SEARCH

SEARCH BY CITATION

Keywords:

  • AIRS;
  • carbon monoxide;
  • biomass burning

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. AIRS CO for INTEX-A/ICARTT
  5. 3. AIRS CO Retrieval Algorithm
  6. 4. AIRS CO Verticality
  7. 5. INTEX-A AIRS CO Comparison
  8. 6. 2004 Yukon Fires and AIRS CO
  9. 7. AIRS CO and Trajectories
  10. 8. AIRS CO and Lidars
  11. 9. AIRS CO Versus MODIS AOD
  12. 10. Conclusions
  13. Acknowledgments
  14. References
  15. Supporting Information

[1] We present observations of transport of tropospheric carbon monoxide (CO) obtained by the Atmospheric Infrared Sounder (AIRS) on board NASA's Aqua satellite during the Intercontinental Chemical Transport Experiment–North America (INTEX-A) field campaign in the summer of 2004, part of the International Consortium for Atmospheric Research on Transport and Transformation (ICARTT). In situ measurements from NASA's DC-8 provide crucial assessment of AIRS midtropospheric (400–500 mbar) CO retrievals. For the nine best INTEX-A profiles, convolution of the in situ profiles with AIRS verticality functions demonstrates version 4 AIRS CO retrievals between 400 and 500 mbar are biased high by approximately 8% with a standard deviation slightly less than 5%. The 400–500 mbar region is the only portion of the version 4 AIRS CO retrievals that can be validated as presented here. In some cases, AIRS CO retrievals may be sensitive to CO in the lower to midtroposphere (800–500 mbar). Focusing on one major episode, we investigate transport of CO from a large fire outbreak in the Alaskan/Canadian Yukon region from 11 to 14 July 2004 and follow it downwind to the southeastern United States and Europe by 22 July 2004. Comparison of AIRS CO maps and forward trajectories from fire locations reveals substantial variations in fire emissions especially emission injection height. Any useful forecast model must control for such fire emission variabilities to predict correctly the downwind impact. To match the forward trajectory analyses with AIRS CO observations requires some fires to have directly injected emissions to at least 500 mbar and perhaps as high as 300 mbar. Ground-based lidar observations show smoke plume altitudes from 3 to 11 km over Wisconsin and from 1 to 4 km over Maryland in agreement with the forward trajectories. The Wisconsin lidar profiles on the afternoon of 18 July 2004 constrain the altitude of CO-rich smoke observed by AIRS and MODIS to lie between 2 and 5.5 km above the surface, roughly 800 to 500 mbar. We find that changes in the correlation between AIRS CO and MODIS AOD reflect changes in the CO vertical distribution during this event. This finding is confirmed by in situ measurements, meteorological analyses, and forward trajectory analyses.

1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. AIRS CO for INTEX-A/ICARTT
  5. 3. AIRS CO Retrieval Algorithm
  6. 4. AIRS CO Verticality
  7. 5. INTEX-A AIRS CO Comparison
  8. 6. 2004 Yukon Fires and AIRS CO
  9. 7. AIRS CO and Trajectories
  10. 8. AIRS CO and Lidars
  11. 9. AIRS CO Versus MODIS AOD
  12. 10. Conclusions
  13. Acknowledgments
  14. References
  15. Supporting Information

[2] With a capacity for daily global observations of numerous atmospheric parameters, NASA's Atmospheric Infrared Sounder (AIRS) on board the Aqua satellite provided a unique perspective for the Intercontinental Chemical Transport Experiment–North America (INTEX-A) field campaign [Singh et al., 2006] and the larger International Consortium for Atmospheric Research on Transport and Transformation (ICARTT) [Fehsenfeld and et al., 2006] during July and August 2004. In particular, AIRS observations of tropospheric carbon monoxide (CO) during INTEX-A/ICARTT illuminate several North American industrial and biomass burning sources and subsequent transport paths [Thompson et al., 2007; A. J. Soja et al., Description of a ground-based methodology for estimating boreal fire emissions for use in regional- and global-scale transport models, unpublished manuscript, 2008].

[3] CO surface concentrations have long been known to exhibit influences from distant sources based on chemical lifetime estimates [Crutzen et al., 1979; Logan et al., 1981; Badr and Probert, 1994], in situ measurements and modeling [Harriss et al., 1992; Law and Pyle, 1993; Thompson et al., 1994; Chatfield et al., 1996; Hannan et al., 2003] and studies of surface concentrations [Novelli et al., 1992; Parrish et al., 1993]. Over the past decade, the impacts of such long-range transport have become evident not only on the intercontinental scale but as hemispheric and potentially global phenomena [Chatfield et al., 1998; Parrish et al., 1998; Forster et al., 2001; Novelli et al., 2003; Duncan et al., 2003; Damoah et al., 2004; Colarco et al., 2004; Honrath et al., 2004].

[4] Satellite observations commencing with the Measurement of Air Pollution from Space (MAPS) instrument on board the Space Shuttle in 1981 illustrated the global distribution of CO and large-scale biomass burning as a significant global CO source [Reichle et al., 1982, 1990; Connors et al., 1989, 1999]. Since 2000, the wealth of CO retrievals from the Measurement Of Pollution In The Troposphere (MOPITT) instrument on NASA's Terra satellite have shown CO transport on time scales of several days [Deeter et al., 2003; Heald et al., 2003; Lamarque et al., 2003; Li et al., 2005; Liu et al., 2005]. Additionally, MOPITT has provided estimates of CO emissions from the 2004 Alaska/Canada fires [Pfister et al., 2005; Emmons et al., 2007; Turquety et al., 2007] and explored correlations between CO signatures and aerosols [Edwards et al., 2004, 2006; Bremer et al., 2004; Kampe and Sokolik, 2007]. Retrievals of tropospheric CO from airborne remote sensors have contributed to our appreciation of long-range transport [McMillan et al., 1996, 2003; McCourt et al., 2004]. Other than AIRS, the most recent satellite instruments to measure tropospheric CO are the Tropospheric Emission Spectrometer (TES) on board NASA's Aura satellite [Jones et al., 2003] and the Scanning Imaging Absorption Spectrometer for Atmospheric Chartography (SCIAMACHY) on board the European Envisat satellite [de Laat et al., 2007].

[5] Launched on board NASA's Aqua satellite on 4 May 2002, AIRS cross-track scanning grating spectrometer coupled with Aqua's cross-track scanning Advanced Microwave Sounding Unit (AMSU) provide vertical profiles of the atmosphere with a nadir 45 km field of regard (FOR) across a 1650 km swath [Aumann et al., 2003; Chahine et al., 2006]. Designed as a prototype next generation temperature and water vapor sounder, AIRS broad spectral coverage (3.7 to 16 μm with 2378 channels) includes spectral features of O3, CO2, CH4, and CO [Haskins and Kaplan, 1992]. AIRS broad swath, infrared spectral coverage, and cloud clearing [Susskind et al., 2003] enable day/night retrievals over nearly 70% of the planet every day with substantial portions of the globe observed twice daily (ascending and descending orbits). Thus, AIRS readily observes global-scale transport from large CO sources on at least daily time scales [McMillan et al., 2005; Stohl et al., 2007a, 2007b].

[6] With biomass burning and oxidation of naturally occurring volatile hydrocarbons accounting for nearly 50% of tropospheric CO [Logan et al., 1981; Thompson et al., 1994], monitoring changes in these sources, changes in anthropogenic sources, and subsequent transport are key to assessing the impact on tropospheric chemistry and near-surface air quality. CO's relatively long lifetime, on average 1/2 to 3 months in different portions of the troposphere, make it an excellent tracer of transport and source variability [Badr and Probert, 1994]. Increases in global tropospheric CO from 1970 to the late 1980s has been linked to increasing anthropogenic emissions [Khalil and Rasmussen, 1988; Yurganov et al., 1997]. Subsequent studies found global tropospheric CO abundances leveled off and began to decrease from 1990 to 2000 because of tighter controls on automobile emissions [Novelli et al., 1994; Khalil and Rasmussen, 1994; Bakwin et al., 1994; Parrish et al., 2002]. However, several recent studies have postulated possible large-scale increases and variations in biomass burning sources in the boreal region due to climate change [Wotawa et al., 2001; Yurganov et al., 2004; Lapina et al., 2006; Kasischke and Turetsky, 2006].

[7] AIRS and its successor satellite instruments are uniquely capable to assess the global impact of such changes in sources and downwind transport. Similar to the instruments that preceded AIRS to orbit, MAPS and MOPITT, AIRS' spectral resolution yields a CO sensitivity broadly peaking in the midtroposphere (300 to 700 mbar) with limited information on the vertical distribution. Thus, although AIRS readily detects CO, interpreting AIRS CO retrievals requires skill and caution. An observed increase in CO abundance by AIRS may indicate more CO is present (stronger source) or that the CO has been lifted into the midtroposphere where AIRS is more sensitive. The latter situation is documented herein. Typically, we observe less CO enhancement near a source than downwind where it has been lifted above the boundary layer. However, AIRS has some sensitivity to CO in the lower troposphere (800–500 mbar) as demonstrated by comparisons to both in situ and ancillary information that identify where the biomass emissions (CO-rich smoke) are located. A full validation of AIRS CO retrievals is beyond the scope of this paper and is hindered by the limited number of CO validation profiles available from INTEX-A. However, the INTEX-A in situ profiles provide an excellent comparison set to assess the performance and accuracy of AIRS CO retrievals in the midtroposphere.

2. AIRS CO for INTEX-A/ICARTT

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. AIRS CO for INTEX-A/ICARTT
  5. 3. AIRS CO Retrieval Algorithm
  6. 4. AIRS CO Verticality
  7. 5. INTEX-A AIRS CO Comparison
  8. 6. 2004 Yukon Fires and AIRS CO
  9. 7. AIRS CO and Trajectories
  10. 8. AIRS CO and Lidars
  11. 9. AIRS CO Versus MODIS AOD
  12. 10. Conclusions
  13. Acknowledgments
  14. References
  15. Supporting Information

[8] AIRS was designed as a prototype next generation atmospheric sounder for NASA and NOAA polar orbiting satellites [Chahine et al., 2006]. As such, AIRS' performance was optimized to retrieve highly accurate temperature (1K RMS error in 1km tropospheric layers) and water vapor profiles (15% RMS error in 2 km tropospheric layers) [Chahine et al., 2006]. In practice, AIRS temperature and water vapor retrievals exceed the target accuracy for cloud-free scenes [Tobin et al., 2006] and are nominally less accurate for cloud-cleared scenes [Divakarla et al., 2006]. The full AIRS team retrieval algorithm is described by Susskind et al. [2003] which includes a detailed discussion of AIRS cloud clearing, temperature, water vapor, and O3 retrievals. AIRS CO retrievals were produced as a research product in the support product files at the NASA Goddard Earth Sciences (GES) Data Information and Service Center (DISC) until August 2007 using a CO retrieval algorithm unchanged since prelaunch planning in 1997.

[9] At the University of Maryland, Baltimore County (UMBC), we operate a research version of the AIRS team algorithm developed by coauthors at the National Oceanic and Atmospheric Administration (NOAA). For the near-real-time (NRT) processing during INTEX-A, we used an intermediate version, v4.2, between the operational algorithm run at the GES DISC, v4.0.9, and the next generation AIRS team algorithm, v5.0. During INTEX-A, the NOAA coauthors provided NRT access to AIRS L1b radiances, 3 to 24 h after data acquisition. Running in NRT with the AIRS research code is far from a routine operational system.

[10] Each evening, after 0000 Universal Time (UT), we executed full retrievals on the L1b radiances for all AIRS granules within the INTEX-A/ICARTT study area (20°N to 70°N, 180°W to 30°E) from the preceding day (UT). Typically we processed 60 granules out of the total of 240 acquired by AIRS globally each day. By the next morning local U.S. east coast time (1400 UT), automated processing routines compiled the AIRS retrievals into web accessible daily maps separated into ascending (PM) and descending (AM) orbit tracks. The PM and AM designations derive from the Aqua satellite's Sun-synchronous equator crossing times of 1330 local time for ascending and 0130 local time for descending orbit tracks. Online posting made the AM and PM maps of AIRS retrieved CO at 500 mbar, total column, and NetCDF data files available for distribution to the INTEX-A/ICARTT Science Teams. In collaboration with R. Pierce (NASA Langley Research Center), we generated 48-h forward trajectories from regions of enhanced 500 mbar CO. These maps, data files, and forward trajectories were used to assist in INTEX-A/ICARTT flight planning.

[11] Since cessation of INTEX-A/ICARTT, we have reprocessed all the AIRS granules to fill gaps arising from missing data in the NRT processing scheme. All maps and data files are available via online interface for the full INTEX-A/ICARTT campaigns: http://asl.umbc.edu/pub/mcmillan/www/index_INTEXA.html. CO is our standard product release for INTEX-A/ICARTT; however, all AIRS products will be available for all AIRS granules we process.

3. AIRS CO Retrieval Algorithm

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. AIRS CO for INTEX-A/ICARTT
  5. 3. AIRS CO Retrieval Algorithm
  6. 4. AIRS CO Verticality
  7. 5. INTEX-A AIRS CO Comparison
  8. 6. 2004 Yukon Fires and AIRS CO
  9. 7. AIRS CO and Trajectories
  10. 8. AIRS CO and Lidars
  11. 9. AIRS CO Versus MODIS AOD
  12. 10. Conclusions
  13. Acknowledgments
  14. References
  15. Supporting Information

[12] As described by McMillan et al. [2005], tropospheric CO abundances are retrieved from AIRS measured radiances in the 4.58–4.50 μm (2183–2220 cm−1) region of the 1–0 vibration-rotation CO fundamental band through numerical inversion of the radiative transfer equation employing the AIRS fast forward radiative transfer model [Strow et al., 2003]. This fast model performs all computations on a set of 100 layers referred to as the AIRS standard layers.

[13] We continue to utilize the prelaunch CO algorithm described by McMillan et al. [2005] where the tropospheric CO profile is divided into a series of vertically overlapping trapezoidal functions empirically determined from prelaunch simulations. The overall methodology of the AIRS CO retrieval algorithm is analogous to the AIRS O3 retrieval algorithm described by Susskind et al. [2003]. Unlike the present AIRS O3 retrieval which performs a statistical regression to acquire a first guess profile, the CO algorithm begins with a single fixed CO profile as its first guess. The prelaunch AIRS CO algorithm utilizes the CO profile from the Air Force Geophysical Lab 1976 standard atmosphere [Anderson et al., 1986] for this single first guess profile.

[14] Figure 1 presents the four AIRS prelaunch CO trapezoids that form the perturbation functions for the profile in the retrieval algorithm. In the middle of the atmosphere, the trapezoids sum to one. At the top and bottom, they sum to 0.5 to lessen impacts from portions of the atmosphere where AIRS has little CO signal. This set of trapezoids was developed on the basis of a limited set of prelaunch retrieval simulations. These simulations showed the peak sensitivity of AIRS to tropospheric CO occurred between 300 and 500 mbar; thus, the narrowest functions were placed in these locations.

image

Figure 1. Trapezoidal perturbation functions used in the AIRS prelaunch CO retrieval algorithm. These functions overlap to minimize discontinuities in the retrieved profiles.

Download figure to PowerPoint

[15] During our AIRS CO validation research, we discovered the prelaunch functions of Figure 1 are insufficient to fully characterize AIRS CO retrievals and vertical sensitivity. However, in this investigation we use the prelaunch algorithm to provide our best estimation of its accuracy. An optimized AIRS CO retrieval algorithm is described by Comer [2006] and is implemented in the next generation AIRS team algorithm, v5.0.14.0. In v5.0.14.0, CO is a new standard product available from the NASA GES DISC. Full reprocessing of all AIRS data began in August 2007 and was completed in January 2008 (not in time for inclusion in this study).

[16] With similar sensitivity to midtropospheric CO as MAPS and MOPITT, AIRS' unique daily global view provides nearly ten times as many retrievals per day as MOPITT and enables process studies of phenomena on 12 to 24 h time scales [McMillan et al., 2005]. However, AIRS spectral resolution (nearly 2 cm−1 in the CO region) does not provide sufficient information on the vertical distribution of CO. AIRS vertical resolution is similar to MAPS and somewhat less than MOPITT or TES. Globally, this version of the AIRS CO retrieval possesses 0.3 to 1.5 degrees of freedom (sum of the eigenvalues). Often, even the largest eigenvector is partially damped [McMillan et al., 2005]. Thus, like MAPS and MOPITT, AIRS CO retrievals are sensitive to a weighted total column of tropospheric CO. Changes in shape of the weighting functions indicate changes in the vertical distribution of CO, sometimes up to differences of more than 100 mbar in the location of peak sensitivity as discussed further below.

4. AIRS CO Verticality

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. AIRS CO for INTEX-A/ICARTT
  5. 3. AIRS CO Retrieval Algorithm
  6. 4. AIRS CO Verticality
  7. 5. INTEX-A AIRS CO Comparison
  8. 6. 2004 Yukon Fires and AIRS CO
  9. 7. AIRS CO and Trajectories
  10. 8. AIRS CO and Lidars
  11. 9. AIRS CO Versus MODIS AOD
  12. 10. Conclusions
  13. Acknowledgments
  14. References
  15. Supporting Information

[17] In more traditional numerical solutions to inverse problems, the vertical weighting of sensitivity is defined as the averaging kernel, A [Rodgers, 2000; Rodgers and Connor, 2003]. In such a formulation, (1 − A) represents the amount of a priori remaining in the final retrieved solution. Although the AIRS team retrieval algorithm is not formulated as a maximum likelihood problem [Susskind et al., 2003], it does compute a quantity, Φ, similar to an averaging kernel but in the reduced measurement space defined by the trapezoids [Maddy and Barnet, 2008]. The 100 layer AIRS CO retrievals (column CO in each radiative transfer layer) are related to this reduced measurement space by a unitary transformation, equation (32) of Susskind et al. [2003] and Φ as the damping factor in equation (36) of Susskind et al. [2003] and below.

  • equation image

Φ represents the amount of information in the radiances and (1 − Φ) is the fraction of the first guess retained in the retrieval [Susskind et al., 2003; Maddy and Barnet, 2008]. λ are the eigenvalues of the eigenvectors from the unitary transformation damped by Δλ. Thus, in the transformed trapezoid space, Φ is the averaging kernel.

[18] Transforming equation (1) back to the reduced measurement space defined by the trapezoids, we obtain the AIRS averaging kernel matrix

  • equation image

where U are the eigenvectors from the unitary transformation [Susskind et al., 2003]. For the prelaunch algorithm, this is a 4 × 4 matrix with each row corresponding to the averaging kernel for the respective trapezoid.

[19] To assess the accuracy of AIRS retrieved midtropospheric CO mixing ratios, we utilize the AIRS verticality function, V [McMillan et al., 2005] to convolve in situ profiles to represent that which AIRS would have observed. The sum of the rows of an averaging kernel matrix, As, represents the fraction of information in the reduced measurement space determined directly from the radiances, and (IAs) represents the fraction of a priori information retained [Rodgers and Connor, 2003]. The coarse layering scheme of the reduced measurement space does not enable a good comparison with high-resolution in situ data. Thus, we sum the rows of As from equation (2) and expand from the 4 trapezoids to the 100 AIRS layers in the same manner as the AIRS retrieval algorithm expands the CO perturbations on the trapezoids back to 100 layer CO columns. This 100 layer function we define as the verticality, V [McMillan et al., 2005],

  • equation image

where Trap are the four, 100 layer trapezoids from Figure 1, a 4 × 100 matrix, and I is a vector of ones 1 × 4.

[20] The verticality, V, defined in equation (3), up-samples the coarse layer averaging kernel sums, As, onto the 100 layer grid in the same manner that the optimal functions are up-sampled [see Susskind et al., 2003, equation (32)]. This process is similar to the column operator described by Rodgers and Connor [2003] but is complicated by the details of the AIRS team retrieval algorithm.

[21] Although the use of trapezoidal perturbation functions and the corresponding interpolation rule (e.g., equation (3)) ensures a smooth retrieval product, the large extent of the bottom function does not permit us to define specifically where within that trapezoid an AIRS CO signature arises. For comparisons of AIRS CO products with in situ data as presented in the following section, we will focus on the 400 to 500 mbar region where the uncertainty in the extent of the perturbation functions is mitigated by the use of thinner trapezoids. This is the only portion of the v4 AIRS CO retrievals that can be validated.

5. INTEX-A AIRS CO Comparison

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. AIRS CO for INTEX-A/ICARTT
  5. 3. AIRS CO Retrieval Algorithm
  6. 4. AIRS CO Verticality
  7. 5. INTEX-A AIRS CO Comparison
  8. 6. 2004 Yukon Fires and AIRS CO
  9. 7. AIRS CO and Trajectories
  10. 8. AIRS CO and Lidars
  11. 9. AIRS CO Versus MODIS AOD
  12. 10. Conclusions
  13. Acknowledgments
  14. References
  15. Supporting Information

[22] One goal of INTEX-A was to provide in situ measurements for validation of satellite remote sensing observations [Singh et al., 2006]. Of 26 flights of NASA's DC-8 during INTEX-A [Singh et al., 2006], nine included DC-8 spiral profiles (ascent or descent) most suitable for comparison to AIRS CO retrievals. These nine flights and their comparison statistics are summarized in Table 1 where the timing column refers to the proximity of the start of the spiral to AIRS overpass time. Our selection criteria for this comparison required a DC-8 spiral within 1 h of AIRS overpass, at least three acceptable AIRS CO retrievals within 100 km of the spiral's center point, and a vertical extent from at least 250 to 900 mbar. Other DC-8 flights either flew spirals for MOPITT validation, did not fly spiral profiles, flew spirals in areas where we have no AIRS retrievals, or flew spirals with insufficient altitude coverage.

Table 1. INTEX-A DC-8 Profile Matchups to AIRS and Comparisons of in Situ and Proxy Convolved in Situ Mixing Ratios to AIRS CO Retrievalsa
DC-8 NumberDateLocationTiming With AIRSAIRS Overpass Time (UT)DACOM – AIRS (%)Convolved DACOM – AIRS (%)
  • a

    Percent errors were computed for mean mixing ratios between 400 and 500 mbar with DACOM taken as truth. DC-8 numbers correspond to the flight designations given by Singh et al. [2006, Table 5a].

31 JulPacificcoincident21.81−3.0−1.5
58 JulIllinois1 h18.63−21−12
610 JulSouth Carolinacoincident18.39−53−10
712 JulOklahomacoincident19.83−23−2.1
815 JulWisconsin1 h18.75−23−8.5
918 JulAtlantic0.5 h15.98−10−4.4
1020 JulIllinoiscoincident19.03−28−7.3
1122 JulGulf of Maine1 h17.19−19−7.3
1811 AugKentucky0.1 h18.40−50−16
Mean    −25−7.7
σ    174.7

[23] Figures 2a and 2b show the AIRS 500 mbar CO retrievals for two of these nine cases: 1 and 20 July 2004, the DC-8 flight track (magenta), and the locations of the AIRS validation profiles. AIRS pixels are missing either in gaps between orbits or where retrievals were rejected because of clouds. The 1 July flight intercepted a plume of Asian pollution over the Pacific Ocean. The 20 July flight crossed the large Alaskan/Canadian fire smoke plume over the southeastern United States on several occasions. Figures 3a and 3b present the in situ CO measurements from the DC-8 Differential Absorption CO Measurement (DACOM) instrument [Sachse et al., 1987], AIRS CO retrievals, AIRS first guess profile, and convolved DACOM profiles. First, we compare the DACOM profiles (heavy dashed red lines) to the AIRS retrievals (thin blue solid line = mean of AIRS retrievals within 100 km; blue dashed lines = ±1 σ). The 1 July profile clearly indicates a CO-rich plume above 500 mbar with clean air below 600 mbar. In contrast, the 20 July profile shows relatively clean air above 750 mbar and polluted conditions in the boundary layer.

image

Figure 2. The individual AIRS CO retrievals at 500 mbar are plotted along with the flight track of the NASA DC-8 (black and white dashed line) for the DC-8 profile matchups of (a) 1 July 2004 and (b) 20 July 2004. All AIRS retrievals within 100 km of the DC-8 spiral profiles are indicated by the black circles.

Download figure to PowerPoint

image

Figure 3. (a and b) For the matchups mapped in Figure 2, the in situ CO profiles from the DC-8 DACOM (heavy red dashed) along with the closest AIRS CO retrievals mean (thin blue) and standard deviation (thin blue dashed), the AIRS first guess CO profile (heavy black dotted), and the in situ convolved with the verticality functions of Figure 4 (thick green). Differences in terms of ppbv are listed.

Download figure to PowerPoint

[24] Directly comparing the in situ CO measurements to the AIRS CO retrievals in Figures 3a and 3b, we note the in situ measurements fortuitously intersect the AIRS retrievals on 1 July between 400 and 500 mbar. On 20 July, the AIRS retrievals are larger than the in situ measurements down to the boundary layer. As previously mentioned, AIRS CO sensitivity generally peaks between 400 and 500 mbar. Comparing the mean in situ in this layer to the mean retrievals, we see AIRS is larger in both cases, 3.8 ppbv (3%) on 1 July and 24.4 ppbv (28%) on 20 July. Similar results for all nine INTEX-A comparisons appear in Column 5 of Table 1. The overall comparison to the raw DACOM data indicates AIRS CO retrievals are biased high by 25% between 400 and 500 mbar and have much larger errors elsewhere.

[25] Our next step in comparing in situ and retrieved CO requires convolving the in situ profile with the AIRS CO verticality function. Figure 4 shows the mean AIRS CO verticality functions for all CO retrievals within a 100 km radius of the center of the respective DC-8 spiral profile. These retrievals are marked in Figures 2a and 2b by black circles. Though the mean verticality functions have similar peak magnitudes and similar values near 400 mbar, for these two cases they have different shapes. The 1 July verticality function (dashed line) peaks more sharply at 400 mbar, and the 20 July verticality function (solid line) is more broad, peaks at 500 mbar, and is larger below 500 mbar. In part, these shape changes reflect the differences in the two true in situ profiles. The extent of the bottom trapezoid (500 mbar to the surface) precludes further specification in the verticality.

image

Figure 4. Mean CO verticality functions for the closest AIRS retrievals to the locations of the DC-8 spirals on 1 July (dashed) and 20 July (solid).

Download figure to PowerPoint

[26] Convolving the in situ mixing ratio profile with the verticality function yields the thick solid green lines in Figure 3. In both cases, this convolution brings the in situ measurements closer to the AIRS retrievals. It is noteworthy that the convolved profile of 1 July dramatically changes shape below 500 mbar and changes slope below 700 mbar. These changes reflect the lack of information for these AIRS CO retrievals in this atmospheric region because the convolved in situ has adopted the general shape of the AIRS first guess profile at pressures greater than 700 mbar. However, the fact that the 20 July convolved in situ does not change shape similarly below 700 mbar could indicate those AIRS CO retrievals contain information in the lower troposphere and perhaps to the boundary layer. Here, the extent of the bottom trapezoid limits AIRS ability to further specify the vertical sensitivity. Convolving the in situ with the verticality function reduces the mean 400 to 500 mbar CO mixing ratio differences to 1.9 ppbv (1.5%) on 1 July and 7.7 ppbv (7.3%) on 20 July. Although AIRS is still biased high, the bias is reduced to less than 10%. Results for all nine INTEX-A comparisons appear in column 6 of Table 1.

[27] These nine INTEX-A matchups form a limited set of validation cases and do not cover a global range of conditions. However, these first comparison results point to the reliability of v4 AIRS CO retrievals only in the 400–500 mbar region and if interpreted in a geophysical context. A more detailed and comprehensive validation effort is underway as part of optimization and improvement of the AIRS CO retrieval algorithm for v5.0.14.0 of the AIRS team algorithm [Comer, 2006; Warner et al., 2007]. Preliminary results show that v5's nine CO trapezoidal perturbation functions and full averaging kernels permit validation of AIRS CO retrievals between 300 and 900 mbar with a bias <10% and a standard deviation of approximately 5% [Evans et al., 2007].

6. 2004 Yukon Fires and AIRS CO

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. AIRS CO for INTEX-A/ICARTT
  5. 3. AIRS CO Retrieval Algorithm
  6. 4. AIRS CO Verticality
  7. 5. INTEX-A AIRS CO Comparison
  8. 6. 2004 Yukon Fires and AIRS CO
  9. 7. AIRS CO and Trajectories
  10. 8. AIRS CO and Lidars
  11. 9. AIRS CO Versus MODIS AOD
  12. 10. Conclusions
  13. Acknowledgments
  14. References
  15. Supporting Information

[28] Lagrangian experiments following polluted air from the east coast of the United States across the Atlantic Ocean to Europe were a major goal of the INTEX-A [Singh et al., 2006], Intercontinental Transport of Ozone and Precursors (ITOP-2004) [Lewis et al., 2007] and the Intercontinental Transport and Chemical Transformation (ITCT-2k4) [Methven et al., 2006] field experiments coordinating observations as part of ICARTT [Fehsenfeld and et al., 2006]. From AIRS' perspective, the most obvious such transport event involved emissions from large fires burning in Alaska and Canada. The 2004 Alaskan fire season was the worst on record with more than 2.7 × 106ha burned, nearly ten times the average [Damoah et al., 2005, 2006; Fuelberg et al., 2007; Kasischke and Turetsky, 2006]. Fuelberg et al. [2007] describe the meteorological conditions leading to the Yukon fires and the ensuing long-range transport of the smoke.

[29] A large fire outbreak occurred in the Alaskan and Canadian Yukon region on 11 to 14 July 2004 [Hoff et al., 2005]. Satellite-detected fire counts in the region increased from a few hundred on 10 July to nearly 6000 on 12 and 13 July as illustrated by the hot spot time series in Figure 5. Hot spot information was obtained from the NOAA Satellite Service Division archive as a composite of all satellite fire detections (http://www.ssd.noaa.gov). The peak at the beginning of July is the end of a previous fire outbreak in June 2004. Numerous fires over other areas of Alaska and Canada burned during different portions of INTEX-A/ICARTT. Here, we focus on this one large episode that produced a several day heterogeneous pulse of CO.

image

Figure 5. July 2004 time series of satellite-detected hot spots in Alaska and the Canadian Yukon.

Download figure to PowerPoint

[30] The Lagrangian chemical transformations of this plume have been explored in recent studies using a combination of ground- and air-based in situ measurements, satellite observations, and several chemical transport models [de Gouw et al., 2006; Val Martín et al., 2006; Cook et al., 2007; Lewis et al., 2007; Ravetta et al., 2007; Real et al., 2007]. Several of these studies find mid-upper tropospheric injection of biomass burning emissions are required to explain the downwind observations [de Gouw et al., 2006; Val Martín et al., 2006; Methven et al., 2006] and several find the downwind plumes mixed into the lower troposphere and boundary layer [de Gouw et al., 2006; Val Martín et al., 2006; Morris et al., 2006; Ravetta et al., 2007]. Studies of an earlier smoke plume in July 2004 also indicate high-altitude injection followed by downwind descent into the boundary layer [Warneke et al., 2006; Duck et al., 2007].

[31] Figure 6 colocates the fire sites on 11 and 14 July 2004 with individual AIRS 500 mbar CO retrievals from afternoon (PM) orbits. The hot spots cluster into the various fire complexes with very dense clustering on the peak days of 12 and 13 July. Some CO enhancement appears over or very near the fire complexes on some days; however, most CO enhancement occurs downwind to the east of the fires until 14 July when much of Alaska is blanketed in enhanced CO. Overlapping AIRS retrievals at this latitude arise from the high orbital inclination of the Aqua satellite.

image

Figure 6. For (a) 11 July 2004 and (b) 14 July 2004, AIRS 500 mbar CO individual retrievals appear as the solid colored circles with the corresponding color scale at the bottom. Locations of each day's satellite-detected hot spots are plotted as open black squares.

Download figure to PowerPoint

7. AIRS CO and Trajectories

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. AIRS CO for INTEX-A/ICARTT
  5. 3. AIRS CO Retrieval Algorithm
  6. 4. AIRS CO Verticality
  7. 5. INTEX-A AIRS CO Comparison
  8. 6. 2004 Yukon Fires and AIRS CO
  9. 7. AIRS CO and Trajectories
  10. 8. AIRS CO and Lidars
  11. 9. AIRS CO Versus MODIS AOD
  12. 10. Conclusions
  13. Acknowledgments
  14. References
  15. Supporting Information

[32] Figure 7 (left) presents 1° × 1° gridded AIRS PM CO retrievals at 500 mbar (near the peak of AIRS tropospheric sensitivity) over the INTEX-A/ICARTT study area for 12, 14, 16, 18, 20, and 22 July 2004. Similar maps were created in NRT in support of INTEX-A/ICARTT flight planning as discussed in section 2. For brevity, the AM maps created in NRT are not shown here. This sequence of maps illustrates the movement of the large CO plume from the aforementioned fires in the Alaskan/Canadian Yukon region starting on 12 July. By 16 July over Ontario, Canada, the plume divides into a southward moving portion, the “U.S. plume,” and an eastward moving portion, the “Atlantic plume.” By 22 July, the U.S. plume stagnated over the southeastern United States as far south as the coast of the Gulf of Mexico, and the Atlantic plume wrapped around a cyclonic system near Ireland producing enhanced CO over coastal western Europe. Gray areas correspond to regions of no AIRS data due to orbital gaps, locations of rejected AIRS retrievals due to thick clouds (greater than 80%), or other retrieval failures.

image

Figure 7. (left) The 1° binned AIRS 500 mbar CO retrievals for PM orbits on 12, 14, 16, 18, 20, and 22 July 2004 over the INTEX-A/ICARTT study area. (right) Location of forward trajectories at 1200 UT on 12, 14, 16, 18, 20, and 22 July 2004 initialized at 500 mbar at 1200 UT over the 12 July satellite-detected hot spots.

Download figure to PowerPoint

[33] To model the three-dimensional transport of CO emitted from the Alaskan/Canadian forest fires we performed a set of forward trajectory simulations. These simulations were computed with the Goddard Kinematic Trajectory Model [Schoeberl and Sparling, 1995] using National Center for Environmental Prediction (NCEP) reanalysis winds [Kalnay et al., 1996]. Figure 7 (right) illustrates a sequence of forward trajectory simulations initialized over the fire locations on 12 July at 1200 UT at 500 mbar with parcel locations shown at 1200 UTC on the subsequent days of 14, 16, 18, 20, and 22 July 2004. Note the similarity between this simple set of trajectories and the gross motions of the CO emissions tracked by AIRS. However, this single set of trajectories initialized at one time and at one altitude above the satellite-detected hot spots hints at the complexity of the transport. Starting from a set of clustered hot spots in Figure 7b, the trajectories show stretching and folding as the parcels ascend and descend downwind.

[34] Beginning at 1200 UT on 11 July, air parcels were initialized every 12 h until 1200 UT on 14 July at four different altitudes over the location of each day's satellite-detected hot spots. Parcels were followed downwind for 10 days. Because of the large number of hot spots in this area each day (see Figure 6), trajectories were launched from every third hot spot to reduce the computational load. Given the number of hot spots associated with each fire complex, this reduced number of trajectories suffices to characterize the downwind flow.

[35] A priori, the injection height for emissions from each hot spot remains an unknown. Therefore, in our forward trajectory simulations, we initialized parcels at four altitudes above the surface, 850, 700, 500, and 300 mbar. By tracking the motions of each group of parcels and comparing them to the AIRS CO maps, we gain insight into the contribution to the downwind fire plume from each emission height above each fire complex. Though only 4 days long, this forest fire transport episode was not a single release of CO into the troposphere. A more detailed analysis of the photochemical and thermodynamic influences on the CO rich smoke plume with mixing into the boundary layer remains the focus of ongoing research.

[36] At 500 mbar, Figure 7 shows initial flow to the southeast as the entire grouping of parcels appear to stretch out. The west end of the trajectories at 1200 UT on 14 July (14.5) complete a loop over Alaska as they are caught in the Alaskan high. At 1200 UT on 16 July (16.5), these parcels have traveled to locations along the northwestern Canadian coast of the Beaufort Sea. The eastern end of the 1200 UT 14 July (14.5) trajectories have moved to the eastern coast of Baffin Island by 1200 UT on 16 July (16.5). The central grouping from 1200 UT on 14 July (14.5) has continued southeast, and the eastern end of this grouping has descended over northern Minnesota by 1200 UT on 16 July (16.5). The center has folded on itself with an east moving bulge just southwest of Hudson Bay.

[37] Figure 8 reveals the divergence of the U.S. and Atlantic plumes as seen in the AIRS CO maps of Figure 7: stretching and folding into multiple streamers of CO. This stretching and folding occurs throughout the atmosphere, but the most southward portion of the U.S. plume arises from parcels initialized at 500 mbar above only some of the fires. The two plumes also differ in their vertical motion. The U.S. plume descends from 500 mbar, and the Atlantic plume exhibits ascent for all trajectories with no parcels moving off the Newfoundland coast below 600 mbar. The 1200 UT 20 July (20.5) and 1200 UT 22 July (22.5) trajectory maps in Figure 7 indicate portions of the Atlantic plume subside as they wrap around a low-pressure system off the west coast of England. Other portions of the Atlantic plume experience uplift and folding on the west side of this low between Greenland and Iceland.

image

Figure 8. Trajectory locations at (left) 1200 UT on 17 July (17.5) and (right) 1200 UT on 19 July (19.5) initialized at all four pressure levels, 300, 500, 700, and 850 mbar, at 1200 UT over the 12 July satellite-detected hot spots.

Download figure to PowerPoint

[38] The trajectories in Figures 7 and 8 reveal the U.S. plume to be a succession of streamers at different altitudes. Although the first streamers descend to at least 800 mbar, subsequent southward incursions occur at higher altitudes. However, all the trajectories that reach the farthest south in the United States started at 500 mbar. Continuing this analysis with the additional complexity of trajectories launched on multiple days, Figure 9 shows emissions from 12 and 13 July contributed to the most southward moving portion of the U.S. plume.

image

Figure 9. Location of parcels at 0000 UT on 21 July for trajectories initialized at 500 mbar over the fires at (a) 1200 UT on 11 July (11.5), (b) 1200 UT on 12 July (12.5), (c) 1200 UT on 13 July (13.5), and (d) 1200 UT on 14 July (14.5), starting from each day's respective satellite-detected hot spots.

Download figure to PowerPoint

[39] Further investigation of all the trajectories demonstrates that only parcels initialized over the most northeastern Alaskan and northern Canadian fires travel to the United States Gulf Coast (Table 2). Therefore, we conclude that emissions from the fire complexes in these northern and eastern areas reached altitudes of at least 500 mbar and potentially to 300 mbar or above. These emission injection heights are consistent with modeling of NOAA P-3 in situ measurements during ICARTT [de Gouw et al., 2006], independent satellite observations from the Total Ozone Mapping Spectrometer (TOMS) and the Polar Ozone and Aerosol Measurement (POAM III) instrument of smoke aerosols in the lower stratosphere from one of the Yukon fires in late June 2004 [Damoah et al., 2006], and inverse modeling of MOPITT CO retrievals for 2004 [Turquety et al., 2007].

Table 2. Locations of Fires That Contribute to Enhanced CO Over the Southeastern United States and Along the Gulf of Mexicoa
Trajectory Start Date/TimeStarting LocationU.S. Ending Location
  • a

    All trajectories starting at 500 mbar.

11 Jul 1200 UT137.75°W, 67.2°NGulf coast
12 Jul 1200 UT146.7°W, 67.25°NGulf coast
12 Jul 1200 UTall from 66–67.25°N and 150–145°Wsoutheast
12 Jul 1200 UT137.8°W, 66.55°Nsoutheast
13 Jul 1200 UT146.75°W, 67.2°NGulf coast
13 Jul 1200 UT145.7°W, 67.28°NGulf coast
13 Jul 1200 UTall others N of 66.25°N and E of 148°Wsoutheast
14 Jul 00-UT138.2°W, 67°NGulf coast

[40] Figure 10 illustrates the overall complexity of the trajectories from our simulations with all parcels initialized every 12 h from 1200 UT on 11 July to 1200 UT on 14 July (11.5–14.5) shown at their respective locations on 21 July. Clearly, only parcels in our simulations originating at 700 and 500 mbar over the Alaskan and Canadian hot spots contributed to CO over the southeastern United States. Moreover, only parcels starting at 500 mbar travel as far south as the Gulf of Mexico. Parcels from all but the 1200 UT 14 July (14.5) launch time travel to both the Gulf of Mexico and the coast of Europe. However, there are distinct filaments of parcels in both regions with origins from different launch times.

image

Figure 10. Location of parcels at 0000 UT on 21 July for all trajectories initialized every 12 h starting 1200 UT on 11 July (11.5) to 1200 UT on 14 July over the respective day's satellite-detected hot spots: parcels starting at (a) 300 mbar, (b) 500 mbar, (c) 700 mbar, and (d) 850 mbar.

Download figure to PowerPoint

[41] Recent modeling efforts with buoyant plume rise parameterizations for South American fires show promising comparisons to AIRS 500 mbar CO retrievals [Freitas et al., 2006]. Accurate forecasts of air quality impacts from such distant sources will require detailed NRT data for emission heights and emission rates from each hot spot.

[42] During the INTEX-A/ICARTT time frame, anthropogenic emissions from North America were overwhelmed by the Yukon smoke plumes. However, a careful examination of the AIRS CO maps for 12 and 14 July in Figure 7 show possible anthropogenic CO enhancements along and off the United States east coast. Similar features appear in other AIRS CO maps from the full INTEX-A/ICARTT period but have yet to be further investigated.

8. AIRS CO and Lidars

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. AIRS CO for INTEX-A/ICARTT
  5. 3. AIRS CO Retrieval Algorithm
  6. 4. AIRS CO Verticality
  7. 5. INTEX-A AIRS CO Comparison
  8. 6. 2004 Yukon Fires and AIRS CO
  9. 7. AIRS CO and Trajectories
  10. 8. AIRS CO and Lidars
  11. 9. AIRS CO Versus MODIS AOD
  12. 10. Conclusions
  13. Acknowledgments
  14. References
  15. Supporting Information

[43] During the INTEX-A/ICARTT field deployment, we relied on data from several ground-based lidar instruments to provide more precise information on the vertical distribution of smoke from the Alaskan/Canadian fires. Fortuitously, as the U.S. plume crossed into the United States over Lake Superior, its eastern edge passed over the continuously operating High Spectral Resolution Lidar (HSRL) at the University of Wisconsin in Madison, Wisconsin [Eloranta, 2005]. As the U.S. plume slowly drifted toward the east along with continuing smoke behind the leading edge, it passed over the Elastic Lidar Facility (ELF) at UMBC near Baltimore, Maryland [Engel-Cox et al., 2006].

[44] Figure 11 presents vertical cross sections of HSRL data from 18 to 19 July in both backscatter cross section and linear depolarization. In the backscatter cross section (Figure 11, top), redder (middle gray) colors indicate thicker aerosol layers. The depolarization is particularly useful for distinguishing between anthropogenic aerosols, larger smoke aerosols, and cirrus ice particles. Generally, spherical anthropogenic aerosols produce little depolarization (darker blues or grays in the lowest 2 to 3 km of Figure 11 (bottom)). The larger and more elongated smoke aerosols produce moderate depolarization (green to orange or middle gray layers in Figure 11 (bottom)), and ice crystals in cirrus clouds strongly depolarize the lidar returns (red or white areas in Figure 11 (bottom) most prevalent between 0 and 6 h on 19 July between 7 and 11 km). Examination of Figure 11 indicates several different smoke plumes crossed the site at different altitudes on 18 and 19 July as suggested by the trajectory simulations discussed in the previous section.

image

Figure 11. Time series of (top) lidar backscatter cross section and (bottom) linear depolarization from the HSRL instrument in Madison, Wisconsin, from 18 to 19 July 2004.

Download figure to PowerPoint

[45] The first smoke appears between 3 and 4 km early on 18 July and appears to mix into the top of the boundary layer beginning at 0800 UT. From 0300 to 1300 UT, several thinner smoke plumes are evident as 0.5 to 2 km thick layers from near 5 km to above 11 km. It is noteworthy that some layers are much more distinct in depolarization when compared to backscatter. The thickest smoke plume appears over Madison commencing approximately at 1700 UT on 18 July and continuing until 1200 UT on 19 July and ranges in altitude from 2 to 5.5 km above the surface, roughly 800 to 500 mbar. This time period covers the AIRS CO observations of 18 July as depicted in Figure 7g and indicates AIRS is sensing CO-rich smoke in the lower midtroposphere. However, the results of the CO validation section, section 5, demonstrate v4 AIRS CO retrievals most likely underestimate the true CO abundances in the lower midtroposphere. Again, there appears to be mixing into the top of the boundary layer. Note the general descent of this thicker plume with time and the decreasing height of the mixed layer and cirrus clouds. The previously discussed trajectory simulations also showed this decrease in height that occurred as the air slid down the west side of the persistent trough across the eastern United States [Fuelberg et al., 2007].

[46] As the U.S. plume moved southward into the central and southeastern United States, it began as several distinct smoke plumes at a range of altitudes on 18 July and ended as a vertically thicker plume in the lower midtroposphere (2–6 km) descending as it traveled south. The lidar reveals the U.S. plume as a series of separate plumes of variable horizontal and vertical extent. As discussed in the previous section, the complexity of the downwind plumes results from the temporal and spatial variations in the number of large, active fires and from the intricate details of the downwind transport during this episode.

[47] Further evidence for the heterogeneity of this smoke event comes from the Polar ELF lidar near Baltimore, Maryland [Engel-Cox et al., 2006]. Figure 12 shows aerosol backscatter from 1150 to 2021 UT on 20 July. Two distinct smoke plumes are evident in this lidar time series. The first plume was from 1200 to 1400 UT between 1 and 2 km, and the second plume appeared near 1600 UT above 4 km. The lower plume may have mixed into the rising boundary layer in the afternoon as evidenced by the decreasing returns from the plume near 2 km between the fair weather cumulus clouds marking the top of the boundary layer (red or white blobs starting near 1 km before 1500 UT and rising to nearly 2 km by 1950 UT).

image

Figure 12. Time series of lidar backscatter cross section from the ELF instrument at UMBC near Baltimore, Maryland, on 20 July 2004.

Download figure to PowerPoint

[48] The smoke mixing into the boundary layer from this event contributed to one of the worst air quality outbreaks in southeastern, mid-Atlantic, and northeastern United States during the summer of 2004. Morris et al. [2006] used a combination of AIRS CO retrievals, Total Ozone Mapping Spectrometer (TOMS) aerosol indices and back trajectory analyses to evaluate the impact of Alaskan/Canadian smoke on boundary layer ozone over Houston, Texas, on 18 and 19 July 2004 as measured by ozonesonde and aircraft in situ O3 profiles. Hoff et al. [2005] found the concentration of fine aerosol particulates, those smaller than 2.5 μm, peaked at nearly 60 μg/m3 on 21 through 23 July in Baltimore, Maryland, as the U.S. plume drifted over the Eastern United States. Although this may be one of the most extensively measured events to date, it is not the first time biomass burning emissions have been shown to impact distant boundary layer locations [Forster et al., 2001; Honrath et al., 2004].

9. AIRS CO Versus MODIS AOD

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. AIRS CO for INTEX-A/ICARTT
  5. 3. AIRS CO Retrieval Algorithm
  6. 4. AIRS CO Verticality
  7. 5. INTEX-A AIRS CO Comparison
  8. 6. 2004 Yukon Fires and AIRS CO
  9. 7. AIRS CO and Trajectories
  10. 8. AIRS CO and Lidars
  11. 9. AIRS CO Versus MODIS AOD
  12. 10. Conclusions
  13. Acknowledgments
  14. References
  15. Supporting Information

[49] The lack of vertical specificity is a major frailty of the previously described AIRS prelaunch CO retrieval algorithm. The averaging kernels do change in response to variations in the CO vertical distribution; however, these mainly manifest as subtle changes in shape (see Figure 4). An apparent increase in CO abundance could indicate either a real increase in CO mixing ratio or ascent of CO-rich air into the midtroposphere (300 to 700 mbar) where AIRS retrievals are most sensitive. In the case of the CO-rich smoke plume evident in Figure 7, there is no significant in situ production of CO in the smoke plume as it moves. For the Atlantic plume, no significant anthropogenic CO sources were encountered. Therefore, the observed downwind increases in CO abundance in the Atlantic plume at 500 mbar indicate changes in the vertical distribution of CO.

[50] The previously discussed trajectory simulations indicate such changes in the vertical distribution of the CO-rich smoke plume occurred as the U.S. and Atlantic plumes diverged between 16 and 20 July, see Figures 7 and 8. To obtain a direct measure of the changes in vertical distribution of the CO-rich smoke, we examined the evolution of the correlation between AIRS 500 mbar CO and the total aerosol optical depth (AOD) as retrieved from the Moderate Resolution Imaging Spectroradiometer (MODIS) also on board Aqua. MODIS AOD are largely insensitive to the altitude distribution of the aerosols [Chu et al., 2003]. Thus, assuming aerosol scavenging is negligible as the plume moves downwind [Singh et al., 2006] and the CO produced by the fires remains in the same air mass as the smoke, we deduce that changes in the correlation of AIRS CO and MODIS AOD reflect changes in the CO vertical distribution. If CO-rich smoke was uplifted into AIRS' region of peak sensitivity, 300–700 mbar, AIRS would retrieve larger CO abundances, MODIS would measure the same AOD, and the slope of the correlation would steepen. Most of the downwind AOD is due to fine smoke particles while the larger flyash fallout is expected near the fires.

[51] The 12 panels in Figure 13 present the daily evolution of the smoke plume from 15 to 20 July 2004 as seen in version 4 MODIS total AOD (left column) and AIRS 500 mbar CO (right column). Because MODIS AOD retrievals rely on visible channels, only daytime maps are available for comparison [Kaufman et al., 1997; Chu et al., 2003; Remer et al., 2005]. For this season and location, the majority of the MODIS retrieved AOD comes from fine mode aerosols including smoke and anthropogenic pollution. The similarities in day-to-day patterns and pattern changes indicate an overall positive correlation between the two data sets.

image

Figure 13. (left) The 1° binned MODIS AOD retrievals and (right) AIRS 500 mbar CO retrievals over the INTEX-A/ICARTT study area from 15 to 20 July 2004 for the afternoon orbits (ascending). The boxed regions correspond to the different areas discussed in the text with correlations plotted in Figure 14.

Download figure to PowerPoint

[52] A more detailed comparison reveals distinct changes in the correlations as the U.S. and Atlantic plumes diverged between 16 and 20 July. The colored boxes in Figure 13 indicate specific regions where we examined AIRS CO versus MODIS AOD as plotted in Figure 14 for 15 to 18 July. The box colors in Figure 13 correspond to the symbol colors in Figure 14 with the exception of the blue boxed area in the AIRS CO maps which appears white in the MODIS AOD maps for clarity. The red box remains centered on the fire locations. The blue/white box remains centered on a source-free region of the mid-Atlantic largely unaffected by transport during this time. The green box first follows the leading edge of the smoke plume until 15 July. Thereafter, the green box shifts to follow the southward moving U.S. plume. The black box first appears on 16 July and shifts to follow the eastward moving Atlantic plume.

image

Figure 14. These scatterplots present the correlation of AIRS 500 mbar CO with MODIS AOD from 15 to 18 July. Colored symbols correspond to 1° bins in the similarly colored box regions in Figure 13. Best fit lines for the correlations are shown for each day with linear correlation coefficients given in the legends.

Download figure to PowerPoint

[53] Prior to and on 15 July, the correlation between AIRS CO and MODIS AOD over the fire region, clean Atlantic, and with the advancing plume appear similar, the red dots, blue asterisks, and open green circles in Figure 14, respectively. In fact, from 12 to 22 July, these three regions maintain similar linear regression slopes β = ΔCOAOD, with significant linear correlation coefficients for each population. As discussed below, of these three, only the slope of the U.S. plume exhibits any trends during this period. Mean values for the liner regression fits are given in Table 3. The y intercepts for the “clean” and “fire” regions crudely reflect marine and continental backgrounds, respectively. The larger scatter for the “fire” and “plume” regions lead to variations in their intercepts. All fits were computed using the MATLAB robust fit linear regression [DuMouchel and O'Brien, 1991].

Table 3. Summary of Mean Statistics for Linear Regressions Between AIRS 500 mbar CO and MODIS AOD
Mean ParameterCleanFireU.S. PlumeAtlantic Plume
Intercept (ppbv)103118107110
Slope (ppbv/AOD)35444074
Correlation coefficient0.840.720.810.82

[54] Starting on 17 July and more evident on 18 July, as the U.S. and Atlantic plumes diverge spatially, their linear regression slopes also diverge, green and black lines in Figure 14. Despite the divergence of the correlation slopes for the U.S. and Atlantic plumes, each population maintains a high linear correlation coefficient as indicated in the legends of Figure 14. Figure 15 illustrates the changes in β for the U.S. and Atlantic plumes. After 16 July, βUS slowly decreases while βAtlantic sharply peaks on 19 July with a slope nearly 3 times that of βUS. The slow decrease of βUS is consistent with subsidence of the CO-rich smoke as it moves southward into the United States as indicated by the trajectory simulations. As the trajectories of Figures 7 and 8 revealed, the Atlantic plume experienced uplift starting on 17 July followed by vertical broadening as portions of the plume experienced subsidence commencing on 20 July. Evidence for the lack of aerosol scavenging in the Atlantic plume can be seen in the relatively constant range of AOD values. The change in slope of the correlation with CO is driven primarily by the increase in CO abundance, compare the plots for 16 and 18 July. This apparent change in slope does not result from differences in the MODIS AOD retrieval over land versus ocean as Figure 15 shows the slope change begins on 17 July as the U.S. and Atlantic plumes diverge over land.

image

Figure 15. Slopes of the best fit lines, β = ΔCOAOD, to the AIRS CO versus MODIS AOD for the advancing U.S. plume (βUS) from 13 to 22 July and the Atlantic plume (βAtlantic) from 16 to 22 July.

Download figure to PowerPoint

[55] Therefore, we conclude that variations in the linear regression slope, β, between AIRS 500 mbar CO and MODIS AOD can indicate changes in the vertical distribution of CO-rich smoke. Furthermore, the increase in βAtlantic when the CO-rich smoke was not only lofted but absent below 600 mbar may imply AIRS retrievals are sensitive to CO at pressures greater than 600 mbar. Our assumption that the CO-rich smoke experienced little loss of aerosols during transit to Europe is confirmed by the findings of [Singh et al., 2006]. Two additional assumptions underlie our interpretation of these variations in β: (1) the downwind CO-rich smoke originated solely from the Yukon fires or (2) other sources along the plumes' travel emitted CO and aerosols in the same ratio as the Yukon fires. We can safely assume there were no large CO sources over the Atlantic Ocean during that plume's lofting on 18 and 19 July and its subsequent vertical broadening.

[56] Our interpretation of the magnitude and variations of the AIRS CO/MODIS AOD correlation slope differs from previous investigations of the correlation between MOPITT total column CO and MODIS fine mode AOD. The major finding of overall positive correlations between tropospheric CO and AOD remains the same; however, some details of the interpretations of the correlations differ. [Edwards et al., 2006] expanded their earlier study [Edwards et al., 2004] to the Southern Hemisphere and derived a mean aerosol lifetime from an observed exponentially decaying correlation of MOPITT total column CO and MODIS fine mode AOD for smoke plumes moving eastward off the coast of southern Africa. Kampe and Sokolik [2007] inferred differences in the biomass burning type from differences in the linear regression slopes of MOPITT total column CO and MODIS fine mode AOD for savanna and tropical forests. These previous studies all examined large geographic areas over more than a week of time encompassing a number of biomass burning plumes. Utilizing the daily broad spatial coverage AIRS, our study tracks a complex biomass burning plume and examines the temporal and spatial evolution of the correlation between AIRS 500 mbar CO and MODIS AOD. From our analysis, vertical redistribution of CO as the plume moves complicates any inferences of aerosol lifetimes or specification of biomass burning type. Future studies of additional biomass burning events are required to ascertain if this is complication also occurs for other types of biomass burning events.

10. Conclusions

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. AIRS CO for INTEX-A/ICARTT
  5. 3. AIRS CO Retrieval Algorithm
  6. 4. AIRS CO Verticality
  7. 5. INTEX-A AIRS CO Comparison
  8. 6. 2004 Yukon Fires and AIRS CO
  9. 7. AIRS CO and Trajectories
  10. 8. AIRS CO and Lidars
  11. 9. AIRS CO Versus MODIS AOD
  12. 10. Conclusions
  13. Acknowledgments
  14. References
  15. Supporting Information

[57] In this investigation, we have demonstrated the accuracy of AIRS midtropospheric (500 mbar) CO retrievals (v4.2) and the capability of AIRS observations to track the long-range transport of biomass burning emissions. We found changes in the correlation slope between AIRS 500 mbar CO and MODIS AOD indicate variations in the vertical distribution of CO-rich smoke. Combined with trajectory simulations, the AIRS CO versus MODIS AOD correlation can diagnose biomass burning injection heights. Comparison to INTEX-A in situ profiles from the DACOM instrument on board the NASA DC-8 establish the accuracy of this set of AIRS midtropospheric CO retrievals: AIRS 400–500 mbar CO is biased 7.7% high with a standard deviation of 4.7%. This comparison is facilitated via convolution of the in situ profiles with the AIRS verticality function defined as the 100 AIRS layer representation of the sum of the rows of the AIRS averaging kernel. In one case presented here, it appears AIRS CO retrievals exhibit sensitivity to CO in the lower troposphere (800 to 500 mbar).

[58] Investigating a major transport episode of emissions from forest fires burning in Alaska and Canada 11–14 July 2004 with extensive trajectory simulations, we demonstrate that a range of emission injection heights is required to explain the observed downwind CO transport to 22 July. These comparisons indicate some of the Yukon fires must have directly injected CO-rich smoke above 500 mbar and perhaps as high as 300 mbar. In order for any forecast model to accurately predict the influence of forest fire emission products on locations and times many days downwind, it must place the emissions at the correct altitude or altitudes. Ground-based lidar observations confirm the general transport heights and complexities evident in the trajectories and indicate AIRS CO retrievals are sensitive to CO in the lower midtroposphere (800 to 500 mbar).

[59] An observed steepening of the correlation slope for the Atlantic plume on 17 to 19 July is caused by an apparent increase in CO abundances observed by AIRS as the CO-rich smoke is uplifted into the region of AIRS peak sensitivity (700 to 300 mbar). Because MODIS AOD retrievals are largely insensitive to the vertical distribution of aerosols, MODIS AOD remains constant; therefore, the correlation slope steepens. However, once the Atlantic plume broadens vertically after 19 July, the correlation slope decreases as CO-rich smoke fills the lower troposphere. Combining this with the AIRS CO retrievals over the U.S. plume as it descends to the boundary layer, reaches the Gulf of Mexico, and stagnates, we infer v4.2 AIRS CO retrievals are sensitive to CO in the lower troposphere well below the nominal peak sensitivity at 500 mbar.

Acknowledgments

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. AIRS CO for INTEX-A/ICARTT
  5. 3. AIRS CO Retrieval Algorithm
  6. 4. AIRS CO Verticality
  7. 5. INTEX-A AIRS CO Comparison
  8. 6. 2004 Yukon Fires and AIRS CO
  9. 7. AIRS CO and Trajectories
  10. 8. AIRS CO and Lidars
  11. 9. AIRS CO Versus MODIS AOD
  12. 10. Conclusions
  13. Acknowledgments
  14. References
  15. Supporting Information

[60] The authors gratefully acknowledge support from the NASA EOS Program for W.W.M., X.J.W., and M.M.C. through NASA grants NAG5-11163, NAG5-11653, NNG04GN42G, and NNG06GB06G. R.M.H. was supported through a grant from the NOAA CREST at CUNY. E.E. is thankful for the support of NSF. We thank Mark Schoeberl for the trajectory code. Special thanks to the entire AIRS Team and Vickie Connors for her diligent review of this manuscript. W.W.M. thanks Rae Force for her patient support. AIRS CO data are available upon request and online.

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. AIRS CO for INTEX-A/ICARTT
  5. 3. AIRS CO Retrieval Algorithm
  6. 4. AIRS CO Verticality
  7. 5. INTEX-A AIRS CO Comparison
  8. 6. 2004 Yukon Fires and AIRS CO
  9. 7. AIRS CO and Trajectories
  10. 8. AIRS CO and Lidars
  11. 9. AIRS CO Versus MODIS AOD
  12. 10. Conclusions
  13. Acknowledgments
  14. References
  15. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. AIRS CO for INTEX-A/ICARTT
  5. 3. AIRS CO Retrieval Algorithm
  6. 4. AIRS CO Verticality
  7. 5. INTEX-A AIRS CO Comparison
  8. 6. 2004 Yukon Fires and AIRS CO
  9. 7. AIRS CO and Trajectories
  10. 8. AIRS CO and Lidars
  11. 9. AIRS CO Versus MODIS AOD
  12. 10. Conclusions
  13. Acknowledgments
  14. References
  15. Supporting Information
FilenameFormatSizeDescription
jgrd14620-sup-0001-t01.txtplain text document1KTab-delimited Table 1.
jgrd14620-sup-0002-t02.txtplain text document1KTab-delimited Table 2.
jgrd14620-sup-0003-t03.txtplain text document0KTab-delimited Table 3.

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.