The effect of aerosols is one of many uncertain factors in projections of future climate. However, the behaviour of mineral dust aerosols (dust) can be investigated within the context of past climate change. The Last Glacial Maximum (LGM) is known to have had enhanced dust deposition in comparison with the present, especially over polar regions. Using the Model for Interdisciplinary Research on Climate Earth System Model (MIROC-ESM), we conducted a standard LGM experiment following the protocol of the Paleoclimate Modelling Intercomparison Project phase 3 and sensitivity experiments. We imposed glaciogenic dust on the standard LGM experiment and investigated the impacts of glaciogenic dust and non-glaciogenic dust on the LGM climate. Global mean radiative perturbations by glaciogenic and non-glaciogenic dust were both negative, consistent with previous studies. However, glaciogenic dust behaved differently in specific regions; e.g. it resulted in less cooling over the polar regions. One of the major reasons for reduced cooling is the ageing of snow or ice, which results in albedo reduction via high dust deposition, especially near sources of high glaciogenic dust emission. Although the net radiative perturbations in the lee of high glaciogenic dust provenances are negative, warming by the ageing of snow overcomes this radiative perturbation in the Northern Hemisphere. In contrast, the radiative perturbation due to high dust loading in the troposphere acts to warm the surface in areas surrounding Antarctica, primarily via the longwave aerosol–cloud interaction of dust, and it is likely the result of the greenhouse effect attributable to the enhanced cloud fraction in the upper troposphere. Although our analysis focused mainly on the results of experiments using the atmospheric part of the MIROC-ESM, we also conducted full MIROC-ESM experiments for an initial examination of the effect of glaciogenic dust on the oceanic general circulation module. A long-term trend of enhanced warming was observed in the Northern Hemisphere with increased glaciogenic dust; however, the level of warming around Antarctica remained almost unchanged, even after extended coupling with the ocean.
The Last Glacial Maximum (ca. 21 000 years before present; LGM), which is the most recent period featuring maximum expansion of the land ice sheets in the Northern Hemisphere, has been investigated thoroughly using various palaeo-proxy records and via modelling studies (Braconnot et al., 2007a, b; Kageyama et al., 2006, 2017). Climate modelling is an essential tool in investigations seeking to clarify the mechanisms of climate change, as stated in the Intergovernmental Panel on Climate Change (IPCC) assessment reports (IPCC, 2013). Therefore, it is especially important to evaluate the capability of numerical models to capture past climatic conditions.
Palaeo-proxy data and modelling studies are both required for a proper understanding of past climates; however, the focus of this study is on modelling. General circulation models (GCMs) are one of the tools used most widely for investigation of the mechanisms of both climate and climate change. The improvement of computational resources has allowed for the development of models with high complexity that permit interactive coupling of various climatic components. In comparison with proxy data, previous modelling experiments targeting the LGM have tended to underestimate the magnitude of cooling, especially over high latitudes (Masson-Delmotte et al., 2006, 2010). The importance of feedback related to dust and vegetation has been identified in Chapter 5 of the IPCC Fifth Assessment Report (IPCC, 2013).
It is recognised that uncertainty over the effect of aerosols is one of the most important factors regarding radiative perturbation in estimates of global warming. Mineral dust is the most abundant atmospheric aerosol, even in the present climate. For example, Mahowald et al. (2010) investigated the trend of the amount of atmospheric dust in the 20th century based on observations and modelling. They reported a correlation between an increase in desert dust and a net negative radiative perturbation. Examination of proxy data has suggested a clear enhancement of dust during the LGM, which was especially pronounced at high latitudes, i.e. reaching levels more than 20 times greater than the present day over Antarctica (Lambert et al., 2008; Lamy et al., 2014; Dome Fuji Ice Core Project members, 2017). Although the enhancement of dust deposition was found less over lower latitudes, it was still several times higher in comparison with the present day (Winckler et al., 2008).
Although earlier studies (Mahowald et al., 1999; Lunt and Valdes, 2002;
Claquin et al., 2003) have estimated higher dust amounts during the LGM in
comparison with the pre-industrial (PI) period, dust amounts over Antarctica
have tended to be underestimated. Claquin et al. (2003) estimated the
radiative perturbation at the top of the atmosphere (TOA). They reported a
cooling effect attributable to dust, but they also found a warming effect
due to dust deposition on snow. Later, Mahowald et al. (2006a, b) estimated
the glaciogenic dust flux and the aerosol–radiation interaction. Their
standard LGM experiment simulated an underestimation of dust deposition flux,
especially over high latitudes, in comparison with the DIRTMAP proxy data
archive (Kohfeld and Harrison, 2001). Then, they considered the effect of
sources of glaciogenic dust surrounding the ice sheets and glaciers. Such
areas are defined to generate substantial amounts of glacial flour during
glacial periods (Bullard et al., 2016). The study considered the emission of
various fluxes of dust from these glaciogenic source areas and a best fit to
the DIRTMAP deposition distribution was obtained. Although this estimate
could conceal other possible and non-introduced processes of dust sources,
it constitutes an important step forward in the determination of a
reasonable representation of both the atmospheric loading and the
depositional distribution of dust during the LGM. However, they did not
estimate the effects of aerosol–cloud interaction. Takemura et al. (2009)
used the Model for Interdisciplinary Research on Climate (MIROC) Atmospheric
GCM (AGCM) with an online aerosol module to determine both the
aerosol–radiation and the aerosol–cloud interactions for the LGM and PI
periods at both the surface and the tropopause. However, they underestimated
the amount of dust deposition over Antarctica, probably because they did not
consider glaciogenic dust. Lambert et al. (2013) used two general
circulation models coupled with online aerosol models and obtained
underestimated dust flux and radiative forcing. This underestimation was
global, but more pronounced over the polar regions, and they suggested the
possibility that it contributes to an underestimation of polar amplification
for LGM and future projections. Yue et al. (2011) used an AGCM to estimate
the aerosol–radiation interaction for dust and they reported an evident
cooling effect. Albani et al. (2014) supposed high erodibility areas to
obtain better representation of LGM dust. They also highlighted the
importance of the optical properties and size distribution of dust aerosols.
In comparison with the control setting, Sagoo and Strelvmo (2017) applied an
emission factor of 3.4 to the dust emissions in an LGM level
Another aspect of dust is related to the ageing of the snow surface, which possibly modulates the surface temperature via albedo reduction. Krinner et al. (2006) discussed the importance of the ageing effect of snow, particularly over eastern Siberia. Their ageing scheme was based on that of Warren and Wiscombe (1980) and Wiscombe and Warren (1980). Moreover, Ganopolski et al. (2010) simulated the glacial–interglacial cycle using an intermediate complexity model, in which the ageing effect was implemented via simple scaling.
Previous studies have not included a dynamic ocean in this context, so the impacts on global ocean circulation are unknown.
In summary, we claim that the evaluation of the total effect of dust on the LGM surface temperature is incomplete. Therefore, this study addresses the problem by incorporating the effects of aerosol–radiation interaction, aerosol–cloud interaction, snow ageing, and dust–ocean interaction. We undertook AGCM simulations and full ESM simulations of the LGM with sensitivity experiments targeting the effects of dust on climate.
The following section explains the modelling and experimental set-ups. The resulting estimations of dust amount and dust depositional distribution are presented in Sect. 3.1, and the influence of dust on surface temperature is described in Sect. 3.2. To investigate how dust might modulate the atmospheric state, the radiative perturbation attributable to dust is described in Sect. 3.3 and the effect of glaciogenic dust on the ocean is discussed in Sect. 3.4. The results of the simulations are summarised and discussed in Sect. 4.
The MIROC-ESM (Watanabe et al., 2011) used in this study was the version
submitted to both the Coupled Model Intercomparison Project phase 5 (CMIP5)
and the Paleoclimate Modelling Intercomparison Project phase 3 (PMIP3). The
resolution of the atmosphere in the model is T42 with 80 vertical levels,
while that of the ocean is about 1
In the MATSIRO module, the effect of dirt in snow (i.e. snow ageing) was considered based on the work of both Yang et al. (1997) and Warren and Wiscombe (1981). The magnitude of dirt concentration at the snow surface was varied to fit an observed relation between snow albedo and dirt concentration (Aoki et al., 2006). The dirt concentration in snow was calculated from the deposition fluxes of dust and soot calculated in the SPRINTARS module. The relative strength of the absorption coefficients for dust and soot were weighted as a function of the deposition fluxes to obtain radiatively effective amounts of dirt in the snow.
We performed eight experiments: five using the AGCM part of the MIROC-ESM and three using the full MIROC-ESM. The specific experiments labelled PI.a and PI.e represent the AD 1850 control climate of the PI era, with PI.e having been submitted to CMIP5. The previous 100-year climatology of sea surface temperature (SST) and sea ice of the period submitted to CMIP5 was used as a boundary condition for PI.a. The experiments labelled LGM.e and LGM.a represent the LGM climate following the PMIP3 protocol (Abe-Ouchi et al., 2015). The LGM.e experiment was submitted to CMIP5–PMIP3 (Sueyoshi et al., 2013). The LGM.a experiment was the AGCM experiment using the SST and sea ice taken from the PMIP3 LGM experiment (LGM.e). The LGM.e experiment was extended for a further 800 years beyond the PMIP3 period (Fig. 1). The LGMglac.a experiment was a new experiment based on the same conditions as LGM.a, but with an additional glaciogenic dust flux following Mahowald et al. (2006a). The LGMglac.naging.a and LGM.naging.a experiments had the same settings as LGMglac.a and LGM.a, but without the effect of snow ageing. The LGMglac.e experiment was the full ESM version of LGMglac.a, which branched from the LGM.e experiment 40 years prior to the period submitted to CMIP5–PMIP3 (Fig. 1). The glaciogenic dust flux from each area was set identical to the estimates of Mahowald et al. (2006a) and the emission areas were defined as shown in Fig. S1 in the Supplement to follow their work as closely as possible; i.e. the three areas of strongest emission were the pampas of South America, central North America, and eastern Siberia. In contrast to non-glaciogenic dust, the emission of glaciogenic dust was independent of dust emission conditions and it was emitted constantly for consistency with the dust flux in Mahowald et al. (2006a) (Table 1b). Once emitted into the atmosphere, the treatment of glaciogenic dust was identical to non-glaciogenic dust. The integration of LGMglac.e was performed for 940 years. Table 2 lists the details of all the experiments.
Time series
of
The emission flux of dust (g m
List of experiments.
The global dust budget can be compared with the findings of previous
studies. Hopcroft et al. (2015) summarised it in their Table 1. They
clarified that the dust amount is highly dependent on the model, not only
for the LGM experiments but also for the PI experiments. Our emission and
load values fall in the middle of the ranges determined by previous studies.
However, they are close to those of Takemura et al. (2009) for PI.a and
LGM.a, probably because the models adopted are from the same model family
and use the same aerosol module. The emission of LGMglac.a is close to that
of Mahowald et al. (2006a), most likely because we adopted their glaciogenic
dust, but the load of LGMglac.a (39 Tg) is about 60 % of Mahowald's
loading (62 Tg), which suggests overestimation of immediate dust deposition
rates near the source areas (Fig. 4) attributable to our assumption of the
independence of dust emission from wind speed. The change in the zonal mean
dust loading in the atmosphere for the ratios LGM.a
Dust emission flux (g m
The surface temperature anomaly for LGM.a–PI.a is presented in Fig. 6a.
The cooling is about 2.3
All panels are zonal mean height plots. Ratio of the dust mass
concentration for
Now, we focus on imposed glaciogenic dust. The surface temperature at the height of 2 m is influenced by glaciogenic dust and the difference in LGMglac.a relative to LGM.a is presented in Fig. 6c. The warming (i.e. less cooling compared with the PI.a results) is pronounced in the high latitudes in contrast to the expectation of the likely cooling effect of the dust (IPCC, 2013).
Model–data comparison of dust deposition flux (g m
Model–data comparison of the ratio of dust deposition flux
estimated from the ice and sediment core data archives obtained from Kohfeld
et al. (2013) and Albani et al. (2014):
The changes in the LGMglac.a result relative to the LGM.a result for the net, longwave, and shortwave downward radiation at the surface are presented in Fig. 7a, d, and g. The figures represent the total effect of the atmospheric loading of glaciogenic dust on radiation toward the Earth's surface. Figure 7g shows a negative anomaly in shortwave radiation near the strong sources of glaciogenic dust, as well as in the northern high latitudes and the edge of Antarctica. In contrast, a positive anomaly of longwave radiation in the LGMglac.a experiment is pronounced around Antarctica and in the northern high latitudes (Fig. 7d). While the negative anomaly in shortwave radiation dominates the net change near the areas of glaciogenic dust emission, the positive longwave anomaly dominates the region surrounding Antarctica. The radiative perturbation attributable to glaciogenic dust is detailed in the next section.
Difference in surface temperature at 2 m of height (
Figure 8 shows that warming of LGMglac.a–LGM.a south of 55
Change in
The aerosol–radiation and aerosol–cloud interactions were estimated using the same method as Takemura et al. (2009). The aerosol–radiation interaction was estimated based on the difference between a standard experiment including dust impacts and another experiment under the same conditions but without the dust affecting radiation. The aerosol–cloud interaction was estimated based on the difference between a standard experiment and another experiment under the same condition but without dust at all.
Difference in 2 m air temperature between LGMglac and LGM. Red line denotes LGMglac.a–LGM.a. Green line denotes LGMglac.naging.a–LGM.naging.a, which means the change is not attributable to the ageing effect of snow. Thin and thick black lines denote LGMglac.e–LGM.e at the beginning (average of year 1 to 100 in Fig. 1) and the end (average of year 701 to 900) of the experiments, respectively. Shading represents the year-to-year standard deviation.
The net global mean radiative perturbation (aerosol–radiation and
aerosol–cloud) of dust is one of cooling at the Earth's surface for all the
experiments; i.e. PI.a:
Change in net radiative perturbation by dust at the top of
the atmosphere (TOA):
Figure 9 shows the spatial distribution of radiative perturbation by dust at
the TOA, which has a smaller difference between the LGMglac.a and LGM.a
results compared with the surface (Fig. 10a). At the TOA, although the
influence of glaciogenic dust from the pampas region is distributed over the
Southern Ocean, the positive longwave and negative shortwave radiation
almost cancel each other out. There are local negative effects over the
strong sources of glaciogenic dust but the amplitudes are much smaller than
at the surface (Figs. 9a and 10a). Figure S3 in the Supplement shows the
LGMglac.a–LGM.a anomaly of aerosol–radiation and aerosol–cloud
interactions for the TOA and the surface; it also presents the same
information but without the snow ageing effect. The panels clarify the fact that the
effect of snow ageing is independent from radiative perturbation by dust
load in the atmosphere. The figure also clarifies the fact that the anomaly of the
aerosol–radiation interaction tends to be significant at the level of 0.1 W m
LGMglac.a–PI.a and LGM.a–PI.a changes in global
mean radiative perturbation by dust
Change in net radiative perturbation by dust at the
surface:
Figure 10 shows the change in the net radiative perturbation due to dust at the surface for the LGMglac.a–LGM.a, LGMglac.a–PI.a, and LGM.a–PI.a experiments. The aerosol–radiation interaction dominates near the massive dust sources, e.g. the Sahara Desert. Except for such regions, the aerosol–cloud interaction dominates the radiative perturbation. The addition of glaciogenic dust acts to reduce shortwave radiation. The negative radiative perturbation is distinct near the emission areas. In contrast, for longwave radiation, a general positive radiative perturbation resulting from glaciogenic dust is obvious, especially near the strong sources of dust and at the edge of Antarctica. The negative shortwave radiation forcing overcomes the positive longwave radiation forcing near the sources of glaciogenic dust. However, the positive longwave radiative perturbation plays a role in the regions surrounding Antarctica. The higher dust loading in the higher troposphere in the Southern Hemisphere promotes the generation of cloud ice nucleation and high-level clouds, especially in the regions surrounding Antarctica, likely resulting in an enhanced greenhouse effect, which warms the lower troposphere (Figs. 3c and 11). Because the dust deposition flux of the standard LGM.a experiment is higher than the PI.a experiment in the Northern Hemisphere but lower in the Southern Hemisphere, the impact of glaciogenic dust might be more efficient in the Southern Hemisphere. Sagoo and Strelvmo (2017) reported global mean cooling in a “high” dust experiment, consistent with our results (Table 3). The discrepancies could arise because of different cloud ice nuclei schemes, their experimental setting (no change in land from their control), and/or because their sources of high dust emission were located mainly in desert areas, whereas our glaciogenic dust sources are located in the high latitudes.
Averaged value height plot (60–80
We extended the LGM.e experiment by 800 years beyond the original PMIP3
period (Fig. 1) and the LGMglac.e experiment was conducted for 940 years.
Because the temperatures become quasi-stable after year 600 in Fig. 1, the
average of the final 300 years is used for the analyses. The strength of the
Atlantic Meridional Overturning Circulation (AMOC) of LGM.e was reduced by about
10 Sv in the analysis period compared with the spin-up period and LGMglac.e.
The strength of the abyssal cells (Fig. S4 in the Supplement) is more stable but
with differences of a few Sverdrups between LGM.e and LGMglac.e, reflecting
the AMOC state. The surface air temperature and SST changes according to the
LGMglac.e–LGM.e results are presented in Fig. 12. The zonal mean anomaly of
air temperature over land and scatter plots of the anomaly in the proxy data
(Bartlein et al., 2011) and the anomaly of the corresponding model grids
are shown in Fig. S5 in the Supplement. It illustrates the level of agreement
between the model and the proxy archives. Pronounced discrepancy is evident
in the northern high latitudes around 70
Difference in surface temperature at 2 m of height:
Warming of the SST by the increased air temperature for LGMglac.e compared to
LGM.e is obvious in the northern high latitudes, but the magnitude of the
SST change is mostly below 0.5
The SST anomaly in both the LGM.e-PI.e and the LGMglac.e–PI.e experiments appears reasonable in comparison with the LGM SST reconstruction shown by coloured circles (MARGO project members, 2009) (Fig. 12d and e). Local cooling of the ocean temperature is seen in the lee of the source of glaciogenic dust in Argentina, which would be caused by the negative radiative perturbation (Figs. 7 and 10a).
The zonal mean potential temperature and salinity anomalies in the Atlantic
and Pacific oceans for the LGM.e–PI.e and LGMglac.e–PI.e experiments are
presented in Figs. S6 and S7 in the Supplement. The positive anomalies in the
Northern Hemisphere in Figs. S6c and 7c in the Supplement are attributable to
the difference in the strength of the AMOC between LGM.e and LGMglac.e. The
minor negative anomaly in the upper 100 m around 30
This study used the MIROC-ESM to investigate the effect of mineral dust
aerosols on the glacial climate. The representations of climatology by the
PI.a and PI.e simulations are considered reasonable for a state-of-the-art
ESM (Watanabe et al., 2011). The cooling evident in the LGM.e experiment in
comparison with the PI.e results is also generally comparable with
palaeo-proxy archives (Fig. 12). The net radiative effect of global mean dust
during the LGM is negative, which is the same trend as reported in previous
studies (Albani et al., 2014; Hopcroft 2015; Mahowald et al., 2006b; Sagoo and
Strelvmo, 2017). The global mean value is dominated by a high emission of dust
from subtropical deserts. Takemura et al. (2009) suggested an LGM-PI anomaly
of
The focus of this study was on the high latitudes, with investigation of the effect of glaciogenic dust based on new LGMglac.a and LGM.a experiments using the AGCM part of the MIROC-ESM. The effect of the addition of glaciogenic dust on climate is evident mainly as warming in the high latitudes. The effect of mineral dust aerosol on climate is highly uncertain but cooling is relatively likely (IPCC, 2013). Our results suggest the effect of dust on climate is dependent on background condition. However, our glaciogenic dust worked differently from that demonstrated by Mahowald et al. (2006b) in the zonal mean. Especially for the northern high latitudes, areas are warmed via albedo reduction because of snow ageing and because of the prolonged disappearance of snow at certain periods, which is especially pronounced in eastern Siberia. Although the longwave radiative perturbation is negative near the strong sources of glaciogenic dust flux, the snow ageing effect overcomes this cooling, resulting in a net increase in temperature. The possibility of overestimating the ageing of snow effect or our simple emission method may influence the result.
The warming effect resulting from the addition of glaciogenic dust is also seen in areas surrounding Antarctica; however, it is not attributable to snow ageing but to longwave aerosol–cloud interactions. Accounting for this effect would alter the distribution of the scatter evident in Fig. 5.5d in the IPCC Fifth Assessment Report, which shows the correlation of eastern Antarctic cooling during the LGM with the future projection.
We adopted additional dust sources from Mahowald et al. (2006a, b) as a first step in which their glaciogenic dust flux was identified as a best fit to the DIRTMAP data archive. Nevertheless, as noted, their deposition flux does not correspond well to new proxy data at locations in the Southern Ocean. However, in our case, this mismatch can also be attributed to a feature of our model, i.e. insufficient dust emission from Australia and South Africa, which is caused mainly by an overestimation of soil moisture and the resulting excess of vegetation. It should be noted that there is still a possibility of contamination by ice-rafted debris at the edge of the sea ice extent. Our study draws attention to the high dust loading over the Southern Ocean that affects the increase in surface temperature in areas surrounding Antarctica, implying the necessity of investigating climate sensitivity to the amount of dust emission in future work. However, over the Southern Ocean, SST is affected minimally (Fig. 8) by the surface radiation change (Figs. 7a and 10a), probably because of the large heat capacity of the ocean.
Glaciogenic dust was imposed constantly in this study, which is not realistic. In reality, temporal variability of glaciogenic dust should be dependent on changes both in wind speed and the threshold wind friction velocity at which dust emission is initiated. Thus, the independence of dust emission from wind speed might cause an overestimation of dust deposition rates at the grids close to emission areas and under low atmospheric loading. However, our results are in good agreement with the measurements of deposition flux in general. It will be necessary to implement a better scheme for glaciogenic dust in subsequent research. Sagoo and Strevmo (2017) prescribed a globally “idealised high” dust emission factor for their LGM-like experiment. Because our glaciogenic dust sources are located in the high latitudes, the influence of glaciogenic dust emission on the surface temperature around Antarctica is likely more pronounced in our simulation results.
In the tropics, the effect of enhanced dust input on the surface temperature is similar to what Mahowald et al. (2010) reported in their study of the middle to late 20th century but with contrasting effects at high latitudes. The major difference is that dust is enhanced at low latitudes, i.e. the Sahara–Sahel drought in the 20th century perturbation compared with the additional high dust inputs at high latitudes in our study, at which the background albedo is high because of the extended areas of snow and ice cover.
In the MIROC-ESM, snow cover in the PI.e (PI.a) experiment tends to persist in boreal spring over Siberia in comparison with reanalysis data (Fig. S8 in the Supplement). This positive bias might influence the change we see in the LGM.e (LGM.a) and LGMglac.e (LGMglac.a) experiments.
The strong effect of snow ageing is especially significant in the Northern
Hemisphere. Because snow ageing has been tuned to fit modern observations in
Hokkaido, Japan (Aoki et al., 2003, 2006), in the MIROC-ESM, a strong dust
provenance near snow-covered areas is lacking, e.g. as in the glaciogenic
dust situation seen in eastern Siberia. Therefore, evaluation of the
quantitative influence of snow ageing using various observational sites is
needed. The albedo impurity relationship provided by Aoki et al. (2003,
2006), in which ageing starts to work when the impurity is
Although we were unable to treat the effect of Fe supply to the ocean in
this model, activating the Fe fertilisation effect and enhancing the amount
of plankton would influence
Plant functional types are considered in the dynamic vegetation module but not returned to the land module in the MIROC-ESM; i.e. the climate–vegetation interaction is limited. The importance of full vegetation coupling was highlighted by O'ishi and Abe-Ouchi (2013), who suggested the necessity for future models to evaluate changes in plant functional types, especially their effect on dust cycles.
Under global warming, the amount of dust emission remains uncertain (Tegen et al., 2004; Woodward et al., 2005; Mahowald et al., 2006a; Jacobson and Streets, 2009; Liao et al., 2009; Ito and Kok, 2017). Therefore, improving the understanding of dust processes in models of the past climate would be a practical way to reduce the uncertainty of projections into the future.
PI.e and LGM.e experiment results can be downloaded from
the
ESGF server (
The supplement related to this article is available online at:
RuO, AAO, and RO discussed the palaeoclimate motivations for LGM dust and experimental design. RuO designed, conducted, and analysed experiments and prepared the paper. TT and AI advised the analyses on the aerosol module. SW, TH, and MK developed MIROC-ESM. All authors contributed to discussions.
The authors declare that they have no conflict of interest.
This article is part of the special issue “Paleoclimate Modelling Intercomparison Project phase 4 (PMIP4) (CP/GMD inter-journal SI)”. It is not associated with a conference.
This research was supported by the Integrated Research Program for
Advancing Climate Models (TOUGOU programme) from the Ministry of Education,
Culture, Sports, Science and Technology (MEXT), Japan, and was also partly
supported by the Japan Society for the Promotion of Science (JSPS) KAKENHI
under grant numbers 17H06104 and JP17H06323. The model experiments were
conducted on the Earth Simulator of JAMSTEC. The authors are grateful for
the help and inspiring discussions offered by the MIROC development team of
JAMSTEC/U and Tokyo/NIES, especially Kumiko Takata for her help in
understanding
the MATSIRO land module. We appreciate two anonymous reviewers for their
constructive and shrewd comments, which improved this study significantly. We
also thank James Buxton MSc from the Edanz Group (