Methanol (CH3OH) is, besides methane, the most abundant organic compound present in the atmosphere e.g., due to its fairly long atmospheric lifetime, of the order of 5 d and to its large global production dominated by an important biogenic emission flux, of magnitude (∼100Tgyr-1 globally) equivalent to 18 %–23 % of the global isoprene source e.g.. Other methanol sources include minor contributions from vegetation fires and anthropogenic emissions, each of the order of 10 Tgyr-1 globally e.g.; photochemical production (∼30–60 Tgyr-1, e.g.) from reactions of CH3O2 with organic peroxy radicals and with the hydroxyl radical OH e.g.; as well as a large and uncertain marine biospheric source of which global magnitude 24–85 Tgyr-1, e.g. is more than offset by ocean uptake (38–101 Tgyr-1).

The importance of methanol for atmospheric chemistry stems primarily from its main atmospheric sink, namely oxidation by OH e.g., which is an important source of carbon monoxide and formaldehyde and has minor impacts on tropospheric ozone and the oxidizing capacity of the atmosphere . Methanol is also removed from the atmosphere through wet scavenging and uptake by oceans (see above) and vegetated areas .

The terrestrial biogenic emission of methanol is primarily associated with the growth of cell walls in plant leaves , while other processes such as grassland cutting and plant decay also contribute but are considered minor. The emissions are dependent on leaf temperature and light, and are higher in young and growing leaves than in mature and senescent leaves . The estimated global biogenic source of methanol, of the order of 100 Tgyr-1 according to the Model of Emissions of Gases and Aerosols from Nature version 2.1 (MEGANv2.1) , agrees with top-down estimates constrained by in situ (primarily airborne) measurements or spaceborne retrievals of CH3OH columns from the Infrared Atmospheric Sounding Interferometer IASI,. These results are also consistent with the total terrestrial surface source of methanol (∼120Tgyr-1) estimated based on column data from the Tropospheric Emission Spectrometer TES,. Nevertheless, the confrontation of models with satellite data suggest substantial deviations from the MEGANv2.1 distributions, such as large underestimations over semi-arid regions (shrubland and savannas), overestimations over rainforests over Central Africa and parts of Amazonia, and a shift of the seasonal peak of biogenic emissions towards the spring at mid-latitudes .

Top-down emission estimates based on satellite data bear uncertainties for several reasons. Firstly, although the IASI- and TES-based inverse modelling of emissions improved model comparisons against independent data , significant underestimations persisted in comparisons with aircraft and ground-based measurements, suggesting potential biases in the satellite data. Recent studies showed that satellite retrievals may present biases with respect to independent datasets, e.g. for HCHO from UV-Visible sensors e.g. and for several organic compounds from IR sensors including acetone and carboxylic acids from IASI and methanol and other species from the Cross-track Infrared Sounder CrIS,. The characterisation of satellite data biases can be used to derive bias-corrected datasets for use in inverse modelling, as has been done for HCHO and NO2 .

Secondly, potential inconsistencies between the vertical concentration profile from the model and assumed in the satellite retrieval might lead to biases in the comparison of total columns, due to vertical variations in the sensitivity of the chemical compound. This issue can be addressed through the application of averaging kernels , which were however not available in previous works based on methanol IASI retrievals.

Thirdly, the inverse modelling of terrestrial methanol emissions is sensitive to the representation of other key budget components. Although very uncertain, marine emissions have little relevance due to their very small impact over land . Of higher importance is the production due to the CH3O2+OH reaction, which was ignored in the earlier studies, including and . Most importantly, the parametrisation of methanol uptake on land surfaces was based on only few dry deposition data, despite the well-established bidirectional nature of biosphere/atmosphere exchange of methanol . The adopted dry deposition velocities, typically below 0.6 cms-1 , are significantly lower than average values reported in many field campaigns . In some cases, the net methanol flux to the atmosphere is close to zero or even negative during a large part of the year .

The present study aims to address the above issues in several ways. We present a newly developed version of the IASI CH3OH retrieval, IASIv4, including several methodological advances, among which the provision of averaging kernels. Next, we validate this product using aircraft measurements of methanol from several campaigns, and we use the results to propose a correction formula. The bias-corrected IASI dataset is then used to optimise terrestrial emissions in the global chemistry-transport model MAGRITTE . This model incorporates methanol formation due to CH3O2+OH, as well as a detailed representation of methanol uptake. The deposition scheme over land is adjusted based on field campaign data. Finally, the optimisations are evaluated against a broad range of independent observations, including surface and airborne in situ data as well as Fourier-transform infrared (FTIR) column measurements.

The manuscript is structured as follows. Sections – describe the IASIv4 retrieval, the airborne and surface in situ concentration datasets and the network of FTIR data. Section  provides a brief description of the MAGRITTE model focusing on the parametrisation of methanol sources and sinks; in particular, Sect.  and Appendix  describe the implementation of methanol dry deposition, including an evaluation of this scheme against observation-based estimates. Sections  and present the methodology used for IASI validation and for emission optimisation based on either aircraft or satellite data. Section  presents the evaluation of IASI biases using aircraft data, and proposes a bias-correction formula for use in inverse modelling. Section  presents an assessment of top-down terrestrial emissions based on IASI, while Sect.  provides an evaluation of the optimised results against independent data; finally, Sect. 6 presents the conclusions of this study.

Here we evaluate IASI against aircraft data, using MAGRITTE as transfer standard. Figure  illustrates the geographical distribution of vertically-averaged CH3OH mixing ratios from the three campaigns. By far the highest values were observed during SENEX, largely because of the higher proportion of low-altitude measurements in this campaign (72 % below 1.5 km) compared to DC3 (15 %) and SEAC⁴RS (35 %). The vertical distribution of methanol (Fig. ) shows indeed a maximum (ca. 4–7 ppbv) in the boundary layer, and a substantial decline in the free troposphere, down to 1–2 ppbv above 6 km, a feature well reproduced by the model. However, the simulation using a priori emissions underestimates the observations by ∼20%–50 % during SENEX and DC3. During SEAC⁴RS, model overestimations are seen over the southeast, and underestimations elsewhere. The largest underestimations are found over the US. midwest, reaching a factor of ∼3 during DC3 and SEAC⁴RS (Fig. ). Similar, or even larger underestimations were obtained in previous model evaluations against aircraft campaigns over western US .

The optimised model using adjusted CH3OH emissions reproduces very well the observations (Figs.  and ), with spatial correlation coefficients of 0.97–0.98 for all campaigns, and negative biases of 1 %–3 % for SENEX and SEAC⁴RS, and 7.5 % for DC3. This agreement is achieved through a substantial increase of summertime methanol emissions over the western US, reaching factors of 2–3 between western Texas and Wyoming (Fig. S4). Small decreases are inferred over large parts of eastern US. Since the emission parameters are under-constrained by the inversion due to the poor coverage of the observations, the optimised emissions have limited reliability and are strongly dependent on the a priori inventories and inversion setup. Nevertheless, the excellent agreement of the optimised model with not only the observational datasets used as constraint in the inversion (Fig. a, c, and d), but also with the TOGA CH3OH measurements on board the GV aircraft during the DC3 campaign (Fig. b) demonstrates that the optimisation successfully derived a methanol distribution closely reproducing the airborne observations.

The linear regression of the observed and simulated concentrations yields a slope of almost 1 (0.98) and a correlation coefficient of 0.98. However, the comparison of IASI and co-located aircraft-constrained model columns (Fig. ) shows significant biases. High IASI columns (>∼25×1015molec.cm-2) are underestimated by up to a factor of ∼1.4. This underestimation of high columns is consistent across the three campaigns. The statistics of the comparison are improved when the averaging kernels are applied to the model profiles: in particular, the correlation coefficient increases from ∼0.81 to ∼0.85.

An ordinary linear regression of IASI and aircraft-constrained model columns yields 17ΩIASI=0.46Ωairc+10.6×1015, where ΩIASI and Ωairc are the CH3OH columns (molec.cm-2) from IASI and from the aircraft-constrained model simulation, respectively. The 1–σ uncertainty is 0.03 for the slope and 1.1×1015 molec.cm-2 for the intercept. The regression suggests a moderate overestimation of IASI columns in the range (15–20)×1015 molec.cm-2, although the data is too limited to draw firm conclusions. Below that range, the bias remains uncharacterised by the aircraft data used in this study.

The reasons for the IASI biases with respect to aircraft in situ data and for their dependence on the magnitude of the columns are yet unclear. Qualitatively similar biases were derived from the evaluation of OMI and TROPOMI CH2O columns against aircraft and FTIR data . The estimated in situ measurement uncertainties are clearly too low (∼20%, see Sect. ) to fully account for the biases derived above, although they could contribute; furthermore, the model biases against the PTR-Q-MS data of the DC3 campaign are validated by the good consistency between the model evaluation against PTR-Q-MS (DC8) and TOGA (GV) measurements from this campaign (Fig. a and b). We verified that the above results are only minimally affected by the filtering of urban and pyrogenic plumes mentioned in Sect. . Without these filters, the slope (0.45) and intercept (10.6×1015) of the above relationship are essentially unchanged. Evaluation against measurements in other regions and using other techniques would be needed to confirm and refine the biases derived in this work.

Here, we derive top-down methanol emissions based on bias-corrected IASI columns (ΩIASI,BC) calculated (see Eq. ) with 18ΩIASI,BC=(ΩIASI-10.6×1015)/0.46.

Figure  displays the seasonally averaged CH3OH columns from IASI (bias-corrected), the a priori model simulation and the IASI-based emission optimisation. The seasonal cycle of the columns over large regions is shown on Fig. S5. The a priori model succeeds in reproducing the general features of the satellite observations, such as high columns (∼40–90×1015molec.cm-2s-1) throughout the year over tropical continents, and a pronounced summertime peak at extratropical latitudes, consistent with previous spaceborne methanol distributions . Both the a priori model and the IASI data display a substantial longitudinal gradient of methanol columns over northern Eurasia during summer, with low values over western Europe and a broad maximum over eastern Siberia. There are also important differences between IASI and the a priori model, most notably a large model underestimation at extratropical northern latitudes during all seasons, reaching a factor of about 2 over Central Asia, Siberia and Canada during summer, and an overestimation of the columns over Amazonia near the end of the wet season (May–July, see Fig. S5). Furthermore, although the a priori model columns peak at the same month as the satellite data at mid-latitudes (most often July), the model underestimations are more pronounced during spring and early summer (i.e. May–July) than in the following months (August–October), in particular over the US, China and Europe (Fig. S5).

The emission optimisation successfully closes the gap between the model and the observations, in particular over tropical regions and at extratropical latitudes during summer (Figs.  and S5). During winter at high northern latitudes, however, the large a priori model underestimation remains unchanged after optimisation. This is explained by the weakness of methanol emissions and by the low number of IASI measurements used in the optimisation at these latitudes during winter, compared to other latitudes and seasons (Fig. S6). Similar results were obtained by in their emission optimisation based on CH3OH column data from TES. Interestingly, although the focus of our study is on continental areas, the agreement of MAGRITTE methanol columns with IASI is also substantially improved over oceanic areas (Fig. S5p–s) after inversion, especially at extratropical latitudes (except in winter).

This improved agreement with IASI data is primarily achieved through changes in the distribution of biogenic methanol emissions (Fig. ). The biogenic emissions are strongly enhanced over North America and most of Eurasia after inversion, while biogenic emissions due to tropical forests are generally decreased, in particular over Amazonia and Indonesia, and emissions due to tropical savanna over Africa, Australia and eastern Brazil are increased (Fig. ). As in previous inversion studies , the strongest enhancements (up to a factor 5) are derived over arid and semi-arid landscapes such as Central Asia, Western US and the Sahel region (Figs.  and ). This underestimation might partly result from the neglect of soil emissions in MEGAN. Soils (including litter decomposition) are indeed a known methanol source , and although their contribution is generally considered to be small, typically 1–2 orders of magnitude lower than foliage emissions , they might be more significant over sparsely vegetated areas characterised by low LAI.

The biogenic emission enhancement at mid-latitudes is highest in spring (Fig. S7), and especially in May (Fig. ). The underestimation of springtime emissions was previously noted by e.g. . The resulting top-down biogenic emissions peak earlier than in the MEGAN inventory, in particular over Europe, Eastern US, China and Central Asia. Boreal regions do not follow this trend, with emission enhancements of similar magnitudes being derived over spring, summer and fall over these regions (Fig. S7). However, the large emission increase during fall (and to a lesser extent during summer) inferred over boreal forests is partly explained by the strong deposition sink (Fig. S3). Since the deposition velocities might be overestimated over boreal forests (Sect. ), the top-down emissions might be also too high, especially during fall. In fact, in spite of the large emission enhancement derived over Siberia, the net emission flux over this region (Fig. ) is lower than the a priori (MEGAN) gross flux.

The seasonal cycle of terrestrial emissions undergoes important changes after optimisation over tropical ecosystems (Fig. ). Over both Northern Hemisphere (NH) Africa and southern Hemisphere (SH) Africa, the biogenic emissions are decreased (by ∼25 %) at the start of the biomass burning season (November–December in NH, June–July in SH), while these emissions are strongly enhanced (by up to 70 %–100 %) in the following months (February–June in NH, August–January in SH), until after the end of the burning season. The optimisation also shifts by one month the seasonal peak of pyrogenic emission over SH Africa (from July in the a priori to August in the optimisation, see Fig. ), although, as explained above (Sect. ), the dominance of the biogenic flux makes the top-down results uncertain for biomass burning emissions.

The top-down biogenic emissions over tropical ecosystems are strongly correlated with temperature and especially solar radiation. Over each of the 5 tropical regions shown on Fig. f–g, the two least-emitting months according to the inversion are the months with the lowest visible radiation fluxes, based on the ERA5 reanalysis. For example, over Amazonia, the lowest monthly biogenic fluxes (0.54 and 0.61 Tgmonth-1, about a factor of two below the annual average) are derived in May and June, which are the months with the lowest visible radiation fluxes (∼85Wm-2, 12 % below the annual average). The same holds for S.-E. Asia and NH Africa (panels f–g), for which the minimum occurs in November–December, and for Equatorial and Southern Africa (h–i), which have their minimum in June–July. At all 5 tropical regions, the top-down monthly biogenic emissions correlate strongly with solar visible radiation fluxes, with Pearson's correlation coefficients ranging between 0.79 (Amazonia) and 0.94 (SH Africa). A strong correlation is also found between biogenic emissions and near-surface temperature over NH Africa (0.92) and SH Africa (0.84). At the other regions (panels f, h, and j), the temperature variations are weak (standard deviation of ∼0.6K) and therefore likely less relevant for biogenic emission variability.

Radiation fluxes and temperature appear to exert a stronger control on biogenic emissions of methanol than is currently accounted for in MEGAN. This control is likely indirect, i.e. phenological changes associated with the seasonal cycle of meteorological variables likely cause variations in the emissions that are currently not represented in the model parametrisations. Over Amazonia, leaf flushing during the wet-to-dry transition period has been suggested to explain a strong reduction of isoprene emissions around May every year , and was also proposed to decrease methanol emissions in July . The growth of new leaves after the wet-to-dry transition period might cause an enhancement of methanol emissions, since young leaves are known to emit at higher rates than mature leaves. However, the MODIS LAI dataset indicates only a moderate and progressive increase of LAI during this period, from ∼4.2 to ∼5.2m2m-2 between February and September. Since the parametrisation of the leaf age response factor in MEGAN (γage) relies on the temporal variation of LAI between time steps, the proportion of new or growing leaves calculated in this way is very small, and γage is close to unity. More work is needed to understand the impact of phenological changes on methanol emissions, and how these changes can be represented in emission models.

The global top-down biogenic emission flux is 160 Tgyr-1, i.e. 23 % higher than our a priori from MEGAN (130 Tgyr-1), and almost 60 % higher than previous top-down estimates based on in situ data or spaceborne IASI columns (Table ). The total terrestrial emissions, amounting to 178 Tgyr-1 globally, are also 46 % higher than the top-down best estimate of 122 Tgyr-1 based on TES column retrievals . The optimisation leads to very small changes in the anthropogenic and pyrogenic emission categories, not exceeding a few percent at the global scale (Table ).

Despite the large enhancement of methanol emissions inferred in this study, the global atmospheric burden of methanol, 3.4 Tg in our optimisation, is only slightly higher (by 9 %–17 %) than in previous modelling studies constrained by observations (Table ). The larger methanol loading (by 17 %) compared to the IASI-based inversion by is largely due to the bias correction of IASI data (Eq. ), leading to column increases of the order of 30 % over source regions (Sect. ). The main reason for the larger enhancement of terrestrial emissions, compared to previous inversion studies, is the sink due to dry deposition over land, 72 Tgyr-1 globally, about a factor of 2.6 larger than in the inversion studies of and , but very close to a global sink estimate based on in situ CH3OH measurements and a dry deposition velocity of 0.4 cms-1 (70 Tgyr-1, ). The global lifetime of atmospheric methanol with respect to dry deposition over land is ∼17 d, well below the range of reported values, 26–38 d . Due to this strong sink, the net terrestrial source of methanol inferred here is only slightly (+10 %) larger than in the inversion studies of and . As seen on Fig. , the gross top-down emission fluxes over forested regions such as Amazonia, Equatorial Africa and Siberia (14.7, 7.3, and 9.4 Tgyr-1, respectively) are up to a factor of 3 higher than the net surface fluxes accounting for dry deposition (5.4, 2.6, and 3.2 Tgyr-1). Over less productive ecosystems, including regions with large biomass burning fluxes, the gross and net fluxes are more similar, but still significantly different, e.g. by factors of 1.4–1.6 for Central Asia, North Africa and South Africa. At global scale, dry deposition over land offsets about 45 % of the biogenic emission flux (Table ).

The deposition velocities computed in this work, typically between 0.2–1.6 cms-1 over vegetated areas (Fig. S3) are well-supported by measurement-based estimates (Fig. ), except for a large overestimation at a boreal forest site (Hyytiälä) (Sect. ). The large dry deposition sink inferred by the model over boreal forests is therefore likely overestimated, and the biogenic emission enhancement at high latitudes (Fig. ) might also be too high. Elsewhere, however, the strong dry deposition sink is consistent with available data. Over the tropical forests of Amazonia, Central Africa, Indonesia and southeast Asia, and even over Europe and eastern US, dry deposition is found to be a stronger sink of methanol than chemical oxidation due to reaction with OH (Fig. ).

The optimised methanol budget presented in Table  bears uncertainties due to potential errors in the IASI data used as constraints and because, while terrestrial emissions are optimised, the other productions and the sinks of methanol have their own uncertainties. In particular, oceanic emissions depend on assumed seawater methanol concentrations for which available field campaign data show a very strong variability . Replacing the seawater concentration adopted in the model (Sect. ) by the value of 61 nmolL-1 determined by based on an analysis of ATom data would decrease the oceanic emission flux from 47 Tgyr-1 globally to 24 Tgyr-1, in excellent agreement with . The photochemical production of methanol due to the CH3O2+OH reaction is also uncertain; for example, adoption of a fixed methanol yield of 13 % from the reaction , in place of the current model assumptions (Sect. ), would increase the global CH3OH production by ∼5Tgyr-1. However, the impact of these uncertainties on the optimisation of continental emissions is very small. A sensitivity inversion performed for one year (2017) for which the marine source and the methanol yield from CH3O2+OH both follow the recommendations of leads to negligible impacts on top-down terrestrial emissions (+1.1% compared to the standard inversion) and on dry deposition fluxes over land (+0.4%), in spite of more sizeable impacts on the global CH3OH burden (-4.9%, to 3.25 Tgyr-1) and on global oceanic uptake (-19%, to 4.9 Tgyr-1).

Twelve years of IASIv4 global methanol column data are used in an inverse modelling framework built on the MAGRITTEv1.2 model to propose an updated assessment of CH3OH global distribution and terrestrial emissions. The IASIv4 dataset is generated using the ANNI version 4 retrieval framework, which incorporates several methodological advances compared to previous versions. In particular, the dataset includes total-column averaging kernels, essential to minimise the impact of vertical-profile differences in the comparisons between IASI retrievals and MAGRITTE outputs.

In a first step, in situ methanol observations from three extensive aircraft campaigns over the US (DC3, SENEX and SEAC⁴RS) are assimilated into the model to derive aircraft-constrained model distributions used to evaluate the IASI columns. The results suggest an underestimation of large IASI columns, reaching a factor of 1.41 for IASI columns of ∼30×1015molec.cm-2. The in situ measurement uncertainties are too low to account for these biases, which therefore remain unexplained.

The bias of IASI with respect to aircraft data is tentatively corrected through a linear relationship, and the bias-corrected IASI columns are used as constraints to optimise the terrestrial CH3OH emissions in the MAGRITTE global model over 2008–2019. Model evaluation against nine aircraft datasets spanning 2008–2018 shows that the emission optimisation leads to a large reduction of the average bias against aircraft observations over land, from -23% in the prior simulation to -8.4% in the optimisation; the model agreement is also improved over oceans. Similarly, the model performance against a broad compilation of surface in situ data (67 campaigns) is greatly improved, as seen from the resulting high correlation and low biases globally and regionally (less than 20 % bias over tropical forests, US, Europe, East Asia and marine areas). The model performance (bias and RMSD) is also improved at seven of the eight FTIR stations.

Nevertheless, closer examination of the comparisons points to important regional differences. Most noticeably, the optimisation leads to substantial model overestimations (by 40 %) against summertime measurements over the Canadian boreal forest (ARCTAS-July campaign) and in northern European Russia (St Petersburg), suggesting that the bias correction of IASI columns is unwarranted at these latitudes. Over tropical ecosystems, the comparisons with in situ data (10 campaigns) and FTIR data (at Porto Velho and Reunion island) suggests a small positive bias (∼15%), despite a substantial reduction of biases, in comparison to the prior simulation. Future work should aim at a better characterisation of IASI biases using aircraft and surface in situ data (especially over boreal and tropical ecosystems, poorly represented in the present study) and FTIR data (in all environments), considering the small number of stations where methanol is being retrieved.

The emission inversion suggests largely increased biogenic emissions over North America and most of Eurasia as well as decreased emissions over tropical forests. Strong enhancements, by up to a factor 5, are found over semi-arid ecosystems, consistent with previous inversion studies and possibly due to soil emissions currently overlooked in MEGAN. The seasonal cycle of biogenic emissions undergoes significant changes. At mid-latitudes, the optimised emissions peak earlier than in the MEGAN inventory. Over tropical ecosystems, emission increases are inferred during warm and sunny periods, while decreases are derived during colder, less sunny months. Temperature and visible radiation fluxes appear to exert a stronger control of biogenic emissions than can be accounted for in MEGAN, for reasons still unclear. A revision of the parametrisation of the leaf age response factor is likely needed for tropical environments.

The global top-down biogenic emission flux (160 Tgyr-1) is almost 60 % higher than previous top-down estimates , due to mainly two reasons. The first reason is the bias correction of IASI columns, corroborated by the improved model performance against a wide range of observations, except over boreal continental regions, as noted above. The total biogenic flux due to boreal forests is increased by a factor of 2.4, from 9.4 to 22.8 Tgyr-1. Even without the contribution of boreal forests, the global top-down biogenic flux would therefore still be much higher than previous estimates. The second reason is the stronger sink due to dry deposition in our model, with a global lifetime with respect to this process of 17 d, well below the range of estimates from previous modelling studies. Dry deposition is estimated here to offset 45 % of the global biogenic emission flux. The deposition velocities are calculated using a Wesely-type parametrisation adjusted based on estimates from 13 field campaign studies. The calculated values range typically between 0.2–1.6 cms-1 over vegetated areas, in generally good agreement with the field studies. A notable exception is the boreal forest site of Hyytiälä, where the deposition velocity is largely overestimated. Therefore, the dry deposition sink (and also the top-down biogenic gross flux) might be similarly overestimated over these forests. Clearly, more field campaign data are needed to provide a better assessment of both methanol abundances and dry deposition velocities in this environment, and more generally over terrestrial ecosystems.