope Standard Error Intercept Standard Error 1 Winter Spring Summer Fall Winter Spring Summer Fall Winter Spring Summer Fall Winter Spring Summer Fall Winter Spring Summer Fall Winter Spring Summer Fall Winter Spring Summer Fall Winter Spring Summer Fall Winter Spring Summer Fall Winter Spring Summer Fall using Terra AOD (August 2, 2007) is shown in Figure 4. To calculate PM 2.5, the slopes and intercepts for summer from Table 3 and the estimated PM 2.5 from the two versions of AOD retrievals shown in Figure 4, a and b, are used. The corresponding in situ PM 2.5 measurements are shown in Figure 4c. As seen from these figures, the PM 2.5 distribution from v4.0.1 and v5.2.6 MODIS AOD from Terra agrees well with the in situ measurement over the eastern United States. Calculated PM 2.5 and in situ measurements observe high values over the eastern United States. The peaks of the pollution extend along Arkansas, Illinois, Kentucky, Indiana, Ohio, and Pennsylvania. Because of their difference in cloud masks, the calculated PM 2.5 field from v4.0.1 is more continuous and covers more areas than that from v This is the one negative impact from the change in version for the IDEA product. Discontinuity in magnitude at the boundaries between EPA regions is also observed because of the difference in the slopes and intercepts in the AOD-PM 2.5 relations. The left two plots in Figure 4d show the scatterplots of the calculated PM 2.5 versus the in situ PM 2.5. The slopes for both versions are 1.07 and the intercepts are less than 1. The correlation coefficients and root mean square (RMS) errors are also similar at approximately 0.7 and 9 g m 3, respectively. For comparison, the right two plots in Figure 4d are scatterplots for PM 2.5 calculated using a fixed ratio of 1:62 (which is used by IDEA) versus in situ PM 2.5. It can be seen that the calculated PM 2.5 using a fixed ratio tends to overestimate; the slopes are approximately 0.52 and 0.58 for v4.0.1 and v5.2.6, respectively. This points out that reduction of bias in the production of IDEA may be a much more important reason to tune the PM 2.5 predictions Volume 59 November 2009 Journal of the Air & Waste Management Association 1363

8 Figure 4. An example prediction of PM 2.5 from MODIS AOD from Terra for August 2, (a) Calculated PM 2.5 from v4.0.1 MODIS AOD. (b) Calculated PM 2.5 from v5.2.6 MODIS AOD. (c) In situ surface PM 2.5 measurements. (d) Scatterplots of the calculated PM 2.5 and in situ PM 2.5. In the right two plots, the PM 2.5 values are calculated using a fixed ratio between AOD and PM 2.5 of 1:62 from IDEA. than the correlation coefficients, which did not improve. The fixed ratio calculated PM 2.5 also has larger RMS errors ( 12 g m 3 ) than those using the regression relations ( 9 g m 3 ). The in situ PM 2.5 results do show a few stations with higher actual PM 2.5 measurements; for example, 14 of 366 match-ups have 20 g m 3 higher in situ PM 2.5 values than calculated PM 2.5 for the v4.0.1 case (11 of these are within urban areas). Seven of these high PM 2.5 sites also have larger individual slopes and intercepts in AOD-PM 2.5 regression than those in their corresponding regions. These sites are usually found to be high in PM 2.5 with frequent particulate matter criteria exceedences. In the urban regions, the spatial variations of particulate matter are usually large. The MODIS aerosol retrieval algorithm may underestimate AOD in these areas because the AOD retrievals are derived from the pixels within a 20- by 20-box with 20 50% brightness in the red channel, 30 and pixels with high aerosol concentration have high probabilities to be in the upper 50% so that they do not contribute to the retrievals. This may be the reason for the larger slopes and intercepts in these regions. Table 4 shows results of the other three selected Table 4. Relation between in situ PM 2.5 and those estimated from MODIS AOD using different methods for the four cases in Terra Collection 4 Terra Collection 5 Aqua Collection 5 Case Method RMS R Slope Intercept N RMS R Slope Intercept N RMS R Slope Intercept N January 26, 2007 Regression Fixed ratio May 23, 2007 Regression Fixed ratio August 2, 2007 Regression Fixed ratio September 27, 2007 Regression Fixed ratio Journal of the Air & Waste Management Association Volume 59 November 2009

9 cases in 2007 along with the above case. The spring and fall cases are chosen from pollution events that occurred over the eastern United States during those days. For the winter case, a day with most of the United States free from cloud cover was selected because there were no widespread winter high pollution days in this dataset. The fall case is similar to the summer case above with a good improvement in RMS errors, correlation coefficients, slopes, and intercepts by using the regression relations. The spring case also shows an improvement in RMS errors and correlation coefficients, but the regression method tends to underestimate PM 2.5. The poorer result in the spring case than in the summer and the fall cases is in agreement with the poorer correlations found in Table 3 in this season. As expected, the winter case is the worst in terms of the correlation coefficients because of the least variation in all cases. Similar to the correlation relations, the slopes and intercepts vary from region to region and season to season. This makes the combinations of the regions difficult. The authors have observed the abrupt change at some of the boundaries in the authors estimation of cases. One way to introduce smooth transfer from region to region is to use interpolation or averaging of the slopes and intercepts close to the boundaries. The interpolation between widely varying regions will require further testing before it is implemented. IDEA will be implementing a revision of the method that uses regional and seasonal slopes where they are significant (high r 2 ) and where they do not add to the bias in the representation. The Differences between Two Versions of AOD Retrievals Because the v4.0.1 AOD retrievals from Terra were used in the IDEA product during at the University of Wisconsin and were switched to v5.2.6 when it was transferred to NOAA NESDIS in 2007, it is necessary to find out the impact of this change on the AOD-PM 2.5 relation over the United States. MODIS aerosol products treat land and ocean differently, and the AOD are retrieved at 10-km spatial resolution on the basis of 0.25-, 0.5-, and 1-km resolution reflectance data. The original algorithm over land is given by Kaufman et al., 25 and that over ocean is given in Tanré et al. 37 The land algorithm 25,26,30 makes use of reflectance from three channels (0.47, 0.66, and 2.12 m) and produces aerosol product in 10-km spatial resolution. The snow mask and cloud mask is first applied, then dark pixels are selected from the remaining for retrieval. The land algorithm has evolved from Collection 3 to the current Collection 5 over the past several years and has been modified in the cloud mask and the assumptions in the retrieval algorithm. The cloud mask derived from the MODIS product MOD35 38 is used for cloud screening in the versions before v A new cloud mask has been implemented since v The new cloud mask uses a variability test in a moving 3- by 3-window on a and m channel, and the corresponding pixel is determined to be cloud if the standard deviation of the reflectance in this window is above a certain threshold. 39 For versions before v5.2 using the original Kaufman algorithm, the m channel is assumed to be free from the influence of aerosol so that the surface reflectance in this channel is the same as the reflectance measured from the top of the atmosphere (TOA). The surface reflectance in the other two channels is derived from that in the m channel with a fixed ratio. Several modifications have been made in v5.2. The surface reflectance relations among the three channels are changed to a more sophisticated empirical parameterization accounting for the surface type and the solar satellite geometry. The transparency assumption in the m channel is abandoned because it is found that this channel is not totally free from the influence of aerosol. Instead of doing the inversion separately for each channel as in previous versions, v5.2 uses a multichannel reflectance inversion that yields three parameters simultaneously (i.e., AOD at 0.55 m, a fraction of aerosol in the fine-dominated mode at 0.55 m, and surface reflectance at 2.12 m). The look-up table is modified on the basis of a set of new assumed spherical fine aerosol models. 40 The method for surface elevation corrections for Rayleigh scattering is also modified, and negative AODs are allowed to account for positive and negative noise in MODIS observations. The MODIS AOD product was validated through comparison with the Aerosol Robotic Network (AERO- NET) AOD, which is considered as ground truth. The v5.2 AODs demonstrate a better agreement with the AERONET AOD than the previous versions. 26,30,41 Here, the additional impact of the changes from v4.0.1 to v5.2.6 is investigated, especially the relation with PM 2.5. As an example, Figures 5a 5c show the RGB image, the AOD retrieved from v4.0.1, and the AOD retrieved from v5.2.6 from Terra on August 1, 2006 at 16:45 coordinated universal time (UTC), respectively. The most significant difference is the data coverage. There is greater data coverage in v4.0.1 than in v5.2.6 mainly because of the difference in the cloud-screening algorithm. In Pennsylvania, West Virginia, Ohio, Kentucky, Virginia, and Tennessee, large areas are screened out without AOD retrievals in v5.2.6, whereas most of these areas are not screened in v To the south, in Louisiana, Mississippi, and Alabama, v4.0.1 also has less area screened. In western Texas, eastern Kansas, and at the boundary between Illinois and Indiana, v5.2.6 has more coverage. The data coverage is 48.8% for v4.0.1 and 37.6% for v5.2.6 in this case. Figure 5d shows the difference between the AODs for these two versions (i.e., , where represents AOD). The AOD retrievals from v5.2.6 are higher than those of v4.0.1 in some areas (e.g., Indiana, Illinois, Ohio, Kansas, and Texas) and are close to those of v4.0.1 in the other areas. In the areas where v4.0.1 and v5.2.6 have retrievals, the average AOD from v5.2.6 is higher than that from v4.0.1 and the RMS difference is The two versions of the MODIS AOD from Terra are compared in terms of their inversion algorithm and the cloud-screening algorithm. The comparison of inversion algorithm is implemented by picking out those data available from v4.0.1 and v5.2.6 (i.e., coincident data with one-to-one correspondences between v4.0.1 and v5.2.6). By doing this, the effect of the cloud-screening algorithm can be ignored. The comparison of the screening algorithm is implemented by comparing the whole dataset Volume 59 November 2009 Journal of the Air & Waste Management Association 1365

10 Figure 5. AOD over the eastern United States region on August 1, 2006 from Terra: (a) RGB image, (b) v4.0.1 AOD, (c) v5.2.6 AOD, and (d) and the subset obtained above that has one-to-one correspondences between the two versions. Table 5 shows the linear regression between Terra and in different geographical regions for the dataset in which v4.0.1 and v5.2.6 AOD retrievals have one-to-one correspondence. The slope ranges from 0.71 to 0.87, indicating that AOD retrievals from v4.0.1 are smaller than those from v5.2.6 by approximately 13 29%. The intercepts are all small and have positive and negative values. The regression relation of this dataset for Terra is shown in the second line of each row in Table 2. The correlation coefficients have slight improvements of approximately 0.01 Table 5. Linear regression between and from Terra, where is on the vertical axis and is on the horizontal axis. EPA Region Slope Intercept r N , , , , , , , , , ,375 Maximum Minimum in most of the regions for v In region 6 (south), the correlation coefficient of v5.2.6 is less than that of v The difference in this region may be caused by the difference in retrieval algorithms. However, the correlation in this region is moderate, and it only explains a small fraction of the variations of data (20% in region 6). The difference is more likely to be caused by some unknown variables that are buried in the other 80% of variances. Because the set of data with one-to-one correspondence between v4.0.1 and v5.2.6 passes the screening algorithm for both versions, it has the least possibility to be contaminated by the cloud. However, because it is only a subset of the whole dataset for each version, the whole dataset should be examined to have a full view of the AOD and PM 2.5 relation for each version. Figure 6 shows the monthly average number of match-ups per site over different EPA regions. The northern regions such as regions 1 3, 5, 7, 8, and 10 demonstrate seasonal variations of the average number per site and have the largest numbers of match-ups during summer and the smallest numbers of match-ups during winter. The southern regions 4 and 6 do not have such variations. These seasonal variations are caused by the lack of retrievals of AOD due to the larger surface reflectance and more cloud cover during winter. Over all of the regions and all of the months, the numbers of match-ups for v4.0.1 (n ) are larger than those for v5.2.6 (n ), and the numbers of match-ups for v5.2.6 are very close to those with one-to-one correspondence between the two versions (n both ) Journal of the Air & Waste Management Association Volume 59 November 2009

11 there must be areas that have retrievals for v5.2.6 but no retrievals for v4.0.1, which is also noticed in Figure 5. Because the difference between n and n both is often large, the relationship between and PM 2.5 may also vary substantially depending on whether it is determined using observations from the entire dataset (n ) or the dataset of observations with a one-to-one correspondence to the v5.2.6 observations. On the other hand, because the differences between n and n both are small, small differences between the relationships for the whole dataset and the subset having one-to-one correspondence with are expected. These differences can be observed by comparing results in Table 2 from the two datasets. For v4.0.1, the linear correlation coefficients are smaller for the whole dataset than those for the subset in all of the regions. The differences range from 0.05 to The slopes are also smaller for the whole dataset than those for the subset in all of the regions with differences ranging from 5 to 14; this indicates that AOD retrievals for the whole dataset are larger than those for the subset. The differences in the intercepts are small in all of the regions. For v5.2.6, the two datasets have much smaller differences than those for v The differences in correlation coefficients are less than or equal to 0.02 in all of the regions except for region 2, which has a difference of Except region 10, the slopes are all observed to be smaller for the whole dataset, with differences ranging from 0.5 to 3. In region 10, the slope is slightly larger for the whole dataset. Similar to v4.0.1, small differences in intercepts are also observed for all of the regions. For the whole dataset, AOD retrievals from v4.0.1 have many more match-ups than those from v5.2.6, with a percentage differences between 25 and 48% over different regions. However, the correlation coefficients for AOD-PM 2.5 are much larger for v5.2.6 than those for v4.0.1, with differences from 0.07 to Unlike the subset with one-to-one correspondence, the slopes for v4.0.1 are all smaller than those for v There are some occasional higher correlation coefficients for v4.0.1 in some seasons and regions than those of v5.2.6, but with more coincidences, which occur mostly during winter when the correlations are poor. Figure 7 shows the average AOD and the average of the corresponding matched PM 2.5. For v4.0.1, the average AODs for the whole dataset are larger than those for the subset in all of the regions. This indicates that the AODs screened out by the match-up to v5.2.6 have high values. However, the corresponding PM 2.5 values are not larger Figure 6. MODIS AOD-PM 2.5 monthly average number of matchups per site by EPA regions 1 10 (a h, respectively). The statistics show that the average number of match-ups per month per site is between 7.1 and 11.1 for v4.0.1 (n ), between 3.8 and 7.9 for v5.2.6 (n ), and between 3.4 and 7.5 for those with one-to-one correspondence (n both ). The percentage differences between n and n both vary from 19 to 49%. The percentage differences between n and n both are much smaller than those between n and n both usually less than 10% although it can occasionally be as high as 23%. Because the numbers of match-ups are not the same for n and n both, than average values because the graph for average PM 2.5 shows that the average PM 2.5 values matched to the whole dataset are smaller than or similar to those matched to the subset. It can be deduced that the screened AOD retrievals for v4.0.1 are located above the PBL or contaminated by cloud cover. A similar conclusion about the effect of cloud screening was recently published by Hutchison et al. 18 Much smaller differences are observed in most regions between the whole set of v5.2.6 AOD retrievals and the corresponding subset because of the small differences in their numbers. One exception is region 2, where the averages of the two datasets appear to have a large difference. The average AOD retrievals of v5.2.6 are larger than those of v4.0.1 for the subset with Volume 59 November 2009 Journal of the Air & Waste Management Association 1367

12 Figure 7. Average (a) AOD and (b) PM 2.5 for different datasets. In the graph of average AOD, the whole datasets are represented by a 1 in parentheses i.e., v4.0.1 (1) and v5.2.6 (1) and the datasets with one-to-one correspondence are represented by a 2 in parentheses. In the graph of average PM 2.5, v4.0.1 represents PM 2.5 matched to the whole dataset of Terra v4.0.1, v5.2.6 represents PM 2.5 matched to the whole dataset of Terra v5.2.6, and both represents PM 2.5 matched to both sets with one-to-one correspondence. one-to-one correspondence; however, for the whole dataset the average AOD retrievals for v5.2.6 are larger than those for v4.0.1 only in some regions (i.e., regions 2, 5, 7, and 8). In regions 1, 4, 6, and 9, the average AOD retrievals for v4.0.1 are larger than those for v In regions 3 and 10, the average values are very close to each other. This indicates that the cloud-contaminated pixels in v4.0.1 are brighter in regions 1, 4, 6, and 9 than those in the other regions. In general, the changes in the retrieval algorithm and the cloud-screening algorithm from v4.0.1 to v5.2.6 have improved the correlation coefficients in most of the regions and seasons. The change of the surface reflectance assumption and the aerosol models has a relatively small effect on the correlation over most of the regions. With the improved cloud-screening algorithm in v5.2.6, the AOD-PM 2.5 correlation improves a great deal, but it also has considerable reduction in data coverage. CONCLUSIONS The relation between the MODIS AOD and PM 2.5 over the United States for the 10 EPA geographical regions was investigated. The AOD retrievals demonstrate geographical and seasonal variations in their relation with PM 2.5. Good correlations are mostly observed over the eastern United States in summer and fall. Region 4 in the southeast has the highest correlation coefficients at more than 0.6. Region 9 in the southwest has the lowest correlation coefficient at approximately 0.2. The seasonal regression relations are derived for each region so that one is able to use the regression relations to estimate PM 2.5 from the AOD retrievals. Case studies from the independent set show that this method provides a better estimate of PM 2.5 than that from IDEA using a fixed ratio. The seasonal and spatial regression relationships will be implemented in IDEA in the future; however, interpolation and smoothing will need to be used between widely varying neighboring regions. The AOD retrievals from Terra and Aqua sensors have similar relations with PM 2.5, and they also have similar data coverage. Comparisons of v4.0.1 and v5.2.6 of the MODIS AOD from Terra were also made in relation to surface PM 2.5 measurements based on the statistics of a 2-yr match-up dataset of MODIS AOD from Terra and PM 2.5 measurements. The comparisons are performed in terms of their inversion methods and cloud-screening algorithm. If only their inversion methods are considered, AODs from v4.0.1 have lower values than those from v5.2.6 over all EPA regions. The correlation coefficients between AOD and PM 2.5 are higher for v5.2.6 than those for v4.0.1 over all EPA regions except region 6 (south). It is possible that the changes of surface reflectance assumptions and aerosol models in the new version do not work well over region 6. The cloud-screening algorithms are found to have much larger impacts on the AOD and PM 2.5 relation. The AOD retrieval coverages are much larger for v4.0.1 than those for v5.2.6 because of the different cloudscreening algorithm. The correlation coefficients are much smaller for the whole dataset for v4.0.1 than those for the subset having one-to-one correspondence with v5.2.6, which indicates that the extra data in v4.0.1 are most likely contaminated by the cloud cover. Overall, AODs from v5.2.6 demonstrate a higher correlation with surface PM 2.5 ; however, the shortcoming of v5.2.6 is its less data coverage because users may not find it very useful if their areas are not covered most of time. ACKNOWLEDGMENTS This work was supported by NASA Cooperative Agreement NNS06AA02A and NOAA contract DG133E07CN0285. The authors thank Shobha Kondragunta of NOAA/NESDIS and Anthony Wimmers of the University of Wisconsin Madison for their constructive suggestions and discussions. The authors thank the reviewers for comments that have improved the paper. REFERENCES 1. Pope, C.A., III; Burnett, R.T.; Thun, M.J.; Calle, E.E.; Krewski, D.; Ito, K.; Thurston, G.D. Lung Cancer, Cardiopulmonary Mortality and Long-Term Exposure to Fine Particulate Air Pollution; JAMA 2002, 287, Journal of the Air & Waste Management Association Volume 59 November 2009

13 2. Technology Transfer Network (TTN) Air Quality System (AQS); U.S. Environmental Protection Agency; available at airsaqs (accessed November 2008). 3. AIRNOW; U.S. Environmental Protection Agency; available at (accessed November 2008). 4. Koelemeijer, R.B.A.; Homan, C.D.; Matthijsen, J. Comparison of Spatial and Temporal Variations of Aerosol Optical Thickness and Particulate Matter over Europe; Atmos. Environ. 2006, 40, Hoff, R.M.; Christopher, S Critical Review Remote Sensing of Particulate Pollution from Space: Have We Reached the Promised Land? J. Air & Waste Manage. Assoc. 2009, 59, ; doi: / Engel-Cox, J.; Holloman, C.H.; Coutant, B.W.; Hoff, R.M. Qualitative and Quantitative Evaluation of MODIS Satellite Sensor Data for Regional and Urban Scale Air Quality; Atmos. Environ. 2004, 38, Al-Saadi, J.; Szykman, J.; Pierce, R.B.; Kittaka, C.; Neil, D.; Chu, D.A.; Remer, L.; Gumley, L.; Prins, E.; Weinstock, L.; MacDonald, C.; Wayland, R.; Dimmick, F.; Fishman, J. Improving National Air Quality Forecasts with Satellite Aerosol Observations; Bull. Am. Meteor. Soc. 2005, 86, Liu, Y.; Sarnat, J.A.; Kilaru, V.; Jacob, D.J.; Koutrakis, P. Estimating Ground Level PM 2.5 in the Eastern United States Using Satellite Remote Sensing; Environ. Sci. Technol. 2005, 39, Liu, Y.; Franklin, M.; Kahn, R.; Koutrakis, P. Using Aerosol Optical Thickness to Predict Ground-Level PM 2.5 Concentrations in the St. Louis Area: a Comparison between MISR and MODIS; Remote Sens. Environ. 2007, 107, Pelletier, B.; Santer, R.; Vidot, J. Retrieving of Particulate Matter from Optical Measurements: a Semiparametric Approach; J. Geophys. Res. 2007, 112, D Liu, Y.; Park, R.J.; Jacob, D.J.; Li, Q.; Kilaru, V.; Sarnat, J.A. Mapping Annual Mean Ground-Level PM 2.5 Concentrations Using Multiangle Imaging Spectroradiometer Aerosol Optical Thickness over the Contiguous United States; J. Geophys. Res. 2004, 109, D Liu, Y.; Kahn, R.; Koutrakis, P. Estimating Fine Particulate Matter Component Concentrations and Size Distributions Using Satellite Retrieved Fractional Aerosol Optical Depth: Part I Method Development; J. Air & Waste Manage. Assoc. 2007, 57, ; doi: / Liu, Y.; Koutrakis, P; Kahn, R.; Turquety, S.; Yantosca, R.M. Estimating Fine Particulate Matter Component Concentrations and Size Distributions Using Satellite Retrieved Fractional Aerosol Optical Depth: Part II a Case Study; J. Air & Waste Manage. Assoc. 2007, 57, ; doi: / Gupta, P.; Christopher, S.A.; Wang, J.; Gehrig, R.; Lee, Y.; Kumar, N. Satellite Remote Sensing of Particulate Matter and Air Quality Assessment over Global Cities; Atmos. Environ. 2006, 40, Gupta, P.; Christopher, S.A. Seven Year Particulate Matter Air Quality Assessment from Surface and Satellite Measurements; Atmos. Chem. Phys. 2008, 8, Hutchison, K.; Smith, S.; Faruqui, S.J. Correlating MODIS Aerosol Optical Thickness Data with Ground-Based PM 2.5 Observations across Texas for Use in a Real-Time Air Quality Prediction System; Atmos. Environ. 2005, 39, Engel-Cox, J.; Hoff, R.; Rogers, R.; Dimmick, F.; Rush, A.; Szykman, J.; al-saadi, J.; Chu, D.; Zell, E. Integrating Lidar and Satellite Optical Depth with Ambient Monitoring for 3-Dimensional Particulate Characterization; Atmos. Environ. 2006, 40, Hutchison, K.D.; Faruqui, S.J.; Smith, S. Improving Correlations between MODIS Aerosol Optical Thickness and Ground-Based PM 2.5 Observations through 3D Spatial Analyses; Atmos. Environ. 2008, 42, Gupta, P. Ph.D. Thesis, University of Alabama Huntsville, Huntsville, AL, Herman, J.R.; Bhartia, P.K.; Torres, O.; Hsu, C.; Seftor, C.; Celarier, E. Global Distribution of UV-Absorbing Aerosols from Nimbus-TOMS Data; J. Geophys. Res. 1997, 102, Levelt, P.F.; Hilsenrath, E.; Leppelmeier, G.W.; van den Oord, G.H.J.; Bhartia, P.K.; Tamminen, J.; de Haan, J.F.; Veefkind, J.P. Science Objectives of the Ozone Monitoring Instrument; IEEE Trans. Geosci. Remote Sens. 2006, 44, Herman, M.; Deuze, J.L.; Devaux, C.; Goloub, P.; Breon, F.M.; Tanre, D. Remote Sensing of Aerosols over Land Surfaces Including Polarization Measurements and Application to POLDER Measurements; J. Geophys. Res. 1997, 102, North, P.R.J. Estimation of Aerosol Opacity and Land Surface Bidirectional Reflectance from ATSR-2 Dual-Angle Imagery: Operational Method and Validation; J. Geophys. Res. 2002, 107, Liu, Y.; Samat, A.; Coull, B.A.; Koutrakis, P.; Jacob, D. Validation of Multiangle Imaging Spectroradiometer (MISR) Aerosol Optical Thickness Measurements Using Aerosol Robotic Network (AERONET) Observations over the Contiguous United States; J. Geophys. Res. 2004, 109, D Kaufman, Y.; Tanre, J.D.; Remer, L.A.; Vermote, E.F.; Chu, A.; Holben, B.N. Operational Remote Sensing of Tropospheric Aerosols over Land from EOS Moderate Imaging Spectro-Radiometer; J. Geophys. Res. 1997, 102, Remer, L.A.; Kaufman, Y.J.; Tanre, D.; Mattoo, S.; Chu, D. A.; Martins, J. V.; Li, R.-R.; Ichoku, C.; Levy, R.C.; Kleidman, R.G.; Eck, T.F.; Vermote, E.; Holben, B.N. The MODIS Aerosol Algorithm, Products, and Validation; J. Atmos. Sci. 2005, 62, Knapp, K.R.; Vonder Haar, T.H.; Kaufman, Y. Aerosol Optical Depth Retrieval from GOES-8: Uncertainty Study and Retrieval Validation over South America; J. Geophys. Res. 2002, 107, Prados, A.I.; Kondragunta, S.; Laszlo, I.; Ciren, P.; Knapp, K. The GOES Aerosol/Smoke Product (GASP) over North America: Comparisons to AERONET and MODIS Observations; J. Geophys. Res. 2007, 112, D Kokhanovsky, A.A.; Breon, F.-M.; Cacciari, A.; Carboni, E.; Diner, D.; Di Nicolantonio, W.; Grainger, R.G.; Grey, W.M.F.; Holler, R.; Lee, K.-H.; Li, Z.; North, P.R.J.; Sayer, A.M.; Thomas, G.E.; von Hoyningen- Huene, W. Aerosol Remote Sensing over Land: a Comparison of Satellite Retrievals Using Different Algorithms and Instruments; Atmos. Res. 2007, 85, Levy, R.C.; Remer, L.A.; Mattoo, S.; Vermote, E.; Kaufman, Y.J. Second- Generation Operational Algorithm: Retrieval of Aerosol Properties over Land from Inversion of Moderate Resolution Imaging Spectroradiometer Spectral Reflectance; J. Geophys. Res. 2007, 112, D MOD04 Revision History; Goddard Space Flight Center; available at (accessed February 2007). 32. Hoff, R.; Zhang, H.; Jordan, N.; Prados, A.; Engel-Cox, J.; Huff, A.; Weber, S.; Zell, E.; Kondragunta, S.; Szykman, J.; Johns, B.; Dimmick, F.; Wimmers, A.; Al-Saadi, J.; Kittaka, C. Applications of the Three- Dimensional Air Quality System (3D-AQS) to Western U.S. Air Quality: IDEA, Smog Blog, Smog Stories, and AirQuest; J. Air & Waste Manage. Assoc. 2009, 59, ; doi: / Infusing Satellite Data into Environmental Applications; available at (accessed February 2009). 34. Smog Blog; University of Maryland Baltimore County; available at (accessed November 2008). 35. IDEA Product at University of Wisconsin; available at wisc.edu (accessed February 2007). 36. LADDS Web: Level 1 and Atmosphere Archive and Distribution System; National Aeronautics and Space Administration; available at ladsweb.nascom.nasa.gov (accessed February 2007). 37. Tanré, D.; Kaufman, Y.; Herman, M.; Mattoo, S. Remote Sensing of Aerosol Properties over Oceans Using the MODIS/EOS Spectral Radiances; J. Geophys. Res. 1997, 102, Ackerman, S.A.; Strabala, K.I.; Menzel, W.P.; Frey, R.A.; Moeller, C.C.; Gumley, L.E. Discriminating Clear Sky from Clouds with MODIS; J. Geophys. Res. 1998, 103, Martins, J.V.; Tanre, D.; Remer, L.A.; Kaufman, Y.J.; Mattoo, S.; Levy, R. MODIS Cloud Screening for Remote Sensing of Aerosol over Oceans Using Spatial Variability; Geophys. Res. Lett. 2002, 29, Levy, R.C.; Remer, L.A.; Dubovik, O. Global Aerosol Optical Properties and Application to Moderate Resolution Imaging Spectroradiometer Aerosol Retrieval over Land; J. Geophys. Res. 2007, 112, D Chu, D.A.; Kaufman, Y.J.; Ichoku, C.; Remer, L.; Tanre, D.; Holben, B.N. Validation of MODIS Aerosol Optical Depth Retrieval over Land; J. Geophys. Res. 2002, 29, About the Authors Hai Zhang is a research associate at the Joint Center for Earth Systems Technology (JCET) at the University of Maryland, Baltimore County. Raymond M. Hoff is a professor in the Department of Physics and Director of JCET and the Goddard Earth Sciences and Technology Center at the University of Maryland, Baltimore County. Jill Engel-Cox is a senior research scientist at Battelle Memorial Institute in Arlington, VA. Please address correspondence to: Hai Zhang, University of Maryland, Baltimore County/Joint Center for Earth Systems Technology, 5523 Research Park Drive, Baltimore, MD 21228; phone: ; fax: ; hazhang@umbc.edu. Volume 59 November 2009 Journal of the Air & Waste Management Association 1369