A decade of global volcanic SO2 emissions measured from space

The global flux of sulfur dioxide (SO2) emitted by passive volcanic degassing is a key parameter that constrains the fluxes of other volcanic gases (including carbon dioxide, CO2) and toxic trace metals (e.g., mercury). It is also a required input for atmospheric chemistry and climate models, since it impacts the tropospheric burden of sulfate aerosol, a major climate-forcing species. Despite its significance, an inventory of passive volcanic degassing is very difficult to produce, due largely to the patchy spatial and temporal coverage of ground-based SO2 measurements. We report here the first volcanic SO2 emissions inventory derived from global, coincident satellite measurements, made by the Ozone Monitoring Instrument (OMI) on NASA’s Aura satellite in 2005–2015. The OMI measurements permit estimation of SO2 emissions from over 90 volcanoes, including new constraints on fluxes from Indonesia, Papua New Guinea, the Aleutian Islands, the Kuril Islands and Kamchatka. On average over the past decade, the volcanic SO2 sources consistently detected from space have discharged a total of ~63 kt/day SO2 during passive degassing, or ~23 ± 2 Tg/yr. We find that ~30% of the sources show significant decadal trends in SO2 emissions, with positive trends observed at multiple volcanoes in some regions including Vanuatu, southern Japan, Peru and Chile.

SO 2 sources over the course of a long-term (multi-decadal) satellite mission. The use of a single instrument permits relatively straightforward reprocessing of archived data as SO 2 retrieval algorithms improve, offering increasing sensitivity to volcanic SO 2 19 . Furthermore, unlike many spectroscopic instruments used for ground-based SO 2 measurements, satellite instruments such as OMI are also subject to intensive calibration and validation 20 .
Although satellites have been used to measure eruptive SO 2 emissions for several decades 3,8,21,22 , their use for quantification of passive volcanic degassing is relatively recent and concurrent with the advent of sufficiently sensitive space-borne instruments, such as OMI [23][24][25] . Previous application of OMI SO 2 data to detection of non-eruptive volcanic degassing has focused on the stronger SO 2 sources, detectable from space on a near-daily basis 3,16,25 . As recently demonstrated [26][27][28][29][30] , with specialized data processing techniques it is possible to enhance the sensitivity of ultraviolet (UV) satellite SO 2 measurements to enable detection of persistent anthropogenic SO 2 sources emitting on the order of 30 kilotons/year (kt/yr; equivalent to ~80 tons/day [t/d]), with the detection limit expected to be even lower for SO 2 sources located at high elevation (including many volcanoes). Here, we present a new global volcanic SO 2 emissions inventory derived from application of these techniques to more than a decade of OMI observations (2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015), which represents a timely replacement for existing databases 13,31 . We also compare the satellite-based SO 2 fluxes to a recent compilation of independent ground-based measurements 31 , and other sources, and examine the global distribution of volcanic SO 2 fluxes to reveal regional-and arc-scale trends in volcanic degassing.

Data and Methods
Volcanic SO 2 emissions were estimated using a new operational OMI planetary boundary layer (PBL) SO 2 column dataset produced using a principal component analysis (PCA) algorithm 32 . A detailed description of the techniques used to identify SO 2 sources (both anthropogenic and volcanic) and calculate emissions is given in ref. 29, and is briefly summarized here. The OMI PCA SO 2 data used in the analysis were restricted to 'clear sky' conditions by including only those OMI pixels with a cloud radiance fraction below 20%; solar zenith angles were also restricted to < 70° to reduce noise at high latitudes. In addition, all pixels affected by the OMI row anomaly data gap since 2007 (see: http://www.knmi.nl/omi/research/product/rowanomaly-background.php) were excluded. After pixel screening, an OMI pixel averaging or oversampling procedure 26,27 is used to resolve potential locations of SO 2 emissions and produce global maps similar to those shown in Fig. 1. To further enhance K ra k a ta u P a p a n d a y a n S la m e t M e ra p i B ro m o -S e m e ru Ij e n -R a u n g R in ja n i S a n g e a n g A p i P a lu w e h   3,10 . The Indonesian map also shows anthropogenic SO 2 sources in Singapore and central Sulawesi, but does not show volcanic SO 2 emissions from Sinabung, Rinjani and Sangeang Api, which first appeared after 2007. Maps were generated using Interactive Data Language (IDL) version 8.5.1 (http://www.harrisgeospatial.com/). the SO 2 signal and identify sources, a wind rotation technique is applied to align all the OMI SO 2 observations for each source along the same wind vector 30 , and then SO 2 emissions are estimated by fitting an exponentially modified Gaussian function to the OMI data 33 . The variable altitude of passive volcanic SO 2 plumes is accounted for by applying an air mass factor (AMF) correction to the OMI PBL SO 2 columns based on volcano altitude. To calculate accurate estimates of the SO 2 PVF, the effects of volcanic eruptions generating transient, large SO 2 column amounts are removed by applying a threshold SO 2 column amount of 5-15 Dobson Units (DU) to the OMI SO 2 data. This threshold was selected based on typical SO 2 column amounts measured by OMI in passive and eruptive volcanic plumes. However, we note that at some volcanoes it may be impossible to completely separate passive (i.e., involving no coincident eruption of magma) from eruptive SO 2 emissions, or even to establish which mode of degassing dominates at any given time. This is particularly problematic at volcanoes undergoing lava dome extrusion (e.g., Merapi, Indonesia; Soufriere Hills, Montserrat) or persistent Vulcanian or Strombolian activity (e.g., Stromboli, Italy; Fuego, Guatemala; Sakura-jima, Japan; Yasur, Vanuatu). Hence, while we believe that passive SO 2 degassing is the dominant process responsible for the emissions reported here, a contribution from eruptive degassing is inevitable at some volcanoes, as is the case for previous SO 2 emissions inventories 13,31 . Total uncertainties (including contributions from AMF, SO 2 mass, SO 2 lifetime, and wind speed uncertainty) on annual SO 2 flux estimates are 55% and > 67% for sources emitting more than 100 kt/yr and under 50 kt/yr, respectively 29 . Some of the largest individual sources of error are systematic and hence will introduce a bias in absolute SO 2 flux values but will not affect relative inter-annual flux variability (Fig. 2).
For the inventory presented here, volcanic SO 2 sources were identified based on 3-year averages of OMI data for 2005-2007, 2008-2010 and 2011-2014, then annual emissions were calculated for each source for the entire 11-year period studied (2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015). Note that the aforementioned 30 kt/yr (~80 t/d) detection limit was determined based on OMI observations of power plant SO 2 emissions in the eastern USA 28 , which are typically confined to the PBL. The higher altitude of volcanic SO 2 plumes translates into a higher AMF (greater sensitivity), which reduces the detection limit to values as low as ~6 kt/yr (~16 t/d). The detection limit will be lowest for low-latitude volcanoes, which benefit from more satellite observations under optimal conditions (e.g., low solar zenith angles). To assess the presence of significant decadal trends in the SO 2 emissions, we applied a weighted linear regression fit to the annual SO 2 emissions for each source, using the 1σ emission uncertainties (Supplementary Table S1) to weight the data, to derive a trend and linear correlation coefficient (r). Although it   Continued is possible to use satellite data to estimate SO 2 fluxes on much shorter timescales for strong sources 9,25 , the focus here is on long-term average emissions and trends rather than short-term variations. Future updates to the volcanic SO 2 emissions inventory will benefit from the recent release of new OMI PCA SO 2 products tailored to the variable injection height of volcanic plumes 19 , which should further reduce the uncertainties.

Results and Discussion
A total of 91 persistently degassing volcanic SO 2 sources have been detected in OMI measurements between 2005 and 2015 (Table 1; Supplementary Fig. S1). However, some of the detected SO 2 signals originate from paired sources (see below), so the actual number of volcanoes contributing to the detected SO 2 emissions is probably at least 100. For comparison, the Andres and Kasgnoc (1998) inventory 13 includes 49 continuously emitting sources. Since 3-year averages of OMI SO 2 data were used to identify the sources, the main criterion for detection is persistent emissions on that timescale. Hence it is possible that volcanoes exhibiting shorter-duration episodes of passive degassing may elude detection, but may subsequently be identified in more detailed analysis of shorter time periods. Table 1 lists the volcanic SO 2 sources, ranked according to their mean SO 2 flux for the entire 11-year period analyzed. Maps of the volcanic SO 2 sources are shown in Fig. 1 and Supplementary Figs S1-S8.   Table S1).
One of the disadvantages of UV satellite measurements is low spatial resolution, and as a result SO 2 emissions from clustered degassing volcanoes (within ~50 km) cannot be distinguished. Hence, some SO 2 emissions in the inventory are attributed to paired sources (e.g., Fig. 2), such as Nyiragongo-Nyamuragira (DR Congo), Bromo-Semeru (East Java, Indonesia) and Batu Tara -Lewotolo (Lesser Sunda Islands, Indonesia). Kamchatka (Russia) is another region where assignment of SO 2 emissions to specific volcanoes can be problematic (e.g., Mutnovsky -Gorely). Emissions reported for Chikurachki in the northern Kuril Islands may include a contribution from Ebeko (Table 1), where SO 2 emissions of ~100 t/d have been reported 34 . Resolving these merged SO 2 sources will require further field-based measurements in some regions, or the use of satellite data with higher spatial resolution 35 .
Notwithstanding some drawbacks, the strength of a satellite-derived emissions inventory is the global coverage. Most of the dominant sources (e.g., Ambrym, Kilauea, Bagana, Etna) are well established from prior measurements 14,16,36,37 . However, the OMI measurements (Table 1; Fig. 2) reveal, in some cases for the first time, significant, persistent SO 2 degassing at remote volcanoes in the South Sandwich Islands (Michael and Montagu), the Kuriles (Ketoi, Kudriavy), the Aleutians (Gareloi, Korovin), Indonesia (e.g., Dukono, Batu Tara -Lewotolo, Sirung, Ebulobo), and the southwest Pacific (e.g., Tofua, Tinakula). Gas emissions from Erebus (Antarctica) are also detected from space for the first time (Table 1; Supplementary Fig. S16). The OMI database thus provides what is the first truly global picture of contemporary volcanic SO 2 degassing, including sources where acquisition of frequent ground-based data will remain highly challenging.
Weighted linear regression reveals a range of temporal trends in the SO 2 fluxes ( Fig. 2; Supplementary Figs S9-S16). We acknowledge that a simple linear trend may not be applicable to many of the volcanic SO 2 sources (indicated by a low correlation coefficient, − 0.5 ≤ r ≤ 0.5; Fig. 2; Supplementary Figs S9-S16), but a detailed exploration of the trends in SO 2 emissions at each volcano is beyond the scope of this study. Nevertheless, the SO 2 data for some sources clearly indicate a long-term decline in SO 2 discharge (e.g., Miyakejima, Manam, Soufriere Hills; Fig. 2, Supplementary Fig. S9). A weak or insignificant trend in SO 2 emissions likely reflects relatively stable emissions (e.g., Bagana, Etna; Fig. 2), or more pulsatory degassing (e.g., Tavurvur, Anatahan, Huila; Fig. 2, Supplementary Figs S9 and S10); the latter could reflect cycles of magma intrusion followed by protracted gas release. Three of the top four sources feature active basaltic lava lakes (Ambrym, Kilauea and Nyiragongo-Nyamuragira), and in these cases the peak SO 2 discharge can be clearly linked to the establishment of new and/or larger lava lakes (e.g., at Kilauea in 2008 38 and Nyamuragira in 2012 35 ). The significance of the observed trends in SO 2 emissions is discussed further below.
In summing the SO 2 emissions from all detected sources, we find that the total annual SO 2 PVF is remarkably stable at 23.0 ± 2.3 Tg/yr (the highest annual total in the past decade was ~26 Tg in 2010). Andres and Kasgnoc (1998) 13 estimated a total non-eruptive volcanic SO 2 flux of ~12 Tg/yr for the 1970-1997 period (including a power-law extrapolation to estimate the contribution from unmeasured volcanoes); our higher estimate reflects the inclusion of more strong sources emitting > 1000 t/d SO 2 (Table 1). A comparison with eruptive SO 2 fluxes 3,10 confirms the common assumption that the SO 2 PVF is typically around an order of magnitude larger (Fig. 3), except during years with major SO 2 -rich eruptions such as at Bárðarbunga-Holuhraun (Iceland) in 2014 39 . The average total SO 2 PVF from all detectable sources is ~63 kt/day (2005-2015 mean; Table 1), which is broadly commensurate with a global SO 2 PVF of ~50.6 kt/day estimated by ref. 31 using a sparser dataset. Fluxes of SO 2 during large eruptions (e.g., Holuhraun 39 ) can greatly exceed the total PVF on short timescales.
The new volcanic SO 2 emissions inventory includes numerous previously unquantified sources. Based on SO 2 data reported in the literature (and we acknowledge that a substantial amount of SO 2 emissions data collected by volcano observatories may not be published), we find that 36 of the 91 sources (i.e., ~40%) have no previously reported SO 2 flux. The most prominent of these is Dukono (Halmahera, Indonesia), ranked 8 th in our inventory (Table 1  recent compilations 31 , 38 volcanoes (i.e., ~68% of the 56 volcanoes with prior measurements) have reported SO 2 fluxes within the 1σ fitting uncertainty of the OMI-derived fluxes. For ~41% of the sources with prior measurements, the OMI-derived SO 2 flux exceeds the independent estimate by at least 20%, and for ~36% the reverse is true (Fig. 2; Supplementary Figs S9-S16), whilst for the remainder (e.g., Ulawun, San Cristobal, Satsuma-Iwojima, Masaya, Fuego; Supplementary Figs S9-S16) the satellite-and ground-based SO 2 emission rates show excellent agreement (to within 20%). Nonetheless, it is notable that the OMI-derived SO 2 fluxes for most of the strongest sources are higher than previous estimates (Table 1; Fig. 2). For several sources (e.g., Ambrym, Bagana, Aoba, Manam) we believe that this is real and a result of infrequent prior measurements at these very active volcanoes coupled with significant variability in SO 2 emissions. Furthermore, at Kilauea, where a significant discrepancy is observed (Fig. 2), it has recently been shown 37 that ground-based techniques can underestimate SO 2 emissions by a factor of 2 or more in dense plumes. However, with the exception of the high-flux volcanoes, we observe no significant high or low bias in the OMI-derived SO 2 fluxes, but more detailed validation of the derived SO 2 emissions is certainly required.
In addition to Dukono, the new database sheds considerable light on the SO 2 flux from other Indonesian volcanoes (Fig. 1), which is noteworthy given the generally poor constraints on volcanic emissions in the archipelago 40,41 . Dukono (Halmahera) is the strongest volcanic SO 2 source with no prior constraints on its SO 2 flux (Figs 1 and 2). The SO 2 signal in the Sunda Strait near Krakatau volcano (Fig. 1) was previously assigned to the Suralaya power plant in Cilegon, West Java 29 , but we now assume this to be dominated by volcanic emissions from Krakatau (Table 1). SO 2 emissions of 190 ± 40 t/d were reported at Krakatau in 2014 42 , well above the satellite detection limit, so if this is a sustained SO 2 flux then it seems likely that most of the detected SO 2 is volcanic. The OMI-derived SO 2 flux for Krakatau is 303 ± 252 t/d (Table 1), i.e., within the range of ground-based measurements 42 . Degassing from Papandayan (West Java 40 ) may also be detected in the OMI data (Fig. 1), although it is difficult to isolate from the larger SO 2 signal associated with Slamet and hence is not treated as a separate source here. As noted earlier, several Indonesian volcanoes in East Java and the Lesser Sunda Islands are difficult to resolve using the OMI measurements, thus the reported emissions for Bromo and Semeru, Raung and Ijen, and Batu Tara and Lewotolo represent aggregated fluxes (Table 1). Ground-based SO 2 measurements in Indonesia are also increasing in frequency and coverage [40][41][42][43][44] . The OMI-derived average SO 2 flux from Bromo-Semeru (775 ± 298 t/d; Table 1) is higher than combined ground-based estimates for these volcanoes (~200 t/d 41,44 ), but the ground-based campaigns only cover a few days of degassing. It is also possible that the satellite measurements are more effective than ground-based techniques at constraining SO 2 flux at volcanoes that exhibit transitions from purely passive degassing to degassing via Vulcanian explosions (e.g., Semeru), due to the difficulty of measuring SO 2 in proximal ash-laden plumes 44 .
Another notable feature apparent in the map of Indonesian SO 2 sources is that some regions show lower emissions or an absence of subaerial SO 2 degassing, despite the presence of numerous Holocene volcanoes; e.g., southern Sumatra and the western Lesser Sunda Islands (Fig. 1). It is perhaps no coincidence that the latter region is the location of several volcanoes responsible for large SO 2 -rich explosive eruptions (linked to significant climate impacts 5 ) including Agung (1963) 45 , Samalas (1257) 46 and Tambora (1815) 47 . The identification of such degassing gaps, where stored gas may be accumulating in magma reservoirs rather than being released to the atmosphere, could assist hazard mitigation and identification of potential sites of future explosive eruptions. The mutually exclusive relationship between strong subaerial SO 2 degassing and large explosive eruptions during the past decade is also apparent in the Aleutian Islands (Fig. 1).
Further corroboration of the OMI-derived SO 2 emissions is possible based on data collected at Japanese volcanoes. A recent assessment 48 showed that 94% of the total volcanic SO 2 flux in Japan originates from 6 volcanoes: Tokachi, Asama, Aso, Sakurajima, Satsuma-Iwojima, and Suwanosejima; plus Mijake-jima after 2000. A total of 17 degassing volcanoes are documented in Japan 48 . OMI is able to detect all seven of the strongest sources (Table 1), yielding a time-averaged total SO 2 flux for Japan of 1.73 Tg/yr in 2005-2015, which is commensurate with a total SO 2 flux of 2.2 Tg/yr (including the intense degassing from Miyake-jima after 2000, which continues to subside) or 1.4 Tg/yr pre-2000 based on ground-based data 48 . Thus the OMI measurements represent an accurate estimate of total volcanic SO 2 emissions from Japan during the ongoing waning phase of Miyake-jima's degassing activity.
Examination of the frequency-flux relationship of volcanic SO 2 fluxes in Japan reveals that they do not fit a power law distribution 48 , as had been previously suggested for the global flux distribution 49 . A frequency-flux plot for the OMI-derived SO 2 emissions confirms that the global volcanic SO 2 sources also do not follow a power law distribution (Fig. 4). We also find a clear 'roll-off ' of the distribution at an SO 2 flux of ~500-600 t/d, remarkably similar to that found in the ground-based Japanese SO 2 flux data 48 . This important result shows that the distribution of volcanic SO 2 emissions on the scale of individual arcs can indeed mimic the global distribution, provided that large flux datasets are available from a range of source strengths (i.e., including very strong emitters such as Miyake-jima). It also indicates that the global volcanic SO 2 flux is dominated by the ~30 largest sources (Table 1; Fig. 4), and quantifying the flux from these volcanoes would provide a good estimate of the global SO 2 flux (in our database the 30 strongest sources emit ~80% of the total flux).
Arc-scale trends in volcanic degassing. Another significant application of the global satellite SO 2 measurements is the potential for detection of arc-scale trends in gas flux. Global, consistent SO 2 measurements such as the OMI-derived database presented here pave the way to new insights into arc-scale volcanic processes, including correlations between volcanic SO 2 emissions and other geophysical parameters such as arc length and subduction rate, since they provide a synoptic perspective on degassing that is not easily obtained from other techniques. The application of pattern recognition techniques to global SO 2 emissions data, such as the example in Fig. 5 (also see Supplementary Fig. S17), will permit an epidemiological approach whereby analogous degassing patterns may be identified at similar volcanic systems on regional or global scales. Interpretation of SO 2 data at individual volcanic systems can be ambiguous 50 , but analysis of arc-scale SO 2 measurements potentially allows the identification of correlated trends at multiple volcanoes that can be more confidently ascribed to similar volcanic processes.
The recent status of SO 2 emissions at the detected volcanic sources can be straightforwardly assessed by comparing the most recently measured annual mean SO 2 flux (for 2015) with the decadal mean flux (Supplementary Figs S1 and S17). This simple metric shows some notable arc-scale consistency in several regions; for example, all the detected volcanic SO 2 sources in Peru and Chile (Isluga, Villarrica, Lastarria, Ubinas, Copahue, and Sabancaya) have measured emissions in 2015 that are above the long-term average (Supplementary Figs S1 and S17). In southern Peru, both Ubinas and Sabancaya show particularly anomalous SO 2 emissions in 2015 (Table 1; Supplementary Figs S1 and S17), suggesting that these volcanoes are currently in a period of elevated activity. In contrast, the volcanoes of Papua New Guinea (Tavurvur, Langila, Bagana, Manam, and Ulawun) all show recent SO 2 emissions close to or below the decadal mean (Supplementary Figs S1 and S17).
A more rigorous evaluation of trends in SO 2 emissions must be restricted to those sources with annual SO 2 emissions showing a significant positive or negative linear correlation coefficient (i.e., r ≤ − 0.5 or r ≥ 0.5; Fig. 5). Using this criterion, 32 volcanoes show significant decadal trends in SO 2 emissions (Fig. 5), and although we highlight some potential arc-scale correlations here, further detailed analyses and other measurements are required to evaluate these findings. Trend analysis reveals that most volcanoes in the Vanuatu arc (Ambrym, Aoba and Yasur) show increased degassing in 2005-2015 (Fig. 5), and the only other detectable volcanic SO 2 source in Vanuatu (Gaua) also shows a positive trend but with a weaker correlation coefficient (r = 0.38; Supplementary Fig. S11). Both Ebulobo and Paluweh (Flores, Indonesia) show significant positive trends (Fig. 5) and are located in the same region of the Sunda arc (Fig. 1). In the Ryukyu Islands and Kyushu regions of Japan, SO 2 emissions from Satsuma-Iwojima, Sakura-jima, and Aso all show significant positive trends in 2005-2015 (Fig. 5), and the only other detected volcanic SO 2 source in this region (Suwanose-jima) also shows a positive trend with a lower correlation coefficient (r = 0.39; Supplementary Fig. S9). In addition, there is independent evidence for increased volcanic activity in the Ryukyu Islands and Kyushu region, including a significant eruption at Aso in October 2016, and elevated unrest at Sakura-jima 51 . A recent study 51 presents geophysical evidence for magma accumulation at Sakura-jima in the 1996-2007 period, with potential for a repeat of its 1914 Plinian eruption in ~25-30 years. The OMI SO 2 observations show a substantial increase in SO 2 degassing from Sakura-jima, particularly in 2011-13 (Fig. 5), indicating that the volcano was releasing more gas in this period largely via an increased frequency of vulcanian eruptions 51 . However, since 2013 the SO 2 emissions from Sakura-jima have declined below the decadal mean (Fig. 5), and so the future evolution of its activity is unclear. Nevertheless, the observed degassing over the past decade may have important implications for future activity at Sakura-jima. For example, the sustained release of SO 2 could be 'defusing' the potential climate impact of a future Plinian eruption, and/or could render a combined explosive-effusive eruption (such as the 1914 event) more likely due to limited gas supply. Gas overpressure and compressibility are rarely factored into models of volcano deformation 52 and the SO 2 emissions could also indicate a contribution to the deformation signal due to volatile overpressure in the magma reservoir.    In summary, while the correlated trends in SO 2 emissions observed in some arcs could be purely coincidental, possible links to underlying regional-or arc-scale geophysical processes (e.g., a coincident pulse in shallow magma supply) merit further investigation but cannot be confirmed on the basis of SO 2 emissions alone. Regardless of the underlying cause, our trend analysis (Fig. 5) provides new insight into the locations of increased volcanic SO 2 degassing over the past decade, which would be good targets for increased monitoring (if not already in place), and into volcanoes undergoing long-term decline.
Pre-eruptive volcanic degassing. Global satellite-based SO 2 surveillance also offers the potential for detection of pre-eruptive degassing at reawakening volcanoes. As noted above, increased SO 2 emissions at Aso (Japan) beginning in 2011 ( Fig. 5; Supplementary Fig. S10) preceded eruptions in 2014-2016 53 . SO 2 emissions were detected at Sarychev Peak (Kuril Islands) in 2005-2008 and showed a modest increase prior to its large eruption in June 2009 3 (Supplementary Fig. S13). At Alu-Dalafilla (Ethiopia), weak but detectable SO 2 emissions were present in 2005-2007 ( Supplementary Fig. S16) prior to an unexpected eruption in November 2008 3 . A shallow (~1 km deep) magma chamber has been identified at Alu-Dalafilla 54 , refilling after the 2008 eruption, which is a likely source of the pre-eruptive SO 2 emissions. Ground deformation data and the longevity of the magmatic system are consistent with the existence of a relatively thick sill 54 ; the persistent low SO 2 flux detected from 2005-2014 ( Supplementary Fig. S16) also supports this, although it is possible that some of the SO 2 detected by OMI may originate from nearby Erta ' Ale volcano. Continued analysis of global space-based SO 2 measurements will thus be valuable for volcanic hazard assessment, particularly at unmonitored volcanoes. Although the low temporal resolution of annual mean SO 2 emissions precludes timely identification of pre-eruptive unrest (unless it spans several years), one possible approach would be to calculate SO 2 emissions for all volcanic sources based on a 12-month moving average of satellite SO 2 measurements (or shorter for stronger sources). This would conserve the sensitivity of the technique to the weak SO 2 degassing expected in the initial stages of pre-eruptive unrest, whilst permitting more timely identification of increased emissions.
Missing sources and global volcanic CO 2 emissions. Inevitably, an undetermined number of weaker SO 2 sources, populating the tail of global SO 2 flux distribution (Fig. 4), are missing from the inventory. Continued ground-based SO 2 measurements at low-flux volcanoes 43,55,56 are required to constrain these sources. Such measurements are also needed to improve the relatively poor constraints on the component of global volcanic CO 2 emissions discharged in volcanic plumes 11 , which requires in-situ determination of the CO 2 /SO 2 ratio in the emissions. As shown by Fig. 4, through coordinated efforts such as the Deep Carbon Observatory (DCO; https:// deepcarbon.net/) 57 significant progress has been made towards improving the spatial coverage of CO 2 /SO 2 measurements, and around 50% of the detected SO 2 sources in Table 1 have characterized CO 2 /SO 2 ratios, including many of the strongest sources (Fig. 4), although the frequency of some measurements remains low. Based on our assessment, particular efforts should be made to pursue further CO 2 /SO 2 measurements in regions such as Indonesia, Papua New Guinea and Kamchatka, in order to improve constraints on the global volcanic CO 2 flux.

Conclusions
We believe that the volcanic SO 2 emissions inventory described here represents the most accurate assessment of contemporary global volcanic SO 2 degassing, and we encourage its use by the volcanological and atmospheric science communities as a substitute for existing databases 13,31 . Techniques such as this represent a major step forward in monitoring global volcanic degassing and ensure that few, if any, significant sources of volcanic SO 2 will remain undetected in the future, provided that satellite instruments with comparable sensitivity to OMI continue to be deployed (e.g., the Tropospheric Monitoring Instrument [TROPOMI], scheduled for launch on board the Copernicus Sentinel 5-Precursor satellite in 2017; http://www.tropomi.eu). Efforts to further characterize and validate the derived SO 2 emissions are strongly encouraged, particularly at those sources with no prior recorded measurements.
We have highlighted several potential applications of the new inventory, including the identification of regional-and arc-scale trends in SO 2 emissions, and improvement of constraints on global volcanic CO 2 emissions via measurement of CO 2 /SO 2 ratios (and their temporal variation) at sources where this information is currently lacking. Ongoing updates to the inventory will potentially provide opportunities to identify pre-eruptive degassing at reawakening volcanoes, and correlate SO 2 flux data with other geophysical data (e.g., ground deformation measured by InSAR) on a larger scale to elucidate volcanic processes. As a final point, the inventory demonstrates the remarkable persistence of passive volcanic degassing, and as anthropogenic SO 2 emissions continue to steadily decline, the volcanic contribution to atmospheric sulfur loading will inexorably increase.