Intensification of the East Asian summer monsoon lifecycle based on observation and CMIP6

Long-term changes in the East Asian summer monsoon (EASM) lifecycle since 1979 are analyzed based on observational datasets and historical simulations of the Coupled Model Intercomparison Project Phase 6 (CMIP6). According to the observation, the active and break phases of EASM have intensified resulting in a shorter but stronger rainy season, followed by a longer dry spell. This intensification in the active-phase precipitation is accompanied by increased lower tropospheric southwesterly wind and subsequent convergence of water vapor flux. These changes are accompanied by the widely reported westward extension of the North Pacific Subtropical High, which has been associated with the warming climate. CMIP6 models generally underestimated the observed intensification of the EASM lifecycle and the monsoon precipitation. However, some of the models did simulate the intensified EASM lifecycle similar to that observed. The result highlights the reasonable performance on EASM shown in some CMIP6 models and those simulations lend support to a dynamically-driven intensification of the EASM lifecycle in the warmer climate.


Introduction
Throughout East Asia, summer precipitation variability in intraseasonal timescales has been increasing, resulting in more frequent floods and intense dry spells seemingly coexisted during the same season. Summer rainfall in northeast Asia is driven by the distinct lifecycle of the East Asian Summer Monsoon (EASM). The EASM lifecycle includes a stepwise northward and northeastward advance, derived from successive passages through (i) its rainbands (Changma, Meiyu, or Baiu), (ii) the western Pacific subtropical high, and (iii) tropical cyclone activity (Chen et al 2004, Yihui andChan 2005). As a result, the EASM features active, break, and revival phases, with intraseasonal timing of these phases dependent upon location. The intensification of this lifecycle was manifest in the 2018 Japan flood-heat wave succession event, in which more than 1000 people were killed in one month (Imada et al 2019, Wang et al 2019. The temporal and spatial variability associated with the EASM lifecycle makes trend analysis difficult and produces geographically inconsistent results. Several studies have indicated that summer precipitation over East Asia has increased (Kim et al 2005, Kornhuber et al 2019, but rainfall trends also vary by month. In Korea, Kim et al (2005) found an overall increase in summer precipitation, while In et al (2014) found that only July and August saw increased precipitation.
The impact of global warming on the EASM is an ongoing dispute. Several studies have asserted that global warming has increased EASM precipitation (Yun et al 2008, Seo et   These seemingly conflicting claims arguably result from the limitation of climate models in simulating EASM features (Liu et al 2018, Tian et al 2019. Additionally, the coexistence of various weather systems in East Asia (Utsumi et al 2017) and their nonlinear changes with global warming have further complicated the matter (Utsumi et al 2016). Meanwhile, Li et al (2010) have suggested the chance of both drought and flood over East Asia would increase together due to global warming, a proposition that may reconcile these notions.
In this paper, we analyzed the variability of the EASM lifecycle and subsequently evaluated the performance of coupled climate models in CMIP6. In section 2, we described the observational and modeling datasets as well as the analysis methods. Section 3 shows the observed intensification to the EASM lifecycle and the variability in the CMIP6 models. A conclusion is provided in section 4.

Data
Daily precipitation data is obtained from the Climate Prediction Center Global Unified Precipitation dataset provided by the NOAA Earth System Research Laboratory's Physical Sciences Division at https://www.esrl.noaa.gov/psd/. Spatial coverage is the global terrestrial area on a half degree grid. Daily outgoing longwave radiation (OLR), daily precipitation data from Asian Precipitation Highly Resolved Observational Data Integration Towards Evaluation (APHRODITE), and from GPCP Version 1.2 are used to check the robustness of the change in unified CPC (table 1). Global 6-hourly meteorological data (850mb geopotential height, u-wind, and v-wind, and column integral of horizontal water vapor flux) at the 1.25 • × 1.25 • resolution is obtained from JRA-55 (Kobayashi et al 2015, Harada et al 2016. We also used the ERA-interim dataset for comparison (Dee et al 2011).
To assess the CMIP6 model performance, we adopted a single ensemble member from the historical simulations of 32 different models retrieved from the PCMDI database; these models are summarized in table 2. Following the sampling approach of Knutti et al (2010), only one member from each model experiment was selected to avoid potential systematic bias when calculating linear trends from the multi-model ensemble mean. CMIP6 outputs are linearly interpolated to 0.5 × 0.5 • , same as the unified CPC rainfall dataset and ocean masking is applied. The analysis period is from 1979 to 2010 for the CMIP6 historical runs and from 1979 to 2017 for the reanalysis data. In case of precipitation from APH-RODITE and GPCP, the analysis period is from 1979 to 2015 and from 1997 to 2015 respectively, due to data accessibility.

Methods
This study categorizes East Asia into three subregions, including the Korean peninsula (125 • -130 • E, 33 • -38 • N), the central-eastern part of China (115 • -120 • E, 27 • -33 • N), and the south-western part of Japan (130 • -139 • E, 31 • -38 • N) as shown in figure  1 inset. The EASM lifecycle has three-phases (active, break, and revival) but the variability of revival is mixed with tropical cyclone activity. Thus, this study focuses on active and break phases only. Various meteorological variables have been used to define the onset and withdrawal of the monsoon rainband (Ellis et al 2004, Yihui and Chan 2005, Seo et al 2011, Pradhan et al 2017; however, for simplicity, we used daily climatology of precipitation to define the active and break phases. Researchers have previously defined the Changma onset as the first day of more than 3 d of at least 5 mm d −1 rainfall (Seo et al 2011), but a defined active period based on this criteria is not always aligned with the operational active period. For example, the active period in Korea under this criteria is defined from 17 June to 5 September, but operationally the Korean Meteorological Agency defined the active phase from 19 June to 24 July (Seo et al 2011) and the Japan Meteorological Agency defined the active phase in Kyushu from 5 June to 19 July. 1 Here, the active phase is defined by 3 consecutive days of rainfall greater than 6 mm d −1 and the break phase is defined as the 14 d after the active phase. Figures 1(a)-(c) display the rainfall pattern and timing of the active phase and subsequent break over Korea, centraleastern China, and southwestern Japan. The dates for active, peak and break phases are listed in table 3, which is comparable to those used in the regional operational centers. In order to examine the extreme precipitation, 'peak phase' is defined as the days in which rainfall exceeds 8 mm d −1 over 1 standard deviation in all three regions. To estimate the dry spell associated with the break phase, we have introduced the Dry Severity Index (DSI), which is defined as the product of the dry spell length and precipitation deficit. Dry spell length (DSL) reflects the duration of the dry event and is the length of the successive dry days (P = 0). Precipitation deficit (PD) reflects the magnitude of a dry event (Tallaksen et al 1997) and is defined as three times the long-term mean precipitation for break periods over time (P) minus annual precipitation (P(t)) divided by mean precipitation. (Multiplying the long-term mean precipitation by three ensures PD is always positive and thus simplifies the interpretation.) In the atmosphere, the amount of water vapor (E-P) is the sum of the water vapor mass change depending on time and water vapor flux transfer (Barnes 1965) (equation (3)). The sum of the water vapor changes ( ∂w ∂t ) is small and negligible. At the surface, the convergence of water vapor flux (-∇ · Q) infers the amount of moisture pooling (equation (4)).

Observational analysis
Linear trends of rainfall in the active and break phases were examined in figure 2 for the three target regions. During the active phase, positive precipitation trends prevail in most areas (figure 2(a)). Active phase precipitation over East Asia has increased (figure 2(c)) and the trend is more pronounced in the peak rainfall of the active phase (figure 2(e)). In Japan, a strong positive trend in the southwest (Kyushu) dominates weak negative trends in the Osaka and Nagoya regions. For both active and peak phases, the precipitation trend over Japan is consistently significant (p<0.05) but in Korea, the increasing trend is only significant in the peak in active phase (p<0.1). This contrasts with precipitation trend over China, which is only significant in the active phase (p<0.1) but not in its peak phase. Nonetheless, the active phase precipitation has increased and this observation suggests enhanced EASM rainfall in active over time. This result is in agreement with a recent study on Taiwan's EASM active rainfall (Tung et al 2020).
During the monsoon break phase, figure 2(b) shows a negative trend in precipitation in Korea and central-eastern China, while a weak but positive trend is exhibited in southwest Japan. In terms of the break phases over each region, the linear trend is weak and statistically insignificant (figure 2(d)). To examine the reduced precipitation amount during the break phase, we calculated the dry severity index (DSI) described above to examine the intensity and duration of precipitation reduction together (figure S1 (available online at stacks.iop.org/ERL/15/0940b9/mmedia)). Although the trend is not statistically significant, dry spells have become drier and have lasted longer, with the exception of central-eastern China. The severity of dry periods in Korea and southwest Japan has increased. Combined, these findings indicate that the overall EASM lifecycle has enhanced, leading to a higher chance of both wet and dry extremes within a single summer season. This is consistent with the results of Singh et al (2014) and Li et al (2010) for the long-term trends, as well as the consecutive floodheat wave events over Japan during the summer of 2018 (Wang et al 2019).  Temporal and latitudinal evolution of rainfall is investigated using a latitude-time Hovmöller diagram ( figure 3(a)). Climatologically, the rainband propagates northward and passes East Asia from early June to mid-July. As the rainband passes East Asia, it causes the Meiyu-Baiu-Changma season from central eastern China to southwest Japan to Korea, respectively, with increased precipitation observed along this migrating rainband in the Meiyu stage (~25˚N). A slight drying trend is observed in the periods before and after this rainband passage, suggesting the EASM rainband has intensified and this intensification is associated with adjunct drying. There is dry signal over East Asia from mid-July, proposing enhanced dry signal in break phase. The result is robust regardless of the other datasets used (see figure S2 for the validation analysis). ∆OLR, which is defined as 235-OLR, values larger than zero is used in this analysis as an indicator of deep convection. The trend of precipitation from GPCP is not calculated because the analysis period is too short. APHRODITE, GPCP, and OLR all show consistent result with that from the unified CPC precipitation. Since the characteristics of each precipitation data are unique, their consistent trends lend support to the robustness of the life-cycle change of the EASM precipitation.
The rainfall migration in summer over East Asia is closely related with the low-level southwesterlies (Seo et al 2013), so we defined the temporal evolution of the 850 hPa south westerlies as SW = u * sin (45 • ) + v * cos (45 • ) and the convergence of water vapor. As shown in the Hovmöller diagram (figures 3(b)), the rainband's evolution coincides with that of the southwesterly wind and their trends are in phase, suggesting that the increased precipitation along the EASM rainband may be dynamically driven. Next, we plotted the migration of convergence of water vapor in figure 3(c) and found that the migrating pattern of − ∇ · Q agrees well with that of the precipitation and southwesterly wind, both in the climatology and the respective trends. Noteworthy is the similar but opposite trends in − ∇ · Q suggesting that the drying trends adjacent to the EASM rainband are accompanied by increased divergence of water vapor flux. The climatology and trend of the south westerlies and − ∇ · Q were crosschecked with the ERA-interim dataset (figure S3), and the result is robust regardless of which dataset used. The cause of these systematic changes may be related to the documented westward expansion of the Western North Pacific Subtropical High (WNPSH) in the warmer world. As shown in figure S4, the WNPSH ] from JRA-55 and averaged from 115E to 139E covering target area. Contour is climatology and distinct value is plotted. (a. 6~10mm day -1 ; b. 3~5m s -1 ; c. -8~-4 *10 -5 kg m -1 s -1 ). The trend is shaded ranging (a) from -0.2mm day -1 yr -1 to 0.2mm day -1 yr -1 , (b) from -0.15m s -1 yr -1 to 0.15m s -1 yr -1 and (c) -0.3*10 -5 kg m -1 s -1 yr -1 to 0.3*10 -5 kg m -1 s -1 yr -1 , respectively. Dotted area is significant at 95% confidence level. has expanded westward in recent decades causing the high pressure to arrive earlier and reach farther in East Asia, hence increasing the south-westerlies. This finding echoes the observation by Woo et al (2017) that the Changma rainfall has increased due to the expansion of WNPSH modulating the jet-stream over Korea and inducing anomalous anti-cyclones south of Korea. Intensified southerly and southwesterly winds also cause moist convection to increase, eventually enhancing the EASM precipitation (Seo et al 2013).

Simulation by CMIP6 models
Next, we examined the historical simulations of CMIP6 for evaluation. The similarity of the observations and the simulated climatological features of EASM lifecycle is quantitatively evaluated using the pattern correlation that is illustrated on a Hovmöller diagram of precipitation (here, x-and y-axis are for time and latitude, respectively) of precipitation. For active and break phases over East Asia (latitude of 27˚N-38˚N, and time period from 7 June to 2 August in figure 4), the pattern correlation between observations and CMIP6 models was calculated to examine how the models catch the migrating pattern. However, it does not consider the intensity of the precipitation, so we included that averaged precipitation amount as an additional measure. CMIP6 models with R 2 larger than 0.5 and the total averaged precipitation for active and break phases over East Asia within ± 20% of observations (7.54) are categorized as 'group A' and the others as 'group B' . In the end, 8 models and 24 models are in group A and B, respectively (table 4). Interestingly, strong correlation between climatological rainfall and its long-term change measured by pattern correlation (figure S5) strongly infers importance of better climatological rainfall in the EASM. Table 4. Pattern correlation coefficient in CMIP6 model. Correlation coefficient larger than 0.70 (r 2 > 0.5) and total precipitation for active and break phase over East Asia is within ± 20% of observation (6.03-9.05) is classified as group A, and the others are in group B. The climatology of the spatiotemporal evolution of precipitation (solid line) is reasonably simulated (figure 4), though simulated precipitation in models both in group A and B tends to be underestimated. Models in group A depict the climatological rainfall onset and its migration to the north well (figure 4(a)). However, the intensity of precipitation is weak. There is a slightly positive trend at the beginning of the rainband, but a negative trend exists throughout the monsoon period, implying underestimation of the EASM precipitation. Precipitation time-series for the overlapping active, break and peak phases in three regions are shown in figure S6. Observations show a distinct increasing trend in the active and peak phases and decreasing trend in break phase, supporting the observed intensification of the EASM lifecycle. Precipitation of group A models shows a distinct decreasing trend in the break phase and slightly increasing trend in active and peak phases. Despite the underestimated precipitation in group A models, northward propagation of the rainband, intraseasonal characteristic, is realistically simulated. Furthermore, the overall pattern of wetting trend (drying trend) in the earlier (later) part of the rainy season is well simulated. All models in group A do simulate climatological propagation of the rainband but the timing and the intensity of the monsoon onset are different (figure S7). Nevertheless, some models in group A (CESM2, CESM2-WACCM, CESM2-WACCM-FV2) reproduce the intensification in both the active and break phases (figure S9). When it comes to models in group B, the rainfall peak in summer is captured but the amount of precipitation is underestimated and they mostly fail to simulate the migrating pattern of EASM ( figure 4(b)). Nevertheless, some models depict the climatological propagation of the rainband, but the rainband reaches to the East Asia lately or its intensity is weak (CNRM-CM6-1, CNRM-CM6-1-HR, CNRM-ESM2-1, EC-Earth3, EC-Earth3-Veg, INM-CM4-8, IPSL-CM6A-LR, MIROC-ES2L, MIROC6, NorCPM1) (figure S8). Also, most models fail to simulate the precipitation trend in either phase except for EC-Earth3, MIROC6, CNRM-CM6-1 (figure S8). These models catch the wetting trend (drying trend) in the earlier (later) part of the rainy season, but the timing of the rainy season is wrong. The climatological propagation of the rainband simulated by EC-Earth3 and MIROC6 is too fast and the rainband reproduced by CNRM-CM6-1 is stagnated at the lower latitude. Time-series of precipitation for overlapped active, break, and peak of MIROC6 phases show similar trend with that of observation (figure S10).
According to previous studies of Park et al (2020) and Li et al (2020) using CMIP5 models, it was found that certain features of the EASM were simulated but its more detailed characteristics were not properly captured, such as the second peak and northward propagation of precipitation. On the other hand, some of CMIP6 models, those in group A simulate the EASM lifecycles and its long-term change, which is a positive sign.

Concluding remarks
In this paper, the long-term change in the EASM lifecycle was analyzed based on observational datasets, and the capacity of CMIP6 models to reproduce the mechanisms of the long-term trends. It was found that precipitation in the active phase has increased and the increasing trend is more significant in its peak, which means that the monsoon peak or extremes is getting stronger. This precipitation trend coincides with the increasing trends of southwesterly wind and the convergence of water vapor flux accompanying the rainband. The WNPSH and its westward extension help intensify the southwesterly wind. It was also found that the break phase tends to be enhanced after the rainband passage, and this tendency is further illustrated by the DSI index, showing the combined intensity and duration of the dry spell. Thus, it can be concluded that the EASM lifecycle has intensified, making it more likely for southwest Japan and Korea to experience heavy rainfall followed by enhanced drying. This result is in agreement with previous studies (Kim et al 2005, Seo et al 2013 in terms of a generally intensified EASM system, but adds further details in that both the rainy season and dry spell within the EASM lifecycle have intensified. Apparently, the observed EASM variability is more complex than just the summer mean value.
The performance of CMIP6 models in simulating the EASM climatological evolution and its variability is still limited, but some models are showing sign of an improved performance. In general, the intensity of monsoon precipitation tends to be underestimated, but group A models realistically simulate the latitude-time propagation of the rainband and the general pattern of the long-term intensification of EASM. Not only that, group A models also depict the precipitation time-series at the active and break phase that is consistent with the observational dataset. While models in group B did not accurately simulate the progression of the summer rainband and they underestimated the EASM rainfall, they still depicted the rainfall peak in summer. This evaluation highlights some of CMIP6 models that exhibit a marked performance in simulating intraseasonal characteristic of the EASM as well as its long-term change.