Unexpected long-range transport of glyoxal and formaldehyde observed from the Copernicus Sentinel-5 Precursor satellite during the 2018 Canadian wildfires

Glyoxal (CHO.CHO) and formaldehyde (HCHO) are intermediate products in the oxidation of the majority of volatile organic compounds (VOC). CHO.CHO is also a precursor of secondary organic aerosol (SOA) formation in the atmosphere. These VOCs are released from biogenic, anthropogenic, and pyrogenic sources. CHO.CHO and HCHO tropospheric lifetimes are short during the daytime and at mid-latitudes (few hours), as they are rapidly removed from the atmosphere by their photolysis, oxidation by OH, and uptake on particles/deposition. During nighttime or at high latitudes, lifetime can be 5 prolonged to many hours or even days. Previous studies demonstrated that CHO.CHO and HCHO can be retrieved from space-borne observations using the DOAS method. In this study, we present CHO.CHO and HCHO columns retrieved from measurements of the TROPOMI instrument, launched recently on the Sentinel-5 Precursor (S5P) platform in October 2017. Strongly elevated amounts of CHO.CHO and HCHO are observed during the fire season in British Columbia, Canada, where a large number of fires occurred in August 2018. CHO.CHO and HCHO plumes from individual fire hot-spots are observed 10 in air masses travelling over distances of up to 1500 km, i.e. much longer than expected for the short atmospheric lifetime of CHO.CHO and HCHO. However, the temporal evolution of the plume differs for both species. Comparison with Lagrangianbased FLEXPART simulations for particles with different lifetimes shows that effective lifetimes of 20 hours and more are needed to explain the observations, indicating that CHO.CHO and HCHO are either efficiently recycled during transport or, continuously formed from the oxidation of longer-lived precursors present in the plume. 15

fires. Consequently the retrieved vertical column densities of carbon monoxide (CO) retrievals also CO and NO 2 from the TROPOMI instrument and true color images from VIIRS satelliteinstrument on NPP, which measures near simultaneously 95 with TROPOMI are used as complementary information in order to interpret theour interpretation of the apparent enhanced lifetime of CHO.CHO and HCHO in the plume. Additionally, interpretation of the main source of these species is performed by computing the ratio of CHOCHO toCHO.CHO and HCHO forin the specific events. plume.

CHO.CHO and HCHO observations 100
The Differential Optical Absorption Spectroscopy (DOAS) method has been successfully applied to retrieve atmospheric columns of trace gases having fingerprint narrow absorption bands in the solar spectral range from space-borne instruments (e.g. Burrows et al., 1999). As noted above, there are several studies describing retrievals of OVOC and their use for the identification of VOC sources emission and their emissions Wittrock et al., 2006;Vrekoussis et al., 2009Vrekoussis et al., , 2010De Smedt et al., 2008, 2018Hewson et al., 105 2013;González Abad et al., 2015;Chan Miller et al., 2014;. Algorithms for the retrieval of HCHO and CHO.CHO have been developed for measurements from the SCanning Imaging Absorption spectroMeter for Atmospheric CHartographY (SCIAMACHY) , the Ozone Monitoring Instrument (OMI) , and the second Global Ozone Monitoring Experiment on MetOp-A and -B (GOME2-A and-B) (Munro et al., 2016), which in combination provide a continuous dataset covering a period of more than 110 20 years. In this study, measurements from the TROPOMI instrument on board of the Sentinel-5 pPrecursor (Veefkind et al., 2012) are used to retrieve atmospheric column amounts of CHO.CHO and HCHO. A brief instrument description and relevant details of the retrieval of CHO.CHO and HCHO are given below.

The TROPOMI instrument
The TROPOspheric Monitoring Instrument (TROPOMI) onboard of the Copernicus Sentinal-5 Precursor satellite was 115 launched on 13 October 2017. It has a spectral range in the UV-VIS-NIR-SWIR covering wavelengths from 270 to 500 nm in the UV-VIS, from 675 to 775 nm in the NIR and in a SWIR band from 2305 to 2385 nm. These bands allow the observation of several relevant atmospheric species, including CHO.CHO, HCHO, NO 2 and CO. TROPOMI provides nearly global coverage each day at a spatial resolution of currentlywhich in August 2018 was 3.5 km×7 km (7 km×7 km in the SWIR). The equator crossing time is 13:0030 LT (ascending node). Similar to OMI, TROPOMI is a nadir-viewing imaging 120 spectrograph, which consists ofemploying a two-dimensional CCD, one dimension collecting the spectral information, the other being used for the spatial information. The TROPOMI instrument onboard ofon board the S5P satellite provides data since November 2017 (Veefkind et al., 2012).

CHO.CHO retrieval from TROPOMI measurements
In recent years, several improvements on the retrieval of CHO.CHO have been reported. In 2014, Chan Miller et al. (2014 and  presented new CHO.CHO retrieval algorithms applied to OMI measurements. These studies, similar to previous studies on GOME-2A data, introduced approaches to reduce interference by other absorbers, such as liquid water and nitrogen dioxide (NO 2 ). In this study, an optimized retrieval algorithm for CHO.CHO was developed, building on the heritage from the OMI CHO.CHO retrieval presented by , extended and applied to S5P measurements. Previous studies have shown that cross-correlations between references cross-sections, as well as 130 instrumental structures or features and shifts in the wavelength calibration can introduce systematic errors in the retrieval. As a result, a strong dependence on the fitting window was identified in the retrieved CHO.CHO slant column densities, SCDs (Chan Miller et al., 2014;. In this study, a fitting window from 433 to 465 nm wasis chosen, which is slightly larger than windows used in previous investigations (Vrekoussis et al., 2010;. This fitting window, which enables the liquid water absorption to be retrieved, leads to a reduction in the number of negative CHO.CHO 135 SCsSCDs over oceanic regions in comparison to a shorter fitting window (e.g. 434-458 nm), as well as a reduction in the residuals. The wavelength range selected covers the strong absorption bands of CHO.CHO (452-457 nm), which have already been used in the past to retrieve CHO.CHO from ground and ship-based DOAS configurations as well as from satellites (Sinreich et al., 2007(Sinreich et al., , 2010Wittrock et al., 2006;Vrekoussis et al., 2009Vrekoussis et al., , 2010Chan Miller et al., 2014;. In order to optimize the quality of the retrievals, a row-dependent daily mean Pacific 140 spectrum from the region 50°S, 160°E -50°N, 135°W is used as a background spectrum , which is computed by averaging over the whole latitude range (50° S -50° N) for each across-track viewing direction independently.
In addition, the mean CHO.CHO SCD over the region 30°S, 150°W -30°N, 150°E is computed each day and subtracted from all SCDs to correct for possible offsets. A summary of the selected absorption cross-sections, and other parameters used in the retrieval, as well as a list of the species included in the retrieval, is shown in Table 1. 145 SCDs depend on observation geometry. Vertical column densities (VCDs) are derived from the SCDs by use of so-called air mass factorss (AMFs), which depends on the trace gas profile, air pressure, surface albedo, temperature, aerosols, clouds, and on solar zenith angle and measurement geometry. As the focus of this study is the observation of CHO.CHO in biomass burning emissions, a simple CHO.CHO profile with a Gaussian distribution having its maximum peak at the altitude of the aerosol layer is used (see Figure 1-A). This is based on the assumption that CHO.CHO is found at the same location as the 150 main plume of aerosol and other trace gases. The altitude of the aerosol layer was estimated from profiles retrieved by the Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO) (Vaughan et al., 2004) (Figure 1-B). These aerosol extinction coefficients (k ext ) profiles retrieved at 532 nm are also used in the calculation of the AMFs by the radiative transfer model SCIATRAN . The computations have been performed on a daily basis, assuming a single scattering albedo of 0.92 and a homogenous distribution of aerosols characterized by the mean profile in the whole 6 region of study. The latter is computed from the average of all aerosol profiles, for every single latitude and longitude, and by taken over the region after removing cloud-contaminated pixels (see Figure 1-C). Clouds are not explicitly accounted for in the CHO.CHO and HCHO retrievals but data are filtered for the presence of clouds using an intensity criterion corresponding to a cloud radiance fraction of about 50%. HCHO (Meller and Mootgat, 2000) CHO.CHO (Volkamer et al., 2005) NO 2 (Vandaele et al., 1998) O 4 (Thalman et at., 2013) O 3 (Serduchenko et al., 2014) BrO (Fleischmann et al., 2004) H 2 O (Rothman et al., 2013) Liquid water (Mason et al., 2016) (Vountas et al., 1998) absorption cross-sections (O 3 * λ + (O 3 ) 2 ) from Taylor expansion  Yes No Iterative spike removal (Richter et al., 2011) Applied Intensity offset correction Linear offset (I/I0) Background spectrum Pacific region (50° N, 135° W -50° S, 160°)

HCHO retrieval from TROPOMI measurements
The accuracy of DOAS retrievals of HCHO is limited by cross-correlations with strong absorbers in the UV (e.g. O 3 ) and the signal to noise ratio of the radiance spectra measured by the instrument. Here, an updated and optimized version of the formaldehyde retrieval developed by Wittrock et al. (2006) and Vrekoussis et al. (2010) is used, which applies a slightly larger fitting window extending from 323.5 nm to 361 nm, resulting in a reduction in the noise of the retrieved slant 175 columns.column densities. At wavelengths shorter than 336 nm, interference with O 3 is observed due to the strong absorptionsmall optical depth of the latter.HCHO, which is about three orders of magnitude smaller. This effect is compensated by applying the method described by Pukīte et al. (2010), which consistconsists of adding two additional pseudo-cross-sections to the fit (λσ O3 and σ 2 O3 ) (Pukīte et al., 2010;De Smedt et al., 2008Smedt et al., 2018).
The cross-sections of interfering species are included in the fit as listed in Table 1. Similar 1. In a manner similar to 180 glyoxalthe retrieval of CHO.CHO, a synthetic ring spectrum (Vountas et al., 1998) is used to account for the Ring effect and a row-dependent daily mean Pacific spectrum from the region 50°S, 135°W -50°N, 16°E is used as background spectrum.
Also, aA latitude dependent offset correction based on SCDs from longitudes between 180° E and 160° E is applied to the data. As for CHO.CHO, VCDs, are computed using AMFs, assuming a Gaussian shape for the distribution of HCHO at the layer where the aerosols are located. in the plume. Figures 2-A and 2-B show examples of CHO.CHO and HCHO fit results 185 for the 10th of10 August 2018, compared to the differential reference cross-section for a single measurement. For an individual CHO.CHO measurement, the detection limit is of the order of molec.cm -2 , which is about 10 times smaller than the columns detected from emissions of the wildfires over the British Columbia region of Canada. For HCHO, the detection limit is an order of magnitude higher ( molec.cm -2 ). The detection limit of a single S5P measurement in this study has been estimated in a manner similar to that explained in Alvarado et al. (2004). 190  and 4.6x10 16 molec.cm −2 for HCHO, respectively. The detection limit for a single measurement from S5P is estimated to be 5.0x10 14 molec.cm −2 and 54.5x10 15 molec.cm −2 for CHO.CHO and HCHO, respectively.

Simulation of tracer transport with FLEXPART
In order to simulate the transport of emissions from the Canadian wildfires, forward simulations with version 10.3 of the FLEXible PARTicle dispersion model FLEXPART (Stohl et al., 2005;Pisso et al., 2019) have been performed. The model 200 was driven by using hourly wind fields from the ECMWF ERA5 reanalysis (C3S) at 0.25° horizontal resolution. As a transport model, FLEXPART cannot adequatelydoes not simulate the complete set of chemical transformations leading to the observed lifetimes of trace gases in the biomass burning plumes. NeverthelessHowever, performing simulations for tracers having different mean lifetimes and comparing these to the observed columns yields a valuable insight into the effective lifetime, which the emitted substances have to have in the plume,piece of information in order to explain 205 understand the observed plume evolution. An effective mean lifetime can be estimated by comparing the observed behaviour of the CHO.CHO and HCHO with different FLEXPART simulations of different assumed tracer lifetimes.
In FLEXPART, the mean effective mean lifetime τ of an emitted tracer is treated as exponential decay with a given half-life (t 0.5 ); τ can then be calculated according to τ = t 0.5 /ln(2). As part of this study, FLEXPART simulations were carried out with half-life times of 2, 4, 6, 8, 10, 12, 14, 16, 18, and 20 hours, corresponding to effective mean lifetimes of ∼2.9, 5.8, 8.7, 11.5, 210 14.4, 17.3, 20.2, 23.1, 26.0, and 28.9 hours, respectively. As the exact emissions rates from the wildfires are highly uncertain, the emission fluxes from the Canadian wildfires are assumed to be proportional to fire radiative power (FRP, see below for more details). This means that effectively, emissions are prescribed in the model in arbitrary mass units; scaling the individual emission sources with FRP ensures that when aggregating the simulation results from all fires, each fire's relative contribution to the simulated columns is retained, and the results can be compared to the observed columns. The 215 emissions, prescribed in the model, are taken from the Global Fire Assimilation System (GFAS) daily FRP and plume height data (Rémy et al., (2017). Simulations were performed on a daily basis for the period 6 Aug to 23 August 2018. For each day, all fires from the GFAS data, which had an FRP of more than 3 W/.cm −2 were gridded to a 0.350° horizontal pattern.
The model was then run forward in time for 120 hours, releasing the tracer for the first 24 hours (the full UTC day) from each of the 0.350° grid cells, assuming no temporal variation throughout the day. Vertically, the emissions within the grid 220 cells were evenly distributed over the range of mean altitude of maximum injection heights given by the GFAS data for the respective grid cell. The output of the simulation contains gridded mass concentrations for each time step. Here, a grid with a horizontal resolution of 0.03125° was chosen, to match the resolution of the gridded satellite observations. Hourly output from the simulation was recorded and then vertically integrated to yield simulated tracer columns. In a post-processing step, for one specific mean lifetime, all simulation results (i.e., simulations for all fires on all days) were aggregated into one 225 dataset. While the absolute tracer column density from the model output cannot be simply compared to the measurements, a comparison of the plume patterns and relative distribution between satellite observation and model output gives an indication about the meaningfulness of the prescribed mean lifetime. At this point, the aggregated model output for one effective mean lifetime consists of hourly latitude-longitude grids of vertical tracer columns throughout the whole study period. For comparison to the satellite observations, the hourly time slice closest to the time of overpass at 53°N was chosen. 230

Results and discussion
During August 2018, a high-temperature anomaly led to the outbreak of many fires in the Canadian Western province of British Columbia, resulting in the emission of large quantities of particles and traces gases that in turn affected air quality in the region. As shown in Figure 3-A and -B, the monthly average of CHO.CHO and HCHO vertical columns from S5P show strongly enhanced values over the fire region, suggesting that these fires were a large direct and/or indirect source of 235 CHO.CHO and HCHO. Surprisingly, the CHO.CHO and HCHO enhancements are not limited to the main fire region but extend over large parts of Canada, where only a few fires were observed. In order to investigate the sources of CHO.CHO and HCHO and their distributions, 24-hour assimilation data of fire radiative power from the Global Fire Assimilation System  are analysed. Briefly, FRP is a measure of outgoing radiant heat from fires, measured in units of W.cm −2 and retrieved from space by the MODerate resolution Imaging Spectroradiometers (MODIS) on board of Terra and 240 Aqua satellites (Justice et al., 2002). The assimilated FRP spatially aggregates all valid fire and non-fire observations from both MODIS instruments onto a horizontal resolution of 0.1° x 0.1° and computes the total FRP sums for each grid bin (Justice et al., 2002). The FRP is also used as input in the FLEXPART simulation as described in section 2.5 as a proxy for emission strength.  The highest CHO.CHO VCD values are found over the locations of the most intense fires, as intuitively expected. The HCHO distribution over the fire regions is similar to that of CHO.CHO, but with some differences in the relative 250 distribution. In addition, enhanced CHO.CHO and HCHO columns are also apparent over the southeasternsoutheasternsouth-eastern US, where large isoprene emissions occur. CHO.CHO and HCHO are also detected in plumes crossing central and eastern Canada, where no fires are identified in the FRP map. This pattern points tois best explained by the transport of CHO.CHO and HCHO emanating from the wildfires. However, it is well-known that CHO.CHO and HCHO have been reported to have short atmospheric lifetimes of about ∼2.2 and ∼4.0 hours during daytime, 255 respectively (Atkinson, 1998;Volkamer, et al., 2005a), which (Atkinson, 2000;Volkamer et al., 2005a). Assuming that the lifetime in the plume is not sufficientsimilar to explain thethat observed pattern.at the ground, we expect that CHO.CHO would be removed reasonably close to the fire sources. HCHO would be transported further but we would also expect that it would be transported no more than approximately twice as far as CHO.CHO.
Earlier studies by Wittrock et al. (2006) and Vrekoussis et al. (2009Vrekoussis et al. ( , 2010 showed that CHO.CHO is also observed over 260 oceanic regions, where no CHO.CHO source is expected. The potential of a) a long-range transport of CHO.CHO and/or of CHO.CHO precursors from continental areas, and b) having an unknown oceanic CHO.CHO source were discussed as a possible explanation in part of these observations, but no firm conclusions could be drawn so far. In the present study, with the support of the simulations using FLEXPART model, evidence of long-range transport of CHO.CHO and HCHO or its precursors from biomass burning emissions is shown.investigated. In order to further investigatethis context the transport of 265 CHO.CHO and HCHO, during two fire plume episodes form different periods of satellite measurements are selected (07-10 and 20-21 of August 2018) andare discussed below. To investigate possible transport pathways, forward simulations of the atmospheric transport with FLEXPART were 275 calculated for the same period as HCHO and ofwhen CHO.CHO observationsand HCHO plumes are observed (see Figure 4-B), assuming a randoman effective lifetime of 14.4 hours. The latter is in contrast to the significantly longer than lifetimes of CHO.CHO and HCHO expected, which are significantly shorter.found in the literature. On the other hand, the simulated pattern of air masses follows the distribution ofsame distinctive path as CHO.CHO and HCHO columns well, showingVCDs. This shows good spatial agreement between simulations and satellite observations. The tracer simulated with 280 FLEXPART spreads over the same area as CHO.CHO, providing evidence for the transport of CHO.CHO and HCHO and their precursors over continental Canada. This is even more evident for the second period of interest in this study, which extends from the 20th20 to 21st21 of August 2018 (see Figure 7). While the spatial match of plume and model is good in this example, it is clear from the figure that even an effective lifetime of 14.4 hours does not describe fully the extent of the trace gas transport,CHO.CHO and usingHCHO transported. Using shorter effective lifetimes for CHO.CHO and HCHO, taken 285 form the literature would not reproduce the observations, however. However both ligetimeslifetimes depend on time of day (daytime and nighttime) and also on photon fluxconditions in the plume: on the diurnal photolysis and OH diurnal patttern.cycles as well as on wet/dry deposition processes and other oxidants. Consequently, comparisons of FLEXPART simulations with different effective lifetimes were performed for two selected days, shich are detail discussed as is shown in Section 3.2. 290  Figure 5 shows the results of FLEXPART simulations assuming effective lifetimes for a surrogate chemical species of ∼2.9, 14.4, 23.1, and 28.9 hours for the 10th10 and 20th20 of August 2018. From this figure, it is clear that only for the simulations having effective lifetimes of 23.1 hours or more, a significant fraction of the tracer emitted is present at the end 13 of the plume as observed in the measurements. This is furtheralso illustrated in Figure 6, depicting CHO.CHO and HCHO maps for the 10th of10 August 2018. On top of these maps, contour lines are shown for the simulated air masses assuming effective lifetimes of ∼2.9, 14.4, and 28.9 hours. It is evident that in both cases the tracer distributions simulated with longer effective lifetimes better describe the observed distribution of glyoxalCHO.CHO and formaldehydeHCHO.

Figures 7-A, -B, -C present a second comparison of daily maps of glyoxalCHO.CHO and formaldehydeHCHO VCDs with a 305
FLEXPART tracer having aan effective lifetime of 28.9 hours for the period for 20th20 and 21st of21 August 2018. It is evident that again, the tracer follows the distribution of CHO.CHO and HCHO observations, similar to the first period studied (see Figure 4). However, on 20th20 and 21st of21 August 2018, the CHO.CHO plume spreadsand HCHO plumes spread over the ocean, where no glyoxal or formaldehyde sources are expected, until it up to the point at which the plume disperses after being transported over a distance of about ∼600 km form from the fires. 310 The observed behaviour of the CHO.CHO and HCHO plumes is in contrast with the short atmospheric lifetimes resulting 315 from their rapid removal by photolysis and reaction with OH. In addition, CHO.CHO oligomerises and thus is a source of SOA formation (Schweitzer et al., 1998;Jang et al., 2002;Liggio et al., 2005;Loeffler et al., 2006;Volkamer et al., 2007;Fu et al., 2007;Myriokefalitakis et al., 2008;Stavrakou et al., 2009b, c). The simplest explanation of the observations of CHO.CHO and HCHO is that, during the fire events, both species are transported and/or produced during transport, over long distances, resulting in an effective lifetime of about 28.9 hours. This would impliesy the transport of the 320 VOC precursors of CHO.CHO and HCHO. One reason for the large dispersionlonger range transport of the CHO.CHO and HCHO plumes is the injection of the 325 biomass burning emissions into the free troposphere, where high wind speeds favour transport over long distances. This is a well-known effect that has also been observed for NO 2 in GOME-2 data (Zien et al., 2014). However, even at high wind speeds, the short lifetime of these species would result in much smaller dispersed plumes than the ones observed. There are three possible explanations for this apparent contradiction: Reason 1: The lifetimes of CHO.CHO and HCHO could be significantly longer than expected in these biomass burning 330 plumes. if the OH mixing ratio and UV and visible radiation within the plume are much lower than outside the plume. There is, however, no indication that this should be the case; on the contrary, OH levels in the biomass burning plume are expected to be enhanced (Folkins et al., 1997), leading to a reduction of the expected CHO.CHO lifetime in the unpolluted troposphere. Folkins et al., 1997), leading to a reduction of the expected CHO.CHO and HCHO lifetimes. In this context, it is interesting to investigate the NO 2 VCD observed. The NO 2 plumes coming from the biomass burning are shown in figure  335 8-B. During daytime NO 2 is removed in the gas phase by reaction with OH. Provided sufficient O 3 is present the photolysis of NO 2 produces NO and O which react respectively with O 3 to make NO 2 and oxygen molecules to make O 3 . This is known as a -do nothing cycle‖. NO 2 appears to decay relatively rapidly in the plumes coming from the fires. Our assumption that oxidation and photolysis of CHO.CHO and HCHO is relatively rapid, is thus not contradicted by the NO 2 decay in the fire plumes. 340 Reason 2: There could be an efficient recycling process between the gas and aerosol phase, resulting in the observed extended effective lifetimes of CHO.CHO and HCHO. However, this reason is considered unlikely, because there is not yet any strong evidence of HCHO being a precursor of SOA formation, and as the shape of the plumes for both trace gases is similar, a similar mechanism is expected for both. Also, evidence for the release of CHO.CHO following the formation of oligomers in the aerosol phase, is limited , and references therein). 345 Reason 3: The plume could contain glyoxalCHO.CHO and formaldehyde HCHO precursors, i.e. VOCs, which are slowly 350 produce additional VOCs oxidized, releasing CHO.CHO and HCHO along the trajectory, resulting. If true, this would result in an apparent increase in lifetime. In order to better assess the CHO.CHO and HCHO spatial distribution seen on the 10th of10 August 2018, two additional S5P TROPOMI retrievals have been taken into account; the column-averaged dry air mole fractions of CO, retrieved by the algorithm described in Schneising et al. (2019), and the VCD NO 2 VCD retrieved using an algorithm similar to the one described for the GOME-2 instrument  and using AMF calculated following 355 the same approach as the one described before for glyoxalCHO.CHO and formaldehyde HCHO (see section 2.3). The CO plume shows a similar spatial behavior as HCHO pattern to those of CHO.CHO and CHO.CHO HCHO (see Figure 8-E).8-E). As CO is a relatively long-lived tracer of fire emissions, having a lifetime with respect to OH of months, this is further confirmation for the fact that supports the fire origin of the VOCs and/or their precursors originate from the fires and then undergo long-range transport.plume. As noted above, NO 2 is removed by OH faster than CHO.CHO and HCHO. The NO 2 ~~ VCD enhancements, in contrast to those of CHO.CHO and HCHO, are limited to the proximity of the fire hot-spots (see Figure 8-D). This is the behaviour expectedagrees with that assumed for a molecule with a short atmospheric lifetime and highlights how unusual the behavior of the VOCs is.. A true-colour image from the Visible Infrared Imaging Radiometer Suite (VIIRS) clearly shows the distribution of smoke and aerosols produced because of the emission from the fires being transported and transformed (Figure 8-F). The distribution of the aerosol appears qualitatively to be similar to the 365 CHO.CHO, HCHO, and CO distributions, indicating. We infer that these speciesthe transported plumes are mixed in the aerosol layer.mixtures of CO, CHO.CHO, HCHO, aerosol and presumably other pollutants released by the fire. It is interesting to note that CHO.CHO and CO follow mainly the main plume, while the HCHO distribution is more diffuseddiffuse and shows enhanced values also over regions where a thinner aerosol plume is visible in the VIIRS image;. This may possibly originatingoriginate from another unidentified fires as this part of the plumeor another unknown source, 370 which is not apparentincluded in the our FLEXPART simulations.
As an additional criterioninformation, the ratio of CHO.CHO glyoxal-to-HCHOformaldehyde (, R GF ), is shownpresented in  (Vrekoussis et al., 2010). This is an indication of enhanced primary emissions of glyoxalCHO.CHO relative to those of HCHO from fires. Progressively lowerLower R GF values are then found untilcloser to the end of the plume suggesting either 375 aimplying a decreasing of glyoxal production over time or a changeof CHO.CHO relative to that forof HCHO during the transport of polluted air in the fire emissions injected into the plume. Another potential explanation would be the mixing in of air from different origins, having lower CHO.CHO and/or higher HCHO different concentrations of the trace gases, during the plume transport. This is however not confirmed by the observed CO behaviour, which shows a similar spatial distribution to CHO.CHO and HCHO. 380 The comparison of retrieved S5P columns and FLEXPART tracer simulations discussed is based on a number of simplifications. First of all, theThe observational conditions of the biomass burning plumes are complex, and aerosol scattering and absorption certainly impact on the sensitivity of the retrievals. While this is taken into account by using air mass factors for elevated plumes positioned at altitudes derived from CALIPSO observations, there isremains considerable uncertainty with respect to absolute values. Aerosol loading and optical properties will vary overalong the plume and thus 385 will the retrieval sensitivities, and this is. This not modelled here.explicitly in this study. However, the differences apparent between the spatial distributions of glyoxalCHO.CHO and NO 2 , which are retrieved in very similar spectral regions, provide evidence for the fact that measurement sensitivity isdoes not explain the driver fordifferences in the observed VCD plume patterns. Another crucial simplification is the assumption of a constant fire emissions and the proportionality of FRP and emission strength in the FLEXPART simulations. In reality, fire emissions will also depend on the type of biomass burned, 390 the age of the fire, the time of the day and the environmental conditions, and this will have an effect on the trace gas distribution along the plume, which reflects both chemical transformation and the history of emissions. Modelling of this time-evolution is complex, if possible at all, and out of the scope of this study. However, the observation that both CHO.CHO and HCHO are present in the biomass burning plume after extended time periods and over long distances is robust and can only beis best explained by continuousthe release of CHO.CHO and HCHO from the transformation of 395 longer-lived precursors and/or efficient recycling processes in the plume as discussed above.

Summary and conlusionsconclusions
The retrieval of formaldehyde and glyoxal total column amounts CHO.CHO and HCHO VCDs from measurements of the 405 TROPOMI instrument on board the Sentinel-5P satellite is reported. This enables the extension ofwill extend the datasets already available from the SCIAMACHY, GOME-2, and OMI instruments, and shows the. The advantage of the high spatial resolution and low noise of TROPOMI for studying specific geophysical phenomena. The is well demonstrated in the features and plumes seen in the CHO.CHO, HCHO, CO, and NO 2 VCDs.
In this case study the satellite data show clear evidence for pyrogenic emissions of CHO.CHO and HCHO and their 410 precursors during the wildfire season in summer 2018 in British Columbia, Canada. The spatial and temporal pattern of the highest retrieved CHO.CHO and HCHO columnsVCDs are associated with areas having high fire radiative power, as observed identified in the MODIS fire data products, indicating. This indicates that in these areas, pyrogenic emissions are the dominant sourcessource of CHO.CHO and HCHO. In addition to local enhancements of NO 2 , CHO.CHO, and HCHO, anclose to the fires, extended plumeplumes of elevated CHO.CHO and HCHO amounts isVCD are observed on some days 415 downwind of the fires. This finding provides evidence of either a) long-range transported VOCs, primarily emitted from the fires or b) their production from precursor species during the transport of. This is in contrast with the plume.behaviour of NO 2 , which is also transported but is short lived. The spatial and temporal VOCCHO.CHO and HCHO distribution observed from satellite follows the samea similar pattern asto that of CO, which is long lived, and that simulated by the FLEXPART dispersion model, initialized by tracer emissions starting at known fire locations. Enhanced CHO.CHO and 420 HCHO columns were found in the S5P data up to 1500 km from their sources.
In order to obtain reasonable agreement between the model results and the measurements, an effective tracer lifetime of more than 20 hours and up to 28.9 hours needs to be assumed in the FLEXPART simulations. This is significantly longer than the anticipated lifetimes of glyoxalCHO.CHO and formaldehyde.HCHO. The long transport of glyoxalCHO.CHO and formaldehyde inHCHO along the length of the plume could be associated with the lifting of glyoxal and formaldehydethese 425 traces gases being lifted from the boundary layer into the free troposphere, where high wind speeds lead to rapid transport.
The long apparent lifetime of CHO.CHO and HCHO in the transported plumes could either be explained by a real increase in atmospherictheir lifetime due toin the specific plume because of photochemical conditions in the biomass burning plume or as we attribute, the presence of longer-lived precursorsplume, which are oxidized to form CHO.CHO and HCHO during transport.we consider unexpected. Based on our current knowledge, the most probable explanation of the apparent long 430 lifetime of CHO.CHO and HCHO would be the latter where formation of glyoxal and formaldehyde within the plume is caused by the oxidation of a mixture of longer-lived emitted VOC speciesprecursors (e.g. methanol, ethanol, acetylene, aromatics, glycolaldehyde, ethylene etc.), that in turn, form CHO.CHO and HCHO at different rates. Further research is needed to investigate how frequent such fire-related long-range transport events of VOCs are, what the exact. The chemical mechanism of the formation within the plume is and howof the CHO.CHO and HCHO in the plumes downwind of the fires 435 needs to be identified. The assessment of the number of such fires events impact on ozoneand their relevance for tropospheric O 3 and aerosol production downwind of the fires and air quality downwind of the firesis required. all authors and developed the glyoxal and formaldehyde and NO 2 retrievals for TROPOMI measurements. M. Vrekoussis,A. 440 Hilboll and A. B. Kalisz Hedegaard have designed and performed the FLEXPART experiment for simulationsimulations of the airmassess assuming different effective lifetimes. O. Schneising has developed the CO retrieval and provided the CO data for the comparison with glyoxal and formaldehyde products.
Acknowledgements. The authors acknowledge financial support provided by the University of Bremen. Copernicus Sentinel-445 5P lv1 data from 2018 were used in this study. This publication contains modified COPERNICUS Sentinel data (2018). We thank the MACC team for providing the GFASv1.0 FRP and injection height products. FLEXPART simulations were conducted on the University of Bremen's HPC cluster Aether, funded by DFG within the scope of the Excellence Initiative.
VIIRS and CALIPSO data were obtained from the NASA Langley Research Center Atmospheric Science Data Center.
FLEXPART simulation results were generated using Copernicus Climate Change Service Information (ERA5). 450 This is in part preparatory work for the analysis of the data from the DFG SPP HALO EMeRGe project, which has one focus on biomass burning and the long range transport of such plumes. We thank Abram Sanders (European Organisation for the  where a large number of fires occurred in August. CHO.CHO and HCHO plumes from individual fire hot-spots are observed in air masses travelling over distances of up to 1500 km, i.e. much longer than expected for the relatively short tropospheric lifetime expected for CHO.CHO and HCHO. Comparison with simulations by the particle dispersion model FLEXPART indicates that effective lifetimes of 20 hours and more are needed to explain the observations of CHO.CHO and HCHO if they decay in an effective first order process. FLEXPART used in the study calculates accurately the transport. In addition 25 an exponential decay, in our case assumed to be photochemical, of a species along the trajectory is added. We have used this simple approach to test our assumption that the CHO.CHO and HCHO are created in the fires and then decay at a constant rate in the plume, as it is transported. This is clearly not the case and we infer that CHOCHO and HCHO are either efficiently recycled during transport, or continuously formed from the oxidation of longer-lived precursors present in the plume, or possibly a mixture of both. We consider the best explanation of the observed CHO.CHO and HCHO VCD in the 2 plumes of the fire is that they are produced by oxidation of longer-lived precursors, also released by the fire and present in the plume.

Formaldehyde (HCHO) is produced in the oxidation of both methane (CH 4 ) and other Volatile Organic Compounds (VOC).
Glyoxal (CHO.CHO) is the smallest alpha-dicarbonyl formed in the oxidation of many VOC containing two or more carbon 35 atoms. Although both CHO.CHO and HCHO, which are known as OVOC (Oxygenated Volatile Organic Compounds) have similar rates of reaction with the hydroxyl radical (OH) in the troposphere, the photolysis frequency of HCHO, which absorbs and is photolysed in the ultraviolet-A (UV-A), is significantly smaller than that of CHO.CHO, which absorbs in the blue. As a result, the atmospheric lifetime of HCHO is longer than that of CHO.CHO (Atkinson, 2000). Both species are short-lived during daytime due to their rapid removal by photolysis and reaction with OH radicals (Atkinson, 2000;40 Volkamer et al., 2007). These processes are the major sinks of CHO.CHO and HCHO, contributing about 69% and 96%, respectively. The remaining part of HCHO is removed by deposition (4%), while for CHO.CHO, 22% are removed by SOA formation and 8% by deposition (Stavrakou et al., 2009a, c). Additionally, HCHO and probably also CHO.CHO during the night are removed by reaction with nitrate (NO 3 ) radicals (Atkinson, 2000). CHO.CHO and HCHO play a key role in tropospheric chemistry because they act as temporary reservoirs releasing carbon monoxide (CO) and HOx (OH and 45 Hydroperoxyl, HO 2 ) free radicals, which participate in catalytic cycles creating and destroying tropospheric ozone (O 3 ).
The slant and vertical column densities of HCHO were first observed from space using measurements from the GOME instrument (e.g. Burrows et al., 1999, and references therein). These columns were later used to estimate the emission strength of precursor VOC Abbot et al., 2003). The simultaneous observation of CHO.CHO and HCHO (Wittrock et al., 2006) enabled an improved assessment of atmospheric VOC levels and the knowledge of the ratio of 50 CHO.CHO-to-HCHO (R GF ), (Vrekoussis et al., 2010), provides some differentiation of source types. Studies have used HCHO, partly in combination with CHO.CHO to estimate the biogenic isoprene emissions (Fu et al., 2007;Stavrakou et al., 2009a, b, c;Liu et al., 2012;Marais et al., 2012). This is the largest natural source of CHO.CHO (Guenther et al., 2006;Fu et al., 2007). The amount of biogenically emitted VOC depends on several factors including, amongst others, the plant species and weather conditions (e.g. temperature and humidity) (Guenther et al., 2000). In urban and rural regions, there are also 55 contributions to the amounts of CHO.CHO from human activities, such as from fossil fuel production, distribution and combustion: the largest anthropogenic source of VOC precursors of CHO.CHO being motor vehicle emissions due to either evaporation or incomplete combustion of fuel . Globally, 55% of CHO.CHO is produced by biogenic precursors, while 27% are from anthropogenic and the remaining 18% from pyrogenic emissions (Stavrakou et al., 2009a).
Fires and vehicle exhausts are thought to be the only two sources, which directly emit CHO.CHO (Stavrakou et al., 2009a; In August 2018, unusually high temperatures caused severe drought in some areas of North America and resulted in the outbreak of many wildfires: the province of British Columbia (BC) in Canada was one of the most affected areas. The 2018 season is the worst on record, with 6826 fires being detected and an area of approximately 22500 km 2 of land burned (Natural Resources Canada, 2018). These fires emitted many different pollutants into the atmosphere, e.g. CO, NOx, VOC, 65 OVOC, O 3 , SO 2 , CO 2 , HCHO, HONO, CH 3 CO.O 2 .NO 2 (PAN) and other toxic species as well as aerosols (Urbanski et al., 2018). During the transport of plumes from fires, photochemical transformation of emitted species occurs. Overall, polluted air is transported to regions where the plumes are dispersed. CHO.CHO and HCHO column amounts are observed by remote sensing from satellite using Differential Optical Absorption Spectroscopy (DOAS) on measurements of the radiances backscattered from the Earth's surface and atmosphere. The global maps of CHO.CHO and HCHO retrieved from 70 SCIAMACHY, GOME-2, and OMI show enhanced CHO.CHO and HCHO over tropical rain forests but also over other regions with high isoprene emissions. In addition, hot-spots of CHO.CHO and HCHO from fire emissions can be detected over large wildfires (Wittrock et al., 2006;Vrekoussis et al., 2009Vrekoussis et al., , 2010Chan Miller et al., 2014;De Smedt et al., 2008, 2018. In this study, we present novel observations of CHO.CHO and HCHO retrieved from the high spatial resolution observations 75 of the instrument TROPOMI on board the S5P platform. On 7 August 2018, strongly elevated amounts of CHO.CHO and HCHO were observed over British Columbia and attributed to being predominantly from fires. Surprisingly, these elevated levels of CHO.CHO and HCHO were not limited to the vicinity of the fires. The fire plumes, which contain both CHO.CHO and HCHO remain visible for several days and appear to travel long distances from the sources. We have used forward simulations of atmospheric transport of air masses calculated using the FLEXPART model (Pisso et al., 2019). The 80 simulations include an effective first order loss, which determines the mean effective lifetime of the tracer. Those tracers emitted over the fire hot-spots with long effective lifetimes reproduce best the evolution of the plumes of CHO.CHO and HCHO for most of the fire events, and thus provide estimates of the effective lifetimes of CHO.CHO and HCHO in the plumes, as is described in the sections below. The R GF provides knowledge about their sources in the plume. CO and nitrogen dioxide, NO 2 , have respectively longer and shorter lifetimes with respect to reaction with OH, and smoke and 85 aerosol are also transported in the plumes from fires. Consequently the retrieved vertical column densities of CO and NO 2 from the TROPOMI instrument and true color images from VIIRS instrument on NPP, which measures near simultaneously with TROPOMI are used as complementary information in our interpretation of the apparent enhanced lifetime of CHO.CHO and HCHO in the plume.

CHO.CHO and HCHO observations
The Differential Optical Absorption Spectroscopy (DOAS) method has been successfully applied to retrieve atmospheric columns of trace gases having fingerprint narrow absorption bands in the solar spectral range from space-borne instruments (e.g. Burrows et al., 1999). As noted above, there are several studies describing retrievals of OVOC and their use for the identification of VOC sources and their emissions Wittrock et al., 2006;Kurosu et 95 al., 2007;Vrekoussis et al., 2009Vrekoussis et al., , 2010De Smedt et al., 2008, 2018Hewson et al., 2013;González Abad et al., 2015;Chan Miller et al., 2014;. Algorithms for the retrieval of HCHO and CHO.CHO have been developed for measurements from the SCanning Imaging Absorption spectroMeter for Atmospheric CHartographY (SCIAMACHY) , the Ozone Monitoring Instrument (OMI) , and the second Global Ozone Monitoring Experiment on MetOp-A and -B (GOME2-A and-B) 100 (Munro et al., 2016), which in combination provide a continuous dataset covering a period of more than 20 years. In this study, measurements from the TROPOMI instrument on board the Sentinel-5 Precursor (Veefkind et al., 2012) are used to retrieve atmospheric column amounts of CHO.CHO and HCHO. A brief instrument description and relevant details of the retrieval of CHO.CHO and HCHO are given below.

The TROPOMI instrument 105
The TROPOspheric Monitoring Instrument (TROPOMI) onboard the Copernicus Sentinal-5 Precursor satellite was launched on 13 October 2017. It has a spectral range in the UV-VIS-NIR-SWIR covering wavelengths from 270 to 500 nm in the UV-VIS, from 675 to 775 nm in the NIR and in a SWIR band from 2305 to 2385 nm. These bands allow the observation of several relevant atmospheric species, including CHO.CHO, HCHO, NO 2 and CO. TROPOMI provides nearly global coverage each day at a spatial resolution which in August 2018 was 3.5 km×7 km (7 km×7 km in the SWIR). The equator 110 crossing time is 13:30 LT (ascending node). Similar to OMI, TROPOMI is a nadir-viewing imaging spectrograph, employing a two-dimensional CCD, one dimension collecting the spectral information, the other being used for the spatial information. The TROPOMI instrument on board the S5P satellite provides data since November 2017 (Veefkind et al., 2012).

CHO.CHO retrieval from TROPOMI measurements 115
In recent years, several improvements on the retrieval of CHO.CHO have been reported. In 2014, Chan Miller et al. (2014) and  presented new CHO.CHO retrieval algorithms applied to OMI measurements. These studies, similar to previous studies on GOME-2A data, introduced approaches to reduce interference by other absorbers, such as liquid water and nitrogen dioxide (NO 2 ). In this study, an optimized retrieval algorithm for CHO.CHO was developed, measurements. Previous studies have shown that cross-correlations between references cross-sections, as well as instrumental structures and shifts in the wavelength calibration can introduce systematic errors in the retrieval. As a result, a strong dependence on the fitting window was identified in the retrieved CHO.CHO slant column densities, SCDs (Chan Miller et al., 2014;. In this study, a fitting window from 433 to 465 nm was chosen, which is slightly larger than windows used in previous investigations (Vrekoussis et al., 2010;. This fitting window, 125 which enables the liquid water absorption to be retrieved, leads to a reduction in the number of negative CHO.CHO SCDs over oceanic regions in comparison to a shorter fitting window (e.g. 434-458 nm), as well as a reduction in the residuals.
The wavelength range selected covers the strong absorption bands of CHO.CHO (452-457 nm), which have already been used in the past to retrieve CHO.CHO from ground and ship-based DOAS configurations as well as from satellites (Sinreich et al., 2007(Sinreich et al., , 2010Wittrock et al., 2006;Vrekoussis et al., 2009Vrekoussis et al., , 2010Chan Miller et al., 2014;Alvarado 130 et al., 2014). In order to optimize the quality of the retrievals, a row-dependent daily mean Pacific spectrum from the region 50°S, 160°E -50°N, 135°W is used as a background spectrum , which is computed by averaging over the whole latitude range (50°S -50°N) for each across-track viewing direction independently. In addition, the mean CHO.CHO SCD over the region 30°S, 150°W -30°N, 150°E is computed each day and subtracted from all SCDs to correct for possible offsets. A summary of the selected absorption cross-sections, and other parameters used in the retrieval, as well as a list of 135 the species included in the retrieval, is shown in Table 1.
SCDs depend on observation geometry. VCDs are derived from the SCDs by use of so-called air mass factors (AMFs), which depend on the trace gas profile, surface albedo, aerosols, clouds, and on solar zenith angle and measurement geometry. As the focus of this study is the observation of CHO.CHO in biomass burning emissions, a simple CHO.CHO profile with a Gaussian distribution having its maximum peak at the altitude of the aerosol layer is used (see Figure 1-A). 140 This is based on the assumption that CHO.CHO is found at the same location as the main plume of aerosol and other trace gases. The altitude of the aerosol layer was estimated from profiles retrieved by the Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO) (Vaughan et al., 2004) (Figure 1-B). These aerosol extinction coefficients (k ext ) profiles retrieved at 532 nm are also used in the calculation of the AMFs by the radiative transfer model SCIATRAN . The computations have been performed on a daily basis, assuming a single scattering albedo of 0.92 145 and a homogenous distribution of aerosols characterized by the mean profile in the whole region of study. The latter is computed from the average of all aerosol profiles taken over the region after removing cloud-contaminated pixels (see Figure 1-C). Clouds are not explicitly accounted for in the CHO.CHO and HCHO retrievals but data are filtered for the presence of clouds using an intensity criterion corresponding to a cloud radiance fraction of about 50%.

HCHO retrieval from TROPOMI measurements 160
The accuracy of DOAS retrievals of HCHO is limited by cross-correlations with strong absorbers in the UV (e.g. O 3 ) and the signal to noise ratio of the radiance spectra measured by the instrument. Here, an updated and optimized version of the formaldehyde retrieval developed by Wittrock et al. (2006) and Vrekoussis et al. (2010) is used, which applies a slightly larger fitting window extending from 323.5 nm to 361 nm, resulting in a reduction in the noise of the retrieved slant column densities. At wavelengths shorter than 336 nm, interference with O 3 is observed due to the small optical depth of HCHO,165 which is about three orders of magnitude smaller. This effect is compensated by applying the method described by Pukīte et al. (2010), which consists of adding two additional pseudo-cross-sections to the fit (λσ O3 and σ 2 O3 ) (Pukīte et al., 2010;De Smedt et al., 2008, 2018. The cross-sections of interfering species are included in the fit as listed in Table 1. In a manner similar to the retrieval of CHO.CHO, a synthetic ring spectrum (Vountas et al., 1998) is used to account for the Ring effect and a row-dependent daily mean Pacific spectrum from the region 50°S, 135°W -50°N, 16°E is used as background 170 spectrum. A latitude dependent offset correction based on SCDs from longitudes between 180° E and 160° E is applied to the data. As for CHO.CHO, VCDs are computed using AMFs, assuming a Gaussian shape for the distribution of HCHO at the layer where the aerosols are located in the plume. Figures 2-A and 2-B show examples of CHO.CHO and HCHO fit results for 10 August 2018, compared to the differential reference cross-section for a single measurement. For an individual CHO.CHO measurement, the detection limit is of the order of molec.cm -2 , which is about 10 times smaller than the 175 columns detected from emissions of the wildfires over the British Columbia region of Canada. For HCHO, the detection limit is an order of magnitude higher ( molec.cm -2 ). The detection limit of a single S5P measurement in this study has been estimated in a manner similar to that explained in Alvarado et al. (2004). can then be calculated according to τ = t 0.5 /ln(2). As part of this study, FLEXPART simulations were carried out with halflife times of 2, 4, 6, 8, 10, 12, 14, 16, 18, and 20 hours, corresponding to effective mean lifetimes of ∼2.9, 5.8, 8.7, 11.5, 14.4, 17.3, 20.2, 23.1, 26.0, and 28.9 hours, respectively. As emission rates from wildfires are highly uncertain, the emission fluxes from the Canadian wildfires are assumed to be proportional to fire radiative power (FRP, see below for more details).
The emissions, prescribed in the model, are taken from the Global Fire Assimilation System (GFAS) daily FRP and plume 200 height data (Rémy et al., 2017). Simulations were performed on a daily basis for the period 6 to 23 August 2018. For each day, all fires from the GFAS data, which had an FRP of more than 3 W.cm −2 were gridded to a 0.350° horizontal pattern.
The model was then run forward in time for 120 hours, releasing the tracer for the first 24 hours (the full UTC day) from each of the 0.350° grid cells, assuming no temporal variation throughout the day. Vertically, the emissions within the grid cells were evenly distributed over the range of mean altitude of maximum injection heights given by the GFAS data for the 205 respective grid cell. The output of the simulation contains gridded mass concentrations for each time step. Here, a grid with a horizontal resolution of 0.03125° was chosen, to match the resolution of the gridded satellite observations. Hourly output from the simulation was recorded and then vertically integrated to yield simulated tracer columns. In a post-processing step, for one specific mean lifetime, all simulation results (i.e., simulations for all fires on all days) were aggregated into one dataset. While the absolute tracer column density from the model output cannot be simply compared to the measurements, a 210 comparison of the plume patterns and relative distribution between satellite observation and model output gives an indication about the meaningfulness of the prescribed mean lifetime. At this point, the aggregated model output for one effective mean lifetime consists of hourly latitude-longitude grids of vertical tracer columns throughout the whole study period. For comparison to the satellite observations, the hourly time slice closest to the time of overpass at 53°N was chosen.

Results and discussion 215
During August 2018, a high-temperature anomaly led to the outbreak of many fires in the Canadian Western province of British Columbia, resulting in the emission of large quantities of particles and traces gases that in turn affected air quality in the region. As shown in Figure 3-A and -B, the monthly average of CHO.CHO and HCHO vertical columns from S5P show strongly enhanced values over the fire region, suggesting that these fires were a large direct and/or indirect source of CHO.CHO and HCHO. Surprisingly, the CHO.CHO and HCHO enhancements are not limited to the main fire region but 220 extend over large parts of Canada, where only a few fires were observed. In order to investigate the sources of CHO.CHO and HCHO and their distributions, 24-hour assimilation data of fire radiative power from the Global Fire Assimilation System  are analysed. Briefly, FRP is a measure of outgoing radiant heat from fires, measured in units of W.cm −2 and retrieved from space by the MODerate resolution Imaging Spectroradiometers (MODIS) on board of Terra and Aqua satellites (Justice et al., 2002). The assimilated FRP spatially aggregates all valid fire and non-fire observations from both MODIS instruments onto a horizontal resolution of 0.1° x 0.1° and computes the total FRP sums for each grid bin (Justice et al., 2002). The FRP is also used as input in the FLEXPART simulation as described in section 2.5 as a proxy for emission strength. The HCHO distribution over the fire regions is similar to that of CHO.CHO, but with some differences in the relative distribution. In addition, enhanced CHO.CHO and HCHO columns are also apparent over the south-eastern US, where large 235 isoprene emissions occur. CHO.CHO and HCHO are also detected in plumes crossing central and eastern Canada, where no fires are identified in the FRP map. This pattern is best explained by the transport of CHO.CHO and HCHO emanating from the wildfires. However, CHO.CHO and HCHO have been reported to have short atmospheric lifetimes of about ∼2.2 and ∼4.0 hours during daytime, respectively (Atkinson, 2000;Volkamer et al., 2005a). Assuming that the lifetime in the plume is similar to that observed at the ground, we expect that CHO.CHO would be removed reasonably close to the fire sources. 240 HCHO would be transported further but we would also expect that it would be transported no more than approximately twice as far as CHO.CHO.
Earlier studies by Wittrock et al. (2006) and Vrekoussis et al. (2009Vrekoussis et al. ( , 2010 showed that CHO.CHO is also observed over oceanic regions, where no CHO.CHO source is expected. The potential of a) a long-range transport of CHO.CHO and/or of CHO.CHO precursors from continental areas, and b) having an unknown oceanic CHO.CHO source were discussed as a 245 possible explanation in part of these observations, but no firm conclusions could be drawn so far. In the present study, with the support of the simulations using FLEXPART, evidence of long-range transport of CHO.CHO and HCHO or its precursors from biomass burning emissions is investigated. In this context the transport of CHO.CHO and HCHO during two fire plume episodes form different periods (07-10 and 20-21 August 2018) are discussed below.  decreases on average from 3x10 15 molec.cm −2 to 3x10 14 molec.cm −2 , while the HCHO has a different variation in the plume but at the end of the plume, it decreases from 3x10 16 molec.cm −2 to 1x10 16 molec.cm −2 . The FLEXPART tracer column decreases from 3x10 6 to 0.3x10 6 for this specific effective lifetime of 14.4 hours.

290
The observed behaviour of the CHO.CHO and HCHO plumes is in contrast with the short atmospheric lifetimes resulting from their rapid removal by photolysis and reaction with OH. In addition, CHO.CHO oligomerises and thus is a source of SOA formation (Schweitzer et al., 1998;Jang et al., 2002;Liggio et al., 2005;Loeffler et al., 2006;Volkamer et al., 2007;Fu et al., 2007;Myriokefalitakis et al., 2008;Stavrakou et al., 2009b, c). The simplest explanation of the observations of CHO.CHO and HCHO is that, during the fire events, both species are transported and/or produced during 295 transport over long distances, resulting in an effective lifetime of about 28.9 hours. This would imply the transport of VOC precursors of CHO.CHO and HCHO.

300
One reason for the longer range transport of the CHO.CHO and HCHO plumes is the injection of the biomass burning emissions into the free troposphere, where high wind speeds favour transport over long distances. This is a well-known effect that has also been observed for NO 2 in GOME-2 data (Zien et al., 2014). However, even at high wind speeds, the short lifetime of these species would result in much smaller dispersed plumes than the ones observed. There are three possible explanations for this apparent contradiction: 305 Reason 1: The lifetimes of CHO.CHO and HCHO could be significantly longer than expected in these biomass burning plumes if the OH mixing ratio and UV and visible radiation within the plume are much lower than outside the plume. There is, however, no indication that this should be the case; on the contrary, OH levels in the biomass burning plume are expected to be enhanced (Folkins et al., 1997), leading to a reduction of the expected CHO.CHO and HCHO lifetimes. In this context, it is interesting to investigate the NO 2 VCD observed. The NO 2 plumes coming from the biomass burning are shown in 310 figure 8-B. During daytime NO 2 is removed in the gas phase by reaction with OH. Provided sufficient O 3 is present the photolysis of NO 2 produces NO and O which react respectively with O 3 to make NO 2 and oxygen molecules to make O 3 . This is known as a -do nothing cycle‖. NO 2 appears to decay relatively rapidly in the plumes coming from the fires. Our assumption that oxidation and photolysis of CHO.CHO and HCHO is relatively rapid, is thus not contradicted by the NO 2 decay in the fire plumes. 315 Reason 2: There could be an efficient recycling process between the gas and aerosol phase, resulting in the observed extended effective lifetimes of CHO.CHO and HCHO. However, this reason is considered unlikely, because there is not yet any strong evidence of HCHO being a precursor of SOA formation, and as the shape of the plumes for both trace gases is similar, a similar mechanism is expected for both. Also, evidence for the release of CHO.CHO following the formation of oligomers in the aerosol phase, is limited , and references therein). 320 Reason 3: The plume could contain CHO.CHO and HCHO precursors, i.e. VOCs, which are slowly oxidized, releasing 325 CHO.CHO and HCHO along the trajectory. If true, this would result in an apparent increase in lifetime. In order to better assess the CHO.CHO and HCHO spatial distribution seen on 10 August 2018, two additional TROPOMI retrievals have been taken into account; the column-averaged dry air mole fractions of CO, retrieved by the algorithm described in Schneising et al. (2019), and the NO 2 VCD retrieved using an algorithm similar to the one described for the GOME-2 instrument  and using AMF calculated following the same approach as the one described before for 330 CHO.CHO and HCHO (see section 2.3). The CO plume shows a similar spatial pattern to those of CHO.CHO and HCHO (see Figure 8-E). As CO is a relatively long-lived tracer of fire emissions, having a lifetime with respect to OH of months, this supports the fire origin of the plume. As noted above, NO 2 is removed by OH faster than CHO.CHO and HCHO. The NO 2 VCD enhancements, in contrast to those of CHO.CHO and HCHO, are limited to the proximity of the fire hot spots (see Figure 8-D). This behaviour agrees with that assumed for a molecule with a short atmospheric lifetime. A true-colour image 335 from the Visible Infrared Imaging Radiometer Suite (VIIRS) clearly shows the distribution of smoke and aerosols produced because of the emission from the fires being transported and transformed (Figure 8-F). The distribution of the aerosol appears qualitatively to be similar to the CHO.CHO, HCHO, and CO distributions. We infer that the transported plumes are mixtures of CO, CHO.CHO, HCHO, aerosol and presumably other pollutants released by the fire. It is interesting to note that CHO.CHO and CO follow mainly the main plume, while the HCHO distribution is more diffuse and shows enhanced values 340 also over regions where a thinner aerosol plume is visible in the VIIRS image. This may possibly originate from other unidentified fires or another unknown source, which is not included in our FLEXPART simulations.
As additional information, the ratio of CHO.CHO-to-HCHO, R GF , is presented in Figure 8-C. Larger values of R GF are found close to the location of the wildfires as already reported in previous publications (Vrekoussis et al., 2010). This is an indication of enhanced primary emissions of CHO.CHO relative to those of HCHO from fires. Lower R GF values are found 345 closer to the end of the plume implying a decreasing production of CHO.CHO relative to that of HCHO during the transport of polluted air in the plume. Another potential explanation would be the mixing in of air from different origins, having lower CHO.CHO and/or higher HCHO concentrations, during the plume transport. This is however not confirmed by the observed CO behaviour, which shows a similar spatial distribution to CHO.CHO and HCHO.
The comparison of retrieved S5P columns and FLEXPART tracer simulations discussed is based on a number of 350 simplifications. The observational conditions of the biomass burning plumes are complex, and aerosol scattering and absorption impact on the sensitivity of the retrievals. While this is taken into account by using air mass factors for elevated plumes positioned at altitudes derived from CALIPSO observations, there remains considerable uncertainty with respect to absolute values. Aerosol loading and optical properties will vary along the plume and thus will the retrieval sensitivities.
This not modelled explicitly in this study. However, the differences apparent between the spatial distributions of CHO.CHO 355 and NO 2 , which are retrieved in similar spectral regions, provide evidence for the fact that measurement sensitivity does not explain the differences in the observed VCD plume patterns. Another crucial simplification is the assumption of a constant fire emissions and the proportionality of FRP and emission strength in the FLEXPART simulations. In reality, fire emissions will also depend on the type of biomass burned, the age of the fire, the time of the day and the environmental conditions, and this will have an effect on the trace gas distribution along the plume, which reflects both chemical transformation and the 360 history of emissions. Modelling of this time-evolution is complex, if possible at all, and out of the scope of this study.
However, the observation that both CHO.CHO and HCHO are present in the biomass burning plume after extended time periods and over long distances is robust and is best explained by the release of CHO.CHO and HCHO from the transformation of longer-lived precursors and/or efficient recycling processes in the plume as discussed above.

Summary and conclusions
The retrieval of CHO.CHO and HCHO VCDs from measurements of the TROPOMI instrument on board the Sentinel-5P satellite is reported. This will extend the datasets already available from the SCIAMACHY, GOME-2, and OMI instruments.
The advantage of the high spatial resolution and low noise of TROPOMI for studying specific geophysical phenomena is 375 well demonstrated in the features and plumes seen in the CHO.CHO, HCHO, CO, and NO 2 VCDs.
In this case study the satellite data show clear evidence for pyrogenic emissions of CHO.CHO and HCHO during the wildfire season in summer 2018 in British Columbia, Canada. The spatial and temporal pattern of the highest retrieved CHO.CHO and HCHO VCDs are associated with areas having high fire radiative power, as identified in the MODIS fire data products. This indicates that in these areas, pyrogenic emissions are the dominant source of CHO.CHO and HCHO. In 380 addition to local enhancements of NO 2 , CHO.CHO, and HCHO, close to the fires, extended plumes of elevated CHO.CHO and HCHO VCD are observed downwind from the fires. This is in contrast with the behaviour of NO 2 , which is also transported but is short lived. The spatial and temporal CHO.CHO and HCHO distribution observed from satellite follows a similar pattern to that of CO, which is long lived, and that simulated by the FLEXPART dispersion model, initialized by tracer emissions starting at known fire locations. Enhanced CHO.CHO and HCHO were found in the S5P data up to 1500 385 km from their sources.
In order to obtain reasonable agreement between the model results and the measurements, an effective tracer lifetime of more than 20 hours and up to 28.9 hours needs to be assumed in the FLEXPART simulations. This is significantly longer than the anticipated lifetimes of CHO.CHO and HCHO. The transport of CHO.CHO and HCHO along the length of the plume could be associated with these trace gases being lifted from the boundary layer into the free troposphere, where high wind speeds 390 lead to rapid transport. The long apparent lifetime of CHO.CHO and HCHO in the transported plumes could be explained by a real increase in their lifetime in the plume because of photochemical conditions in the plume, which we consider unexpected. Based on our current knowledge, the most probable explanation of the apparent long lifetime of CHO.CHO and HCHO would be formation within the plume caused by the oxidation of a mixture of longer-lived emitted VOC precursors (e.g. methanol, ethanol, acetylene, aromatics, glycolaldehyde, ethylene etc.), that form CHO.CHO and HCHO at different 395 rates. Further research is needed to investigate how frequent such fire-related long-range transport events of VOCs are. The chemical mechanism of the formation of the CHO.CHO and HCHO in the plumes downwind of the fires needs to be identified. The assessment of the number of such fires events and their relevance for tropospheric O 3 and aerosol production downwind of the fires and air quality is required.

19
VIIRS and CALIPSO data were obtained from the NASA Langley Research Center Atmospheric Science Data Center.
FLEXPART simulation results were generated using Copernicus Climate Change Service Information (ERA5). This is in part preparatory work for the analysis of the data from the DFG SPP HALO EMeRGe project, which has one focus on biomass burning and the long range transport of such plumes. We thank Abram Sanders (European Organisation for the Exploitation of Meteorological Satellites, Darmstadt) for providing support in the preparation of CALIPSO data for the 415 computation of CHO.CHO and HCHO AMFs.