Introduction

Volcanic emissions of sulfur dioxide can affect significantly atmospheric chemistry and global climate. SO2 is a precursor of sulfate aerosols, important for air quality1, and sulfuric acid, a compound known to alter local ecosystems, and which can cause damage to aircraft engines2,3. Injection of SO2 in the upper-troposphere and lower-stratosphere can lead to significant changes on global climate4 although the role of modest eruptions on volcanic forcing is only partly understood and is still an important subject of research5,6,7.

Measurement of volcanic SO2 degassing is also vitally important for volcano monitoring and to understand the underlying processes that can ultimately lead to an eruption8,9, or for following up an ongoing eruption10. In this respect, remote sensing of SO2 has been widely carried out from ground11,12,13, owing to the ease of SO2 detection, due to its strong absorption in the near UV and its low atmospheric background. Satellite nadir SO2 measurements with their global daily coverage provide complementary information for unmonitored volcanoes and explosive eruptions. Since the early measurements of SO2 from space in the late seventies14,15, advances in instrumentation and retrieval techniques, in both ultraviolet (UV) and thermal infrared (TIR) spectral ranges, have helped better understand volcanic processes, pre-eruptive signs or eruption (together with other geophysical data) and assess the impact of volcanic degassing on the atmosphere in general16. Global SO2 data has been increasingly recognized as a crucial data source and is now being used e.g. for near real-time monitoring of volcanic plumes17.

Satellite observations of volcanic SO2 emissions has mostly been used for estimation of total masses of SO2 emitted during explosive eruptions or from strongly degassing and/or high elevation volcanoes16,18,19,20,21, and to a lesser extent for the inversion of SO2 fluxes22,23,24,25,26,27. For passive degassing volcanoes, which dominate by large the time-averaged global volcanic emission28, space-based constraints on SO2 fluxes are more difficult to obtain. Typically, the SO2 concentrations are lower, the SO2 plume can be sub-pixel sized (especially near the source) and it is generally located in the lower troposphere where the measurement sensitivity is less favourable (and where rapid oxidation or wet deposition can also be a complicating factor). For these reasons, the state-of-the art in global SO2 emissions monitoring had been obtained using the Aura/Ozone Monitoring Instrument29 (OMI), as it combines the advantages of good sensitivity and selectivity to SO2 in the lower troposphere (compared to other sensors), and reasonably high nadir spatial resolution of 13 × 24 km2. Recently, improved methodology to estimate mean SO2 emissions from space measurements has been developed30 and applied to 10 years of OMI SO2 data over many degassing volcanoes, resulting in the most complete inventory of global volcanic SO2 annual emissions28 produced so far.

With a spatial footprint of 7 × 3.5 km² (13 times better than OMI, at least), the Sentinel-5 Precursor (S5P) TROPOMI instrument31 opens new possibilities in the surveillance of volcanic SO2 from space, and in quantifying robustly SO2 emission changes on shorter time intervals. SO2 clouds are detected globally and mapped with unprecedented detail, and even the weaker SO2 degassing plumes are measured nearly on a daily basis. Here, we present the first TROPOMI SO2 results and we examine and demonstrate several advantages of high spatial resolution data for volcano monitoring. We investigate the improvement in detection limit and the implication on the increased frequency of detecting weak emissions. We also investigate the potential of TROPOMI to provide high temporal resolution information on SO2 emission and compare the obtained results with operational ground-based SO2 flux data.

Data and Methods

TROPOMI and OMI SO2 vertical column data

SO2 data used in this study were retrieved from backscattered radiance measurements of TROPOMI using the S5P operational processing algorithm. The retrieval scheme is completely described elsewhere32,33, and is only briefly summarized here. The same algorithm was also applied to measurements obtained by the predecessor OMI sensor to produce a scientific dataset34 (different from the operational NASA SO2 product) for comparison with the presented TROPOMI data. Details on instruments characteristics and retrieval settings are provided in Tables S1 and S2 (Supplementary Material).

SO2 slant column densities, representing the effective optical-path integral of SO2 concentration, are retrieved by applying differential optical absorption spectroscopy35 (DOAS) to ultraviolet spectra using a combination of three fitting windows (312–326 nm: standard, 325–335 nm or 360–390 nm: alternatives to avoid signal saturation for high SO2). To cope with possible biases in the spectral retrieval step, a post-processing correction is then applied to produce so-called background corrected slant column densities (SCDs). The final results of the algorithm are the SO2 vertical column densities (VCDs), corresponding to the number of SO2 molecules in an atmospheric column per unit area (expressed hereafter in Dobson Units [DU] − 1 DU = 2.69 × 1016 molecules/cm²). SO2 VCDs are obtained from SCDs using conversion factors (air mass factors) that account for changes in measurement sensitivity due to observation geometry, total ozone absorption, clouds, and surface reflectivity. Since the measurement sensitivity also varies with altitude and the height of the emitted plumes is not known a-priori, the SO2 VCDs are calculated for three different hypothetical profiles of this gas, corresponding to 1 km thick boxes, at ground level and centered at 7 km and 15 km a.s.l. For illustration purpose, the SO2 VCD maps presented in this study correspond to the 7 km product (unless stated otherwise).

For this work, TROPOMI and OMI SO2 column data are used for the period from November 2017 to July 2018, for the days when both sensors were simultaneously measuring. The data over this measurement period were exploited to produce statistics on SO2 detection rate (hereafter referred as ‘SO2 detection frequency’) over certain volcanoes. For this, we apply a simple scheme independently to each satellite data set. First, a selection is performed on satellite pixels with cloud fraction <0.5, solar zenith angle <70° and central across-track positions (TROPOMI rows 50–400 and OMI rows 5–55). Moreover, we exclude OMI data affected by the row anomaly (http:// projects.knmi.nl/omi/research/product/rowanomaly-background.php). Secondly, all selected pixels for a given day and within a 75 km radius around a given volcano are considered (N in total). The pixels with SO2 SCD > 3 × SCDE (where SCDE is the uncertainty on the fitted SO2 slant column, typically of 0.3 DU for TROPOMI and 0.25 DU for OMI) are counted (Nc) and a detection is considered as plausible if Nc ≥2 and Nc > 0.04 × N. The latter criteria are necessary to avoid false detections and have been determined from statistics of the data over regions not affected by volcanic emissions. Finally, the SO2 detection frequency for a given volcano is simply the daily detection score divided by the total number of usable days for the measurement period.

TROPOMI-based SO2 flux estimation

Volcanic SO2 emission rates (expressed in kg/s) are estimated from daily maps of TROPOMI SO2 VCD applying the well-established traverse technique22,23,36,37. For this, a 3° × 3° square centered at a given volcano is used and the pixels with SO2 SCD >1 DU (and with at least one neighboring pixel satisfying the same criterion) are selected to delineate the plume for the analysis. After, the coordinates of the pixels considered are used to infer a mean plume direction, via a simple line fit in longitude-latitude axis. At this point, the method needs ancillary information of the SO2 plume height and wind speed (representative of the plume altitude), at the location of the volcano and at satellite overpass time. The plume height and wind data are described in the next section. The plume height is used to recalculate SO2 VCDs by linear interpolation of the three SO2 VCD products (0–1 km, 7 km and 15 km). The wind speed v is assumed to be constant over the length scale of the plume and does not account for local variations of the wind field. It is used to estimate, for each pixel, the time t elapsed since emission (plume age) which is approximated by t = d/v (where d is the distance in plume direction from the pixel center to the volcano). Then, for 0.5 h (Δt) plume age bins, total SO2 masses m are calculated by summing up their corresponding SO2 VCDs, and finally values for the SO2 flux (F = m/Δt) are derived, for 3 h long plume age intervals (hence six values of F in total, from ‘plume traverses’ at increasing distances from the volcano). This approach follows previous studies22,23 that have demonstrated the possibility to reconstruct the emission chronology up to several hours before the overpass time using a single satellite SO2 image. We estimate the total relative uncertainty of our SO2 flux estimates to be about 50%23, with a dominant contribution from the uncertainty on the wind.

SO2 flux from ground-based NOVAC observations

Ground-based data is obtained from scanning-DOAS instruments from the Network for Observation of Volcanic and Atmospheric Change (NOVAC) operated by PHIVOLCS/EOS (Mayon) and UNAM (Popocatépetl). NOVAC is a global network of ground-based remote sensors, providing routinely SO2 emissions from several degassing volcanoes. The detailed description of the instruments and analysis is given elsewhere12, and summarized in Tables S1 and S2. In essence, the method to determine SO2 flux is equivalent to the one described for TROPOMI.

Usually 2–3 scanners are installed around a given volcano (3–10 km distance) to cover most directions of plume transport. A full scan takes about 5–10 min to complete, comprising about 50 angular positions. SCDs of SO2 are derived from the measured UV spectra using DOAS (Table S2) and VCDs in the plume (relative to atmospheric background) are obtained assuming a straight-optical path through the plume. From the spatial distribution of VCDs, the total amount of molecules of SO2 in a cross section of the plume is obtained. This quantity can then be converted into fluxes by multiplication with the plume velocity. The standard relative uncertainty of a single SO2 flux measurement from NOVAC depends strongly on measurements conditions12, and is estimated to be within 30%.

The NOVAC data set provides high temporal resolution SO2 flux measurements and includes also ancillary data such as plume height and wind-speed. The latter data are used as input of the TROPOMI SO2 flux retrievals (as described above), to ensure consistency of TROPOMI and NOVAC comparisons. The plume direction and height can be estimated from NOVAC measurements by triangulation, using the angular distributions of SCDs obtained by simultaneous measurements of the different scanners. When only one instrument produces valid measurements at a given time, plume height is assumed to be equal to the difference between the station and volcano summit altitudes and plume direction is derived geometrically. Plume speed for this study is obtained from wind speed data of the NOAA Global Data Assimilation System (GDAS, 1 deg/3 h, accessed through https://www.ready.noaa.gov/) sampled at the coordinates of the volcano summit and interpolated at the measurement time.

Results and Discussion

Monitoring of large SO2 emissions

During the studied period, several eruptions produced spatially extended SO2 plumes that have been successfully detected by TROPOMI. In particular, the Ambae volcano (also known as Aoba) in the archipelago of Vanuatu, continuously emitted copious amounts of SO2. Figure 1 shows TROPOMI retrieved SO2 vertical columns above Vanuatu on 21 November 2017; the corresponding OMI SO2 VCD is shown in inset in Fig. 1 for comparison. The TROPOMI results reveal a stretched SO2 plume of ~1000 km length with remarkably more details than OMI SO2 columns. Owing to its improved spatial resolution, the TROPOMI data resolves small SO2 puffs and local maxima that are likely attributable to short-term variations in volcanic activity. The typical situation with OMI coarse spatial resolution is that the SO2 plume covers only a fraction of the pixel, resulting in reduced measured VCD. This is especially important close to the source where the plume horizontal size can be small. This dilution effect is much less pronounced with TROPOMI; the maximum VCD is 34.7 DU for TROPOMI, while it is only of 6.7 DU for OMI. Also note that signal saturation is probably responsible for some differences in the SO2 VCDs because the alternative fitting window 325–335 nm is more frequently used for TROPOMI than OMI, notably for the pixels close to the source with elevated SO2 columns. As a consequence, TROPOMI measurements are more representative of the volcanic source than OMI data.

Figure 1
figure 1

TROPOMI and OMI SO2 vertical columns over Vanuatu on November 21, 2017, with SO2 emission from Ambae volcano.

Figure 1 demonstrates the spatial resolution at which SO2 can be measured from TROPOMI. Considerable information on emission chronology and volcanic processes (highly relevant for volcanologists) can be extracted from measurements downwind. Ongoing efforts to reconstruct SO2 emission altitude and flux time-series with up to hourly resolution using inverse modeling approaches23,24,27,38,39,40 will likely benefit from TROPOMI SO2 measurements. Enhanced information on short-lived volcanic processes is expected from such analysis, as well as more robust forecasting of dispersion of volcanic clouds (of importance for aviation safety).

Detection of weak SO2 emissions

Whereas TROPOMI has drastically improved spatial resolution compared to OMI (by a factor of 13 better), we find that both sensors have similar spectral quality, e.g. in terms of spectral resolution and radiometric noise. This suggests that TROPOMI also outperforms OMI in detecting weak SO2 plumes typically residing in the lower troposphere.

It is enlightening to compare TROPOMI’s ability of detecting passive degassing plumes not only to OMI, but also to other existing (and past) sensors. Figure 2 compiles typical values of SO2 vertical column detection limit (at 3-σ level) for a tropospheric plume at 3 km height, as a function of pixel size (in km²) of a number of established nadir sensors (references are in caption of Fig. 2). UV measurements utilize SO2 absorption bands in the wavelength range around 310–340 nm and have good sensitivity to the lower troposphere. Figure 2 illustrates the progress of UV remote sensing of tropospheric SO2 since the first measurements performed by TOMS. Although SO2 retrieval algorithms have significantly evolved over the last years34,41, the increase in information on tropospheric SO2 is mainly due to improvement in spatial resolution of operational UV sensors (Fig. 2). TIR measurements of SO2 usually exploit absorption bands at 7.3 µm or 8.6 µm. High spatial resolution SO2 measurements are being carried out using thermal infrared imagers (such as ASTER and MODIS) but with limited sensitivity and accuracy to lower tropospheric SO2, as depicted in Fig. 2. The main reasons are interference by water vapor or volcanic ash (or other types of aerosols), lack of thermal contrast, dependence on variable surface emissivity and low spectral sampling. Better detection limit and selectivity to SO2 can be obtained from hyperspectral TIR instruments, like IASI and AIRS, but with larger footprint sizes. To better assess the different instruments, Fig. 2 also features iso-curves of SO2 mass detection limit (this quantity scales the SO2 VCD detection limit to a unit area of 1 km² and is independent of satellite footprint size). One can see that the ability of TROPOMI to detect volcanic degassing emissions is better than any other space sensor. The only exception is perhaps ASTER, but it is mostly because of its higher spatial resolution of 90 × 90 m². Moreover, ASTER does not allow for global daily coverage (as TROPOMI), and retrievals generally suffer at discriminating SO2 from other spectral features.

Figure 2
figure 2

SO2 VCD detection limit (at 3-σ level) for a tropospheric plume at 3 km height, as a function of pixel size (in km²) for space nadir sensors with proven capability to detected SO2 (values are adapted from the literature and personal communications). Pale orange points are used for thermal infrared imagers: ASTER44, VIIRS (detection limit presumably similar as MODIS), MODIS45,46, SEVIRI47. In blue, the IASI21,48 and AIRS18 thermal infrared hyperspectral sounders. Purple points are used for UV-VIS instruments: TOMS49, GOME50, SCIAMACHY51, GOME-252, OMPS53, OMI34,53, and TROPOMI32,33 (this study). The orange curves are the corresponding iso-lines of SO2 mass detection limit (t/km²).

Moreover, we find that TROPOMI has, due to its improved spatial resolution, about 4 times better SO2 mass detection limit relative to OMI. This is corroborated by Fig. 3, which shows TROPOMI retrieved SO2 vertical columns and corresponding OMI data over Indonesia, on November 29, 2017. On that day, two degassing volcanoes were particularly active in this region, Dukono and Karangetang. While TROPOMI unambiguously measures SO2 emissions from both volcanoes, OMI only detects emissions from Dukono (and with much less detail than TROPOMI). The small SO2 plume from Karangetang is undetected by OMI because of the coarse spatial resolution of the instrument. The example of Fig. 3 is often observed at other weakly emitting volcanoes worldwide (see Supplementary Material, Figs S13).

Figure 3
figure 3

TROPOMI and OMI SO2 vertical columns over Indonesia on November 29, 2017, with SO2 emission from Karengetang and Dukono volcanoes.

Based on global SO2 maps of averaged data over the measurement period (not shown), we have investigated all volcanoes reported in the recent volcanic SO2 emissions inventory28 and found that 58 degassing volcanoes were unambiguously detected in the TROPOMI measurements for the period from November 2017 to July 2018. Following the method described in section 2, we have calculated the frequency of daily SO2 detection for all 58 volcanoes. Figure 4 compares the results for OMI and TROPOMI data. While the presence of the OMI row anomaly data gap is likely to play a role in this picture, we find that TROPOMI is always detecting SO2 more frequently than OMI, as expected from improved spatial resolution. A striking feature of this analysis is that the SO2 detection frequency depends a lot on the volcano considered. For strong SO2 emitters (e.g., Popocatépetl, Ambae), SO2 columns are largely above the limit of detection, and both sensors have comparable performances in detecting SO2, with detection frequency values close to 1. Hence, for these cases, TROPOMI and OMI will detect SO2 on a nearly daily basis (and the advantage of TROPOMI resides mainly in providing more detailed data, as illustrated in Fig. 1). Conversely, for volcanoes with lower emission strengths, the situation is different, and we find that TROPOMI confidently detects SO2 plumes 2–4 times more frequently than OMI. For example, for the Karategang case of Fig. 3, TROPOMI detects SO2 about 3/5 of the time while it is detected rather occasionally by OMI. For volcanoes like Korovin and Tokachi, with emissions of 100–200 t SO2/d (according to the inventory)28, SO2 columns are below or close to the detection limit of TROPOMI and the detection frequency values are about 40% (albeit with large uncertainties). Those plumes are quite rarely detected (or not at all) in daily OMI SO2 images, and only appears in the global OMI inventory because of the pixels binning approach used28.

Figure 4
figure 4

SO2 daily detection frequency derived from TROPOMI and OMI data from November 2017 to July 2018, following the method described in section 2. The points correspond to the degassing volcanoes (based on the global inventory)28 active during this measurement period. Several volcanoes are identified by a color code. The dashed lines correspond to 1:1, 2:1 and 4:1 lines.

More work is needed to better quantify the performance of TROPOMI data as a function of SO2 emission flux but the conclusion of the analysis above is that monitoring of volcanic plumes nearly on a (clear-sky) daily basis is possible with TROPOMI, for an increasing number of (weakly emitting) volcanoes. This finding is an important step forward in monitoring volcanic degassing at global scale, and to improve budgets of global volcanic emission which is largely dominated by quiescently degassing activity28.

SO2 flux retrievals

As abovementioned, the higher-spatial resolution of TROPOMI combined with its increased sensitivity offer the possibility to study more frequently and robustly short-term changes in degassing patterns (at daily or sub-daily resolution), by extracting information on SO2 fluxes from the satellite SO2 imagery itself. Here we demonstrate the potential of TROPOMI to provide hourly to sub-hourly information on SO2 emission rates, hence at a frequency beyond the revisiting time of one day of the satellite. Let us consider a volcanic SO2 plume at a constant height being advected by a steady horizontal wind of 5 m/s oriented in the TROPOMI across-track direction i.e. a nearly zonal wind, which is a common situation for many degassing volcanoes. The across-track dimension of TROPOMI being 3.5 km, each satellite pixel is in principle representative for roughly 12 minutes of emission. Comparatively, OMI with an across-track dimension of at best 24 km has a time resolution limit of about 80 minutes. Noteworthy and regardless of time resolution considerations, another decisive advantage of TROPOMI for inferring SO2 fluxes compared to OMI, is that it resolves and represents much better the SO2 plume near the source, which is typically constrained to a narrow horizontal extent for passive degassing volcanoes.

To validate our approach, we have applied the technique outlined in section 2 to two volcanoes well monitored from the ground, Mayon (Philippines) and Popocatépetl (Mexico), and part of NOVAC. For both volcanoes, we selected the best scenes, i.e. clear-sky satellite overpasses for which TROPOMI detects well-shaped plumes originating from the volcanoes, meaning plumes that are narrow, non-bifurcated and following a main direction, due to stable winds. These conditions are required for validity of the SO2 flux retrieval method23. An example for Mayon on February 11, 2018 is given in Fig. 5a,b. One can see that the TROPOMI SO2 data downwind of the volcano allows the reconstruction of the SO2 emission rate (here at half-hourly sampling) up to three hours before the satellite time of observation (~5h30 UTC). This enables identifying short-term changes in SO2 degassing, as obvious from the excellent agreement between NOVAC and TROPOMI SO2 flux time-series (Fig. 5b). Such results can only be obtained from space using TROPOMI (owing to the combination of high-spatial resolution and high detection limit; Fig. 2).

Figure 5
figure 5

Panels a and b illustrate the TROPOMI SO2 flux retrieval technique for the case of Mayon (13.26°N, 123.68°E) on February 11, 2018. (Panel a) TROPOMI SO2 VCDs (DU) over the island of Luzon (Philippines), the Mayon volcano is symbolized by a blue triangle, the locations of the NOVAC instruments are marked by magenta circles. From the plume direction and wind speed at plume altitude, a plume age (expressed in hour) is assigned to each TROPOMI pixel. (Panel b) TROPOMI SO2 flux (kg/s) obtained at half-hourly sampling from the traverses downwind (hence back in time as illustrated by the gray arrow representing the plume age) and comparison with NOVAC SO2 time-series. (Panels c,d) SO2 fluxes retrieved from selected days of TROPOMI data for February-April 2018 period, for Mayon and Popocatépetl (19.02°N, 98.63°W) respectively, and estimations from NOVAC measurements. Mean SO2 fluxes and standard deviation are depicted for each day.

Note that further application of the SO2 flux retrieval technique to other cases is in general not as good as in Fig. 5b for the SO2 flux reconstruction. This reflects the limitations in the inversion techniques (both from space and ground) and probably also differences in air mass sampling. Nevertheless, for daily averaged values, the TROPOMI-derived SO2 emissions agree very well with the NOVAC estimates, as shown in Fig. 5c,d for a limited number of days (for the February-April 2018 period), for Mayon and Popocatépetl volcanoes, respectively. This corroborates and validates the TROPOMI SO2 results and the ability of the instrument to track changes in SO2 flux at (sub-)daily time scale42.

Conclusions

We have presented the first SO2 vertical column results retrieved from TROPOMI spectral measurements using the S5P operational processing algorithm. With a spatial footprint of 7 × 3.5 km² combined with excellent radiometric characteristics, TROPOMI profiles itself as the best hyperspectral UV imager in space, and we demonstrate in this paper the exceptional level of detail and sensitivity at which SO2 emitted by volcanoes can be measured. Compared to other satellite-based nadir sensors, TROPOMI is arguably the best satellite instrument for daily tracking of small emission plumes at a global scale. The limit of detection for SO2 emissions is a factor of 4 better with TROPOMI than with its predecessor OMI, and we showed that several volcanic SO2 sources that were detectable rather irregularly (and to some extent because of the OMI row anomaly) can now be monitored with TROPOMI on a near-daily basis. We have also investigated the potential of TROPOMI to provide information on volcanic SO2 flux by comparing the results to data from the NOVAC network, and found that day-to-day changes in SO2 emission are well reproduced and that –for the most favorable conditions– the inversion of SO2 flux is possible with a time-resolution of less than to an hour.

The TROPOMI SO2 product introduced here will constitute a key data source for global volcanic surveillance in the next years. In particular, for weakly emitting and poorly or non-monitored volcanoes, TROPOMI will provide a unique source of data for tracking changes in degassing patterns and, in combination with other types of data (e.g., on seismicity or ground deformation), for detecting early signs of eruption. Efforts to further validate and characterize the product quality at well-equipped volcanic sites are therefore critically needed. We anticipate that innovative techniques based on TROPOMI data will be developed in the future to estimate SO2 fluxes more systematically and at more volcanoes. This will improve the study of global volcanic processes and the interplay between volcanic emissions and the atmosphere, and open new possibilities for service applications.