Southern Hemisphere continental temperature responses to major volcanic eruptions since 1883 in CMIP5 models

Although global and Northern Hemisphere temperature responses to volcanic forcing have been extensively investigated, knowledge of such responses over Southern Hemisphere continental regions is still limited. Here we use an ensemble of CMIP5 models to explore Southern Hemisphere temperature responses to four major volcanic eruptions: Krakatau (1883), Santa Maria (1902), Agung (1963) and Pinatubo (1991). Focus is on near-surface temperature responses over southern continental landmasses including southern South America, southern Africa and Australia and their seasonal differences. Findings indicate that for all continents, temperature responses were strongest and lasted longest following the Krakatau eruption. Responses in Australia had the shortest lag time, strongest maximum seasonal response and the most significant monthly anomalies. In contrast, southern South America records the longest lag time, weakest maximum seasonal temperature response and lowest number of monthly negative anomalies following these eruptions. In most cases, the strongest single-season response occurred in austral autumn or winter, and the weakest in summer or spring. We tentatively propose that cooler temperature responses are likely caused, at least in part, by the intensification of the westerlies and associated mid-latitude cyclones and anti-cyclones.


Introduction
The relationship between volcanoes and climate has long been of scientific interest (Humphreys 1913;Gilliland 1982;Robock 2000;Allen et al. 2018), so much so, that Past Global Changes (PAGES) now has a working group examining volcanic impacts on climate and society (VICS).
It is well-recognized that volcanic events emitting c. 5 Tg or more of sulphur-containing gases into the lower stratosphere, where chemical reactions form sulphate aerosols, can affect climate (Timmreck 2018). These aerosols not only change the earth's energy balance by reducing incoming solar radiation through reflection and scattering, but also cause absorption of near infrared and long-wave radiation, with consequential warming of the lower stratosphere (Stenchikov et al. 1998;Langmann 2014). It is understood that changes in radiation induce stronger zonal winds and increase temperature and density gradients between the poles and equator, thereby strengthening the Northern Hemisphere (NH) polar vortex (Perlwitz and Graf 1995;Zambri and Robock 2016). However, given the expanse of oceans in mid-to high latitudes of the Southern Hemisphere (SH), the SH polar vortex and jet streams are more robust. This weakens the expected impact in southerly latitudes (Robock et al. 2007); nonetheless, it too is susceptible to volcanic forcing (Hudson 2012;Kidston et al. 2015).
Major eruptions may cause global annual temperatures to decrease by at least 0.1 °C for 2 to 3 years following the eruption (Robock 2000;Brönnimann et al. 2019). Findings generally show cooling over much of the NH, particularly in boreal summer, and widespread mid-to high latitude winter warming (Stoffel et al. 2015;Zambri and Robock 2016;Zambri et al. 2017). SH temperature responses are somewhat weaker than those in the NH (Man et al. 2014;Raible et al. 2016), with cooling of c. 0.1-0.2 °C for 1-2 years after major eruptions (Mass & Portman 1989;Robock & Mao 1995). Studies into the SH response to volcanic forcing are scarce; however, climate reconstructions can reveal climatic responses to volcanic forcing. For instance, climate proxies have shown cooling occurred after several eruptions between 1833 and 1900 in Lesotho Nash 2010), after Lakigigar (1783) and Tambora (1815) in the western Cape of South Africa (Dunwiddie and LaMarche 1980), and in Patagonia after eruptions between 1640 and 1989 (Roig and Villalba 2008). Despite these findings, no clear link was established between volcanic eruptions and cooling periods in Tasmania, southernmost Australia (Allen et al. 2018).
CMIP5 models have been shown to overestimate responses to volcanic forcing (Driscoll et al. 2012; Barnes et al. 2016;McGraw et al. 2016), which is likely a result of inaccurate parameterization methods for aerosols and ice, and mixed phase clouds, as well as the inaccurate removal processes of aerosols (Chylek et al. 2020;Dhomse et al. 2020). CMIP5 models have simulated a SH autumn/winter cooling of between − 0.19 and − 0.36 °C (Harvey et al. 2020). Somewhat stronger cooling responses (< − 0.25 °C) have been modelled over SH mid-latitudes, but with warming over eastern Antarctica following the Agung (1963), El Chichón (1982 and Pinatubo (1991) eruptions (Ménégoz et al. 2018). Although tropospheric responses to major eruptions still require further detailed investigation, Barnes et al. (2016) have identified stratospheric warming in the tropics and cooling over both poles, following the 1991 Pinatubo eruption. While temperatures cooled over most of South America, strongest negative departures were modelled at higher latitudes following the Pinatubo eruption (Colose et al. 2016). Despite recent work on volcanic forcing effects on SH climate, a large knowledge gap still remains, with many questions still unanswered. For instance, comparisons of seasonal climatic responses to such eruptions between inter-continental SH landmasses, to our knowledge, have yet to be investigated.
Given that many factors (e.g. eruption season, location and strength; pre-condition of the atmosphere, aerosol altitude) influence climatic response to volcanic forcing (Oman et al. 2005;Colose et al. 2015;Predybaylo et al. 2017;Stevenson et al. 2017;Zuo et al. 2018;Sun et al. 2019;Krishnamohan et al. 2019), responses (spatially and temporally) are likely to differ between eruptions. To this end, our aim is to establish how individual major volcanic eruptions since 1883 may have affected SH continental temperatures, and specifically how these responses varied spatially (within and between continents) and temporally (during specific seasons) following these eruptions. Using historical simulations from the Coupled Model Intercomparison Project, phase 5 (CMIP5) (Taylor et al. 2012), temperature responses following the eruptions of Krakatau (1883), Santa Maria (1902), Agung (1963) and Pinatubo (1991) are investigated and then briefly compared to 20th Century Reanalysis V2 data in order to establish whether the overestimation of CMIP5 simulated responses are also found in the SH.

Models and data
Climate model simulations from the Coupled Model Intercomparison Project, phase 5 (CMIP5) (Taylor et al. 2012), are utilized for the purpose of this study (see Table 1). We use a subset of 11 different models (Table 1), selected based on their ability to simulate a satisfactory temperature response (in Santer et al. 2013;Supplementary Fig. S1) to Table 1 Models used in this study and the volcanic forcing dataset that each incorporated (Driscoll et al., 2012;Taylor et al., 2012) Ammann et al. (2007) volcanic forcing in the lower stratosphere (see Barnes et al. (2016) for more details). The first three realizations of each model were used, with the exception of the MIROC-ESM-CHEM model, for which only one realization was available, and the GISS-E2_R model for which two different physics versions were used. Both physics version one (p1) and three (p3) were incorporated due to the different ways in which aerosols are calculated (prescribed for p1 and calculated internally in p3). This enabled the analysis of 31 simulations. For this study, an ensemble of the historical experiments is used, which include both natural (volcanic and solar) and anthropogenic climate forcing and cover the period 1850-2005. A volcanic forcing dataset was not prescribed for CMIP5, so models implemented the forcing from either Sato et al. (1993) or Ammann et al. (2007), except for the MRI-CGCM3 model, which calculated it interactively. Notably, climate models do not perfectly simulate responses to volcanic forcing and may either over-or underestimate the response (Neukom et al. 2019). CMIP5 models generally overestimate the response to volcanic forcing (Driscoll et al. 2012;Lehner et al. 2016;Raible et al. 2016). Results thus reflect what the selection of CMIP5 models show and may not necessarily accurately represent actual near-surface temperature conditions. Results do nevertheless provide valuable indications of relative temperature responses to volcanic forcing and offer an opportunity for future comparison with ground-based instrumental records where these exist. We do, however, compare the CMIP5 results with reanalysis data from NOAA/OAR/ESRL, the 20th Century Reanalysis V2 (20CRv2). Each of the model simulations were linearly regridded to a 2° by 2° lat-long grid before having been combined into a multi-model mean. Near-surface (2 m) land-only temperatures are analysed, and all temperatures mentioned in this paper are a SH mean of each continent (Australia: between 44° S and 0° S, 110° E and 160° E; SAF: 40° S and 0° S, 10° E and 40° E; SSA: 60° S and 0° and 70° W and 30° W). For both the CMIP5 and 20CRv2 datasets, annual, seasonal and monthly mean anomalies are presented with respect to a 20-year running mean. So as to calculate statistical significance, a bootstrap approach (Efron 1979) was used to resample the time series of departures 5000 times. To distinguish between random occurrences and significant climatic responses, the 5th and 95th percentiles of the bootstrapped dataset were incorporated to test for significant temperature responses (Haurwitz and Brier 1981;Rao et al. 2017).

Temperature response to volcanic eruptions:
southern South America (SSA)

SSA: Krakatau
Krakatau (6° S, 105° E) erupted on 27 August 1883, placing 22 Tg of SO 2 into the lower stratosphere (Neely III and Schmidt 2016). In SSA, near-surface temperatures responded significantly ( Fig. 1) (Fig. 2). The strongest response is measured for the second winter (− 0.42 °C), and based on a 3-year mean seasonal temperature departure, the strongest responding season was winter (− 0.36 °C), while the weakest was summer (− 0.27 °C). The first month in which a significant below-normal anomaly is simulated in SSA is May 1884 (− 0.24 °C), 9 months after the eruption (Fig. 3a). While 78% of monthly anomalies from the time of the event until the end of y3 were below normal, 32% were significantly below normal.
Cooling occurred during all seasons over all of SSA in y1 and y2 ( Fig. 4 and Fig. S1, S2, S3 in Online Resource 1). Temperatures had recovered over southernmost South America in y3 but remained below normal over the northern regions of SSA during all seasons (apart from autumn  1901 1902 1903 1904 1905 1906 1995 1962 1963 1964 1965 1967 1966 1990 1991 1992 1993 1994 1˚C 0˚C -1˚C Year Year Year Year The strongest 3-year mean seasonal temperature departure was for austral autumn (− 0.23 °C), and the weakest for austral summer (− 0.17 °C), and overall considerably weaker than that for Krakatau. From the time of the eruption until the end of y3, none of the negative monthly anomalies was statistically significant, although 87% of monthly temperature anomalies were below normal ( Fig. 3a).
Temperatures cooled throughout SSA in both autumn and winter, with northern parts of SSA (30-0°S) experiencing strongest cooling (between − 0.20 and − 0.80 °C) (Fig. 4, Fig. S2) in y1 (1903). In y2, temperatures in autumn and winter warmed to above normal (between 0.00 and 0.60 °C) over the central regions of SSA (35-20°S), while those in northern and southern regions were below normal (between − 0.10 and − 0.60 °C). During autumn and winter of y3, temperatures in northern parts of SSA remained below normal (between − 0.10 and − 0.60 °C), while those in the south returned to near normal (between − 0.20 and + 0.20 °C). During summer and spring seasons, the response is much weaker (between − 0.10 and − 0.40 °C), with partial cooling over SSA.

SSA: Agung
Mt Agung (8° S, 116° E) erupted on 17 March 1963 and emitted 7.5 Tg of SO 2 into the lower stratosphere (Neely III and Schmidt 2016). Only austral winter of 1964 and autumn of 1966 had significant negative temperature departures in SSA (JJA y1: − 0.24 °C; MAM y3: − 0.26 °C) (Fig. 2). The greatest single-season response occurred in autumn of y3, and despite this, the strongest 3-year mean seasonal temperature departure was for austral winter (JJA) (− 0.19 °C), while the weakest was for austral summer (− 0.15 °C). The first significant monthly temperature departure is noted for January 1964 (− 0.27 °C), 10 months after the eruption (Fig. 3a). While 91% of monthly temperature anomalies were below normal from the time of the eruption until the end of y3, only 7% were significantly below normal.
Temperatures cooled over portions of SSA during all seasons in y1 (1964) and y2 ( Fig. 4 and Fig. S1, S2, S3 in Online Resource 1), but it was not until y3 when the entire SSA experienced negative temperature anomalies. Contrary to the response simulated after the Krakatau and Santa Maria eruptions, temperatures in y2 after the Agung eruption were warmer than normal in northern parts of SSA (between 0.00 and 0.60 °C) but remained cooler than normal in southern regions (between − 0.10 and − 0.60 °C).

SSA: Pinatubo
Mt Pinatubo (15° N, 120° E) erupted on 15 June 1991, emitting 18 Tg of SO 2 into the stratosphere (Neely III and Schmidt 2016). Each season exhibited significant mean seasonal responses after the Pinatubo eruption in SSA during y1 (Fig. 1). Austral autumn and winter temperatures responded significantly for one season respectively (y1: − 0.32 °C, − 0.34 °C respectively) (Fig. 2). Austral spring and summer experienced significant negative temperature departures for two seasons in y0 and y1 for spring (y0: − 0.23 °C, y1: − 0.24 °C) and y1 and y3 for summer (y1: − 0.28 °C, y3: − 0.26 °C). While the strongest singleseason response occurred in winter, the strongest 3-year mean seasonal temperature departure occurred in austral summer (− 0.25 °C), and the mean 3-season response for all remaining seasons was − 0.21 °C. The first month for which a significant anomaly occurred was March 1992 (− 0.24 °C) (9 months after the eruption) (Fig. 3a). During the period from the eruption date until the end of y3, 81% of monthly anomalies were below normal, while 23% were significantly below normal. Temperatures cooled over all of SSA in y1 during all seasons ( Fig. 4 and Fig. S1, S2, S3 in Online Resource 1). In y2, temperatures were close to normal for autumn and winter, but summer maintained temperatures below normal (between − 0.20 and − 0.80 °C) in the central and southernmost regions of SSA, as also found by Colose et al. (2016). Northernmost temperatures remained below normal (between − 0.10 and − 0.40 °C) during spring of y2. Although temperatures during autumn and winter seemed to recover in y2, they were cooler again in y3 over most of SSA (between − 0.10 and − 0.60 °C).

SSA summary
In SSA, significant negative annual temperature anomalies are detected after all major eruptions since 1883. The longest and strongest annual response is measured after Krakatau, which had significant annual anomalies for 3 years (3-year average of − 0.30 °C), whereas the annual anomalies following the other three eruptions were only significant for y1, with the weakest response occurring after Agung (3-year average: − 0.12 °C). The strongest seasonal response occurred during austral winter for Krakatau and Pinatubo and autumn for Santa Maria and Agung, while the weakest response generally occurred during summer. Significant monthly responses typically first occur 9-10 months after an eruption. Spatially, cooling occurred over most of SSA in y1 (except Agung, following which temperatures in the tropics did not fall below normal until y3). Strongest cooling often occurs between 25° S and northwards except after Pinatubo. Temperatures typically first recovered in southern parts of SSA following the investigated eruptions.

Temperature response to volcanic eruptions:
southern Africa (SAF)

SAF: Krakatau
Following the Krakatau eruption, SAF temperatures were significantly cooler than normal during all seasons in y1 and y2, and in all except autumn in y3 (Fig. 1). The strongest responses appeared in austral autumn and winter during y1 (− 0.54 °C and − 0.53 °C, respectively). Spring and summer of y1 had anomalies of − 0.40 °C and − 0.35 °C, respectively. Autumn and winter also displayed the strongest 3-year mean seasonal negative temperature departures of − 0.41 °C and − 0.40 °C, respectively (Fig. 2), with the weakest 3-season mean occurring in spring − 0.32 °C). During the second and third years, mean temperature anomalies ranged between − 0.26 and − 0.41 °C for all seasons. The first significant monthly temperature response is recorded in February 1884 (− 0.34 °C), 6 months after the eruption (Fig. 3b). Between the eruption event and the end of y3, 93% of monthly temperature anomalies were negative, and 39% were significantly below normal. All of SAF cooled during y1-y3 after the Krakatau eruption. During austral autumn (Fig. 4), winter (Fig. S2) and spring (Fig. S3) of y1, strong cooling (between − 0.40 and − 1.00 °C) occurred throughout the sub-continent. The same strong anomalies continued into y2 over southernmost SAF (Namibia, Botswana and South Africa) during summer and autumn, shifting over central SAF during winter but decreasing in spring of y2. In y3, strong cooling (between − 0.40 and − 0.80 °C) continued over Namibia, Botswana and surrounding regions during summer and over Namibia and western South Africa during winter and spring (between − 0.40 and − 0.60 °C).

SAF: Santa Maria
Temperature anomalies following the Santa Maria eruption, although negative for all seasons, were only significant for austral summer of y0, autumn and spring of y1 and winter of y2 (− 0.28 °C, − 0.48 °C, − 0.29 °C, − 0.32 °C, respectively) ( Fig. 1). Mean winter temperature departures in y1, though not significant, cooled by − 0.30 °C. The strongest seasonal response occurred in autumn of y1 (− 0.48 °C), while the most pronounced 3-year mean seasonal temperature departure also occurred in autumn (− 0.31 °C) and was weakest (− 0.21°) during summer (Fig. 2). The first month demonstrating a significant temperature departure is February 1903 (− 0.23 °C), 4 months after the eruption (Fig. 3b). While 95% of monthly anomalies from the time of the event until the end of y3 were below normal, only 18% were significantly below normal.
Spatially, all of SAF had below normal temperatures for y1 to y3 during all austral seasons except spring (Fig. S3) in y3 when temperatures over South Africa and the western coast of SAF remained above normal (between 0.20 and 0.40 °C). Strongest cooling occurred over most of SAF during autumn (Fig. 4) of y1 (− 0.40 to − 1.00 °C). Notable strong cooling (− 0.40 to − 0.60 °C) occurred between 20 and 30° S in summer of y1, between 10 and 20° S in winter and spring of y1 and between 15 and 25° S during autumn and winter of y2, which extended to southernmost SAF during spring of y2.

SAF: Agung
Following the Agung eruption, temperature departures were significant for 1 year only and for austral autumn (− 0.34 °C), winter (− 0.39 °C) and spring (− 0.30 °C) of y1 and summer of y2 (− 0.25 °C) (Fig. 1). Although the strongest seasonal response in y1 occurred during winter, the strongest 3-year mean seasonal temperature departure is measured for summer (− 0.23 °C), while the weakest 3-season mean departure occurred during spring (− 0.19 °C) (Fig. 2). Negative temperature anomalies are first significant in September 1963 (− 0.22 °C), 6 months after the eruption (Fig. 3b). From the time of the eruption until the end of y3, 82% of monthly temperature anomalies were below normal; however, only 23% were significantly below normal.
All of SAF had negative temperature departures during all seasons in y1 ( Fig. 4 and Fig. S1, S2, S3 in Online Resource 1). Strong autumn and winter cooling occurred over most of SAF and over southernmost SAF during spring (between − 0.40 and − 0.80 °C). All seasonal temperatures, apart from summer, returned to near normal in y2 and y3 (between − 0.20 and 0.20 °C), with summer temperatures remaining below normal (between − 0.20 and − 0.04 °C).

SAF: Pinatubo
SAF temperature anomalies were significantly below normal after the Pinatubo eruption during all austral seasons of y1 and autumn of y2 (DJF: − 0.33 °C; MAM y1: − 0.35 °C; y2: − 0.37 °C; JJA: − 0.35 °C; SON: − 0.31 °C) (Fig. 1). Autumn had the strongest response in y1 and also the strongest 3-year mean seasonal temperature departure (− 0.30 °C), while the weakest 3-year response occurred in summer − 0.23 °C) (Fig. 2). The first month in which a significant temperature departure occurred was February 1992 (− 0.28 °C), 8 months after the eruption (Fig. 3b). During the period from the eruption date until the end of y3, 91% of monthly anomalies were below normal, of which 28% were significantly below normal.
Throughout y1-y3, all seasonal temperatures cooled to below normal over most of SAF ( Fig. 4 and Fig. S1, S2, S3 in Online Resource 1). Strong cooling (− 0.40 to − 0.60 °C) occurred in southernmost regions during summer, autumn and winter of y1, and over the central regions during spring. Strong cooling (− 0.40 to − 0.80 °C) in y2 only occurred during autumn over the central regions (10 to 25° S).

SAF summary
Mean annual temperatures over southern Africa decreased significantly after all four eruptions in y1 and also in y2 and y3 after Krakatau. The strongest response is detected after Krakatau (3-year average: − 0.37 °C), and the weakest followed Santa Maria (3-year average: − 0.26 °C). Seasonal temperature responses are strongest during austral autumn for most eruptions, apart from Agung, after which winter had the strongest response. The weakest responses occurred in either spring (following Krakatau and Agung) or summer (after Santa Maria and Pinatubo). Significant responses typically occurred 4-8 months after an eruption. Spatially, strongest cooling trends occurred over the central or southern regions of SAF (10° S and southwards).

Australia: Krakatau
Following the eruption of Krakatau, temperatures in Australia were significantly cooler than normal during all seasons in y1 (1884) (Fig. 1). Significant anomalies continued into austral summer of y2 and y3 (y2: − 0.31 °C; y2: − 0.34 °C) and winter of y2 (− 0.23 °C) (Fig. 2). The strongest 3-year mean seasonal departures occurred in autumn (− 0.35 °C) and weakest in spring (− 0.26 °C). The first month recording a significant response (− 0.31 °C) was January 1884 (5 months after the eruption) (Fig. 3c). While 95% of monthly anomalies from the time of the event until the end of y3 were below normal, 49% were significantly below normal. With the exception of spring, cool anomalies occurred over the entire continent during all other seasons in 1884 ( Fig. 4 and Fig. S1, S2, S3 in Online Resource 1). Strong cooling (− 0.40 to − 0.80 °C) commences in southern parts of Australia during summer of y1, which expands to the whole continent and intensifies (− 0.40 to − 1.20 °C) during autumn but then starts to weaken and spatially retract during winter. However, the cooling effect is still notable in northern regions during winter and spring of y1. Temperature anomalies are weaker in y2, with some being above normal during autumn (0.20 °C). Strong cooling (− 0.40 to − 0.60 °C) occurs again in summer and autumn of y3 over central and eastern Australia.

Australia: Santa Maria
After the Santa Maria eruption, Australian temperature anomalies were below normal during austral autumn and winter of y1 (MAM: − 0.31 °C; JJA: − 0.35 °C), spring of y2 (− 0.27 °C) and winter of y3 (− 0.22 °C) (Fig. 1). Summer anomalies were not significant but still cooled by − 0.22 °C in y1 (Fig. 2). Winter had the strongest modelled response in y1 as well as the strongest 3-year mean seasonal temperature departure (− 0.25 °C), with summer recording the weakest 3-year mean departure (− 0.17 °C). Although negative temperature anomalies are modelled for the same month in which Santa Maria erupted, significant cooling (− 0.25 °C) over Australia is only apparent from 6 months (April 1903) after the eruption (Fig. 3c). From the time of the eruption until the end of y3, 97% of monthly temperature anomalies were below normal, but only 20% were significantly below normal.
After Santa Maria, cooling occurred over all of Australia during all seasons of y1 ( Fig. 4 and Fig. S1, S2, S3 in Online Resource 1). Strong cooling (between − 0.40 and − 0.60 °C) is measured over eastern parts of Australia in autumn and spring and over western regions in summer and winter of y1. In y2, most of the continent still experienced below normal anomalies, except during summer when temperatures returned to normal over parts of Australia (max 0.20 °C). Strong cooling (− 0.20 to − 0.80 °C) only occurred over southwestern regions during spring. Notable negative temperature departures (− 0.40 to − 0.80 °C) occurred in eastern regions of Australia, while above normal anomalies (up to 0.40 °C) occurred in the western regions during summer of y3.

Australia: Agung
Australian spring temperatures were significantly lower than normal during the same year that Agung erupted (y0: − 0.42 °C) (Fig. 2). Temperatures then remained significantly below normal during all seasons in y1 (DJF: − 0.38 °C; MAM: − 0.40 °C; JJA: − 0.40 °C; SON: − 0.31 °C) (Fig. 2). Although the strongest single-season response is austral spring of y0, the strongest 3-year mean seasonal temperature departure occurs in autumn (− 0.22 °C), and the weakest is modelled for winter (− 0.14 °C). The first month recording a significantly below normal temperature was October 1963 (− 0.23 °C), 7 months after the eruption (Fig. 3c). While 80% of monthly temperature anomalies were below normal from the time of the eruption until the end of y3, 34% were significantly below normal.
All of Australia cooled during all seasons in 1964 ( Fig. 4  and Fig. S1, S2, S3 in Online Resource 1). There were strong negative temperature departures (− 0.40 to − 0.80 °C) in southeastern regions throughout all seasons. This strong cooling also occurred throughout the south of Australia in summer, throughout the east of Australia during autumn and most of Australia in winter. Below normal anomalies remained throughout Australia in y2, except in summer when some regions were above normal (max 0.20 °C). Temperature anomalies returned to near normal (± 0.20 °C departures) during all seasons of y3.

Australia: Pinatubo
Significant cooling is measured over Australia after the Pinatubo eruption during austral spring of y0 (− 0.52 °C) and autumn (− 0.33 °C), winter (-0.39 °C) and spring (− 0.29 °C) of y1 (Fig. 1). Although spring of y1 experiences the strongest negative anomaly, the strongest 3-year mean seasonal response occurred in both summer and winter (− 0.23 °C), and the weakest in both autumn and spring (− 0.21 °C) (Fig. 2). The first significant negative anomaly is recorded in September 1991 (− 0.29 °C), 3 months after the eruption (Fig. 3c). From the time of the eruption until the end of y3, 95% of monthly anomalies were negative, and 40% significantly below normal.
Below normal anomalies occurred throughout Australia during y1-y3 for all seasons except autumn during which temperatures in the west were above normal (between 0.20 and 0.40 °C) in y1 and y2 ( Fig. 4 and Fig. S1, S2, S3 in Online Resource 1). However, autumn also experienced strong cooling (− 0.40 to − 0.80 °C) over northeastern regions in y1. Similar anomalies are modelled for eastern areas of Australia during the summers of y1 and y3, over central and northern areas in winter of y1 and northern areas during spring of y1.

Australia summary
Mean annual temperatures decreased significantly for 1 year across Australia following each of the major eruptions investigated and 3 years following Krakatau. The strongest (negative) mean annual temperature departures occurred after Krakatau (3-year average: − 0.31 °C) and weakest after Santa Maria (3-year average: − 0.20 °C). Strongest 3-year negative temperature anomalies were for austral autumn (after Krakatau and Agung) and winter (after Santa Maria and Pinatubo). Significant negative departures typically occurred 3-7 months after the eruptions. The entire continent cooled during all seasons of y1. Near normal or above normal temperatures seem to occur more frequently over Australia than SSA or SAF, but no uniform pattern is apparent.

Summary: SH temperature responses to volcanic forcing
Notably, strongest and most widespread SH continental negative temperature anomalies following the four major eruptions occur in austral autumn and to a lesser extent in austral winter, with exceptions following the Agung and Pinatubo eruptions when Australia experienced strongest responses in austral spring. This is in contrast to NH temperatures which warm in mid-high latitude regions during boreal winter (Zambri and Robock 2016;Zambri et al. 2017). While SH land temperatures do not depict warming, there is very little land mass at mid-high SH latitudes. However, Fig. 4 and Fig. S1-S3 show a warming in surface temperatures over the oceans in the mid-high latitudes during all seasons. Over continental regions of the SH (excluding Antarctica), initial significant negative temperature departures following major volcanic eruptions occur most rapidly (within 5 months on average) over Australia, followed closely by SAF (6-month average), but are somewhat more delayed over SSA (9 months on average). Typically, the first significant response is seen between January and May for SSA, in February in SAF and between October and April over Australia. Temperature responses following all eruptions except Santa Maria (for which the response was strongest over SAF) are strongest over Australia and weakest over SSA. However, the mean response of all eruptions is equally strong for both SAF and Australia. The overall percentage of months recording significant negative temperature anomalies over the initial 3 years following major eruptions was highest over Australia (20-49%) and lowest over SSA (0-32%).

CMIP5 comparison to reanalysis data
CMIP5 data were compared to the 20CRv2 reanalysis data (Fig. 5 and Fig. S4, S5, S6 in Online Resource 1). We find that most CMIP5 values, both annual and seasonal, are stronger (cooler) than the 20CRv2 values by an average 0.30 °C (Table 3). Following the Krakatau eruption, all CMIP5 values are cooler (on average by 0.37 °C) than the 20CRv2 values, except during JJA of y1 in SAF and SON of y3 in Australia when the reanalysis temperatures are cooler (by an average of 0.12 °C). After the Santa Maria eruption, the CMIP5 annual anomalies are only warmer than the reanalysis data in y2 and y3 in Australia (by − 0.03 °C and − 0.17 °C, respectively). Seasonal values show that this occurs during DJF, MAM, JJA of y2 and JJA and SON of y3 in Australia but also in SON of y1, and DJF and MAM of y2 in SAF. Modelled CMIP5 anomalies are more often cooler than the 20CRv2 anomalies following the Agung eruption, particularly in SAF in y1 and y2 (annual anomalies by 0.07 °C and 0.24 °C, respectively), and in Australia in y3 (by 0.29 °C). This also applies to SSA during JJA and SON of y1 (by 0.29 °C and 0.20 °C, respectively) and MAM of y2 (by 0.17 °C). Following Pinatubo, CMIP5 annual anomalies are stronger than the 20CRv2 values in Australia during y1 and y2, in DJF (by 0.09 °C and 0.49 °C, respectively) and SON (by 1.20 °C and 0.44 °C, respectively), but also during JJA in SAF (by 0.01 °C and JJA and SON in SAF (by 0.03 °C and 0.04 °C, respectively). There are very few 20CRv2 temperature departures that are significantly below normal following the investigated eruptions. Although no annual anomalies are significantly cooler, SON of y1 following the Pinatubo eruption is significantly cooler over Australia (− 1.49 °C). In SAF, the only significant temperature anomalies based on the 20CRv2 data occur after the Agung eruption, during MAM of y0 (− 0.71 °C) and JJA (− 0.72 °C) and SON (− 1.12 °C) of y1, while no significantly cool anomalies are recorded over SSA.

Discussion
Modelled temperature responses were for the most part weakest for Santa Maria (except in SSA, where the response after Pinatubo was weakest). Such outcomes might be a product of the eruption time (Stevenson et al. 2017;Sun et al. 2019). For example, circulation in the SH is more favourable to aerosol distribution during winter due to strengthening of the Brewer Dobson circulation in both the deep and shallow branches in winter (Perlwitz and Graf 1995;Birner and Bönisch 2010;Young et al. 2011). Agung erupted before austral winter, and Pinatubo during austral winter, allowing for better aerosol distribution soon after the eruptions than Santa Maria, which erupted after the austral winter season when aerosol distribution may have been weaker. Differences in the extent of climatic response between Pinatubo and Santa Maria might also be due to the height at which aerosols reached the stratosphere. Aerosols from the Pinatubo eruption were likely dispersed at a higher altitude given that the plume reached a considerably higher (c. 40 km) altitude than that for Santa Maria (c. 23.7-27 km) (Bonadonna and Costa 2013). Recent work by Krishnamohan et al. (2019) advocates that the higher the altitude of aerosol injection, the greater the potential cooling effect, which would support greater cooling associated with Pinatubo than Santa Maria. However, the space-time spread of aerosols may not have as much of an influence in the CMIP5 model outputs.
Volcanic forcing influences atmospheric circulation even into the stratosphere by affecting stratospheric temperature gradients and wind speed-i.e. the jet streams (Cui et al. 2014;Kidston et al. 2015). Jet streams can affect the Southern Annular Mode (SAM) (Karpechko et al. 2010;Xia et al. 2015), planetary waves, westerly circulation and storm tracks and hence influence surface weather (Gallego et al. 2005;Hudson 2012;Kidston et al. 2015). The SH jet stream may seasonally shift by c. 10° (Tyson et al. 2000;Shulmeister et al. 2004;Gallego et al. 2005;Hudson 2012). Although knowledge on how volcanic forcing impacts the jet streams is still limited, it is understood that jet streams were impacted by both the 1982 El Chichón and 1991 Pinatubo eruptions in the NH (Hudson 2012). Hudsons' results indicate that the subtropical jets in both hemispheres initially shifted poleward, and subsequently equatorward, after both eruptions. The study additionally found that the SH polar jet moved equatorward after the Pinatubo eruption. With mid-latitude circulation being affected by volcanic forcing through major SH circulation variability modes (such as SAM, ENSO, jet streams), it seems that major volcanic eruptions affect mid-latitude cyclones (Picas and Grab 2020). An increased intensity and frequency of cold fronts (associated with mid-latitude cyclones) would unquestionably provide for enhanced cooling over SH continental regions affected by such frontogenesis (see Grab and Simpson 2000). Topographic differences between the SH continental land masses (i.e. high Andes mountain range along western SSA versus relatively flat relief across western, southern and central Australia) would influence the northerly/easterly spread of such cold air wave disturbances. This may, at least in part, Austral Autumn (MAM) Near-Surface Temperature Anomalies explain why cooling is most enhanced over Australia and least so over much of central SSA, following major eruptions. The likely enhanced frequency and/or intensity of mid-latitude cyclones could consequently enhance cooling over southern to mid-latitude continental regions, particularly during austral autumn-winter-spring. Southwestern Western Australia, southeastern Australia (Smith et al. 1982;Buckley and Leslie 2004) and southern SAF (Tyson et al. 2000) are affected by cyclones/cold fronts under normal conditions, which coincidently are areas of strongest volcanically induced cooling anomalies during austral autumn-winter-spring. Such cold surges in SSA are known to bring frost and freezing temperatures to Brazil, oftentimes damaging crops (Vera and Vigliarolo 2000; Espinoza et al. 2013). The cold surges are also responsible for transporting dryer air northwards, which at times has led to drought, as was the case with the severe 1817 drought over Brazil, soon after the 1815 Tambora eruption (Garreaud 2001). Comparison between the CMIP5 data outputs and 20CRv2 reanalysis data suggests that the CMIP5 models have overestimated the cooling response to volcanic eruptions over SH continental regions of SSA, SAF and Australia. The fact that CMIP5 overestimates the response to volcanic forcing has also been noted by Driscoll et al. (2012), Marotzke and Forster (2015) and Chylek et al. (2020). Further, Barnes et al. (2016) suggest that internal climate variability may cause the response to volcanic eruptions seen in the CMIP5 models, to be disguised in the observed response. While the reanalysis data successfully represent most SH atmospheric circulation during the period 1979-2010, data pre 1979 have greater discrepancies due to the reduced number of available observational stations (Ke and Hui 2013;Zhang et al. 2013b). However, while acknowledging that CMIP5 models might not accurately represent SH circulation responses to volcanic forcing, McGraw et al. (2016) argue that the observations do fall within the distributions of modelled SH circulation, in particular the Southern Annular Mode. There is hence much opportunity for CMIP6 to improve the modelled response to volcanic forcing.

Conclusion
Until recently, little attention has been given to volcanic forcing impacts on climates of the SH, with a general perception that such forcing has had relatively minimal impact on the SH in relation to its northern counterpart. Here, we have examined the response of SH temperatures to such forcing at a finer spatial and temporal resolution than earlier approaches (i.e. at continental rather than hemispheric scales and seasonal rather than annual scales). Investigations at finer scales demonstrate that (significant) negative temperature departures over SH land masses typically occur first over Australia (3-7 months) and lag most strongly over SSA (9-10 months). In addition, strongest cooling is simulated over SAF (av = − 0.27 °C) and weakest responses simulated over SSA (av = − 0.21 °C). Strongest responses generally occur during austral autumn/winter and are usually weakest during spring/summer. While all inter-continental regions of the SH responded with strongest cooling following the Krakatau eruption, some spatial disparity in the strength and timing of responses is noted for twentieth-century eruptions. This illustrates the importance of assessing climate responses to volcanic eruptions at fine rather than course spatial and temporal scales.