During the DISCOVER-AQ Texas campaign in September 2013, the NASA King Air B-200 carried the GCAS instrument for remote sensing of trace gases and aerosols, as well as the NASA High Spectral Resolution Lidar-2 (HSRL-2) instrument for the purpose of measuring aerosol profiles below the aircraft. The B-200 typically flew at a cruise altitude of ∼9 km, with occasional descents to avoid cirrus clouds. Table 1 summarizes the 26 GCAS flights (21 for air quality and 5 for ocean color), which took place on 13 days. Most flights were designed to coincide with P-3B flight paths. The B-200 aircraft was based at Ellington Field in southeast Houston and typically flew a morning flight, refueled and then flew an afternoon flight. Each B-200 flight over Houston consisted of two overpasses of nearly the same flight path, so that there are typically four GCAS overpasses of Houston each day. The ocean color flights involved collecting data over the Gulf of Mexico in support of the ocean color component of GEO-CAPE. In this study, we focus only on the air quality flights over the Houston area.

Total column observations of NO2 were made from 15 ground-based Pandora spectrometers viewing in direct-sun (DS) mode at 11 sites during the DISCOVER-AQ Texas campaign. GCAS overflew 14 of these spectrometers at 10 sites, which are summarized in Table 2. Pandora NO2 is determined at a temporal resolution of 90 s using the ratio of direct-sun spectra to a reference spectrum derived by a top-of-the-atmosphere Langley extrapolation using spectra collected on a clear day with low NO2 . Spectra are fit from 400 to 440 nm with NO2 cross sections interpolated to 264 K and O3 at 225 K . At solar zenith angles (SZAs) less than 80∘, the observed DS slant column is converted to vertical total column using a simple geometric air mass factor . Pandora DS NO2 measurements have a nominal precision of 2.7×1014 moleculescm-2 and accuracy of 2.7×1015 moleculescm-2. Pandora observations with fitting root mean square <0.005 and relative error <10 % are included in this study, to exclude possible cloud-contaminated measurements.

Pandoras also operated in multi-axis sky-scanning mode (MAX-DOAS) measuring lower-tropospheric NO2 distribution and tropospheric columns at the La Porte, Moody Tower and Smith Point sites. Pandora head sensors sequentially pointed at 1, 2, 3, 4, 6, 8, 10, 15, 20, 30, 40 and 90∘ elevation angles from the horizon with a field of view of 1.6∘. Azimuth angles were chosen to ensure an unobstructed view down to the horizon and were 320∘ from north at La Porte, 45∘ at Moody Tower and 270∘ at Smith Point. Differential slant column densities of NO2 and O2-O2 within a single scan were calculated using a zenith sky reference spectrum. A temperature-dependent NO2 absorption cross section (linear and constant terms) and Ring, H2O and O2-O2 cross sections (see Table 3 for references) were used in the MAX-DOAS fitting window 425–490 nm. The profile inversion was performed using the maximum a posteriori optimal estimation method with aerosol and gas weighting functions calculated using the Vector Linearized Discrete Ordinate Radiative Transfer (VLIDORT) model . Tropospheric columns were also estimated using a geometrical approach using NO2 and O2-O2 columns derived from 15∘ elevation angle measurements when inversions failed.

The Pandora dataset contains observations from two Pandora instruments placed at the Moody Tower site at the University of Houston, 70 m above the surface. We correct for the column in the bottom 70 m of the atmosphere using in situ observations at the base and top of the towers collected every 5 min by the University of Houston following . The in situ measurements indicated that NO2 within these altitudes was usually well mixed at the overpasses. This correction varies in magnitude from 0.3×1015 to 3.7×1015 moleculescm-2 for different GCAS overpasses.

The P-3B aircraft carried a suite of in situ instruments in order to profile the atmosphere during the campaign. Profiles were collected during aircraft spirals near eight DISCOVER-AQ ground sites, with each site typically overflown two or three times each day. Depending on the site and flight, the aircraft typically flew between a lowermost altitude of 0–300 m and an uppermost altitude of 3.5–5 km. The typical radius of a spiral was 4–5 km.

The National Center for Atmospheric Research's (NCAR) chemiluminescence instrument (P-CL) measured in situ NO2 concentrations from the P-3B. P-CL observations of NO2 have uncertainties of 0.02 ppbv in precision and 10 % in accuracy.

The NCAR Differential Frequency Generation Absorption Spectrometer (DFGAS) measured in situ CH2O concentrations from the P-3B. The DFGAS instrument collects data with a temporal resolution of 1 s, with a 15 s background zero-air addition period every 60 to 120 s. This addition captures and removes both inlet/sample cell CH2O outgassing and optical noise. For a typical spiral, the temporal resolution translates to a vertical resolution of approximately 5 m. The 1 s measurements have a precision of ∼0.08 ppbv (upper limit) and an estimated accuracy of 4 % at the 1σ level.

This study uses model-simulated trace gas profiles for radiative transfer calculations in order to determine vertical column densities from observed slant column densities. Tropospheric simulations are performed with the Environmental Protection Agency's (EPA) Community Multiscale Air Quality (CMAQ) version 5.0.2 modeling system over the campaign domain at a spatial resolution of 4×4 km2 and a temporal resolution of 20 min. The model has 45 vertical levels from the surface to 50 hPa. The model's vertical resolution ranges from 22 m at the surface to ∼200 m at an altitude of 2 km, further increasing to ∼650 m by the aircraft flight altitude. CMAQ simulations are driven by offline meteorology from the Advanced Research Weather and Forecasting (WRF-ARW) model via the Meteorology-Chemistry Interface Processor (MCIP) . describe the CMAQ and WRF modeling approach used for the DISCOVER-AQ Texas campaign in detail.

Stratospheric NO2 profiles used in the study are estimated using the PRATMO chemical box model from simulated profiles provided as a function of month, solar zenith angle and latitude. Stratospheric ozone profiles are from the September 2013 monthly climatology derived at 1∘×1∘ from the OMI ozone profile product up to 0.3 hPa.

We perform spectral fitting to derive slant columns using radiometrically calibrated spectra (Level 1B), which are geolocated and derived from raw (Level 0) data using characterization data collected in the laboratory before the campaign, as described in detail by . The first-guess wavelength calibration was determined from spectra collected in the laboratory using mercury–argon, cadmium, neon and krypton discharge lamps as sources. The pre-flight slit function shape and width as functions of cross-track position, wavelength and temperature were also determined using a tunable laser with an integrating sphere. These laboratory tests indicated the instrument's spectral shift is ∼0.004 nm and the change in the slit function's full width at half maximum (FWHM) is less than 0.0013 nm within the instrument's thermal stability range of ±0.25 ∘C and nominal operating temperature of 20 ∘C. Pressure changes within the instrument may also shift the wavelength calibration through changes in the index of refraction . These changes are minimized in GCAS and are primarily due to changes in ambient temperature, as the instrument is backfilled with gaseous nitrogen and sealed prior to aircraft integration to mitigate moisture. The impact of wavelength shifts on retrievals is further minimized through simultaneous fitting of a wavelength shift for each observation as described in Sect. .

We further refine the instrument spectral registration and slit function calibration using spectra collected during the Texas flights, following our calibration approach previously applied to GOME, GOME-2, OCO-2, ACAM and GeoTASO . As a first step in the spectral fitting, we simultaneously derive a wavelength dispersion and slit function shape by fitting a reference spectrum to a high-spectral-resolution solar atlas . This is similar to the approach employed in our satellite retrievals, but, as the airborne nadir reference spectrum contains atmospheric features (which are not present in a satellite-observed exo-atmospheric reference), we also simultaneously fit preliminary amounts of the atmospheric molecular absorbers listed in Table 3 and the Ring effect (rotational Raman scattering) to account for these spectral features .

We determine a separate wavelength dispersion and slit function shape for each of the 21 cross-track positions. The wavelength dispersion is determined by fitting the coefficients in a fifth-order (NO2) or seventh-order (CH2O) polynomial that represents the wavelength as a function of detector pixel. For NO2, we model the slit function using an asymmetric super-Gaussian . For CH2O, we fit parameters describing the shape and width of an asymmetric Gaussian . While the super-Gaussian works well for NO2, it results in a very small increase (∼5 %) in fitting residuals for CH2O over an asymmetric Gaussian, possibly due to the presence of a double shoulder to one side of the slit function shape, as measured in the laboratory at wavelengths less than 380 nm .

The retrieved slit function in the NO2 fitting window is nearly symmetric and very similar in width to the FWHM = 0.58 nm measured in the laboratory. Using in-flight data, the retrieved FWHM is 0.57 nm at the nadir center position, expanding to 0.58 nm at the edges of the swath. The retrieved slit width in the CH2O fitting window changes in a similar way but is larger at the edges (0.57 nm at swath center, growing to 0.60 nm at the leftmost cross-track position (no. 1) and to 0.63 nm at the rightmost cross-track position (no. 21)). We estimate the uncertainty in the slit width is ∼0.01 nm, primarily due to temperature fluctuations during flight. In-flight data show the center detector pixel-to-wavelength registration for both the NO2 and CH2O fitting windows varies nearly linearly as a function of swath cross-track position across the left-hand side of the swath, varying by ∼0.1 nm between cross-track positions 1 and 11, but remains approximately constant from positions 11 to 21. The retrieved wavelength calibration is stable to ∼0.002 nm after the instrument has thermally stabilized during flight.

For air quality applications, we are interested in the vertical column density, V, of the trace gas (NO2 or CH2O) in the troposphere. The vertical column density can be derived from the slant column density, S, using an air mass factor, A, which describes the mean light path through the atmosphere, by V=SA. In practice, the retrieval algorithm determines a differential slant column ΔS, which is the difference between the slant column S of the absorber in the spectrum of interest and the slant column SR in the reference spectrum. Each of these slant columns is the sum of the slant column of absorber in the light path above (↑) and below (↓) the aircraft, so that ΔS=(S↓+S↑)-(SR↓+SR↑). In terms of the air mass factor and vertical column, the vertical column below the aircraft can then be expressed as V↓=ΔS-V↑A↑+VR↓AR↓+VR↑AR↑A↓, where the vertical columns V↑, VR↓ and VR↑ are typically determined from a model. Because the flight altitude of 9 km is well above the majority of tropospheric NO2 and CH2O, we refer to V↓ and V↑ as the tropospheric and stratospheric trace gas columns. NO2 above the aircraft is dominated by stratospheric NO2, varies primarily by time of day and ranges within V↑=2.3–3.8×1015 moleculescm-2. CH2O in the model is more variable, with the early part of the campaign (4 to 18 September) seeing levels of V↑=2–25×1014 moleculescm-2 and the latter part seeing levels of V↑=1–3×1014 moleculescm-2. For our chosen reference location, the modeled vertical columns below the aircraft are VR↓=2.0×1015 moleculescm-2 for NO2 and VR↓=7.5×1015 moleculescm-2 for CH2O. The modeled vertical columns above the aircraft at the reference location are VR↑=3.6×1015 moleculescm-2 for NO2 and VR↑=7.9×1014 moleculescm-2 for CH2O.

Only cloud-free measurements are used in this study, and the radiative transfer calculations assume cloud-free conditions. Unlike the case of satellite observations with footprints on the order of tens of square kilometers, GCAS observations are of sufficiently high spatial resolution that cloudy pixels can be discarded without loss of a significant amount of data. We flag as cloudy any pixel that has a mean radiance in the NO2 fitting window over a threshold of 2×1013 photonscm-2nm-1s-1sr-1, which is typically only exceeded in the case of a bright cloud. The Ring scattering parameter retrieved simultaneously with NO2 and a color index (the radiance ratio at wavelengths 320 to 440 nm) are also used to flag less bright pixels where clouds likely occur .

Uncertainties in the vertical column density result from uncertainties in (1) the slant column fitting; (2) the air mass factor calculation; and (3) the modeled reference and stratospheric columns needed for determining the vertical column below the aircraft using Eq. ().

We compare the retrieved GCAS NO2 and CH2O columns with columns derived from in situ observations on the P-3B aircraft. The P-3B profiles are converted into column amounts below the top flight altitude (usually 3.5–5 km) using mixing ratios and pressure/temperature profiles measured on board the P-3B aircraft. Comparisons between GCAS and P-3B columns are shown in Fig. .

Figures  and show comparisons of GCAS tropospheric columns with Pandora NO2 columns derived from both DS and MAX-DOAS scattered light retrievals, by day and by site. Figure  shows the Pandora measurements at four sites as a function of time, and GCAS coincidences with those observations. In the case of the Pandora DS observations, we have estimated the tropospheric Pandora column by subtracting the modeled NO2 above the GCAS instrument (typically ∼2.5–4×1015 moleculescm-2) from the closest Pandora observation in time within 3 min. The uncertainty in the stratospheric NO2 column in our model is estimated at 30 % (see Sect. ). In the case of MAX-DOAS comparisons, we compare a single GCAS observation over each site with the closest MAX-DOAS observation within 20 min. For the comparison with DS observations, we have determined the GCAS observation from the mean of GCAS ground pixels intersected by the Pandora line of sight in the bottom 2 km of the atmosphere (shown in Fig. ). This helps to minimize the influence of the Pandora viewing geometry on the comparison. For instance, GCAS consistently measures large columns over the Deer Park site, with some of the largest NO2 often to the north of the site; however, when viewing the sun directly, Pandora always looks south into cleaner air. The use of a GCAS NO2 amount determined along the Pandora DS line of sight reduces the influence of these biased site locations on the results, with an overall reduction in the GCAS-versus-Pandora bias of 20 %. There remain, however, several sites with an obvious difference in GCAS versus Pandora DS measurements, despite considering the field of view.

Overall, GCAS tropospheric NO2 is larger than Pandora (GCAS/Pandora = 1.50 for DS and 1.33 for MAX-DOAS), although the spatial correlations are very good at r2=0.85 (DS) and r2=0.94 (MAX-DOAS). A background offset of ∼2×1015 moleculescm-2 is seen between GCAS and the Pandora DS measurements, similar to that seen in the P-3B comparisons. Again, this is most likely from uncertainties in the modeled stratospheric correction and reference spectrum correction, with a possible contribution from the Pandora reference as well. More surprisingly, GCAS NO2 is 50 % (DS) and 33 % (MAX-DOAS) larger at high NO2 values.

In order to assess the dependence of the GCAS observations on the profile uncertainty, we also apply the P-3B profiles in place of model profiles in the GCAS AMF calculations and compare the new GCAS columns with the P-3B and Pandora columns. When the spiral profiles are applied to the GCAS observations within their vicinity, the use of the observed profiles lowers the overall slope of GCAS tropospheric NO2 columns by 4 % (P-3B) and 2 % (Pandora) and the CH2O columns by 2 % (P-3B) as compared with coincident observations. The NO2 correlations with the P-3B and Pandora remain the same, but the correlation increases to r2=0.62 for P-3B CH2O. Individual coincident observations can change by as much as -50 % to +35 % for NO2 (mean change of +1±10 %) and -15 % to +25 % for CH2O (mean change of +3±8 %). The largest mean changes for a single day occur at the Deer Park site in the Pandora comparisons, where the GCAS NO2 column on 25 September is reduced by 15 % on average.

Overall, the GCAS observations correlate well spatially and temporally with the P-3B and Pandora observations. The GCAS observations also show agreement in magnitude with the P-3B-inferred NO2 and CH2O columns well within the measurement uncertainties. The GCAS NO2 observations are significantly higher than those of the Pandora direct-sun instruments by 50 %. They are also higher than the Pandora MAX-DOAS by 33 %, although there are fewer coincidences with the MAX-DOAS observations, and there is good agreement at the La Porte site.

Differences between GCAS and the P-3B primarily result from errors in the GCAS AMF (surface reflectance, aerosols and profile shape); the inability of the P-3B to capture profiles of near-surface gases below 300 m near the Channelview, Manvel Croix and Moody Tower sites; and spatial variability and P-3B sampling. Much of the variability observed in individual spirals in the NO2 column comparison in Fig.  is due to the large radius of the P-3B spiral, which can mean the P-3B sometimes flies in and out of NO2 plumes, as seen in Fig. .

The differences between GCAS and Pandora NO2 are much larger. GCAS NO2 columns could be influenced by several uncertainties that can result in cumulative biases in the AMF calculation (again, primarily from errors in surface reflectance, aerosols and profile shape). Different factors likely dominate the uncertainties at different sites; some sites are located at locations with very inhomogeneous surface reflectance (Smith Point and Moody Tower), and some at locations with large uncertainties in profile shape. The slope is also dominated by the larger polluted measurements on 25 September, which was a day with complicated meteorology, a morning boundary layer of ∼200 m (according to HSRL data) and uncertain emissions .

also found a large difference between GCAS and Pandora observations during the Texas campaign. By using a Bayesian inversion to constrain the MODIS BRDF, they reduced the overestimation of GCAS relative to Pandora by 23 % through a 0.023 increase in surface albedo, broadly consistent with studies that have found a low bias in MODIS surface reflectance at short wavelengths. The exclusion of aerosols in our AMF calculation may cause the AMF to be underestimated (and therefore the vertical column to be overestimated) in some cases, particularly where scattering aerosols are in the lowest part of the boundary layer (see discussion in Sect. ). We find that the GCAS vertical columns at Pandora coincidences are reduced on average by 10 % when the air mass factor is calculated using the nearest HSRL aerosol optical thickness profiles below the aircraft for scattering aerosols. In the previous section, we saw there is likely a small +2 % bias in the GCAS column from the use of CMAQ-modeled NO2 profile shapes in measurements coincident with Pandora. Therefore, while profile shape may contribute to large errors on individual observations, it is unlikely to produce a large bias in the GCAS observations overall.

Differences in the GCAS and Pandora slant column retrievals themselves may also play a role, including the wavelength fitting region and atmospheric temperature assumptions. Previous comparisons of Pandora DS total column NO2 observations with other ground-based observations have shown good agreement . In contrast, compared a year of Pandora zenith sky stratospheric NO2 slant columns with those from a zenith-looking UV–Vis spectrometer (DOAS M07) from the Network for the Detection of Atmospheric Composition Change (NDACC) using different retrieval settings and found Pandora underestimated the NDACC instrument by 7 %–40 %. The Pandora slant column product used in our study was produced assuming a fixed effective temperature of 264 K, which could result in a low bias in the retrieved Pandora slant column of 10 % .

Despite the sources of uncertainty on the GCAS columns, it should be noted that a large reduction in the GCAS vertical columns from the use of different AMF inputs resulting in better agreement with the Pandora columns would mean a significant underestimation by GCAS of both the NO2 and CH2O P-3B columns. Recent comparisons of Pandora DS NO2 columns and NO2 inferred from the P-3B P-CL instrument for the four DISCOVER-AQ campaigns (Maryland, California, Texas and Colorado) show the P-3B agrees well with Pandora DS measurements for all campaigns except Texas, where Pandora NO2 is significantly underestimated (Sungyeon Choi, personal communication). Previous airborne comparisons with the GeoTASO instrument during DISCOVER-AQ Texas on four relatively unpolluted or cloudy days (13, 14, 18, 24 September) also suggested airborne NO2 larger than Pandora .