The likelihood of a large volcanic eruption in the future provides
the largest uncertainty concerning the evolution of the climate
system on the timescale of a few years, but also an excellent
opportunity to learn about the behavior of the climate system, and
our models thereof. So the following question emerges: how predictable is the
response of the climate system to future eruptions? By this we mean
to what extent will the volcanic perturbation affect decadal climate
predictions and how does the pre-eruption climate state influence
the impact of the volcanic signal on the predictions? To address
these questions, we performed decadal forecasts with the MiKlip
prediction system, which is based on the MPI-ESM, in the
low-resolution configuration for the initialization years 2012 and
2014, which differ in the Pacific Decadal Oscillation (PDO) and
North Atlantic Oscillation (NAO) phase. Each forecast contains an
artificial Pinatubo-like eruption starting in June of the first
prediction year and consists of 10 ensemble members. For the
construction of the aerosol radiative forcing, we used the global
aerosol model ECHAM5-HAM in a version adapted for volcanic
eruptions. We investigate the response of different climate
variables, including near-surface air temperature, precipitation,
frost days, and sea ice area fraction. Our results show that the
average global cooling response over 4 years of about
0.2
More and more attention has been paid to decadal climate prediction in the last decade (Meehl et al., 2009; Smith et al., 2007). This research field tries to fill the gap between short-term (weather to seasonal) predictions on the one hand and long-term climate projections on the other hand. Decadal predictability comes mainly from the multi-year memory of the ocean. The memory in the ocean arises, for instance, from the persistence of ocean heat content anomalies and from properly initialized ocean dynamics and circulation (e.g., Guemas et al., 2012; Matei et al., 2012). A detailed explanation of the principles behind decadal prediction can be found in Kirtman et al. (2013).
A number of studies revealed that there is at least potential prediction skill in near-surface air temperature, precipitation, and three-dimensional variables like air temperature or geopotential height (Goddard et al., 2013; Kadow et al., 2016; Stolzenberger et al., 2016). Recently, some institutions like the UK Met Office and the German project for decadal climate prediction, MiKlip (Marotzke et al., 2016), have started issuing decadal climate forecasts for near-surface air temperature on a regular – but still experimental – basis (metoffice.gov, 2017; Smith et al., 2013; Vamborg et al., 2017).
The skill of decadal predictions is usually evaluated using hindcast
simulations (e.g., Doblas-Reyes et al., 2013; Kim et al., 2012), and it
is assumed that external forcing is known over the whole simulation
period. In a real decadal forecast, this assumption is invalid because
rapid forcing changes like volcanic eruptions cannot be
predicted in advance. Hence, strong tropical volcanic eruptions (SVEs)
are arguably the largest source of uncertainty for this type of
prediction. They increase the stratospheric aerosol load, which leads
to a reduction of global mean surface temperature due to the reduced
incoming solar radiation. For instance, after the tropical Pinatubo
eruption in 1991, a global peak cooling of about 0.4
There are only a few studies that focus on how volcanic forcing impacts decadal climate predictions. Meehl et al. (2015) showed in a multi-model study that the Pinatubo eruption led to a reduction of decadal hindcast skill in Pacific sea surface temperatures. Timmreck et al. (2016) demonstrated that neglecting volcanic aerosol in decadal predictions significantly affects hindcast skill for near-surface air temperature and leads to a skill reduction in most regions up to prediction year 5. Bethke et al. (2017) explored how possible future volcanic eruptions impact climate variability under RCP4.5 and found that the consideration of volcanic forcing enhances climate variability on annual to decadal timescales. However, not every volcanic eruption influences the climate in the same way. Zanchettin et al. (2013) showed in a case study of the Tambora eruption in 1815 that near-surface atmospheric and oceanic dynamics are significantly influenced by climate background conditions. Furthermore, hindcasts with the atmosphere-only HADGEM1 model (Marshall et al., 2009) showed that the climate anomalies in the first post-volcanic winter over Europe are strongly dependent on the stratospheric conditions in early winter.
Historically explosive tropical volcanic eruptions have a statistical recurrence frequency of about 50 to 100 years (Ammann and Naveau, 2003; Self et al., 2006). With the Pinatubo eruption almost 27 years ago and the recent ongoing unrest of Mount Agung in Indonesia, the following question arises: what would happen if a large volcanic eruption occurred in the present and how dependent would the results be on the start year and the associated initial climate state? In this paper, we investigate the response of different climate variables, including near-surface air temperature (TAS), precipitation (PR), number of frost days (FD), and sea ice area fraction (SIC), to an artificial Pinatubo-like eruption happening in June of the first simulation year for two initializations that differ in their initial state. To quantify the volcanic effect we compare our multi-year forecasts with the simulations without a volcanic eruption performed with the MiKlip prediction system (Pohlmann et al., 2013).
In Sect. 2 we describe the models used and our experimental setup, while the results of our analysis for different variables are presented in Sect. 3. In Sect. 4 we draw conclusions and discuss our results.
Time series of 4-year running mean ensemble forecast
of near-surface air temperature (TAS) anomalies. The blue shows
values without volcanic eruption, and red with a Pinatubo-like
eruption.
We perform our forecasts containing a Pinatubo-like eruption with the
baseline1 version of the MiKlip prediction system (Marotzke et al.,
2016; Pohlmann et al., 2013), which is based on the coupled
Max Planck Institute–Earth System Model (MPI-ESM; Giorgetta et al.,
2013; Jungclaus et al., 2013). The MPI-ESM is an Earth system model
with atmosphere, ocean, and dynamic vegetation components. We use the
“low-resolution” (LR) configuration of the baseline1 system, which
has a resolution of T63 with 47 vertical levels in the atmosphere and
an oceanic resolution of 1.5
Here, a two-step modeling approach is applied to consider the effect
of large volcanic eruptions in the MiKlip decadal prediction
system. In a first step, the formation of volcanic sulfate aerosol and
its corresponding optical parameters (aerosol optical depth: AOD;
effective radius: R
For the construction of the aerosol radiative forcing, we use the
global aerosol model ECHAM5–HAM (Stier et al., 2005) in
a version adapted for volcanic studies (Niemeier et al., 2009), which
agrees very well with measurements of AOD in the visible range and the
effective particle radius after the Pinatubo eruption. We decided to
simulate a future Pinatubo-like eruption because the Pinatubo
eruption is the best-observed eruption in recorded history and the second strongest
since 1850. In addition, the likelihood of such an eruption is of the
order of once every 50 to 100 years (e.g., Self, 2006). To compile
volcanic forcing fields, we inject 17 MT
Figure 1a shows the aerosol optical depth (AOD) simulated with the global aerosol model for a Pinatubo-like volcanic eruption. We use the simulated AOD as the forcing component for the decadal prediction system. For our experiment, we perform two decadal forecasts for 10 years with 10 ensemble members each. For ensemble generation we use the lagged-day initialization method, which means that the individual ensemble member is started on different days around 31 December to spread the ensemble. One forecast was initialized around 31 December 2012 (Pinatubo-2012) and the other one around 31 December 2014 (Pinatubo-2014). The Pinatubo-like eruption happens in June of the first prediction year, which is June 2013 in the case of the Pinatubo-2012 experiment and June 2015 for the Pinatubo-2014 experiment. We chose these initialization years because they are relatively close together and therefore have similar greenhouse gas forcing that can be considered close to present day conditions, but they also differ in important climatic conditions. In December 2012 the Pacific Decadal Oscillation (PDO) and the North Atlantic Oscillation (NAO) were in a negative phase, whereas the Pinatubo-2014 experiment is initialized with a positive PDO and a positive NAO (Fig. 1b and c). Other important climate modes like the El Niño–Southern Oscillation (ENSO) or the Atlantic Multidecadal Oscillation (AMO) are in a similar state in both experiments (not shown). PDO and NAO are both important drivers of internal climate variability. A negative PDO phase is associated with below average temperatures in the Pacific Northwest, British Columbia, and Alaska and an above average Indian summer monsoon (e.g., Mantua and Hare, 2002). A positive NAO indicates colder and drier Mediterranean regions and warmer and wetter than average conditions in northern Europe and the eastern United States (e.g., Visbeck et al., 2001). The different phases of the NAO and PDO at initialization time enables us to investigate the influence of initial climate conditions on the volcanic response of the model in a present day setup. As reference datasets, we use the MiKlip baseline1 experiments initialized on the same start dates (b1-2012 and b1-2014) but without volcanic aerosol.
In order to quantify the effect of volcanic aerosols, we calculate the
differences in the ensemble mean between the simulations containing
a Pinatubo-like eruption and the baseline1 simulations
(exp-2012 is Pinatubo-2012 minus b1-2012, exp-2014 is Pinatubo-2014
minus b1-2014). We also calculate the difference between exp-2012 and
exp-2014 to quantify the impact of the different initial
conditions. Statistical significance is determined by using
a two-sided
Differences in ensemble mean forecasts of TAS for prediction
years 1–4
Differences in ensemble mean forecasts of zonal mean air
temperature (TA) for prediction years 1–4.
Figure 2 shows a forecast for the 4-year running mean near-surface
air temperature (TAS) for different regions. The forecast is shown
like it would be issued by the MiKlip project (Vamborg et al.,
2017). In addition, we present it together with our Pinatubo
experiments. The Pinatubo-like eruption leads to a statistically
significant decrease at the 95 % level in global mean temperature
of about 0.2
Differences in ensemble mean forecasts of SIC for prediction
years 1–4,
Same as Fig. 2, but for sea ice area fraction (SIC) maximum
and minimum and different regions.
This disparity is also visible in the global maps in Fig. 3, which
shows TAS for prediction years 1–4 and 7–10. For both initialization
dates, the Pinatubo-like eruption leads to significant cooling over
most parts of the tropics, North America, and the North Atlantic
(Fig. 3a and c) for prediction years 1–4. Generally, the cooling
effect is strongest over the continents and reaches up to 1
Figure 4 shows a cross section of the zonal mean air temperature (TA)
averaged over prediction years 1–4. In both experiments, the cooling
we found at the surface continues in the troposphere and is strongest
in the tropical troposphere between 100 and
400
Differences in ensemble mean forecasts of frost days (FDs) for
prediction years 1–4.
Gagne et al. (2017) recently showed that a decade of increased Arctic
sea ice followed the last three large volcanic eruptions in the 20th
century. Figure 5 shows the differences in the ensemble mean forecasts
of sea ice area fraction (SIC) for prediction years 1–4 for the
sea ice maximum in March (top row) and the sea ice minimum in
September (bottom row). Overall we see increased maximum values of SIC
due to the volcanic eruption in both experiments, but the two
initialization times differ in the affected local areas. On the one
hand, exp-2012 shows increased values of SIC in the Bering Sea of up
to 10 % where there is no evident signal in exp-2014. On the other
hand, the 2014 initialized experiment shows significantly increased
SIC values in the Nordic Sea where the sea ice area fraction of
exp-2012 is only slightly higher. This different behavior is not only
evident in the first four prediction years. Figure 6a–d show the
4-year running mean forecast for maximum SIC in the Nordic Sea area
(30–90
Gagne et al. (2017) stated in their recent study that the sea ice response is dependent on pre-eruption temperature conditions and that a warmer pre-eruption climate leads to a stronger sea ice increase. The results shown in Fig. 6 do not corroborate their findings. In fact, they show a slightly contrary tendency and regions with higher initial sea ice and lower temperature conditions (not shown) react more strongly to the Pinatubo-like eruption. This could be a model-dependent effect or a sampling effect due to the focus on only two initialization times in our study.
Same as Fig. 2, but for precipitation (PR) and different
regions.
Same as Fig. 7, but for precipitation (PR) anomalies.
Top row shows the Niño 4 index and bottom row shows the ENSO precipitation index (ESPI) for the first four prediction years calculated as a 12-month running mean to reduce variance. Left (right) column shows the 2012 (2014) initialized experiments. Error bars show the SD of the ensemble and vertical black lines indicate a significant difference.
It is notable that there is a decreasing trend in all our simulations in the three regions and that the trend is not affected by the Pinatubo-like eruption. If there are increased values of SIC in one experiment (Fig. 6b, c, and f), the difference in SIC values between the Pinatubo and the baseline1 simulations stays nearly constant for all prediction years.
Not only mean temperature values are influenced by volcanic aerosol,
but also the daily temperature minimum. The Expert Team of Climate
Change Indices (ETCCDI; Karl et al., 1999) defines a day as a frost day
if the daily minimum temperature is below 0
A critical aspect is the understanding of the volcanic impact on the
hydrological cycle. It has been demonstrated that volcanoes modulate
the African, Asian, and South American monsoon systems (Liu et al.,
2016; Oman et al., 2006), impacting areas that are now home to
Hence, while there could be some confidence in the general behavior of the post-volcanic changes in the hydrological cycle, the quantitative values of our forecast simulation should be taken with caution. Although the longer-persisting reduction over the ocean is seen in CMIP5 models, it cannot be detected in observations due to the short satellite time period, which covers only two major eruptions (Iles and Hegerl, 2014). The timescale of the precipitation reduction over the ocean is consistent with the response of TAS (Fig. 2). This is in agreement with previous studies (Iles et al., 2013; Joseph and Zeng, 2011).
In the global precipitation maps, we see a reduction of precipitation for both experiments through the volcanic aerosol in large parts, especially over land, in the first four prediction years (Fig. 9). The drying effect is strongest over the tropics, particularly in Southeast Asia, and is even more pronounced in exp-2014. In fact, the tropical precipitation pattern in Southeast Asia and the East Pacific in exp-2014 is very similar to an El Niño response. Recent model studies (Maher et al., 2015; Pausata et al., 2015; Khodri et al., 2017) revealed that volcanic eruptions have a significant impact on ENSO, and there is some ongoing debate about whether a tropical volcanic eruption can trigger an El Niño event (Meehl et al., 2015; Predybaylo et al., 2017; Swingedouw et al., 2017). To further investigate this, we calculated the temperature-based Niño 4 index (Trenberth and Stepaniak, 2001) and the ENSO precipitation index (ESPI; Curtis and Adler, 2000) for both experiments for the first four prediction years (Fig. 10) as 12-month running means to reduce variance. The ensemble initialized in 2014 with a Pinatubo-like eruption shows a tendency towards El Niño conditions, whereas the baseline1 ensemble favors a weak La Niña condition (Fig. 10b and d). The difference between the two experiments in the ESPI is significant until simulation months 18–30 when both indices come back to neutral conditions. In exp-2012 there is no difference evident in the first three prediction years, but in year 4 the baseline1 ensemble starts simulating a La Niña phase (Fig. 10a and c) with a significant difference to the Pinatubo-like experiment. In general, exp-2014 shows a stronger drying response in the tropical region. In contrast, in this experiment, wetter conditions over Western Europe can be found that do not occur in exp-2012.
In this study, we examined the sensitivity of decadal climate
predictions to a tropical volcanic eruption using an artificial
Pinatubo-like eruption as stratospheric forcing. We performed two
decadal forecasts with different initial conditions, each forecast
containing a Pinatubo-like eruption starting in June of the first
prediction year, and compared them to the corresponding simulations
without a volcanic eruption. We chose the initialization years 2012
and 2014 because they differ in important climate indices like the
NAO and the PDO. Other important climate modes like the El Niño–Southern Oscillation (ENSO) or the Atlantic Multidecadal Oscillation
(AMO), which have the potential to influence the volcanic response as
well (e.g., Swingedouw et al., 2017, and references therein), are in
a similar state in both experiments at the time of initialization (not
shown). We have shown that the global near-surface air temperature and
precipitation decrease as a response to the volcanic eruption is
independent of the initial state of the PDO and the NAO and that the
reduction is significant for the whole prediction period in both
forecasts. In our experiments, the global mean temperature reduction
in the first 4 years following a Pinatubo-like eruption is about
0.2
Pre-eruption climate conditions play an important role for decadal predictions on a regional scale. We found significant regional differences between the two initialization experiments in the variables near-surface air temperature, sea ice area fraction, frost days, and precipitation for the whole forecast period. One of the most substantial differences between the experiments can be found in the predictions of minimum and maximum sea ice area fraction. The volcanic eruption in the 2012 initialized simulation has nearly no effect on the 4-yearly minimum SIC, whereas in exp-2014 we see a significant increase of up to 4 %. For maximum SIC, both simulations show increased values, but the increase is concentrated in different regions (2012: the Bering Sea, 2014: the Nordic Sea). This can be explained partly by the different phase of the PDO; a negative PDO, as in the 2012 initialized experiments, brings colder temperatures to Alaska (Wendler et al., 2013) and strengthens the Arctic wintertime warming (Screen and Francis, 2016). In the 2012 experiment the temperature decrease in the North Pacific basin is nearly constant over the whole prediction period, whereas in 2014 the temperature starts recovering after a few years. Additionally, we see a stronger cooling over Europe and a more pronounced drying in the monsoon region in the first four prediction years and a longer-lasting cooling effect in the North Atlantic in the 2014 initialized simulations. We also see a stronger increase in the number of frost days in most regions – except for the Bering Sea – in this experiment. We could not find a clear link between the different initial states of the NAO and any of these changes.
We note a few caveats and possibilities for improvements to this study. We only investigated the volcanic response to different initial conditions of the NAO and PDO. Therefore, our simulations in this study should be extended with experiments starting with other initial conditions like the recent El Niño year 2015–2016. Another factor currently neglected is the phase of the QBO as it changes due to the post-volcanic atmospheric response (e.g., Thomas et al., 2009) and its self-modulation by strong volcanic eruptions (Aquila et al., 2014). The model (MPI-ESM) in the low-resolution version used in this study is not able to develop its own quasi-biennial oscillation (QBO), but the same model with higher vertical resolution shows a predictive skill of the QBO of up to 4 years (Pohlmann et al., 2013). Another aspect is that our results could be model dependent and the analysis should be expanded to a multi-model study. In order to gain a better understanding of the impact of volcanic eruptions on decadal predictions and predictability, a collaboration is planned between the model intercomparison project on the climatic response to volcanic forcing VolMIP (Zanchettin et al., 2016) and the decadal climate prediction project DCPP (Boer et al., 2016). In line with the protocol of the upcoming CMIP6 (Eyring et al., 2016), a set of decadal prediction experiments will be conducted in which, similar to our experiment, the impact of a Pinatubo-like eruption occurring in 2015 will be examined, which provides the unique opportunity to discuss our results in a multi-model framework.
The model output from all simulations described
in this paper will be distributed through the World Data Climate
Center at
The authors declare that they have no conflict of interest.
This article is part of the special issue “The Model Intercomparison Project on the climatic response to Volcanic forcing (VolMIP) (ESD/GMD/ACP/CP inter-journal SI)”. It is not associated with a conference.
We thank Wolfgang Müller and Bereket Berhane whose comments helped to improve this paper. We also thank the three anonymous reviewers for their helpful comments. The research was supported by the German Federal Ministry for Education and Research through the “MiKlip” program (FKZ: 01LP1519B to Sebastian Illing and Christopher Kadow; 01LP1517B to Claudia Timmreck; 01LP1519A to Holger Pohlmann). We acknowledge the use of the European Centre for Medium-Range Weather Forecasts reanalysis data for the initialization (ORAS4, ERA-40, and ERA-Interim). Computations were carried out at the German Climate Computing Centre (DKRZ). Supporting information that may be useful in reproducing the authors' work is available from the authors upon request (sebastian.illing@met.fu-berlin.de). Edited by: Govindasamy Bala Reviewed by: three anonymous referees