Tropical modulation of East Asia air pollution

Understanding air pollution in East Asia is of great importance given its high population density and serious air pollution problems during winter. Here, we show that the day-to-day variability of East Asia air pollution, during the recent 21-year winters, is remotely influenced by the Madden–Julian Oscillation (MJO), a dominant mode of subseasonal variability in the tropics. In particular, the concentration of particulate matter with aerodynamic diameter less than 10 micron (PM10) becomes significantly high when the tropical convections are suppressed over the Indian Ocean (MJO phase 5–6), and becomes significantly low when those convections are enhanced (MJO phase 1–2). The station-averaged PM10 difference between these two MJO phases reaches up to 15% of daily PM10 variability, indicating that MJO is partly responsible for wintertime PM10 variability in East Asia. This finding helps to better understanding the wintertime PM10 variability in East Asia and monitoring high PM10 days.

Severe air pollution events, with high PM 10 concentration, have frequently occurred in East Asia over the past decades, despite the government's air quality regulation [1][2][3] . Such events, which are most frequent in cold season 4 , have often been explained by local emissions and meteorological conditions that control aerosol deposition, accumulation, and transport 5,6 . Although minor, they are also influenced by climate variability on a wide range of time scales. On the interannual time scale, for instance, wintertime PM 10 concentration in East Asia is weakly sensitive to the El Niño-Southern Oscillation (ENSO) 7 and Arctic sea ice variability 8,9 . The relationship between haze days and Pacific Decadal Oscillation (PDO) 10,11 implies that East Asia PM 10 concentration may be affected by climate oscillation on the decadal time scale. East Asia PM 10 concentration also substantially varies on the daily time scale. However, its linkage to subseasonal climate variability has received little attention.
The Madden-Julian Oscillation (MJO) 12,13 is a planetary-scale convection-circulation coupled system in the tropics that typically moves eastward from the tropical Indian Ocean to the western Pacific with a period of 30 to 60 days. As the MJO develops and propagates, it influences the midlatitude circulation and weather events by exciting the Rossby wave train 14,15 . It is reported that the MJO-induced circulation changes can also modulate local air pollutants, such as aerosol and ozone [16][17][18][19][20][21][22] . For example, in Santiago, Chile, the dispersion of air pollutants is influenced by the MJO-related atmospheric circulation 20 . Another example is air pollution in South and Southeast Asia. In these regions, air quality is directly modulated by the MJO 16,21,22 . For instance, the variability of PM 10 , CO, and ozone in Malaysia is significantly affected by the MJO 22 .
However, the remote linkage between MJO and PM 10 concentration in East Asia has not been addressed yet. Since East Asia is on the main pathway of the Rossby wave train, precipitation, temperature, and atmospheric circulations are significantly modulated by the MJO [23][24][25][26] . Therefore, we hypothesize that MJO affects local air pollution over East Asia by changing meteorological conditions via tropics-extratropics teleconnections. In this study, we show that daily variability of East Asia PM 10 concentration is significantly modulated by subseasonal climate variability in the tropics. Specifically, we show that wintertime PM 10 variability in East Asia is to a large extent determined by the MJO.   Fig. 1b by computing daily standard deviation. The daily standard deviation of stationaveraged PM 10 concentration is 52 μg m −3 , which accounts for 54% of the daily climatology (~96 μg m −3 ), indicating a significant day-to-day variability of East Asia PM 10 concentration, especially at the stations with high PM 10 climatology in northern China. Approximately 30% to 50% of this daily variability is due to subseasonal variability which is quantified by a 20-96 days band-pass-filtered PM 10 anomaly (Fig. 1c).

MJO-related PM 10 change in East Asia
The PM 10 concentration in East Asia can be regulated not only by local emissions but also by atmospheric conditions [5][6][7]29 . On the subseasonal time scale, midlatitude circulations are substantially influenced by remote processes. One of the most important remote processes that affect the atmospheric circulations in East Asia is the tropical-extratropical teleconnections driven by the MJO 14 . By exciting the Rossby gyres in the subtropics and the Rossby waves that propagate into the midlatitudes 14,15 , MJO effectively modulates the precipitation and atmospheric circulations in East Asia [23][24][25][26] . When MJO convection is enhanced over the tropical Indian Ocean and suppressed over the western Pacific (MJO phase 1-2; Fig. 2c), a robust anomalous cyclonic circulation, corresponding to a negative anomaly of 500-hPa geopotential height (Z500), develops over East Asia, especially in southern China, Korea, and western Japan. The opposite is true for the mirror phase of MJO, in which convection is suppressed over the tropical Indian Ocean and enhanced over the western Pacific (MJO phase 5-6; Fig. 2d).
These MJO-related atmospheric circulation anomalies significantly modulate the wintertime PM 10 concentration in East Asia. Figure 2a, b show PM 10 anomalies during MJO phases 1-2 and 5-6 at lag 6-10 days (i.e., 6-10 days after the selected MJO phase). This time lag matches with the time scale by which the MJO-induced Rossby waves take to affect in the midlatitudes 30,31 . During MJO phase 1-2, PM 10 concentration becomes anomalously low almost everywhere except in northeastern China (Fig. 2a). The station-averaged PM 10 anomaly is approximately −4 μg m −3 , being statistically significant at the 95% confidence level. In contrast, PM 10 concentration is anomalously high during MJO phase 5-6 ( Fig. 2b) with the station-averaged PM 10 anomaly of approximately 4 μg m −3 . Their spatial distribution is almost the mirror image to MJO phase 1-2 (see Supplementary Fig. 1 for the other MJO phases). This result indicates that MJO systematically modulates PM 10 concentration in East Asia.
Here we emphasize that although the station-averaged PM 10 anomaly is relatively small compared to the daily standard deviation (i.e., ± 15%), its value at each station can be large. The PM 10 difference between MJO phases 1-2 and 5-6 and its ratio to the daily standard deviation are evaluated in Supplementary Fig. 2. It turns out that MJO impact is relatively small in the in-land stations where the daily PM 10 variability is large (e.g., Beijing-Tianjin region). However, its impact is substantially large along the coast. For instance, the MJOrelated PM 10 anomaly in South Korea reaches up to 35% of daily variability. This result suggests that MJO plays a critical role in determining daily PM 10 concentration in central and southern China as well as in South Korea.
The MJO also modulates extreme PM 10 days. Figure 3 shows the frequency distribution of the station-averaged PM 10 concentration. Unfiltered data is used here. The frequency distribution exhibits a long tail with increasing PM 10 concentration (Fig. 3a), indicating that high pollution days occur more frequently than clean air days. A similar frequency distribution, with different mean and extremes, is also found when MJO phases 1-2 and 5-6 are separately considered (Fig. 3b). The daily PM 10 concentration during MJO phase 1-2 is on average around 89 μg m −3 , which is 7% smaller than the wintertime average of about 96 μg m −3 . However, the daily PM 10 concentration during MJO phase 5-6 is frequently observed at higher values, with its mean value of about 100 μg m −3 . This is approximately 1.2 times higher than that during MJO phase 1-2. This result is consistent with Fig. 2 which is based on subseasonally-filtered PM 10 concentration. It is noticeable from Fig. 3b that PM 10 distribution is skewed toward high pollution days during MJO phase 5-6. High PM 10 days, defined as days with PM 10 concentration greater than 140 μg m −3 corresponding to the top 10 percentile, are more frequently observed during MJO phase 5-6. This result indicates that MJO teleconnections modulate not only the mean PM 10 concentration but also the frequency of high PM 10 days in East Asia.
How does the MJO modulate East Asia PM 10 ?
The MJO-related PM 10   midlatitudes 32,33 . They strengthen moisture transport and vertical motion in East Asia, leading to more frequent precipitation events (Fig. 4b). Since precipitation can remove PM 10 via wet scavenging processes 34 , it could contribute to a decrease in PM 10 concentration during MJO phase 1-2. As hypothesized, positive precipitation frequency anomalies (Fig. 4b) indeed match well with PM 10 anomalies, especially in southeastern China (Fig. 2a).
Consistent with previous studies, Z500 shows cyclonic anomaly in the mid-troposphere from eastern China to Korea during MJO phase 1-2 (Fig. 4a). The center of circulation anomaly, which is a part of the Rossby gyre, is associated with updraft that promotes the dispersion of air pollutants 35 . The bulk stability, which is defined by the potential temperature difference between 925 hPa and 1000 hPa, is indeed reduced, especially in eastern China (Fig. 4c). This unstable condition below the planetary boundary layer could reduce PM 10 concentration through strong vertical ventilation. The combined effect of more frequent precipitation events and enhanced atmospheric ventilations could lead to low PM 10 concentration in East Asia during MJO phase 1-2.
The anomalous atmospheric condition switches to opposite sign during MJO phase 5-6. Precipitation events become less frequent in East Asia (Fig. 4e), thus contributes to anomalously high PM 10 concentration (Fig. 2b). However, this may not be sufficient to produce high PM 10 concentration as a relatively dry condition itself does not guarantee more frequent high PM 10 days. Other factors such as atmospheric ventilation changes may also play a role 8,36 . The anticyclonic circulation anomaly drives anomalous downward motion that promotes the accumulation of local pollutants 6 . Along with this, the positive anomaly of bulk stability is observed in eastern China (Fig. 4f). The stable boundary layer can lead to reduced vertical ventilation, increasing PM 10 concentration 8,9 . This together suggests that relatively high PM 10 concentration in East Asia during MJO phase 5-6 is likely caused by the combined effects of less frequent precipitation events and reduced atmospheric ventilations.
The atmospheric condition changes in response to the MJO are further compared with the daily standard deviation in Supplementary  Figs. 5d-f. The MJO-related precipitation change is over 30% of day-today precipitation variability in southern China and over 10% in eastern c, d Same as (a, b) but for anomalies of 500-hPa geopotential height (Z500) (shading) averaged over lag 6-10 days and tropical outgoing longwave radiation (OLR) (purple/green contours represent positive/negative values with 5 W m −2 interval) at lag 0 day from 20°S to 60°N. Values that are statistically significant at the 95% confidence level are dotted. Gray boxes are the analysis domains shown as (a, b).
China and South Korea. The bulk stability change also accounts for over 20% of daily variability in eastern China. It is noteworthy that a few stations in northeastern China show the MJO-related PM 10 anomalies whose sign is opposite to most other stations (Fig. 2). Although the reason is unclear, it is again consistent with the opposite-signed precipitation and near-surface stability responses to the MJO in the region (Fig. 4).
The temporal evolution of the MJO convection and its relationship with the station-averaged PM 10 concentration in East Asia are further demonstrated by considering eight MJO phases in Fig. 5. As shown in Fig. 2, tropical convection is enhanced in the equatorial western Indian Ocean (60-80°E, 10°S-10°N) during MJO phase 1-2 (blue shading at lag 0 day) but suppressed during MJO phase 5-6 (red shading at lag 0 day) (Fig. 5a). These anomalies switch the sign within 15-30 days as the enhanced and suppressed convection couplets move eastward in time across the Maritime Continent. This makes a systematic OLR evolution as shown in Fig. 5a. Figure 5b shows the temporal evolution of PM 10 anomaly in East Asia. As anticipated from Fig. 2, it well follows the evolution of the MJO convection (Fig. 5a). Most importantly, the maximum PM 10 anomaly appears 8-10 days after the maximum OLR anomaly. A large negative PM 10 anomaly at lag 8 day from the MJO phase 1 is connected to the enhanced convection over the western Indian Ocean. Likewise, a large positive PM 10 anomaly at lag 10 day from the MJO phase 5 is linked to the suppressed convection over the western Indian Ocean. Since the time interval from one MJO phase to another is about 5-8 days, these PM 10 anomalies partly include the responses to MJO phases 2 and 6 at a shorter time lag.
The time-lagged remote connections can be again explained by the precipitation and bulk stability changes induced by MJO teleconnections (Fig. 5c, d). Here the grid point closest to each station is used to compute the station-averaged value. Multiple counts are not allowed. The atmospheric circulation anomalies resemble the PM 10 anomalies. The reduced PM 10 concentration, followed by MJO phase 1-2, is closely related with an enhanced precipitation frequency and a reduced bulk stability. In contrast, the increased PM 10 concentration after MJO phase 5-6 is related with a reduced precipitation frequency and an increased bulk stability. The combined effect of precipitation frequency and bulk stability changes in response to the MJO well captures the temporal evolution of East Asia PM 10 anomalies (Fig. 5e).

Discussion
This study reveals that the wintertime PM 10 concentration in East Asia is significantly modulated by the MJO. Specifically, MJO phase 1-2 tends to lead to reduced PM 10 concentration, presumably due to more frequent precipitation events and strong atmospheric ventilation in East Asia. In contrast, during MJO phase 5-6, PM 10 concentration becomes anomalously high possibly due to less frequent precipitation events and weak atmospheric ventilation. This finding is all based on the time average from lag 6-10 days (Figs. 2-4), the former lagging the latter. It is anticipated that when MJO phase 5-6 is detected in the tropics, the PM 10 in East Asia increases with a high chance of heavy pollution days one to two weeks later. Although such a simple relationship does not guarantee the quantitative prediction, it is helpful to better understand the daily variability of East Asia PM 10 concentration and monitor high PM 10 days.
Given the high population density and serious air pollution problems in East Asia, any added competency in understanding PM 10 characteristics is of great significance. Nevertheless, the results presented here need to be interpreted with caution because of several limitations. First, PM 10 data in China are partly derived from the air pollution index 37 . The data conversion may have introduced small but non-negligible biases. Second, PM 10 data are available only for 21 years from 2001 to 2021. This is a rather short-term record, compared to long-term meteorological observations commonly used in the MJO teleconnection studies. To better quantify the MJO-PM 10 relationship, longer data records are required. Third, the effects of local emissions and regional transports of PM 10 are not fully eliminated in this study. To isolate the MJO-teleconnection impacts, it may be useful to employ numerical modeling in which the local processes can be controlled.

Observations and Reanalysis Data
Daily PM 10 concentrations are obtained from air quality monitoring stations in Korea and China for 21 years (2001-2021). The Korean PM 10 data are directly obtained from the Ministry of Environment, Republic of Korea (https://www.airkorea.or.kr), while the long-term record of daily PM 10 concentration in China is obtained from the two different sources. Until December 2012, PM 10 concentration is estimated by converting the air pollution index (API) 37 . The unitless API is determined by the maximum pollutant concentration among the three pollutants, i.e., PM 10 , SO 2 , and NO 2 . When the API is 50 or higher, only the PM 10 -based API is used. This accounts for more than 85% of the total analysis period. However, the API smaller than 50 does not report the principal pollutant. In such a case, the API is simply assumed to be derived from PM 10 . This introduces the uncertainty in low PM 10 concentrations. Despite this uncertainty, the PM 10 concentration derived from the API is closely related with the observed PM 10 concentration with a correlation coefficient of up to 0.9 37,38 .
In March 2012, Ministry of Environmental Protection in China announced an official revision from the API to the air quality index (AQI). This index is based on the six pollutants, i.e., PM 10 , PM 2.5 , SO 2 , NO 2 , O 3 , and CO [39][40][41] . The Air Quality Inspection Platform of China provides the concentration of these six pollutants and the resultant AQI since December 2013 (http://www.aqistudy.cn/historydata/). Therefore, PM 10 concentration after December 2013 are collected from this website. It may be noted that no data are available in China for 11 months from January to November 2013.
The transition from the API to the AQI in 2013 may introduce the uncertainty in PM 10 concentration. To test the sensitivity of our result to data sources, all analyses are repeated for the two different periods, i.e., the early period from 2001/2002 winter to 2012 December using the API-derived PM 10 data, and the later period from 2013/2014 winter to 2020/2021 winter using the AQI-collected PM 10 data. The MJO-dependent PM 10 anomalies in these two periods are quantitatively similar ( Supplementary Fig. 6), indicating that the MJO-PM 10 relationship in China is not sensitive to the choice of data sources.
For both the Korean and Chinese stations, only those that exceed 75% of the temporal coverage are considered. This allows inclusion of a total of 100 stations, with 58 in Korea and 42 in China. For the stationaveraged PM 10 concentration, only the stations located below 40°N, which show a consistent response according to the MJO phase, were used (Fig. 1). Because Korean stations are densely distributed, only six stations (i.e., Busan, Daejeon, Gangneung, Gwangju, Jeju, and Seoul) are subsampled in this calculation. The overall results do not change much when different stations are selected.
The daily meteorological variables, including geopotential height, wind, and temperature, are obtained from European Centre for Medium Range Weather Forecasts Reanalysis 5 (ERA5) 42 . The gauge-based precipitation data from the Climate Prediction Center (CPC) (https://ftp.cpc.ncep.noaa.gov/precip/CPC_UNI_PRCP/) and the outgoing longwave radiation (OLR) data from the National Oceanic and Atmospheric Administration (NOAA) (https://psl.noaa. gov/data/gridded/data.olrcdr.interp.html) are also used. All variables are linearly interpolated into a horizontal resolution of 0.5°l ongitude × 0.5°latitude.
The daily climatology of each variable is derived from a 31-day moving average of the calendar day mean over 21 years. The anomaly is then calculated by subtracting the daily climatology from the raw data. To focus on the subseasonal variability, the low-frequency anomaly is further isolated by applying a 20-96 day Lanczos band-pass-filter to the daily anomaly. The cut-off periods of 20 and 96 days, the frequency associated with the MJO, are determined by the definition of the OLRbased MJO index (OMI) 43 used in this study. The resultant anomaly, which is referred to as the subseasonal anomaly, is used in the composite analysis. The statistical significance of the composite value is tested using Student's t-test. When calculating the degree of freedom, a continuous timeseries of one MJO phase (e.g., phase 1), separated by at least 7 days from the date of another MJO phase 1, is considered to be one degree of freedom 44 . The null hypothesis is that the population mean is zero.
Only the boreal winter (December to February) is considered in this study. When considering the extended winter, the overall results do not change much although some results become slightly weak. There is also no noticeable difference when detrended data are used.

MJO index and precipitation frequency index
The OMI 43 is utilized in this study. This index is directly obtained from NOAA website (https://www.psl.noaa.gov/mjo/mjoindex/). The MJO phase is defined with the two leading principal components (PC) timeseries, which are calculated by projecting 20-96 days filtered OLR anomalies over 20°S-20°N onto the two leading empirical orthogonal functions. Following previous studies, the MJO phase is divided into eight phases, depending on the location of OLR anomalies. The MJO amplitude is defined as below.
Only well-organized MJO days, with the MJO amplitude greater than one, are considered in the composite analysis. Approximately 64% of days in the analysis period are classified as MJO days.
To relate PM 10 concentration to precipitation events, the precipitation frequency index is defined. It is counted as one if the daily precipitation exceeds 0.1 mm day −1 , otherwise, it is set to zero. Although not shown, the key results are not sensitive to the choice of precipitation threshold varying from 0.1 to 5 mm day −1 . It is considered that more frequent precipitation (or a higher precipitation index) is associated with a lower PM 10 concentration.

Reconstruction of East Asia PM 10
East Asia PM 10