Behavior and Transport of Organic Acids in the Troposphere using Observational Data and Models

Formic acid (HFo) and acetic acid (HAc) have both natural and anthropogenic sources and a role in the atmospheric processing of carbon. These organic acids also have an increasing importance in setting the acidity of precipitation as nitrate and sulfate concentrations have decreased. This dissertation examines HFo and HAc tropospheric formation and transport in the continental United States using observations and models. Observational data from two field campaigns were collected with the peroxide chemical ionization mass spectrometer (PCIMS) using iodide clusters for both HFo and HAc recorded at mass-to-charge ratios of 173 and 187. The first campaign, the Deep Convective Clouds and Chemistry Experiment (DC3), was in May and June 2012 and observations extended from the surface to 13 km over the central and eastern United States. The second campaign, the Front Range Air Pollution and Photochemistry Experiment (FRAPPÉ), was in July and August 2014 with measurements from the surface to 7 km over the Colorado Front Range. Post-mission calibration work determined glycolaldehyde (GA) is a significant isobaric interference to HAc with the HAc:GA sensitivity ranging from 1:1 to 1:10. PCIMS HAc data from both campaigns are reported as the acetic acid equivalent sum (AAES). Based on DC3 model work and estimates of secondary production during FRAPPÉ the instrumental sensitivity was closer to a 1:1. Manuscripts 1 and 2 focus on the DC3 May 21 st airmass storm case study at the Alabama/Tennessee border. During this flight a 700 ppt HFo plume at 8 km was observed, approximately 300 ppt in excess of boundary layer air. Different potential reasons for this increase including aqueous production and a pH dependent scavenging were evaluated with the Weather Research and Forecasting model version 3.7 coupled with chemistry (WRF-Chem). Manuscript 1 evaluated the WRF-Chem meteorological reproduction of the airmass storm and the applicability of the Model for Ozone And Related chemical Tracers version 4 and Model for Simulating Aerosol Interactions and Chemistry (MOZART-MOSAIC) compatible microphysics schemes, Morrison and Purdue Lin, in conjunction with a lightning data assimilation (LDA) method. The Morrison microphysics scheme with an LDA temperature range of 261 – 291 K best represented the case study storm. Manuscript 2 showed that there was no difference in WRF-Chem scavenging between a convective complex and isolated convection. It is possible to have cloud top HFo greater than cloud base in a more acidic cloud, pH of 3.5, with multiple HFo aqueous sources, and assuming there is aqueous chemistry up to -40 o C. Manuscript 3 investigated HFo and AAES distributions on the Colorado Front Range using three geographic and four chemical classifications. HFo was highest near predominately biogenic sources with the Denver Metropolitan area as the second highest region. AAES was higher than HFo throughout the campaign with the highest AAES in the Denver Metropolitan area and during the Greeley missed approaches. This dissertation highlights that precipitation chemistry influences organic acids in the upper troposphere. Additionally, HFo and HAc gas phase production are controlled by different emission sources which could provide insight into the atmospheric processing of carbon.


LIST OF TABLES
Formic acid (HFo) and acetic acid (HAc) have both natural and anthropogenic sources and play a significant role in atmospheric chemical processesin particular volatile organic compound (VOC) and oxygenated volatile organic compound (OVOC) processing in the troposphere and precipitation chemistry. Secondary production is a significant source for both acids especially from biogenic precursors, biomass burning, secondary organic aerosols, and photochemical production from VOCs and OVOCs . Both organic acids have been studied for decades however a great deal of uncertainty remains concerning the extent and pathways of their secondary production. Chemical transport model simulations presented by multiple authors highlight the large discrepancies between model and measurements as a result of these unknown pathways Stavrakou et al., 2012;. For example,  reported model HFo results 13 -40 times lower than measurements.  highlighted that increasing secondary sources in order to close the HFo budget, such as from isoprene ozonolysis, requires significant alteration of current product yields to keep the carbon balance. Closing the organic acid budgets will improve our overall understanding of VOC chemistry.
Part of this discrepancy could be from underrepresentation of HFo from the Criegee biradical. The Criegee reaction series is important as it is a major degradation pathway for biogenic (e.g. isoprene and pinenes) and anthropogenic (e.g. ethene and propene) alkenes. The Criegee biradical reacts with H 2 O, NO, SO 2 , and CO leading to a variety of different products which have not all been identified (Neeb et al., 1997).
Organic acids are formed from the reaction of the stabilized biradical with H 2 O: 2 = 2 + 3 → + [ 2 ] * (1) Reaction 3 produces hydroxymethyl hydroperoxide (HOCH 2 OOH) which decomposes under atmospheric conditions to HFo (Neeb et al., 1997). In laboratory experiments the presence of water increased HFo production from the Criegee biradical up to 10 times the dry HFo production (Neeb et al., 1997). HFo formation from the Criegee biradical could explain a great deal of the current discrepancy between models and measurements.
HFo and HAc are fairly soluble species. HFo Henry's Law constant is 8900 M/atm and HAc is 4100 M/atm at 298 K (Johnson et al., 1996); therefore, wet deposition is a dominant sink for both acids especially near the surface. HFo and HAc contribute to the free acidity (portion of total acidity that exists in the form of an acid) of rainwater all over the world. Acid rain (pH < 5.0) is generally considered to be influenced by SO 2 and NO x from anthropogenic emissions though there are several chemicals influencing the pH of rain and cloud water. Emission controls on NO x and SO 2 have led to a reduction in sulfate and nitrate and a consequent increase in precipitation pH. This change in pH is expected to be reflected in aerosol composition and will increase the proportion of the weaker organic acids in these waters. Over thirty years ago HFo and HAc comprised 64% of the volume weighed free acidity at a remote site in Australia (Keene et al., 1983) and 16% in North Carolina (Keene & Galloway, 1984). It is very likely their relative contribution has increased as SO 2 and NO x emissions decreased.
Wet and dry deposition are the largest HFo and HAc sinks leading to a lifetime of a few days for both acids. If they reach the upper troposphere the lifetimes increase to 20+ days because reaction with HO is the dominant gas phase sink. In general, one efficient pathway to move chemical species to the upper troposphere is convection.
HFo and HAc are assumed to be mostly scavenged in convective systems based on their solubilities.  presented model results indicating that the amount of HFo in the outflow depends on the storm type (affecting aqueous phase HFo production) as well as cloud and rainwater pH.  determined that it may be possible to use HFo to detect cloud-processed air though it is highly dependent on cloud conditions and the initial concentration of HFo. Traditionally, HFo and HAc have not been the focus of modeled convective storm chemistry compared to peroxides and formaldehyde. In addition, the majority of available measurements for HFo and HAc in the United States did not sample vertical profiles to the upper troposphere. For example, Jones et al. (2014), Le Breton et al. (2012), , Reiner et al. (1999), and Talbot et al. (1996) have reported vertical profiles for HFo and/or HAc though only Reiner et al. (1999) and Talbot et al. (1996) sampled above 7 km.
This dissertation explores the formation, transportation, and removal of HFo and HAc in the troposphere. Studying organic acids will help us understand precipitation chemistry and atmospheric carbon processing. This dissertation addresses the following questions: 1. Does organic acid scavenging extent differ between a convective multicell complex and an isolated convective cell?
2. Can HFo serve as a tracer of cloud processed air?
3. What HFo potential sources are we not accounting for in models? What does this tell us about the differences in production pathways between HFo and HAc?
4. How do HFo and HAc distributions vary based on natural and anthropogenic sources?
This dissertation presents work from two field campaigns, box models, and the Weather Research and Forecasting with coupled Chemistry (WRF-Chem) regional chemical transport model. A general caveat to the organic acid measurements presented here is the potential cofounding measurement of HAc with glycolaldehyde (GA) with our chemical ionization mass spectrometer. Manuscripts 2 and 3 address this concern in different ways; however, until a quantifiable standard can be prepared DC3 sampled summertime, mid-latitude deep convection in the United States in order to understand how deep convection impacts upper tropospheric composition and chemistry. This was accomplished by sampling active convection inflow and outflow and the upper troposphere 12-48 hours after convection. The May 21 st case study was chosen because there was higher than expected HFo by a few hundred parts per trillion (ppt) above background levels in a region dominated by convective outflow. This HFo increase suggests either transport from the boundary layer or formation within the storm and subsequent release in the outflow. A major goal of this dissertation was to understand why there was elevated HFo detected at high altitude and what this could mean for the production pathways for both acids. WRF-Chem was used in conjunction with observations to test different hypotheses explaining the observations. However, first verification of the reproducibility of the case study storm in WRF-Chem was needed which was the focus of Manuscript 1.
WRF-Chem was unable to produce a storm at the Alabama/Tennessee border without the use of a lightning data assimilation (LDA) method. Manuscript 1 discusses different combinations of cloud microphysics schemes and temperature ranges for the LDA method. This LDA method adjusts the water vapor over a set temperature range to help locate and promote convection by augmenting buoyancy. The microphysics schemes and LDA temperature ranges were evaluated for the smallest domain (0.6 km) using 5 criteria: 1) maximum-column radar reflectivity, 2) vertical wind, 3) maximum cloud top height, 4) cloud mass flux, and 4) hydrometeors' mass and number concentration. The different simulated storms were compared to observations and a previously simulated WRF-Chem storm. New simulations needed to be performed for this case study as the microphysics scheme from the previous WRF-Chem simulations was not compatible with the chemical mechanism desired to produce organic acids. Paulot, F., Wunch, D., Crounse, J. D., Toon, G. C., . Importance of secondary sources in the atmospheric budgets of formic and acetic acids. Atmospheric Chemistry and Physics, 11(5), 1989-2013. https://doi.org/10.5194/acp-11-1989-2011 Pfister, G. G., Reddy, P. J., Barth, M. C., Flocke, F. F., Fried, A., Herndon, S. C., et al. (2017) Yuan, B., Veres, P. R., Warneke, C., Roberts, J. M., . Investigation of secondary formation of formic acid: urban environment vs. oil and gas producing region. Atmospheric Chemistry andPhysics, 15(4), 1975-1993

Background
It has been known for decades that convection has the ability to transport boundary layer (BL) chemicals to the upper troposphere (UT) and will alter the amount of ozone in the UT (e.g., Bertram et al., 2007;Dickerson et al., 1987;Lelieveld and Crutzen, 1994). Despite the work so far we still do not know the full impact of BL chemical precursors on UT ozone formation. The amount of ozone in the UT will impact the radiative budget and the production of radical species that could remove pollutants (e.g., Bertram et al., 2007). Therefore, understanding the deep convective transport and transformation of ozone and its precursors will help improve the UT ozone budget.
Relative to the UT, the BL has slower wind speeds, higher humidity, and warmer temperatures shortening chemical lifetimes (Dickerson et al., 1987 found that 54% of the sampled air between 7.5 and 11.5 km was influenced by convection in the previous 2 days (Bertram et al., 2007).
Traditionally, chemical studies of deep convection have used insoluble tracers such as carbon monoxide and ozone (e.g. Bertram et al., 2007;Dickerson et al., 1987;. Both carbon monoxide and ozone have lifetimes longer than that of a thunderstorm making them ideal tracers to study transport through storms and downwind. As studies of convection's chemical impact increased, soluble chemical species have become more widely used. Three common chemical tracers are formaldehyde (CH 2 O), hydrogen peroxide (H 2 O 2 ), and methyl hydroperoxide (CH 3 OOH). These three chemicals are important reservoirs for odd-hydrogen radicals thus impacting ozone production (Lee et al., 2000). The importance of hydrogen peroxide and methyl hydroperoxide as sources of the odd-hydrogen radicals increases in the upper troposphere because there is low water vapor (Lee et al., 2000).
Furthermore, the ratio of hydrogen peroxide to methyl hydroperoxide serves as a tracer of convective outflow (Prather & Jacob, 1997). There is a greater scavenging of hydrogen peroxide within a storm cloud because the Henry's Law constant of hydrogen peroxide (8.33 x10 4 M atm -1 at 298 K, O' Sullivan et al., 1996) is greater than methyl hydroperoxide (3.11 x 10 2 M atm -1 at 298 K, O 'Sullivan et al., 1996).
Soluble species studies are aimed at understanding how scavenging impacts transport of soluble ozone precursors to the UT. Barth et al. (2001) modeled nonreactive, soluble species to study the impact of scavenging and how liquid versus solid or mixed phase impacted solubility. As liquid freezes it is possible that some soluble species, such as hydrogen peroxide, may be scavenged into ice during the conversion of liquid to ice, snow, or hail (Barth et al., 2001). The other most likely possibility is that the soluble species will degas during the freezing process (Barth et al., 2001). If modeled scavenged species were degassed during conversion of liquid to solid then, regardless of solubility, it was transported to the UT (Barth et al., 2001).
Modeled species with H >10 5 M atm -1 were retained in snow and hail and had a scavenging efficiency of at least 50% (Barth et al., 2001).
Another important component to consider is the chemical transformation of BL chemicals within the storm cloud. Soluble species scavenged by cloud water can undergo aqueous chemistry transforming them into a different species. This eliminates the possibility of degassing the original chemical in the storm outflow. For example, in the aqueous phase formaldehyde can be oxidized to formic acid (HCOOH).  hypothesized that formic acid could be a tracer for cloud processed air as a result of the aqueous formation from formaldehyde. Formic acid is an important contributor in establishing the pH of cloud and precipitation water. Up to 64% of the free acidity of rainwater in remote regions is controlled by formic and acetic acid . If formic acid is lofted to the UT, given the lifetime of 20+ days with respect to HO , it could impact ozone chemistry and other photochemical processes far removed from the BL origin.   and will serve as a test case for the possibility of formic acid as a tracer of cloud processed air in a future study (Manuscript 2).
Manuscript 2 will utilize observational data and numerical experiments together to investigate this hypothesis. In order to accomplish this chemical investigation the Weather Research and Forecasting (WRF) model will be coupled with a more extensive chemical mechanism, including aqueous chemistry, than previously used for this DC3 case . Previous studies  used the WRF Single-Moment 6-Class (WSM6, Hong and Lim 2006) microphysics scheme coupled to the Model for Ozone and Related chemical Tracers version 4 (MOZART-4) gas phase chemistry scheme  and Goddard Chemistry Aerosol Radiation and Transport (GOCART) aerosol scheme (Chin et al., 2002). In order to simulate formic acid and the impacts of aqueous chemistry a different chemical mechanism was chosen which is only compatible with the Morrison and Lin cloud microphysics schemes.
MOZART-4 is a detailed chemical mechanism for tropospheric inorganic chemistry and organic chemistry up to three carbons.  Morrison and Lin to observations and Li et al. (2017) and discusses the influence that the LDA temperature range has on simulated storms. The majority of the work presented is without chemistry included. The inclusion of chemistry into the model altered the storm compared to the meteorology only simulations. Section 5 briefly explores how the storm changed after chemistry was added. Besides the microphysics, the WRF set-up replicated the previous work of  and  that tested different parameterizations in order to accurately represent the observed storm. The 15 km domain used the Grell 3D cumulous parameterization (Grell, 1993;Grell & Dévényi, 2002). The Yonsei University (YSU) scheme was used to simulate planetary boundary layer (PBL) mixing in all three domains . The RRTMG shortwave and longwave radiation schemes (Iacono et al., 2008) and the Noah land-surface parameterization (Koren et al., 1999) were used for all three domains.

Lightning Data Assimilation Method
The lightning data assimilation (LDA) technique used was developed by Fierro et al. (2012;2015) and modified by . This technique nudges the water vapor based on flash counts and graupel. The Buck (1996) (Fierro et al., 2015;Marchand & Fuelberg, 2014). As will be discussed below, the 261 -285 K temperature range was insufficient to reproduce the vertical extent observed. Therefore, the temperature range was increased further to 261 -288 K to root the storm in the boundary layer. This work compares the Li et al.
(2017) and modified temperature range using Morrison.
Lightning data came from the North Alabama Lightning Mapping Array (NALMA) very high frequency (VHF). The LDA adjustment occurred only if the minimum flashes were greater than 5 per 10 min. A dampening option was also used to suppress convection outside the region of interest. Prior to convection initiation, from 17:00-19:00 UTC, the relative humidity was reduced to 75% throughout the domain to prevent the formation of spurious storms . The maximum radar reflectivity profiles are the maximum merged ARMOR-KHTX horizontal radar reflectivity. The radar composites have a 1 km horizontal resolution and a vertical grid spacing of 1 km. The three-dimensional radar composites and vertical velocity were prepared using the methods outlined in .

Observational Description of the Storm
The May 21st storm has been described extensively in , , , Fried et al. (2016), and  and will only be discussed briefly here. There were two rounds of deep moist convection on May 21st in the DC3 Alabama domain . The storm of interest started in Tennessee as two separate updrafts at 19:45 UTC with a third storm on the Alabama/Tennessee border. At 20:04 UTC ( Fig. 1.2a) the northern storm had its first lightning flashes though no lightning was detected in the southern storm until the two merged . At 20:15 UTC ( Fig. 1.2b) the two storms merged into one, commonly called B2, resulting in an increase in lightning and vertical motion . By 20:20 UTC ( Fig. 1.3a) the maximum velocity was 12 m/s and was in the -10 o C layer. The peak flash rate was at 20:23 UTC and the storm started to collapse after 20:30 UTC ( Fig. 1.2c) .
The vertical motion formed a wedge that sloped from west to east likely due to the surface outflow moving faster than the main convection line and a strong cold pool . By 20:50 UTC ( Fig. 1.2d) B2 merged with the southern storm to form a ring-shaped convective complex.

Results and Discussion
This work confirmed that both Morrison and Lin required the LDA water vapor adjustment to produce a storm in the region ( Fig. 1.3). Even with the lightning data, the simulated storms generated about 30 minutes later than the observed storm.

Horizontal Radar Reflectivity
The column-maximum radar reflectivity profiles for the three WRF simulations are in Figs 1.4-1.6. For M285, the two storms that merged to form B2 first appeared at 20:30 UTC ( Fig. 1.4a) and B2 formed by 20:50 ( Fig. 1.4b). As the storm moved south, it elongated to the north forming a "tail" that remained as the storm decayed. There was also the storm on the Alabama/Tennessee border as seen in observations. By 21:30 UTC there was a cluster of storms at the border but not the observed ring-shaped convective complex.
L285 column-maximum radar reflectivity ( Fig. 1.5) contained scattered background reflectivity (less than 10 dBZ) in the surrounding area that was not present with Morrison. Unlike M285, the two initial storms appeared at 20:20 UTC ( Fig. 1.5a) though much weaker than the initial appearance in M285 (20:30 UTC). By 20:30 UTC the storms' reflectivities were similar to M285 and the two storms fully merged by 20:50 UTC ( Fig. 1.5b). As B2 decayed and moved towards the southeast it formed a ring shape similar to the observed storm ( Fig. 1.5e) though it did not merge with the surrounding storms like observed. L285 formed a similar elongated shape as M285 to the north as the storm decayed.
M288 had the highest reflectivity of the three simulations ( Fig. 1.6). Like M285, the two storms appeared at 20:30 UTC ( Fig. 1.6a) though M288 was larger. By 20:40 UTC the two merged but there were clearly two storm cores with reflectivity greater than 55 dBZ ( Fig. 1.3c). The two cores fully merged by 20:50 UTC ( Fig. 1.6b).
M288 produced a larger and stronger storm than either M285 or L285 though the characteristic northern "tail" was present. The border storm was also larger than either of the other simulations. Unlike M285 and L285, M288 did not form a ring as the storm decayed. Instead, smaller storms, with reflectivity greater than 50 dBZ, formed on the edges of B2 ( Fig. 1.6c) resulting in a cluster of small storms and the original B2 was lost.
Overall, L285 was the only simulated storm to form a ring with surrounding storms as it decayed. This is likely helped by the presence of spurious smaller storms in the area. M285 stayed as isolated convection during its decay. M288 created smaller storms near B2 as it decayed but the convection was scattered and never formed a ring.

Vertical Radar and Wind Cross-Sections
ARMOR reflectivity and vertical wind (w) for 20:20 UTC in Fig. 1.7 represents a snapshot during the peak time of B2 . The panels are oriented the same as WRF but the cross-sections are not the same length. The cross-sections are spaced 1 km apart and the south-north transects covered 30 km while west-east covered 25 km. This was to keep the similar storm analysis box used in . Figure S1.1 shows multiple panels in both directions to show the observed storm structure. The maximum reflectivity was less than 60 dBZ and w was 12 m/s. The storm topped at 12 km and was about 10 km wide. The maximum reflectivity reached the surface. The wind contours wedged in the southnorth direction. corresponding with the irregular shape of the storm core in Fig. 1.4. M285 was 9 km wide at its maximum reflectivity center ( Fig. 1.9 top panel) and the maximum vertical extent was 9 km. M285 at 20:40 UTC (not shown) had a higher reflectivity in the core (greater than 60 dBZ) but this is before the full formation of B2 as indicated in Fig.  1.4b. M288 clearly produced the strongest storm ( Fig. 1.10

Cloud Top Height, Maximum Vertical Motion, and Storm Core Volume
It is essential to use other parameters, besides radar images, to compare the simulated and observed storm. observations. This encompasses the B2 storm for the time period that includes the two storms that formed B2 and follows B2 through its lifetime. As the storms moved, L285 and M285 remained as isolated storms longer than M288. It is not possible to distinguish the M288 B2 from other storms in the region after 21:10 UTC. For this reason, the M288 time series ends before the other two simulated storms.
The cloud top height ( Fig. 1.11a) Fig. 1.11b). The second increase in M288 w max to 17 m/s at 21:10 UTC could be caused by two small storms that formed to the west of B2. The observed storm has a peak w max of 17 m/s just before the storm's decay . For the majority of observed B2's lifetime the w max was 10 -13 m/s. The 35 dBZ volume ( Fig. 1.11c) is defined as the volume of grid cells of the storm with reflectivity equal to or greater than 35 dBZ. This is used as a proxy for graupel/hail. Overall   The initial convection in L285 began at 20:20 UTC while Morrison initial convection began at 20:30 UTC. Though the L285 simulated cloud appeared first, the three simulated storms' updraft started at the same time (20:10 UTC). This implies that the LDA increase in water vapor resulted in a similar timing in the initial perturbation in the virtual potential temperature leading to the updraft (Fierro et al., 2012). Lin cloud microphysics scheme responded faster to the injection of additional water vapor and formed a cloud. All three storms initiation times were delayed relative to the observed storm by about 30 minutes. This is likely an effect of the assimilation method. Adding water vapor based on observed lightning strikes requires the observed storm to have developed enough to produce lightning. If there is no simulated storm present without adding additional water vapor, as opposed to just enhancing a weakly simulated storm, the simulated storm will be delayed relative to observations. Future users may want to consider changing the timing of the LDA based adjustments.

Cumulus Updraft Mass Flux
Since the goal of Manuscript 2 is to use modeled aqueous chemistry to study chemical fate and transport through this storm it is important to reasonably estimate the mass flux through the storm. The mean cumulus updraft mass flux for storm B2 was calculated for the WRF simulations and the observed storm ( Fig. 1.12 The varying mass flux maximum peaks for the WRF simulations were a function of the changing area meeting the criteria ( Fig. 1.13). ARMOR (at 20:23 UTC) had the greatest storm area meeting the criteria while all three storms had substantially smaller areas ( Fig. 1.13). M288 had the largest area meeting the criteria of the three simulations. The shape of the area meeting the criteria results in the multiple mass fluxes peaks observed for the simulations leading to the multiple peaks in the mass flux for the simulations. For example, the M285 and M288 peak below 2 km is reflected in the area plot.

Water Hydrometeors
A final comparison shows how water is divided between the hydrometeors for the different meteorological set-ups. WRF reports ice, snow, graupel/hail, cloud water, and rain (g/kg for all). The graupel/hail hydrometeor represents graupel for L285 and hail for M285 and M288. The hydrometeor maximum for B2 is shown from 20:00 UTC to 21:00 UTC in Fig. 1.14. Using the maximum value represents the dominant hydrometeor at a given model time and shows trends between the set-ups. Total condensed water (q total ) is the sum of cloud water, rain, ice, snow, and graupel/hail.
Please note that q total shows maximum of the sum and not the sum of the maximums.
As there is no possible quantitative comparison for ARMOR, the maximum hydrometeors are compared to the WSM6 data from  Morrison bulk density of hail is 0.9 g/cm 3 compared to 0.4 g/cm 3 for graupel (Morrison et al., 2005). This analysis does not account for altitude differences; therefore, the high M288 hail maximum is likely much closer to the ground than the graupel maximum in WSM6.
Overall, M288 had the highest maximums in each category of the Morrison and Lin simulations. This is expected as M288 added water vapor over the largest temperature range, corresponding to the greatest number of grid cells. The different microphysics schemes also clearly impact how water is divided among the hydrometeors. M285 and L285 had nearly identical q total maximums; however, they did not partition the water among the categories in a similar way. M285 started with a higher cloud maximum though by 20:40 UTC they were comparable. L285 had a higher graupel maximum than M285 for the duration of B2's lifetime. The heavier hail in M285 would fall and rain out potentially explaining M285's higher rain maximum earlier in the storm. This discussion highlights how different microphysics schemes partition water vapor among hydrometeors. The similarities between L285 and WSM6 can be attributed to both being single moment schemes.

Meteorological Versus Chemical Simulations
Of the three simulations discussed above, M288 best captured the horizontal and vertical structure of the storm as well as the total mass transport through the storm. Since the ultimate objective was to use WRF-Chem with MOZART-MOSAIC it was important to confirm that adding the chemistry module did not significantly impact the storm structure. The addition of chemistry adds an aerosol scheme which allows for a more complex interaction between radiation and aerosols which could impact the storm structure. The WRF-Chem run with M288 produced a simulated storm that was weaker than M288 and in fact more reflected M285 in horizontal size and storm strength. It is possible that aerosols impacted the radiation strongly enough to dampen the storm. However, after running with and without aerosol direct effects there was minimal improvement in storm strength.
Due to the similarity to M285 we increased the LDA temperature adjustment range another 3 K (261 K to 291 K) into the boundary layer to see if that strengthened the storm. This improved the storm structure, better reflecting the M288 meteorological simulation. This suggests that there were potential dampening effects in the chemistry run not present in the meteorological run involving interactions between aerosols, chemistry, and radiation. Aerosol-radiation interactions impact the sensible and latent heat fluxes which impacts the thermodynamic structure of the atmosphere. Furthermore, with an aerosol scheme there are aerosol-cloud interactions that will impact cloud condensation nuclei concentrations.
When chemistry was included, other storms appeared in the 0.6 km domain that were not present in the meteorological simulation including a squall line near the western boundary. This suggests that the baseline conditions were different for the chemistry and meteorological simulations. In general, the water vapor adjustment is not applied if the relative humidity is greater than 93%. It is assumed that above this relative humidity a storm would form "naturally" and not need the additional water vapor. The chemistry simulation had a vertical layer (around 2.5 km) with relative humidity greater than 90% adjacent to B2 that was not present in the meteorological run. It is possible the adjacent peak in relative humidity altered the B2 storm region enough that a smaller amount of water was being added with the chemistry run because, in theory, the relative humidity was high enough to produce the storm.
However, because B2 was not formed without the LDA addition the amount of water added with the M288 chemistry run was insufficient to root the storm in the boundary layer thus producing a similar storm to M285. The M288 chemistry simulation had half the total condensed water (q total ) maximum as M288 and M291 with chemistry.
For LDA to be effective and reproduce B2 it needed a larger addition of water vapor to increase the buoyancy even more than the meteorological run. This is likely due to aerosol effects. With only meteorology, particles had a prescribed number and size distribution. When chemistry is added, the emissions and chemistry impact the number, composition, and size distribution of the aerosols. WRF-Chem passes the cloud droplet number source (resulting from aerosol activation) and cloud droplet number between the chemistry and physics modules (Chapman et al., 2009). Not using aerosol direct effects slightly improved the simulation though not substantially. It is possible that indirect effects or other chemical and meteorological interactions impacted the overall domain making it more challenging for the LDA adjustment to root the storm in the BL.
Furthermore, there were unforeseen interactions with the addition of observed formic acid (HFo). WRF-Chem HFo mixing ratio was two orders of magnitude lower than observed. Aircraft HFo data was implemented into WRF-Chem to evaluate HFo scavenging (Manuscript 2) following the approach of . Peroxide chemical ionization mass spectrometer HFo median altitude profiles in the boundary layer and free troposphere were used as the HFo altitude profile implemented into WRF-Chem. The boundary layer was defined as altitude < 2 km and q total < 0.01 g/kg from 21:00 -22:00 UTC which was during a spiral before returning to base. The free troposphere was defined as altitude > 3km, q total <0.01 g/kg, and O 3 /CO<1.25 with aircraft data from 17:00-19:00 UTC which was before the storm sampling. The lowest altitude median HFo value was applied to all layers below available GV measurements down to the surface. The chemistry of these simulations is discussed in detail in Manuscript 2. This change was implemented into the WRF boundary files and the WRF restart file at 19:00 UTC.
The WRF-Chem simulation with LDA adjustment from 261 -291 K and the aircraft HFo mixing ratio is referred to as M291-Chem. Since this version will be used in Manuscript 2 it is important to understand how this version of the storm compares to observations. The structure and timing of M291-Chem differed from the above discussed simulated storms. The critical understanding for M291-Chem is how this storm impacts the chemical transport through the storm in order to evaluate different hypotheses explaining the elevated HFo near storm outflows. A brief discussion is presented with 4 of the criteria discussed above: 1) maximum-column radar reflectivity, 2) vertical wind, 3) maximum cloud top height, and 4) cloud mass flux.
This comparison focuses on ARMOR and M291-Chem during the outflow period  Despite the differences in storm structure compared to ARMOR and the simulations discussed above the mass flux was similar to observed. The M291-Chem mass flux had a similar maximum flux to the later portions of ARMOR B2 after 20:30 UTC in Figure 1.12. The transition between convergence and divergence was at a similar altitude to ARMOR at 20:30 UTC. This is before the observed outflow period.
The outflow period ARMOR cumulus updraft mass flux is compared to M291-Chem in Fig. 1.16. This comparison is important when considering the transport through a storm and scavenging efficiency in Manuscript 2. The same outflow times are shown as in Figs. S1.5 and S1.6. Despite the higher cloud top height in Fig. 1.15 there is no mass flux above 6 km for ARMOR. Overall, the ARMOR mass flux decreased with altitude and is half of the M291-Chem flux. The sampled outflow period was likely not directly from the surface. There is cumulus mass flux up to almost 10 km for M291-Chem. The difference between the cloud top height and mass flux tops for ARMOR relates to vertical velocity. There must be part of ARMOR with reflectivity greater than 20 dBZ, a factor for mass flux and cloud top height, but vertical velocity less than 1 m/s. Therefore, the cumulus mass flux is lower than the cloud top height.
During a review on the status of atmospheric modeling in Europe, Baklanov et al. (2014) discussed the need to improve our conceptual understanding of the aerosolradiation-cloud-chemistry interactions. As more is understood about these interactions the modeled parameterizations will also need to be improved. Baklanov et al. (2014) cautioned that as more chemistry and meteorological parameters are added we need to be aware of how these will affect each other. Feedbacks between chemistry, radiation, and aerosols for WRF-Chem with MOZART-MOSAIC appear to have altered the meteorology of the domain. A more extensive exploration of this interaction for the May 21 st case is beyond the scope of this work. However, as discussed here, when data assimilation is required to produce a storm it is possible that even more energy is needed to overcome the natural interactions between chemistry, aerosols, and meteorology.

Summary
WRF simulations were performed with Morrison and Lin microphysics schemes to replicate the DC3 May 21 st airmass storm at the Alabama-Tennessee border. In order to produce a storm in the region with WRF a LDA method was needed that added water vapor in a defined temperature range. Originally, the Morrison and Lin simulations were performed over the temperature range 261 -285 K to replicate WSM6 simulations in  and . This resulted in a storm with a similar maximum vertical velocity and radar reflectivity to the observed storm. However, the cloud top (10 km) was lower than observed (12 km) and there was a smaller cumulus mass flux. The LDA temperature range was extended from 285 K to 288 K for Morrison resulting in a cloud top height within 1 km of observations. However, the maximum vertical wind was overestimated by 18 m/s.
Clearly extending the temperature region to 288 K made the simulated storm too vigorous. Despite the differences to the observed storm, it produced a similar storm to the WSM6 simulations based on the 35 dBZ volume, cloud top height, and maximum vertical velocity.
As the goal of this study was to simulate the chemical transport through this storm, the amount of mass being moved through the storm was important. M285 and L285 had similar maximum mass fluxes. M288, because of its higher cloud top height, was the only simulated storm to have mass flux reach 12 km, like observed. None of the WRF simulations produced a perfect replica of the observed May 21st storm. Of the options available M288 did the best at capturing the horizontal and vertical structure of the storm as well as the total mass transport through the storm. M288 also best reflected previously published WRF work using WSM6.
A final adjustment was made with the addition of chemistry. M288 used with the MOZART-MOSAIC chemistry set-up resulted in a storm that was weaker than the meteorological M288 run. The chemical M288 simulation produced a similar storm to M285; therefore, the LDA temperature region was increased another 3 K (261 -291 K) in order to provide more energy to the system. Adjusting the temperature range          were redistributed and transformed to the UT . This redistribution is affected by scavenging, aqueous production, lightning NO x , entrainment/detrainment, and ice processes. A simple schematic of a mature single cell thunderstorm highlighting some of these processes is presented in Figure 2.1. In the layer between 0 and -40 o C is a mixed-phase ice and water region within which both cloud water and ice exist. Above -40 o C there are only ice particles. Heavy precipitation is in the downdraft.
Scavenging efficiency (SE) is commonly used to quantify the removal of a soluble BL constituent by precipitation during its transport through a storm. SE quantifies the fraction of soluble species removed by precipitation in a storm by measuring soluble species in the storm inflow and outflow. SE determination is impacted by processes, other than removal by precipitation, which alter a species composition such as aqueous and ice chemistry, entrainment, and detrainment (purple arrows in Figure 2.1). Entrainment and detrainment are typically accounted for using an insoluble tracer, e.g. n-butane, that is not altered by chemistry or scavenging over a storm's lifetime, by normalizing the soluble species to the insoluble tracer thus accounting for entrainment and detrainment (e.g., Fried et al., 2016).
Transport and SE of the soluble trace gas species hydrogen peroxide (H 2 O 2 ), methyl hydroperoxide (CH 3 OOH), and formaldehyde (CH 2 O) were evaluated for different DC3 storms Fried et al., 2016).
One case study was an airmass storm sampled at the Alabama/Tennessee border on May 21, 2012 Fried et al., 2016). Fried et al. (2016) found a significantly higher formaldehyde SE (81+5%) for the airmass storm relative to larger DC3 storms (48-67%). Fried et al. (2016) remarked the potential disconnect between inflow and outflow measurements based on butane and pentane ratios. It was also suggested the high formaldehyde SE could be due to aqueous chemistry.
Formaldehyde is considered the primary aqueous chemistry source for formic acid (HFo); therefore, HFo production would be a formaldehyde sink. While a formaldehyde aqueous chemistry sink was mentioned, Fried et al. (2016) simulations did not include this formaldehyde loss and HFo source.
Prior modeling work suggested that clouds could be a significant source of HFo (Jacob, 1986) via an aqueous CH 2 O -HO reaction although observational evidence is mixed. Laj et al. (1997) reported a pH dependent aqueous production of HFo and release to the gas phase. Keene et al. (1995) found that clouds are a net sink for HFo and fail to provide significant production. HFo and the other dominate carboxylic acid, acetic acid (HAc), are considered to be significantly scavenged in clouds via wet deposition  but this scavenging is pH dependent.
When the pH is lower than 5, aqueous HFo is volatilized to the gas phase (Jacob, 1986) and cloud production could provide a UT HFo source. For completeness, the primary aqueous production of HAc is from pyruvic acid (Carlton et al., 2006) and aqueous HAc production is not considered to be a significant source of HAc (Jacob & Wofsy, 1988) to the atmosphere.
While there are uncertainties regarding the possibility of aqueous phase production, both organic acids have natural and anthropogenic gas phase sources including biogenic emissions, motor vehicle exhaust, agricultural emissions, and biomass burning . Secondary gas phase production from biogenic precursors is considered to be a significant source for both organic acids though full reaction mechanisms and yields remain uncertain . In addition, both acids help set the natural acidity of rainwater. HFo and HAc comprise 16% and 64% of the volume weighted free acidity of precipitation in urban (Keene & Galloway, 1984) and remote (Keene et al., 1983) regions, respectively. It is likely that with decreasing SO x and NO x emissions the acid composition of precipitation has changed with organic acids playing a larger role even in urban regions.
Wet and dry deposition are the dominant organic acid sinks . Because of this, traditionally HFo and HAc are thought to be removed through convective systems. If HFo and HAc were transported to the UT, their lifetime would increase from a few days to 20+ days as the main gas phase sink is reaction with HO which is quite slow . To our knowledge, there have been no convectively influenced UT measurements of either organic acid in the United States on storm timescales. Talbot et al. (1990) reported a 200+ ppt increase in HFo measured in convective outflow over the Amazon that was not seen in HAc.
Convective outflow was sampled on two separate occasions but there was only an increase in HFo for one flight (Talbot et al., 1990). This suggests that there were different processes for the two storms which impacted HFo scavenging differently. This paper examines the potential for HFo, HAc, and GA transport and transformation to the UT for the airmass case study using observations and model results. Here we combine observations and model results from the Weather Research and Forecasting Model coupled with chemistry (WRF-Chem) and a photochemical box model (Barth et al., 2003; to investigate potential causes for the HFo peak noted aloft. WRF-Chem simulations provided insight into potential SE changes for organic acids for isolated convection compared to a convective complex. The WRF-Chem simulations are also used to discuss the likely HAc and GA contributions to the AAES measurements and how both are transported through the airmass storm. The box model is used to discuss the influence of pH and HFo aqueous production on HFo gas phase mixing ratios.

Campaign Info
The (KHTX). The NEXRAD horizontal radar profiles presented were prepared using the NOAA Weather and Climate Toolbox and Google Earth with the NEXRAD data.

Aircraft Measurements
Aircraft measurements for chemical, aerosol, and cloud physics parameters were collected by the NASA DC8 and NSF/NCAR GV. A list of the species used in this study, instruments, and detection information are in Tables (Lee et al., 2014;Neuman et al., 2002Neuman et al., , 2010 below 1000 m between 35 and 36 o N and -85 and -87.5 o W were averaged to represent the inflow region.

Simulation Set-up and Model Description
This study used the Weather Research and Forecasting Model version 3.7 coupled with chemistry (WRF-Chem). The model set-up details are presented in Table   2.3 and briefly described here. Simulations were run with one-way nesting for the three domains. There were 40 vertical levels with a model top of 70 hPa. The simulation was started before the analysis time to allow the chemistry and physics to stabilize. The 15 km domain used the Grell 3D cumulous parameterization (Grell, 1993;Grell & Dévényi, 2002) and the 3 km and 0.6 km treated convection explicitly.
The WRF-Chem simulation set-up was the same as  and  with the exception of the lightning data assimilation temperature range, cloud microphysics scheme, and chemical mechanism.
The Morrison double moment scheme was used for cloud microphysics for reasons outlined in Manuscript 1. Lightning data assimilation was used following the WRF-Chem was unable to produce sufficient HFo (Section 4.1). Peroxide chemical ionization mass spectrometer HFo median altitude profiles in the boundary layer (BL) and free troposphere (FT) were used as the HFo altitude profile implemented into WRF-Chem following the same method as . BL was defined as altitude < 2 km and q total < 0.01 g/kg from 21:00 -22:00 UTC which was during a spiral before returning to base. FT was defined as altitude > 3km, q total <0.01 g/kg, and O 3 /CO<1.25, from 17:00-19:00 UTC. The lowest altitude median was applied for all model layers below GV measurements down to the surface. The GV focused on storm outflow therefore not all altitude bins had HFo measurements. For altitude bins without measurements, the median from the bin below was used. This is why the HFo background profiles (Figure 2.3) decrease < 20 ppt from 3-7 km. For altitudes above the aircraft, results from the global chemistry transport model MOZART  were used.

Observational Scavenging Efficiency Calculation
The observed inflow and outflow times are from  and    Fried et al., 2016). This determines SE based on the inflow and outflow soluble species mixing ratios relative to n-butane. n-Butane is transported and not changed chemically during the time period of transport from the BL to the top of the storm. However, entrainment will alter n-butane mixing ratios as air moves from cloud base to top. Therefore its mixing ratio will not be the same in the inflow and outflow regions because of entrainment. The n-butane method calculates SE using the average ratio of the soluble species to n-butane for the inflow and outflow for all observations meeting the above criteria as shown below. It requires a background air profile (obtained from the DC8 for the whole flight), the inflow mixing ratio, and outflow mixing ratio. Through iterations the entrainment rate is found for n-butane, i-butane, n-pentane, and i-pentane. Fried et al. (2016) only used n-butane for the altitude entrainment method for the May 21st storm due to discrepancies between the inflow and outflow mixing ratios for the other species.
This study only determined the observed SE with the n-butane method. The WRF-Chem entrainment rate will be compared to the Fried et al. (2016) entrainment rate in Section 4.4.

WRF-Chem Scavenging Efficiency Calculation
The WRF-Chem SE method used here differs from  and Barth et al. (2016). They calculated SE for the WRF-Chem grid points corresponding to the location and timing of the observed outflow sampled by the GV.
Due to the difference in microphysics scheme and chemical mechanism the location and timing of the simulated storm differed from observed storm and, as a result, the observed outflow region was not near the simulated storm. In addition, as discussed in Manuscript 1 the simulation set-up differences resulted in the primary storm being produced 30 minutes after the observed storm. For these reasons WRF-Chem SE used a chemical characterization based on HNO 3 and NH 3 for a defined storm region. A chemical threshold for HNO 3 (SE>60%) and NH 3 (SE>=10%) was used. SE values were determined for a box encompassing the whole storm and only included model grid points where the summed mass mixing ratio of cloud water (q cloud ) and ice (q ice ) was greater than 0.01 g/kg.
Historically, HNO 3 and SO 2 oxidation (H 2 SO 4 ) control the pH of precipitation with some neutralization by NH 3 (Charlson & Rodhe, 1982;Galloway et al., 1982). SE was calculated for a box encompassing the storm and extending vertically from 8 -11 km for the convective complex and 6 -9 km for the isolated convection.
This expands the SE area from previous literature work that used the grid cells closest to the aircraft sampled outflow location. WRF-Chem SE values for species j were calculated using the mean outflow mixing ratios (q) for a simulation with precipitation scavenging on (q j,scav ) and a simulation with precipitation scavenging off (q j,noscav ).

WRF Passive Tracers
Passive tracers were used to calculate the entrainment ratio through the simulated storm. Tracers were released in 1-km bins from the surface to 20 km . Tracers were set to 1.0 in clear air outside the storm defined by total water mixing ratio (q total ) < 0.01 g/kg . This was done for one time step an hour before the end of the outflow period for a box encompassing the storm. For example, the convective complex tracers were released from 20:40-20:50 UTC for the 21:30-21:50 UTC outflow period. Entrainment was calculated by taking the average percent contribution of each 1 km altitude layer within the outflow box described in Section 2.5. This was done for all points that met the chemical scavenging criteria (HNO 3 SE>60%, NH 3 SE>=10%, and q cloud + q ice > 0.01 g/kg) and at the top of the storm core which is defined as within the 40 dbz radar contour.

Box Model
The impact of aqueous chemistry, temperature, and pH on organic acid distribution were evaluated using a gas-aqueous box model described by Barth et al. (2003; It is likely both models underpredicted HFo.

Meteorological and Chemical Description of the Observed Storm
The Alabama airmass storm meteorology has been described in detail in , , , and Manuscript 1, among others. The meteorological foundation of the Alabama storm simulation was discussed in Manuscript 1 including comparisons to previous simulations  and observations . This was a critical step necessary before the chemical discussion presented here. Thus, only a brief meteorological overview is presented with the necessary context to understand the chemistry.
On May 21, 2012, Research Flight 03 (RF 03), the objective was to sample airmass storms in the Alabama region. The storm of interest, referred to as B2 here and in , was part of the second round of convection that day.   . This spiral will be used later for a scavenging efficiency comparison (Sections 4.5 and 4.6). AAES was highly variable throughout the whole flight and was greater than HFo with one exception. During the southeast-northwest wall pattern, the GV flew past an isolated storm northeast of B2 with an enhancement in HFo (20:00 UTC) but with no similar peak detected in available GV measurements. While this storm was not the primary target and was on the edge of the KHTX radar it appears to be a roughly similar size and strength to B2 based on maximum column radar reflectivity. Figure 2.3 shows the HFo measured from 20:00-21:00 UTC overlaid on a Google map with the maximum column radar reflectivity for 19:46 UTC. This is 15 minutes before the HFo peak and shows a storm just south of the elevated HFo that was decaying by 20:00 UTC. The wind direction was from the storm towards the GV during the elevated HFo sampling period.

Formic acid model and observation discrepancy
WRF-Chem HFo mixing ratios are substantially lower than expected, less than in the southern U.S. during SENEX. Both authors discuss that this discrepancy is mostly due to models missing secondary formation sources for HFo.
WRF-Chem MOZART gas phase mechanism has 10 gas phase production reactions for HFo. Biogenic precursor reactions with ozone include a-pinene, bpinene, limonene, myrcene, methylbutenol (MBO), and b-caryophyllene. All biogenic reactions have a 5% yield of HFo with the exception of MBO which has a 25% HFo yield. The two non-biogenic sources are ethene and ethyne (acetylene). Ethene reacts with ozone to form HFo with a 50% yield. Acetylene reacts with HO to form HFo with a 35% yield. The last two reactions involve the radical HOCH 2 OO, from formaldehyde, reacting with NO and HO 2 in a 1:1 reaction. The only gas phase HFo sink is the reaction with HO.
A HFo pseudo steady state (pss) mixing ratio was calculated using the MOZART reactions and available low altitude aircraft data. DC8 Whole Air Sampler (WAS) data during the inflow period were used for ethene and acetylene. GV TOGA data during the spiral were used for a-pinene, b-pinene, and MBO. The pss mixing ratio was determined for each HFo formation reaction and the total production based on DC8 and GV data were summed separately. There was no wet or dry deposition accounted for in this approximation. Some minor source species measurements were unavailable during the inflow period, e.g. limonene, mycrene, beta-caryophyllene, and HOCH 2 OO. However, measurements of the main MOZART gas phase sources, e.g.
ethene, acetylene, and MBO, were available. The majority of the pss HFo came from ethene and acetylene (205 + 12 ppt) with an additional 19 + 0.4 ppt attributed to biogenic sources. For comparison, the measured HFo during the GV spiral was 332 + 15 ppt. While the MOZART mechanism appears to do reasonable job of capturing the HFo mixing ratio it was still lower than observed. If deposition was included, the calculated pss would be even lower still, further increasing the discrepancy between observed and calculated.  estimated an HFo combined wet and dry deposition sink that was 4 times the photochemical sink. The HO reaction photochemical loss rate was 0.014 day -1 using GV data. Based on , the combined wet and dry deposition was 0.056 day -1 .
The biogenic pss was recalculated assuming a 31% HFo yield from the isoprene ozonolysis . The new mixing ratio (570 + 196 ppt) was substantially higher though with a much larger standard deviation. This large standard deviation is due to only having two isoprene measurements (602 and 292 ppt) during the GV spiral as a result of TOGA's two minute sampling rate. If only the lower isoprene measurement is used, the biogenic HFo pss is 363 + 13 ppt -much closer to the observed HFo mixing ratio (332 + 15 ppt). The isoprene ozonolysis reaction produces the Criegee biradical which reacts with water to form HFo. This improved production supports previous work, e.g.  and , showing that the Criegee biradical is important for HFo secondary production.
This analysis was expanded beyond the low altitude data to encompass the whole flight. HFo pss mixing ratio was calculated the same as above with the addition of limonene. The HFo pss mixing ratio presented here used the maximum value representing an upper limit. This includes the MOZART reactions plus the 31% yield from isoprene ozonolysis. Figure 2.4 compares the observed HFo to the maximum HFo pss mixing ratio for the isoprene ozonolysis, anthropogenic precursors (ethene and acetylene), and biogenic precursors (a-pinene, b-pinene, limonene, and MBO).
The production from the combined biogenic sources is negligible for the majority of the HFo range. The total calculated HFo is overestimated when the observed HFo was between 300-400 ppt. This is not surprising as there was no deposition. Yet the modeled production sources do not capture the observed high end of HFo (greater than 500 ppt) which includes the high altitude peak in HFo. If there was no isoprene production, as is the case for MOZART, the maximum possible calculated HFo was ~200 pptless than half what was observed.  saw similar discrepancies for the high HFo measurements (greater than 1 ppb). Without the addition of isoprene HO oxidation, GEOS-Chem did not exceed 500 ppt and even with isoprene simulated HFo was up to 5 times underpredicted .
While it is likely there are missing HFo source terms, it is also important to consider how WRF-Chem replicates the mixing ratios for the available sources. Using observational precursors resulted in a similar order of magnitude between calculated and observed HFo. Consequently, it is likely that part of large inconsistency between WRF-Chem and observations could be due to underestimating a precursor source as well. Based on the pss calculations acetylene dominates the HFo production (159+10 ppt). DC8 inflow data for acetylene (~1 km) was 293 + 21 ppt. However, for the simulated 1-2 km bin acetylene was 0.029 + 0.0876 ppt. Clearly, this is quite a large difference in mixing ratios for a dominate HFo source in the MOZART mechanism.
Acetylene comes from NEI anthropogenic emission data. Acetylene's lifetime is greater than the timescale of the simulation (20+ days); therefore, if acetylene emissions were accurately represented it should have been significantly closer to observations. In comparison, the ethene simulated mixing ratio in the 1-2 km bin was 105 + 87 ppt and DC8 inflow average was 89+14 ppt. Therefore, while there are certainly missing gas phase sources in MOZART, if a precursor is underestimated it will impact secondary formation reactions. Until there is a better understanding of HFo sources WRF-Chem should be modified with observational data when available.
Here WRF-Chem was modified with a GV HFo median altitude profile described in Section 2.3.

HAc and GA
The HAc PCIMS signal is sensitive to an isobaric interference from GA and the mass signal associated with HAc is reported as the acetic acid equivalent sum HAc:GA ratio for the Southeastern United States based on literature surface data ranged from 0.9 to 10 and when using aircraft data from 1 to 14 Lee et al., 1995;Talbot et al. 1995). This indicates that previously collected data had at least equal mixing ratios with HAc likely to be greater than GA.
Here we estimate HAc and GA mixing ratios measured as AAES based on literature data. Table 2 Table 2.4. AAES could represent equal measurements of HAc and GA of 0.24 ppb but more likely it is in between the extremes presented here.
In addition to the literature measurements, relative proportions of HAc and GA were estimated with WRF-Chem results using both the background profiles defined in Section 4.3 and the convective outflow average mixing ratios for the convective complex and isolated convection. The convective outflow HAc and GA mixing ratios were within 100 ppt of each other thus the ratios were about 1:1. Figure S2.1 shows the HAc:GA ratio for the WRF-Chem north and south background profiles discussed in Section 4.3 with a 1:1 line. The ratio is greater than 1:1 below 8 km. Only above 8 km does the ratio fall below the 1:1 line in the northern profile corresponding with the GA increase (Figure 2.5a). The increase in the ratio above 11 km corresponds to less than 10 ppt for both so does not reflect what would be measurable. WRF-Chem results support literature measurements that found similar mixing ratios between HAc and GA with it more likely that HAc would be greater than GA.
If AAES was only GA and there was a 1:10 sensitivity then GV GA median would be 0.045 ppb which is lower than the previously reported data for the region and what was predicted by WRF-Chem. Based on this assessment the PCIMS sensitivity likely represents closer to a true sum. Previous measurements and WRF-Chem results suggest greater HAc relative to GA in the BL but equal proportions above the BL.

WRF-Chem Background Profiles
Non-convective air chemical profiles are important as they help us understand the general chemical composition for the region and what could be entrained into the storm. The data chosen were in clear air (q cloud <0.01 g/kg and q ice < 0.01 g/kg) for a 324 km 2 box. Two regions were chosen that were to the north and south of modeled convective storms to give a general representation of the region. Figure 2.5 shows WRF-Chem background mixing ratios for HAc, HFo, and GA for the northern profile ( Figure 2.5a) and southern profile (Figure 2.5b). Figure S2.2 has the background mixing ratios for HP, MHP, and CH 2 O for both the northern ( Figure S2.2a) and southern ( Figure S2.2b) regions. The increase in HFo at 11 km could be the result of using GV data with the drop-off above 11 km the switch to MOZART global data (above 12 km).
Background air appeared to be convectively influenced, e.g. GA increased around 10 km in both profiles and HAc increased in the southern profile around 10 km. This convective influence and mixing are why two profiles are used to characterize the region. As a check that this is not UT gas phase chemistry, a simulation without gas phase chemistry was performed. The background profile structure was similar with and without gas phase chemistry. This indicates that the chemical profile was due to mixing and not in situ production.

Entrainment Fraction
WRF-Chem passive tracers are used to calculate the lateral entrainment rate through the simulated storms as defined in Section 2.6. The storm core entrainment rate was calculated for the convection complex ( The majority (~0.7) of the entrainment for the isolated convection came from the bottom 2 km. The rest of the entrainment came from mostly the 4-5 km bin. The isolated convection average entrainment rate (0.11+0.17/km) was similar to the convective complex. The cloud top height for the isolated convection was 6-8 km which is why there was no entrainment at higher altitudes.  Fried et al. (2016) also discussed the disconnect between the inflow and outflow measurements for the May 21 st storm based on the pentane ratios. Butane and pentane ratios are commonly used to match inflow and outflow periods because they have a low reactivity for the convective transport scale (less than 30 minutes) and low solubility. They also have lower mixing ratios in the UT than BL with few chemical sources outside the BL . Given that the GV sampled the outflow of a convective complex and not an isolated storm an hour after the inflow measurements it is reasonable that there was a disconnect between the inflow and outflow chemical signatures. For this reason observational SE calculations were performed for the downwind SE instead of tracing back to the storm core.

Scavenging Efficiency of Convective Complex
Given the change in storm structure and time between the DC8 inflow measurements and GV outflow measurements, the DC8 inflow serves as a characterization of the region. Species that were only present onboard the GV, ex.
HFo and AAES, were not measured during the DC8 inflow period; therefore, the GV spiral is used as a second representative inflow. GV measurements from the spiral on the way back to base (170 km west of storm) provided inflow measurements over Tennessee for HFo, HAc, CH 2 O, and HP. MHP was not available. The DC8 inflow period was before the convection and that from the GV was after. The GV inflow measurements may be lower than DC8 for more soluble species. The discrepancy in pentane ratio (Table 2.5) between the DC8 inflow and GV outflow reflected potentially different airmass source . When using the butane and pentane ratios for the GV spiral, the inflow and outflow pentane ratio align (Table   2.5). Therefore it is reasonable to assume that the GV spiral inflow region was of a similar chemical composition to the airmass sampled in the outflow.  criterion. The choice of 10% ensured that the grid cells had some basic chemical scavenged from the gas phase preventing the inclusion of model points with unrealistic scavenging. Adding the NH 3 criterion did not substantially change the overall SE but it did lower the standard deviation by removing unrealistic SE values.
As mentioned in Section 2.5 all chemicals in these WRF simulations were prescribed to have an ice retention of zero.  discussed different retention factors and how the soluble species responded differently to the ice retention.  did not determine the best soluble species ice retention coefficients for the May 21st storm due to discrepancies between the inflow and outflow chemical signatures for the observed storm and large standard deviation for the SE of the simulated storm. This study did not use the same observed location to determine the SE so a direct comparison cannot be made between this work and . In addition,  simulations and this one used different cloud microphysics schemes as discussed in Manuscript 1 which impact the hydrometeors-and thus the trace gas scavenging.  reported the WRF-Chem predicted SE for the different ice retention factors. The 0 ice retention simulation had MHP (0+45%), HP (8+28%), CH 2 O (3+34%), HNO 3 (20+22%), and SO 2 (-14 + 148%). Based on HNO 3 this work aligns with an ice retention of 0.5 -1.  found that CH 2 O and HP were not sensitive to ice retention as CH 2 O was nearly completely scavenged, and HP was completely scavenged, when the retention factor was greater than or equal to 0.25.  found that MHP was very sensitive to ice retention. Unfortunately, there were no MHP observations during this flight to compare to either simulation.
Our WRF-Chem MHP SE aligned better with retention factor 0-0.25 than 0.5 though both  and this work reported WRF-Chem MHP SE with standard deviations larger than the average.
The CH 2 O and HP WRF-Chem convective complex SEs were higher than  reported for the 0 ice retention factor simulation. However, CH 2 O and HP were not completely scavenged like  found for any ice retention factor greater than 0. The DC8 inflow HP SE was the same as the simulation with cloud chemistry and within the standard deviation of the simulation without cloud chemistry. The HP GV spiral SE was 90+1.4% and greater than WRF-Chem even when accounting for the standard deviation. The GV spiral inflow HP mixing ratio was double the DC8 inflow (Table S2.2). One potential reason for the different HP SE with the DC8 and GV inflow was a difference in reported measurements for peroxides on the two aircraft . Another possibility is there is a difference in chemical composition between the two regions. Both observed CH 2 O SEs were higher than WRF-Chem. The WRF-Chem CH 2 O SE reflects the range that Fried et al. (2016) expected for the DC3 storms. As mentioned above, the airmass storm had the highest CH 2 O SE of any of the DC3 test cases by ~30% . Since this difference is for both observed SE it is potentially the result of in situ chemistry and not just the disconnect between inflow and outflow.
The WRF-Chem SO 2 SE was higher than  simulations and with a smaller standard deviation. The 0.25 retention factor simulation (41+92%) was closest to our WRF-Chem simulations. The SO 2 SE was higher when using cloud chemistry but given the large standard deviations there was no appreciable difference between the simulation with cloud chemistry (46+22%) or without (31+23%). Future work should compare different ice retention factors with the larger SE sampling range of this study to quantify the impact of ice retention on the larger scale. It is possible this chemical criteria SE method based on HNO 3 scavenging accounted for part of the impact of ice retention because the WRF-Chem SE was bounded by observational HNO 3 SE.
The HFo GV spiral SE was negative (-22 + 17%). A negative SE means there was more HFo in the outflow than the inflow suggesting in situ aqueous production.
The difference between the inflow and outflow was small, 30 ppt, with the outflow higher than inflow. The HFo SENEX SE was around 80% and within the one standard deviation for both WRF-Chem simulations. The drastic difference between the observe HFo SE is due in part to the difference in sampling conditions. SENEX's goal was to quantify the impact of natural and anthropogenic emissions on tropospheric O 3 and aerosol formation; therefore, sampling on warm, dry days was preferable. The GV spiral was at the end of the flight, after several hours of convective activity in the region that would have washed out some HFo. The SENEX and GV spiral provide potential upper and lower bounds to HFo inflow. The WRF-Chem HFo SE was in between SENEX and GV.
The almost identical HFo in the storm inflow and outflow using the GV inflow suggests that there is some in situ production countering the scavenging. WRF-Chem was unable to replicate the observed CH 2 O SE. The similarity between the DC8 and GV CH 2 O SE implies it is not a difference of mismatch between inflow and outflow but of storm dynamics or chemistry. The similarity of butane and pentane ratios ( Table   2.5) for the GV data support that this was not a mismatch between the inflow and outflow. It is possible that the higher than expected CH 2 O SE was because of in situ production of HFo and therefore loss of CH 2 O.
The AAES GV spiral SE is lower than HAc and GA WRF-Chem SE. As discussed in  the PCIMS sensitivity to HAc relative to GA is somewhere between 1:1 and 1:10. The HAc and GA WRF-Chem SE are almost identical and was unexpected given the GA Henry's Law constant is an order of magnitude higher. Though the SEs are similar the outflow average is almost two times higher for HAc than GA. The similar SE is surprising based on the Henry's Law constants for the two but some of the difference might be lost due to the substantial standard deviations.

Isolated Convection Scavenging Efficiency
The SE for a second simulated storm along the Alabama/Tennessee border was determined to investigate how different storm sizes and dynamics could influence organic acid SE (Figure 2.8). The WRF-Chem SE for this storm was determined from 21:00 -21:20 UTC while the storm was still an isolated convective cell. The WRF-Chem SE was determined using the box in Figure S2.3. The wind is confirmed to be moving from SW-NE at high altitude so the box is "sampling" outflow. The q ice and q cloud , stratospheric air removal, HNO 3 >60%, and NH 3 >10% criteria were applied. The observed HNO 3 SE was ~60%. To confirm that 60% HNO 3 wasn't too strict the criterion was lowered to 40%. This slightly lowered the WRF-Chem SE averages and the standard deviation magnitudes increased.
This simulated storm is assumed to be similar to the isolated convection to the northeast of the convective complex (B2) where the HFo plume was sampled (Section 3). The HFo plume was encountered on the second pass further out from the storm (20:00 UTC). Observations are replicated using the DC8, SENEX, and GV spiral inflow data from Section 4.5. The DC8 inflow sampling area was near the isolated storm (to the west of the storm) and 20 minutes before the isolated storm outflow period. The outflow measurements are from near isolated convection to the northeast of B2. NEXRAD radar (Figure 2.3) and aircraft flight videos show that there was isolated convection that was decaying by 20:00 UTC near the HFo plume (720 ppt).
The outflow measurements are GV data between 19:49 -19:57 UTC coinciding with flying by the storm outflow as seen on the aircraft video. The outflow data still meet the stratospheric air (O 3 /CO < 1.25) and q cloud > 0.01 g/kg criteria. There is only one measurement available during this period for the butanes and pentanes. The i/n butane ratio was 0.387 which is closer to the DC8 inflow butane ratio (0.363 + 0.0198) though there was only one measurement (Table 2.5). The i/n pentane ratio is lower than either inflow region suggesting a disconnect between the inflow and outflow.
There was limited data for organic acids during this time.  (Calvert et al., 1985).

HFo Aqueous Production
The WRF-Chem aqueous chemistry mechanism used has only one aqueous source for HFothe CH 2 O + HO reaction. However, other aqueous reactions have been suggested including glyoxal and HO, HP and glyoxal, and HP and glyoxylic acid. These are included in Barth's box model (Table S2.1) as well as a second loss reaction of HP and HFo though this reaction is slow. The difference in outflow HFo is compared between just CH 2 O production and the additional production and loss sources. Both experiments were performed with the WRF-Chem initial mixing ratios, a pH of 5, and aqueous production to 258 K (-15 o C). This would prevent changes in HFo production and loss due to pH.
In the CH 2 O only experiment, HFo at the top of the cloud decreased by 50% for the isolated convection and 80% for the convective complex compared to the inflow. The maximum amount of cloud water was similar between the two storms but the convective complex had a higher cloud top height. This allows for more removal of the soluble species to the cloud water, the analogy for scavenging in this model.
With the addition of other HFo aqueous production sources, the gas phase HFo was the same as the CH 2 O only experiment before the cloud evaporated. HFo increased ~10 ppt for both convective systems relative to the inflow value when the cloud evaporated. This is a very small increase in the outflow that still cannot account for the observed HFo plume. There was no difference in HAc and GA as a result of these changes. The next step is to test different pH values to see how that impacts HFo, HAc, and GA.

Cloud Water pH
Four pH experiments were performed at 5.5, 4.5, 3.5, and 2.5 to cover the pKa range for HFo (3.75) and HAc (4.75). This also covered the natural precipitation range which can be between 5 -5.5 and acid rain range which is typically 4.2 -4.4 (EPA, accessed 3/3/19). The pH experiment used the WRF-Chem near surface mixing ratios and stopped aqueous production at temperatures lower than 258 K (-15 o C). This also only used the CH 2 O HFo production source.
There was no difference in GA mixing ratios for the different pH values even though the aqueous source was HAc which is pH dependent. HFo acted as expected for the pH experiments with the highest pH having the greatest scavenging impact. At the higher pH the equilibrium shifts towards formate allowing more HFo uptake. HFo was almost completely scavenged in the convective complex at the highest pH (5.5).
HFo decreased 80% in the isolated convection for the 5.5 pH experiment. Isolated convection had the biggest HFo increase in the outflow when the pH dropped from 5.5 to 4.5. The HFo outflow gas phase mixing ratio increased almost 100 ppt. The HFo outflow for the convective complex increased ~30 ppt for each drop in pH from 5.5 to 3.5. There was very little difference in HFo for either storm when the pH was lowered from 3.5 to 2.5. Previous model work found the maximum total HFo at a pH of 3.5 (Jacob, 1986). The HFo outflow increased only a few ppt when the pH was lowered to 2.5 relative to the 3.5 pH outflow. The isolated convection HFo outflow was almost back to the inflow value when the pH was 3.5 or lower. The convective complex HFo outflow never reached the inflow value. There was ~a 25% decrease between the inflow and outflow even at a pH of 2.5.
The surprise came with HAc. At the top of the cloud there appeared to be a maximum HAc at a pH greater than the pKa. The 5.5 simulation had the greatest gas phase mixing ratio in the outflow. There is HAc gas phase production, unlike HFo, though it doesn't produce a drastic enough difference in mixing ratio to cause this. The primary HAc aqueous source is pyruvic acid. The other aqueous reaction is acetaldehyde with HO. A pH of 5.5 had the maximum total (gas + aqueous) pyruvic acid of all the pH runs for both storms. The gas phase mixing ratio of pyruvic acid is small (4 orders of magnitude lower than total). The majority is in the aqueous phase and the maximum pyruvic acid is at 5.5 therefore the HAc maximum is a function of pyruvic acid formation. In the model aqueous pyruvic acid is formed by reaction of ozone and MVK.

Aqueous Chemistry Temperature Minimum
WRF-Chem assumes that all aqueous chemistry occurs while the temperature is greater than 258 K (-15 o C) but this may be an underestimation. HFo, HAc, and GA production is compared at a constant pH (5.0) for three different temperature ranges: 258 K (-15 o C), 248 K (-20 o C) and 233 K (-40 o C). This experiment also used the WRF-Chem gas phase near surface values and only the CH 2 O source of HFo. The temperature profile was the same for the two storms. The 258 K simulation stopped aqueous chemistry at 6200 m. The 248 K simulation stopped aqueous chemistry at 7600 m. The 233 K simulation stopped aqueous chemistry at 9400 m. The 233 K simulation was chosen as the upper limit of aqueous chemistry. This is the lowest temperature usually considered for supercooled droplets to exist. As there was no cloud water above 7000 m for the isolated convection, and therefore no aqueous production, this comparison will only be for the convective complex.
There was no difference in GA for the three temperature experiments. GA has one aqueous loss and formation reaction in the box model and the reaction rate between the aqueous formation (k = 1.2x10 9 M s -1 ) and loss are similar though the loss is temperature dependent (A = 1.2x10 9 M s -1 , E/R = -1.3x10 3 K). Consequently, it is expected that there would be little change in concentration as a function of temperature. The gas phase mixing ratio decreased in the cloud for all three simulations but the total concentration remained the nearly constant throughout the storm. This indicates that for GA the box model both gas and aqueous phase chemistry played a very small role compared to scavenging.
There was also no difference in HFo between the temperature simulations.
Again, there is only one aqueous production and loss source. The CH 2 O reaction rate constant is about an order of magnitude higher than the HFo loss rate constant in cloud temperature range so there should have been net production. Both rate constants decrease with temperature and the rate constant at 233 K is about half the 258 K rate constant. The 233 K experiment allows for aqueous chemistry for 3200 m more than the 258 K experiment. This does not compensate for the decreasing reaction rate enough to cause a notable gas phase mixing ratio increase at the cloud top because the cloud water mixing ratio is so small.
In comparison to HFo and GA, HAc has multiple aqueous formation reactions and one loss. The HAc cloud top mixing ratio was higher than the cloud bottom for all box model simulations indicating the HAc aqueous reactions rates were faster than HFo or GA leading to more production. There was a 12% increase in HAc at the cloud top relative to the cloud base for convective complex 233 K simulation. This was the largest increase in the HAc outflow. The cloud top HAc increased 8% for the 248 K simulation and 4% for the 233 K simulation. There was only a 16 ppt difference between the 258 K and 248 K simulations even though this corresponds with a 1400 m altitude difference. Gas phase production was small enough to not impact this temperature difference (less than 4 ppt). The 258 K WRF-Chem temperature constraint is likely an overly conservative estimate for the aqueous chemistry range within a storm. However, if there isn't substantial liquid water content to produce a desired species in the aqueous phase increasing the temperature range is irrelevant.

Best Combined Scenario
In the experiments described above, only one factor was altered at a time to see what the influence was on HFo, HAc, and GA cloud top mixing ratios. None of the above experiments found a similar relative outflow HFo mixing ratio to explain the HFo plume of 700 ppt. The HFo peak observed during RF03 is likely not the product of one difference but a combination of situations that would increase HFo. The final experiment used multiple HFo aqueous reactions at a constant pH of 3.5 with aqueous chemistry up to 233 K. In essence this assumes aqueous production for the whole cloud for both storms. While it is not the same large increase as observed, the modified box model produced higher HFo at the cloud top than cloud bottom for the isolated convection.
This was not possible for the convective complex. The cloud top HAc was ~40 ppt higher than the cloud bottom for both storms. In other words, there was no appreciable difference in HAc in the "outflow" between the storms but there was for HFo which is similar to observations even if the magnitude could not be replicated. Another possibility not considered here was the updraft region had a significantly higher HFo mixing ratio than sampled during the GV spiral. The observed HFo plume is likely a combination of higher HFo inflow than accounted for here as well as in situ aqueous production and subsequent release to the gas phase when the cloud evaporated.

Summary
The primary goal of this study was to understand how convective storms affect organic acids and what potential there is for either organic acid (HFo and HAc) to be transported to the UT. The photochemical lifetime of HFo and HAc increases to 20+ days in the UT; therefore, if they are transported they could influence chemistry far from the original source. Conventionally, because both organic acids are fairly soluble, they are treated as scavenged in a storm and thus not present in the convective outflow. Observational evidence from DC3 found elevated HFo (700 ppt) at 8 km near isolated convection. In comparison, there was ~300 ppt HFo sampled in the outflow of a nearby convective complex. There was not sufficient observational data to explore the cause for this HFo peak and the influence that different storm structures have on scavenging.
A regional chemical model (WRF-Chem) and box model provided insight to how organic acids are altered in convective storms by comparing a convective complex and isolated convection. There was no difference in scavenging efficiency after accounting for the standard deviation for either organic acid between the two storms with or without aqueous chemistry. The impact of pH, adding HFo aqueous production sources, and aqueous chemistry temperature range were all tested with a box model. Cloud top HFo for the isolated convection was higher than the cloud bottom with a pH of 3.5, multiple aqueous chemistry sources, and allowing aqueous chemistry to occur for the whole cloud. The convective complex still had an overall decrease in HFo. HAc at the cloud top was greater than the cloud base for all experiments with the majority of formation from pyruvic acid. There was little difference in the cloud top HAc between the two storms reflecting observations which had similar HAc outflow mixing ratio for both storms. GA was not altered significantly in the box model experiments and was lower than HAc in both models.        Observational formic acid (HFo) mixing ratios (ppt) compared to a pseudo steady state maximum HFo (ppt) calculation using DC3 aircraft data and the MOZART gas phase reactions. The anthropogenic sources include ethene and acetylene. The biogenic sources are a-pinene, b-pinene, limonene, and MBO. A 31% HFo production from isoprene ozonolysis is shown as well. The only gas phase loss is HFo+HO. Figure 2.5: WRF-Chem non-convective, or background, air altitude profiles (km) to the north (a) and south (b) of the main storm for formic acid (HFo), acetic acid (HAc), and glycolaldehyde (GA) (ppt). Background air is defined as having an ice and cloud mixing ratio < 0.01 g/kg and radar reflectivity of zero. Data are averaged over 1 km bins from the surface to 12 km which is the maximum altitude of the simulated storm outflow. Error bars represent 1 standard deviation.         Figure S2.1: HAc:GA ratio in the WRF-Chem background profile (a) to the north of the storms and (b) to the south of the storms. Background air is defined as having a cloud mixing ratio < 0.01 g/kg and radar reflectivity of zero. Data are averaged over 1 km bins from the surface to 12 km which is the maximum altitude of the simulated storm outflow. Error bars represent 1 standard deviation. Figure S2.2: WRF-Chem background air altitude profiles (km) to the north (a) and south (b) of the main storm for hydrogen peroxide, methyl hydroperoxide, and formaldehyde (ppt). Background air is defined as having a cloud mixing ratio < 0.01 g/kg and radar reflectivity of zero. Data are averaged over 1 km bins from the surface to 12 km which is the maximum altitude of the simulated storm outflow. Error bars represent 1 standard deviation. controlled GA production. GA estimated production was lower than HAc and GA loss was an order of magnitude higher. Based on gas phase processes HAc represented a greater portion of AAES than did GA.

Introduction
Formic (HFo) and acetic (HAc) acid are the most abundant carboxylic acids in the troposphere yet their production pathways remain uncertain. Both acids are underestimated in models which has been linked to missing sources Stavrakou et al., 2012;. Organic acids contribute up to 60% of the natural acidity of precipitation in remote regions and 16% in urban areas . With the decrease of NO x and SO x emissions organic acids contributions to acidity should increase in urban to rural areas. Satellite evidence indicates that HFo represents between 30-50% of continental United States summertime rain acidity (Stavrakou et al., 2012). Both organic acids are photochemically long-lived (20+ days with respect to HO) but subject to dry deposition and episodic wet deposition at the Earth's surface resulting in a 1-10 day lifetime in the boundary layer. They represent a relatively long-lived intermediate product in the oxidation of organic matter and we need to understand their sources.
Primary emissions for both acids include vegetation, agriculture, and motor vehicle emissions . Substantial HAc and HFo emissions are also associated with intensive animal farming (i.e. concentrated animal feedlot operations, CAFOs) (Mårtensson et al., 1999). The high ammonia emitted from CAFOs may enhance uptake of HFo and HAc onto particulate matter as well.
Both organic acids have been measured as primary emissions in motor vehicular exhaust with greater HAc relative to HFo (Kawamura et al., 1985). It is hypothesized that both organic acids are released as a result of incomplete combustion which is supported by the high organic acid concentration measured in used oil (Kawamura et al., 1985). Additionally, biomass burning plumes contain organic acids and these are the third most important emitted carbon reservoir (Yokelson et al. 2009).
Secondary production is also a significant source for both acids especially from biomass burning gases, secondary organic aerosols, and photochemical production from volatile organic compounds (VOCs) and oxygenated volatile organic compounds (OVOCs) of natural and anthropogenic origin . (1) . This pathway impacts ozone photochemistry and is potentially a significant HO x source in urban environments (Finlayson-Pitts & Pitts Jr, 1997).
HAc measurements with our chemical ionization mass spectrometer are sensitive to an isobaric interference with hydroxyacetaldehyde, or glycolaldehyde . Glycolaldehyde (GA) is directly emitted in smoldering biomass burning plumes (Yokelson et al., 1997) and there are no other reported primary sources. Secondary production is important for GA with the highest reported mixing ratios associated with biogenic precursors including isoprene, MVK, and 2methyl-3-buten-2-ol (MBO) Tuazon & Atkinson, 1989), biomass burning emissions (Johnson et al., 2013;Yokelson et al., 1997), and the HO oxidation of unsaturated anthropogenic VOCs such as ethene (Niki et al., 1981). Peroxyacetyl  . FRAPPÉ flew multiple upslope-downslope flights Sullivan et al., 2016). Pfister et al. (2017) identified strong upslope events that were sampled by the C-130 including August 12 th (RF 12) which is presented here. presented in which upslope flow was predicted with spillover into Granby on the western slope. Secondary production of HFo, HAc, and GA was explored using established secondary reaction pathways and aircraft measurements.

FRAPPÉ Field Campaign
In situ measurements during FRAPPÉ were made onboard the NSF/NCAR  Altitude is reported here as above ground level (a.g.l). This is the difference between the gps altitude and the elevation measured by the aircraft. This puts the altitude in relation to the surface regardless if along the plains or in the foothills. Some flights were missing elevation data and they are eliminated from any boundary layer calculations.

Organic Acid Instrumentation
Organic acids were measured with the PCIMS (Peroxide Chemical Ionization Mass Spectrometer). PCIMS is a quadrupole negative ion mass spectrometer. PCIMS set-up including calibration, blanks, and ion cluster chemistry is explained in detail in Heikes et al. (2018), O 'Sullivan et al. (2018), and and is only briefly discussed here. In flight, air was sampled with a HIAPER Modular Inlet (HIMIL) hard mounted on the fuselage, extending beyond the aircraft boundary layer.
The HIMIL inlet surfaces were lined with PFA Teflon® tubing. The HIMIL and gas transfer lines were heated to 343K during FRAPPÉ to minimize artifacts caused by the adsorption and/or release of the target gases from or onto the inlet surface.
This was the first field deployment of the PCIMS explicitly using a dual-ionization scheme involving CO 2 and CH 3 I. Reagent CO 2 (400 ppm, 0.080 slpm) in ultrapure air was mixed with a second reagent gas CH 3 I (5 ppm in N 2 , 0.0005 slpm) and carried by an N 2 stream. The reagent gas blend of CH 3 I, CO 2 , O 2 , and N 2 yielded responses for water vapor, organic acids, hydroxyacetaldehyde (discussed below), and peroxides. Organic acids were quantified using the Iclusters, I -(HFo) and I -(HAc) . Peroxides, H 2 O 2 and CH 3 OOH, were quantified using O 2 -(CO 2 ) and Iclusters .
The PCIMS measurement cycle was 720 sec. and included a 90 sec. blank, followed by a 75 sec. gas standard addition, and then a 555 sec. measurement period.
Sixteen mass-to-charge ratios were serially sampled once every 3.5 seconds. Standard additions to ambient air were performed by evaporating aqueous organic acids ( 200® followed by a NaOH trap and was found to be the most effective way to remove both peroxides and organic acids ).
Post-campaign, hydroxyacetaldehyde, or glycolaldehyde (GA), was found to be an isobaric interferent for HAc at m/z 187. Its PCIMS sensitivity relative to HAc was between 1:1 and 1:10 for HAc:GA . HAc and GA are necessarily reported as the Acetic Acid Equivalent Sum (AAES) since only HAc was used in the field as a calibrant. Two other potential interferences examined post-mission were ethanol and propanol. Ethanol has a PCIMS response that was 3.3% that of HFo and was subtracted from all HFo data reported here . 1-and 2-propanol each gave a 1% response relative to HAc. 2-propanol was measured by the TOGA instrument (Rebecca Hornbrook personal communication) and has a large uncertainty (+100%). We used the 2-propanol to estimate the potential interference and subtracted 2%, representing 1-and 2-propanol, of the campaign 2-propanol maximum (509 ppt) from the AAES data.

Source Characterization
The Colorado Front Range was characterized using geographic and chemical parameters. Two potential boundary layer thickness were defined, 1000 m and 2300 m a.g.l. (Bahreini et al., 2018;Vu et al., 2016). A map ( Figure S3.1) highlights three defined geographic emission source regions representing primarily urban sources in the Denver Metropolitan area (red), forest vegetation including Rocky Mountain National Park (green), and CAFOs/oil and gas production in the Greeley area (blue).
CAFO and oil and natural gas (O&NG) were co-located geographically and could not be separated.
A chemical identification scheme was developed to better identify source types and separate O&NG from CAFOs. The source type parameters (Table S3.1) were developed from literature work (Baker et al., 2008;Eilerman et al., 2016;Gilman et al., 2013;Hornbrook et al., 2015Hornbrook et al., , 2017Kim et al., 2010). Different authors have suggested and used different species and thresholds to develop source characterization.  represents natural biogenic sources such as grasses, forest, etc. This eliminates biogenic sources that are co-located with anthropogenic sources. These chemical characterizations are likely more stringent than necessary either with the number of chemical markers needed for each source type or the mixing ratio ranges used.
However, the Front Range is a mixture of multiple sources in a relatively small region and narrow chemical classification ranges limits sampling of multiple source types together.

Chemical Production
HFo, HAc, and GA chemical production on the Front Range was evaluated using the Master Chemical Mechanism, MCM v 3.3.1 (MCM, 2018), via website: http://mcm.leeds.ac.uk/MCM. HFo (Table 3.2), HAc (Table 3.3), and GA production and loss gas phase reactions (

Results and Discussion
The goal of this study is to understand the distribution of HFo and AAES on the Colorado Front Range. Figure 3.1 shows HFo and AAES altitude profiles for the whole campaign. Data are grouped in 1 km bins where the diamond represents the median mixing ratio, the thick line the interquartile, and the thin line the 10 th -90 th percentile for each altitude bin. The maximum altitude sampled during FRAPPÉ was ~7.5 km above sea level thus the altitude profile is from 0 to 8 km. This is the only altitude data not presented as a.g.l in order to use all available data. As discussed in Section 2.1 the elevation measurement is missing for some flights thus a.g.l could not be calculated. AAES mixing ratios were higher than HFo for the whole campaign. The maximum HFo and AAES were in the 3-4 km bin which is between ~1.5-2.5 km a.g.l.
HFo and AAES mixing ratios were < 1 ppb above 6 km. The high altitude measurements have smaller interquartile ranges reflective of cleaner background air and fewer data points than near the surface (Table S3.2). FRAPPÉ focused on low altitude sampling because the goal was to characterize emissions on the Colorado Front Range. As a result, the majority of flights were in the BL. The BL was often turbulent which resulted in aircraft vibrations that affected the PCIMS mass flow controllers and thereby impacted instrument response . AAES had greater variance, represented by the length of the 10 th -90 th percentile line. This could in part be due to the instrumental noise and chemical contamination in the airport hangar before flights .

Geographic and Chemical Divisions
To better understand HFo and AAES distribution, the Front Range was divided geographically and chemically. The geographic divisions ( Figure S3.  (Schobesberger et al., 2016). Net upward HFo fluxes over a boreal forest found that there was a missing primary source or fast high-yield production from monoterpenes (Schobesberger et al., 2016). Geographically, all three HFo medians are statistically different (p<0.001, Kruskal-Wallis (1952)) and the Denver Metropolitan area had the second highest HFo. Chemically, there was no HFo data available with the CAFO and urban source types. The HFo IQR and median for the combined geographic O&NG/CAFO and chemically classified O&NG region were quite similar ( HFo had a clear decrease in mixing ratio above 2 km representative of background air. The highest AAES (> 10 ppb, Figure 3.5) was during the low altitude legs over the eastern portion of the FRAPPÉ domain. The highest HFo (Figure 3.4) was during the low altitude legs over the Denver Metropolitan area and foothills.
AAES was at least 9 ppb during the Greeley missed approach (altitude < 0.35 km a.g.l) with a median of 12 ppb. In comparison, HFo median during the missed approach was 0.82 ppb with a maximum of 0.99 ppb (Figure 3.4). Greeley has joint emissions from O&NG and CAFO sources. Ethane (C 2 H 6 ), representative of oil and gas operations, maximum was 23 ppb during the Greeley missed approach. Ammonia (NH 3 ), representative of CAFOs, was up to 22 ppb during the Greeley missed approach. Using the median values, the AAES/NH 3 ratio was 0.67 ppb/ppb. This is significantly higher than the cattle HAc/NH 3 ratios found by Yuan et al. (2017) and Ngwabie et al. (2008). The AAES/NH 3 ratio reported here includes O&NG as well. A chemical CAFO AAES/NH 3 ratio could not be determined as there were no data points for the chemical criteria during this missed approach. It is also likely the sampled CAFO emissions were a mix of multiple sources. The high AAES at the start of the flight, including the missed approach, might also be due to contamination in the airport hangar. The PCIMS was flushed with N 2 gas before each flight inside the hangar though this was not always completely effective in preventing contamination.
Even with the minimum reported AAES during the missed approach (9.3 ppb) and the maximum NH 3 the ratio was 0.42, higher than previously reported ratios.
High HFo (>1 ppb) and AAES (>10 ppb) along the Denver Metropolitan track corresponded with elevated NO x (8 ppb). The highest isoprene (0.40 ppb) was also encountered in the Denver Metropolitan area. In general, the primary biogenic tracers, isoprene, MBO, and MVK, were all highest during the Greeley missed approach and in the Denver Metropolitan area and not the foothills transect (-105.4 E). This serves as further evidence that the Colorado Front Range is a mixture of source types.
The first flight segment flew stacked legs over Greeley, the Denver Metropolitan area, and foothills similar to the day before. After refueling, a second stack was flown over Denver and the foothills before two legs over the Continental Divide and a missed approach into Granby (-106 E). AAES (Figure 3.5b) variation was similar to the 11 th with higher AAES near anthropogenic sources in Greeley (14 ppb) and Denver (>10 ppb). The highest southwest over the mountains (-105.6 E). There was ~200 ppt isoprene along this same track though the highest was in Greeley similar to the day before. The highest MBO (~200 ppt) was below 1.5 km along the foothills track. HFo has a similar pattern to ozone on the foothills track. Ozone was ~10 ppb higher on August 12 th than the day before with 75 ppb in Granby and at least 70 ppb along the foothills. This had the potential to be an ozone exceedance event. In fact, the next day was an ozone exceedance day at two Denver area EPA sites based on the current 0.07 ppm 8-hour maximum (EPA 8 Hour Ozone Data, accessed 3/1/2019).
These two flights highlight the complicated chemistry and meteorology of the Front Range and also where similar patterns exist. HFo mixing ratios were similar both days along the Denver Metropolitan pass and on the plains. However, HFo was several hundred ppt higher on August 12 th in the foothills and mountains. This difference is due to both the fact that there was less transport from the plains on the 11 th as it was the weaker upslope day and the presence of stratospherically influenced air which would have a lower HFo mixing ratio. AAES had a similar pattern between the two days though higher mixing ratios on the 11 th . AAES during the Greeley missed approach was similar between days even though ammonia was substantially higher on the 12 th . The anthropogenic portions of the flights over the plains had higher AAES than the foothills and mountains. These patterns suggest that HFo formation is influenced more by biogenic secondary sources rather than anthropogenic sources while AAES is the opposite. In the next section, HFo and AAES secondary production is evaluated using the Master Chemical Mechanism.

HFo and AAES Modeled Production
HFo, HAc, and GA gas phase production was assessed using the Master The majority of HFo production came from Criegee B, C, and E (Figure 3.6).
There was little difference in the breakdown of the three main Criegee sources between the two days. HFo secondary formation is a mixture of biogenic (isoprene, MBO, and MVK) and anthropogenic (ibutene, propene, and acrolein). Criegee B is the dominant source representing 44% of HFo production for August 11 th and 37% for August 12 th . It is likely the dominant source because it has multiple precursors (Table   3.2). Isoprene ozonolysis (Criegee E) represents 26% and 32% of HFo production for August 11 th and 12 th . HFo has been shown by other authors (e.g.  to be underestimated by chemical mechanisms.  modified MCM 3.2 and added additional HFo production reactions. No additional reactions were added here because of the assumption that HFo is the only result from Criegee loss though this is unlikely. HFo hourly production rate was estimated for both flights and the production rate was less than 10 ppt/hr for the majority of the flight. The highest production rate, 30 ppt/hr on August 12 th and 20 ppt/hr on August 11 th , was during the Greeley missed approachclose to sources. Propene represented a negligible HAc source (<1%). Peroxyacetyl nitrate (PAN) and acetaldehyde were the two dominate HAc sources (Table 3.3). Overall PAN dominated HAc production with 61% on August 11 th and 84% on August 12 th ( Figure 3.7). PAN and acetaldehyde have primarily anthropogenic sources (Fischer et al., 2014). PAN is a secondary product of VOCs and nitrogen oxides in photochemical smog. The higher portion on the 12 th may reflect the higher mixing ratios of other anthropogenic markers in the region that day as discussed above.
GA is produced with various intermediates (Table 3.4) coming from MBO, MVK, ethene, and acetaldehyde. Unlike the organic acids these are reactions are all with HO and not O 3 or some combination. A substantial fraction of GA production was from biogenic sources on both days. GA formation is primarily from MBO, MVK, and acetaldehyde (Figure 3.8). Half of GA production was from acetaldehyde on August 11 th (50%) with comparable proportions of MBO (22%) and MVK (25%).
As discussed in Section 2.2, a caveat to the PCIMS measurements is that HAc and GA must be accounted for together and reported as the operationally defined AAES. The contribution of HAc and GA to AAES are evaluated with the MCM. The HAc production rate was higher for both days. The HAc median production rate on the 12 th was 44 ppt/hr and GA was 1 ppt/hr. The GA HO loss rate was an order of magnitude greater than HAc. Given the larger production HAc production rate and slower loss rate it is likely that HAc represented a greater portion of AAES measurements when assuming the mixing ratios are controlled by gas phase processes.

Conclusion
HFo and AAES along the Colorado Front Range highlight the impact of sources on formation. HFo was highest near biogenic sources regardless of a chemical or geographic classification though there were anthropogenic HFo sources. The Denver Metropolitan area had the second highest HFo. HFo natural and anthropogenic formation mixture was also found with MCM estimated production. HFo secondary formation was split between biogenic and anthropogenic sources. Isoprene ozonolysis alone accounted for a third of the total HFo production. AAES was highest near anthropogenic sources, in particular oil and natural gas. The highest AAES encountered during the two case studies was during the Greeley missed approachan area dominated by oil and gas and concentrated animal feeding operations. MCM estimated HAc production showed that HAc production was controlled by PAN and acetaldehyde. MCM GA production was lower than MCM HAc production and the GA loss rate was an order of magnitude higher. Working in this gas phase framework, i.e. ignoring deposition, AAES represents a greater portion of HAc to GA. This work contributes to our understanding of HFo, HAc, and GA formation and provides insight into VOC reaction pathways which impact ozone and HO formation.      with enough available data. The geographic regions are: forest (over Rocky Mountain National Park), the combined oil and natural gas (O&NG) and concentrated animal feeding operations (CAFO), and the Denver Metropolitan Area. The two chemical source types with enough data (see Table 3.6) are biogenic and O&NG. The red line is the median and the notches are the median 95% confidence interval. The box shows the interquartile range from the 25 th (q 1 ) to 75 th (q 3 ) percentile. The whiskers (w) extend to the most extreme values not defined as outliers. Outlier data (red plus) are greater than q 3 + w(q 3 -q 1 ) or less than q 1 + w(q 3 -q 1 ).
(b) (a) Figure 3.3: Same as Figure 3.2 except for the acetic acid equivalent sum (AAES). The three chemical source types with enough data (see Table 3.6) are biogenic, concentrated animal feeding operations (CAFO), and oil and natural gas (O&NG).   Figure 3.7: Acetic acid secondary production sources using the Master Chemical Mechanism v 3.3.1 for (a) August 11 th and (b) August 12 th . The reaction pathways are listed in Table 3.3.    Chemistry (WRF-Chem) regional chemical transport model addressed the following questions: 1. Does organic acid scavenging extent differ between a convective multicell complex and an isolated convective cell?
2. Can HFo serve as a tracer of cloud processed air?
3. What HFo potential sources are we not accounting for in models? What does this tell us about the differences in production pathways between HFo and HAc?
4. How do HFo and HAc distributions vary based on natural and anthropogenic sources?
In addition to the questions above, manuscripts 2 and 3 explored the relative contributions of HAc and glycolaldehyde (GA) to the operationally defined acetic acid equivalent sum (AAES). Based on model simulations and previous literature measurements, the AAES HAc:GA instrumental sensitivity was closer to a 1:1 than 1:10 though it is likely the sensitivity is in between these two extremes.
Manuscripts 1 and 2 analyzed an airmass case study, Research Flight 03 on May 21, 2012, from the Deep Convective Clouds and Chemistry Experiment (DC3).
The May 21 st case study was chosen because there was higher than expected HFo by a few hundred parts per trillion (ppt) above background levels in a region dominated by convective outflow. This HFo increase suggests either transport from the boundary layer or formation within the storm and subsequent release in the outflow.
Manuscript 1  Manuscript 2 further explored the possibility of aqueous production and release and the most conducive conditions to produce the HFo plume noted at high altitude. Photochemical box model results suggested this HFo upper troposphere plume could be possible if there were multiple HFo aqueous sources and the cloud evaporated releasing HFo to the gas phase. It is also possible that the cloud water in this storm was more acidic preventing formate from reacting with HO and thus an HFo sink. Confounding this discuss was the absence of HFo storm inflow measurements to assess if this plume could have been from a local region of elevated surface HFo that was ingested into the storm, as was noted in sulfur dioxide. There is still no clear evidence for the feasibility of using HFo to detect cloud processed air.
Manuscript 2 used observed HFo measurements to constrain WRF-Chem as this dissertation corroborated previous work showing that organic acids, HFo in particular, are underpredicted in chemical transport models. WRF-Chem HFo mixing ratios were substantially lower than expected, less than 10 ppt, while observed HFo ranged from 28 -724 ppt. As a result, WRF-Chem and the MOZART gas phase chemical mechanism were evaluated as to the causes of this underprediction.
MOZART estimated HFo mixing ratios were no more than a third of observations without the addition of isoprene ozonolysis with a 31% HFo yield. Further, acetylene, a major HFo precursor in MOZART, was four orders of magnitude lower than that observed and it was concluded that acetylene was underrepresented in the WRF-Chem emission files. As a consequence secondary production of HFo was significantly underrepresented in the model.

Manuscript 3 focused on the Front Range Atmospheric Pollution and
Photochemistry Experiment (FRAPPÉ) field campaign and in particular HFo and AAES source characterization. FRAPPÉ was designed to explore ozone over the Colorado Front Range and there is a potential role for organic acids to aid in characterizing carbon processing and ozone chemistry. HFo was highest in forested regions and near biogenic emissions. AAES was highest near anthropogenic sources including the Denver Metropolitan Area and near oil and gas operations. In addition to a campaign wide characterization, two case study flights, August 11 th and 12 th , were analyzed. These two were chosen because both were forecasted to be upslope, or mountain-valley, circulation flights; although, this was only observed on August 12 th .
August 12 th upslope flow resulted in a "spillover" event in which Front Range air made it up and over the divide to Granby, CO located on the western slope. Elevated HFo was measured in the upslope flow and the highest HFo measured on the 12 th was during the spillover event. The same pattern was not observed in AAES.
HFo, HAc, and GA secondary production for the August 11 th and 12 th flights was estimated using VOC measurements of the acid precursors also obtained on the C-130. The Master Chemical Mechanism was interrogated to identify these precursors. Both anthropogenic and biogenic sources for HFo were present though the majority was from biogenic precursors. A third of HFo production was attributed to isoprene ozonolysis alone. HAc production was found to be controlled by anthropogenic sources with at least 60% from peroxyacetyl nitrate (PAN), a secondary product of VOCs and nitrogen oxides in photochemical smog. GA production was determined to be a mix of anthropogenic and biogenic sources with at least 50% of its production from MBO and MVK. MVK is an isoprene secondary product and MBO emissions are linked to coniferous trees which are found in the Rocky Mountains.
FRAPPÉ measurements and MCM production estimates underscored the breakdown between biogenic and anthropogenic sources for HFo and HAc on the Colorado Front Range.
This work reinforces prior analyses showing organic acids are underpredicted in chemical transport models and highlights gaps in understanding atmospheric carbon processing. As shown here, both organic acids are lofted to the upper troposphere through midlatitude deep convection which will transport them far from their emission source. Moving forward, these measurements can be used to better constrain model reactions to improve our understanding of carbon processing in the atmosphere.