Exploring the elevated water vapor signal associated with the free-tropospheric biomass burning plume over the southeast Atlantic Ocean

. In southern Africa, widespread agricultural ﬁres produce substantial biomass burning (BB) emissions over the region. The seasonal smoke plumes associated with these emissions are then advected westward over the persistent stratocumulus cloud deck in the Southeast Atlantic (SEA) Ocean, resulting in aerosol effects which vary with time and location. Much work has focused on the effects of these aerosol plumes, but previous studies have also described an elevated free-tropospheric water vapor signal over the SEA. Water vapor inﬂuences climate in its own right, and it is especially important to consider 5 atmospheric water vapor when quantifying aerosol-cloud interactions and aerosol radiative effects. Here we present airborne observations made during the NASA ORACLES (ObseRvations of Aerosols above CLouds and their intEractionS) campaign over the SEA Ocean. In observations collected from multiple independent instruments on the NASA P-3 aircraft (from near-surface to 6-7km), we observe a strongly linear correlation between pollution indicators (carbon monoxide (CO) and aerosol loading) and atmospheric water vapor content, seen at all altitudes above the boundary layer. The focus of the current study 10 is on the especially strong correlation observed during the ORACLES-2016 deployment (out of Walvis Bay, Namibia), but a similar relationship is also observed in the August 2017 and October 2018 ORACLES deployments. Using ECMWF and MERRA-2 reanalyses and specialized WRF-Chem simulations, we trace the plume-vapor relationship to an initial humid, smoky continental source region, where it is subjected to conditions of strong westward advection, namely the South African Easterly Jet (AEJ-S). Our analysis indicates that airmasses likely left the continent with the same relationship between water vapor and carbon monoxide as was observed by aircraft. This linear relationship developed over the continent due to daytime convection within a deep continental boundary layer (up to ∼ 5-6km) which produced fairly consistent vertical gradients in CO and water vapor, decreasing with altitude and varying in time, but does not originate as a product of the BB combustion itself. Due to a combination of conditions and mixing between the smoky, moist continental boundary layer and the 5 dry and fairly clean upper-troposphere air above ( ∼ 6 km), the smoky, humid air is transported by strong zonal winds and then advected over the SEA (to the ORACLES ﬂight region) following largely isentropic trajectories. HYSPLIT back trajectories support this interpretation. Better understanding of the conditions and processes which cause the water vapor to covary with plume strength is important to accurately quantify direct, semi-direct, and indirect aerosol effects in this region.

that the SE-to-NW diagonal (passing through Zones 1, 3, 5, 7, and 8) includes six "routine flights" overlaying one another. Reddish circles indicate locations of the 95 partial or full aircraft vertical profiles during all flights which will be discussed in more detail in Section 3.3. The blue boxes indicate the regional subsets (labeled Zones 1-8) used in the spatially-subdivided aircraft analysis in Section 3, and the lavender boxes show the oceanic and continental regions used for the reanalysis analysis in Section 3.4.
In this paper, we use recent aircraft measurements over the SEA Ocean, combined with large-scale meteorological reanalyses and specialized models, to identify and explore this feature of co-located humidity and BB plume. With the new aircraft-based observations discussed here, we are able to gain a better understanding of this relationship than was previously possible.
In the bulk of our analysis, we use carbon monoxide (CO) as a tracer of biomass burning emissions. CO is not aged or removed by cloud processes as the BB aerosols are, and thus is a more reliable indicator to determine air mass origin than 5 the aerosols themselves. Modeled CO is also more robust than modeled outputs of individual aerosol species (e.g., Shinozuka et al., 2020), and thus allows analysis of airmass origins and trajectories by comparing to these products. However, the results we show using observed CO are largely consistent with results using aircraft-measured aerosol extinction or scattering.
The NASA ORACLES (ObseRvations of Aerosols above CLouds and their intEractionS) campaign was a 5-year, multiinstitutional project to study the effects of biomass burning aerosols and their interactions with the southeast Atlantic stratocu-10 mulus deck Redemann et al., 2020). ORACLES had three field deployments during the African biomass burning season: out of Walvis Bay, Namibia, in September 2016, and out of São Tomé, São Tomé e Príncipe in August 2017 and October 2018. Each of these deployments used a NASA P-3 aircraft for tropospheric sampling (roughly 0-7km), although 4 https://doi.org /10.5194/acp-2020-1322 Preprint. Discussion started: 21 January 2021 c Author(s) 2021. CC BY 4.0 License. the 2016 deployment had an additional high-flying ER-2 aircraft (above 20km) for downward-looking remote sensing measurements. There are significant logistical and meteorological differences between each deployment; due to the different seasonal timing (by design of the ORACLES campaign) and the different deployment locations of 2016 versus 2017/2018, we analyze each deployment separately. In this work, we focus on data from the P-3 aircraft during the September 2016 deployment, with some discussion of the August 2017 and October 2018 observations to provide insight into the multi-year context. A more 5 detailed discussion of the three ORACLES deployments may be found in Redemann et al. (2020). Figure 1 shows all P-3 flight paths and the locations of aircraft profiles for 2016 (i.e., the main focus of the present paper), as well as some key spatial delineations which we will use.
In Section 2, we introduce the instruments, data, and reanalysis and model products used. In Section 3.1, we present analysis of the atmospheric humidity as measured by three independent instruments aboard the P-3 aircraft, and in Section 3.2 we 10 discuss how the water vapor relates to the presence of the biomass burning plume over the SEA in ORACLES-2016. We next compare our observations to reanalysis products and model outputs over the SEA (Section 3.3), and over the continental source region (Section 3.4). In Section 3.5 we briefly discuss the 2017 and 2018 results and how they differ from the 2016 deployment.
In Section 4, we synthesize the results before discussing potential causes of the observed patterns, their context within previous studies of the region, and their potential radiative implications. 15

Instruments and Methods
In this work we use observational data from ORACLES in conjunction with large-scale atmospheric reanalysis and the outputs of specialized model configurations, as described below.

Aircraft instrumentation
The observational data considered here are from the ORACLES dataset. The full dataset is archived at https://doi.org /10.5067/ 20 Suborbital/ORACLES/P3/2016_V1 for the 2016 deployment, https://doi.org/10.5067/Suborbital/ORACLES/P3/2017_V1 for 2017 and https://doi.org/10.5067/Suborbital/ORACLES/P3/2018_V1 for 2018. All instruments used here were deployed on the P-3 aircraft during all three ORACLES deployments. 1 Hz measurements are used unless otherwise indicated. Individual flights were classified as either "routine flights," which in 2016 extended along a diagonal flight path from (20 • S, 10 • E) to (10 • S, 0 • E), or "flights of opportunity," which focused on specific science objectives and were largely nearer to the Namibian/Angolan 25 coast ( Figure 1). A more complete overview of the ORACLES operations and major results can be found in Redemann et al. (2020).

COMA
In all ORACLES deployments, volume mixing ratios of carbon monoxide (CO), carbon dioxide (CO 2 ), and water vapor (q) 5 were measured by a Los Gatos Research CO/CO 2 /H 2 O Analyzer (known as COMA), modified for flight operations. It uses off-axis integrated cavity output spectroscopy (ICOS) technology to make stable cavity enhanced absorption measurements of CO, CO 2 , and H 2 O in the infrared spectral region, technology that previously flew on other airborne research platforms with a precision of 0.5 ppbv over 10s (Liu et al., 2017;Provencal et al., 2005). Water vapor measurements of less than 50 ppmv (∼0.03 g/kg) were removed due to instrument limitations, but this has minimal effect on the data considered here. 10 The CO measured during ORACLES is used in the present work as a tracer for air masses originating from combustion.
While a major focus of ORACLES is the radiative effects of aerosols, CO will be conserved even under cloud processing which may affect the aerosol concentrations from biomass burning, and thus provides valuable information on air mass origin (and simplifies the comparison to modeled parameters). 15 Atmospheric water vapor was also measured as part of the Water Isotope System for Precipitation and Entrainment Research (WISPER) data (reporting H 2 O concentration, D/H and 18 O/ 16 O isotope ratios). For ORACLES WISPER was continued to use a pair of gas phase isotopic analyzers based on the Picarro Incorporated L2120-i Water Vapor Isotopic Analyzer (Gupta et al., 2009). Coupled to the near-isokinetic SDI inlet, the system reports total water (vapor plus condensate), which can be interpreted as vapor when out of cloud. Air was sampled from the inlet flow at 2.5 slpm via a 6 meter long thermally-insulated 20 copper transfer line heated to 50 • C to minimize any wall effects and avoid possible condensation in the lines. The exterior portion of the SDI inlet was unheated. Two different Picarro L2120-i instruments were used during the 2016 campaign, one for the dates up to and including 04 September, and another for later dates. The switch was associated with an instrument failure that led to poor data recovery on 3 of the 14 flights (Table 1). The instrument used in the first part of the campaign reports data at 5Hz while the instrument used later in the campaign reports at 0.5 Hz. Both data are aggregated onto a 1Hz common 25 time using simple binning, and synchronized to the data system using cloud probes timing when entering/exiting clouds. Time synchronization has an uncertainty of about 1 second. Calibration of the system based on pre-campaign lab calibration using a LI-COR Model 610 dew point generator at a fixed temperature, with air diluted with ultra-zero grade dry air to span low concentration range using quantitatively calibrated mass flow controllers. The water vapor measurements are valid to 10ppmv (0.016 g/kg) and precision was typically reported as between 9-50ppmv (0.01-0.08 g/kg), with greater values corresponding to 30 lower absolute water vapor amount.

P-3 aircraft data
The P-3 aircraft is equipped with instrumentation to make a number of standard on-board measurements of environmental data such as temperature, pressure, relative humidity, and wind speed. A full description of the on-board instrumentation may be found in Section 4.6 of the aircraft handbook at https://airbornescience.nasa.gov/sites/default/files/P-3B%20Experimenter% 20Handbook%20548-HDBK-0001.pdf. The aircraft-based specific humidity (q) considered here was calculated from the re-5 ported dew point temperature (from an EdgeTech Model 137 aircraft dew point hygrometer) and static pressure (from a Rosemount MADT 2014 sensor) values following Vaisala (2013): where p meas is the measured static pressure, m r is the ratio of the molecular weight of water vapor to air (18.015/28.97), and p ws is the simplified formula for water vapor saturation pressure over water given as 10 p ws = A × 10 mT dp /(T dp +Tn) , where T dp is the measured dew point temperature and the constants A, m, and T n are 6.116441 hPa, 7.591386, and 240.7263 • C, respectively (Vaisala, 2013). The static pressure measurements have a precision of 0.5 hPa and an accuracy of ±2.5 hPa. For the dew point hygrometer, measurement precision was 0.1 • C and an accuracy of 0.2 • C nominally, with greater uncertainty below 0 • C and during profiles with large δT dp /δT .

Large-scale reanalyses and models
In conjunction with these observations, we select two large-scale reanalyses, which assimilate satellite observations and thus should be consistent with conditions observed by aircraft; and two free-running models, which, due to their unconstrained nature, may help to diagnose which processes are/not in play. The reanalyses considered are the latest iteration of the European Centre for Medium-Range Weather Forecasts (ECMWF) reanalysis, ERA5 (CDS, 2017) and NASA's Modern-Era Retrospec-20 tive analysis for Research and Applications, Version 2 (MERRA-2; Gelaro et al., 2017). The former was chosen due to its exceptionally good agreement with the ORACLES observations (Section 3.3), and the latter was chosen as it incorporates aerosol observations, the lack of which is a shortcoming of the ERA product. We also briefly show results using the previous ERA-Interim (Dee et al., 2011) for continuity with previous work.
For the models, we consider two different specialized configurations of WRF developed in support of the ORACLES mis-25 sion, termed WRF-CAM5 and WRF-Chem for consistency with previous studies (e.g., Shinozuka et al., 2020). The similarities and differences between each of these products is not the focus of the present paper, but the results of the differences between each product allows us to diagnose the influence of potential drivers in the real world.

ECMWF reanalyses
The European Centre for Medium-Range Weather Forecasts (ECMWF) has developed global atmospheric reanalysis products 30 for several decades, with the ERA-Interim (Dee et al., 2011) serving as the primary reanalysis product through mid-2019, before being surpassed by the recently-released ERA5 (Hersbach et al., 2019). ERA5 is considered at 0.25-degree, hourly resolution in the comparison with ORACLES flights (Section 3.3), and 0.25-degree, 3-hourly resolution in the continental analysis (Sections 3.4 and 4.1). ERA5 does not report atmospheric chemistry or aerosols nor does it directly incorporate aerosol effects, though satellite measurements of aerosol-influenced radiances are incorporated into the reanalysis. ERA-Interim was only available at 3-hourly resolution. Due to the timing in the ERA5 dataset release, we explored results using both of these products in Section 5 3.3, and found ERA5 performs generally better compared to the observations. In Supplementary Materials (Figures S1 and S2) we provide selected comparisons between ERA5, ERA-Interim, and observations over the SEA.

MERRA-2
The Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2) is an atmospheric reanalysis produced by NASA's Global Modeling and Assimilation Office (GMAO) (Gelaro et al., 2017;Randles et al., 2017;Buchard 10 et al., 2017). MERRA-2 assimilates observations of meteorological parameters from multiple satellite platforms, as well as aerosol optical depth from satellites (MODIS, AVHRR) and ground-based (AERONET) measurements, into a comprehensive atmospheric model, with explicit accounting of aerosol radiative effects. MERRA-2 datasets are given on a nominal 50 km horizontal resolution (0.5 • × 0.5 • ) with 72 vertical layers from the surface to 0.01 hPa. An additional goal of the ORACLES campaign was to evaluate chemical transport models and reanalysis products such as MERRA-2, and to this end the complete 15 set of MERRA-2 files have been sampled up to 1-second resolution along every ORACLES flight (Collow et al., 2020). These products are available online at https://portal.nccs.nasa.gov/datashare/iesa/campaigns/ORACLES/. Over the larger continental and oceanic domain, both MERRA-2 and ERA5 are considered at 3-hourly temporal resolution.

WRF-CAM5
The WRF-CAM5 configuration was run at 36km horizontal resolution over the month of September were turned on, and a smoke plume rise process was enabled.
Initial and boundary conditions from ERA5 and CAMS reanalysis were used to account for the meteorology and chemistry and aerosols, respectively. Simulations were initialized every day at 00Z and ran for 30 hours. The first 6 hours were discarded to account for the meteorology spin up. We consider the period between 15-31 August (17 days) as a spin up for chemistry and aerosols. CAMS was used for boundary conditions during the whole simulation to account for possible intrusion of aerosols outside the domain (e.g. Saharan dust, smoke from Madagascar, sea salt). CAMS was used only for the 15 August initialization, and subsequent simulations were initialized by recycling the chemistry and aerosols from the previous run. In this manner, we can assume that all chemistry and aerosols used here are explicitly calculated by the model. In contrast, ERA5 was used for 10 initialization and boundary conditions throughout the whole simulation (i.e., at daily reinitialization).

NOAA HYSPLIT trajectories
We ran NOAA's Hybrid Single-Particle Lagrangian Integrated Trajectory model (HYSPLIT; Stein et al., 2016) to trace airmasses sampled by aircraft profiles towards their origins. Runs were computed offline using a standard HYSPLIT backtrajectory configuration. As ERA5 is not currently available as a HYSPLIT meteorological input, the meteorology used is from 15 the National Centers for Environmental Prediction (NCEP) Global Data Assimilation System (GDAS) 0.5-degree model, provided directly by NOAA HYSPLIT (ftp://arlftp.arlhq.noaa.gov/pub/archives/gdas0p5), which is the highest resolution available for 2016. Trajectories are run using vertical motion determined alternately by the default "model motion" (kinematic trajectories using winds from the GDAS meteorology) and using isentropic pathways calculated from GDAS potential temperature fields (https://www.arl.noaa.gov/documents/workshop/NAQC2007/HTML_Docs/trajvert.html). 20 3 Results

Measured humidity from different ORACLES instruments
Before presenting the analysis of the BB plume as it relates to the humidity, we first show the robustness of the water vapor measurements by comparing the three independent instruments available during ORACLES: COMA, WISPER, and the dewpoint hygrometer ("Aircraft") data from an onboard hygrometer (Table 1), as were described in Section 2.
25 Figure 2 shows measurements from the three water vapor instruments for all 2016 flights at 1s resolution, for the full dataset and for specific subsets based on altitude (i.e., excluding layers which are clearly boundary layer altitudes) or water vapor gradient (i.e., to minimize the effect of varying instrument response times). The correlations in all cases are robust and statistically significant (R 2 > 0.97 for all data; Table 2), likely due in part to the large amount of data collected. Significant deviations from the 1:1 line (grey dots) occur either during high humidity conditions within the planetary boundary layer, or 30 during a rapid change in water vapor conditions, which can be explained in part as due to inlet differences and related issues  of differing instrument response times. Deviations are expected during transitions from in-cloud to out-of-cloud conditions since each inlet system has differing heating to manage (or otherwise avoid) condensation artifacts. Focusing specifically on the in-plume conditions, we find that the instruments shows quantitatively consistent water vapor measurements, with slopes of total least-squares fits between 0.98 and 1.01. These strong correlations between independent instruments on the same platform indicate that the observed water vapor signal is robust.

5
Having established that we have good confidence in the robustness of the water vapor data measured by multiple instruments, in the following sections, we focus largely on COMA water vapor. This instrument measures q with greater precision than the aircraft probe, and more data are available from COMA than from WISPER for flight times either within the biomass burning plume, or profiling the full atmosphere (Table 1), which are the sampling times on which we focus in this study. Additionally, while temporal corrections have been applied to synchronize the various instruments against one another, COMA CO and q

ORACLES in-plume measurements
Examining the correlation between the biomass burning tracer CO and the water vapor content q within the plume layer (i.e., excluding boundary layer altitudes, here defined as below 2km), we see a consistent pattern of elevated humidity with high CO. Figure 3 shows correlations between CO and q for each individual flight, for all altitudes above 2km (to isolate the plume altitudes from those with boundary layer influence). We find similar results for a variety of spatial, altitudinal, and temporal 10 subsets above the planetary boundary layer; in other words, there does not appear to be a single altitudinal or latitudinal range which dominates this relationship for the dataset as a whole. We note that the results are also similar for aerosol extinction and scattering coefficient (Supplementary Figure S3). The amount of water vapor seen here is consistent with the 5g/kg moisture levels reported by Deaconu et al. (2019) for August and September (compared with 2.5g/kg in June/July), and indeed during ORACLES we frequently see values of 6g/kg or greater in the free troposphere. 15 Each of the flight days in Figure 3 shows a robust linear correlation, and some of the flights show especially linear correlations between CO and q, specifically the flights on 8, 10, 12, and 14 September 2016 (middle row). The first three of these flights were along the routine diagonal covering a fairly significant portion of the SEA extending northward to 10 • S, 13.5 • S, and 9.5 • S, respectively. The flight on 14 September is classified as a radiation flight of opportunity, and while it did not follow the routine path, it still sampled somewhat diagonally from Walvis Bay (out to 16 • S and a maximum westward extension of 20 7.5 • E). While the correlations appear as generally stronger for routine flights, most of the flights of opportunity show strong correlations as well (R 2 > 0.8; Table 3). The notable exceptions to this are the flights of opportunity on 20 and 24 September and the red line shows the fit through all flights combined. All correlation coefficients are significant to two decimals (p < 0.01).
which were both particularly close to the coast; when subdivided spatially over all flights, the relationship is more variable for the more southern coastal areas ( Figure 4). On 20 September, dust was also observed during a portion of the flight; this could indicate that the air mass sampled on these days had a different origin and different trajectory upon exiting the continent (i.e., directly easterly), compared with the typical conditions of the elevated biomass burning aerosol layer (i.e., a more northwesterly recirculation from an origin at AEJ-S latitudes). On 24 September, an unusually high boundary layer height was observed with 5 westerlies below 3.5km; this anomalous meteorological condition between 15-20 • S may be responsible for the slightly weaker correlation that day. The routine flight on 25 September also has a lower correlation than the other flights. Examining the data shows that there is a shift in the CO-q slope with altitude during this day, which is not seen on other flights; for smaller altitude subsets during this flight, the correlation is stronger. We also note that these flights are the last flights of the deployment, and thus may be capturing an expected seasonal shift compared with the flights earlier in September.
10 Figure 4 shows a frequency distribution of these same data subdivided spatially, highlighting how remarkably consistent the slope of this relationship is. Humid, smoky air exits the continent at roughly 10 • S in the south Atlantic Easterly Jet (AEJ-S), but we see that even at higher latitudes (lower rows), farther from the latitudes of the AEJ-S, the CO-q relationship is strongly Table 3. Correlations between free tropospheric CO and q (z > 2km) from in situ instruments as shown in Figure 3, and correlations between AOD and CWV (z > 1.3km) from 4STAR as shown in Figure 5, by flight for ORACLES 2016. All correlations are significant to p < 0.001.
Note that the different altitude limits are due to different methodologies. "Routine" and "opportunity" indicate whether the flights were along the northwest diagonal or near-coast (Section 2.1).
date flight CO vs q 2-6.3km AOD vs CWV 1.3-5km observed are present as a given airmass exits the continent, and is not progressively diluted via mixing during transport.

Column ORACLES measurements
The 4STAR retrievals of AOD and column water vapor (CWV) are measured along the aircraft-to-sun light path and thus 5 represent the full above-aircraft airmass, rather than the values at the aircraft altitude. While some impact of ambient humidity is to be expected due to hygroscopic swelling of aerosols (increasing AOD), it is nonetheless still instructive to examine these parameters as they compare to inlet-based instruments. A prior study of the ORACLES-2016 data (Shinozuka et al., 2020) estimated that the effect of aerosol hygroscopic swelling on extinction was fairly minimal in the free troposphere, with an ambient-RH/dry ratio less than 1.2 for 90% of measurements, suggesting the same may be true for AOD. Figure 5 shows the 10 correlation between the 4STAR AOD at 500nm and the CWV for 1s data from all 2016 flights from above the boundary layer (here, > 1.3 km) to upper plume level (≤ 5 km). The 4STAR instrument provides a different geometric perspective from that of the inlet-based measurements described above, yet shows similar results, providing additional evidence of the observed linearity between q and smoke concentration. The different altitude ranges compared with Figure (Table 3) and only slightly lower for all flights combined (R 2 = 0.82). instrument requirements and capabilities, i.e., 4STAR observations from within the plume give only partial vertical profiles as 4STAR measures only the airmass above the aircraft at a given time. Thus measurements from below the BB aerosol plume are valid and even preferred for 4STAR, whereas the inlet-based instruments are less useful at lower altitudes when there is a lack of plume loading. The largest range in AOD (and CWV) is seen on 24 September, near the coast, consistent with Figures 3 and 4.

5
The 4STAR observations demonstrate that the plume/vapor relationship is consistent through the plume column and not solely at the instantaneous altitudes and locations as seen by the inlet-based instruments. We note this is also consistent with the results of Adebiyi et al. (2015) who showed that upper-level (∼ 700hPa, roughly 3.2km) humidity from radiosondes corresponded to conditions of high AOD from satellites, albeit this was farther offshore at St Helena Island. The fact that we see a strong linear correlation between markers of the biomass burning plume and atmospheric water vapor from multiple 10 instruments and over multiple flights is a strong indication of the robustness of this relationship over this region during the ORACLES-2016 time period.
3.3 Do reanalyses/models capture the relationship seen in the observations?
Having established the robust CO-q relationship over the southeast Atlantic Ocean as seen in these observations during ORACLES-2016, we next seek to explore the larger mechanisms by which this relationship has developed. The source region for ORACLES BB observations includes widespread seasonal grassland savannah fires over central and southern Africa (e.g., van der Werf et al., 2010;Redemann et al., 2020) and sees little variability in either fuel source or combustion efficiency 5 (e.g., Vakkari et al., 2018). We wish to take a broader perspective which incorporates these continental regions, which is not possible solely using the over-ocean ORACLES aircraft data alone. Thus, we turn to reanalyses and model simulations. Figure 6 shows the ORACLES flight data from aircraft profiles aggregated and subset to the times and altitudes of the ERA5, ERA-Interim, and MERRA-2 reanalyses, and the WRF-CAM5 and WRF-Chem models, with different reanalysis altitude ranges distinguished by color and shape. We note that this is a subset of the data shown in Figure 3, but the CO-q relationship 10 shown here is consistent with that of the full dataset. For each of the altitude ranges-boundary layer (square), boundary layerinfluenced (triangle), or plume-level (circle)-there is good agreement between ERA5 and the aircraft observations (Figure 6a) from the surface through the plume level. An exception is at altitudes at the top of the boundary layer ( ∼570m; squares), where ERA5 often underestimates water vapor, perhaps due to difficulties in determining boundary layer height over the ocean surface. Despite this, the humidity at surface level agrees well with the observations and, more importantly in the context of 15 this study, the existence, magnitude, and location of elevated water vapor for plume altitudes is also well represented in ERA5.
It is reassuring that this newest ECMWF product agrees so well with the aircraft observations (R 2 = 0.79 for z > 2km), and this gives us confidence that the ERA5 meteorology may be consistent with real-world meteorology over the continental source region as well. Figure 6b and Figure 6c show the comparisons between aircraft-observed q and ERA-Interim and MERRA-2 reanalysis q, respectively. Both of these correlations are rather weaker than that for the ERA5 reanalysis (R 2 = 0.53 and 0.40 20 for ERA-Interim and MERRA-2, respectively), but both still largely capture the presence of an elevated water vapor signal in the altitudes above the boundary layer. However, both these products also often report this high-humidity air as being at a lower altitude than what was observed by the aircraft observations (an example is shown in Figure S2).
Finally, Figure 6d and 6e show the two configurations of WRF described in Sections 2.2.3 and 2.2.4. WRF-Chem q shows a strong correlation with the observed q, in line with that of ERA5 (R 2 = 0.79 for both products for all altitudes > 2km), 25 which is not surprising due to WRF-Chem's daily initialization with ERA5 reanalysis meteorology. The WRF-CAM5 water vapor is more weakly correlated with the observed water vapor (R 2 = 0.48, more in line with the results from MERRA-2 and ERA-Interim). This difference is likely due in part to the different meteorological fields used (NCEP versus ERA5), and also to WRF-CAM5's less frequent initializations (5-day versus 1-day), allowing it to drift farther from the "actual" meteorology and chemistry conditions between initializations. Given these results alone, one might be discouraged by the possibility of using 30 MERRA-2 or either WRF configuration in this analysis, but this isn't the full story. Although the water vapor co-location is poor, we find that the relationship between CO and q does hold over the flight path ( Figure 7). Here, interestingly, the results are flipped: MERRA-2 and WRF-CAM5 show comparatively better correlations between CO and q (R 2 = 0.56 and 0.71, respectively, compared with R 2 = 0.78 in the observations), while WRF-Chem now shows more variability in CO-q conditions d. e. and thus a poorer correlation between the two (R 2 = 0.49). The fact that the CO/q correlation is fairly high for MERRA-2 and WRF-CAM5 even while the observed/modeled q correlation is low essentially indicates that while these two products aren't placing a given airmass exactly where and when it is observed by the aircraft, the consistent relationship between the plume and water vapor is maintained in the alternate location. We must also consider the differences in model emissions and meteorological configurations to potentially explain this. MERRA-2 and both WRF models use QFED emissions, albeit with 5 different implementation in each. Because WRF-CAM5 has the best correlation between CO and q, and the longest independent run length, it seems plausible that the periodic reinitialization of each model's meteorology independent of its emissions weakens the correlation between the two. This would be because the reinitialization will "correct" the meteorology (water vapor) towards the reanalysis, while the chemistry (CO) will be adjusted independent of the meteorology, and to a different degree. This would explain why the 3-day runs (5-day minus 2-day spin-up) of WRF-CAM5 show a stronger correlation 10 than WRF-Chem (with daily reinitialization) or the MERRA-2 reanalysis. We also note that MERRA-2 and WRF-CAM5 report lower CO for higher water vapor (i.e., the slope between the two variables is steeper than in the observations) whereas the opposite is true for WRF-Chem. Overall, this pattern suggests that the CO-q relationship is sustained through dynamics affecting both properties equally, i.e., not diabatic processes such as cloud formation which could decrease the water vapor, a. b.
c. d. and moreover that this is also true within the considered models. Given this context, we conclude that, while not perfect, the different strengths (and limitations) of each of these models may be useful in understanding the mechanisms involved in the real world. Figure 8 shows vertical profiles of water vapor from COMA subdivided spatially by latitude and longitude grids according to the boxes shown in Figure 1 (the same divisions used in Figure 4), with routine flight paths in the left column and coastal 5 flights on the right. Each subplot shows profiles of the nearest co-located ERA5 reanalysis points, for comparison. This spatial division by aircraft profile highlights both the consistency in the vertical structure of the plume observed by aircraft and shown by ERA5, and the differences in this vertical structure in different regions of the SEA. In terms of the spatial differences, Zone 2 (top right) has consistently the highest measured water vapor (4-11 g/kg) and CO, possibly due to its proximity to the location of the AEJ-S (∼ 10 • S). Also, along the routine diagonal (i.e., farther from the coast), we more frequently see a dry/clean gap 10 between the humid plume and the more humid boundary layer, plus a greater plume strength compared with the near-coast regions at the same latitude (see also Figure 4). In contrast, the more coastal flights often see either more humid, higher-CO air masses at lower altitudes, or constant CO and q at all altitudes ( Figure 4). Finally, we note that Figure 8 shows again how consistently well the ERA5 reanalysis performs when compared to the aircraft observations, even in the case of varying profile type. There is a good deal of variability in this structure in different latitude/longitude ranges (e.g., high-and low-altitude 15 plumes with substantial vertical variation or a fairly consistent magnitude with altitude) but these differences are consistent between both ERA5 and the observations. Our results thus far are consistent with previous satellite-and reanalysis-based work which described both the same pattern of elevated water vapor coinciding with biomass burning aerosols over different parts of the SEA (e.g., Adebiyi et al., 2015;Deaconu et al., 2019), and the importance of the south African easterly jet (AEJ-S) in transporting continental airmasses over the southeast Atlantic Ocean (e.g., Adebiyi and Zuidema, 2016). Having shown that several models and reanalyses are able, to 5 some degree, to capture the presence of an upper-level water vapor signal during ORACLES-2016, in this section we focus on the reanalyses to gain more insight into the origins of this pattern over the biomass burning source region. Specifically, we may reasonably expect that due to the excellent agreement between the ERA5 reanalysis and the observations in the ORACLES SEA sampling region, ERA5 may give an accurate picture of meteorological context for the airmass origin over the continent and its evolution during its westward transport. MERRA-2, while not as directly translatable to aircraft measurements, may yet 10 allow us to complete the picture by showing how q relates to CO concentration.  Figure 1). These q contours are overlaid with average horizontal wind vectors at the same altitude. A few features are 15 obvious from this reanalysis: first, multi-day episodes of high water vapor conditions are seen to originate over the continent and are advected westward when zonal wind speeds are high. That is, an elevated water vapor signal is frequently present up to 5km over the continent and these humid airmasses are transported in the easterly jet only under conditions of high zonal wind speeds. Second, we note that there is a notable diurnal cycle in q over the continent, likely driven by the diurnal cycle in the continental boundary layer development. The timing of the diurnal maximum q varies substantially with altitude (as will 20 be discussed shortly). While Figure 9 shows the 600hPa pressure level, the results are largely the same for pressure altitudes 700-500hPa (i.e., the range of the AEJ-S; Adebiyi and Zuidema, 2016), and for latitude subsets within this range. For more southern latitudes, the reanalysis shows much weaker zonal winds, less water vapor at higher altitudes, and no direct connection between continental and over-ocean conditions at the same latitude; the direct east-west transport is not observed. While the AEJ-S ranges from 5-15 • S, between ∼ 5 and 8 • S there is likely a combination of dry and moist convection present, whereas 25 dry convection is likely to dominate south of 10 • S. Either type of convection will result in elevated q at the AEJ-S altitudes.
This pattern of transport is consistent with the BB source region being at more equatorial latitudes even for the more southern ORACLES observations, i.e., recirculation of smoky, humid air from the north to the south, as was also shown by Adebiyi and Zuidema (2016). The broader meteorological features were discussed in more detail in Redemann et al. (2020).
A similar pattern is seen in CO reported by MERRA-2 ( Figure 9b): periodic events of westward CO transport are co-located 30 with water vapor transport events, driven by the zonal winds. Both the zonal winds and water vapor are generally similar between the MERRA-2 and ERA5 reanalysis. The water vapor and CO are qualitatively similar in the WRF models as well, although we observe a distinct discontinuity in the timeseries of these models which corresponds to the (daily for WRF-Chem, or 3-daily for WRF-CAM5; Figure S4)   Wind vectors do not scale between the two panels, although the patterns are seen to be largely similar between the two reanalyses.
responsible for the weaker correlations in these products (WRF-Chem in particular; Figure 7), as q and CO are adjusted to differing degrees during this process. The fact that the correlations persist between reinitializations but then are lost again suggests that any removal/mixing processes over the SEA Ocean are affecting CO and q equally; i.e., the air is not subject to significant diabatic processes or cloud formation during transport, which could lower q without affecting CO.   boundary layer influence reaching to above 5km, propagating upward in time. While the presence of the AEJ-S over the SEA corresponds to significant carbon monoxide, we also see how this high-CO airmass may disperse out into the broader region (e.g., the episode starting around 4 September at 3km over the ocean region is transported down to 1km by 6 September in the absence of the strong zonal winds). The direct comparison between MERRA-2 profiles and aircraft observations suggested a potentially too-strong subsidence, resulting in a lower-altitude q maximum (Figures 6, S2); indeed, Das et al. (2017) previously 5 documented a subsidence in MERRA-2 which was greater than that inferred from satellite observations. For this particular instance there was a sustained downward motion at 700hPa in both ERA5 and MERRA-2 between 4-6 September, which may be responsible for this episode seen in both reanalyses (Figures 10a, 11a). Regardless, even in a case of too-strong subsidence in MERRA-2, this issue itself will not affect the relationship between CO and q once it's over the SEA, but rather just its location. It is clear from the two reanalyses that continentally-influenced air over the SEA remains for a sustained period of 10 time and is transported both horizontally and vertically throughout the region while retaining high-q and high-CO amounts.
Further insight can be gained by examining the diurnal cycle directly at individual pressure levels. Figure 12 shows time series of key meteorological parameters: zonal winds, water vapor, pressure vertical velocity, and potential temperature (u, q, ω, and θ, respectively) from ERA5, and the same parameters plus CO from MERRA-2, at constant pressure levels of 550 and 650hPa (approximately 5.1 and 3.7km; just above and below the AEJ-S maximum). The bottom panels of Figure 12 show the 15 diurnal cycles of each day normalized to scale between a unitless 0 and 1, and then averaged over all days in September 2016.
While this doesn't provide any information on the magnitude (this is captured in the panels above), it does illustrate the relative timing of the minima and maxima of each variable through the diurnal cycle, as well as providing a qualitative idea of the strength of this diurnal cycle throughout the month (i.e., when the maximum in an average curve approaches 1 as u 650hPa at 15Z, this is an indication that wind speed consistently peaks at that time each day, and in contrast, the flatter curve of CO 650hPa 20 shows that the diurnal cycle either does not vary throughout the day, or peaks at different times on different days; from the above panel for CO, we can see in this case it's the former). Taken together with the upper panels, this visualization allows us to examine the the strength of the diurnal variations compared with multi-day events, how each of these parameters at a given altitude is offset from the others at the same height, and thus the range of airmass conditions which exit the continent in the AEJ-S.

25
In the previous figures, we saw a daily upward propagation in the continental water vapor (Figure 10b) and the similar feature in CO (Figure 11b), likely due to diurnal heating causing daytime boundary layer growth over the land. This convection allows the surface air to mix upward and reach strikingly high altitudes (∼ 5km) during the day, but the vertical motion is influenced by upper-level subsidence at night. In Figure 12, we note that this pattern propagates upward with a delay: while daily maximum humidity at 750hPa (∼ 2.6km) was generally around 9-12Z, the maximum at 650hPa varies between 12-18Z, 30 and at 550hPa it is still later, between 15-21Z. Again we note there is both daily variation and multi-day episodes, which both vary with altitude. Specifically, the diurnal variability in q is strong at both 650 and 550 hPa, whereas for CO, there is a distinctly stronger diurnal cycle at 550 hPa; the reverse is true for u, which has larger daily variation at 650 hPa. The diurnal cycle also varies throughout the month, with a somewhat weaker diurnal cycle in both CO and q when the zonal winds are strongest (e.g., 19-21 September).  Figure 12. Time series of (top to bottom) zonal winds (u, m/s), specific humidity (q, g/kg), CO (ppb), potential temperature (θ, • C), and vertical velocity (ω, Pa/s), at 650hPa (left) and 550hPa (right) for MERRA-2 (colored lines) and ERA5 (black lines) reanalyses. Each parameter has a distinct diurnal cycle except CO at 650hPa. The 650 and 550hPa panels for a given parameter are on the same scale so as to highlight differences in diurnal cycle magnitudes with altitude, though shifted to capture the full range at each level. Shading indicates night (6pm-6am). The horizontal dashed line in the u panel shows the 6m/s AEJ-S wind speed threshhold (Adebiyi and Zuidema, 2016), and the horizontal dashed line in the ω panel shows the 0 Pa/s threshhold which separates rising (−ω) from sinking (+ω) vertical motion. Note that easterly u-winds are given by negative values. The bottom panel shows the composite diurnal cycle for each variable from MERRA-2 (solid) and ERA5 (dashed) overlaid on one another (colors the same as above), normalized to a diurnal minimum of 0 and maximum of 1, and then averaged over all September days.
We note that while the water vapor over the African continent shows a strong diurnal cycle due to solar heating, the fire strength also has a diurnal cycle following the anthropogenic burning patterns (Roberts et al., 2009). While these timings vary based on location, they generally peak in the late afternoon and are almost entirely extinguished by nightfall (Roberts et al., 2009), which is fairly similar to the timing of daily evaporation and convection over the continent. As mentioned earlier, in this region, the fire characteristics themselves are fairly consistent over this period (fuel type, combustion efficiency, and burn condition). While the multi-day CO variation does not closely track with that in q, the timing of the peaks for an individual day is largely consistent with one another at both levels (minima at 09Z, maxima between 15-18Z; Figure 12 bottom row). CO at the lower altitude varies substantially over the course of several days (∼ 100ppbv), the 550hPa CO consistently varies by 50-100ppb within a 24-hour cycle, with the maximum CO between 18Z-00Z, suggesting frequent influence from dry, clean air above. This suggests that the 550hPa level is influenced by upper level subsidence and mixing on a daily basis, whereas the 10 values at 650hPa are mostly affected by transport in the AEJ-S.
Another piece to the puzzle is the dynamics. Daytime vertical motion over the continent is dominated by solar heating and subsequent convection, as is seen in the substantial daytime increase in potential temperature and the upward propagation of both humid and high-CO air. Overnight, convection is reduced and (when the AEJ-S is active) the zonal wind generally increases, advecting this air to the west. During times of weak-to-no AEJ-S (e.g., first week of September 2016), the decreasing 15 q and CO overnight at 550 hPa is accompanied by frequent strong subsidence and increasing θ (due to the subsidence from above in the absence of solar heating), which suggests increased stratification which would inhibit vertical mixing. The vertical velocities in Figure 12 show more frequent subsidence (+ω) at 550hPa versus 650hPa, and ω at both levels has a maximum (downward velocity) in the early morning (06Z) and a minimum (upward motion) in the late afternoon (15-18Z), which is consistent with convection caused by diurnal heating. In contrast, during times of strong jet activity (e.g., 18-22 September), 20 the jet still largely strengthens overnight, q and CO decrease, but potential temperature also decreases. Since CO and q still generally decrease during this time, this may indicate that increased shear mixing is happening when the jet is strong, which decreases the CO and q values by mixing the more humid and smoky continentally-influenced air with dry, clean upper-level air. When AEJ-S conditions are weak, and when the potential temperature is relatively high, large-scale subsidence dominates and stabilizes the atmosphere without much mixing at this interface.

25
This distinction between high-jet and low-jet conditions is corroborated by Figure 13. This figure shows the CO-q correlations from the MERRA-2 reanalysis along one longitude line over each of (right) the continental source region and (left) the oceanic ORACLES sampling region for the surface-influenced altitudes and for the free-troposphere, respectively. For all data throughout the continental boundary layer over land (top right), the relationship is not as coherent as that observed during ORACLES, and at individual altitude levels below 550hPa the linear relationship is nonexistent ( Figure S5); the low-CO, low-q 30 data are almost entirely driven by the higher altitudes (>600hPa). In Figure 13, there is also a frequent condition of (relatively) high-q (∼ 12g/kg) and low-CO (< 300ppb) which does not correspond to any particular altitude level. In other words, this humid air with a wide range of CO values is frequently present at AEJ-S altitudes, rather than being confined closer to the surface ( Figure S5), yet was not observed during ORACLES. At the same time, the linear relationship is seen over the SEA Ocean for these same latitudes (Figure 13, top left); this is puzzling, since based on our previous analysis (e.g., Figure 9), we expect  Figure 7). A vertically-resolved version of this plot is shown in Figure S5. The results are largely similar for WRF-CAM5 and WRF-Chem ( Figure S6).
the eastern continental region to be the direct source for the western oceanic region. When we consider only the conditions of strong easterly transport (Figure 13, bottom), the situation becomes clearer: now, the CO-q relationship over the continent is much closer to the linear relationship observed over the ORACLES region, and over the ocean it is largely similar. Similar patterns are seen in both WRF configurations ( Figure S6), with a stronger high-q, low-CO feature, likely due to differences in biomass burning implementation between each model.

5
It is notable that if we consider the CO-q relationship of Figure 13b only for one jet level (e.g., the jet maximum of 600hPa), there is no obvious linear CO-q relationship at all over land ( Figure S5). Only starting at the 550hPa level does a linear relationship begin to emerge primarily driven by low-q, low-CO conditions. These higher altitudes are at times alternately influenced by both clean, dry upper troposphere air and by humid, smoky surface-influenced air ( Figure 12). According to MERRA-2, these values decrease in altitude (as expected) from 5 to 12 g/kg in q and 200 to 500 ppb in CO at 700hPa, to 0 to 5 g/kg in q and 60 to 300ppb in CO at 500hPa. While the maximum q continues to decrease above 500hPa, dropping to 1 g/kg at 400hPa, even at this high altitude the CO doesn't fall below 60ppb. While this may be due to the emissions schema used rather than physical reasons, this is nonetheless consistent with the minimum CO observed by aircraft during ORACLES, suggesting accurate background CO is used by the models.
It thus seems plausible that the mixing between surface and upper troposphere air is occurring over the continent, resulting

Results from the 2017 and 2018 deployments
As the ORACLES-2016 data represent only about one third of the data collected during ORACLES, we wish to briefly discuss the context of the latter two ORACLES deployments. As discussed in Section 1, the ORACLES-2017 and -2018 deployments shifts geographically through the season, the plume itself, and the prevailing meteorology, would have been different even if the flights had occurred from the same base in all three years (Redemann et al., 2020). Aside from this, the ORACLES analysis found that there was significant interannnual variability from year to year such that some years saw a peak in BB in September 25 and some saw the peak in August. A more detailed discussion of the broader meteorological and aerosol contexts may be found in Redemann et al. (2020). observations, particularly in October 2018, indicating boundary layer influence extends to a higher altitude than in 2016. The differences between the three deployments is likely due to the anticyclonic atmospheric circulation at AEJ-S latitudes towards the south. In other words, seasonal variation aside, the 2016 deployment simply sampled more airmasses which were influenced primarily by the BB plume, rather than other more northerly origins of the latter two years. August 2017 more frequently saw higher-CO airmasses with relatively lower water vapor compared with the other two deployments. August climatologically 5 sees more northern convection (compared to that in September and October, when the convection migrates south with the end of winter) and also has a much weaker AEJ-S; the AEJ-S was especially weak in 2017 (Redemann et al., 2020), which may also be a factor in the weaker correlations during this deployment. Of the three years, the correlation coefficients between the two measurements are highest in 2016.
The weaker correlations and more humid conditions are thus likely caused by a combination of upward mixing of the oceanic

Discussion
Thus far, we have established that (1) there is a robust linear correlation between water vapor and BB plume strength as measured from several distinct aircraft instruments; (2) this elevated water vapor feature appears, with varying fidelity, in both meteorological reanalyses and free-running climate models; (3) there is frequent deep boundary layer daytime convection over the continent which causes humid/smoky air to be lofted to the altitude of the AEJ-S, which transports it westward; and (4) 5 the linear CO-q relationship is seen over the continent, but only concurrent with a strong AEJ-S condition. We now attempt to synthesize these findings to paint a coherent picture of the evolution of this condition between its source on the African continent and its observation with the ORACLES aircraft. Then, we will briefly explore whether the high water vapor content may be due to some characteristic of the biomass burning itself, or due to some other cause. Figure 16 shows the example of an ORACLES aircraft profile (ramp) from 10 September 2016 at approximately 10Z (09:58:50-10:10:33 UTC) centered at 15.6 • S, 5.6 • E (south of the AEJ-S range; Zone 3 in Figure 1). We choose this profile as, first, it showed multiple plume layers of varying strength: a main plume layer starts around 3.5km, strengthens to 4km, and continues above the aircraft range (∼4.2km in this case), with a secondary peak in CO and q around 2.4 km, and a layer of low-CO/low-q 5 between the two (∼2.8 to 3.2 km). Below the second plume layer, there is a gap of much cleaner air around 1.5km, just above the boundary layer. The second reason we choose this profile is that since the ERA5 reanalysis captures these features fairly well at this time and place, including the smaller secondary q below 3km. (We note that MERRA-2 shows this feature as well (purple dashed line), although the main plume layer is too low in altitude compared with the observations).

Trajectories from emission to observation
Next, the map in Figure 16 (top left-center) shows HYSPLIT back trajectories from three locations within this profile: 4km 10 (the maximum plume), 3.1km (the local minimum), and 2.4km (the smaller local maximum). Back trajectories are run for 6 days for both isentropic (constant θ) pathways and using the GDAS "model motion" (kinematic trajectories using vertical winds from the GDAS meteorology). For this case, at the two higher altitudes, these trajectories (while over the SEA Ocean) are remarkably similar to one another in terms of latitude and longitude, which allows us to explore the implications of each configuration. For a given initial altitude, the two trajectories diverge in trajectory altitude, with the kinematic trajectories 15 showing consistent subsidence (when looking forward in time) and the isentropic trajectories being fairly constant in altitude (at least after they depart the continent), but the two trajectories are very similar in terms of horizontal location, at least after exiting the continent (beyond that point, the trajectories become more uncertain due to convection over land).
Finally, the right-hand panel in Figure 16 shows these trajectories overlaid on the ERA5 reanalysis fields of water vapor (blue shading) and potential temperatures (θ, grey contours show isentropes at 3K intervals), following the location of the isentropic 20 trajectories. Here we can clearly see the differences between the two trajectories are most pronounced in the vertical. We note that the isentropic trajectories as given by HYSPLIT (circle-lines) correspond to isentropic contours from ERA5 (grey curves) at all altitudes until the trajectory reaches (or rather, exits) the continent: on 8 September for the 4 km trajectory, and on 7 September for the 3.1 km trajectory. The 2.4 km trajectory is over the ocean during the entire trajectory and thus follows the isentropes this entire period. Once trajectories are determined to be over the continent, they exhibit more variability in terms of 25 altitude, as would be expected due to the strong convection in this region. This also likely indicates the trajectory analysis is less reliable beyond this point, but the trajectories are nonetheless consistent with airmasses originating from a diurnally-varying deep continental boundary layer.
The kinematic (nonisentropic) trajectories, in contrast, are seen to cross many θ curves during this time, but this is not necessarily inconsistent with the ERA5 reanalysis: in terms of the water vapor, these back trajectories calculated using GDAS reanalysis is considered along the remainder of the 2.4km kinematic trajectory (i.e., at the HYSPLIT-indicated latitudes and longitudes), this trajectory too remains within the top of the water vapor plume until 3 September.
Taken together and considering the analytical caveats of each, these three perspectives on one sampling instance suggests that the airmass transport leading up to the aircraft observations may be somewhere in between the results of these two trajectories.
We remember that a too-strong subsidence is an issue in models over this region; Das et al. (2017) showed that vertical velocities 5 in several different models were frequently too large compared with CALIOP satellite observations, especially once airmasses exit the continent. This is consistent with what we see here regarding very strong subsidence in the GDAS vertical motion, and suggests that the isentropic trajectories may be closer to the observed conditions. Yet the fact that the kinematic trajectories continue to follow the humid layer even with this strong subsidence indicates it is possible that these model trajectories are in the famous model category of "wrong, but useful." Or rather, while the air masses sampled during ORACLES largely follow isentropic back trajectory with altitude, z 0 =5km, over ERA5 q reanalysis along trajectory isentropic back trajectories with altitude, z 0 =4km, over ERA5 q reanalysis along trajectory isentropic back trajectories with altitude, z 0 =3km, over ERA5 q reanalysis along trajectory isentropic pathways, there is some influence from clean, dry free tropospheric air especially over the continent. Indeed, this would be consistent with what we see in Figures 12 and 13: the linear relationship between CO and water vapor over the continent is largely driven by higher-altitude air which is only periodically influenced by continental sources; without these influences, the conditions of low-CO, low-q would not be as prevalent in the air which is advected over the SEA.
As a final example, we consider the case shown in Figure 17, for back trajectories initialized at the aircraft profile sampled 5 just before 13Z (12:35:21 to 12:50:14) on 31 August 2016, centered on 15.3 • S, 5.1 • E, in the same general area (Zone 3) as Figure 16. In contrast to the previous figure which was a very layered profile, this profile was fairly uniform in both q and CO with altitude; this is corroborated by ERA5. Here, when we run the HYSPLIT back trajectories using model motion and isentropic motion initialized at three altitudes (3km, 4km, and 5km), we find that the two configurations diverge much more rapidly. Again we find that the model motion trajectories (from GDAS meteorology) show very strong subsidence while the 10 isentropic trajectories actually show the opposite: rising motion going forward in time. Spatially, the two trajectories diverge in latitude/longitude much earlier than did Figure 16, though both methods end up in largely the same location for the 4km and 32 https://doi.org/10.5194/acp-2020-1322 Preprint. Discussion started: 21 January 2021 c Author(s) 2021. CC BY 4.0 License. 5km trajectories. Looking at the ERA5 reanalysis along these trajectories, we find that the isentropic trajectories agree fairly well with the presence of the elevated water vapor plume, and some altitudes with fairly constant θ, which may indicate these trajectories are less reliable, causing the discrepancy. This highlights the limitations of this type of analysis.

Sources of continental plume water vapor
We now briefly discuss the initial source of this continental water vapor. There are several potential explanations for the 5 correlation between water vapor and the SEA BB plume, including direct emission of water vapor as a product of combustion; water vapor co-emission due to fuel properties; enhanced surface evaporation or evapotranspiration from the burning regions; or simple meteorological coincidence between plume air and already-humidified ambient air. As both smoke and water vapor have their source in the continental boundary layer, it may purely be coincidence of this source and further mixing with dry and clean free-tropospheric air, but we briefly explore the other possibilities. 10 To the first point: some amount of water vapor is co-emitted with other gases and aerosols during combustion. Parmar et al. (2008) measured the ratio of enhanced water vapor to carbon dioxide and emissions ((∆H 2 O)/(∆CO+∆CO 2 )) for different vegetation types: for savannah grasses this ratio is ∼1.2-1.6 and for some trees it reaches up to ∼3. For the sake of argument, even for a relatively high ratio of 3 (which should be an overestimate of the amount of water we should expect from burning of savannah grass), this means that a 2 g/kg enhancement in water vapor would be accompanied by an enhancement of ∼ 15 800 − 1000ppm of ∆CO+∆CO 2 .
For all three ORACLES deployments, the vast majority of CO 2 concentrations were measured as between 400 and 460 ppm and there were no measurements above 500 ppm. Based on these ratios and the CO 2 and water vapor concentrations observed during ORACLES, burning biomass could only have increased atmospheric water vapor by a tiny fraction of what was observed. Unless either the estimates of the ratio of water vapor emitted per carbon dioxide and carbon monoxide or 20 of the typical ∆CO+∆CO 2 plume enhancement are too low by orders of magnitude, it is not plausible that the linear CO-q relationship seen in ORACLES-2016 or the general moistness of the smoke plume are due to the co-emission of water vapor during biomass burning. The fact that the elevated water vapor (∼ 2 − 4g/kg) observed during ORACLES is not associated with a significant elevated CO 2 over the same region (on the order of 2000ppmv) suggests that the water vapor at least is not a direct product of combustion.

25
Another possibility is that the moisture of the fuel itself could be evaporated during combustion; however, Potter (2005) suggested that for woody fuels, the fuel moisture would constitute no more than a third of the water vapor emitted by combustion, which would not account for the magnitude of the signal we observe. It is still plausible that some amount of the enhanced atmospheric water vapor near the fire sites could simply be a result of moist fuels releasing water vapor under the higher fire temperatures; alternately, the observed q could result entirely from surface evaporation/evapotranspiration independent of the 30 fire conditions. Clements et al. (2006) also measured higher sensible and latent heat fluxes and increased turbulent mixing associated with the smoke plumes from small grass fires, and concluded that vapor emissions from such fires would have measurable impacts on local atmospheric dynamics, which may also be in play here. However, to these last points: since we find that models consistently reproduce some level of elevated q without including either a source of water vapor co-emitted from biomass burning, or an enhanced evaporation due to the higher surface temperatures in fire conditions, this suggests that these factors are not primary.
Thus, it seems likely that we can rule out direct co-emission of water vapor as the primary cause of the humid plume, and a meteorological coincidence seems to be the most likely explanation behind the observed correlations.

5
In the aircraft observations collected during the ORACLES field campaign over the southeast Atlantic Ocean, we find a robust correlation between plume strength, as indicated by both inlet-based CO concentration and column AOD, and water vapor concentration. The correlations are highly robust and linear in the September 2016 data and somewhat weaker in the more equatorial observations from August 2017 and October 2018. This could be due to a variety of factors, including the difference in season, deployment location, and sampling patterns over the SEA (e.g., routine diagonal versus routine north-south leg). 10 The ERA5 reanalysis is particularly accurate in placing its high humidity to be coincident with the higher humidity measured by ORACLES flights. All the other reanalyses/models showed a similar pattern, although these models show water vapor content which is more weakly correlated with q from the aircraft observations. For the products which report CO, the COwater vapor relationship shows the opposite pattern: the product which best corresponds to observed q (WRF-Chem) shows the least consistent correlation between CO and q. In contrast, WRF-CAM5 and MERRA-2 both show somewhat better correlation 15 between CO and q, but poorer correlation between modeled and observed q. This suggests that the CO-q relationship overall is better represented in a free-running model (versus one which is frequently reinitialized) likely due to the different effect of this reinitialization on water vapor versus chemistry. However, such a free-running model results in a greater mismatch in the location of a given airmass compared with the observations (in latitude/longitude and in altitude). On the regional scale, the ERA5 reanalysis shows humid air reaching high altitudes (700-500hPa; 3-6km) over the continent, 20 albeit with a lag time from the surface. This is corroborated by other products. The analysis from MERRA-2 also indicates that the CO and q are in phase with one another at the plume level, despite day to day variability in the actual magnitudes of each. Large-scale analysis thus suggests the air masses sampled over the ocean in ORACLES left the continent with the same relationship between water vapor and carbon monoxide as is observed by aircraft. This linear relationship develops over the continent due to diurnal upward mixing within the deep continental boundary layer (max height ∼5-6km) which produces 25 fairly consistent q and CO vertical gradients (decreasing with altitude) which vary in time. Due to a combination of conditions including differential advection at different levels, daytime convection, nighttime subsidence, and resulting mixing between the smoky, moist continental boundary layer and the dry and fairly clean upper-troposphere air above (∼ 6km), the verticallyaligned gradients effectively get stretched horizontally and into layer-like structures over the ocean. For conditions of strong zonal wind, the smoky, humid air is advected over the SEA following largely isentropic trajectories, where it persists, circulates, 30 and in this case was sampled by ORACLES.
Water vapor, particularly when co-located with absorbing aerosols, will have significant impacts on both atmospheric radiative transfer (shortwave heating and longwave cooling) and cloud macrophysics and dynamics. An analysis which builds upon our results here -and other components of the ORACLES dataset-to quantify the radiative impacts of this water vapor on the atmosphere over the broader SEA may thus help to clarify or corroborate previous studies of these effects. Future work will examine the year-to-year variation in this relationship, and the contributions of the BB plume and the humid layer to atmospheric radiative heating and aerosol-cloud interactions within this stratocumulus deck.
Code and data availability. The data used in this paper are publicly available at http://dx.doi.org/10.5067/Suborbital/ORACLES/P3/2016_