2021 Volume 99 Issue 1 Pages 79-100
In this study, future changes in the rainy season in East Asia are projected based on massive ensemble simulations of about 100 members with a 60-km mesh global atmospheric model (the 60-km model hereinafter) called the “Database for Policy Decision-Making for Future Climate Change (d4PDF)”. For the present-climate, historical observed sea surface temperatures (SSTs) are prescribed to the 60-km model. In the future, 4°C warmer climate relative to the preindustrial climate, six different SST distributions projected by Atmosphere–Ocean General Circulation Models of the fifth phase of the Coupled Model Intercomparison Project (CMIP5) are given to the 60-km model. In the future, summer precipitation will generally increase in most regions of East Asia, but will decrease over western Japan. Precipitation decreases in June around 30–35°N over China, Korea and Japan. The Probability Density Function directly was derived from the massive ensemble simulations at each grid point in June and revealed that the most intense precipitation increase will occur in some regions where moderate precipitation decreases will take place in terms of the simple ensemble average. In western Japan, the onset of rainy season will delay and the retreat will occur earlier, resulting in a shorter rainy season. The decrease of precipitation in June over western Japan may be attributed to the counter-effect of the convergence of moisture to the south of Japan, originating in the southward shift of the western North Pacific subtropical high. The projected decrease in June precipitation over western Japan is confirmed to be robust, regardless of model differences in horizontal resolution, convection schemes, and with/without air-sea interactions.
In the fourth assessment report of the Intergovermental Panel on Climate Change (IPCC AR4; Intergovermental Panel on Climate Change 2007), future climate change is summarized on the base of a large amount of global warming projections produced by Atmosphere–Ocean General Circulation Models (AOGCMs) which have contributed to the third phase of the Coupled Model Intercomparison Project (CMIP3). However, the AOGCMs of CMIP3 have deficits in that they underestimate precipitation during the East Asia summer monsoon season (Kusunoki and Arakawa 2012). Additionally, the AOGCMs of CMIP3 also tend to underestimate extreme precipitation events in the East Asian region (Kusunoki and Arakawa 2012), the southern part of China (Tu et al. 2009) and the equatorial region (Dai 2006).
In the fifth assessment report of the IPCC (AR5; Intergovermental Panel on Climate Change 2013), future climate change is described with reference to global warming projections by AOGCMs that participated in the fifth phase of CMIP (CMIP5). Flato et al. (2013) demonstrated that CMIP5 AOGCMs perform better than CMIP3 AOGCMs in simulating a variety of phenomena in climate system. As for Asian summer monsoon season, CMIP5 AOGCMs also perform better than CMIP3 AOGCMs (Sperber et al. 2013; Ogata et al. 2015; Kusunoki and Arakawa 2015).
The reproducibility of present-day climatology by Atmospheric General Circulation Models (AGCMs) is evaluated by prescribing the observed sea surface temperature (SST) as a lower boundary condition. This kind of simulation is called an Atmosphere Model Intercomparison (AMIP)-type experiment. Previously, many studies on the AMIP-type experiments by AGCMs have pointed out that AGCMs underestimate precipitation in the East Asia summer monsoon season (Lau et al. 1996; Lau and Yang 1996; Liang et al. 2001; Kusunoki et al. 2001; Kusunoki 2018a). Moreover, Kang et al. (2002) and Kusunoki (2018a) revealed that AGCMs cannot reproduce the northward migration of rain bands over East Asia.
Nevertheless, the higher horizontal resolution of the AGCM performs better than the lower horizontal resolution AGCM in simulating precipitation in the East Asia summer monsoon season (Kusunoki et al. 2006; Kitoh and Kusunoki 2008; Kusunoki 2018a). Furthermore, the higher horizontal resolution AGCM accurately reproduces the intense precipitation in the East Asian summer rainy season compared to the lower horizontal resolution AGCM (Kusunoki et al. 2006; Randall et al. 2007).
Considering the advantages of higher horizontal resolution models over lower ones in simulating the climate of the East Asia region, we have conducted a series of global warming projections using global atmospheric models with 20-km and 60-km grid sizes (Kusunoki et al. 2006, 2011; Kusunoki and Mizuta 2008, 2012, 2013; Endo et al. 2012; Okada et al. 2017; Kusunoki 2017, 2018b; Chen et al. 2019; Lui et al. 2019). Table 1 summarizes the experimental design, ensemble size, and the main results. Also, future changes in precipitation intensity are investigated using the same models (Kamiguchi et al. 2006; Kitoh et al. 2009; Kusunoki et al. 2006; Kusunoki and Mizuta 2008; Endo et al. 2012; Kitoh and Endo 2016; Kusunoki 2017, 2018b; Lui et al. 2019). High-resolution model data are provided to researchers conducting impact studies, such as river discharge modeling (Kim et al. 2010). However, owing to the huge computer resources required for the time integration of these high-resolution climate models, the ensemble sizes of experiments have been limited, hindering the full evaluation of the uncertainty of future projections (See the ensemble size in Table 1).
To enhance the reliability of future projection, we conducted massive ensemble simulations, which are referred to as the “Database for Policy Decision-Making for Future Climate Change (d4PDF)” and that were generated by integrating the global model with a 60-km grid size over a period of 5000 years (Mizuta et al. 2017; Ishii and Mori 2020). As higher resolution models more accurately reproduce extreme events than lower ones, the d4PDF simulations are far more suitable for projecting future changes in extreme events than conventional lower-resolution simulations. In addition, the high horizontal resolution of the d4PDF simulations meets the requirements from impact assessment researchers who demand a high horizontal resolution for their local impact studies on global warming. Another striking advantage of the d4PDF simulations is unprecedentedly large ensemble size, which enables us to directly estimate the Probability Density Function (PDF) of the target phenomena.
Endo et al. (2017) analyzed the d4PDF simulations in order to investigate the increase in extreme precipitation in East Asia. They also identified the source of uncertainty in future projections in SST distributions through the modification of tropical cyclone activity. However, they did not investigate future changes in the rainy season in East Asia.
The purpose of this study is to rectify this and project future changes in the rainy season in East Asia including Japan, Korea and China. Also, we intend to enhance the reliability of future change in the rainy season in East Asia, as previous studies have suffered from large uncertainties due to lower-resolution of climate models and small ensemble sizes.
We used the 60 km grid size version of the Meteorological Research Institute – Atmospheric General Circulation Model (MRI-AGCM3.2H; Mizuta et al. 2012). Hereinafter, we refer to this as “the 60-km model”. It has 60 levels, with the top level at 0.01 hPa (about 80 km in altitude). For cumulus convection, we implemented the “Yoshimura scheme (YS)” (Yoshimura et al. 2015). See Mizuta et al. (2012) for further detail. The 60-km model has been used in previous studies on future precipitation change in East Asia (Endo et al. 2012; Kusunoki and Mizuta 2013; Kusunoki 2018b, c; Okada et al. 2017).
2.2 Experimental designIn this study, we adopted “the time-slice experiment”, wherein the atmospheric model is forced with an SST as the lower boundary condition (Bengtsson et al. 2009). For the present-day climate, we forced the 60-km model by observed historical SST and sea ice concentrations from the Centennial Observation-Based Estimates of SST, version 2 (COBE-SST2; Hirahara et al. 2014). The target period was 60 years, spanning 1951 to 2010. In order to create 100-member ensemble simulations, space and time varying small perturbations (δSSTs) based on SST analysis error were added to the SST distributions together with different initial atmospheric conditions.
For the future climate, the target period is around the 2090s, when the global average surface air temperature will rise by 4.1°C relative to preindustrial levels, corresponding to the Representative Concentration Pathways 8.5 (RCP8.5) scenario (Collins et al. 2013). The climatological SST differences (ΔSSTs) for the 1991–2010 period in the historical experiments and 2080–2099 in the RCP8.5 experiments simulated by the CMIP5 AOGCMs are added to the observed SST after removing the long-term trend. Six different ΔSSTs from CCSM4, GFDL-CM3, HadGEM2-AO, MIROC5, MPI-ESM-MR and MRI-CGCM3 were chosen through a cluster analysis focusing on tropical ΔSST distributions (Table 2). With respect to each of the six ΔSSTs, 15-member ensemble simulations were integrated for a 60-year period around the end of the 21st century with respect to different δSSTs and different initial atmospheric conditions, yielding a total of 6 × 15 = 90 members. Table 3 defines the simulation names used in this paper. For further technical details on the experimental design, see Mizuta et al. (2017).
Since the study by Kusunoki (2018b) and this current study use the atmospheric model, air-sea interaction is not considered. In order to evaluate the effect of air-sea coupling, we conducted a “quasi-coupled” simulation, combing the ocean model MRI.COM3 (Tsujino et al. 2010) to the MRI-AGCM3.2. To realize realistic SST year-to-year variability, flux adjustment was used (Ogata et al. 2015). The RCP8.5 scenario is assumed. See Table S1 for further technical details on the experimental design.
In order to evaluate the performance of the 60-km model, we used the One-Degree Daily data (1dd) of the Global Precipitation Climatology Project (GPCP) v1.3 covering 22 years from 1997 to 2018 (Huffman et al. 2001). The horizontal resolution is 1.0 degree in longitude and latitude which is equivalent to a longitudinal interval of about 91 km at around 35°N.
As observations have some uncertainties (Sperber et al. 2013), we used an additional dataset. We used the monthly data of GPCP v2.3 covering 25 years from 1979 to 2003 (Adler et al. 2003). The horizontal resolution is 2.5 degrees which is equivalent to a longitudinal interval of about 228 km at around 35°N.
Also, we used monthly data from the Climate prediction center Merged Analysis of Precipitation (CMAP) v1705 covering the 25 years from 1979 to 2003 (Xie and Arkin 1997). The horizontal resolution is 2.5 degrees.
As the longitudinal grid size of the 60-km model at around 35°N equates to about 51 km, the model's performances were validated against the highest resolution data of GPCP 1ddv1.3 although it did cover the entire target period of simulations from 1979 through 2003. Before the calculation of the skill scores, the model data was interpolated onto the 1-degree grid system of the GPCP 1dd data. Table 4 summarizes the features of the three observations.
Observations of summer (June to August) precipitation (Figs. 1a–c) were used to validate model simulations (Figs. 1d–f). In the observation based on the GPCP 1dd data (Fig. 1a), larger precipitations were distributed over the southeastern part of China, the Taiwan Island, the Korean peninsula, and western Japan; these originate in stationary front in the rainy season as well as the passage of extratropical and tropical cyclones. Other observations (Figs. 1b, c) also show similar patterns of precipitation, but the amount of precipitation tends to be less than the GPCP 1dd data (Fig. 1a). In particular, the large amount of precipitation over the Korean peninsula shown in Fig. 1a is not well represented in Fig. 1c. The precipitation distribution of a specific single simulation (Fig. 1d) generally resembles the GPCP 1dd data (Fig. 1a), but observed the large amount of precipitation observed over the southern part of China, Taiwan Island and the Korean peninsula (Fig. 1a) is underestimated. The situation is almost the same for the ensemble average of all the 100-member HPBs (Fig. 1e). In contrast, the Multi-Model Ensemble (MME) average of the 24 CMIP5 AGCMs (Table 5), forced with observed historical SST, generally underestimates precipitation, especially over Japan. In terms of bias as defined by simulation minus observation, the 60-km model underestimates precipitation over China, but overestimates it over Japan (Figs. 1g, h). As for CMIP5 AGCMs (Fig. 1i), the underestimation of precipitation over China, Korea, and Japan is far more evident than in the 60-km model (Figs. 1g, h).
The climatology of summer (June–August) precipitation over East Asia. Unit is mm day−1. (a) Observation by GPCP 1ddv1.3 (Table 4). Climatology of 22-years from 1997 to 2018. (b) Observation by GPCP v2.3 (Table 4). Climatology of 25-years from 1979 to 2003. (c) Same as (b) but for CMAP v1705 (Table 4). (d) The simulation of HPB_m001 averaged for 1979–2003 (25 years). (e) Same as (d) but for the average of all the 100 member HPB_ mxxx (xxx = 001 to 100); HPB. (f) Same as (d) but for 24 CMIP5 AGCMs MME average (Table 5). (g) Bias of HPB_m001 verified against GPCP 1ddv1.3. (h) Same as (g) but for HPB. (i) Same as (g) but for CMIP5 MME average (Table 5).
Figure 2 quantitatively illustrates the reproducibility of summer precipitation by models verified against the GPCP 1dd data (green circle). In Fig. 2a, root mean square errors (RMSEs) of all of the 60-km model simulations (black X) are smaller than those of all of CMIP5 AGCMs (black characters). The biases of the 60-km models are small but negative, but the absolute magnitudes of bias of the 60-km models are generally smaller than those of the CMIP5 AGCMs. In terms of RMSE and bias (Fig. 2a), the skill difference between the 60-km model and CMIP5 AGCMs is much larger than the uncertainty of observations as is estimated by the spread of the three green marks. The red circle denotes the skill of the MME average of the 60-models (red circle), where precipitations by 100-member simulations are averaged at each grid point, and skill scores then calculated. The red square denotes the average of 100 skills of the individual model simulations (AVM). The same calculation procedure was applied to the CMIP5 AGCMs (black circle and square). In terms of the skill of MME (circle) and AVM (square), the RMSE and bias of the 60-models (red) were smaller than CMIP5 AGCMs (black).
Skill of summer precipitation simulated by models verified against the GPCP 1dd v1.3 data (green circle) over East Asia (110–150°E, 20–50°N). The target domain is the same as in Fig. 1. Green square and diamond denote other observations (Table 4). X mark show HPB_mxxx (xxx = 001 to 100). Red circle shows HPB. Red square indicate the average of skill scores of all the 100 member (AVM). (a) The Root Mean Square Error (RMSE) and bias. Unit is mm day−1. The domain average of observation is shown above the panel. (b) The Taylor diagram for displaying pattern statistics (Taylor 2001). The radial distance from the origin is proportional to the standard deviation of a simulated pattern normalized by the observed standard deviation. The spatial correlation coefficient between the observed and simulated fields is given by angle from y-axis. The standard deviation of the observation in the domain is shown above the panel.
Figure 2b shows the Taylor diagram which measures the spatial pattern and spatial variance (Taylor 2001) against the GPCP 1dd data (green circle). The plots of all of the 60-km model simulations (black X) are much closer to the observation (green circle) than those of all of the CMIP5 AGCMs (black characters). This means the spatial structure simulated by the 60-km model is far superior to the CMIP5 AGCMs. This is also true in terms of the skill of the MME (circle) and AVM (square).
The advantage of the 60-km models over the CMIP5 AGCMs in simulating precipitation in East Asia is consistent with previous studies (Kusunoki 2018a, b). For further detailed the verification of simulated precipitation in East Asia by the 60-km model, see Kusunoki (2018a).
Figure 3 displays future (25 years, around the 2090s) precipitation changes in June when rainy zones stagnate over China, Korea and Japan. In the case of HFB_4K_CC (Fig. 3a), the precipitation increases in most regions in East Asia, but decreases over Japan. Decreasing the tendency of precipitation over Japan is most striking in the case of HFB_4K_GF (Fig. 3b), whereas precipitation increases in Japan in the case of HFB_4K_HA (Fig. 3c). For other SST conditions (Figs. 3d–f), the spatial change pattern is almost similar to HFB_4K_CC (Fig. 3a), with some differences. As a result, the average of all simulations of HFB_4K (Fig. 3g) shows an increase in precipitation in most regions in East Asia, but precipitation decreases over Japan. However, a statistically significant (hatched) decrease in precipitation is restricted to the western part of Japan.
Future changes (25 years around 2090s) in June precipitation (%) from the ensemble average of the presentday climatology HPB (1979–2003). Change is normalized by HPB. Hatched regions show changes above the 95 % significance level based on Student's t-test. (a) The average of all the 15-member HFB_4K_CC_myyy ( yyy = 101–115); HFB_4K_CC. (b) Same as (a) but for HFB_4K_GF. (c) Same as (a) but for HFB_4K_HA. (d) Same as (a) but for HFB_4K_MI. (e) Same as (a) but for HFB_4K_MP. (f) Same as (a) but for HFB_4K_MR. (g) Same as (a) but for the average of all the simulations HFB_4K_SS (SS = CC, GF, HA, MI, MP, MR); HFB_4K.
The striking advantage of massive ensemble simulations is that the frequency distribution of precipitation, namely the Probability Density Function (PDF), can be evaluated at each grid point. Figure 4 shows the frequency distributions of June precipitation for the present-day climate (thick blue line) and future climate (thick red line) at a grid point near Kyoto, Japan (X mark in Fig. 5a). The distribution of the present-day simulation consists of 100 members, each of which features a 25-year average precipitation at the grid point, whereas that of the future simulation consists of 90 members with 25-year average precipitation. The PDF of the future climate shifts toward less precipitation compared to that of the present-day climate. As a result, the average of the future PDF (5.33 mm day−1) is smaller than that of the present-day PDF (5.77 mm day−1). This is consistent with decreases in precipitation in terms of the ensemble average of all members, as revealed in Figs. 3g and 5a. However, the maximum future PDF (7.86 mm day−1) is larger than that of the present-day PDF (7.34 mm day−1). This extension of the future PDF toward a greater quantity of precipitation is caused by the large increase of precipitation in the simulations of HFB_4K_HA (thin magenta line). As a result, the standard deviation of the future PDF (0.87 mm day−1) becomes larger than that of the present-day PDF (0.69 mm day−1).
Probability Distribution Function (PDF) of June precipitation at a grid point near Kyoto (135.5°E, 35.5°N; X mark in Fig. 5a). The unit of frequency is %. Thick blue line indicates the frequency distribution based on 100 members of 25-year average climatology for presentday climate (HPB). The blue vertical line shows the average value of all members. Thick red line indicates the frequency distribution based on 90-member climatology for future climate (HFB_4K). Thin lines indicate the frequency distribution of 15-member future climatology depending on different SST patterns. The red vertical line shows the average value of all members. Bin width is 0.5 mm day−1. The first bin is defined as the range for 0.0 ≤ precipitation < 0.5. In the top-left corner, statistics of average (av), standard deviation (sd), maximum (max) and minimum (min) are shown for PDF of present-day (Pre) and future (Fut) climates.
Comparison between present-day PDF (HPB) and future PDF (HFB_4K) for June at each grid point. (a) Future precipitation change rate (%) based on averages of present-day and future PDFs. Hatched regions show changes above the 95 % significance level. Same as Fig. 3g but for smaller domain. X mark indicates the location of target grid for Fig. 4. (b) Probability (%) of future PDF exceeding the average value of present-day PDF. Black dots indicate the grid point where the probability is below 50 %, but at least one member of future simulations exceeds the maximum value of member in present-day simulations. (c) Probability (%) of future PDF exceeding the maximum value of member in present-day simulations. White color means the probability of 0 %.
The PDF can provide additional detailed information on future changes, as well as change obtained from the simple ensemble average of all members (Fig. 5). The shading in Fig. 5b shows the probability of the future PDF exceeding the average value of the present-day PDF. For instance, in the case of Fig. 4, the number of members of future simulations (thick red line) that exceed the average of the present-day simulations (blue vertical line; 5.77 mm day−1) is 26. As the ensemble size of future simulations is 90, the probability of the future PDF exceeding the average value of the present-day one amounts to 26/90 × 100 = 28.89 % at this grid point. The distribution of less than 50 % probability as indicated (the yellow and red regions in Fig. 5b) matches that of the decreased precipitation based on the simple ensemble average of all members (the brown region in Fig. 5a). This is reasonable because the future PDF shifts to less precipitation compared to the present-day PDF.
As another example of additional information derived from the PDF, the probability of the future PDF exceeding the maximum value of the member of present-day simulations is depicted in Fig. 5c. For instance, in the case of Fig. 4, the number of members in future simulations (thick red line) that exceed the maximum of the present-day simulations (7.34 mm day−1) is 3. As the ensemble size of future simulations is 90, the probability of the future PDF exceeding the maximum value of the present-day PDF amounts to 3/90 × 100 = 3.33 % at this grid point. This probability is almost positive over the target region (Fig. 5c), which is due to the increase in the spread of the future PDF (figure not shown).
At some grid points, even though the average precipitation decreases in the future, larger precipitation that exceeds the maximum value of the present-climate emerges in the future (the black dots in Fig. 5b). This suggests the possibility of an increase in natural disasters associated with increasingly intense precipitation even if the average precipitation decreases. In this way, the PDFs derived from massive ensemble simulation can provide additional useful information as well as conventional information on future climate change based on simple ensemble averages.
5.3 Precipitation change for all monthsFigure 6 shows the precipitation change through the average of all simulations HFB_4K for all 12 months. In January to May, September, October, and December, precipitation increases over China, Korea and Japan, but decreases to the south of Japan. In June to August, November, precipitation increases over China and Korea, but decreases over Japan. In particular, the precipitation increase over China is large and statistically significant.
Future changes in monthly precipitation projected by the all simulations HFB_4K. Unit is %. Hatched regions show changes above the 95 % significance level. Black boxes in (f) indicate the three target region of China (110–120°E, 20–45°N), Korea (120–130°E, 20–45°N) and Japan (130–142°E, 20–45°N) in Figs. 7 and 8. Red box indicates the target region of western Japan (130–142°E, 30–38°N) in Fig. 9.
We applied a similar analysis as in Fig. 5b to all 12 months. We also found in all of the months that larger quantities of precipitation exceed the maximum value of that in the present-day climate in the future over some regions where average precipitation decreases in the future (figure not shown).
5.4 Rainy seasonFigure 7 shows future changes in the seasonal march of the rainy season over Japan (130–142°E, 20–45°N; Black box in Fig. 6f). The contour lines of 8 mm day−1 in the present-day simulation HPB (Fig. S1a) are also plotted as a reference for the rainy season in Japan. The seasonal march of precipitation in the future simulations are also illustrated in Figs. S1b–h. Precipitation decreases at the beginning to middle of June around 30°N, except for HPB_4K_CC (Fig. 7a). Also, precipitation decreases in the middle of July around 35°N except for HPB_4K_MP (Fig. 7e). As a result, the ensemble average of all of the simulations (Fig. 7g) shows a precipitation decrease of around 30–35°N in June as well as around 35°N in July. In the region of Japan around 30–35°N, the rainy season starts at the beginning of June and ends in the middle to end of July. Therefore, Fig. 7g suggests that the rainy season in Japan starts later and ends faster in the future climate, resulting in a shorter rainy season.
Future changes in the time-latitude cross section of pentad mean precipitation averaged over Japan for longitude 130–142°E. Contours indicate the precipitation amount of 8 mm day−1 in the present-day climatology HPB (Fig. S1a). Unit is %. Black dots show changes above the 95 % significance level. The target region of Japan is indicated by the black box labeled with ‘Japan’ in Fig. 6f. The target period is from pentad 27 (11–15 May) to 46 (14–18 August). Seasonal march of precipitation in the present-day and future simulations are shown in Fig. S1.
Figure 8 compares the future change in the seasonal march of the rainy season across three regions. In China (Fig. 8a), precipitation generally increases in the target time-latitude domain, but decreases around 30–35°N in June and around 25°N in July. Similar change can be observed in Korea, with some differences (Fig. 8b). In China (Fig. 8a) and Korea (Fig. 8b) in July and August, precipitation increases around the 30–35°N region, indicating a delay in the termination of the rainy season. Focusing on the rainy season defined by the contour lines of 6 mm day−1 in the present-day simulation HPB (Figs. S2a, S3a), precipitation persistently increases during the rainy season in China (Fig. 8a) and Korea (Fig. 8b). These changes in the timing of the rainy season are partially consistent with Kusunioki (2018c), which objectively and rigorously defines the rainy season in terms of grid point bases. In contrast, an inverse change in precipitation appears over Japan in July and August.
Future changes in the time-latitude cross section of pentad mean precipitation averaged for three longitudinal zones (black boxes in Fig. 6f). The target pentad period is the same as Fig. 7. All the simulation HFB_4K is used. Unit is %. Black dots show changes above the 95 % significance level. (a) China (110–120°E, 20–45°N). Contours indicates the precipitation amount of 6 mm day−1 in the present-day climatology HPB. See Fig. S2 for the seasonal march of precipitation in the present-day and future simulations. (b) Same as (a) but for Korea (120–130°E, 20–45°N). See Fig. S3 for the seasonal march of precipitation in the present-day and future simulations. (c) Japan (130–142°E, 20–45°N). Same as Fig. 7g. Contours indicates the precipitation amount of 8 mm day−1 in the present-day climatology HPB.
Figure 9 depicts the time evolution of pentad precipitation averaged over western Japan (red box in Fig. 6f), with precipitation decreasing in June. Figure 9a compares the observed precipitation profiles with simulations in HPB. In observations (black), precipitation starts to increase at about the end of May and reaches a maximum at around the end of June to the beginning of July. Finally, precipitation decreases to a minimum around the end of July. Roughly speaking, the rainy season in this region begins at the beginning of June and ends at the end of July. Simulations of HPB (blue) accurately reproduce the time evolution of the rainy season, although the peak value of the precipitation is somewhat underestimated. Figure 9b compares the present-day simulations in HPB (blue) with the future simulations in HFB_4K (red). In the future, the rainy season starts earlier and ends faster which is consistent with Fig. 7g. This means that the period of the rainy season becomes shorter in the future. Another striking feature is the increased peak value of precipitation at the beginning of July. These future changes are also confirmed by the precipitation change illustrated in Fig. 9c. Moreover, these changes are statistically significant (closed green circles in Fig. 9c).
Time evolutions of pentad mean precipitation averaged over western Japan. The target region (130–142°E, 30–38°N) is indicated by the red box in Fig. 6f. Unit is mm day−1. The target period is from pentad 27 (11–15 May) to 46 (14–18 August). (a) Observations by GPCP 1ddv1.3, GPCP v2.3 and CMAP v1705 (black; Table 4) and the present-day climatology HPB (blue). (b) The present-day climatology HPB (blue) and the future climatology HFB_4K (red). (c) Future change (green). Closed circles show changes above the 95 % significance level.
Kusunoiki (2018b) investigated future precipitation changes in East Asia based on ensemble simulations with the 20-km and 60-km models, assuming the same RCP8.5 scenario as in this study. However, the SST setting in Kusunoiki (2018b) differed from the current study. The delay in the onset of the rainy season is consistent with Fig. 8 of Kusunoki (2018b), in which the 20-km and 60-km models were used, but the more rapid retreat and increased peak precipitation is not found in Kusunoki (2018b). The delay in the termination of the rainy season indicated in ensemble simulations with the 20-km model for different SSTs (Okada et al. 2017) is not found in this study. The discrepancies between previous studies and this one can be attributed to differences in experimental settings.
The time evolutions of the regional average precipitation over 15 regions of East Asia are also investigated (Figs. S4–6). The model reproduce the seasonality of precipitation with reasonable accuracy, but precipitation in the summer season tends to be underestimated in some regions (Fig. S4; regions 9–11, 14–15, and 17). Future precipitation generally tends to increase across all seasons, especially in summer, except for the region over western Japan (Fig. S5; region 11). Some regions show statistically significant decrease in precipitation during summer (Fig. S6; regions 7, 8, 11, and 18).
5.6 Extreme precipitation eventsIncreased peak precipitation in the middle of the rainy season, indicated in Fig. 9b, suggests an increase in intense precipitation during the rainy season, which might enhance the potential for natural disasters. Extreme precipitation events often lead to flooding, landslides, and drought. Figures 10b–d show future changes in extreme events (Table 6), which are defined by Frich et al. (2002). In the case of non-leap years, the number of 5-day cumulative precipitation events (unit is mm), ranging from 1–5 January (1st to 5th days of the year; 1st data) to 27–31 December (361st to 365th days of the year; 361st data), amounts to 361. R5d is defined as the maximum of these 361 5-day cumulative precipitation datasets. In each year, the daily precipitation data amounts to a total of 365. PMAX is defined as the maximum of these 365 daily datasets. For comparison, the annual average precipitation is also shown (Fig. 10a). Note that the indices of these extreme events are based on annual statistics from January through December. Annual precipitation PAVE increases across almost all domains in East Asia except for regions to the southeast of Japan, which is consistent with Fig. 6. A decrease in the PAVE around Japan (Fig, 10a) is mainly attributed to the decrease in monthly precipitation in June to August around Japan (Figs. 6f–h).
Future changes in extreme precipitation events (annual statistics, Table 6) projected by the all simulations. Unit is %. Hatched regions show changes above the 95 % significance level. (a) Annual precipitation (PAVE). (b) Maximum 5-day precipitation total (R5d). (c) Maximum 1-day precipitation (PMAX). (d) Consecutive dry days (CDD). Note that color bar is reversed for CDD. Values of RA in the captions of each panels denote domain average.
The maximum 5-day precipitation total R5d increases in almost all domains in East Asia (Fig. 10b). Similarly, the maximum 1-day precipitation total PMAX increases over almost all domains in East Asia (Fig. 10c). The domain averages of the precipitation changes for PAVE, R5d, and PMAX are 15.0, 21.7, and 25.1 %, respectively. The increase in the most intense precipitation PMAX (25.1 %) is larger than that of moderate and weak precipitation PAVE (15.0 %). This is consistent with previous studies (Kusunoki and Mizuta 2013; Kusunoki 2018b). The increase in intense precipitation in East Asia is also consistent with Endo et al. (2017), based on the d4PDF simulations.
Consecutive dry days of CDD decrease over the midlatitudes and high latitudes of China, but increase in the southern part of that country, as well as Taiwan island, the East China Sea and Japan (Fig. 10d). Focusing on Japan, intense precipitation increases (Figs. 10b, c), but the possibility of drought also does. This leads to a concentration of rainfall events within a short period of time, which could cause natural disaster. This is consistent with Kusunoki (2018b).
We investigated the reason why precipitation decreases over western Japan in June. The mean sea level pressure increases to the south of 20°N (Fig. 11a), implying a southward shift in the western North Pacific subtropical high (WNPSH). This causes clockwise moisture transport around 10–30°N, as is indicated by the arrows in Fig. 11b. Consequently, moisture converges to the south of Japan around 20–30°N, where precipitation increases (Fig. 11c). Aa a counter-effect of convergence, moisture diverges over Japan (Fig. 11b), finally resulting in a decrease in precipitation over western Japan (Fig. 11c). This mechanism is consistent with the results of Kusunoki (2018b).
Future changes in different physical variables for June. The ensemble averages of all simulations are illustrated. (a) Mean sea level pressure (hPa). Hatched regions show changes above the 95 % significance level. (b) The vertically integrated water vapor flux (arrow; kg m−1 s−1) and its convergence (shading; mm day−1). The unit of convergence is converted to mm day−1 assuming the density of liquid water as 1 g cm−3. Arrow is plotted only if change is above the 95 % significance level. (c) Precipitation. Hatched regions show changes above the 95 % significance level. Black indicates the target region (130–142°E, 25–45°N) for Figs. 7 and 12.
The upward motion originating from the horizontal advection of warm air along westerly jets contributes to the formation and maintenance of rainbands over East Asia (Sampe and Xie 2010). The model accurately reproduces the northward migration of the 200 hPa zonal component of wind (U200) and precipitation from April to July in East Asia, although the model underestimates the amount of precipitation (Fig. S7). Figure 12 illustrates the present-day climatology HPB, the future climatology HFB_4K and future changes. See Figs. S8 and S9 for the SST dependence. U200 increases in June around 20–35°N. Also, precipitation increases in June around 20–30°N, which suggests that the intensification of the westerly jet also contributes to the intensification of the rain band around 20–30°N to the south of Japan in June. Therefore, both the southward shift of the WPSH in June (Fig. 11) and the intensification of U200 in June (Fig. 12) may cause the increase in precipitation around 20–30°N to the south of Japan, resulting in the decrease in precipitation over western Japan.
Seasonal march of monthly zonal wind velocity at 200 hPa (U200, contour interval 10 m s−1) and monthly precipitation (shade) averaged for 130–142°E. Target region is indicated by black box in Fig. 11c. (a) The present- day climatology HPB. (b) The future climatology HFB_4K. (c) Future change HFB_4K - HPB. Contour interval is 1 m s−1. Green contour indicates the zero value of U200.
Here we compare the distribution of precipitation in June revealed in this study with previous studies. Figure 13a shows future precipitation change in June by the 20-km model with the YS convection scheme forced with the MME average of SST projected by CMIP5 AOGCMs (Kusunoiki 2018b). The ensemble size is one. The precipitation strikingly declines in western Japan, but increases in other regions, which is partly consistent with our results (Fig. 13g). Figure 13b is the same as Fig. 13a but for the 60-km model. The change pattern in Fig. 13b far more closely resembles our results (Fig. 13g) than the 20-km results (Fig. 13a). This is reasonable because both experiments use the same 60-km model. Figures 13 c and d are the same as Fig. 13a but for different convection schemes of Arakawa-Schubert (AS; Randall and Pan 1993), Kain- Fritsch (KF; Kain and Fritsch 1990), respectively. A decrease in precipitation also appears in Fig. 13c, but is weak and not clear in Fig. 13d.
Comparison of future precipitation change for June between previous studies and this study. The length of target periods for the present-day and future simulations are 25 years for all the simulations. Hatched regions show changes above the 95 % significance level. (a) The 20-km model with the YS convection scheme forced with CMIP5 AOGCM MME average SST (Mizuta et al. 2014; Fig. 2a). Ensemble size is one. Kusunoki (2018b). The period for the present-day climatology is 25 years from 1979 to 2003. The period for the future climatology is 25 years from 2075 to 2099. (b) Same as (a) but for the 60-km model. (c) Same as (b) but for the AS convection scheme. (d) Same as (b) but for the KF convection scheme. (e) The 20-km model with the YS convection scheme coupled to the ocean model. Ogata et al. (2015). The period for the present-day climatology is 25 years from 1979 to 2003. The period for the future climatology is 25 years from 2075 to 2099. See Table S1 for details. (f) Same as (e) but for the 60-km model. (g) This study. Same as Figs. 3g and 6f.
Since the study by Kusunoki (2018b) and this current study use an atmospheric model, air-sea interaction is not considered. In order to evaluate the effect of air-sea coupling, we conducted a coupled simulation using the ocean model MRI.COM3 (Tsujino et al. 2010). In order to realize realistic SST variability, flux adjustment was employed (Ogata et al. 2015). The RCP8.5 scenario was assumed. The coupled simulation using the 20-km model (Fig. 13e) with the YS scheme also showed a decrease in precipitation over western Japan. The decrease in precipitation over western Japan by the 60-km model (Fig. 13f) was weaker than by the 20-km model (Fig. 13e). Regardless of differences in horizontal resolutions, convection schemes, and with/without air-sea interaction, we confirmed that the decrease in precipitation in June over western Japan is robust.
Our results are also compared to the fully coupled AOGCM experiments (Figs. S10f–h) for June. Simulations by the MRI-CGCM3 (Fig. S10f) and MRIESM1 (Fig. S10g) also show a decrease in precipitation in western Japan, but the simulation by the MRIESM2 (Fig. S10h) shows an increase. Overall, we again confirmed the robustness of the precipitation in decrease in western Japan in June with some exceptions.
The following results were obtained in this paper.
Supplement includes Figures S1–S10 and Table S1.
Figure S1. The seasonal march of pentad precipitation averaged over Japan for longitude 130–142°E. The target period is from pentad 27 (11–15 May) to 46 (14–18 August). Unit is mm day−1. The target region of Japan is indicated by the black box labeled with ‘Japan’ in Fig. 6f. (a) The present-day climatology HPB. The ensemble average of all 100 members. Contour indicates the precipitation amount of 8 mm day−1. (b) The future climatology; HFB_4K_CC. Ensemble average of all 15 members. The contour is the same as in (a). (c) HFB_4K_GF. (d) HFB_4K_HA. (e) HFB_4K_MI. (f) HFB_4K_MP. (g) HFB_4K_MR. (h) HPB. The ensemble average of all 90 members.
Figure S2. Same as Fig. S1 but for over China for longitude 110–120°E and contour of 6 mm day−1.
Figure S3. Same as Fig. S1 but for over Korea for longitude 120–130°E and contour of 6 mm day−1.
Figure S4. Time evolution of pentad precipitation over 15 regions over East Asia. Unit is mm day−1. Black lines show observations (Table 4). GPCP 1ddv1.3; solid line. GPCP v2.3; log dash line. CMAP v1705; short dash line. Blue lines show the present-day climatology HPB. All the 100-member ensemble average.
Figure S5. Same as Fig. S4 but for the model climatology only. Blue lines show the present-day climatology HPB. Red lines show the future climatology HFB_4K. All the member ensemble average.
Figure S6. Same as Fig. S5 but for future change as HFB_4K - HPB. Closed circles show changes above the 95 % significance level.
Figure S7. Seasonal march of monthly zonal wind velocity at 200 hPa (U200, contour interval 10 m s−1) and monthly precipitation (shade, mm day−1) averaged over Japan for longitude 130–142°E. Target region is indicated by black box in Fig. 11c. The 25 year climatology from 1979 to 2003. (a) Observations. U200; The Japanese 55-year Reanalysis (JRA-55; Kobayashi et al. 2015). The 25 year climatology from 1979 to 2003. Precipitation; The GPCP 1ddv1.3. The 22 year climatology from 1997 to 2018. (b) The present-day climatology HPB. The 25 year climatology from 1979 to 2003. 3-dimensional daily wind of model data were not archived because of huge data amount. Here we used monthly data for U200 and precipitation in this analysis.
Figure S8. Same as Fig. S7 but for the model future simulations.
Figure S9. Same as Fig. S8 but for the model future changes. Contour interval is 1 m s−1. Green contour indicates the zero value of U200.
Figure S10. Comparison of future precipitation change for June between previous studies and this study. The present-day climatology; 1979–2003 (25 years). The future climatology; 2075–2099 (25 years), except for (e) in which 25-year climatology around the 2090s is used. Hatched regions show changes above the 95 % significance level. (a) The 20-km model MRIAGCM3.2S with the YS convection scheme forced with CMIP5 AOGCM MME average SST (Mizuta et al. 2014; Fig. 2a). Ensemble size is one. Kusunoki et al. (2018b). (b) Same as (a) but for the 60-km grid model MRI-AGCM3.2H. (c) The 20-km model MRIAGCM3.2S with the YS convection scheme coupled to the ocean model. Ogata et al. (2015). See Table S1 for details. (d) Same as (c) but for the 60-km model MRI-AGCM3.2H. (e) This study. Same as Figs. 3g and 6f. (f) MRI-CGCM3 (Yukimoto et al. 2012). RCP8.5. Convection scheme is the Arakawa-Schubert scheme (Randall and Pan 1993). Grid size is 120 km. (g) Same as (f) but for MRI-ESM1 (Yukimoto et al. 2012). (h) MRI-ESM2 (Yukimoto et al. 2019). The Shared Socioeconomic Pathway (SSP) 5–8.5 (O'Neill et al. 2016). Convection scheme is the YS scheme. Grid size is 120 km. The grid 80 size for panels in the first column (a, c) is 20 km. The grid size for panels in the second column (b, d, e) is 60 km. The grid size for panels in the third column (f, g, h) is 20 km.
Table S1. Experimantal design of “quasi-coupled” simulations.
This work was supported by the Integrated Research Program for Advancing Climate Models (TOUGOU), Grant Number JPMXD0717935561, from the Ministry of Education, Culture, Sports, Science and Technology (MEXT), Japan. This study utilized the database for Policy Decision making for Future climate change (d4PDF), which was produced under the SOUSEI program. We acknowledge the international modeling groups for providing model data for our analysis, the Program for Climate Model Diagnosis and Intercomparison (PCMDI) for collecting and archiving the CMIP5 multi-model data, and the Joint Scientific Committee (JSC)/CLIVAR Working Group on Coupled Modeling (WGCM). The data archive at the Lawrence Livermore National Laboratory (LLNL) is supported by the Office of Science, U.S. Department of Energy. We appreciate the advice and comments by anonymous reviewers which enhanced the quality of the manuscript.
