Tao WANG, Jiapeng MIAO, Huijun WANG,,4, and Jianqi SUN

1 Climate Change Research Center, Institute of Atmospheric Physics, Chinese Academy of Sciences, Beijing 100029

2 Collaborative Innovation Center on Forecast and Evaluation of Meteorological Disasters, Nanjing University of Information Science &Technology, Nanjing 210044

3 Nansen–Zhu International Research Center, Institute of Atmospheric Physics, Chinese Academy of Sciences, Beijing 100029

4 Key Laboratory of Meteorological Disaster, Nanjing University of Information Science &Technology, Nanjing 210044

ABSTRACT This study investigates the influences of strong tropical volcanic eruptions (SVEs) on daily temperature and precipitation extreme events using long-term simulations from the Hadley Centre Coupled Model version 3 (HadCM3) and the Community Earth System Model version 1.1 (CESM1). The results indicate that the occurrences of daily hot extremes and daily heavy precipitation extremes decrease over most parts of the world in the peak forcing years of SVEs. Due to the volcanic cooling effect, the average probability of daily hot extremes decreases by approximately 50% across the globe. The decrease in intensity is stronger for midlatitude land regions and tropical South America.In contrast, daily cold extremes occur more frequently over most parts of continental regions. Globally, a cold extreme event expected once every 3 years under non-volcanic conditions can be expected every 1.5 years on average in the peak forcing years. Overall, the SVE-induced cooling effect plays a dominant role in regulating daily cold and hot extremes. Over high-latitude Eurasian regions, in contrast to other continental regions, the probability and intensity of daily cold extremes decrease due to an SVE-strengthened polar vortex and the associated temperature advection anomalies. With regard to daily heavy precipitation extremes, the probability and intensity both decrease over most monsoon areas. Further analysis suggests that the reduced probability and intensity of daily heavy precipitation extremes are mainly due to the SVE-induced global decrease in the water-holding capacity.

Key words: strong tropical volcanic eruption (SVE), daily extremes, temperature, precipitation

1. Introduction

Volcanic eruptions, which can have adverse impacts on ecosystems, people’s lives and property, and economies, are recognized as one of the most catastrophic types of natural disasters. Additionally, they have significant influences on the local weather. In particular, after a strong volcanic eruption, the sulfur gas can be ejected into the stratosphere, where it is converted into sulfate aerosols, which can cause global climate change over timescales ranging from seasonal to decadal (e.g., Robock,2000; Wang et al., 2012; Zanchettin et al., 2012, 2013).Based on observational and modeling studies (e.g., Robock, 2000; Otter?, 2008), volcanic sulfate aerosol is known to cool the surface and troposphere by reflecting and scattering incoming solar radiation, whilst at the same time, it heats the stratosphere by absorbing incoming near-infrared and outgoing longwave radiation.Therefore, the volcanic forcing can lead to regional or global climate changes through inducing energy imbalances in the climate system.

Notably, strong tropical volcanic eruptions (SVEs)have particularly wide impacts on climate (Timmreck,2012). For instance, numerous studies have noted that SVEs can weaken the East Asian and African summer monsoons (e.g., Oman et al., 2006; Cui et al., 2014) and strengthen the East Asian winter monsoon (Miao et al.,2016). As a result, via the anomalous atmospheric circulation and associated water vapor transport, monsoonal precipitation is reduced in the first year after such an eruption (e.g., Peng et al., 2010; Joseph and Zeng, 2011;Liu et al., 2016; Zambri and Robock, 2016). Over ocean areas, the precipitation response has been shown in simulations to last longer than that over land (Iles et al., 2013;Iles and Hegerl, 2014). Additionally, SVEs can lead to a strong polar vortex and a positive phase of the Arctic Oscillation/North Atlantic Oscillation (AO/NAO) by enhancing the stratospheric meridional temperature gradient(e.g., Stenchikov et al., 2002, 2006; Fischer et al., 2007).Moreover, there is evidence to suggest that SVEs can cause the El Ni?o–Southern Oscillation (ENSO), which is one of the most important climate modes, to shift into a positive phase (e.g., Emile-Geay et al., 2008; Lim et al.,2016; Khodri et al., 2017; Liu et al., 2018; Wang et al.,2018) and then rebound into a negative phase (Adams et al., 2003; Wang et al., 2018). It has been suggested that SVEs have long-term regulatory effects on the evolution of ENSO (Wang et al., 2018).

At a longer timescale, it has been suggested that SVEs might modulate the phases of the Atlantic Multidecadal Oscillation (Otter? et al., 2010), Pacific Decadal Oscillation (Wang et al., 2012), fluctuations of Arctic sea-ice extent (Slawinska and Robock, 2018), and other largescale climate modes (Zanchettin et al., 2012). Through these processes, volcanic signals and related influences can be retained in the climate system for long periods.Timmreck et al. (2016) found that a volcanic imprint on decadal variability was detectable at a regional scale. Indeed, a modeling study even suggested that the abrupt onset of the Little Ice Age was triggered by a spate of large sulfur-rich explosive eruptions (Miller et al., 2012).Therefore, volcanic eruptions are one of the most important natural external forcings and play a significant role in regulating global climate.

Recently, extreme temperature and precipitation events have started to occur more frequently, and with severe impacts on natural ecosystems and society substantial socioeconomic losses (Field et al., 2012; Jiang et al., 2012, 2015; Sun et al., 2014; Chen and Sun, 2015;Donat et al., 2016; Lu and Chen, 2016). Accordingly,studies on the attribution of extreme climate events have become increasingly more common since the turn of the century (Mitchell et al., 2001; Chen and Sun, 2017a), attracting a great deal of interest within and outside the scientific community. More specifically, much attention has been paid to examining the impacts of humans on climate extremes, with a large number of studies suggesting that anthropogenic forcing has probably contributed to the observed increase in the frequency of extreme temperature and precipitation events (e.g., Min et al., 2011;Stocker et al., 2013; Zhang et al., 2013; Chen and Sun,2017b; Paik et al., 2020a). However, how climate extremes respond to volcanism, particularly SVEs, has received less attention and needs to be studied in more depth. Paik and Min (2018) assessed the impacts of volcanic eruptions on climate extremes using CMIP5(Coupled Model Intercomparison Project Phase 5) historical simulations from 1850 to 2005 and found that the intensities of extreme temperature and precipitation both decreased in response to volcanic forcing. However, other types of external forcing (e.g., the rapid increases in greenhouse gas concentrations, regional anthropogenic aerosols, and transient solar radiation) were also included in these historical simulations, which can also have an influence on regional or global extremes. As such, how volcanic eruptions influence the frequency of the climate extremes still needs further investigation.

In this paper, we investigate the influences of SVEs on daily temperature and precipitation extremes on the global scale using eight long-term volcanic forcing simulation outputs from the Hadley Centre Coupled Model version 3 (HadCM3) and the Community Earth System model version 1.1 (CESM1). In particular, we focus on the changes in both the frequency and intensity of extreme climate events due to volcanic forcing. Following this introduction, Section 2 provides information on the models and long-term simulations, and the methods are described in detail. In Section 3, the influences of SVEs on climate extremes are investigated, and then conclusions are presented and discussed in Section 4.

2. Model, simulations, and methods

Long-term simulations from HadCM3 and CESM1 are used to investigate the influences of volcanic forcing on daily temperature and precipitation extremes. Both Had-CM3 (Gordon et al., 2000) and CESM1 (Otto-Bliesner et al., 2016) are coupled ocean–atmosphere models. The atmospheric component of HadCM3 has a horizontal resolution of 2.5° latitude × 3.75° longitude, while its oceanic component has a grid resolution of 1.25° × 1.25° and a total of 20 vertical levels. The version of CESM1 used here has a resolution of ～2° (1.9° × 2.5°) in the atmospheric component and of ～1° in the ocean and sea-ice components. Its resolution is higher than that of HadCM3.

In this paper, we analyzed eight long-term simulations. Three are from HadCM3, covering from 1401 to 1999. They were initialized with the ocean conditions in model-year 1400 derived from a long-term all-forced simulation, but with different atmospheric conditions near model-year 1400. In this experiment, the three simulations included only transient volcanic forcing during this period, which was based on the reconstructed volcanic aerosol optical depth (AOD; Fig. 1a) in Crowley et al. (2008). The other types of forcing (i.e., solar forcing,well-mixed greenhouse gases, land use, and orbital forcing) were set equal to the conditions in 1400, while the ozone forcing was set to preindustrial levels. Additionally, five simulations from CESM1 are analyzed, covering a longer period from 850 to 2005. As with the Had-CM3 simulations, these five long-term simulations had different initial atmospheric conditions, and the only forcing present was transient volcanic forcing. The ozone forcing was set to preindustrial levels, and the other types of external forcing were set to their values in the year 850. The volcanic forcing used in the CESM1 simulations was different; it was from ice-core volcanic index 2(Fig. 1b; Gao et al., 2008). In total, these eight long-term simulations included 7577 model years. Daily surface air temperature (SAT), daily precipitation, daily maximum/minimum SAT, and related monthly data from these long-term simulations are analyzed in this paper.

Fig. 1. Global averaged volcanic forcing: (a) aerosol optical depth (AOD) from Crowley et al. (2008) and (b) global total volcanic aerosol (Tg)from Gao et al. (2008). The strong tropical volcanic eruptions (SVEs) analyzed in this paper are marked by the red arrows.

In Crowley et al. (2008), the volcanic forcing (i.e., reconstructed AOD) was supplied in four bands (30o–90oN,0–30oN, 0–30oS, and 30o–90oS), and employed in the HadCM3 simulations. In contrast, the volcanic forcing in Gao et al. (2008) comprised ice-core-derived estimates of aerosol loadings as a function of latitude, altitude, and month. In the CESM1 simulations, volcanic aerosol was prescribed as a fixed, single-size distribution in the three layers in the lower stratosphere. In the present analysis,an SVE is defined as an eruption located in the tropics,i.e., the tropical mean (30oN–30oS) volcanic forcing is greater than both the Northern and Southern Hemispheric extratropical (30o–90oN/30o–90oS) mean forcings. At the same time, the intensity of an SVE is defined by the annual globally averaged AOD being 0.05 larger than that in Crowley et al. (2008) (for the HadCM3 simulations) or by the annual global total stratospheric sulfate aerosol injection being 14 Tg larger than that in Gao et al. (2008)(for the CESM1 simulations). This selection mainly considers the intensities of SVEs observed in the 20th century: the Agung eruption in 1963 (AOD 0.05, 17 Tg), the El Chichon eruption in 1982 (AOD 0.05, 14 Tg), and the Pinatubo eruption in 1991 (AOD 0.13, 30 Tg). According to this threshold, there are 18 SVEs in each HadCM3 simulation and 32 SVEs in each CESM1 simulation.Generally, the volcanic forcing can last for 2–3 yr after strong eruptions. In this paper, the strongest forcing years(i.e., the peak forcing years) for each SVE are analyzed.Therefore, 18 and 32 peak forcing years were selected from each of the HadCM3 and CESM1 simulations (for a total of 54 peak forcing years in the HadCM3 simulations and 160 peak forcing years in the CESM1 simulations). In the HadCM3 simulations, the selection of the peak forcing years was based on the maximum reconstructed annual tropical AOD in Crowley et al. (2008).Due to the starting months of SVEs being different among the SVEs in the HadCM3 simulation, the peak forcing year was often the first year after the eruption year. However, the situation was different for the CESM1 simulations, in which the starting month for the majority of eruptions was assumed to be April (Otto-Bliesner et al., 2016) and the peak forcing period was usually from May of the eruption year to the following April. Therefore, we defined this period as the peak forcing year of the SVE in the CESM1 simulations. In this paper, we mainly examine the responses of climate extremes and related changes in mean climate in the peak forcing years relative to the periods without volcanic forcing.

To investigate changes in the intensity and frequency of climate extremes during the years of peak volcanic forcing, six indices were used in this study. In terms of the intensity of extreme climate, the annual warmest day(TXx) and coldest night (TNn), as well as the maximum 5-day consecutive precipitation amount (Rx5day), are examined in the peak forcing years and no volcanic forcing years. These three climate extreme indices were developed by the Expert Team on Climate Change Detection and Indices (ETCCDI; http://etccdi.pacificclimate.org/list_27_indices.shtml). Note that TXx and TNn are measured by the absolute values of SAT, and thus, daily hot extremes are only going to be found in local summer and daily cold extremes in local winter. To better understand the possible mechanisms responsible for changes in daily hot/cold extremes, the corresponding seasonal mean climates are also examined, wherein boreal winter(i.e., austral summer) refers to as the first post-eruption December–January–February in both the HadCM3 and CESM1 simulations and boreal summer (i.e., austral winter) of June–July–August in the peak forcing year.

In terms of frequency, following Fischer and Knutti(2015), we define an extreme temperature (or precipitation) event as one that exceeds the 99.9th percentile of the daily temperature (or precipitation). In the model, the corresponding threshold is estimated at each individual grid points. To avoid any influence from volcanic forcing on the thresholds, we calculated them across all days of no volcanic forcing years. The extreme events of interest herein include daily cold extremes, daily hot extremes, and daily heavy precipitation extremes. Taking daily hot extremes as an example, all days of no volcanic forcing years were sorted by temperature. Then, the value of highest 0.1% quantile was defined as the threshold at that grid point, meaning that the hot extreme event could be expected to occur once in every 3 years,approximately (once every 1000 days). In the analysis of daily heavy precipitation extremes, the days without rain are excluded. The probability ratio (PR) is defined asP1/P0, whereP0is the probability of exceeding the threshold during the reference period andP1is the probability of exceeding the same threshold during a given period (Stott et al., 2004). In this study, the no volcanic forcing years were chosen as the reference period. Thus,P0was equal to 0.1% for all the grid points. The peak forcing years were used as the given period. Therefore,the PR indicates the probability that a daily extreme event has changed due to volcanic forcing. A PR > (<)1.0 indicates a higher (lower) likelihood of occurrence for that particular climatic extreme. Additionally, composite analysis was employed in this study, and the statistical significance was assessed by using the Student’sttest. The statistical significance of PR was determined by a standard Monte Carlo randomization procedure(1000 times).

3. Results

3.1 Changes in daily cold/hot extremes

During the peak forcing years, the annual SAT over most of the world decreases significantly, especially over land regions (Fig. 2). The decreased annual SAT over most mid–low-latitude land regions and high-latitude North America exceeds 0.5°C. The cooling over the oceans is approximately 0.2°C. In addition, in the Northern Hemisphere, the cooling over midlatitude land regions is stronger, which is consistent with many other models’ results (Chai et al., 2020). SVE-induced surface cooling also increases the probability of daily cold extremes over most of the world (Figs. 3a, c). From a global average perspective, PR increases by factors of 2.42 and 2.89 in the HadCM3 and CESM1 simulations, respectively (Table 1). This implies that over land, a cold extreme event expected once every 1000 days (approximately 3 yr) under normal conditions can be expected on average once every year. Additionally, the globally averaged TNn over land in the HadCM3/CESM1 simulations decreases by 0.54/0.26°C in the peak volcanic forcing years relative to no volcanic forcing years (Table 2).Overall, both the frequency and the intensity of daily cold extremes increase due to volcanic forcing.

Fig. 2. Simulated differences in (a) annual, (c) boreal winter, and (e) boreal summer SAT (°C) between the volcanic peak forcing years and no volcanic forcing years in the HadCM3 simulations. (b, d, f) As in (a, c, e), but for CESM1 simulations. Areas that exceed the 95% confidence level are denoted by dots.

Due to the lower internal variability of the climate over the tropics, the volcanic signal is stronger. Therefore, the increasing probability of daily cold extremes is generally higher in the tropics than at other latitudes. The average PR is greater than 3 for tropical land regions(3.68/4.81 in Table 1). This means that an extreme cold event expected once every 3 years under normal conditions can be expected to occur more than once in a peak volcanic forcing year. Over tropical Africa and tropical South America, PR even increases by a factor ranging from approximately 5 to 20 (Figs. 3a, c), which suggests that the frequency of daily cold extremes increases by 5 to 20 times. Daily cold extremes can occur more than twice in a peak volcanic forcing year in these regions.The TNn values of the tropical regions show responses similar to those of the PRs (Figs. 4a, c). On average, TNn decreases by approximately 0.5°C. Particularly in the HadCM3 simulations, the decrease of TNn in tropical regions is larger.

Table 1. Averaged probability ratio (PR) values in the HadCM3/CESM1 simulations over land regions in the peak eruption years for daily cold/hot/heavy precipitation extremes exceeding the 99.9th percentile of no volcanic forcing years

Table 2. Averaged differences (°C) in TNn, TXx, and annual mean SAT in the HadCM3/CESM1 simulations over land regions between the volcanic peak forcing years and no volcanic forcing years

In midlatitude continental regions, the probability of daily cold extremes also increases by a factor of approximately 2 in northeastern Asia and eastern Europe during the peak volcanic forcing years in the HadCM3 simulations. However, it changes little over North America. Unlike the simulated PR in the HadCM3 simulations, the CESM1-simulated probability of daily cold extremes increases significantly for North America and 40° south of the Asian landmass. Nevertheless, the probability of daily cold extremes shows no change or even decreases for Northeast Asia. On average, PR increases by a factor of approximately 1.4 for midlatitude land regions in the Northern Hemisphere. In terms of the intensity of daily cold extremes, the HadCM3-simulated TNn decreases by 0.59°C for the Northern Hemispheric midlatitude regions, whereas the CESM1-simulated TNn only decreases by 0.21°C (Table 2). The main reason for this difference is that the CESM1-simulated TNn values increase significantly over Northeast Asia, which implies that the daily cold extreme is relatively weaker. The maximum increase in TNn can exceed 1°C (Fig. 4c).However, for North America, the CESM1-simulated TNn values decrease by more than 1°C, which means that this region experiences the strongest changes in daily cold extremes after an SVE.

Fig. 3. Simulated spatial distributions of probability ratio (PR) in the volcanic peak forcing years for (a) daily cold extremes and (b) daily hot extremes in HadCM3. (c, d) As in (a, b), but in CESM1. Areas that exceed the 95% (90%) confidence level are denoted by white dots in the dark blue areas and black (small) dots in the light shaded areas.

For Antarctica and Greenland, the two models simulate different responses in terms of the frequency and intensity of daily cold extremes following an SVE. In the HadCM3 simulations, the frequency and intensity of daily cold extremes exhibit no significant changes. In contrast, the probability of daily cold extremes in the CESM1 simulations increases by a factor of approximately 1.5 in most parts of these two regions. The TNn values also decrease by approximately 0.3°C over Antarctica (Table 2), and by much more over Greenland(Fig. 4c). However, the two models simulate consistent responses in terms of daily cold extremes in Siberia. The probability of daily cold extremes decreases by approximately 50% and TNn increases by approximately 0.5°C in the HadCM3 simulations, and by more than 1°C in the CESM1 simulations (Figs. 3, 4). In particular, the CESM1-simulated positive TNn anomalies extend throughout the high-latitude Eurasian land region and Northeast China,and increase by approximately 1°C in most regions (Fig.4c). This suggests that the intensity of daily cold extremes is likely to decrease in high-latitude Eurasian land regions and that the probability of daily cold extremes decreases significantly over Siberia in the peak forcing years. Actually, the responses of the daily cold extremes are consistent with those of the mean SAT in the highlatitude Eurasian land regions. In boreal winter during peak forcing years, weaker cooling or even warming can be seen in high-latitude Eurasian land regions (Figs. 2c,d). In the CESM1 simulations in particular, the warming is much stronger. Based on observational and modeling studies (e.g., Robock, 2000; Zanchettin et al., 2012;Wang et al., 2012), we know that an SVE can lead to a stronger polar vortex and produces a characteristic stationary wave pattern of tropospheric circulation. It can decrease the amount of cold air from the polar region. At the same time, the associated stronger zonal winds advect warmer maritime air over the continents (Robock and Mao, 1992). In the HadCM3 and CESM1 simulations, a strengthened polar vortex and anomalously positive AO are evident in the peak forcing years (Fig. 5).Similar low-level circulation responses can be found in high-latitude Eurasia. In the CESM1 simulations in particular, significant southwesterly wind anomalies are evident over northern Europe and Siberia. During boreal winter, the incoming radiation is very low over the highlatitude regions of the Northern Hemisphere. The SVEinduced cooling effect is very weak over there. Thus,these southwesterly wind anomalies can cause a reduction in polar cold-air advection and an enhancement of maritime warm-air advection, which together play a dominant role in increasing the mean winter SAT over high-latitude Eurasian continental regions. As a result,the frequency and intensity of daily cold extremes decrease, showing lower TNn and PR values for some parts of high-latitude Eurasia. In fact, a similar winter warming and strengthened polar vortex can be found only in a small number of models (Zambri and Robock, 2016;Zambri et al., 2017), being underestimated by most (Liu et al., 2020). Stronger responses of global circulation to the SVEs are possibly an important reason why the models simulate winter warming over the high-latitude regions of continental Eurasia (Fig. S1, available online).

Fig. 4. Simulated differences in (a) TNn (°C) and (b) TXx (°C) between the volcanic peak forcing years and no volcanic forcing years in Had-CM3. (c, d) As in (a, b), but in CESM1. Areas exceeding the 95% confidence level are denoted by white dots in the dark blue areas and black dots in the light shading areas.

Fig. 5. Simulated differences in boreal winter (a) sea level pressure (hPa), (c) zonal wind (m s–1) at 50 hPa, and (e) wind field (m s–1) at 850 hPa between the volcanic peak forcing years and no volcanic forcing years in the HadCM3 simulations. (b, d, f) As in (a, c, e), but for the CESM1 simulations. Areas that exceed the 95% confidence level are denoted by dots or gray shading.

Fig. 6. Simulated differences in annual total precipitation (mm) between the volcanic peak forcing years and no volcanic forcing years in (a)HadCM3 and (b) CESM1. Areas that exceed the 95% confidence level are denoted by dots.

For daily hot extremes, the PR decreases for most land regions, which implies a lower probability of hot extremes in the peak forcing years after SVEs (Figs. 3b, d).The global PRs are 0.33 and 0.53 in the HadCM3 and CESM1 simulations, respectively, which means that the frequency of daily hot extremes decreases by more than 60% and 40%, respectively, in these two models.Moreover, the probability of a daily hot extreme decreases for tropical land regions. In particular, the PR is zero for most parts of tropical South America in the Had-CM3 simulations, which suggests that a once-in-threeyear daily high temperature cannot happen there during the peak forcing years. The average daily hot extreme PRs for tropical land areas are 0.20 and 0.44 in the Had-CM3 and CESM1 simulations, respectively (Table 1). In contrast, the PR for midlatitude land regions ranges from～0.3 to ～0.5, which makes it a little larger than that in the tropics and much closer to 1 in high-latitude land regions. This means that in the peak forcing years, the decrease in the probability of a daily hot extreme is the greatest in the tropics and smaller at high latitudes. In Antarctica, the probability of a daily hot extreme only decreases in some small areas.

In terms of the intensity of daily hot extremes, a global decrease in TXx occurs in peak forcing years (Figs. 4b,d). The spatial patterns of the TXx anomalies are consistent over North and South America in these two models.The maxima of the negative TXx anomalies are both located in midlatitude North America and central South America. In the Eastern Hemisphere, the maxima of the negative TXx anomalies are located in midlatitude regions of Eurasia and North Africa in the HadCM3 simulations. However, in the CESM1 simulations, the negative TXx anomalies are relatively uniform and slightly stronger in southern China, southern Australia, and the area north of the Mediterranean. For Antarctica and Greenland, the negative TXx anomalies are stronger in the CESM1 simulations than in the HadCM3 simulations. However, the stronger negative TXx anomalies in the HadCM3 simulations occur in high-latitude regions of Eurasia. On average, TXx decreases by 0.85/0.69°C in the tropics in the HadCM3/CESM1 simulations (Table 2). In contrast, it decreases by 1.24/0.72°C in midlatitude regions of the Northern Hemisphere, which is greater than the decrease in the tropics.

3.2 Changes in daily heavy precipitation extremes

Both modeling and observational studies have indicated that SVEs can decrease global precipitation, especially in monsoon areas (Peng et al., 2010; Joseph and Zeng, 2011; Liu et al., 2016). An SVE can weaken the monsoonal circulation (Cui et al., 2014). At the same time, the water-holding capacity of air decreases in response to volcanic cooling. These two factors lead to reduced precipitation after an SVE. The same results are found in the HadCM3 and CESM1 simulations. As shown in Fig. 6, the annual total precipitation decreases significantly in East and Northeast Asia, tropical Africa,Australia, and some parts of North and South America during the peak forcing years. Some of these reductions in precipitation can last 2 or 3 yr after the SVEs in the HadCM3 (Iles et al., 2013) and CESM1 simulations (Liu et al., 2016).

With regard to daily heavy precipitation extremes, the PR distribution is similar to the spatial pattern of the anomalous annual total precipitation (Fig. 7). The probability of heavy precipitation extremes decreases in most regions of the world. Globally, the average PR is approximately 0.90 in both model simulations. Across different latitudinal bands, the PR ranges from approximately 0.7 to 1, which suggests that the response of the heavy precipitation extremes to the volcanic forcing is generally consistent for land regions at different latitudes.

Regionally, the decrease in the probability of daily heavy precipitation is the strongest over the South American monsoon area in the HadCM3 simulations. The average PR is 0.76 (Table 3), which suggests that the probability is 25% lower during peak forcing years. Nevertheless, the strongest CESM1-simulated decreases in the probability occur in the African monsoon regions (Fig.7). The average PRs are 0.72 and 0.75 for the North and South African monsoon areas, respectively. In other monsoon areas, the two models simulate PRs ranging from 0.8 to 0.9. However, the HadCM3-simulated PRs are significantly larger than 1 over some southern Indian subcontinental regions (Fig. 7a). This means that the probability increases in these regions. In terms of the intensity of daily heavy precipitation, significant negative Rx5day anomalies are evident in most parts of the global monsoon areas in both model simulations (Fig. 8). These anomaly patterns resemble the simulated distributions of the PR. In the HadCM3 simulations, positive Rx5day anomalies also occur in southern India, which is consistent with the frequency of daily heavy precipitation. On average, Rx5day decreases by 6.44/4.90 mm and 2.42/5.36 mm over the East and South Asian monsoon areas in the peak forcing years, respectively (Table 3). These reductions are stronger than those of other monsoon areas. In the CESM1 simulations, the average decrease in Rx5day for the African monsoon region is also strong, at up to 3 mm.

Table 3. Averaged PR values and averaged differences in Rx5day and annual total precipitation (mm) in the HadCM3/CESM1 simulations over land monsoon regions (Christensen et al., 2013)

Fig. 7. Simulated spatial distributions of PR in the volcanic peak forcing years for daily heavy precipitation in (a) HadCM3 and (b) CESM1.Areas that exceed the 95% (90%) confidence level are denoted by white and black (small) dots.

Fig. 8. Simulated differences in Rx5day (mm) between the volcanic peak forcing years and no volcanic forcing years in (a) HadCM3 and (b)CESM1. Areas that exceed the 95% confidence level are denoted by dots.

Atmospheric stability is an important driver of heavy precipitation extremes. In this paper, to depict the stability of the atmosphere, we use theK-index, which was developed by George (1960). TheK-index is based on temperature and humidity. It is widely used to measure the instability of the troposphere and to predict thunderstorms and heavy precipitation (e.g., Marinaki et al.,2006; Chen et al., 2012; Barfus and Bernhofer, 2015;also see the National Weather Service website at https://www.weather.gov/ffc/gloss2). TheK-index is defined as follows:

whereTiis the air temperature at pressure leveli(hPa),andTdiis the dew-point temperature at pressure leveli(hPa), which is calculated based on the relative humidity and the actual air temperature using the Magnus–Tetens approximation (Lawrence, 2005). On the right-hand side of Eq. (1), the first term represents the temperature lapse rate from 850 to 500 hPa (geopotential heights), the second term represents the moisture conditions in the lower troposphere, and the third term reflects the saturation of the atmosphere and the depth of the moisture layer in the middle level (700-hPa geopotential height). Generally, theK-index is an integrative indicator of atmospheric stability and moisture conditions. A larger value ofKsuggests a less stable atmosphere, which is conducive to more frequent daily heavy precipitation extreme events(Fig. S2, available online).

As shown in Figs. 9a, e, theK-index is lower over most land regions in the peak forcing years, which suggests that the troposphere is more stable after an SVE and explains the lower frequency and intensity of daily heavy precipitation extremes in almost all the monsoon areas. The main reason for the large-scale decrease in theK-index is the SVE-induced decrease in the water-holding capacity of lower-tropospheric air on a global scale(Figs. 9c, g). The weaker temperature lapse rate from the lower troposphere to the middle troposphere over the Indian and American monsoon areas also contributes to the decrease in theK-index (Figs. 9b, f). In contrast, a decrease in the tropospheric temperature can make it easier for air to become saturated, which increases the instability of the troposphere. As a result of this factor, theK-index increases in North Africa and over the Iranian Plateau in both model simulations. Correspondingly, the frequency and intensity of daily heavy precipitation increase over these regions.

In the HadCM3 simulations, the daily heavy precipitation and the mean precipitation both increase over southern India (Fig. 6a). Therefore, we examine the changes in the circulation during boreal summer in peak forcing years. As shown in Fig. 10, easterly wind anomalies can be seen over the Indian subcontinent and Indo-China Peninsula, which suggests slight weakening of the summer monsoon circulation. As a result, the precipitation is weakened over central–northern India, where the boreal summer climatological precipitation has a widespread maximum (Bollasina et al., 2011). Due to changes of monsoon circulation, the rainfall belt shifts southward and causes an increase in the mean precipitation in southern India, where it can also contribute to the increase in the daily heavy precipitation there.

Fig. 9. HadCM3-simulated differences in the (a) annual K-index, (b) temperature lapse rate, (c) moisture conditions, and (d) saturation of the atmosphere between the volcanic peak forcing years and no volcanic forcing years. (e–h) As in (a–d), but for the CESM1 simulations. Areas that exceed the 95% confidence level are denoted by dots. The regions with elevations higher than 1500 m are blank.

Fig. 10. HadCM3-simulated differences in the boreal summer wind field (m s?1) at 850 hPa between the volcanic peak forcing years and no volcanic forcing years. Areas that exceed the 95% confidence level are shaded gray.

4. Conclusions and discussion

In this study, the influences of SVEs on daily cold extremes, hot extremes, and heavy precipitation extremes were investigated in parallel using HadCM3 and CESM1 long-term simulations. One metric, namely, PR, was used to assess the probabilities of daily extremes under SVEinfluenced conditions. Additionally, the TNn, TXx, and Rx5day indices were examined to explore the responses of the intensities of daily extremes to volcanic forcing.

The results show that the probability of daily cold extremes increases significantly. Globally, PR increases, on average, by a factor of more than 2. This suggests that an extreme cold event expected once every 3 years without volcanic forcing can be expected every 1.5 years when there is volcanic forcing. In most parts of the tropical regions, the probability of a daily cold extreme increases.Over tropical Africa and South America in particular, PR increases by a factor of 5–20 in the two model simulations, which suggests a much higher frequency of daily cold extremes. In terms of intensity, TNn decreases significantly in tropical and midlatitude land regions. In contrast, the probability and intensity of daily cold extremes decrease over Siberia in the HadCM3 simulations and large areas of high-latitude Eurasia in the CESM1 simulations. These are mainly caused by the SVEstrengthened polar vortex and the associated temperature advection anomalies.

Due to volcanic cooling, the probability of daily hot extremes decreases around the world, but especially in tropical regions. Daily hot extremes do not occur over most parts of the tropics in peak forcing years. On average, the frequency of daily hot extremes decreases by more than 60% and 40% in the peak forcing years in the HadCM3 and CESM1 simulations, respectively. In addition, the intensity of daily hot extremes decreases around the world. For midlatitude land regions in the Northern Hemisphere and tropical South America especially, TXx decreases by more than 1°C after an SVE. Overall, the SVE-induced cooling effect plays a dominant role in regulating daily cold and hot extremes.

SVEs can decrease the global monsoon precipitation and daily heavy precipitation extremes. In almost all monsoon areas, the probability of a daily heavy precipitation extreme decreases. In particular, due to the influences of SVEs, the probability of daily heavy precipitation extremes decreases by more than 20% in the East Asian, South African, and South American monsoon areas, which is slightly higher than that in other monsoon areas. Correspondingly, the intensity of daily heavy precipitation also decreases. The strongest decrease can be found in the East Asian monsoon area. Based on an analysis of changes in theK-index, this decrease is mainly caused by an SVE-induced global decrease in the water-holding capacity of lower tropospheric air.However, the frequency and intensity of daily heavy precipitation extremes increase over the southern Indian subcontinent, Iranian Plateau, and North Africa in one or both model simulations. These enhancements to the daily heavy precipitation extremes are mainly caused by SVEdecreased dew points in the low-level troposphere. In the HadCM3 simulations, the SVE-weakened Indian monsoon circulation can increase the mean precipitation over southern India, which can also contribute to the increase in daily heavy precipitation. Therefore, SVEs have important influences on the daily heavy precipitation. SVEinduced changes in the water-holding capacity and the dew-point temperature in the low-level troposphere are the main factors affecting daily heavy precipitation extremes. On the global scale, changes in the water-holding capacity play a dominant role in decreasing the frequency and intensity of daily heavy precipitation extremes.

In terms of temperature, volcanic forcing has similar influences on mean climate and extreme climate. The direct cooling effect of volcanic forcing is the main reason for the increase in the daily cold extremes and decrease in the daily hot extremes around the world. However,during boreal winter, the high-latitude regions of the Northern Hemisphere receive less solar radiation. Therefore, the volcanic cooling effect is limited. SVE-induced changes in high-latitude circulation can overcome the cooling effect and result in warmer winters in Siberia and even larger regions of high-latitude Eurasia. Correspondingly, the frequency and intensity of daily cold extremes are lower. This means that SVE-induced changes in circulation can also play an important role in regulating temperature extremes over high-latitude regions.

Additionally, changes in the frequency and intensity of extreme temperatures are qualitatively consistent, but inconsistent in range for some tropical and midlatitude regions. Taking the change in the daily cold extremes as an example, the probability of occurrence increases for the tropical regions, where it is higher than for midlatitude regions (Fig. 3b). However, the biggest change in TNn is evident in the midlatitude regions of Eurasia, rather than the tropics (Fig. 4b). In fact, stronger cooling happens over the midlatitude regions following the SVEs (Fig. 2).This is why we find the biggest change of TNn in the midlatitude regions of the Northern Hemisphere.However, the distribution of the daily SAT over the midlatitude regions is much wider than that over the tropics(Fig. S3, available online). That means that for the midlatitude regions, the same SAT change from the tail can explain a smaller percentage of the SAT distribution than that for the tropics. Therefore, with regard to frequency of daily cold extremes, bigger changes can be seen over the tropics, rather than the midlatitude regions. The daily hot extremes are the same.

ENSO is one of the most important coupled modes affecting global and regional climate and climate extremes.In HadCM3 and CESM1, an El Ni?o-like response can be found following SVEs (Stevenson et al., 2016; Liu et al., 2018; Miao et al., 2018). Thus, the influences from SVE-induced ENSO evolution can also regulate the global climate extremes during the peak forcing period, particularly over the tropics. As shown in Fig. 6, the annual total precipitation is increased over the central–western tropical Pacific, showing an El Ni?o-like precipitation response.This is consistent with our previous Bergen climate model’s output (Wang et al., 2018), as well as CESM1 and CMIP5 results (Stevenson et al., 2016; Paik et al.,2020b). It also suggests that this ENSO response can influence the extreme climate over tropical continental areas in the peak forcing years. In fact, during this period,the responses of climate and extreme climate to ENSO’s variation are also caused by the SVEs. It is difficult to distinguish them from the direct impacts of SVEs.However, based on the present study’s findings, this kind of indirect influence from anomalous ENSO on tropical climate extremes is likely smaller than that directly from volcanic forcing. For high-latitude climate extremes, influences from SVE-induced anomalous high-latitude circulation or modes are more important.

There are some different responses of extreme temperature and precipitation to SVEs between the HadCM3 and CESM1 simulations. As described in Section 2, there are many differences in the forcing added into the two models. In HadCM3, the added volcanic forcing is reconstructed AOD, whereas the forcing for CESM1 is reconstructed stratospheric sulfate aerosol. In addition, the intensity and meridional difference of the forcing also differ. Therefore, these distinctions in volcanic forcing may partly contribute to the different responses of extreme climate to SVEs between the two models. Of course,there may be some other reasons, such as dynamic processes and related feedbacks of the climate system in the models. However, it is difficult to draw conclusions in this respect based on the present data. To further explore the possible causes of the different climate responses between the two models, more sensitivity simulations(e.g., identical volcanic forcing in HadCM3 and CESM1)are needed. Additionally, influences from the season in which the eruption took place may also constitute an important factor affecting global climate (Stevenson et al.,2017). As shown in Figs. 2c–f, there are large differences in the seasonal responses of SAT to the SVEs between the HadCM3 and CESM1 simulations. In the CESM1 simulations, the start time for the majority of eruptions is April, whereas in the HadCM3 simulations,it varies amongst the different SVEs. Thus, boreal summer period analyzed in the CESM1 simulations was June–July–August, which closely followed the start times. The intensities of volcanic forcing and responses of the climate system usually do not reach the peak (Figs.1, 3 in Otter?, 2008). Thus, the boreal summer responses of SAT in the CESM1 simulations are smaller than those in the HadCM3 simulations (Figs. 2e, f), whereas the boreal winter responses in CESM1 are greater than those in the HadCM3 simulations (Figs. 2c, d).

Focusing on the five SVEs that occurred in the past 100 years, Paik and Min (2018) investigated the impact of strong eruptions on the intensity of climate extremes using CMIP5 multimodel results. In this study, the largescale responses of climate extremes in HadCM3 and CESM1 are qualitatively consistent with Paik and Min(2018). However, some differences still exist in some regions. For example, compared with their results (Paik and Min, 2018), the high-latitude winter warming and the associated changes in the daily cold extremes are more pronounced in this study. In fact, similar winter warming can be found in observations (e.g., Robock, 2000) and a previous modeling study (e.g., Zambri and Robock,2016). Some responses of extreme climate are possibly hidden within the process of the multimodel ensemble mean. Therefore, analysis of individual model results is also very necessary. Additionally, the present study further examined the responses of the frequency of climate extremes to SVEs. Based on the HadCM3 and CESM1 results, they are different from the responses of the intensity of climate extremes in some continental regions.

As mentioned above, volcanic eruptions constitute an important external forcing of the climate. However, there are some different ways to added volcanic forcing into a model. Additionally, there are large uncertainties in the start time and intensity of radiative forcing in reconstructed volcanic records. These factors can all markedly influence the responses of the model’s climate system to the forcing, thereby causing considerable uncertainty in the research findings. Therefore, how to accurately express volcanic activity and associated forcing in the model is a key and urgent issue in investigating volcanic impacts on climate. Within the Model Intercomparison Project on the climatic response to Volcanic forcing(VolMIP), a coordinated set of idealized volcanic perturbation experiments have been carried out (Zanchettin et al., 2016). In future, more studies that employ Vol-MIP multimodel results are needed.

Acknowledgments.We thank three anonymous reviewers and the editor for their valuable comments and suggestions, which have helped improve the quality of this paper significantly.