Summertime land–sea thermal contrast and atmospheric circulation over East Asia in a warming climate—Part II: Importance of CO₂-induced continental warming

Abstract

In this the second of a two-part study, we examine the physical mechanisms responsible for the increasing contrast of the land–sea surface air temperature (SAT) in summertime over the Far East, as observed in recent decades and revealed in future climate projections obtained from a series of transient warming and sensitivity experiments conducted under the umbrella of the Coupled Model Intercomparison Project phase 5. On a global perspective, a strengthening of land–sea SAT contrast in the transient warming simulations of coupled atmosphere–ocean general circulation models is attributed to an increase in sea surface temperature (SST). However, in boreal summer, the strengthened contrast over the Far East is reproduced only by increasing atmospheric CO₂ concentration. In response to SST increase alone, the tropospheric warming over the interior of the mid- to high-latitude continents including Eurasia are weaker than those over the surrounding oceans, leading to a weakening of the land–sea SAT contrast over the Far East. Thus, the increasing contrast and associated change in atmospheric circulation over East Asia is explained by CO₂-induced continental warming. The degree of strengthening of the land–sea SAT contrast varies in different transient warming scenarios, but is reproduced through a combination of the CO₂-induced positive and SST-induced negative contributions to the land–sea contrast. These results imply that changes of climate patterns over the land–ocean boundary regions are sensitive to future scenarios of CO₂ concentration pathways including extreme cases.

1 Introduction

Recent modeling studies have shown that the increase of surface air temperature (SAT) over land is larger than that over the ocean in global warming experiments (e.g. Manabe et al. 1991; Dommenget 2009; Dong et al. 2009; Boer 2011). This robust feature, namely the change in land–sea thermal contrast, has also been confirmed in the recent trend of observational data (e.g. Lambert and Chiang 2007; Sutton et al. 2007; Jones et al. 2013). The projected future change in land–sea SAT contrast could influence regional atmospheric circulations in a warming climate (e.g. Kamae et al. 2014, hereafter Part I).

The physical mechanisms for the increasing global land–sea SAT contrast in a warming climate have been investigated by using coupled atmosphere–ocean general circulation models (CGCMs) and atmosphere-only general circulation models (AGCMs) in previous studies (Lambert and Chiang 2007; Joshi et al. 2008; Dommenget 2009; Dong et al. 2009; Boer 2011; Lambert et al. 2011; Joshi et al. 2013). The land–sea contrast is strengthened in both transient and equilibrium responses to external forcing (Manabe et al. 1991). However, the increasing land–sea SAT contrast in a warming climate is not simply a consequence of the different thermal inertias between land and ocean, but originates from the different properties of the surface and boundary layer between the land and ocean (Manabe et al. 1991; Sutton et al. 2007; Joshi et al. 2008; Dong et al. 2009). In response to ocean surface warming, enhanced evaporation results in a smaller SAT change compared with the land surface (Manabe et al. 1991; Sutton et al. 2007). Joshi et al. (2008) showed that temperature change in the lower free troposphere above the boundary layer (i.e. 600–700 hPa level) was similar over land and ocean and that the land–sea thermal contrast in a warming climate was confined to the boundary layer. They then suggested that the lapse rate change in the boundary layer over the ocean, in response to sea surface temperature (SST) increase, is larger than that over land because of the difference in relative humidity and the associated lapse rate between land and ocean (see Fig. 1 in Joshi et al. 2013). Compo and Sardeshmukh (2009) demonstrated that the bulk of the global-scale annual-mean land surface warming in recent decades could be largely reproduced by AGCM experiments forced by observed SST increase. In addition, stomatal closure in response to CO₂ increase and the associated decrease of cloud amount also affect the land–sea SAT contrast in future projections (e.g. Dong et al. 2009; Doutriaux-Boucher et al. 2009).

The change in the land–sea SAT contrast can influence atmospheric thermodynamic structures and associated circulation patterns in a warming climate (e.g. Ueda et al. 2006; Dong et al. 2009; Sun et al. 2010; Fasullo 2012; Li et al. 2012; Wu et al. 2012; Zhu et al. 2012; Bayr and Dommenget 2013). For example, the land–sea SAT contrast over the Far East in boreal summer has increased in recent decades (0.11 K decade⁻¹, Part I). This trend is expected to continue in future warming projection obtained from general circulation models (GCMs) conducted in Coupled Model Intercomparison Project phase 5 (CMIP5; Taylor et al. 2012). In the future climate projections, it was also revealed that the land SAT and geopotential height in the mid-troposphere over the Far East increase in a GCM accompanying anomalous pressure and atmospheric circulation patterns over East Asia. This manifests as a stronger surface anticyclone over the Okhotsk high area and around the Philippines, an increase of cold northeasterly wind over northern Japan called Yamase, and strengthened rainfall over the Meiyu–Baiu front (Kimoto 2005; Arai and Kimoto 2008; Part I).

Kimoto (2005) showed that the increase of CO₂ concentration in the atmosphere alone induces a continental warming and an associated summertime atmospheric circulation pattern over East Asia in an AGCM simulation (Fig. 4b in Kimoto 2005). In transient warming experiments conducted by CGCMs, the global-mean SAT (SAT _g ) increases gradually with time, in response to the increase of CO₂ concentration in the atmosphere. However, the direct radiative effect of continuously increasing amounts of CO₂ and the effect of an increase of SST have played distinct roles in recent climate change (e.g. Folland et al. 1998; Bracco et al. 2004; Deser and Phillips 2009). The warming of the troposphere and land surface by CO₂ radiative forcing plays an important role in changes to the hydrological cycle and atmospheric circulations (e.g. Mitchell et al. 1987; Allen and Ingram 2002; Gregory and Webb 2008; Cao et al. 2012; Bony et al. 2013; Chadwick et al. 2014), cloud, and radiation (e.g. Gregory and Webb 2008; Kamae and Watanabe 2012, 2013; Tomassini et al. 2013) in a warming climate. Kamae and Watanabe (2013) revealed the changes in cloud, precipitation, and atmospheric circulation through the change in land–sea SAT contrast in the fast response to the atmospheric CO₂ quadrupling. However, the seasonal and regional responses in the atmospheric circulation, their mechanisms, and their consistency among different GCMs have not been examined sufficiently. In this paper, we investigate the possible roles of the SST effect and CO₂ direct radiative effect on projected future changes over the Far East during boreal summer (Part I) and their consistencies by using the CMIP5 multi-model dataset. In CMIP5, we can investigate the roles of CO₂ and SST by using sensitivity experiments prescribing the CO₂ concentration and SST with multi-AGCMs. We also attempt to decompose the contributions of the two effects in idealized transient experiments and future scenario experiments by using a reconstruction method detailed below. Section 2 describes the model, experiments and methods. Section 3 presents the roles of CO₂ and SST on the changes in idealized simulations forced by the transient CO₂ increase. In Sect. 4, the roles of them in future scenario experiments are also examined. Section 5 presents a discussion and implication. Section 6 describes the conclusions of Parts I and II.

2 Data and methods

2.1 CMIP5 model ensemble and experiments

In this study, we used the CMIP5 multi-model dataset (Taylor et al. 2012) to analyze the results of idealized transient experiments, as in Part I. In Part II, the abrupt change in response to CO₂ quadrupling and the equilibrium responses in atmosphere-only model experiments are also investigated, in order to quantify the contributions of different forcings (see Sect. 2.2).

We used nine models for which all the outputs of transient and sensitivity runs required are available (Table 1). For the idealized transient experiments, a 1 %/year CO₂ increase experiment (hereafter 1 % CO₂) is used. In 1 % CO₂, the atmospheric CO₂ concentration is prescribed and increased by 1 %/year starting from 1850 of a pre-industrial control simulation until a quadrupling of the CO₂ concentration is achieved with respect to the pre-industrial levels after 140 years. All the other external forcings such as aerosols are kept at their pre-industrial values. Anomaly Δ of any variables in 1 % CO₂ run shown in this study is defined by the difference between averages over the last 30 years in 1 % CO₂ run (111–140) and corresponding 30 years in the pre-industrial control simulation. To quantify the possible contributions of atmospheric CO₂ and SST forcing on the results of the 1 % CO₂ run, we analyzed an Atmospheric Model Intercomparison Project (AMIP) type control (amip) and two sensitivity experiments; amip4 × CO₂ and amipFuture (Bony et al. 2011; Taylor et al. 2012). In amip and amip4 × CO₂, the models were driven by observation-based SST and sea–ice concentration for 1979–2008 with atmospheric CO₂ concentration set to 280 and 1,120 ppmv, respectively. In amipFuture, a spatially patterned SST increase, derived from a multi-model ensemble mean of the 1 % CO₂ experiment in CMIP phase 3, is superimposed on the observed SST prescribed in amip. In Sect. 3.1, we also analyzed amip4 K run in which spatially-uniform +4 K SST is superimposed on the SST to examine a sensitivity of results on the spatial pattern of the imposed SST. Global-mean anomaly of the SST in amipFuture is equal to that in amip4 K (i.e. 4 K). Anomaly Δ is defined by the difference of 30-year averages between sensitivity runs (amip4× CO₂, amip4 K, and amipFuture) and control run (amip). These sensitivity experiments were also utilized in the previous studies examining climate feedbacks (Webb and Lock 2013; Demoto et al. 2013), changes in the hydrological cycle (Bony et al. 2013; Huang et al. 2013; Ma and Xie 2013; Widlandky et al. 2013), and tropospheric and land adjustments to increasing CO₂ (Kamae and Watanabe 2012, 2013; Tomassini et al. 2013).

Table 1 Changes in global-mean SAT (ΔSAT _g ), land–sea SAT (ΔSAT _lnd  − ΔSAT _ocn ) and thickness contrasts (ΔZ3085 _lnd  − ΔZ3085 _ocn ) over the Far East (110°E–170°E, 30°N–70°N), and orthogonal wind speed at 150°E, 500 hPa level (ΔU _orth 500 _150E ) simulated in 1 % CO₂ run (CGCM) with CMIP5 nine models (similar to Table 1 in Part I)

We also examined the results of historical (1850–2005) and future scenario (2006–2100) experiments; the latter forced by a representative concentration pathway (RCP) of future anthropogenic changes in atmospheric compositions, by using the MIROC5 model (Watanabe et al. 2010). The RCP runs have four types of scenarios (RCP2.6, RCP4.5, RCP6.0, and RCP8.5, Meinshausen et al. 2011). Δ in RCP runs are calculated by anomalies of 30-year averages (2041–2070 and 2071–2100) from 111-year average (1850–1960) in the historical simulation (Table 2). Furthermore, we investigated the rapid change due to abrupt CO₂ increase, and slow change associated with SAT _g increase, by using the abrupt4× CO₂ run (hereafter step experiment), which is branched off from the control run by instantaneously quadrupling the CO₂ concentration from its pre-industrial level and holding it fixed. Δ is calculated by anomaly from the corresponding mean state of the pre-industrial control simulation.

Table 2 Global-mean radiative forcing (ΔF), ΔSAT _g , and the land–sea contrasts over the Far East simulated in MIROC5

2.2 Reconstruction of idealized transient and future scenario runs with CO₂ step experiment

In the AMIP-type experiments, we can determine explicitly two factors for the changes in land–sea contrast: direct radiative forcing due to CO₂ increase (CO₂ effect) and surface warming (SST effect). As for the responses in the idealized transient experiments using CGCMs, the simulated changes in land–sea contrast are composed of the two factors. In Sect. 4, we attempt to decompose the two by using a method below.

First, we reconstruct the response y _i in the 1 % CO₂ with the step experiment at year i, by using a method proposed in Good et al. (2011, 2013):

$$y_{i} = \sum\limits_{j = o}^{i} {w_{i - j} x_{j} ,}$$

(1a)

$$w_{i - j} = \frac{{\Delta F_{i - j} }}{{\Delta F_{{4 \times {\rm{CO}}_{2} }} }},$$

(1b)

where x _j is the response of the same variable at year j of the step experiment. w _i−j is a scaling factor of the annual-mean radiative forcing change from year i to year j of the given scenario (ΔF _i−j ) relative to the radiative forcing due to CO₂ quadrupling ($\Delta F_{{4 \times {\text{CO}}_{2} }}$). We obtained $\Delta F_{{4 \times {\text{CO}}_{2} }}$ = 8.70 W m⁻² by the amip4× CO₂ experiment (Kamae and Watanabe 2013).

As for the RCP runs, we follow the method of Forster and Taylor (2006) to estimate w _i−j from the output of the model experiments [Eq. (2) in Good et al. 2013] to take into account the CO₂ and non-CO₂ forcings, such as aerosols and land use changes. The global-mean net radiative flux at the top of atmosphere (N, positive downward) is approximated by global-mean radiative forcing and ΔSAT _g as:

$$F = N + \lambda \Delta SAT_{g} ,$$

(2)

where F is the radiative forcing and λ is the climate feedback parameter. We obtain λ = 1.69 W m⁻² K⁻¹ from the least-square regression coefficient of N to ΔSAT _g in the step experiment (Kamae and Watanabe 2013) by following the method of Gregory et al. (2004). Equation (2) can be used to estimate the time-varying F from the simulated time-series of N and ΔSAT _g . The applicability of this method to the Far East land–sea contrast is discussed in Sects. 4 and 5.

3 Distinct roles of CO₂ forcing and SST increase on the land–sea contrast

3.1 Global and regional changes

As a prelude, we reconfirm the spatial pattern of ΔSAT in global warming experiment shown in Part I. Figure 1a shows the June–August (JJA)-mean ΔSAT in the 1 % CO₂ run by the nine CMIP5 models. Similar to the earlier studies, positive ΔSAT is generally larger over land than over the ocean. An anomalous warming over the Arctic Ocean, the so-called arctic amplification (e.g. Serreze and Barry 2011), is not found during boreal summer (Fig. 1a). The maximum warming associated with decreasing sea ice cover occurs over the Barents Sea and Antarctic Ocean. In low latitudes, the warming is large in the eastern equatorial Pacific and western equatorial Indian Ocean associated with the spatial pattern of projected SST increase (e.g. Collins et al. 2010). Figure 1e shows the change in temperature (T) at the 700 hPa level (ΔT ₇₀₀ ), representing the free tropospheric response. In contrast to ΔSAT, ΔT ₇₀₀ represents a zonally uniform increase independent on the distribution of land and ocean. The rest panels of Fig. 1 show the effects of CO₂ quadrupling, uniform and patterned SST +4 K identified in the AMIP-type sensitivity experiments (amip4× CO₂ minus amip, amip4 K minus amip and amipFuture minus amip, respectively). Consistent with earlier reports (Joshi et al. 2008; Compo and Sardeshmukh 2009), SST increase alone without radiative forcing could establish the land–sea ΔSAT contrast in low latitudes (Fig. 1c, d), whereas the land–sea ΔT ₇₀₀ contrast is not clearly found (Fig. 1g, h). In the amip4× CO₂ run, both ΔSAT and ΔT ₇₀₀ over land (ocean) increase (change little) owing to the CO₂-induced continental warming in contrast to the fixed ocean temperature (Fig. 1b, f). Note that the continental warming due to the CO₂ increase is also found in the other seasons (figure not shown).

Fig. 1

CMIP5 nine models ensemble means of JJA-mean a–d ΔSAT (K) and e–h difference of temperature (ΔT) at 700 hPa (ΔT ₇₀₀ , K). Model means in a, c, d, e, g, h are calculated with normalized values (K⁻¹) by global-mean ΔSAT (ΔSAT _g ) in the individual models. The anomalies are calculated by averages in 30 years of a, e 1 % CO₂ run (111–140) relative to control run, b, f amip4× CO₂, c, g amip4 K and d, h amipFuture run relative to amip run

Some differences are found in the responses between amip4 K and amipFuture. For example, the surface warming over the eastern equatorial Pacific and western equatorial Indian Ocean (over Southern Hemisphere mid- and high-latitude) are larger (smaller) in amipFuture relative to amip4 K associated with the spatial pattern of imposed SST anomaly in amipFuture (e.g. Fig. 1 in Lu et al. 2008). These differences are important for atmospheric circulations, regional moisture convergence, and precipitation responses in a warmer climate over the tropics and subtropics (Huang et al. 2013; Ma and Xie 2013; Widlandky et al. 2013). In contrast, the warming over land area (both at the surface and 700 hPa) over Northern Hemisphere mid- and high-latitude (40°N–80°N) both in amip4 K and amipFuture runs are relatively smaller than those over the ocean (Fig. 1c, d, g, h). These results indicate that the sign of change in the free-tropospheric (and surface) land–sea temperature contrast does not depend on whether the imposed SST is uniform or patterned. We only show the results derived from amipFuture run hereafter because the spatial pattern of the imposed SST in amipFuture is rather similar to that of SST anomaly simulated in 1 % CO₂ run than the uniform increase. Note that spatial patterns of the SST anomaly simulated in 1 % CO₂ in CMIP5 models show some differences with the imposed SST pattern in amipFuture run (see Sects. 3.2, 5).

Figure 2 shows the zonally-averaged land- and ocean-mean ΔSAT during JJA in the 1 % CO₂, AMIP-type CO₂ quadrupling and patterned +4 K SST experiments. In amipFuture run, the land-mean ΔSAT is larger than the ocean-mean ΔSAT in low latitudes (35°S–40°N, Fig. 2c). Hence the established land–sea ΔSAT contrast is confined to low latitudes (35°S–40°N), which is inconsistent with the results of the 1 % CO₂ run. The ΔSAT contrast in mid- and high-latitudes in the Northern Hemisphere (40°N–70°N) found in the 1 % CO₂ run (Figs. 1a, 2a) exhibits in the experiment with CO₂ radiative forcing (Figs. 1b, 2b).

Fig. 2

Zonal-mean ΔSAT in JJA over land (black) and ocean (blue) in a 1 % CO₂ run relative to control run, b amip4× CO₂ and c amipFuture run relative to amip run. a, c Normalized by ΔSAT _g . Shadings represent ±1 SDs in nine models

On a global perspective, previous studies have revealed that the spatially uniform warming in the free troposphere in response to SST increase induces land–sea ΔSAT contrast through the difference in lapse rate in the boundary layer between land and ocean (Joshi et al. 2008; Compo and Sardeshmukh 2009; Dommenget 2009). However, this mechanism does not work efficiently in mid- and high-latitudes in the Northern Hemisphere during boreal summer. Figure 3 shows the vertical profile of difference between land-mean ΔT and ocean-mean ΔT in response to patterned +4 K SST increase (amipFuture minus amip) in global-mean and the Far East (110°E–170°E, 30°N–70°N). On a global perspective (Joshi et al. 2008), ΔT ₇₀₀ over both land and ocean shows similar values, resulting in a land–sea temperature contrast in the boundary layer due to the difference of lapse-rate between the land and the ocean (Figs. 1d, h, 3). In contrast, the land-mean ΔT ₇₀₀ in response to patterned +4 K SST increase is smaller than that over the ocean in mid- and high-latitudes in the Northern Hemisphere including the Far East (Figs. 1d, h, 3). In the tropics, the free-tropospheric warming tends to be homogenized because of the large Rossby radius of deformation, often referred to as the weak temperature gradient (Sobel et al. 2001). However, free-tropospheric temperature over the continental interiors in mid- and high-latitude is less sensitive to the SST increase owing to the smaller Rossby radius of deformation than that in low latitude (Figs. 1g, h, 3). The smaller warming of free troposphere over land than ocean results in a negative or no land–sea ΔSAT contrast over mid- and high-latitude including the Far East (Figs. 1, 2, 3). The response of the warming contrast over East Asia to the uniform SST increase is similar to that to the patterned SST increase (Fig. 1c, d, g, h), indicating that the sign of the response of the land–sea contrast is not sensitive to whether the imposed SST is uniform or patterned.

Fig. 3

Vertical profile of normalized global-mean (orange) and the Far East mean (blue) ΔT difference between land and ocean (K K⁻¹) in amipFuture run relative to amip run. Solid curves and shadings represent the nine models ensemble means and ±1 SDs. Crosses at bottom of the panel represent the normalized ΔSAT difference between land and ocean (K K⁻¹)

3.2 Thermodynamic structure over the Far East

Given the essential component of the CO₂-induced continental warming for the land–sea ΔSAT contrast in the global warming projection, we focused on the thermodynamic structure in the troposphere over the Far East. We also examined changes in atmospheric circulation which is tightly associated with the change of the tropospheric thermodynamic structure over East Asia (Part I). Figures 4 and 5 compare the changes in thickness and geopotential height in the 1 % CO₂ and two AMIP sensitivity runs. The change in thickness between 300 and 850 hPa (ΔZ3085) can be used as an index of tropospheric temperature change. As shown in Part I, the tropospheric warming over the Eurasian continent induces an anticyclonic anomaly in the upper troposphere, the land–sea ΔZ3085 contrast (Fig. 4a), and an associated change in the East Asian atmospheric circulation (Fig. 4; Table 1). Figure 4b, c show the effects of CO₂ quadrupling and patterned +4 K SST increase. Tropospheric warming over the ocean is larger than that over land in response to the SST increase (Fig. 4c). This result is consistent with the larger ΔSAT over the ocean than the land (Fig. 1d). The anomalous tropospheric warming over the ocean induces negative land–sea thermal and geopotential height contrasts in the upper troposphere (Figs. 4c, 5c and curve in Fig. 5f). In contrast, CO₂ quadrupling induces the anomalous tropospheric warming over the continent and associated positive contrasts (Figs. 4b, 5b, e), which is consistent with that simulated in the 1 % CO₂ run (Figs. 4a, 5a, d). Note that the maximum warming in the upper troposphere, particularly over low latitude (Fig. 5a) is attributed to the SST increase (Fig. 5c). Spatial distribution of horizontal wind at 850 hPa level shows northeasterly anomaly over East Asia in 1 % CO₂ run (Fig. 4a, Part I). Figure 4 also shows that the spatial distributions of changes in the dynamical circulation are tightly coupled with the spatial patterns of tropospheric thermodynamic structure over East Asia in the three runs (1 % CO₂, amip4× CO₂, and amipFuture). An index of dynamical atmospheric circulation over East Asia used in Part I, axis-orthogonal wind (U _orth , positive southwesterly) defined as the speed of the horizontal wind vector along orthogonal axis to the land-to-ocean line (130°E, 65°N–160°E, 35°N, Part I), shows negative (northeasterly) anomalies in 1 % CO₂ and amip4× CO₂ and a positive (southwesterly) anomaly in amipFuture (Fig. 4; Table 1).

Fig. 4

Similar to Fig. 1a, b, d but for thickness between 300 and 850 hPa (ΔZ3085, shade, m), horizontal wind anomaly at 850 hPa (m s⁻¹), and ΔZ _eddy at 300 hPa level (ΔZ _eddy 300, m) over the Far East and western North Pacific

Fig. 5

Similar to Fig. 4 but for vertical structures of a–c ΔT and d–f difference of geopotential height on the land-to-ocean line (130°E, 65°N–160°E, 35°N, blue line in Fig. 4a). The lines in d–f represent ΔZ3085 (m K⁻¹ in d, f, m in e, right axis)

The above results represent that the effects of CO₂ and SST play the distinct roles on the changes simulated in 1 % CO₂ run over the Far East. Previous studies (e.g. Lambert et al. 2011) revealed that the general characteristics of climate response to sum of the CO₂ and SST forcing can be largely simulated by the sum of the individual climate responses to the individual forcings although a nonlinearity term is not negligible. Now we examine the relationship of the responses simulated in the three experiments among the multi-models before we test an assumption of additivity of the CO₂ and SST forcings (see Sect. 4). Figure 6 shows relationship of the land–sea ΔSAT and ΔZ3085 contrasts in the individual models. All the values are also listed in Table 1. The multi-model ensemble means in the individual simulations and their spreads reveal that: (1) the CO₂ and SST effects act as positive and negative contributions for the ΔSAT and ΔZ3085 contrasts in the 1 % CO₂ run; and (2) inter-model variances show clear positive correlations between land–sea ΔSAT and ΔZ3085 contrast in all the ensembles (correlation coefficients are 0.63, 0.74, and 0.58 in the 1 % CO₂, amip4× CO₂, and amipFuture, respectively). This means that the positive changes in the land–sea contrasts of the Far East in global warming experiment are determined by the large positive change due to CO₂ forcing and the small negative change due to SST increase.

Fig. 6

A scatter plot of the land–sea ΔSAT (K) and ΔZ3085 contrasts (m) in nine models of 1 % CO₂ (red), amip4× CO₂ (open grey circle), and amipFuture (filled grey circle). Crosses represent the nine models mean and the ranges of their ±1SDs. All the values are listed in Table 1. The lines represent least-square regressions. R means correlation coefficient

It is also apparent that the responses in 1 % CO₂ runs are larger than the sum of the responses simulated in amip4× CO₂ and amipFuture. One factor is that the amplitude of global warming is generally larger in the amipFuture run than 1 % CO₂ run (multi-model means of ΔSAT _g are 3.91 and 4.55 K in 1 % CO₂ and amipFuture respectively, Table 1). The responses in amipFuture should be scaled by the ΔSAT _g ratio between amipFuture and 1 % CO₂ in the individual models when we compare the responses. Another factor for the discrepancy is a difference in spatial patterns of SST anomaly between 1 % CO₂ and amipFuture runs. The imposed SST anomaly in amipFuture shows anomalous warming in the mid-latitude western North Pacific and Sea of Okhotsk (e.g. Fig. 1 in Lu et al. 2008) but some models simulate weaker warming in that region in 1 % CO₂ run than the imposed SST. The difference of SST anomaly can influence the land–sea ΔSAT and tropospheric warming contrast over the Far East. For example, SST effects estimated by the responses in amipFuture run in some models including MPI-ESM-MR are overestimates (Table 1; Fig. 6) because those models simulate smaller warming in the western North Pacific in 1 % CO₂ run than the prescribed-SST. However, other models including MIROC5 simulate similar patterns of SST in 1 % CO₂ run to the prescribed-SST (see Sect. 4), resulting in better additivity of the CO₂ and SST effects than the other models (Table 1). We should note that this method has a limitation in decomposing the two effects quantitatively by using the results of AMIP-type sensitivity experiments conducted under the CMIP5 protocol. In the next section, we further examine the additivity of CO₂- and SST-induced influences on the projected changes in the transient warming experiments including RCP-type future scenario runs by using MIROC5 model.

4 Decomposing effects of radiative forcing and surface temperature increase

It is suggested that the changes in land–sea contrast over the Far East in any runs of 1 % CO₂, RCPs, or the step experiment, could largely be decomposed into the positive effect of ΔF and the negative effect of other factors (e.g. SST increase). To confirm this hypothesis, we next investigate ΔF-dependent and other components of changes determined in the step experiment conducted with MIROC5 model. We also should take into account possible factors for errors in linearity (see Sect. 5).

Figure 7a shows the change in the land–sea ΔSAT contrast over the Far East in the single-member step experiment (black crosses) and 12-member ensemble experiments (red crosses) starting from initial states 1 month apart to remove seasonality and increase the signal-to-noise ratio (Sect. 2.2. in Kamae and Watanabe 2013). The plot located near the y-axis (ΔSAT _g  ≈ 0) means the effect of CO₂ forcing which is independent of ΔSAT _g . To quantify these effects, we utilize a regression method proposed in Gregory et al. (2004):

$$\Delta X \approx \Delta X_{F} + \alpha \Delta SAT_{g} ,$$

(3)

where ΔX is the change in a given variable in the step experiment, α is the slope of the regression line against ΔSAT _g , and ΔX _F is the y-axis intercept. ΔX _F and α at a given grid point determined by this method, represent the ΔF effect and ΔSAT _g effect, respectively. Note that the ΔF effect is equal to the CO₂ effect in any experiments with CO₂ forcing (1 % CO₂, step experiment, and amip4× CO₂) except the historical and those RCP runs including non-CO₂ forcings. This method has been adopted widely in studies on climate feedback, energy budgets, and the hydrological cycle in step experiments (e.g. Gregory and Webb 2008; Andrews et al. 2009; Bala et al. 2010; Watanabe et al. 2010; Kamae and Watanabe 2013).

Fig. 7

MIROC5 step experiment and its application to reconstruction of the 1 % CO₂ run. a Scatter plot of the land–sea ΔSAT contrasts (K) in step experiment. Black plots represent a result of single member over 30 years. Red crosses are 12-member ensemble mean over 5 years and its ±1 SDs. The lines represent the least-square regressions. b Time series of ΔSAT land–sea contrast in 1 % CO₂ run (gray) and its 30-year running mean (black). Blue curve shows the reconstruction by the step experiment using the method detailed in Sect. 2.2. Light blue and purple curves are ΔF and ΔSAT _g effects on the reconstructed time series estimated by Eq. 2. c Comparison of four values averaged in year 41–70 (2× CO₂) and 111–140 (4× CO₂). Error bars are ±1 SDs of interannual variabilities in 30 years

The ΔF effect, represented by the fast response to CO₂ quadrupling, is consistent qualitatively (positive sign) and comparable quantitatively (2.06 K) with the CO₂ quadrupling as effect estimated by the amip4× CO₂ run (Table 1). This result indicates that the ΔF effect estimated by the AMIP-type simulation (amip4× CO₂) can largely reproduce the ΔF effect identified by the fast response to abrupt CO₂ increase in the CGCM run (abrupt4× CO₂). Although the ΔSAT contrast in the single-member run swings largely during the long-term integration because of the large interannual variability, the least-square regression lines in both of the two ensembles show clear declines of the ΔSAT contrast in response to the positive change in ΔSAT _g . The ΔSAT _g effect (−0.39 K K⁻¹) is determined by the slope of the regression line. Next, we attempt to reconstruct the results of the 1 % CO₂ run by the response in the step experiment, following Good et al. (2011, 2013) method detailed in Sect. 2.2. Figure 7b shows the ΔSAT contrast in the 1 % CO₂ run and its reconstruction by the step experiment. The reconstruction (blue curve) generally captures the 30-year running-mean of the 1 % CO₂ run (black curve). Note that the reconstruction skill is limited in the first 40 years because of weak forcing during this period. However, the values in the periods of CO₂ doubling (years 41–70) and quadrupling (years 111–140) are well reconstructed. In addition, the spatial pattern of ΔSAT in years 111–140, shown in Fig. 8a, is also well reconstructed by this method, indicating that the reconstruction skill is good when the forcing is sufficiently strong. Note that MIROC5 simulates the warming peak in the mid-latitude western North Pacific in 1 % CO₂ run (Fig. 8a) which is consistent with the prescribed-SST in amipFuture run (see Sect. 3.2).

Fig. 8

Map of ΔF and ΔSAT _g effects on the reconstructed JJA-mean ΔSAT (K) in MIROC5 1 % CO₂ run. a Result of 1 % CO₂ run (contour, interval = 1.5 K) and its reconstruction (shading). b ΔF and c ΔSAT _g effects. d Fraction (%) of the ΔF and e ΔSAT _g effects on the reconstructed ΔSAT. Contours represent ±25 %

According to the reconstruction, we decompose the reconstructed values in the 1 % CO₂ run into the ΔF effect (light blue) and ΔSAT _g effect (purple, Fig. 7b, c). The ΔF and ΔSAT _g effects contribute to the 1 % CO₂ run positively and negatively, and the former is larger than the latter both in the transient (Fig. 7b) and equilibrium state (Table 2), resulting in the positive sign in the reconstruction. The negative ΔU _orth in the 1 % CO₂ run is also attributed to the ΔF effect (Table 2). In principle, using this method, we can compare the ΔF and ΔSAT _g effects for any period in both the transient and equilibrium experiment; however, some potential problems might exist, which will be discussed in Sect. 5.

Figure 8 shows spatial patterns of the ΔF and ΔSAT _g effects on the reconstructed ΔSAT in the 1 % CO₂ run. They show positive and comparable contributions for positive ΔSAT over land in mid- and high-latitudes. The fraction of ΔF effects on the Far East ΔSAT is about 40 % over land (Fig. 8d) in contrast to the ΔSAT _g contribution on the ΔSAT over the ocean (Fig. 8c, e). Note that the relative contributions of the two are also comparable in the other seasons but the land–sea contrasts are relatively unclear than JJA (figure not shown). Further investigations would be needed to explain the spatial patterns of the contributions in the other seasons with particularly attention for influences of sea–ice reductions in the surrounding oceans.

Figure 9 shows the land- and ocean-mean ΔSAT during JJA over the Far East in relation to the annual-mean ΔSAT _g and ΔF in the individual experiments. Note that the transient changes shown in Fig. 9 (small circles) are the reconstructed 1 % CO₂ run (blue curve in Fig. 7b) which represent well the low-frequency variability in the 1 % CO₂ run (Fig. 7b, c). In the transient warming experiment (1 % CO₂ and RCP runs), both ΔSAT _g and ΔF increase gradually and the plotted marks shift to the upper right of the panels. In contrast, the plotted marks of the AMIP-type sensitivity experiments are located on the x-axis or near the y-axis because of fixed ΔF or ΔSST in these experiments (0 W m⁻² and 0 K, respectively). The ΔSAT _g in the step experiment increases rapidly with constant ΔF owing to the CO₂ quadrupling (8.70 W m⁻²). In the individual runs, the increase of the land-mean (ocean-mean) ΔSAT over the Far East region (color) follows largely the sum of the ΔF and ΔSAT _g effects (contour) obtained by the regression method using the results of the step experiment (Eq. 2). Namely, ΔX _F corresponds to the ΔF effect and αΔSAT _g represents the ΔSAT _g effect. We assume linearity and additivity of the forcings, which is discussed in Sect. 5. The land-mean ΔSAT increases in response to ΔSAT _g and ΔF result in the estimated contours being diagonally downwards to the right. In contrast, the estimated contour of the ocean-mean ΔSAT lies vertically because it is mostly determined by ΔSAT _g .

Fig. 9

Scatter diagram of ΔSAT _g and global-mean ΔF in the individual runs simulated in MIROC5. Colors in plotting marks are a land-mean, b ocean-mean ΔSAT over the Far East. Sets of gray and black squares are averages in the period 2041–2070 and 2071–2100 in RCP runs. Three large circles are the first year and 2–5 years mean in 12-member ensemble means, and 6–30 years mean in single-member, respectively (left–right). Small circles are the reconstructed 1 % CO₂ run (blue curve in Fig. 7b). The isolines represent a land-mean, b ocean-mean ΔSAT over the Far East estimated by the ΔF and ΔSAT _g effects (Eq. 3)

Figure 10 shows the land–sea ΔSAT contrast over the Far East in the individual runs with the ΔSAT _g and ΔF, i.e. the difference between Fig. 9a, b. The ΔSAT contrast shows negative and positive signs in response to increasing ΔSAT _g and ΔF respectively, resulting in the contour (sum of the two effects) being diagonally downwards to the left. The individual transient and sensitivity runs follow this relationship: the large positive change in amip4× CO₂ and the step experiments, the gradual increase in the 1 % CO₂ and RCP runs, and the large negative change in amipFuture. The signs of the transient runs are positive, according to the larger increase in the land-mean SAT compared with the ocean-mean (Fig. 9).

Fig. 10

Similar to Fig. 9 but for the land–sea ΔSAT contrast over the Far East (Table 2)

Figure 11 shows the ΔZ3085 land–sea contrast over the Far East. It also shows similar features to those of the ΔSAT contrast (Fig. 10), indicating that both the ΔSAT and ΔZ3085 contrasts over the Far East can be largely interpreted by the sum of the positive and negative effects of ΔF and ΔSAT _g . Note that some discrepancies between the sum of the two effects and the results of the runs are found. For example, the sign of change in SAT contrast in the 2041–2070 period of the RCP2.6 run is opposite to that estimated (Fig. 10). Some quantitative mismatches are also found in the ΔSAT and ΔZ3085 land–sea contrasts (Figs. 10, 11), indicating some limitations and problems in the estimation method as discussed below.

Fig. 11

Similar to Fig. 10 but for the ΔZ3085 land–sea contrast over the Far East (Table 2)

5 Discussion

The opposing effects of ΔF and ΔSAT _g on the land–sea contrast over the Far East contribute positively and negatively to the projected change under global warming. The sum of these effects can reproduce the results of the simulations largely including the AMIP-type experiments (amip4× CO₂, amipFuture), transient warming experiments (1 % CO₂, RCP runs), and step experiment (abrupt4× CO₂, Figs. 9, 10, 11). Does a nonlinear term have any contributions on the results of those simulations? Lambert et al. (2011) revealed that no substantial nonlinearities were found in indices of atmospheric circulation and cloud radiative effect when the effects of ΔF associated with the doubling of CO₂ concentration and spatially uniform ΔSST were added to an AGCM. Deser and Phillips (2009) also revealed a high degree of linearity in the response to ΔF and ΔSST in atmospheric circulation trends during 1950–2000, simulated in an AGCM. As noted in Sect. 4, some inconsistencies are found between the linear summed estimation of the ΔF and ΔSAT _g effects on the land–sea contrast and the results of the simulations (Figs. 9, 10, 11). Possible factors contributing the discrepancies are discussed below.

One problem in the linear summed estimation method is the limited signal to noise ratio owing to the limited ensemble members and substantial interannual variability over the Far East (Fig. 7b, Part I). The limited signal to noise ratio may contribute to uncertainties in the estimated ΔF and ΔSAT _g effects particularly during the period when the imposed forcing and the response are weak (Fig. 7b, c). Another factor for the inconsistencies between the changes simulated in the individual runs (colors in Figs. 9, 10, 11) and the sum of the ΔF and ΔSAT _g effects (contours in Figs. 9, 10, 11) is the difference in the simulated spatial patterns of SST among the simulations. Although MIROC5 simulates similar SST pattern in 1 % CO₂ run generally to the prescribed SST in amipFuture (e.g. warming peak in the mid-latitude western North Pacific, Fig. 8a), some differences in regional SST anomalies can lead to a difference of land–sea warming contrasts over the Far East between the simulations. In addition, differences of increasing SST simulated in the RCP runs with 1 % CO₂ run can also be factors for the inconsistencies. It is needed to examine how do these SST influence on responses of global and regional land–sea contrast in future works.

In Sect. 4, we apply the method reconstructing the transient responses from the step experiment. The reconstruction method of Good et al. (2013) applied to the RCP runs might not be valid when the non-CO₂ forcing is dominant in the given scenarios. The effects of regional non-CO₂ forcing, including aerosols in the RCP runs, on the land–sea contrast over the Far East cannot be reproduced by the method in this study. For example, differences of imposed forcing owing to changes of the atmospheric concentration of aerosols and land use might contribute to the regional ΔSAT, ΔZ3085 and resultant changes of the land–sea contrast. In fact, some discrepancies are found, as detailed above. The possible influence of regional non-CO₂ forcing on the simulated change is another factor for the inconsistency between the changes simulated in the individual runs and the sum of the ΔF and ΔSAT _g effects.

The method assuming linear additivity by ΔF and ΔSAT _g effects used in this study has some problems described above. However, the reconstructions reproduce well the relationship that the run with stronger ΔF shows a larger change in land–sea contrast. The reconstruction has good skill in the scenarios with strong ΔF (i.e. 1 % CO₂ and RCP8.5) owing to the dominant effect of the much-increased ΔF due to increasing CO₂. We have to pay attention to the fact that this reconstruction cannot be applied when the non-CO₂ forcing has a relatively large contribution to the total change. We should also note that the different patterns of increasing SST simulated in given experiments can contribute to the simulated change in regional land–sea warming contrast substantially. We cannot apply this method to explain all the regional climate responses simulated in the transient warming experiments. However, we can conclude that the ΔF effect plays an important role in the land–sea contrast over the Far East during boreal summer in the projected future climate simulated in the CMIP5 models.

In this study, we tried to interpret the physical mechanism of the change in land–sea contrast over East Asia by decomposing ΔF and ΔSAT _g effects. The change of the land–sea contrast over East Asia can be largely explained by a part of the global characteristics that the relative importance of ΔF and ΔSAT _g effect depends on latitude (Figs. 1, 2, 8). However, the responses in the land–sea thermal contrast over mid- and high-latitude simulated in the SST experiments (amip4 K and amipFuture) show some regionalities (Fig. 1). The weaker warming of the free-troposphere relative to the surrounding regions is found over eastern and western sides of the Eurasian Continent. Further studies are needed to examine possible mechanisms and influences on projected regional climate change in the future scenario experiments.

The increase in summertime land–sea contrast over the Far East is not attributed to ΔSAT _g but CO₂-induced ΔF. This result suggests a possibility that the anomalous warming over land relative to the ocean in the mid- and high-latitude Northern Hemisphere could be used as an index of ΔF due to changes in atmospheric concentrations of greenhouse gases under given emission pathways. The changes in land–sea SAT contrast and associated atmospheric circulation in the future climate projections may be sensitive to given CO₂ concentration pathways including the cases of decreasing concentration (e.g. Cao et al. 2011). In addition, the kind of ΔF (CO₂ or solar radiation) in geoengineering method may affect the sign and magnitude of the changes. Bony et al. (2013) suggested that changes in dynamical component of tropical precipitation due to the direct effect of CO₂ increase may not be avoided by ΔSAT _g control with solar radiation management. The change in the land–sea contrast over the Far East gives another example of the importance of the direct effect by CO₂ increase. It is needed to assess effects of other types of forcings (anthropogenic aerosol emissions, volcanic eruptions, and solar radiation) and its additivity on changes of the thermodynamic structure and atmospheric circulation over the Far East in future works.

6 Conclusions

In this two-part paper, we examined the changes in the land–sea thermal contrast and summertime atmospheric circulation over East Asia under the global warming with observations and CMIP5 dataset. ΔSAT over land shows relatively larger increase in the recent decades than that over ocean, indicating the possible impact on the land–sea thickness contrast and associated atmospheric circulation pattern that accompanies a stronger northeasterly to northeastern Japan. The observed long-term changes of the land–sea SAT contrast in the recent decades are reproduced by CMIP5 multi-models, but with somewhat larger magnitude. The projected positive changes in land–sea SAT contrasts among the models are correlated with the change in the thickness contrast between them, indicating the positive changes in SAT and thickness contrasts are the part of the consistent responses to global warming among CMIP5 models.

The AMIP-type sensitivity experiments show positive and negative contributions of CO₂ quadrupling and SST increase to the projected future change in land–sea contrast over the Far East. The projected positive ΔSST contributes to strengthen the SAT contrast in low latitudes but not in the mid- and high-latitude because of limited warming in the free troposphere over the interior of the continent. The transient changes of land–sea contrast in 1 % CO₂ and RCP experiments can be approximated by the sum of the positive effect of global ΔF due to CO₂ increase and negative effect of ΔSAT _g . The experiment with larger ΔF shows larger increase in land–sea contrast than that with the smaller ΔF experiment due to the larger effect summed of the two (ΔF and ΔSAT _g effects) over land relative to ocean.

This study gives an example that the direct effect of CO₂ increase is essential for the regional climate change projected in the future scenario experiments. Bony et al. (2013) also pointed its importance on the projected change in the dynamical tropical precipitation. The distinct influences of the increases in atmospheric CO₂ and SST have mainly been focused on the changes of the mean climate states. In future studies, respective roles of the imposed forcings on climate variability and frequency of extreme events should be examined to assess possible risks to ecosystems and society associated with severe climate events due to the different climate forcings.