Saharan Dust and Wildfire Smoke: An Evaluation of Enhancing AirNow with NASA Satellite Data

Adam N. Pasch1a, Patrick H. Zahn1, Jennifer L. DeWinter1, Michael D. Haderman1, James J. Szykman2, John E. White3, Phillip Dickerson3, Tim S. Dye1, Aaron van Donkelaar4, and Randall V. Martin4
aCorresponding author
1Sonoma Technology Inc., Petaluma, California, USA
2NASA Langley Research Center, Hampton, Virginia, USA
3U.S. Environmental Protection Agency, Research Triangle Park, North Carolina, USA
4Dalhousie University, Halifax, Nova Scotia, Canada

We present two case studies as a way to evaluate the performance of the AirNow Satellite Data Processor (ASDP): a Saharan dust transport event and a wildfire smoke event.

Abstract

The U.S. Environmental Protection Agency’s (EPA) AirNow program provides the public with real-time and forecasted air quality conditions. Millions of people each day use information from AirNow to protect their health. Each hour, AirNow systems generate thousands of products including maps, Keyhole Markup Language (KML) files, data files, and animations to support EPA’s mission to protect public health. The usefulness of the AirNow air quality maps depends on the accuracy and spatial coverage of air quality measurements. Currently, AirNow uses only ground-based measurements, which have significant gaps in coverage in some parts of the United States. As a result, uncertainty is high for contoured air quality index (AQI) levels in regions far from monitors. To improve the accuracy of air quality maps, scientists at EPA, Dalhousie University, and Sonoma Technology Inc., in collaboration with NASA and the National Oceanic and Atmospheric Administration (NOAA), have fused satellite-estimated surface PM2.5 concentrations with ground-based observations via the AirNow Satellite Data Processor (ASDP). The ASDP uses uncertainty information about each data set to create a fused product based on uncertainty weighted-averaging. We present two case studies as a way to evaluate the performance of the ASDP: a Saharan dust transport event and a wildfire smoke event.

Introduction

Figure 1. AirNow surface PM2.5 observations (black dots). Shaded red areas indicate regions for which AirNow provides no air quality information because the prediction standard error is above 4.8 μg/m3 with the currently used inverse distance weighting. Image Credit: Google Earth.

Figure 1. AirNow surface PM2.5 observations (black dots). Shaded red areas indicate regions for which AirNow provides no air quality information because the prediction standard error is above 4.8 μg/m3 with the currently used inverse distance weighting. Image Credit: Google Earth.

The U.S. Environmental Protection Agency’s (EPA) AirNow program uses the Air Quality Index (AQI) to provide the public with easy access to national ambient air quality information. The AQI is a standardized index for reporting air quality based on health effects for five major air pollutants: ground-level ozone (O3), fine particulate matter (PM2.5), carbon monoxide (CO), sulfur dioxide (SO2), and nitrogen dioxide (NO2). AirNow presents near-real-time hourly AQI conditions and daily AQI forecasts with maps of interpolated AQI levels on national, regional, and local spatial scales. The AirNow program has developed a system called the AirNow Information Management System (IMS) (Dye et al., 2008) that blends (or fuses) different data sets.

Exposure to elevated ambient PM2.5 concentrations is associated with adverse cardiovascular and respiratory health effects, and currently more than 42 million people reside in locations with no air quality information. Thus, providing air quality information nationwide is important; however, there are some challenges to providing a gridded nationwide AQI map based on surface observations. Most notably, the United States surface air quality monitoring network is sparse in many areas. In Figure 1, a map of the current AirNow PM2.5 monitors, shaded red areas indicate regions for which the current point-to-grid interpolation method within AirNow (inverse distance weighting) has a prediction standard error above 4.8 μg/m3 (microgram per cubic meter).The large prediction standard error results in such high uncertainty in the interpolated values that no air quality information is provided.

From a public health perspective, there are substantial health benefits for people who take protective action to avoid exposure to high outdoor PM2.5 concentrations. One way to provide additional PM2.5 information is to fill the surface monitoring gaps with PM2.5 estimates derived from satellite aerosol optical depth (AOD) data  (Satellite Remote Sensing and Air Quality video). Satellite retrievals of total column AOD, a measure of extinction of light passing through an atmospheric column, are able to estimate surface PM2.5 concentrations (Zhang et al., 2009; van Donkelaar et al., 2006; 2011; Wang and Christopher, 2003; Engel-Cox et al., 2004).

To improve the accuracy of air quality maps, scientists at EPA, Dalhousie University, and Sonoma Technology Inc., have been working in collaboration with NASA Applied Sciences and NOAA’s Satellite and Information Service (NESDIS) to incorporate satellite-estimated surface PM2.5 concentrations into the EPA’s AQI maps via the AirNow Satellite Data Processor (ASDP Project Overview video). To derive these satellite estimates, we use NASA/NOAA satellite AOD retrievals and GEOS-Chem modeled ratios of surface PM2.5 concentrations to AOD.  GEOS-Chem is a three-dimensional chemical transport model for atmospheric composition driven by meteorological input from the Goddard Earth Observing System (GEOS). The ASDP can fuse satellite-derived PM2.5 concentrations with surface observed PM2.5 concentrations to generate AQI maps with improved estimates.

The goal of ASDP is to provide more detailed AQI information in monitor-sparse locations and augment monitor-dense locations with more information. The case studies described in this paper can be used to evaluate the ASDP’s performance in meeting this goal.

Methodology

Figure 2. AirNow 24-hour average PM2.5 concentrations (colored dots), MODIS Aqua True Color imagery, and 24-hour wind roses (blue triangles) for Aug. 8, 2013 (obtained from AirNow-Tech). The circle highlights the location of the Saharan Dust.

Figure 2. AirNow 24-hour average PM2.5 concentrations (colored dots), MODIS Aqua True Color imagery, and 24-hour wind roses (blue triangles) for Aug. 8, 2013 (obtained from AirNow-Tech). The circle highlights the location of the Saharan Dust.

The ASDP fusion program uses a weighted average approach (Equation 1, below). The satellite-derived PM2.5 concentrations and the interpolated observed PM2.5 data are weighted according to their respective uncertainties (E) at each grid point (i). The lower (higher) the uncertainty, the higher (lower) the degree of confidence in the predicted value; thus, in locations with surface monitors, data from the monitors are weighted more heavily than the satellite estimates, and the fused value equals the observed value (Equation 1, below).

 

equation 1

 

 

 

 

 

For each data set, gridded concentration estimates and the associated uncertainty are required. The uncertainty in the satellite-derived PM2.5 concentrations is described with a one-sigma error envelope (van Donkelaar et al., 2012), whereas the interpolation of the observed PM2.5 concentration data obtained from AirNow uses a variance of prediction (VOP) grid (The VOP is a product of the kriging interpolation, which is a measure of the variance in the predicted value at each grid cell). As a result, uncertainty from the satellite estimates is represented with a one-sigma error envelope, and uncertainty from the observed data is represented as a variance in the daily prediction. However, since the successful application of the weighted average method requires a consistent representation of the uncertainties from each data set, we developed a method to consistently relate the uncertainties in the in situ observation interpolation (kriging) to the uncertainties in the satellite estimates under a variety of different VOP and PM2.5 conditions.

Case Study Results

One way to evaluate the performance of the ASDP fusion program is through case studies. Case studies help to evaluate the program’s day-to-day performance under different atmospheric conditions. The two case studies presented below examine very different types of events: transport of Saharan dust and transport of wildfire smoke. Videos of these case studies are also available as Case Study 1 and Case Study 2.

Saharan Dust:  Aug. 8, 2013

Figure 3. AirNow 24-hour average PM2.5 concentrations (colored dots), MODIS Terra and Aqua AOD (colored pixels), and MODIS Aqua True Color imagery for Aug, 8, 2013 (obtained from AirNow-Tech).

Figure 3. AirNow 24-hour average PM2.5 concentrations (colored dots), MODIS Terra and Aqua AOD (colored pixels), and MODIS Aqua True Color imagery for Aug, 8, 2013 (obtained from AirNow-Tech).

On Aug. 8, 2013, dust from the Sahara Desert was transported thousands of miles across the Atlantic Ocean. Figure 2 shows MODIS True Color imagery of the Gulf of Mexico region, AirNow daily average PM2.5 concentrations (colored dots), and 24-hour wind roses (blue petals indicate wind direction) for selected locations along the Gulf Coast.

As seen in Figure 2, PM2.5 concentrations in Texas and Louisiana were Moderate to Unhealthy for Sensitive Groups (USG); these were well beyond what is expected with southerly winds, which typically transport cool, moist, clean air into the Gulf Coast region. MODIS AOD also indicated aerosols in the region (Figure 3).

The SKIRON is an atmospheric dust forecast model produced by the University of Athens, Greece, predicted dust transport from the Sahara Desert across the Atlantic several days in advance (Figure 4). Figure 4 shows SKIRON-predicted dust concentrations greater than 102 μg/m3 in Texas and Louisiana. Because satellites provide a vertical column average of data, it is also necessary to determine whether the aerosols shown are near the surface. Another NASA satellite, Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO), provided additional information on the vertical structure of the aerosols. The CALIPSO satellite overpass occurs immediately after the MODIS satellites, so the information from CALIPSO can be used to validate the MODIS retrievals. Figure 5 shows dust over the Gulf Coast region. As indicated from the CALIPSO data, dust was present over the Gulf Coast region in the lower 2 km of the atmosphere.

 

 

Figure 4. SKIRON dust model prediction obtained from University of Athens (AM&WFG) for Aug. 8, 2013; shading represents near-ground dust concentrations, shown in μg/m3.

Figure 4. SKIRON dust model prediction obtained from University of Athens (AM&WFG) for Aug. 8, 2013; shading represents near-ground dust concentrations, shown in μg/m3.

Figure 5. Vertical cross-sections of aerosol subtyping obtained from morning and afternoon CALIPSO overpasses (shown in pink [top] and green [bottom] in the inset image) on Aug. 8, 2013. Black circles indicates dust over southern Texas and Louisiana.

Figure 5. Vertical cross-sections of aerosol subtyping obtained from morning and afternoon CALIPSO overpasses (shown in pink [top] and green [bottom] in the inset image) on Aug. 8, 2013. Black circles indicates dust over southern Texas and Louisiana.

 

 

 

 

 

 

 

 

 

 

 

 

 

How did the ASDP perform on this day when we know there was dust present and this dust was near the surface? In this case, the gaps in the ground monitoring network resulted in the interpolation of high PM2.5 concentrations across a wide swath of central Louisiana and eastern Texas (Figure 6a). By comparison, the satellite-estimated surface PM2.5 concentrations were between 8 and 12 μg/m3 for much of the region, with concentrations in some areas in northeastern Texas, southeastern Oklahoma, and Arkansas above 18 μg/m3 (Figure 6b). Since the monitoring network is sparse in some regions, the uncertainty in the interpolated observations is high, and thus the weighting is low (Figure 6c). In contrast, the uncertainty in the satellite estimates is low, and thus the weighting is high in regions far from monitors (Figure 6d). As a result, the fused product (Figure 6e) was weighted more heavily toward the satellite estimates, effectively constraining the interpolated concentrations in eastern Texas. In addition, the regions of USG PM2.5 concentrations were constrained to locations near the monitors where the USG PM2.5 concentrations were measured in northeastern Louisiana and north-central Texas.

Figure 6. Interpolated 24-hour ground-observed PM2.5 concentrations (a) and weights (d); satellite-estimated surface PM2.5 concentrations (b) and weights (e); and fused PM2.5 concentrations (c) for Aug. 8, 2013. The 24-hour ground observed PM2.5 concentrations are denoted by colored dot. The black circles show affected regions in Texas and Louisiana. Fusion can be performed only over areas with satellite data—where no satellite data exist, observed data are used. Image Credit: Google Earth.

Figure 6. Interpolated 24-hour ground-observed PM2.5 concentrations (a) and weights (d); satellite-estimated surface PM2.5 concentrations (b) and weights (e); and fused PM2.5 concentrations (c) for Aug. 8, 2013. The 24-hour ground observed PM2.5 concentrations are denoted by colored dot. The black circles show affected regions in Texas and Louisiana. Fusion can be performed only over areas with satellite data—where no satellite data exist, observed data are used. Image Credit: Google Earth.

Long-range dust transport cases of this magnitude are relatively uncommon and provide a particular challenge for the ASDP. In this case, the ASDP performed well, filling gaps in the monitoring network and limiting the interpolation of high PM2.5 concentrations across areas lacking monitors.

Wildfire Smoke:  Sept. 4, 2013

On Sept. 4, 2013, fires in Kansas and Missouri produced a smoke plume over central and northern Missouri. Figure 7 shows the fire detections and smoke plumes from the Hazard Mapping System (HMS), the MODIS True Color imagery, and 24-hour average PM2.5 concentrations from AirNow.

In this case, the gaps in the ground monitoring network caused the smoke plume to go largely undetected (Figure 8a). However, satellite estimates of ground-level PM2.5 indicated a region of low-Moderate AQI levels extending from north-central Missouri into Iowa and southern Minnesota, coinciding with the region of smoke (Figure 8b). Since, northern Missouri lacks ground monitor coverage, the satellite estimates were used to increase the predicted PM2.5 concentration from low-Good AQI levels (0-4.0 μg/m3) to high-Good (8.0-12.0 μg/m3) (Figures 8c and 8d). The resulting fusion product was able to combine the satellite information with surface monitors and thus account for the smoke plume (Figure 8e). CALIPSO data indicated that smoke and polluted dust was likely near the surface during the morning hours and then aloft in the afternoon (Figure 9).

 

Figure 7. Smoke plumes and fire location (red triangle) from HMS, 24-hour average PM2.5 concentrations from AirNow (colored dots), and MODIS True Color imagery for Sept. 4, 2013 (obtained from AirNow-Tech).

Figure 7. Smoke plumes and fire location (red triangle) from HMS, 24-hour average PM2.5 concentrations from AirNow (colored dots), and MODIS True Color imagery for Sept. 4, 2013 (obtained from AirNow-Tech).

Figure 8. Interpolated 24-hour ground-observed PM2.5 concentrations (a) and weights (d), satellite-estimated surface PM2.5 concentrations (b) and weights (e), and fused PM2.5 concentrations (c) for Sept. 4, 2013. The 24-hour ground observed PM2.5 concentrations are denoted by colored dots. Fusion can be performed only over areas with satellite data—where no satellite data exists, observed data is used. Image Credit: Google Earth.

Figure 8. Interpolated 24-hour ground-observed PM2.5 concentrations (a) and weights (d), satellite-estimated surface PM2.5 concentrations (b) and weights (e), and fused PM2.5 concentrations (c) for Sept. 4, 2013. The 24-hour ground observed PM2.5 concentrations are denoted by colored dots. Fusion can be performed only over areas with satellite data—where no satellite data exists, observed data is used. Image Credit: Google Earth.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Smoke plumes and other high-PM2.5 events over regions with sparse ground monitors provide the perfect opportunity for the ASDP system to fill in gaps by fusing satellite-estimated PM2.5 with interpolated ground observations.

Conclusions

The two case studies, each presenting a unique challenge and an opportunity to evaluate the performance of the ASDP, showed that satellite data can be used successfully to fill in gaps in the current monitoring network. Case Study 1, the Saharan dust transport event, demonstrated the ability of the ASDP and satellite data to help constrain the interpolation of monitoring data. Case Study 2, a wildfire smoke plume in unmonitored areas, demonstrated the ability of the ASDP to provide air quality information in unmonitored locations. In both cases, satellite remote sensing data were used to provide additional information.

More information about the ASDP, including daily ASDP model runs and additional outreach videos and materials, can be found at the project website at asdp.airnowtech.org. This project was funded by NASA Applied Sciences.

 

Figure 9. Vertical cross-sections of aerosol subtyping obtained from morning and afternoon CALIPSO overpasses (shown in pink [top] and green [bottom] in the inset image) on Sept. 4, 2013. Black circles indicates smoke and polluted dust over Missouri and Iowa.

Figure 9. Vertical cross-sections of aerosol subtyping obtained from morning and afternoon CALIPSO overpasses (shown in pink [top] and green [bottom] in the inset image) on Sept. 4, 2013. Black circles indicates smoke and polluted dust over Missouri and Iowa.

 

 

 

 

References

Dye T.S., Chan A.C., Gray E.A., Reed J.E., and Pryden D.A. (2008) AIRNow-International (AIRNow-I) draft specifications document. Software specifications prepared for the U.S. Environmental Protection Agency, Research Triangle Park, NC, by Sonoma Technology, Inc., Petaluma, CA, STI-905403.06-3282, May.

Engel-Cox J.A., Holloman C.H., Coutant B.W., and Hoff R.M. (2004) Qualitative and quantitative evaluation of MODIS satellite sensor data for regional and urban scale air quality. Atmos. Environ, 38(16), 2495-2509, May.

van Donkelaar A., Martin R.V., and Park R.J. (2006) Estimating ground-level PM2.5 using aerosol optical depth determined from satellite remote sensing. Journal of Geophysical Research (Atmospheres), 111, D21201, doi: 10.1029/2005JD006996,2006.

van Donkelaar A., Martin R.V., Levy R.C., da Silva A.M., Krzyzanowski M., Chubarova N., Semutnikova E., and Cohen A.J. (2011) Satellite-based estimates of ground-level fine particulate matter during extreme events: A case study of the Moscow fires in 2010. Atmos. Environ., 45, doi: 10.1016/j.atmosenv.2011.07.068.

van Donkelaar A., Martin R.V., Pasch A.N., Szykman J.J., Zhang L., Wang Y.X., and Chen D. (2012) Improving the accuracy of daily satellite-derived ground-level fine aerosol concentration estimates for North America. Environ. Sci. Technol., (10.1021/es3025319), September. Available at http://pubs.acs.org/doi/abs/10.1021/es3025319.

Wang J. and Christopher S.A. (2003) Intercomparison between satellite-derived aerosol optical thickness and PM2.5 mass:  implications for air quality studies. Geophysical Research Letters, 30(21), 2095-2098.

Zhang H., Hoff R.M., and Engel-Cox J.A. (2009) The relation between Moderate Resolution Imaging Spectroradiometer (MODIS) aerosol optical depth and PM2.5 over the United States: a geographical comparison by U.S. Environmental Protection Agency regions. J. Air Waste Manage., 59(11), 1358-1369, doi: 10.3155/1047-3289.59.11.1358. Available at http://dx.doi.org/10.3155/1047-3289.59.11.1358.

Topic: