Corrigendum: Ocean warming pattern effects on future changes in East Asian atmospheric rivers (2019 Environ. Res. Lett. 7 054019)

Atmospheric rivers ( ARs ) , intense water vapor transports associated with extra-tropical cyclones, frequently bring heavy rainfalls over mid-latitudes. Over East Asia, landfalling ARs result in major socio-economic impacts including widespread ﬂ oods and landslides; for example, western Japan heavy rainfall in July 2018 killed more than 200 people. Using results of high-resolution atmospheric model ensemble simulations, we examine projected future change in summertime AR frequency over East Asia. Different sea surface temperature ( SST ) warming patterns derived from six atmosphere – ocean coupled model simulations were assumed to represent uncertainty in future SST projections. The rate of increase in the frequency of landfalling ARs over summertime East Asia is on average 0.9% K – 1 and is dependent on SST warming patterns. Stronger warming over the North Indian Ocean and South China Sea or weaker warming over the tropical central Paci ﬁ c produce more frequent landfalling ARs over East Asia. These patterns are similar to the co-variability of SST, atmospheric circulation, and ARs over the western North Paci ﬁ c found on the interannual time scale. The results of this study suggest that the natural disaster risk related to landfalling ARs should increase over East Asia under global warming and SSTs over the Indo-Paci ﬁ c region holds the key for a

In Kamae et al (2019), we made mistakes explaining the method for detecting atmospheric rivers (ARs) in the global warming simulations. The anomalies in vertically integrated water vapor transport (IVT) used in the AR detection were calculated by subtracting IVT climatology in the global warming simulations, not IVT climatology in the current climate simulation.
Section 2.2 'Future projection' 1st paragraph, the last sentence should read 'The AR detection scheme described in section 2.1 and used in Kamae et al (2017b) was applied to the outputs of global warming simulation. Note that the IVT anomaly used in the AR detection was calculated by subtracting IVT climatology in the global warming simulations.' . Section 3.1 'AR response to climate warming' 1st paragraph, the last two sentences 'The general increasing trend of ARs under climate warming can be attributed to thermodynamic effect (e.g. , Gao et al 2015. Atmospheric water vapor increase and resultant positive ∆IVT should result in more AR count because the anomalous IVT threshold (section 2.1) is identical between current climate and global warming simulations.' were incorrect and should be skipped.
Section 4 'Summary and discussion' 1st paragraph, the second and third sentences should read 'Generally, ARs become more frequent over the Northern Hemisphere including summertime East Asia increasing trend, dynamic effect is important for the spreads of East-Asian ∆ARs among different SST warming experiments via changing atmospheric circulation patterns over the Indo-Pacific region.' . The corrections do not affect the major conclusions of the paper.
In Kamae et al (2019), we made mistakes explaining the method for detecting atmospheric rivers (ARs) in the global warming simulations. The anomalies in vertically integrated water vapor transport (IVT) used in the AR detection were calculated by subtracting IVT climatology in the global warming simulations, not IVT climatology in the current climate simulation.
Section 2.2 'Future projection' 1st paragraph, the last sentence should read 'The AR detection scheme described in section 2.1 and used in Kamae et al (2017b) was applied to the outputs of global warming simulation. Note that the IVT anomaly used in the AR detection was calculated by subtracting IVT climatology in the global warming simulations.' . Section 3.1 'AR response to climate warming' 1st paragraph, the last two sentences 'The general increasing trend of ARs under climate warming can be attributed to thermodynamic effect (e.g. , Gao et al 2015. Atmospheric water vapor increase and resultant positive ∆IVT should result in more AR count because the anomalous IVT threshold (section 2.1) is identical between current climate and global warming simulations.' were incorrect and should be skipped.
Section 4 'Summary and discussion' 1st paragraph, the second and third sentences should read 'Generally, ARs become more frequent over the Northern Hemisphere including summertime East Asia increasing trend, dynamic effect is important for the spreads of East-Asian ∆ARs among different SST warming experiments via changing atmospheric circulation patterns over the Indo-Pacific region.' . The corrections do not affect the major conclusions of the paper.

Introduction
Atmospheric rivers (ARs), narrow corridors of enhanced moisture transport extending for thousands of kilometers, often cause natural disasters over the middle latitudes including East Asia (e.g. , Gimeno et al 2016, Kamae et al 2017c. For example, an AR-like enhanced water vapor transport was closely related to occurrence of the heavy rain event of July 2018 in Japan that caused 221 fatalities due to widespread floods and landslides (Tsuguti et al 2019). Many previous studies have been devoted to investigating future change in AR activity because it is the key for projecting water resources and natural disaster risk. Increased water vapor transports associated with ARs are robustly projected , Payne and Magnusdottir 2015, Gao et al 2015, Warner et al 2015, Espinoza et al 2018 because of atmospheric water vapor increase in response to global warming (O'Gorman et al 2012 and references therein). In addition to this thermodynamic effect, future change in atmospheric circulation (dynamic effect; e.g. Ma and Xie 2013, Shepherd 2014) is important for projecting regional changes in AR activity (Payne and Magnusdottir 2015, Gao et al 2015, Shields and Kiehl 2016, Espinoza et al 2018. Previous works examined projected future changes in AR activity based on multi-model outputs from the Coupled Model Intercomparison Project phase 5 (CMIP5; Taylor et al 2012). Regional projections of AR frequency in a warming world exhibit large spreads among models (Payne and Magnusdottir 2015, Espinoza et al 2018. Some of the models have limited skill in representing current AR Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. behavior and its statistics (frequency, intensity, and variability), limiting reliability of AR projections using such models (Payne andMagnusdottir 2015, Guan and. In general, higher resolution is necessary to represent more realistic structure (and statistics) of ARs (models with horizontal resolution of 0.7°-1.3°have generally higher skills than models with 1.9°-3.7°; figure  . However, it is not easy to evaluate the effect of SST warming pattern uncertainty, because other factors (e.g. model resolution, physics, and biases) also affect the results obtained from the CMIP multiple model ensemble.
In this study, we employ ensemble simulations by a high-resolution atmospheric general circulation model (AGCM) to evaluate the dynamic component of uncertainties in projected AR changes, which is primarily related to the effect of SST pattern. The model shows skills in reproducing climatology and variability of AR activity over the northwestern Pacific, and was forced by different SST warming patterns. agents (greenhouse gases, aerosols, and ozone). The ensemble simulations were developed with random initial and SST perturbations (Hirahara et al 2014). In this study, a 10-member ensemble mean was examined to reduce the effect of atmospheric internal variability (Kamae et al 2017a(Kamae et al , 2017b.

Data and method
We detect ARs using a methodology modified from Mundhenk et al (2016). First, vertically integrated water vapor transport (IVT) is calculated using 6-hr atmospheric variables. IVT anomaly is calculated by comparing with its climatology. Next, shape (e.g. length>1500 km, area>7.8×10 5 km 2 , lengthwidth ratio>1.325) and intensity (140 kg m -1 s -1 ) of anomalous IVT are checked to detect ARs. The detection scheme can filter out circular IVT anomalies associated with tropical cyclones via the shape criterion (see figure A2 of Mundhenk et al 2016). Note that ARlike IVT anomalies examined in this study are sometimes influenced by remote effects of tropical cyclone (Kamae et al 2017c). AR frequency examined in this study is defined as the fraction of time periods during which an AR exists to the total number of periods. Kamae et al (2017b) reported that this model can realistically simulate the climatology, seasonal cycle, and interannual variability of North-Pacific AR compared to an atmospheric reanalysis.

Future projection
In addition to the current climate simulation, outputs from global warming simulations (Mizuta et al 2017) were also examined to evaluate future change in AR activity. In this set of simulations, the model was forced by constant radiative forcing, SST warming patterns, and anomalous sea ice concentration. Anomalous radiative forcing corresponding to the level of year 2090 in the Representative Concentration Pathway (RCP) 8.5 scenario adopted in CMIP5 was added to the radiative forcing in the current climate simulation. Six SST warming patterns derived from cluster analyses of CMIP5 multi-model results (figure S1 is available online: stacks.iop.org/ERL/14/054019/ mmedia; Mizuta et al 2014) were added to historical SST. The six SST warming patterns were based on RCP8.5 simulations by CCSM4 (hereafter CC), GFDL-CM3 (GF), HadGEM2-AO (HA), MIROC5 (MI), MPI-ESM-MR (MP), and MRI-CGCM3 (MR). Each SST warming pattern was scaled so that the simulated global-mean surface air warming from the pre-industrial level is equal to 4 K (Mizuta et al 2017). The six patterns exhibit different zonal and meridional gradients over the Pacific, Indian and Atlantic Oceans (table S1, figure S1). For example, CC run was forced by La Niña-like SST pattern in the tropical Pacific (larger warming over the western equatorial Pacific than central-to-eastern equatorial Pacific), in contrast to El Niño-like warming pattern prescribed in other runs (figure S1; see section 3.3). Among the six runs, tropical Atlantic warming and meridional warming contrast between 30°-60°N and EQ-30°N in the North Pacific are strongest in MI and GF runs, respectively (table S1, figures S1, S2). In each run, ensemble was developed in a similar way to the current climate simulation (section 2.1). The AR detection scheme described in section 2.1 and used in Kamae et al (2017b) was applied to the outputs of global warming simulation.
To evaluate the forced response to radiative forcing and SST warming, 10-member ensemble mean, 60-yr climatology of AR frequency in each run is compared. In this study, projected changes between global warming and current climate simulations scaled by tropical-mean (30°S-30°N) SST anomaly are represented as Δ(K -1 ). Average of Δ among six warming simulations can be regarded as a robust response to global warming simulated in this model. Deviations of Δ in individual warming simulations from the averaged Δ can be considered as the effect of ΔSST deviation from the averaged warming pattern (figure S2).

AR response to climate warming
We first examine simulated changes in atmospheric circulation and AR frequency in each season robustly found among the six warming simulations. Figure 1 shows six-run-mean sea level pressure (SLP) response (ΔSLP), change in AR frequency, and prescribed ΔSST. Averaged ΔSST pattern is almost identical to the ensemble mean of CMIP models: stronger warming over the mid-and high-latitude North Pacific than the tropics, and stronger warming over the tropical eastern Pacific than western Pacific ARs become more frequent over East Asia (0.91±0.15 % K -1 ; table 1), western North America, eastern North America, and Europe (figures 1(b), (d), (f), (h)). During boreal summer, large increases in AR frequency over the subtropical North Pacific and East Asia (25°-40°N) are concurrent with anomalous lowlevel circulations associated with the weakened NPH and the enhanced WNPSH. Anomalous moist southwesterlies on the southeastern flank of the weakened NPH and on the northwestern flank of the enhanced WNPSH increase AR occurrences over those regions (figure 1(e); detailed in section 3.2). Such circulation-AR relationship is consistent with what found in interannual variability (Kamae et al 2017b; detailed in section 3.3). Such a dynamic effect of atmospheric circulation on regional changes in AR frequency (e.g. Payne and Magnusdottir 2015, Shields and Kiehl 2016; see section 1) suggests the importance of ΔSST pattern in future projections of North-Pacific AR activity. In this study, we mainly focus on North-Pacific ARs during boreal summer. AR responses during other seasons and over other regions will be examined in future studies.

Uncertainty related to tropical SST warming pattern
Next, we examine sensitivity to ΔSST pattern. Figure 2 shows changes in boreal summer AR frequency over East Asia and the North Pacific simulated in the six runs. All the simulations produce a general increase in North-Pacific AR frequency, but the circulation and AR responses exhibit large spreads among the six runs: limited enhancement of WNPSH (limited weakening of NPH) in CC, GF, MP and MR runs (CC run) compared to other runs (table 1). The differences in the quantitative changes in East-Asian ARs (table 1) are partly related to the spreads of the WNPSH response. For example, MI run shows the largest enhancement of WNPSH among the six runs and correspondingly the largest increase in East-Asian AR occurrence (table 1) especially over southeastern China (figure 2(d)), while the changes are relatively weak in GF and MP runs (table 1, figures 2(b), (e), S3). These results suggest that a significant portion of the uncertainty in ΔARs is due to the SST-mediated dynamic effect (see section 3.3).
However, the WNPSH response cannot explain all the spreads of East Asian ΔARs. For example, CC run (figure 2(a)) produces larger East-Asian ΔARs than GF, HA, MP and MR runs (table 1) but not a larger enhancement of WNPSH. Another feature found in CC run is a limited weakening of NPH and associated North-Pacific circulation response (figure S3). Here circulation anomaly associated with the NPH response is important for ΔAR pattern over the North Pacific. GF run shows the largest weakening of NPH, causing a southeastward shift of seasonal-mean AR frequency over the North Pacific: a moderate increase over the middle latitudes on the northwestern flank of the anomalous NPH with a greater increase over the subtropics on the southeastern flank ( figure 2(b)). Such a meridional shift is not apparent in CC run (figure 2(a)), suggesting that the NPH response is also an important factor for the spatial pattern of ΔARs over the North Pacific.
The spreads in circulation and AR responses result from differences in ΔSST prescribed in these runs. Here SST-circulation-AR relationship found in the global warming simulations appears consistent with that observed in interannual variability. In the next subsection, we examine the relationship found in interannual variability and compare it with the results shown above. Figure 3 shows major modes of interannual variability of East-Asian summertime ARs in the current climate simulation for 1951-2010 based on empirical orthogonal function (EOF) analysis of 10-member ensemble Table 1. Simulated changes in sea level pressure (ΔSLP; hPa K -1 ) averaged over the North Pacific High (NPH) region (25°-45°N, 180°-130°W) and the western North Pacific subtropical high (WNPSH) region (10°-25°N, 110°-130°E) and frequency of atmospheric rivers (ΔARs; % K -1 ) averaged over the East Asian region (20°-45°N, 100°-150°E; figure 2) under global warming in each experiment. Uncertainty range represents ±1 standard deviation among 10-member ensemble simulations. Bottom low indicates ensemble mean and ±1 standard deviation of the six simulations (60 members).

Experiments
NPH ΔSLP (hPa K -1 ) WNPSH ΔSLP (hPa K -1 ) East Asian ΔARs (% K -1 )  In contrast to the tripolar pattern, the EOF2 is characterized by a dipole pattern of AR frequency over the North Pacific: an increase over the extratropics and a decrease over the subtropics ( figure 3(b)). Atmospheric circulation associated with EOF2 exhibits an anticyclonic anomaly over the North Pacific, indicating an enhanced NPH. The circulation-AR relationship, an increase (decrease) in ARs on the northwestern (southeastern) flank of the anomalous anticyclone, is similar to that in EOF1 although positions and horizontal extent are different. The anomalous circulation ( figure 3(b)) facilitates more frequent landfalling ARs over East Asia (especially northeastern China, Korean Peninsula, and northern Japan) and suppresses ARs off the Pacific coasts of Japan. Figure 3(d) shows the anomalous SST pattern corresponding to EOF2. Negative SST anomalies are found over the central and eastern tropical Pacific, indicating a summertime development of La Niña. The relationship between summertime SST anomaly and the NPH is consistent with previous studies (e.g. Wang et al 2013, Paek et al 2018. Details of La Niña-induced anomalies in atmospheric circulation and ARs over the North Pacific are currently being extensively examined and will be reported in a separated paper.

CC
The SST-circulation-AR relationship found here is helpful for detecting the SST pattern effect on North-Pacific ΔARs under global warming. The SST forcing prescribed in MI run exhibits larger warming over the NIO and the South China Sea and weaker warming over the central equatorial Pacific than the six-run mean (table S1, figures 4, S1, S2). The larger enhancement of WNPSH under global warming in MI run is consistent with CMIP5 inter-model spreads (He and Zhou 2015) and the SST-circulation relationship found in interannual variability (figure 3). CC run was forced by La Niña-like ΔSST pattern (zonal ΔSST gradient between the western equatorial Pacific and central equatorial Pacific; table S1, figures S1, S2), in contrast to El Niño-like Pacific ΔSST found in the The ΔSST in CC run is consistent with the weak NPH response compared to other runs (table 1, figures 2(a), S3(a)). These results suggest that the anomalous NIO and South China Sea warming and La Niña-like SST gradient are important factors for the inter-model spreads in ΔARs over East Asia and the North Pacific. Note that ΔSST over other regions including the Atlantic Ocean (table S1, figures S1, 2) may also play some roles (e.g. Chen et al 2014, Hong et al 2014, which is not discussed here since our focus in this study is on the tropical Indian and Pacific Ocean. Figure 4(a) shows the interannual relationship among NIO SST (averaged over 50°-125°E, 5°-25°N), zonal SST gradient between the tropical western Pacific (TWP) and tropical central Pacific (TCP; 100°-150°E, 5 S°-5°N minus 150E°-140°W, 5 S°-5°N), and East-Asian summertime AR frequency for 1951-2010. Removing linear trends for 1951-2010 from these variables does not change the results significantly. Here the two SST indices are not significantly correlated (r=0.07). If we use these two as explanatory variables, East Asian ARs (%) can be approximated using a multiple linear regression (R 2 =0.57) as follows: where X 1 is NIO SST (°C) and X 2 is TWP SST minus TCP SST (K). Years with higher NIO SST tend to produce more frequent East Asian ARs as reported in Kamae et al (2017b). In addition, warmer TWP than TCP (La Niña-like SST gradient) is associated with increased East Asian ARs. Similarly, SSTs in these two regions are important for the spreads of East-Asian ΔARs among the six warming simulations. Figure 4(b) shows relationship between NIO ΔSST, TWP ΔSST minus TCP ΔSST, and East Asian ΔARs in these runs. As a reference, we also plot ΔSST simulated in CMIP5 models (Taylor et al 2012;  where X 1 is NIO ΔSST (K) and X 2 is TWP ΔSST minus TCP ΔSST (K). R 2 of equation (2) is 0.66. Among the d4PDF runs, the larger increases in SST over the NIO and the South China Sea in MI and MR runs (table S1, figure 4(b), S2) are associated with larger increases in East-Asian ΔARs (table 1, figures 2, 4(b)). In addition, the La Niña-like ΔSST gradients in CC and MI runs (figure 4(b), S2) are also related to greater increases in East-Asian ΔARs (table 1, figures 2, 4(b)). Based on these results, we suggest that the two ΔSST indices are important sources of uncertainty in East-Asian ΔARs. Note that the regression coefficients obtained in equation (2) based on the limited samples size (n=6) may be sensitive to increasing sample size. Relative importance of NIO SST (X 1 ) compared to the zonal SST gradient (X 2 ) under global warming should be further tested using sensitivity experiments prescribed with idealized SST warming patterns.

Summary and discussion
Future changes in AR activity over East Asia and the North Pacific and their uncertainty related to SST warming pattern are investigated using large ensemble simulations by a high-resolution AGCM. Generally, ARs become more frequent over the Northern Hemisphere including summertime East Asia due to increased water vapor in a warming climate. In addition to the robust thermodynamic effect, dynamic effect is important for the spreads of East-Asian ΔARs among different SST warming experiments via changing atmospheric circulation patterns over the Indo-Pacific region. A warmer NIO-South China Sea results in an enhanced WNPSH, and the strengthened moist southwesterlies increase AR occurrence on the northwestern flank. In addition, La Niña-like tropical Pacific SST gradient is important for the NPH response and a resultant increase in East-Asian ARs. These SSTcirculation-AR relationships are similar to those found in interannual variability. Our results suggest that the tropical ΔSST patterns are important sources of uncertainty in future projections of natural disaster risk via ARs over summertime East Asia.
The investigation based on a single climate model enables us to examine effects of imposed forcing (e.g. ΔSST pattern) on ARs. However, the uncertainty range is not necessarily covered by the single-AGCM experiments even though imposed ΔSST sufficiently covers the spreads of ΔSST projected by different models. In addition, the quantitative results obtained in this study are possibly dependent on choices of thresholds employed in the AR detection (e.g. IVT threshold; Waliser 2015, Shields et al 2018). Recently, intercomparison projects on high-resolution climate models (Haarsma et al 2018) and on AR detection methods (Shields et al 2018) are proposed for comprehensive evaluation of uncertainties rooted from climate models and AR detection methods. Further research is needed to evaluate robustness of the results obtained in this study. This study only focuses on summertime AR activity over East Asia and the North Pacific. Robust and uncertain parts of future projections of AR activity during other seasons and over other regions are also worth examining in future studies.