In the summer of 2007, the NASA DC-8 aircraft took part in the Tropical Composition, Cloud and Climate Coupling campaign based in San Jose, Costa Rica. During this campaign, multiple in situ and remote-sensing instruments aboard the aircraft measured the atmospheric composition of the tropical tropopause layer (TTL) in the equatorial region around Central and South America. During the 17 July flight off the Ecuadorian coast, well-defined “bubbles” of anomalously low-ozone concentration (less than 75 ppbv) were detected above the aircraft in the TTL at the altitude near 365 K (between 14 and 16 km) and at ∼3°S and ∼82°W. Backward trajectories from meteorological analyses and the aircraft in situ measurements suggest that the ozone-depleted air mass originated from deep convection in the equatorial eastern Pacific and/or Panama Bight regions at least 5 days before observation by the DC-8; this was not a feature produced by local convection. Given uncertainties known in regard to trajectories calculated from global reanalysis, it is not possible to identify the exact convective system that produced this particular low-ozone anomaly, but only the general origin from a region of high convective activity. However, the fact that the feature apparently maintained its coherency for at least 5 days suggests a significant contribution to the chemical composition of the tropical upper troposphere portion of the TTL from convective systems followed by quasi-horizontal transport. It also suggests that mixing time scales for these relatively small spatial features are greater than 5 days.
 Ozone changes in the tropical lower stratosphere are important for determining the magnitude and sign of the ozone radiative forcing [IPCC, 2001]. Additionally, as noted by Gettelman et al. , modeled tropopause height levels and cold point temperatures are sensitive to the amount of ozone near the tropopause. The photochemical lifetime of ozone in the tropical tropopause layer (TTL) with respect to the chemical production and loss is several months, so transport is the primary cause of changes in ozone mixing ratios [Folkins et al., 2002; Fueglistaler et al., 2009; Rivière et al., 2006; Wennberg et al., 1998]. Because the TTL serves as the gateway for air entering the stratosphere, it is of interest to study processes that impact the distribution of radiatively and chemically important gases in that region. In this study, we examine a specific case in which low-ozone air from low altitudes is transported into the tropical upper troposphere and maintains its integrity for a number of days, demonstrating the importance of convective transport to establishing species concentrations in the tropical upper troposphere.
 In the summer of 2007, the NASA DC-8 aircraft took part in the Tropical Composition, Cloud and Climate Coupling (TC4) campaign based in Costa Rica [Pfister et al., 2010; Toon et al., 2010]. Multiple in situ and remote-sensing instruments aboard the aircraft were flown to measure atmospheric composition of the TTL. The layer was first defined by Highwood and Hoskins  and Folkins et al.  as the transitional layer in the tropics (located at ∼12–18 km altitude) that is significantly impacted by deep convection, and the chemical composition of which is transitional between the “convectively dominated tropical troposphere and the radiatively controlled stratosphere” [Gettelman and Forster, 2002]. Fueglistaler et al.  defined the layer slightly differently, extending it further into the tropical stratosphere.
 In this study, observations are described only in the tropospheric portion of the TTL, from ∼150 hPa or 15 km to the tropopause. On the 17 July 2007 TC4 flight of the NASA DC-8, a bubble of depleted ozone between 14 and 16 km was observed near the coast of Ecuador by several instruments. A similar ozone feature is found in a satellite ozone profile and is also consistent with satellite measured total ozone column data on that day. Analyses of an ozone sounding data climatology over several Intertropical Convergence Zone (ITCZ) ground stations often show the typical S-shaped vertical structure near 15 km that is indicative of the ongoing convective process [Thompson et al., 2010]. This indicates that such phenomena are not an uncommon occurrence and hence likely contribute to the overall ozone budget in the UT. Convective signatures are not necessarily local and recent; however, advection and wave activity also play a role in ozone structure near the TTL [Selkirk et al., 2010; Thompson et al., 2010]. Thus, it is of interest to study this particular case observed by the DC-8 to focus on questions concerning the impact of convection and time scales for chemistry and transport affecting the composition of the TTL [Fueglistaler et al., 2009]. Sections 2 and 3 describe ground and aircraft data associated with the observed low-ozone event. Section 4 describes the back trajectory analysis of the low-ozone parcel transport before the observation. Section 5 presents satellite data for confirmation of the spatial and temporal extend of the low-ozone parcel. Further exploration of the long-range isentropic ozone transport hypothesis is done in section 6 by using reverse domain filling (RDF) calculations driven by assimilated meteorological analyses and initialized with satellite ozone. Section 7 summarizes TC4 results discussed in other TC4 special issue papers that are relevant to the TTL composition affected by local convection processes, wave activity, pollution from the boundary layer, and long-term ozone climatology. Section 8 concludes that local convection can be excluded as a possible explanation for the low-ozone bubbles observed in the TTL by the TC4 aircraft instruments.
2. Measurements and Data
 The TC4 campaign was based out of San Jose, Costa Rica, during July and August 2007. Measurements were coordinated between three aircraft, providing coverage of the stratosphere (ER-2), upper troposphere (WB-57), and low to middle troposphere (DC-8). In this study, we exclusively use measurements from the DC-8, as the flight of interest did not have simultaneous measurements from the other two aircraft. The DC-8 aircraft was equipped with both in situ and remote-sensing instruments. Measurements of ozone, aerosol, cloud, water vapor, and other trace gasses were taken from aboard the aircraft. A key measurement for this study was from the DC-8-borne CCD-based Actinic Flux Spectroradiometer (CAFS) [Petropavlovskikh et al., 2007], which has been used in previous satellite validation campaigns [Kroon et al., 2008; Petropavlovskikh et al., 2008]. Other relevant DC-8 measurements included nadir and zenith ozone, aerosol, and depolarization profiles from the differential absorption lidar (DIAL) system [Browell, 1981; 1989; Browell et al., 1998]; in situ ozone mixing ratios from the FastOz system [Avery et al., 2010; Pearson and Stedman, 1980] using the fast-response nitric oxide chemiluminesence method; carbon monoxide (CO) measurements from the differential absorption CO measurement instrument [Sachse et al., 1987]; and tropical ozonesondes launched from Las Tablas, Panama (7.8°N, 80°W) [Thompson et al., 2010, G. Morris et al., Observations of ozone production in a dissipating tropical convective cell during TC-4, submitted to Atmospheric Chemistry and Physics Discussion, 2010], Juan Santa Maria airport, Costa Rica (10°N, 84°W) [Selkirk et al., 2010], San Cristobal, Galapagos Islands (1°S, 90°W), and Paramaribo, Ecuador (5.8°N, 55°W). The Costa Rica, San Cristobal, and Paramaribo sites are part of the Southern Hemisphere Additional Ozonesondes network. In addition to the daily ozonesondes at Panama, continuous surface ozone measurements from the Nittany Atmospheric Trailer and Integrated Validation Experiment were also taken. Measurements from all of these instruments have been used in past satellite validation campaigns [Froidevaux and Douglass, 2001; Newman et al., 2001] and for scientific process studies.
 The DC-8 sampled the atmosphere over a very large region near equatorial Central and South America [Toon et al., 2010]. However, the phenomenon we discuss in this study, a low ozone bubble near the coast of Ecuador, was the only one observed by the DC-8 during the TC4 mission. Figure 1 shows the NASA DC-8 flight tracks on 17 July 2007 plotted over the map of coincident Aura Ozone Monitoring Instrument (OMI) total ozone data (orbit 15982) [Levelt et al., 2006a, 2006b]. There are two ozone products that are available from the OMI ultraviolet-visible images of the Earth [Kroon et al., 2008; McPeters et al., 2008]. One method (OMI-Total Ozone Mapping Spectrometer (TOMS)) is based on the traditional TOMS retrieval algorithm that utilizes six discrete wavelengths from 306 to 380 nm [McPeters et al., 1998], while another method (OMI-differential optical absorption spectroscopy (DOAS)) is based on the differential optical absorption spectroscopy that takes advantage of the hyperspectral measurements from 270 to 500 nm at an average resolution of 0.5 nm [Veefkind et al., 2006]. Figure 1 displays the OMI-DOAS-derived total ozone field (collection 2 data).
 We use GOES-12 satellite (located over the equator at 75°W) images from the TC4 region for identification of deep convection and for moisture and temperature analysis. Both visible (channel 1) and infrared (IR) (channel 4) images are used in this study. Brightness temperatures (from channel 4) below −35°C are designated by colors with −10°C for each color change (green is between −65°C and −75°C). All images used here have been degraded to a 6 km resolution from the original 1 km visible and 4 km IR data.
 This study focuses on analysis of measurements taken during one DC-8 flight. The flight was on 17 July 2007, and the track is shown in dark gray in Figure 1. In particular, we examine one unusual ozone feature and analyze its origins.
3. NASA DC-8 Observations
 A depleted ozone column above the DC-8 aircraft was detected by both DIAL and CAFS near the Ecuador coast on 17 July 2007. The total ozone column was also measured by the OMI aboard the Aura satellite. The OMI surface tracks on 17 July 2009 were located in close proximity to the depleted ozone episode location (see Figure 1). The OMI data were then interpolated to the latitude of the DC-8 flight tracks. Figure 2 shows time series of CAFS ozone columns (green) derived above the altitude of the DC-8 for the 17 July 2007 flight. In addition, the colocated OMI-TOMS version 2.2 data are shown as total ozone column above the clouds (magenta) and ozone columns above the surface (blue). The depleted ozone column (by ∼10 DU) above the DC-8 aircraft is found at about 1700 UT (Figure 2, vertical red line). The extension of the CAFS-derived partial ozone column data (green) with ozone climatology [Bhartia and Wellemeyer, 2002] estimated below the DC-8 altitude (orange) creates a total ozone column data set (black symbols) that matches a similar reduction in the OMI-TOMS total ozone column time series (seen in both blue and magenta symbols). It suggests that the reduction in OMI-TOMS total ozone column is entirely confined to the altitudes above the aircraft.
 This anomaly in the CAFS and OMI ozone column observations occurs at the same time that the DIAL vertical profile data shows a bubble of depleted ozone between 14 and 16 km. Figure 3 shows the cross-section contour plot of the aerosol scattering ratio (a) and ozone mixing ratio (b) distributions above and below the aircraft level as measured by the DIAL system during the part of the flight between 1600 and 1720 UT. This portion of the flight was flown at an altitude of 11.3 km (until about 1702 UT when the aircraft turned and descended to 10.3 km altitude). The black line in the middle of Figure 3a indicates the DC-8 altitude. Note that FastOz in situ ozone data are included in Figure 3b and are shown as a color-coded thin line at the altitude of the NASA DC-8 aircraft. The ozone anomaly between 14 and 16 km altitude is measured twice in DIAL time series (centered around 1653 UT on the southbound leg and again around 1705 on a parallel track heading north 0.4° longitude farther to the west). The temporal extension of the low-ozone feature is about 12 min (for both episodes). The NASA DC-8 cruises with a true air speed between 787 and 908 km/h (or between 425 and 490 knots). The DC-8 speed was recorded at ∼821 km/h during the low-ozone bubble encounter. Therefore, the transverse dimension of the detected air mass can be estimated at ∼165 km. The variation in ozone noted here is approximately a factor of 2, from a high value in the 15 km region of ∼0.125 ppmv (or 125 ppb) to a low value of ∼0.06 ppmv (60 ppb). To estimate the partial column of ozone between 14 and 16 km, we use basic profile information from the ozonesonde launched at San Cristobal, Galapagos Islands (1°S, 90°W that is in relatively close vicinity to the low-ozone air mass location at ∼2°S, 82°W) on 16 July 2007. That sonde recorded ∼154 and ∼110 hPa air pressure at 14 and 16 km altitude, respectively. Assuming ozone is well mixed in that altitude range, as indicated by the DIAL curtain, and using the 60 ppbv difference between the low-ozone feature and the background measured by the DIAL instrument at 15 km, we thus estimate a partial column ozone reduction of ∼2 DU. Therefore, it appears to be about 20% of the 10 DU reduction detected in both the CAFS partial and OMI total column ozone data (Figure 2) while crossing the area with low-ozone air mass. There has to be an additional ozone decrease, possibly related to the difference in the tropopause altitude in the area of ozone bubble compared to the climatological mean.
 To determine the origin of the ozone anomaly at 15 km altitude, the coincident in situ ozone measurements were examined for signs of the local deep convection at 11.3 km flight level. In situ ozone measurements were collected by the FastOz instrument [Avery et al., 2010] aboard the DC-8. During most of this flight, an average 38 ± 6 ppbv of ozone mixing ratio was measured while sampling inside the cloud, and 63 ± 2 ppbv was measured when out of the cloud as determined by the condensed water content measurements from the National Center for Atmospheric Research Counterflow Virtual Impactor instrument that flew on the DC-8 [Noone et al., 1988; Twohy et al., 1997]. Figure 4 shows FastOz data during the portion of the flight when the DIAL observed the low-ozone events at 15 km altitude. Note that the FastOz data are plotted at various aircraft altitudes at 11.3 km from 1600 to 1702 UT and at 10.3 km from 1705 to 1715 UT (the period of the DC-8 descent is indicated by two solid vertical lines). An intermediate ozone concentration during the flight near the Ecuador coast (marked by two vertical dashed lines between 1642 and 1655 UT) at 11.3 km altitude is in the range of 50 ppbv. The absence of large gradients in the ozone mixing ratio suggests mixed air and not fresh convection. Therefore, this implies that the depleted ozone at 15 km altitude is not related to the local convection that would have altered the ozone mixing ratios at the 11.3 km flight level.
 Another chemical measured aboard the NASA DC-8 is CO, which can be used as an indicator of vertical transport. CO has significant effects on hydroxyl (OH) radicals in the atmosphere by reducing their abundance and increasing tropospheric ozone concentration [Andreae et al., 1988; Crutzen and Andreae, 1990]. Elevated CO concentrations in troposphere in the tropics can be a consequence of biomass burning [Lee et al., 1997; Wennberg et al., 1998]. At the same time, the tropospheric ozone distribution in the tropics is also altered through interactions of pollution with large-scale circulation and deep convection [Newell et al., 1997; Thompson et al., 2003].
Figure 4 shows CO mixing ratios remaining unchanged when sampled directly below the depleted ozone features (the low-ozone time period is indicated by the first two vertical dashed lines). CO was either uncorrelated or anticorrelated with ozone during most of the flight. The lack of the elevated CO concentrations in the upper troposphere before 1705 UT, while high concentration (∼90 ppbv) levels were measured near the surface (spiral portion of the DC-8 flight between ∼1720 and 1840 UT; data not shown), suggests that DC-8-sampled air mass at 11.3 km was different from the polluted marine boundary layer. Below the DIAL-detected ozone minimum at ∼15 km, FastOz instrument measurements show an intermediate ozone concentration in the range of 50 ppbv, which suggests mixed air. The period of the DC-8 flight between 1705 and 1710 is coincident in time with the second DIAL sampling of the low-ozone feature (between the second solid vertical line and the right edge of the plot) where an ozone minimum of 30 ppbv coincides with elevated CO readings. This period occurs right after a short descent from 11.3 to 10.3 km (indicated by two solid lines), which suggests possible convective influence at the DC-8 aircraft level. Since the DC-8 was at a lower altitude and different longitude for the second pass, it likely encountered different dynamical conditions. Although high clouds were seen in the nadir-looking DIAL aerosol channel (with cloud top heights just below 10 km) up to 1642 UT and after 1705 UT (Figure 3a), the satellite images near the time of the aircraft flight do not indicate any deep convection reaching up to the 14–16 km levels (see section 4 for more discussion). The depleted ozone at the NASA DC-8 level appears to be a narrow layer located above the cloud tops and just above a slight enhancement in the DIAL nadir aerosol image (Figure 3a). The aircraft seemed to intercept the upper outer fringe of this layer at 1657 UT but was on the lower outer fringe when it leveled out at 10.3 km at 1705 UT. There could very likely be the influence of shallower convection at the DC-8 levels, with the possibility of transport from the east (see section 4 on back trajectories), but that convection is not getting up to the levels where the depleted ozone is detected. Moreover, the DIAL data show a break in the vertical distribution with increased ozone mixing ratios at 13 km (Figure 3b). Therefore, the CO observations at the NASA DC-8 aircraft flight level provide supporting evidence that the depleted ozone is not related to local vertical transport.
4. Back Trajectory Analysis of the Low-Ozone Air Mass
 In this study, backward trajectory calculations and satellite data are analyzed to examine the evolution and identify the likely source region of the low-ozone air mass observed from the DC-8 on the 17 July flight. Back trajectories are started at the geo-location and time of the DC-8 DIAL interception of the low-ozone bubble event (near the Ecuador coast (3°S, 82°W) between 1630 and 1700 UT).
 At first, following the approach of Pfister et al. [2001, 2010], a combination of National Center for Environmental Prediction (NCEP) reanalysis meteorological fields and GOES images were used to create convective influence plots for the area under question [Pfister et al., 2001]. The back trajectories were run for 8 days before the event on 17 July 2007 and were stopped when it was determined that parcel had encountered convection as noted on the satellite images. The geo-location of the air parcels was checked against the GOES images for bright clouds (see Figures 6 and 7) that are indicative of deep convection events. The brightness temperatures were also adjusted by as much as 6° according to the findings of Sherwood et al.  and Minnis et al. ; details regarding this correction are given by Pfister et al. . Whenever a back trajectory parcel was found to be at least as high as the altitude of the intercepted cold cloud, it was considered to be convectively influenced [Pfister et al., 2010]. For example, Figure 5 shows the GOES-12 satellite IR image at 1745 UT taken on 17 July 2007. The Aura High Resolution Dynamics Limb Sounder (H), Tropospheric Emission Spectrometer (+), and Microwave Limb Sounder/Ozone Monitoring Instrument (M) instrument sampling tracks are also shown. IR channel 4 typically “sees” the surface unless it is obstructed by clouds. In this and subsequent images, brightness temperatures below −35°C have been marked with bright green colors representing the area of deep convection. The three most prominent areas of deep convection are found in IR images over the Pacific coast of Mexico, Panama, and northern South America. Moreover, based on the DIAL ozone curtain plots (Figure 3), it appears that the depleted ozone is found between about 14.9 km (or ∼49 kft) and 15.7 km (or ∼52 kft) geometric altitude. Therefore, the lower limit for trajectories was placed at a pressure of 134 hPa, while the upper limit was extended to 117 hPa.
 Additional tests were performed to investigate sensitivity of the back trajectory analysis to the meteorological data fields and transport assumptions. For the analysis presented in Figures 5–67, 10 day back trajectory calculations were performed using the HYbrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model [Draxler and Hess, 1997, 1998; Draxler, 2003]. The input meteorological data for HYSPLIT were selected from the NCEP Global Data Assimilation System (GDAS) (1° × 1° resolution; http://www.emc.ncep.noaa.gov/modelinfo/index.html) model output. Isentropic back trajectories were initialized at 1700 UT on 17 July 2007 over a matrix of nine latitudes and eight longitudes spanning the bounding latitude/longitude box of the region over which the DC-8 observed the low-ozone bubble (near the Ecuador coast (3°S, 82°W) between 1630 and 1700 UT). Additional kinematic model runs (data not shown) were performed using combinations of the GDAS omega (vertical velocity in pressure coordinates) as well as different initialization times (1600 and 1800 UT) to investigate the sensitivity of the results. Although the end points of these runs differ slightly from the nominal run, the results below are not sensitive to these small perturbations in initialization time or vertical velocity used in the analysis. The similarity between the adiabatic and omega trajectories points to the relative insignificance of diabatic heating on the trajectory of the low-ozone parcel within the TTL over the time scale considered here. A complementary view is that vertical velocities and vertical wind shear in the TTL are small compared to other regions of the troposphere, and hence have little effect on the trajectory end points over the period of 1 week. In addition, the trajectories were stopped whenever the convective influence analyses (described above) suggested the intercept with the deep convective system.
Figure 5 shows the back trajectory runs initialized at the altitude of the low-ozone bubble (top, 15.4 km) and at the mean altitude of the DC-8 during the low-ozone bubble measurement (bottom, 10.4 km). The backward trajectory analysis plots illustrate that the air masses at the DC-8 altitude and 15 km are of significantly different origin, with the flight-level air mass originating from the east over South America and the 15 km air mass originating from the west and ultimately north-northeast over the Panama Bight region.
 The 15 km back trajectories were tracked backward in time to the point at which they horizontally intercepted convection in the GOES channel 4 imagery (defined by low brightness temperatures <238 K, or blue colors in the Figures 5–7) to obtain a qualitative understanding of where the trajectories likely originated. In addition, they were also checked for consistency with the above described analysis of the convectively influenced parcels [Pfister et al., 2001, 2010]. As the trajectories make their turn and head north, we find convective influence in the time range >5 days old. All trajectories intercepted convection between 10 and 12 July, 5–7 days before being measured by the DC-8. The southernmost trajectories (black through light blue colors) encountered convection in the Panama Bight and East Pacific region, whereas the northernmost trajectories encountered convection off of the east coast of Colombia, over Colombia, and over Venezuela (green through red colors). Although the convective source of the trajectories at a given initialization latitude is somewhat sensitive to the start time and vertical transport used (i.e., isentropic trajectories as opposed to those including assimilated vertical motion output), all of the combinations of backward trajectories yield convective sources in the vicinity of Panama and Colombia. Because the GOES brightness temperatures are consistently lower over Panama for the trajectories considered here, we hypothesize that the low-ozone bubble air mass originated over this region. Overall, these back trajectories support the idea that low-ozone air detrained from deep convection over the South and Central America regions can be transported through the TTL over long times (∼7 days) and distances (∼1000 km) in a coherent manner.
 Detailed analyses of ozone soundings during the TC4 campaign period are presented by Thompson et al. . They find that between 40% and 50% of ozone in the TTL is influenced by convective transport. For example, Figure 3b in that paper [Thompson et al., 2010] presents ozonesonde measurements at the Las Tables, Panama, site during the TC4 campaign. The mean ozone mixing ratio between the 3 and 10 km altitude range (active convection zone according to the study by Avery et al. ) is about 50 ± 10 ppbv. Since the average mixing ratio in the DIAL-measured low-ozone bubble is about 60 ± 10 ppbv, we can expect that ∼20% of the mixing might have occurred in the air mass that was transported through the UT during convection.
5. Satellite Observations
 In support of our hypothesis of long-range transport from the Panama region, we present data from the two coincident times over the course of the 10 day trajectories in which the air parcels were located in the proximity of Aura/High Resolution Dynamics Limb Sounder (HIRDLS) measurements [Gille et al., 2008; Khosravi et al., 2009]. The trajectory locations over the GOES IR imagery are shown in Figures 6 (top) and 7, whereas latitude cross sections of HIRDLS ozone data for the first time segment are shown in Figure 6 (bottom).
 The HIRDLS V4 ozone profile data cover a wider range of latitudes, while profiles are about 100 km apart, so its resolution does not contain the fine horizontal details observed by DIAL. However, the vertical resolution of HIRDLS is about 1 km, which should be sufficient for identifying the vertical ozone gradient. The HIRDLS ozone pressure-latitude cross section is shown in Figure 6 (bottom). It is accompanied by the GOES IR images and trajectories shown in Figure 6 (top). Yellow arrows and the vertical dashed line in Figure 6 (top) point to the trajectory locations on 16 July over the eastern Pacific ocean, far from the regions of persistent convection off the coasts of Central America. Also, a region of low-ozone air close in space and time to the altitude of the back trajectories (∼15.25 km) is present in the HIRDLS data taken at ∼2120 UT (Figure 6, bottom, yellow horizontal line and arrows). Because of the lack of convective clouds in this region of low ozone, as evidenced by the absence of cold colors in the GOES imagery (Figure 6, top), we hypothesize that this low-ozone region in the HIRDLS data is the same air mass measured on 17 July by the DC-8. The plotted HIRDLS data have been screened as recommended by the HIRDLS data document (HIRDLS Data Description and Quality Document, http://disc.sci.gsfc.nasa.gov/Aura/data-holdings/HIRDLS/hirdls2_004.shtml); this includes eliminating data points with negative precision flags (which are dominated by a priori information) and data points earthward of the detected cloud top pressure. The region to the south of the trajectory location (between 10°S and 5°S) and below 17 km where no data exist (Figure 6, white) may be due to the presence of clouds or when a retrieval error covariance becomes greater than half of the a priori error covariance [Nardi et al., 2008]. However, the deep gradient in the HIRDLS ozone profiles between 2°S and 5°S latitude and between ∼13 and 17 km altitude shows the depleted ozone area coincident with location of the transported low-ozone bubble as suggested by our back trajectory analysis.
Figure 7 shows the trajectory locations at 2100 UT on 10 July plotted over the GOES-12 IR image, as well as the HIRDLS overpass at ∼2025 UT. This time is within 1 day after the trajectories (purple and blue) coincide with the IR image cold colors indicative of the deep convective clouds north of 10° along the HIRDLS track. It is likely that this large convective region from 5°–15°N seen in the GOES image on 10 July contributed to the low-ozone air mass observed on 17 July by the DC-8, although a definitive attribution is not possible. Further discussion on variability of ozone in the TTL region in Central America will be addressed with the analysis of ozone soundings in Panama and Costa Rica launched during the TC4 campaign.
 The HIRDLS latitude-pressure ozone cross sections provide only qualitative support to the hypothesis discussed above of long-range low-ozone transport and absence of significant mixing in the TTL. The quality of the HIRDLS ozone satellite data in the TTL region is often affected by cloud interferences that can potentially introduce errors in the ozone profile retrievals. On the other hand, coherent spatial and temporal ozone structures in the HIRDLS ozone data support the hypothesis of low ozone transport around the TC4-covered region of the tropics; quantitative agreement is not essential to this argument.
 Thus, the above results tend to support our hypothesis that the episodes of low ozone found in the DC-8 measurements in the nonconvective region originate from long-range transport of convectively influenced, low-ozone air that has maintained some integrity for several days. These results suggest that quasi-horizontal mixing processes in the upper tropical troposphere are relatively slow. Deep convection over Panama is likely the source of the observed low-ozone “bubble” (see further discussion of ozone sounding data).
6. RDF Analysis
 The back trajectory analysis discussed above suggested the possibility of the long-range quasi-horizontal transport of a low-concentration ozone bubble detected by the DIAL and CAFS instruments between 14 and 16 km near the Ecuador coast from the Panama-Bight region, where it was generated by the deep convection mechanism. The DIAL and CAFS instruments detected a low-ozone bubble between 14 and 16 km altitude around 1700 UT during the NASA DC-8 aircraft flight on 17 July 2007. Here we explore the long-range transport hypothesis further by using reverse domain filling calculations [Manney et al., 1998; Sutton et al., 1994] driven by GEOS-5 data assimilation system meteorological analyses [Reinecker et al., 2008] and initialized with Aura Microwave Limb Sounder (MLS) data [Waters et al., 2006] to infer the isentropic transport of ozone features noted here.
 For this analysis, trajectory calculations using assimilated winds from the GEOS-5 analysis were started on a dense grid (0.25° latitude by 0.40° longitude) at several isentropic levels and run back 8 days; at that time, gridded MLS (or HIRDLS) data were interpolated to the parcel locations to provide an estimate of the ozone that was transported to the starting locations of the trajectories. A second set of similar calculations was done, but with the parcels initialized on a dense vertical grid (100 levels equally spaced in log-potential temperature between 330 and 530 K) at the 5 minute average positions of the aircraft flight track. Thus, the RDF maps/profiles are based on transport by GEOS-5 winds and initialization with a single day of gridded MLS data. Calculations were also initialized with HIRDLS V4 ozone data and generally showed similar results to those initialized with MLS data; since the MLS version 2.2 data are better characterized in the Upper Troposphere/Lower Stratosphere (UTLS) [Livesey et al., 2008] and are less frequently affected by clouds than HIRDLS, we show only the MLS results here. Comparable results were obtained using preliminary version 3 MLS data. Calculations initialized with equivalent latitude [Butchart and Remsberg, 1986] also provide a consistent picture of the air parcels' origins. While the MLS (and to a lesser degree, HIRDLS) fields used for initialization have much coarser horizontal and vertical resolution than the RDF grids, previous studies using similar data sets [Manney et al., 1998, 2000, and references therein] have demonstrated good skill in reproducing small-scale features in observed profiles that result from differential advection by the large-scale wind fields.
Figure 8a shows RDF ozone from an MLS-initialized run, which shows low-ozone features at locations consistent with where the DC-8 sampled low-ozone bubbles near Ecuador on 17 July 2007. The feature also appears at 360 K, consistent with the vertical extent of the low-ozone bubble in the aircraft observations (G.M.). Gradients comparable to those observed from the DC-8 are seen in the RDF-generated ozone field, indicating that transport over 8 days can indeed generate features like those observed. RDF calculations were also initialized with several other chemical species measured by MLS (H2O, CO, and HNO3; results are not shown); these show strong consistency in the morphology of the RDF fields with those for ozone, suggesting that the RDF calculations are largely showing transport of real atmospheric features, since transported “noise” (i.e., spurious values) would be less likely to be correlated among all the species.
 Results in Figure 8a suggest that the low-ozone feature in the RDF analysis near the coast of Ecuador at the 360 K potential temperature level (near 15 km altitude; see Figure 3) are similar to the DIAL-observed low-ozone mixing ratios between 14 and 16 km altitude. The light blue-colored filament of ozone that represents the low-ozone mixing ratio of 50 ppbv extends to the west from the Ecuador coast (DC-8 tracks are marked by the white line), loops under the red-colored (higher mixing ratio) ozone feature in the middle of the plot, and then extends to the north up to the coast of Mexico. The vertical extend of ozone feature may be smoothed out because of the initialization with MLS ozone profiles that have low vertical, horizontal, and temporal resolution, but it is still indicative of the transport-related ozone residuals between 14 and 16 km altitude.
 Results of an RDF analysis “transporting” equivalent latitude [Butchart and Remsberg, 1986] are shown in Figure 8b. The equivalent latitudes are the latitudes that would enclose the same area as the potential vorticity contours, thus showing at what equivalent latitude the air at each point in the plot originated 8 days previously. The light sand-colored filaments seen near the south end of the DC-8 flight track (shown as a white line) suggest that the low-ozone bubble most likely originated at about 10°N. The RDF analyses support our hypothesis of the origination of the low-ozone bubble in the ITCZ as the region of the deep convective processes at low northern latitudes.
 Finally, MLS-initialized RDF calculations on the dense vertical grid were used to produce a cross section similar to Figure 3, albeit with much coarser time resolution. Figure 8c shows time-altitude distribution of the ozone mixing ratio from the RDF analysis (described above) that is adjusted by 25 ppbv to match the ozone range in Figure 3. Results are plotted between 340 and 375 K potential temperature levels and between 1600 and 1720 UT range that is similar to the spatial and temporal distribution of the DIAL ozone mixing ratio measured aboard the NASA DC-8 flight on 17 July 2007 (Figure 3b). Distinct ozone minima are seen near 365 K at approximately the times the ozone bubbles were sampled by the aircraft. The lowest ozone in the RDF calculations occurs slightly earlier in time along the flight track than observed. This is likely related to the coarse time resolution in the RDF calculations, but the low-ozone feature is apparent during the entire period it was observed by the aircraft. The vertical position and extent of the feature are very close to those in the aircraft observations. RDF calculations were also done using HIRDLS data; however, because of data quality issues (section 5), HIRDLS coverage below ∼370 K is insufficient for the initialization. A HIRDLS RDF section similar to Figure 8c shows a hint of a low-ozone feature just below 370 K, consistent with the MLS-derived feature in Figure 8c. Note that layers of enhanced ozone are seen both above and below the simulated low-ozone bubbles. The combination of RDF maps and cross sections suggests that the low-ozone bubbles are, in fact, the result of sampling across the narrow dimension of long streamers of ozone transported relatively intact over long distances and times. Note that the RDF calculations, by definition, include no mixing; the reasonable reproduction of the observed low-ozone bubbles in the RDF fields is thus an additional indication that the air comprising them has been transported over long distances with little mixing.
 While the TTL is a region that is infrequently perturbed by convection [Gettelman et al., 2002], chemical composition of the tropospheric part of the TTL, which lies below the tropopause, is more impacted by convective processes than that above the tropopause [Ricaud et al., 2007; Schiller et al., 2009, and references therein]. Consequently, ozone mixing ratios in the tropospheric part of the TTL are significantly greater (Figure 3, red color in DIAL data) than in the free tropical troposphere below 14 km, a region heavily influenced by air brought up via convection from near the surface, where ozone mixing ratios are low. This is shown by DIAL data in Figure 3b. A typical tropical ozone profile during the TC4 campaign showed ozone 50–60 ppbv through the bulk of the troposphere (see Figure 3 in the work by Thompson et al. ), with a sharp gradient increasing to maximum values beginning at 15 km.
 All ozone measurements from the Las Tablas, Panama, site taken during TC4 are shown in Figure 9. Other studies using TC4 data have shown that convection impacts ozone profiles [Avery et al., 2010; Selkirk et al., 2010; Thompson et al., 2010]. Figure 9 shows significant variability near 15 km, not unlike the range noted in the single DC-8 flight with the DIAL measurements. There was no Panama or Costa Rica ozonesonde launch on 17 July 2007, but the ozone mixing ratio was <40 ppbv, or 20–30% below the mean, at 12–14 km in the Costa Rica sonde from 16 July (Figure 2 in the work by Thompson et al. ). The variation at Panama is comparable to the spatial variation seen in the DIAL measurements, with minima on the order of 0.05 ppmv and maxima as high at 0.15 ppmv. Five of the Panama sondes launched during TC4 show low values (below 0.07 ppmv) in the 14–16 km level, which is ∼20% of the time. Moreover, the vertical distribution of ozone at Alajuella, Costa Rica (10°N, 84°W), shows the typical S-shape vertical structure near 15 km that is indicative of ongoing convective process [Thompson et al., 2010]. This indicates that such phenomena, as observed from the DC-8 in this case study, are not uncommon occurrences and hence likely contribute to the overall ozone budget in the UT.
 One goal of the TC4 mission was to characterize the chemical boundary conditions below the TTL, particularly for ozone. In the tropical upper troposphere, the chemical lifetime of ozone is about 50 days, which is much longer than the mixing time because of frequent strenuous ITCZ-related convection. In situ measurements from aircraft were analyzed to characterize the statistical vertical distribution of ozone created by the convective redistribution of ozone [Avery et al., 2010]. Avery et al.  suggested that very fresh convective outflow at 10–11 km altitude chemically more closely resembles in situ ozone sampled at about 3 km than it does at the surface, with very low variability in the measurements seen during updrafts. With vertical transport time scales on the order of 10–20 min, it seems unlikely for storms to efficiently entrain and mix midtropospheric air. This suggests that vertical transport in the middle troposphere is predominantly from 2–3 to 10–11 km and that vertical transport is more complicated than just moving boundary layer air up to the tropopause. Evidence from the sondes shows that local convection predominantly impacts ozone up to 11 km.
 For the case we have examined in this work, shortly after the first encounter of the low-ozone bubble event, the in situ and DIAL ozone mixing ratio data indicate an intercept of the thin depleted ozone layer located just above the deep convective clouds. It suggests the influence of shallower local convection at the DC-8 levels at ∼10.3 km, with the possibility of transport from the east. However, DIAL data show a break in vertical distribution with increased ozone mixing ratios at 13 km. Thus, the local convection does not reach up to the levels where the low-ozone bubble is detected. Therefore, the depleted ozone at 14–16 km was not the result of local uplift. The back trajectory runs indicate that the low-ozone features observed by DIAL were the result of deep convective upwelling in the ITCZ, followed by quasi-horizontal transport to south of the equator. It is somewhat surprising that the low-ozone features were so pronounced after moving around the UT for ≥1 week, which may reveal information on mixing time scales in the TTL. Our results are confined to assessing mixing times from our limited trajectories; in this case, we would infer mixing times greater than 1 week, which is not inconsistent with the results of James and Legras .
 Evidence for low-ozone bubbles is limited. In the DC-8 flights, the transit from California to Costa Rica also showed evidence of a thin low-ozone layer above the principal convective outflow signal that is very prominent around 10 km (Figure 6b in the work of Thompson et al. ). Indeed, analysis of wave signatures by Thompson et al.  in the Panama soundings reveals both convective and advective flows affect ozone structure in the TTL. Unfortunately, very few TC4 flights were able to measure the TTL ozone outside of the convectively influenced region near the Panama Bight and without major cloud interference. However, such low-ozone values in the TTL between 14 and 16 km have been observed in past experiments; low-ozone values in the TTL were seen in DIAL data during the Pacific Exploratory Mission A (PEM-A) and PEM-B campaigns [Browell et al., 2001, 2003]. However, it should be noted that the previous observations did not observe a similar, spatially coherent (∼2 km thick in vertical and ∼100 km long in horizontal dimensions) low-ozone “blob” such as seen during TC4. Because of the winds in this case, there is a contribution from convection that is north of the equator and likely from the Panama Bight region.
 Some other points may be useful for further discussion. For example, the DIAL ozone data on 17 July at 1710 UT show large ozone values just below the ozone bubble, larger than at similar altitudes on either side of the low-ozone area. High-ozone air parcels likely have stratospheric origin, either via local vertical exchange or quasi horizontal exchange across the subtropical jet. Because the vertical and horizontal winds at the TTL levels were very slow, quasi-horizontal transport from higher latitudes is the likely cause for the high ozone layer. This is supported by the RDF calculations, which showed alternating narrow layers of high and low ozone resulting from isentropic transport.
 There are other possibilities we have considered that could have caused the low-ozone features noted above 14 km on 17 July in the DIAL measurements. The DC-8 was flying in close proximity to the volcanic outflow from Ecuador. The chemical reaction involving volcanic SO2→H2SO4 particles could be the reason for the ozone destruction through the activation of chlorine. However, if that were the case, we would expect to see aerosols present in the low-ozone air mass. The contours in Figure 3 show aerosols in the regions of elevated ozone in the 14–16 km levels, indicative of an aged air mass, but there are no aerosols detected by DIAL in the low-ozone bubble, hence supporting the idea that it is a relatively “freshly” pumped up air mass.
 Yet another mechanism of ozone destruction could be related to the combination of high H2O mixing ratios and occurrence of clouds below (apparent from backscatter in DIAL aerosol backscatter data). It could lead to high photochemical destruction of ozone through photochemical loss J(O3)→O(1D) + H2O→OH, followed by additional ozone loss through the HO2/OH catalytic cycle. However, the DIAL/LASE system did not detect high mixing water vapor above the DC-8 level, negating this possibility. Moreover, NO abundance in the UTLS is high enough for supporting ozone production mechanism and thus negating the ozone loss processes. Therefore, enhancements in HOX tend to improve chemical ozone production even further.
 To summarize, based on our back trajectory analysis, Figure 5 (top) shows that the coherent low-ozone bubble observed in the south part of the DC-8 flight of 17 July 2009 by the DIAL instrument at ∼1700 UT between 14 and 16 km (Figure 3, green) most likely resulted from nonlocal convection occurring near Panama. Supporting evidence for the nonlocality and subsequent transport comes from the fact that a low-ozone region crossing the back trajectories is seen in satellite observations in the absence of convection several days after the convective event (Figure 6). There is a distinct difference in the direction of trajectories derived above aircraft level (Figure 5, top) and at aircraft level (Figure 5, bottom, South America), also supporting the nonlocality of the source. Therefore, ozone depletion above the aircraft and elevated/reduced ozone mixing ratios at the aircraft level, and increased ozone below the aircraft, are governed by different processes. The picture of complementary convective and advective influences on ozone structure in the UT and the TTL is consistent with analysis based on wave patterns in the ozone sounding data over Panama and Costa Rica during the TC4 campaign [Thompson et al., 2010].
 This work was supported by the NASA Headquarters Atmospheric Composition Focus Area including the Upper Atmospheric Research Program (Michael Kurylo, program manager), the Radiation Science Program (Hal Maring, program manager), and the Tropospheric Chemistry Program (Jim Crawford, program manager). We gratefully acknowledge helpful discussions with R. McPeters (NASA, Goddard), K. Chance (Harvard University), and E. Hilsenrath (NASA Headquarters). We also emphasize the crucial contributions of the pilots and crew of the NASA DC-8 aircrafts. We extend our gratitude to mission scientists (Brian Toon and Dave Starr) and DC-8 platform scientists (Mark Schoeberl and Paul Wennberg) for planning and successfully executing the TC4 campaign. We greatly appreciate support from the Aura HIRDLS, OMI, and MLS teams for providing us with the coincident data. We extend our special thanks to J. Gille and S. Karol (HIRDLS, NCAR) for help with data quality, analysis, and discussion. We also thank Marc Kroon (OMI, KNMI) for help with the OMI DOAS data analysis, updates, and conscientious figures. The OMI-TOMS and OMI-DOAS total ozone data were obtained from the NASA Goddard Earth Sciences (GES) Data and Information Services Center, home of the GES Distributed Active Archive Center. Work at the Jet Propulsion Laboratory, California Institute of Technology, was done under contract with NASA. We extend special thanks to Kurt Severance (NASA, Langley) and William Daffer (NASA, JPL) for help preparing the 3-D graphics in record short time. Finally, we acknowledge the hard work by Gary A. Morris (Valparaiso University) and Alex Bryan and David Lutz (Valparaiso University undergraduates), who were responsible for all 25 ozonesonde launches from Las Tablas. Without their effort, we would have no balloon data from Panama.