Atmospheric Rivers Bring More Frequent and Intense Extreme Rainfall Events Over East Asia Under Global Warming

Portions of East Asia often experienced extremely heavy rainfall events over the last decade. Intense atmospheric rivers (ARs), eddy transports of moisture over the middle latitudes, contributed significantly to these events. Although previous studies pointed out that landfalling ARs will become more frequent under global warming, the extent to which ARs produce extreme rainfall over East Asia in a warmer climate remains unclear. Here we evaluate changes in the frequency and intensity of AR‐related extreme heavy rainfall under global warming using a set of high‐resolution global and regional atmospheric simulations. We find that both the AR‐related water vapor transport and rainfall intensify over the southern and western slopes of mountains over East Asia in a warmer climate. ARs are responsible for a large fraction of the increase in the occurrence of extreme rainfall in boreal spring and summer. ARs will bring unprecedented extreme rainfall over East Asia under global warming.

• Atmospheric rivers bring extreme rainfall over East Asia in current and future climates • Atmospheric rivers are responsible for a large fraction of the increase in extreme rainfall over East Asia under global warming • Intensified water vapor transports by atmospheric rivers cause recordbreaking extreme rainfall events in a warmer climate

Supporting Information:
Supporting Information may be found in the online version of this article.

Current and Future Climate Simulations
We use a set of ensemble simulations with an AGCM and a RCM from the database for Policy Decision making for Future climate change (d4PDF, Ishii & Mori, 2020;Mizuta et al., 2017). The AGCM used in the d4PDF is the Meteorological Research Institute AGCM version 3.2 (MRI-AGCM3.2; Mizuta et al., 2012) with 60-km horizontal resolution and 64 vertical layers. This model realistically simulates rainfall (Ito et al., 2020) and AR activity  over East Asia. For the past simulations, the AGCM was driven by historical radiative forcing (greenhouse gases, aerosols, and ozone), sea surface temperature (SST) and sea ice (Hirahara et al., 2014) for 1951-2010. In addition to the PAST simulations, outputs from 60-year future simulations (PLUS4K simulations hereafter) were also examined to evaluate future change in extreme rainfall associated with ARs. In this set of simulations, the AGCM was driven by anomalous radiative forcing corresponding to the level of year 2090 in the Representative Concentration Pathway (RCP) 8.5, warmed SST and changes in sea ice. Six SST warming patterns obtained from CMIP5 multi-model results (Mizuta et al., 2014) were used to perform six-type PLUS4K simulations. The six SST warming patterns were based on RCP8.5 simulations by CCSM4, GFDL-CM3, HadGEM2-AO, MIROC5, MPI-ESM-MR, and MRI-CGCM3 (CC, GF, HA, MI, MP, and MR hereafter). The SST warming was scaled so that the simulated global-mean surface air temperature is 4 K warmer than the pre-industrial level . In this study, a 60-year climatology in the PAST and PLUS4K is compared to evaluate the effects of global warming on AR activity and extreme rainfall. Using the d4PDF AGCM output,  found an increase in AR frequency over East Asia under global warming. They also showed that the projected increase in AR activity is related to the patterns of tropical SST changes and the associated changes in WNPSH intensity. In the present study, we further evaluate AR contributions to extreme rainfall using output from the downscaling simulations described below.
The d4PDF also provides outputs from regional downscaling simulations covering Japan, the Korean Peninsula and eastern China ( Figure S2a in Supporting Information S1). The d4PDF RCM simulations were performed by the Nonhydrostatic Regional Climate Model (NHRCM; Sasaki et al., 2008) with 20-km grid spacing and 40 vertical layers . Model outputs at every one hour with 20-km horizontal resolution are more instrumental in examining extreme rainfall than the 6-hr outputs from the d4PDF AGCM simulations with a 60km horizontal resolution. The combination of the d4PDF AGCM and RCM simulations greatly facilitates the evaluation of the changes in extreme rainfall associated with ARs under climate change (Fujita et al., 2019;Imada et al., 2020;Kawase et al., 2019Kawase et al., , 2020Miyasaka et al., 2020). In this study, we defined extreme rainfall events in the PAST simulations as upper 0.1% cases of wet hours (rainfall >0.1 mm hr −1 ) using 1-hr outputs of the RCM simulations. This threshold obtained from the PAST simulations was also used to define extreme rainfall events in the PLUS4K simulations and evaluate changes in frequency of events under global warming (see Figure 1c of Schär et al., 2016). We calculated a 60-year climatology for each simulation, then averaged across ensemble members (10 members for PAST and 30 members for PLUS4K simulations; 5 members for each of the six-type PLUS4K simulations) to minimize the effects of atmospheric internal variability Kamae, Shiogama, et al., 2017).

AR Detection
Using outputs from the d4PDF AGCM runs, ARs were detected at 6-hr intervals based on vertically integrated water vapor transport (IVT) as calculated below:- where q is specific humidity, u is zonal wind, v is meridional wind, p is air pressure, and g is the acceleration due to gravity. ARs were detected using IVT field following the method described in , which is modified from the method proposed in Mundhenk et al. (2016). Anomalous IVT in the PAST simulations was obtained by comparing with its daily climatology. For the PLUS4K simulations, daily climatology in the PLUS4K simulations was subtracted to obtain IVT anomaly (Kamae et al., , 2021. ARs were detected based on shapes (length >1,500 km, area >7.8 × 10 5 km 2 , length-width ratio >1.325) and intensity (140 kg m −1 s −1 ) of the anomalous IVT.  showed that the climatology and interannual variability of AR occurrence frequency in the d4PDF AGCM simulations are generally consistent with those in observations. Details on future change in the frequency of the detected ARs in the d4PDF AGCM simulations can be found in Kamae et al. ( , 2021. We compared the AR data obtained from the AGCM runs with rainfall simulated in the RCM runs to evaluate AR-related rainfall and its extremes. The 6-hr, 60-km resolution data of AGCM-based AR were linearly interpolated to the 1-hr, 20 km-resolution. We confirmed that the positions of the detected ARs in the AGCM runs are generally consistent with those in the RCM runs ( Figure S3 in Supporting Information S1). Figure 1 shows the climatology of AR-related rainfall for individual seasons. As shown in , AR-related rainfall pattern manifests a strong orographic effect. Peaks in AR-related rainfall are found on the southern and western slopes of mountains over East Asia (western Japan, central Japan, northern Japan, Taiwan, the Korean Peninsula, and northeastern China). In particular, AR rainfall dominates at the southwestern slope of Japan's Alps ( Figure S2b in Supporting Information S1) in all seasons, therefore we focus on this region in the latter part of this study. The spatial pattern of AR rainfall climatology in the simulations is similar to that obtained from observations and an atmospheric reanalysis , indicating that the AR-rainfall relationship over East Asia is well reproduced in the AGCM/RCM. Table 1 summarizes the climatology of AR-related rainfall averaged over the "southern-western slope" regions over East Asia ( Figure  S2b in Supporting Information S1). ARs produce rainfall of 22.8-27.4 mm day −1 on average, contributing to a large fraction of total rainfall in these regions ( Figure S4 in Supporting Information S1). Table 1 also summarizes AR fraction in the occurrence of extreme rainfall (Section 2.1). The relative fraction is highest in MAM (67.7%) and relatively lower in JJA and SON (41.7% and 49.2%, respectively). This seasonal variation is also found in observations and can be interpreted as a result of the dominance of Meiyu-Baiu rainband and tropical cyclones in JJA and SON .

Climatology of AR Rainfall and Its Change Under Global Warming
Before examining change in AR-related rainfall under global warming, we compare the past and future climatology of AR occurrence frequency. Black contours in Figure 2 show the seasonal-mean climatology of AR frequency over the western North Pacific. In the PAST simulations, ARs are most active during JJA (Figure 2b), particularly over the Pacific coast of Japan . No ARs are detected over western China in DJF, owing to sparce moisture in the air (Figures 3a and 3d in . As examined in , East-Asian ARs become more frequent in a warmer climate. During MAM, for example, low-level westerly is enhanced over the middle latitudes (the area between anomalous subtropical high pressure and subpolar low pressure; Figure 1c of Kamae et al., 2019) and AR frequency increases (shading in Figure 2a) over southeastern  Figure S2 in Supporting Information S1 Note. Top and bottom rows indicate climatology in AR-related rainfall (mm day −1 ) and fraction (%) of occurrence frequency of AR-related extreme rainfall (f_AR P ) to that of all extreme rainfall (f P ) in the PAST simulations for each season averaged over "southern-western slope" regions over East Asia ( Figure S2b in Supporting Information S1). Extreme rainfall events are defined as upper 0.1% cases of wet hours (rainfall >0.1 mm hr −1 ). Uncertainty ranges indicate 95% confidence intervals. China (+3-6%), Korean Peninsula (+2-3%), Japan (+2-4%) and northeastern China (+1-3%). The increase is found among all six-type global warming simulations, suggesting a significant and robust increase in occurrence of AR-related rainfall over East Asia.
In the d4PDF simulations, East-Asian extreme rainfall becomes more frequent and intense under global warming, as demonstrated in previous studies (e.g., Endo et al., 2017;Fujita et al., 2019). Seasonal-mean changes in frequency of extreme rainfall (PLUS4K minus PAST) averaged over the "southern-western slope" regions are larger during MAM and DJF than JJA and SON (Table S1 in Supporting Information S1). The changes in frequency during JJA and SON over East Asia ( Figure S5 in Supporting Information S1) are significantly influenced by tropical cyclone activity, as demonstrated in Endo et al. (2017). In MAM, the frequency of extreme rainfall triples (+206 ± 11%) with non-negligible spread among simulations with different SST warming patterns (for example, +159% in CC, +244% in MI). The spread may be attributed to the differences in the dynamic atmospheric circulation (Pfahl et al., 2017), which in turn are associated with the differences in SST warming pattern ( Figure S6 in Supporting Information S1). Kamae et al. (2019) revealed that the spread in East-Asian atmospheric circulation response to SST warming in the d4PDF simulations (0.70%-1.14% per 1 K of tropical-mean SST warming) is primarily related to the warming patterns over the Indian and Pacific Oceans.
It is expected that ARs play an important role in increasing the occurrence of extreme rainfall under global warming. Figure 3 shows the contributions of ARs to the increase in the occurrence of extreme rainfall. ARs contribute greatly to the large increase in extreme rainfall, especially in MAM and DJF. Prominent differences exist in the spatial pattern of the contributions (for example, between western and eastern Kyushu, and between south-west slope and east slope of the Japan's Alps), indicating a strong orographic effect on AR-related extreme rainfall. Table 2 summarizes the averaged contributions over the "southern-western slope" regions. ARs explain 76.6 ± 0.6% of the increase in the occurrence of extreme rainfall during MAM. The AR contribution becomes lower in JJA and SON (46.2 ± 0.9% and 55.4 ± 1.1%, respectively), largely owing to the significant contributions of Meiyu-Baiu rainband and tropical cyclones (Endo et al., 2017). The spread in AR contributions among , where f P is occurrence frequency of extreme rainfall in the PAST simulation, f P4 is that of the PLUS4K simulation (see Figure S5 in Supporting Information S1), f_AR P and f_AR P4 are those of AR-related extreme rainfall in the PAST and PLUS4K simulations, respectively. the six-type warming simulations is modest (for example, 74.2%-78.0% in MAM and 43.5%-50.4% in JJA), suggesting the robust AR contributions to the increasing occurrence of extreme rainfall.

Water Vapor Transport and Rainfall
Landfalling ARs bring large-scale water vapor flow to East Asia and cause heavy rainfall in the region. Under global warming, strengthened water vapor transport combined with orographic uplift over steep mountains should result in more intense local rainfall. In this subsection, we examine the relationship between the large-scale IVT simulated in the AGCM and the local rainfall resolved in the RCM. Figure 4 shows the joint probability distribution of IVT and rainfall over the southwestern slope of Japan's Alps ( Figure  S2b in Supporting Information S1) in MAM. We chose this region as a representative of the regions with strong AR-related orographic rainfall (Figures 1  and 3; and Figure S3 in Supporting Information S1). In the PAST simulations (Figure 4a), the largest probability is found in IVT cases weaker than 200 kg m −1 s −1 (horizontal axis) with rainfall less than 1 mm hr −1 (vertical axis), and stronger IVT cases have a larger chance of producing heavy rainfall. For example, IVT of 400 kg m −1 s −1 have a 0.01% chance of generating rainfall of around 9 mm hr −1 , whereas the chance is 0.001% for IVTs of 150 kg m −1 s −1 . This IVT-rainfall relationship is tied to the strong orographic effect over the southwestern slope of Japan's Alps . In the PAST simulations, the upper-end of IVT and rainfall are 1200 kg m −1 s −1 and 29 mm hr −1 , respectively.
Under global warming, both the IVT and rainfall are enhanced over this region. Figure 4b shows the IVT-rainfall relationship in the PLUS4K simulations. The probability distribution generally shifts toward stronger sides of IVT and rainfall compared to the PAST simulations. From this panel, we can also identify extreme IVT or rainfall in a warmer climate above the upper-end level in the PAST simulations (i.e., IVT >1200 kg m −1 s −1 and rainfall >29 mm hr −1 ; Figure 4a). If we limit to AR-related cases, these probability distributions are only found in stronger IVT bins, in line with the definition of ARs ( Figure S7 in Supporting Information S1). Over central Japan, AR fraction in each bin shown in Figure 4c clearly demonstrates a great importance of ARs in increasing extreme rainfall in a warmer climate. As summarized in Table 2 Note. The ratio (%) is represented as 100(f_AR P4 -f_AR P )/(f P4 -f P ) as shown in Figure 3. "All" indicates ensemble mean of the six simulations. Uncertainty ranges indicate 95% confidence intervals.   1600 1800 IVT kg m s -1 -1 extreme rainfall stronger than 29 mm hr −1 in the PLUS4K simulations, not occurred in the PAST simulation, is due primarily to ARs (89.3%).

Summary and Discussion
The ensemble simulations based on the AGCM and RCM indicate that more ARs pass through East Asia in a warmer climate. In the future simulations, ARs often bring more water vapor and more intense rainfall over the southern and western slopes of mountains over western Japan, central Japan, northern Japan, Taiwan, the Korean Peninsula, and northeastern mainland China, compared to the PAST simulations. The increase in the occurrence frequency of AR-related extreme rainfall peaks in MAM because of the contributions of other phenomena (including tropical cyclones) to extreme rainfall in JJA and SON. The results of this study indicate that more record-breaking intense rainfall events will occur in a warmer climate and are greatly contributed by ARs.
In this study, we focus on rainfall over East Asia, because of the limitation in computational resources in integrating both the AGCM and RCM for past and future ensemble simulations. However, the findings of this study should also apply to other mid-latitude regions. Over western North America or Europe, for example, landfalling ARs often bring heavy rainfall when they encontour the steep mountains. The AR-related orographic rainfall should be intensified in a warmer climate because of increased water vapor transport (e.g., Gao et al., 2016;Gershunov et al., 2019;Hagos et al., 2016;Huang et al., 2020;Radić et al., 2015;Warner et al., 2015). Note that the uncertainty in the future changes of AR activity and of AR contributions to extreme rainfall are not necessarily fully covered by the ensemble simulations with a single AGCM. Also, the presented results may partly depend on the algorithm of AR detection (Ryu et al., 2021;Shields et al., 2018). Further studies using multiple climate model outputs and different detection algorithms are needed to evaluate the robustness of the future changes in AR frequency, AR intensity, and AR-related extreme rainfall.