The drilling site for the WD ice core (79.48∘ S, 112.11∘ W; 1766 m a.s.l.) was selected to obtain a precisely dated, high-time-resolution ice-core record that would be the Southern Hemisphere equivalent of the deep Greenland ice cores (WAIS Divide Project Members, 2013, 2015). The 3404 m long WD ice core was collected from a cold (mean annual temperature -31 ∘C), high-snowfall (200 kg m-2 yr-1) West Antarctic site. Within the European Project for Ice Coring in Antarctica (EPICA), more deep ice cores were drilled in Antarctica (EPICA-Community-Members, 2004, 2006). The ice core from Dronning Maud Land (EDML; 75.00∘ S, 00.07∘ E; 2882 m a.s.l.) has, at 68 kg m-2 yr-1, a 2–3 times higher accumulation rate than the ice core at Dome C (EDC; 75.10∘ S, 123.35∘ E; 3233 m a.s.l.). Multiple deep ice cores have also been drilled in Greenland, including the Greenland Ice Core Project (GRIP; 72.60∘ N, 35.80∘ W; 3232 m a.s.l.), the Greenland Ice Sheet Project 2 (GISP2; 72.58∘ N, 38.47∘ W; 3053 m a.s.l.) and the North Greenland Ice Core Project (NGRIP or NorthGRIP) ice cores, providing continuous records of atmospheric impurities over the Holocene (Seierstad et al., 2014). Figure S1 (in the Supplement) summarizes the depth–age relation for these deep ice cores on the common, annual-layer-counted WD2014 chronology (Sigl et al., 2016) after applying volcanic synchronization during the glacial (Buizert et al., 2018) and Holocene (this study). The specific datasets used for aligning these chronologies are shown in Table 1.

Sulfur concentrations between 1300 and 2003 m depth (4060–10 000 BCE, 6010–11 950 years BP) covering the Early to Middle Holocene were analyzed using trace element continuous flow analysis (TE-CFA) at the Desert Research Institute (DRI) in Reno, USA. The DRI Ultra-Trace Chemistry Laboratory used a method that allowed continuous, simultaneous measurement of a large number of trace elements at very high depth resolution (Cole-Dai et al., 2021; McConnell, 2002; McConnell et al., 2017, 2018). The depth resolution for sulfur achieved with this system is 1 cm in ice, allowing for the achievement of a nominal monthly time resolution over the entire Holocene. Sulfate concentrations between 577 and 1300 m depth (396–4060 BCE) covering the Middle to Late Holocene and the brittle zone (Neff, 2014) of the WD ice core were analyzed using ion chromatography in discrete and continuous flow analysis mode (Cole-Dai et al., 2021, 2013, 2006) at the South Dakota State University, USA. The depth resolution for sulfate was 2 cm. High sampling resolution throughout the Holocene permitted the detection of annual cycles in impurity data, allowing for the precise and accurate annual-layer dating of the ice-core records during the entire Holocene (Sigl et al., 2016). For consistency with Toohey and Sigl (2017), we report the calendar ages using the ISO 8601 international standard, which does (in contrast to the historical Gregorian calendar) include a year zero. For key events and time periods, we also report ages as years before present (BP, years before 1950) – a notation used frequently in archeology, geology and other scientific disciplines.

Synchronization is based on matching volcanic sulfate, sulfur, acidity or conductivity peaks of the dependent core to equivalent peaks in an independently dated reference core, and it is used to transfer or to synchronize ice-core timescales. It is widely used in the ice-core community to align ice-core chronologies on a common reference chronology (Langway et al., 1995; Sigl et al., 2014; Svensson et al., 2020). For Greenland and the Arctic, many ice cores (e.g., NGRIP, GRIP and GISP2) have been synchronized (Rasmussen et al., 2013; Seierstad et al., 2014) on the Greenland Ice Core Chronology 2005 – GICC05 (Rasmussen et al., 2006; Svensson et al., 2008; Vinther et al., 2006), whereas the WD2014 chronology (Buizert et al., 2015; Sigl et al., 2016) serves as the reference chronology for Antarctica (Buizert et al., 2021, 2018; Sigl et al., 2015; Winski et al., 2019). Ice cores have also been synchronized across the hemispheres (Langway et al., 1995), but the density and certainty of these match points have been much lower owing to hitherto low dating precision in ice cores from Antarctica and the large abundance of volcanic eruption signals in the Greenland ice core from high-latitude volcanic eruptions (e.g., Iceland, Alaska and the Kamchatka Peninsula) hampering reliable source attribution. In an attempt to synchronize ice cores from Greenland and Antarctica over the entire Holocene, a total of 74 match points have been suggested between the NGRIP and EDML ice cores (Veres et al., 2013), about as many as were identified between WD and EDML during the Common Era (Sigl et al., 2014).

The accuracy of stratigraphic matches further depends on the volcanic signal (e.g., temporal resolution) and ice-core site-specific properties (e.g., accumulation rate variability) of both the dependent and reference ice-core records through time. We synchronized ice-core records in this study using an iterative approach. First, volcanic signals with outstanding magnitudes and characteristic temporal spacing that are virtually certain (e.g., in the 17th century BCE, 2910 BCE, the 45th–43th century BCE and the 67th–63th century BCE) were synchronized (major tie points). Confidence in these match points was derived from the combination of (1) a temporal sequence of distinctive signals, (2) comparable magnitudes, (3) a uniform evolution of layer thickness between stratigraphic tie points, (4) a distinctive shape of the common signals in some cases and, finally, (5) independent age constraints from 10Be that reflected variations in cosmic ray flux (Adolphi and Muscheler, 2016; Sigl et al., 2016). Using linear interpolation of the derived initial mean annual layer thickness calculated between age markers, secondary stratigraphic links from moderate volcanic eruptions became obvious and were matched to WD2014 (see Fig. S2 in the Supplement). Relative accumulation rates in Antarctica calculated for longer time periods usually show low variability, narrowing the window for potential stratigraphic tie points between the two records. We applied volcanic synchronization against WD first to EDML and EDC and verified that the individual selected tie points were consistent with the previous volcanic synchronization between EDML and EDC (Severi et al., 2007). We performed several iterations allowing for 218 (EDML) and 148 (EDC) volcanic matches with WD (Fig. 1, Table 2). We repeated this approach for the two Greenland ice-core records of sulfate from GISP2 (Mayewski et al., 1997) and dielectric profiling (DEP) from GRIP (Wolff et al., 1997). Confidence in the match points was derived from the combination of (1) a distinctive sequence of common signals, (2) a uniform evolution of layer thickness between stratigraphic tie points, (3) sequential annual-layer counts between volcanic age markers and (4) constraints from 10Be matching. We performed several iterations allowing for 164 (GISP2) and 93 (GRIP) volcanic matches with WD (Fig. 2, Table 2). We verified that all major bipolar tie points identified in GRIP and GISP2 are consistent with the previous synchronization between GRIP and GISP2 (Seierstad et al., 2014).

Sporadic volcanic sulfate deposition at the ice-core sites is superimposed on the background deposition of marine sulfate and other sulfuric species (e.g., methanesulfonic acid). This background is seasonally variable; however, without volcanic input, it has very limited variability between years (Cole-Dai, 2010). Therefore, a method to differentiate between volcanic sulfur or sulfate and the nonvolcanic background requires quantification of the background and its variability (Traufetter et al., 2004). To detect and quantify volcanic sulfate deposition, we used established methods (Cole-Dai, 2010; Cole-Dai et al., 2021; Gao et al., 2007; Sigl et al., 2013, 2014), which are summarized below. Sulfate deposition over both polar ice sheets varies systematically with snow accumulation and may be further modified randomly by site-specific post-depositional effects (such as redistribution of snow through wind). To account for these random and systematic differences, we selected a stacking approach for Antarctica (for which three continuous ice-core records are available) similar to the one used for the past 2000 years (Sigl et al., 2014) and in accordance with previous work (Gao et al., 2008; Crowley and Unterman, 2013).

Over the last 2500 years, the localities and the timing of several (N=31; Toohey and Sigl, 2017) stratospheric sulfur injections reconstructed from ice cores could be assigned based on matching the ice-core inventory with observed historical eruptions using the Volcanoes of the World online database (Global Volcanism Program, 2013) and other sources of information. There is some degree of uncertainty and subjectivity associated with such matchings. However, for certain cases, geochemical analysis of tephra from ice cores has been used to establish or strengthen the matches, including Veiðivötn in 1477 CE (Abbott et al., 2021b), Samalas in 1257 CE (Lavigne et al., 2013), Changbaishan in 946 CE (Oppenheimer et al., 2017; Sun et al., 2014), Eldgjá in 939–940 CE (Oppenheimer et al., 2018; Zielinski, 1995), Mt Churchill in 853 CE (Jensen et al., 2014), Katla in 822 CE (Büntgen et al., 2017; Plunkett et al., 2020) and Ilopango in 431 CE (Smith et al., 2020). Attributing locations to ice-core eruption signals over the full Holocene is even more difficult due to the increasing incompleteness and decreasing dating precision (often based on radiocarbon dating) over time of the volcanic eruption inventory derived from proximal geological evidence (Brown et al., 2014; Crosweller et al., 2012). To date, only a handful of ice-core sulfate peaks in the Holocene have been linked geochemically to known eruptions, including the caldera-forming 43 BCE Okmok II eruption in Alaska (McConnell et al., 2020a), the ca. 1628 BCE Aniakchak II eruption in Alaska (McAneney and Baillie, 2019; Pearce et al., 2004; Pearson et al., 2022; Plunkett and Pilcher, 2018), the caldera-forming 5677±150 BCE Crater Lake “Mazama” eruption in Oregon (Zdanowicz et al., 1999), the 5922±50 BCE Khangar eruption on the Kamchatka Peninsula (Cook et al., 2018) and the ca. 10 ka Grímsvötn “Saksunarvatn Ash” eruption series from Iceland (Oladottir et al., 2020). The vast majority of large eruptions, such as the caldera-forming Bronze Age Thera/Santorini eruption or the ca. 6440±25 BCE caldera-forming Kurile Lake eruption on the Kamchatka Peninsula, has remained unidentified in the ice-core record. We assigned approximate eruption latitudes for most sulfate signals in ice cores that cannot be immediately attributed to a known eruption, based on the presence or absence of simultaneous signals in the Greenland and Antarctic ice cores. Volcanic sulfate deposition identified synchronously (within small possible dating errors) in both Greenland and the Antarctic composites are attributed to eruptions in the tropics, whereas signals that occur in only one hemisphere are assumed to be of extratropical origin, as described in Sigl et al. (2015). Characteristic latitudes for unidentified eruptions are inferred from the latitudinal distribution of known eruptions. Using the mean distribution of all VEI ≥4 (where VEI refers to the volcanic explosivity index) eruptions from a Holocene eruption database (Global Volcanism Program, 2013), we assigned average latitudes of 48∘ N, 37∘ S and 5∘ N to all unidentified extratropical eruptions in the Northern and Southern hemispheres and in the tropics, respectively. We further attributed all volcanic events for which volcanic sulfate deposition to Greenland persisted for more than 10 years to Icelandic source eruptions, most likely from the Katla, Bárðarbunga and Grímsvötn volcanic systems or from shield volcanoes in the Western Volcanic Zone (Hjartarson, 2003; Sinton et al., 2005; Thordarson et al., 2003), and assigned 64∘ N as the default latitude for these prolonged episodes (Fig. 5; Tables 3, S1). We are aware that the use of default latitudes paints a crude picture of the geographical distribution of past global volcanic activity (see Fig. 6), but we currently lack the necessary knowledge to more precisely assign individual volcanoes to ice-core signals. We also note that, in the construction of aerosol optical properties and radiative forcing using the Easy Volcanic Aerosol (EVA) forcing generator (see Sect. 2.7), only the broad region of the eruption site is important (i.e., tropics, NHET and SHET), as the exact latitude has no impact on the generated aerosol properties. When sulfur emissions are directly used in aerosol–climate models, differences in aerosol evolution depending on the latitude of the eruption within these broad regions may be relevant (see Toohey et al., 2019; Marshall et al., 2021), and our choice of the mean latitudes helps to minimize any potential bias in the long-term mean radiative forcing. Consistent with Toohey and Sigl (2017), an eruption date of 1 January is assigned to unidentified eruptions.

Stratospheric sulfate injections are estimated from the ice-sheet sulfate flux composites using a method described in detail by Toohey and Sigl (2017). Briefly, ice-sheet average sulfate fluxes for Antarctica fA and Greenland fG are related to injected sulfur mass MS following Eq. (1): 1MS=13LGfG+LAfA, where LG and LA are transfer functions accounting for the spatial distribution of sulfate deposition over each hemisphere. Based on the analysis of the spread and deposition of nuclides from a nuclear bomb test, sulfate from prior volcanic eruptions and atmospheric model simulations (Gao et al., 2007), the transfer functions LG and LA are estimated to be 1×109 km2 for tropical eruptions and 0.57×109 km2 for extratropical eruptions. The method described above is based on the assumption that the ice-core sulfate deposition is proportional to the stratospheric sulfur emission. In fact, some of the sulfate deposited may originate from volcanic sulfur emissions into the troposphere, especially when volcanic eruptions are situated upwind of (e.g., in Alaska) or in close proximity to the Greenland ice sheet (e.g., in Iceland). Recently, sulfur isotopes from Greenland ice-core records have been used to detect the presence of sulfate deposited in Greenland via both the stratospheric and tropospheric transport pathways following the large VEI = 6 eruption of Katmai/Novarupta in Alaska, upwind of the Greenland ice sheet (Burke et al., 2019). Of particular importance are long-lasting, effusive (i.e., nonexplosive) eruptions from Iceland, which may produce significant sulfate deposition over Greenland, even when the stratospheric injection is minimal. The two largest fissure eruptions in Iceland in historical times (Eldgjá 939–940 and Laki 1783–1784 AD) are the most prominent examples, and the extent to which sulfate was injected into the lower stratosphere during these eruptions is the subject of ongoing research (Lanciki et al., 2012; Schmidt et al., 2012; Zambri et al., 2019). Only for very recent volcanic eruptions are direct observations of key eruption source parameters (e.g., plume height, SO2 dispersion height, duration, season) available, which determine how much sulfur gets injected into the stratosphere. Detailed volcanological fieldwork could delineate 10–11 distinctive eruptive episodes during the Laki 1783–1784 event (Thordarson and Self, 2003), allowing for the development of detailed SO2 emission scenarios for modeling the climatic impact of this episode (Schmidt et al., 2010; Zambri et al., 2019). However, such detailed information is not available for other large fissure eruptions in Iceland, of which at least 14 are known to have occurred over the Holocene (Thordarson and Larsen, 2007; Thordarson et al., 2003). To correct for a significant proportion of tropospheric sulfate when estimating stratospheric sulfur emissions, Crowley and Unterman (2013) adjusted Greenlandic sulfate depositions following the Laki eruption in 1783–1784 and derived a ratio of stratospheric to total sulfate deposition of 0.15. Due to a lack of data on the stratospheric vs. tropospheric distribution of injected sulfur for other major Icelandic eruptions of the Holocene, we adopted this approach and used a transfer function of 0.10×109 km2 for those long-lasting deposition signals that we assume to be from prolonged volcanic eruptions in Iceland. We implemented the sulfur injections within these eruptive episodes using the biannual resolution of the GISP2 ice-core record (i.e., a 16-year-long episode is implemented as eight subsequent injections), and the durations are reported so that injection can be spread uniformly over time in simulations. We stress that additional objective criteria to detect proximal eruption signals, correctly attribute these to specific source eruptions and subsequently correct the VSSI estimates are urgently needed.

Estimates of VSSI have significant uncertainty due to three major sources of potential errors: (1) random errors in the ice-core flux measurements, (2) uncertainties in the transfer functions used to translate the ice-core sulfate data to estimates of VSSI and (3) potential errors in the estimation of the latitudinal position (and explosivity) of the eruption (i.e., tropical vs. extratropical explosive vs. extratropical effusive). VSSI uncertainties are included in the HolVol dataset, with the aim of estimating the uncertainties from the first two terms. Uncertainty related to the limited number of ice cores and related sampling of the ice sheets has been estimated (see Sect. 2.4.3). As in Toohey and Sigl (2017), this uncertainty is added in quadrature to an estimate of the uncertainty related to using ice-sheet deposition to estimate hemispheric deposition. Based on an ensemble of aerosol model simulations (Toohey et al., 2013), this term is estimated to contribute ∼16 % and 9 % of uncertainty to the Northern Hemisphere (NH) and Southern Hemisphere (SH) transfer functions (LNH and LSH), respectively, but these estimates may be model-dependent, and recent work has pointed to potentially larger values (Marshall et al., 2019, 2018). It remains difficult to quantify errors arising from a potentially incorrect attribution of the source location for individual eruptions. VSSI from an eruption erroneously attributed to a tropical source, which in reality may have been from two different eruptions in the high latitudes of both hemispheres, will be overestimated by 43 %. As another example, sulfate deposition in Greenland resulting from a potential cluster of several subsequent volcanic eruptions in the NHET may not be recognized as separate eruptions in the biannual-resolution GISP2 record; thus, it may be erroneously attributed to a prolonged eruptive period when sulfate levels remain increased for >10 years. Under such a scenario, the true VSSI would be underestimated by up to 80 %. To account for the latter case, reported VSSI uncertainties for prolonged eruptions have been inflated to include magnitudes that would be calculated if the eruption was not prolonged, which results in uncertainties of over 100 %. This large error also signifies a relatively low confidence in the adjustment to the transfer function used for prolonged eruptions compared with explosive extratropical eruptions. We note that only specific eruptions may be subject to errors caused by incorrect attributions which can be subsequently assessed and corrected in future updates of this database if independent constraints for source locations from cryptotephra, sulfur isotope and trace metal analyses of archived and new ice cores become available (Burke et al., 2019; Gautier et al., 2019; McConnell et al., 2017, 2020a). A primary source of systematic error in the VSSI estimates is likely to originate from the uncertainty in the transfer functions (LG and LA) used to estimate hemispheric stratospheric sulfate aerosol burdens. These are originally derived using ice-sheet sulfate fluxes in Antarctica and observed from the stratospheric sulfate burden of the Pinatubo 1991 eruption as well as deposited nuclear bomb test fallout in Greenland and climate model simulations (Gao et al., 2007). Continued efforts to constrain observational uncertainties with aerosol model simulations have been unsuccessful due to significant inter-model differences – for example in the simulation of aerosol spread and deposition after the Tambora 1815 eruption (Marshall et al., 2018).

The Easy Volcanic Aerosol (EVA) version 1.2 forcing generator (Toohey et al., 2016b) is employed to convert sulfur emissions into optical properties of volcanic aerosols. We specifically consider the variation in the stratospheric aerosol optical depth (SAOD) at 550 nm. Using a time series of VSSI and eruption latitudes as input, EVA generates aerosol optical properties as required for use in climate model simulations. The spatiotemporal structure of the EVA output fields is based on a simple three-box model of stratospheric transport that is optimized to produce agreement with observations of the aerosol cloud from the eruption of Pinatubo in Indonesia in 1991. Internally, EVA first computes the transport of sulfate mass and then scales the sulfate mass to the SAOD. While this scaling is linear for most eruptions, following Crowley and Unterman (2013), a nonlinear scaling between mass and SAOD is adopted for very large eruptions (i.e., eruptions with a VSSI in excess of that of Tambora in 1815). Furthermore, to account for the self-limiting effect of aerosol growth on the stratospheric lifetime of aerosol after large eruptions implied in model studies (Pinto et al., 1989; Timmreck et al., 2009), a simple parameterization of variable removal time has been implemented in EVA based on ECHAM5-HAM aerosol model simulations of eruptions with a wide range of magnitudes (Metzner et al., 2014). Based on the model results, the stratospheric aerosol removal timescale is varied between its nominal value of 11 months and a minimum of 6 months as the global stratospheric sulfate burden rises above 10 Tg S. We refer to the SAOD results presented below that were generated by the EVA forcing generator using the HolVol VSSI database as “EVA(HolVol)”. This naming convention emphasizes the two-stage procedure of the SAOD reconstruction, with HolVol used as an input to EVA. SAOD time series are shown as monthly, annual or centennial averages. Therefore, peak SAOD values can differ significantly depending on the time resolution of the time series.

Nominal age uncertainty for the WD2014 chronology due to ambiguities in the interpretation of annual layering has been estimated to linearly increase with age over most of the Holocene (Sigl et al., 2016). Constrained at 775 CE using 10Be in ice cores (Mekhaldi et al., 2015) and 14C in tree rings (Büntgen et al., 2018) to detect the distinctive 774/775 CE solar proton event (Miyake et al., 2012), the age error from annual-layer interpretation down to 3000 BCE (5 ka) is estimated to be better than ±20 years. During the Early Holocene, at 9500 BCE (11.5 ka), the WD2014 age uncertainty was estimated at ±66 years. Matching the common production signal in cosmogenic isotopes (14C, 10Be) has further allowed us to assess the WD2014 ages relative to the radiocarbon calibration curve, which is based on dendrochronology during the Holocene and, thus, has virtually no age uncertainty (Sigl et al., 2016). The best fit necessary to align both chronologies had been found to vary by small margins of less than ±15 years throughout the Holocene, suggesting that the cumulative error estimated from the annual-layer counting of WD2014 is very conservative. There is a tendency for WD2014 ages to be slightly too young during the Early Holocene and slightly too old between 7000 BCE and 1 CE (Sigl et al., 2016). No 10Be measurements from WD were previously available; thus, no assessment of the WD2014 timescale was possible for the time period between 3500 and 500 BCE. To fill this gap, we employ a multi-millennial compilation of the occurrence of ring-width minima and frost rings in a bristlecone pine chronology from the southwestern USA covering the past 5000 years (Salzer and Hughes, 2007). A strong association between frost-ring formation and climatically effective volcanic eruptions has been previously noted (Baillie, 2010; Lamarche and Hirschboeck, 1984; McAneney and Baillie, 2019; Salzer and Hughes, 2007; Sigl et al., 2015). To assess the temporal relation between major volcanic eruptions reconstructed with HolVol and cooling extremes indicated by the tree-ring series, we extract all marker events from the compilation by Salzer and Hughes (2007) (N=10) in which at least two consecutive ring-width minima corresponded to a frost-damaged ring within an error margin of ±1 year (Table 4). Additional marker years in which a frost ring corresponds to a ring-width minima within ±1 year are provided in the Supplement (Table S2).

The HolVol v.1.0 VSSI time series is shown in Figs. 6 and 7. With 100 % data coverage for Antarctica and 95 % data coverage for Greenland, we consider this record to be virtually complete for all volcanic eruptions with a strong climate impact potential (i.e., VSSI comparable to or larger than the 1991 Pinatubo eruption). The ability to detect and quantify smaller events is primarily limited by data gaps (equivalent to 560 years) and the coarse (biannual) temporal resolution of the GISP2 record from Greenland. Therefore, there is the possibility that smaller eruptions situated in the NHET are under-recorded. With only 88 % data coverage between 3000 and 1000 BCE obtained within the “brittle zone” of the GISP2 record, under-recording and ambiguities in matching and correctly attributing source latitudes pose some limitations at the moment.

HolVol v.1.0 contains a total of 1189 volcanic eruptions, resulting in a cumulative VSSI of 7412 Tg S between 9500 BCE and 1900 CE. On average, a detected eruption occurred once every 10 years. Of these eruptions, 850 injected at least 1 Tg S into the stratosphere, the equivalent of the eruptions of Nabro (Eritrea 2011) and Kasatochi (Alaska 2008), which were implicated to have contributed to the slowdown of warming in the 21st century (Carn et al., 2016; Ridley et al., 2014; Santer et al., 2014). Figures 6 and S6 show the latitudinal distribution for each reconstructed VSSI. Figure 8 summarizes their mean distribution over the Holocene. A total of 40 % of the eruptions (with VSSI >1 Tg S) are attributed to tropical eruptions (30∘ N–30∘ S), 5 % of eruptions are attributed to effusive Icelandic events (>63∘ N), 45 % of eruptions are attributed to other NHET (30–60∘ N) events and 10 % of eruptions are attributed to SHET (30–90∘ S) events. The mean frequency of reconstructed volcanic eruptions >1 Tg S is 0.074 yr-1 (i.e., an eruption every 14 years on average). These 850 eruptions injected a total of 7260 Tg S into the stratosphere, of which 75 % was emitted in the tropics, 4 % was emitted in Iceland, 18 % was emitted in the NHET and 4 % was emitted in the SH (Fig. 8).

The number of eruptions and the cumulative centennial VSSI varied within the Holocene (Figs. 6, 7). In general, the number of eruptions and the cumulative VSSI were enhanced during the Early to Middle Holocene (10th to 5th millennium BCE; 76 Tg S per century on average). Between the 4th millennium BCE and the present, both the average number of eruptions and the cumulative VSSI (45 Tg S per century) were 21 % and 41 % lower, respectively (Figs. 6, S4, S7). The period from 9500 to 7000 BCE, when glacial ice sheets were retreating rapidly and were widespread (Carlson and Clark, 2012), is characterized by the highest frequency of eruptions as well as the largest cumulative VSSI over the entire Holocene. With an average of 90 Tg S injected in the stratosphere per century, this period – which we term the “Deglaciation Active Period” – is 43 % above the Holocene mean VSSI rate of 63 Tg S. An increased VSSI rate is noted for eruptions in the NHET as well as in the tropics, but this feature is absent for the SHET (see Fig. S6 in the Supplement). The majority of the events that we attributed to prolonged eruptive episodes (379 years of cumulative duration) also falls into this time period (Fig. 6, Table 3).

The window from 4000 to 1000 BCE has the lowest frequency of eruptions and the smallest VSSI rates in the Holocene. With 21 % less eruptions and 36 % smaller VSSI rates than the Holocene mean, we term this period the “Holocene Quiet Period”, in analogy to other time periods with reduced volcanic activity such the “Medieval Quiet Period” (700–1100 CE) and the “Roman Quiet Period” (40 BCE–200 CE). Throughout the Holocene, the longest subsequent time period without an eruption with VSSI >1 Tg S is 77 years (ending in 3206 BCE); the longest period without VSSI >10 Tg is 317 years (ending in 2258 BCE). These volcanically quiet periods are slightly longer compared with the same metrics for the Medieval Quiet Period (55 and 217 years, respectively) and the Roman Quiet Period (71 and 212 years, respectively).

A total of 8 of the 10 largest VSSI injections are recorded between 6700 and 4300 BCE, all exceeding the VSSI of the largest known volcanic eruptions of the Common Era except for Samalas (Lavigne et al., 2013; Vidal et al., 2016) in 1257 CE (ranked ninth). The highest recorded VSSI reach values >150 Tg S. With a VSSI rate 22 % above the Holocene mean, we term this period the “Mid-Holocene Active Period”.

The global mean SAOD from the EVA(HolVol) reconstruction is shown in Fig. 9. The SAOD closely follows the spatiotemporal structure of VSSI in the Holocene, albeit with relatively less pronounced peaks for the largest eruptions due to the nonlinear parameterizations used in EVA. The global mean SAOD over the Holocene was 0.0153; SAOD over the NH (0.0182) was almost 50 % higher than that over the SH (0.0124). The global mean SAOD between 9500 and 4500 BCE was 48 % higher than between 4500 BCE and 1900 CE. The difference in SAOD between these two time windows was stronger (+57 %) when integrating over the NH (0–90∘ N), whereas it was less pronounced (+21 %) when integrating over the SHET (30–90∘ S). The largest global annual SAOD reached 0.85; the largest SAOD over the NHET reached 1.45 following the Crater Lake eruption (Oregon, USA). For comparison, the largest eruption during the Common Era, Samalas 1257 CE, is estimated in eVolv2k to have produced a global annual SAOD of 0.50; the largest nontropical eruption of the Common Era in 536 CE produced a NHET SAOD of 0.43. We stress that for such large eruptions, which are significantly larger than any eruption observed in the instrumental era, uncertainties in the SAOD should be understood to be large.

The EVA(HolVol) reconstruction is compared to that of Kobashi et al. (2017) in Fig. 10. As the Kobashi et al. (2017) reconstruction contains estimates of radiative forcing (RF, in units of W m-2), the EVA(HolVol) SAOD values are converted to RF using the linear scaling (RF = -25 ⋅ SAOD) of Hansen et al. (2005), which has been employed in prior Intergovernmental Panel on Climate Change (IPCC) reports (Myhre et al., 2013). We note that several recent studies have suggested that consideration of rapid adjustments (e.g., in cloud formation) leads to a reduction in the scaling factor in the order of 20 % (Marshall et al., 2020; Schmidt et al., 2018). The major difference between the multi-ice-core HolVol reconstruction and the single-ice-core (GISP2) reconstruction from Greenland is the smaller magnitudes (minima of -21 W m-2) of RF for large volcanic eruptions in HolVol compared with values as strong as -50 W m-2, as reconstructed by Kobashi et al. (2017), which we attribute to applying a nonlinear scaling to HolVol. Furthermore, HolVol RF values are constrained by ice-core records from Antarctica, whereas Kobashi et al. (2017) assumed that the GISP2 sulfate record from Greenland is representative of the global volcanic sulfate burden, thereby inevitably overestimating RF for all eruptions with unipolar sulfate distribution (e.g., eruptions from Iceland) or eruptions with a strong asymmetry of the sulfate burden in the NH. While the negative radiative forcing from large events is very likely overestimated by Kobashi et al. (2017), the negative RF from smaller and moderate eruptions (that are often not detected in the single-ice-core reconstruction from Greenland) are underestimated. Some of the difference is also due to the additional inclusion of a nonzero background SAOD in the EVA(HolVol) reconstruction. The effect of these methodological differences is that HolVol depicts smaller variability than the previous reconstruction of global RF (Kobashi et al., 2017).

The results of the assessment of dating accuracy and precision are summarized in Fig. 11 (see Table 4 and Table S1 in the Supplement for details). The previous assessment based on correlating multidecadal- to centennial-scale production rates in the cosmogenic radionuclides 10Be and 14C showed only slightly varying age-scale differences over most of the Holocene (Sigl et al., 2016). This indicates a high accuracy and precision of the WD2014 timescale, in contrast to other annual-layer-dated chronologies (e.g., GICC05) that consistently overestimated (via overcounting annual layers) ages throughout most of the Holocene (Adolphi and Muscheler, 2016; Muscheler et al., 2014). Age differences between ice-core-indicated sulfate deposition and tree-ring-indicated summer cooling extremes (i.e., frost-ring formation co-occurring with reduced ring width) are calculated between 3000 BCE and 1640 CE for 14 major volcanic sulfate signals. The characteristic spacing of sulfate peaks in ice-core-indicated and tree-ring-indicated cooling events has previously been proposed as strong evidence of an age-scale bias in GICC05 over the Late Holocene (Baillie, 2008, 2010; McAneney and Baillie, 2019), including some of the very same tree-ring marker years (e.g., 1627 BCE, 43 BCE, 542 CE) that we now correlate against the WD ice-core record. Before 1 CE, the age differences show only subtle variations within very narrow margins of 3–5 years, with WD2014 ages being a few years too old on average. This indicates that the WD2014 ice-core timescale and, thus, the HolVol v.1.0 eruption database are highly accurate as well as precise for at least the past 5 kyr (and probably also over the full Holocene, given that WD ice-core quality and data resolution improve again) below the brittle zone (i.e., before 4000 BCE) of the WD ice core (Sigl et al., 2016).

The effect of deglaciation on mantle melting beneath Iceland leading to increased eruption frequencies has been recognized since the 1990s (Hardarson and Fitton, 1991; Jull and McKenzie, 1996). Globally, ice sheets reached their maximum extent during the Last Glacial Maximum (LGM), before retreating rapidly during the deglaciation until the Early Holocene (Clark et al., 2012). It is generally thought that the postglacial ice retreat and mass unloading after the LGM resulted in regionally increased frequencies of subaerial eruptions in volcanically active areas, due to increased mantle melting rates (Huybers and Langmuir, 2009). While quantifying the increase in some of these areas (e.g., southern Andes, Cascades and the Kamchatka Peninsula) remains difficult due to the incomplete nature of the geologic eruption record (Watt et al., 2013), the evidence is particularly strong for Iceland (Maclennan et al., 2002; Sinton et al., 2005). Coming out of the glacial, the final deglaciation of Iceland was dominated by rapid ice unloading that peaked between 9800 and 8300 BCE and coincided with an increase in volcanic eruption rates (mass discharge per time) which were 30–50 times higher than the present day. These high eruption rates persisted for over 1000 years after the deglaciation in each area investigated (Maclennan et al., 2002).

When reconstructing a 110 kyr volcanic eruption record from the GISP2 ice core, Zielinski et al. (1996) and Lin et al. (2022) noted a strong increase in the number, magnitude and duration of volcanic sulfate peaks during the termination of the Last Glacial Period and the Early Holocene; they tentatively linked this to crustal responses following the deglaciation. A similarly long-lasting sulfate signal dated to 3160 BCE in HolVol was – in the absence of known volcanic eruptions at the time or confirmative data from other ice cores – attributed by Zielinski et al. (1994) to anomalous marine biogenic emissions. In light of the independent verification of long-lasting acid deposition (see Fig. 5) in 3160 BCE from the GRIP core (Wolff et al., 1997) and new ice-core records that corroborate the fact that volcanic sulfate emissions can be sustained for centuries (McConnell et al., 2017), we interpret the increased frequency and duration of volcanic sulfate deposition in Greenland as additional evidence of the increased volcanic activity, predominantly from Iceland, following the large-scale warming during the deglaciation (Geirsdottir et al., 2009). The spatiotemporal structure of volcanic emissions in HolVol, with 75 % higher SAOD during the Early Holocene (9500–7500 BCE) than between 4000 BCE and 1000 CE and the increased SAOD concentrating in the NH, is consistent with a causal coupling of subaerial volcanism and rapid deglaciation in formerly glaciated volcanic regions.

The increased VSSI rates during the Early Holocene, also from eruptions that we have attributed to the tropics, seem to be at odds with the idea of a coupling between volcanic activity and rapid ice unloading during the deglaciation, as these areas were not glaciated during the glacial period. To investigate this further, we compared the relative distribution of sulfur deposition between Greenland and Antarctica (defined as the asymmetry ratio) for all eruptions in HolVol v.1.0 with those of known tropical eruptions (Fig. S10). We find that the mean asymmetry ratio for attributed tropical eruptions between 9500 and 7000 BCE (N=98) is significantly (p<0.05) different (i.e., indicating a stronger asymmetry of sulfate burden towards Greenland) from the mean asymmetry ratio for attributed tropical eruptions between 7000 BCE and 1900 CE (N=218). This difference arises due to a larger frequency of events with a large (>0.75) asymmetry ratio. We interpret this result to be an indication that the former group (during deglaciation) contains more eruptions that occurred further north than the latter group. The mean asymmetry ratios of both of these groups of bipolar eruptions are significantly (p<0.01) different (i.e., again indicating a stronger asymmetry of the sulfate burden towards Greenland) from the mean asymmetry ratios calculated for the limited number (N=11) of known tropical volcanic eruptions in ice-core records in both HolVol v.1.0 and eVolv2k. Based on this we hypothesize that the apparent increased volcanic activity in the tropics is likely an artifact from our applied and traditionally used source attribution (i.e., bipolar signal equals tropical source). We further hypothesize that some of the increased activity actually took place in glaciated NHET regions (e.g., Iceland, Alaska and the Kamchatka Peninsula), but contemporaneous volcanic sulfate detected in Antarctica had caused us to incorrectly attribute the eruption source to the tropics.

A consequence of this is that we would have overestimated the true frequency of tropical eruptions and underestimated the true frequency of NHET eruptions (some of which plausibly originated from volcanic areas experiencing ice unloading). This hypothesis is gaining increasing support from new tephra identifications for bipolar volcanic events that can be geochemically assigned to eruptions in Iceland (Lin et al., 2022; Svensson et al., 2020) or Alaska (McConnell et al., 2020a; Pearson et al., 2022). Consequently, in recently published studies, “bipolarity” is no longer a definitive criterion for tropical volcanic sources (Abbott et al., 2021a; Lin et al., 2022; Pearson et al., 2022). Linking specific volcanic eruptions with ice-core-indicated sulfate signals in HolVol, however, remains difficult due to the often low age precision of proximal volcanic deposits and the scarcity of tephra to fingerprint and identify eruptions in ice cores during the Holocene (Abbott and Davies, 2012). The 10 ka Grímsvötn tephra series (i.e., the Saksunarvatn Ash, found in numerous Greenland ice cores; see Fig. 5) is one of the few exceptions; however, these Grímsvötn tephra layers are increasingly considered to represent a time interval marker spanning approximately 500 years, rather than a sharp marker horizon, further complicating the alignment of climate proxy and volcanic records in the North Atlantic region (Oladottir et al., 2020). Although difficult to date precisely, large volume (≥2 km3) fissure eruptions and lava shield eruptions from the Western Iceland Zone (e.g., Hallmundarhraun, Leitahraun, Skjaldbreiður and Þingvallahraun) overlap within age uncertainties with some prominent prolonged sulfate signals in the Greenland ice cores (Fig. 5, Table S1). Since Holuhraun, a comparable small fissure eruption producing slightly above 1 km3 of lava in 2014–2015 (Bonny et al., 2018) is readily detectable in snow samples from Greenland; despite increased sulfate background from industrial emissions (Du et al., 2019), the prolonged sulfate signals over the Holocene may plausibly be linked to the aforementioned long-lived eruptions of larger volume with their characteristically low average effusion rates (Sinton et al., 2005).

Although carbon emissions from volcanoes are dwarfed by human emissions (Fischer et al., 2019; Le Quéré et al., 2018), several studies have suggested that post-deglaciation increases in subaerial volcanism evoked potential feedbacks in the climate system primarily through the co-emission of greenhouse gases (e.g., carbon dioxide (Huybers and Langmuir, 2009; Kutterolf et al., 2013). Observations have further shown that individual volcanoes such as Katla in Iceland can act as large point sources of CO2, emitting 12–24 kt d-1, even during quiescent time periods (Ilyinskaya et al., 2018). Potential gas emissions from prolonged volcanic eruptions lasting for decades are likely several orders of magnitude larger. As estimating the global volcanic CO2 flux to the atmosphere often involves relating volcanic fluxes of sulfur dioxide with measured or estimated C/S molar ratios (Fischer et al., 2019; Werner et al., 2019), our comprehensive HolVol reconstruction of volcanic sulfate now provides a basis for further research to advance our understanding of the coupling between climate and volcanism during the last deglaciation.

Reconstructions of volcanic aerosol forcing commonly assume that the vast majority of the volcanic sulfate deposited on the polar ice sheets derives from fallout from the stratosphere. Observations of volcanic SO2 emissions using remote sensing show that a large proportion of the total volcanic sulfur emissions remains within the troposphere. Between 1979 and 2018, a total of 44 Tg SO2 was emitted globally from effusive eruptions (Carn et al., 2016), of which only 5 % was stratospheric emissions. The mean plume height of these effusive eruptions was <7 km. Over the same time period, 54 Tg SO2 was emitted from explosive eruptions, of which >80 % was stratospheric emissions with a mean plume height of 16 km (Carn et al., 2016).

For the HolVol reconstruction, as in earlier reconstructions, stratospheric sulfur injection is estimated from the ice-core sulfate fluxes using simple scaling factors, which are assumed to be unbiased in an average sense but uncertain for individual eruptions. Indeed, some sulfur spikes recorded in ice cores may be the result of purely tropospheric emissions. Remote sensing suggests that 1–2 Tg S was emitted up to 1–3 km high during the 6-month-long fissure eruption of Holuhraun starting in September 2014 (Schmidt et al., 2015). An increase in volcanic sulfate and fluoride dated to late 2014 in a northeastern Greenland snow-pit sample indicates that volcanic fallout from this effusive eruption is preserved on the Greenland ice sheet (Du et al., 2019), but the deep ice cores used to estimate volcanic fallout over the Holocene in this study have not been updated to the present. Future work linking recent observed eruptions to ice cores may help to constrain how much of the Icelandic tropospheric emissions from effusive eruptions can be deposited over Greenland and to improve our interpretation of ice-core sulfate records in terms of stratospheric vs. tropospheric content.

While tropospheric sulfur emissions from eruptions lead to aerosols with shorter atmospheric lifetimes, the climate impacts of such emissions may not be negligible, especially for the largest such eruptions. In terms of atmospheric sulfur mass injection, the prolonged fissure eruptions of Lakigar, with 61 Tg S (Thordarson and Self, 2003), and of Eldgjá, with 110 Tg S (Thordarson et al., 2001), quantified using the petrological method, exceeded that of most stratospheric eruptions in the Common Era (Thordarson and Self, 2003; Thordarson et al., 2003). For comparison, these sulfur emissions are twice as high as present-day global annual sulfur emissions from fossil fuel burning and industrial processes (Lamarque et al., 2010). Thus, such extreme eruptions can be seen as natural analogues for the massive tropospheric sulfur emissions that occurred during the 1970s–1980s in industrialized Europe and North America, leading to increased SAOD and reduced solar irradiance at the surface known as “global dimming” (Wild, 2009). In particular, during the Early Holocene, when these types of emissions were more frequent and persistent, as reconstructed from proximal geologic records (Maclennan et al., 2002; Sinton et al., 2005) and from HolVol, our understanding of stratospheric and tropospheric volcanic sulfur emissions remains fragmentary, calling for the development and application of new research approaches and more specific diagnostic ice-core proxies for discrimination.

Comprising the eruption year and the VSSI, detailed information about the two primary eruption source parameters that define the eruptions' climatic impact are provided in HolVol v.1.0. However, observations and climate modeling suggest that additional eruption source parameters are important (Aubry et al., 2020; Marshall et al., 2019). Specifically, the location, the season of the eruption and the plume height of the SO2 emission are important, as they have an effect on the specific climate footprint of a given eruption. Refining these secondary – and often interlinked – eruption parameters and quantifying their effect constitute a challenge that is ideally addressed using a multidisciplinary approach with evidence from classical proximal deposits (geologic records), distal fallout (ice cores) and climate models. The detection of volcanic ash (i.e., cryptotephra) preceding the sulfate deposition in Greenland in 43 BCE, for example, allowed for the eruption to be geochemically pinpointed to the caldera-forming Okmok II event in Alaska (McConnell et al., 2020a). The known location not only gave access to key eruption source parameters, such as the magnitude or the volcanic plume height from the proximal deposits (Burgisser, 2005; Crosweller et al., 2012), but also helped to narrow down the eruption date to the winter season, due to the shorter (< weeks) atmospheric lifetime of ash compared with sulfate. The known location (53∘ N) can, in turn, be employed to evaluate the performance of aerosol–climate models (and statistical emulators) used to project the radiative effects of volcanic eruptions over a wide covarying range of SO2 emission magnitudes, injection heights and eruption latitudes (e.g., (Marshall et al., 2019). In the case of Okmok II, model simulations constrained by ice-core deposition values from Greenland and Antarctica would imply a most likely eruption latitude between 44∘ N and 9∘ S for an eruption in January with a 26.5 km plume height and a 50 Tg S injection (Marshall et al., 2019, 2021). This is only slightly south of the actual location of the eruption at 53∘ N. Besides information from cryptotephra, information about the potential volcanic sources may also been drawn from the halogen content or the trace element chemistry, as several case studies have demonstrated (Clausen et al., 1997; Kellerhals et al., 2010; McConnell et al., 2017). In addition, sulfur isotopes (33S and 34S) have been established as a powerful tool (Baroni et al., 2007; Gautier et al., 2019) to differentiate sulfate produced above the ozone layer (i.e., stratospheric) from sulfate forming below the ozone layer (i.e., tropospheric), and they are now applied to ice-core records at a sub-annual time resolution (Burke et al., 2019). Thus, further elucidating eruption parameters using novel methodology, such as targeted cryptotephra analyses and high-resolution S-isotope and trace element measurements, holds great potential with respect to substantially refining and improving HolVol v.1.0 as well as reducing existing uncertainties in the future.