Volcanic sulfur inventory for CMIP7 charts 274 years of eruptions

When a large volcano erupts, sulfur dioxide lofted into the stratosphere converts into sulfate aerosol particles, making volcanic sulfur one of the most important natural drivers of climate variability. Capturing that forcing in climate models has always depended on how well scientists can reconstruct what past eruptions actually injected into the upper atmosphere. A team led by Thomas Aubry of the University of Oxford and the University of Exeter has now delivered a new inventory of upper-tropospheric and stratospheric volcanic sulfur dioxide emissions covering 1750 to 2023, built specifically for Phase 7 of the Coupled Model Intercomparison Project, or CMIP7, the benchmark simulation effort that underpins upcoming international climate assessments. The inventory, published in Geoscientific Model Development, catalogs 463 explosive eruptions that together injected 428.2 teragrams of sulfur dioxide, an average flux of 1.56 teragrams per year.

The dataset matters because two different classes of climate models need it. Most models cannot simulate stratospheric aerosols interactively from precursor emissions, so they rely on prescribed aerosol optical properties, which the team’s companion paper derives directly from this emission inventory for the pre-satellite period of 1750 to 1978. A growing number of models, however, do simulate stratospheric aerosol interactively, and for them the emissions themselves are the required input. Those interactive schemes also allow researchers to probe the controls on volcanic radiative forcing, to evaluate proposals for solar radiation modification through deliberate stratospheric aerosol injection, and to predict how increasingly intense wildfires and their smoke-driven thunderclouds might alter stratospheric composition and Earth’s radiative balance.

Building a consistent record across nearly three centuries required stitching together fundamentally different data sources. Since October 1978, when NASA’s Nimbus-7 Total Ozone Mapping Spectrometer began daily ultraviolet measurements, satellite sensors including TOMS, the Ozone Monitoring Instrument, the Ozone Mapping and Profiler Suite and TROPOMI have quantified volcanic sulfur dioxide, with infrared sensors filling in high-latitude eruptions during polar winter when ultraviolet data are unavailable. The team used the Multi-Satellite Volcanic Sulfur Dioxide L4 database, which records the mass, injection altitude, location and date of every eruption detectable from orbit. Before 1979, the record rests on polar ice cores: the bipolar arrays of Sigl and colleagues and the eVolv2k product, which detect eruptions injecting roughly 10 teragrams of sulfur dioxide or more, supplemented by a high-resolution Greenland ice core known as D4i and by the Smithsonian Global Volcanism Program’s geological database, which together support the inclusion of much smaller eruptions injecting on the order of 0.1 to 1 teragram.

That focus on small-to-moderate eruptions, defined here as those injecting 3 teragrams of sulfur dioxide or less, is the inventory’s central ambition. Around 25 percent of pre-satellite CMIP7 volcanic sulfur emissions come from these events, which many available ice-core arrays simply cannot detect. In the previous model intercomparison, CMIP6, the forcing from such eruptions was likely strongly underestimated, and because pre-industrial and scenario simulations use an aerosol climatology averaged over the historical period, that bias propagated into projections. The underrepresentation in CMIP6 stemmed partly from the difficulty of detecting small eruptions in sparse and uncertain pyrheliometer records of atmospheric optical thickness, and partly from the limits of ice-core detection itself.

Matching sulfate deposition peaks in ice cores to specific historical eruptions is one of the dataset’s most delicate tasks, and the team graded every proposed match on a four-level confidence scale. The highest confidence, flag 4, requires geochemical fingerprinting that links tephra shards found in the ice to a specific volcano, as was achieved for the previously unidentified 1831 eruption of Zavaritskii in the Kurile Islands. That reattribution matters: CMIP7 assigns 25.9 teragrams of sulfur dioxide to 1831, compared with 17.9 in the VolcanEESM inventory and just 8.3 in IVI2, where the Antarctic signal had gone unresolved. The team also reassessed older matches, retaining, for example, the 1824 eruption of Shishaldin as the source of a 0.8-teragram deposition event in 1825 after weighing alternative candidates.

Where direct measurements are missing, the authors derived new empirical relationships. Eruption latitude is estimated from the asymmetry of sulfate deposition between Greenland and Antarctica through a linear fit, latitude equal to 7.2 plus 48 times the deposition asymmetry factor, which explains 75 percent of the variance in eruptions with known locations and carries a prediction uncertainty of about 40 degrees. Injection height is estimated from sulfur dioxide mass through a power law, height equal to 17.3 times the mass raised to the power 0.12, with a relative uncertainty of 66 percent. For eruptions drawn from the geological database with no constrained sulfur mass, the team calibrated a default of 0.08 teragrams for VEI 4 eruptions so that the modeled pre-satellite aerosol optical depth anomaly from small eruptions matches the satellite-observed average of 0.003 over 1998 to 2023, a period with no large eruptions.

Some eruptions demanded bespoke treatment. The Laki eruption of 1783 to 1784 was split into ten explosive phases with a total ice-core-derived mass of 41.6 teragrams of sulfur dioxide, with phase-specific heights scaling with emission rate and ranging from 8.1 to 18.1 kilometers above sea level. Agung in 1963 was split into two phases of 4.9 and 5.2 teragrams, though the team notes the two masses were accidentally swapped, an error with minimal impact given how close the phases were. The authors also disclose a coding error that assigned a default date of 1 December instead of 1 January to 13 D4i-derived eruptions between 1750 and 1849, plus eruptions in 1868, 1885 and 1894, placing those events roughly a year too late; the flaw was detected too late for correction in the CMIP7 Fast Track.

The comparison with alternative inventories lays bare how uncertain historical volcanic forcing remains. For the pre-satellite era, CMIP7 contains roughly twice as many years with volcanic injections as either IVI2 or VolcanEESM, reflecting the deliberate effort to capture small eruptions. Even in the satellite era, discrepancies of 20 to 30 percent persist for recent small eruptions such as Nabro in 2011. The team acknowledges that residual biases remain: eruptions injecting less than 1 teragram appear about one order of magnitude weaker before 1979 than after, likely because ice cores miss them, while eruptions of 1 to 3 teragrams appear twice as strong pre-satellite, possibly because the single-core, Greenland-only D4i record overestimates stratospheric injections or captures sulfate of tropospheric origin. Tropical and Southern Hemisphere small eruptions remain particularly under-recorded.

Looking forward, the authors identify transfer functions converting ice-sheet deposition to atmospheric loadings that may be biased by up to 20 percent, sulfur isotope measurements showing that only about 60 percent of deposited sulfur was stratospheric for some Alaskan eruptions versus essentially all of it for tropical ones, and the growing contamination of Greenland records by anthropogenic sulfate after 1920 as priority targets. They also stress that the inventory covers sulfur alone: volcanic halogens and water vapor, dramatically illustrated by the 2022 Hunga Tonga eruption, which injected only 0.5 to 1 teragram of sulfur dioxide but 146 teragrams of water vapor into the stratosphere, as well as emerging non-volcanic sources such as wildfire-driven pyrocumulonimbus and aerosols from the space industry, all await inclusion. To standardize implementation, the team recommends injecting each eruption’s sulfur over 24 hours into a single model grid cell, with a Gaussian vertical profile centered on the provided injection altitude, and asks modeling groups to document any deviation from those prescriptions.

Subject of Research: Construction and uncertainties of the CMIP7 volcanic sulfur dioxide emission inventory for stratospheric aerosol forcing

Article Title: Stratospheric aerosol forcing for CMIP7 – Part 2: Volcanic sulfur dioxide emissions

Article References: Aubry, T. J., Sigl, M., Toohey, M., Chim, M. M., Verkerk, M., Schmidt, A., & Carn, S. A. (2026). Stratospheric aerosol forcing for CMIP7 – Part 2: Volcanic sulfur dioxide emissions. Geoscientific Model Development, 19(20), 10155-10184. https://doi.org/10.5194/gmd-19-10155-2026

Image Credits: AI Generated

DOI: 10.5194/gmd-19-10155-2026

Keywords: CMIP7, volcanic sulfur dioxide, stratospheric aerosol, ice cores, climate modeling, volcanic forcing, sulfate aerosols, satellite remote sensing, paleoclimate reconstruction, eruption source parameters, Global Volcanism Program, stratospheric aerosol optical depth