Accepted Manuscript

Multi-scale correlations between air quality and meteorology in the Guangdong−Hong

Kong−Macau Greater Bay Area of China during 2015–2017

Xingqin Fang, Qi Fan, Haowen Li, Zhiheng Liao, Jielan Xie, Shaojia Fan

PII: S1352-2310(18)30533-8
DOI: 10.1016/j.atmosenv.2018.08.018
Reference: AEA 16186

To appear in: Atmospheric Environment

Received Date: 15 April 2018

Revised Date: 7 August 2018
Accepted Date: 9 August 2018

Please cite this article as: Fang, X., Fan, Q., Li, H., Liao, Z., Xie, J., Fan, S., Multi-scale correlations
between air quality and meteorology in the Guangdong−Hong Kong−Macau Greater Bay Area of China
during 2015–2017, Atmospheric Environment (2018), doi: 10.1016/j.atmosenv.2018.08.018.

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to
our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please
note that during the production process errors may be discovered which could affect the content, and all
legal disclaimers that apply to the journal pertain.
 ACCEPTED MANUSCRIPT

2 Multi-scale correlations between air quality and meteorology

3 in the Guangdong−Hong Kong−Macau Greater Bay Area of China during 2015−2017

PT
4

RI
5 Xingqin Fang*, Qi Fan, Haowen Li, Zhiheng Liao, Jielan Xie, Shaojia Fan*

SC
6

7
U
School of the Atmospheric Sciences/Guangdong Province Key Laboratory for Climate
AN
8 Change and Natural Disaster Studies, Sun Yat-sen University, Guangzhou, 510275,
M

9 China
D

10
TE

11 Submitted to Atmospheric Environment

EP

12 2018/08/07
C

13
AC

14 Corresponding to:

15 Dr. Xingqin Fang (fangxq7@mail.sysu.edu.cn)

16 Dr. Shaojia Fan (eesfsj@mail.sysu.edu.cn)

 ACCEPTED MANUSCRIPT

17

18 Abstract

19 A multi-scale correlation analysis method is introduced and a combined analysis of

PT
20 multi-scale correlations between air quality and meteorology in Guangdong-Hong

RI
21 Kong-Macao Greater Bay Area (GHMGBA) of China is performed using 3-yr daily

SC
22 series of particulate matter (PM) fractions and O3 observations and collocated

23
U
meteorological reanalysis data during 2015-2017. PM and O3 have significant seasonal
AN
24 oscillation and oscillations of 2-3 days, 5-7 days, quasi two weeks, and 20-30 days,
M

25 which are typical periods in east-west wind (U), north-south wind (V), wind speed (WS),
D

26 temperature (T), relative humidity (RH), 925-hPa specific humidity (Q), and surface air
TE

27 pressure (SP), although different types of meteorological parameters have different

EP

28 significant periods. In GHMGBA, the East Asian Monsoon has different significant
C

29 impacts on PM and O3. The relationships of air quality with meteorology on higher
AC

30 frequency modes are usually very different from those on lower frequency modes since

31 different mechanisms make effects. Correlation details may also differ on different

32 synoptic scales. Especially, PM has consistent significant negative correlations with V, T,

2
 ACCEPTED MANUSCRIPT

33 RH, and Q on seasonal oscillation but has large positive (negative) correlations with V,

34 T, and Q (WS and SP) on oscillations of 2-3 days, 5-7 days, and quasi two weeks in

35 winter half year. The latter synoptic correlations reflect the impact of the typical

PT
36 quasi-stationary front and cold front activities on PM, which usually includes two

RI
37 sequential asymmetric processes: a long no-precipitation stagnation period when

SC
38 ventilation, dilution, and hygroscopic growth make effects; and a short precipitation

39
U
period when ventilation, dilution, and wet deposition make effects. In a word, local
AN
40 synoptic scales correlations cannot be directly extended to estimate the response of air
M

41 quality to climate change projections.

D

42
TE

43 Keywords:
EP

44 Multi-scale correlations; Air quality; Air pollution meteorology; Particulate matter;

C

45 Ozone; Guangdong−Hong Kong−Macau Greater Bay Area

AC

46

3
 ACCEPTED MANUSCRIPT

47

48 1 Introduction

49

PT
50 The Guangdong−Hong Kong−Macao Greater Bay Area (GHMGBA, Fig.1) of

RI
51 China includes nine cities of Guangdong province – Guangzhou, Shenzhen, Zhuhai,

SC
52 Foshan, Huizhou, Dongguan, Zhongshan, Jiangmen and Zhaoqing – as well as the Hong

53
U
Kong SAR and the Macao SAR. It covers an area of 56,000 square kilometers, and has
AN
54 68 million permanent residents. This area has been observed the most rapid economic
M

55 development in China during the last two decades. In GHMGBA, complex air pollution
D

56 issues, represented by high levels of urban and regional fine particulate matter (PM) and
TE

57 Ozone (O3) concentrations associated with photochemical reactions, are getting more
EP

58 and more concern (Wu et al., 2014a). It has been a consensus to pursue sustainable
C

59 economic development and avoid fatal air pollution events that happened decades ago
AC

60 during the old style of urbanization and industrialization. Many cooperative efforts from

61 the local governments have been made to monitor air quality and control air pollution in

62 this area.

4
 ACCEPTED MANUSCRIPT

63 Emissions of primary pollutants, ambient natural environments such as topography

64 and sea-land contrast, and meteorological conditions are all important impact factors of

65 air quality. In general, pollutant sources and ambient natural environments do not

PT
66 change much during a short period; however, the atmospheric circulations have

RI
67 complicated multi-scale coherent structures and variations (Fang and Kuo, 2015).

SC
68 Therefore, the air quality at any specific location has very complicated variations that

69
U
could not be explained only by sources and meteorological conditions play an import
AN
70 role (Galindo et al., 2010; Yadav et al., 2014; Zhang et al., 2016; Zhou et al., 2013).
M

71 Complicated climate variability and diverse aerosol feedbacks (Jacobson, 2001;

D

72 Rosenfeld et al., 2014) make it uneasy to see instant air quality improvement from
TE

73 emission control. In order to respond to public health concern, it is necessary to interpret

EP

74 air quality variabilities from meteorological aspect. The investigation of relationships

C

75 between air quality and meteorological situations also helps to build up air quality
AC

76 forecast capacity and contributes to coordinated air pollution control.

77 One approach for projecting the influence of climate change on air quality is using

78 the general circulation model (GCM) coupled with chemical transport model (CTM).

5
 ACCEPTED MANUSCRIPT

79 But this way is expensive and there are many uncertainties. An alternative approach is

80 using statistical models. The observational air quality-meteorology relationships are not

81 only essential for the statistical approach but also useful for testing parameters in the

PT
82 GCM-CTM approach. Many studies on air quality-meteorology correlations have been

RI
83 performed on various specific scales, but based on our knowledge, this paper is the first

SC
84 attempt to address the combined multi-scale air quality-meteorology correlations.

85
U
Some previous studies intended to test and calibrate the chemical transport model
AN
86 or extend the synoptic scale correlation to simulate or estimate the response of aerosol
M

87 to climate change (Dawson et al., 2007; Tai et al., 2010); while others tried to identify
D

88 significant meteorological impact factors on air quality, in terms of either individual

TE

89 parameters or certain synoptic patterns, to build up efficient day-to-day air quality

EP

90 statistic forecasts. The meteorological influences vary significantly across seasons and
C

91 locations (Zhang et al., 2015). As indicated in Li et al. (2017), the relationship of PM10
AC

92 and PM2.5 with meteorology is complicated. They found that high air temperature favors

93 the transformation of secondary PM through photochemical processes in summer, but

94 also results in efficient vertical dispersion of pollutants in autumn and winter. Large

6
 ACCEPTED MANUSCRIPT

95 relative humidity usually causes increases in PM concentrations due to the hygroscopic

96 effect of aerosols, but not for PM10 in spring and summer, mainly due to the suppression

97 of dust emissions under wet air conditions in spring and the effects of wet scavenging

PT
98 under high summer rainfall. Recently, Leung et al. (2017) suggested that the apparent

RI
99 correlations with individual meteorological variables may arise from common

SC
100 association with synoptic systems and recommended a process-based modeling. Zhang

101
U
et al. (2012) investigated the air quality characteristics of different circulation patterns
AN
102 derived from surface level pressure over a region around Beijing, China and suggested
M

103 that the circulation pattern types are the primary drivers of day-to-day variations of air
D

104 quality. Liao et al. (2018) examined the regulating effects of different atmospheric
TE

105 boundary layer (ABL) types on Beijing’s air quality and otherwise suggested that the
EP

106 ABL types are the primary drivers of day-to-day variations in Beijing’s air quality.
C

107 Some studies (Fan et al., 2008; Fan et al., 2011; Wu et al., 2013; Wu et al., 2014b) also
AC

108 found that the ABL characteristics are important for identifying the air pollution

109 problem in GHMGBA. Recently, Li et al. (2018) examined the combined effects of the

110 recirculation index and stable energy on air quality at Foshan for two typical episodes.

7
 ACCEPTED MANUSCRIPT

111 Actually, either the horizontal circulation or the vertical structure is a coherent part

112 of the whole atmospheric system. It is a good idea to investigate air quality-meteorology

113 correlations with a coherent multi-scale continuum of atmosphere in mind. A combined

PT
114 analysis of air quality-meteorology correlations not only involves multivariate

RI
115 correlations with meteorology on various timescales, but also considers behaviors of

SC
116 different coexisting pollutants as well as air quality diversity among neighboring cities.

117
U
It is expected that based on the general emission information, some details on physical
AN
118 and chemical mechanisms behind air quality variations might be derived from a
M

119 combined analysis.

D

120 For cities in GHMGBA of China, Ji et al. (2006) applied wavelet analysis to the air
TE

121 quality time series and found that all of the investigated air quality concentrations
EP

122 exhibit significant annual cycles and oscillations of 5−7 days, 10−20 days, and 30−60
C

123 days. However, they did not perform multi-scale correlation analysis. In this paper, we
AC

124 attempt to study the multi-scale correlations between air quality and meteorology in

125 GHMGBA using daily mean time series of air quality observations of PM10, PM2.5, and

126 O3 and relevant meteorological reanalysis data during 2015−2017. The Ensemble

8
 ACCEPTED MANUSCRIPT

127 Empirical Mode Decomposition (EEMD) method (Wu et al., 2016; Wu and Huang,

128 2009) is applied to get the multi-scale intrinsic mode functions (IMFs) of both air

129 quality and meteorology datasets. By applying the correlation analysis on decomposed

PT
130 modes, it is convenient to separate correlations on different scales without any other

RI
131 subjective filtering or smoothing. The scale separation also benefits for the investigation

SC
132 of correlation mechanisms between air quality and meteorology on individual scale. The

133
U
paper is organized as follows. Section 2 introduces the data and method. Section 3
AN
134 performs multi-scale correlation analysis of air quality and meteorology and interprets
M

135 correlation mechanisms scale by scale. Finally, Section 4 gives conclusions.

D

136
TE

137 2 Data and Method

EP

138
C

139 2.1 Data

AC

140

141 In this paper, a particular emphasis is paid on a region FGD in GHMGBA which is

142 composed of three neighboring cities Foshan (FS), Guangzhou (GZ), and Dongguan

9
 ACCEPTED MANUSCRIPT

143 (DG). The air quality variables studied include PM10, PM2.5, and O3. The collocated

144 meteorological parameters investigated include surface air pressure (SP), 10-m

145 east-west wind (U), 10-m north-south wind (V), 10-m wind speed (WS), 2-m

PT
146 temperature (T), 2-m relative humidity (RH), and 925-hPa specific humidity (Q).

RI
147 Daily mean air quality observations at FS, GZ, and DG during 2015−2017 from a

SC
148 research website (https://www.zq12369.com) are used for time series analysis. The

149
U
percentage of missing data is less than 1% and they are all rescued using a simple linear
AN
150 interpolation method. The small scale air quality variations within a city are omitted and
M

151 the mean value derived from the monitoring stations (featured as traffic, urban, industry,
D

152 background, etc.) carefully laid by environment monitoring authorities is taken as the
TE

153 representative air quality value for a citywide area. The loci of observation stations are
EP

154 shown in Fig. 1. There are generally about 5-10 stations scattered within a city. Since
C

155 the observation stations are not evenly distributed (with more stations located in the
AC

156 urban areas), the city mean air quality may have different representativeness at different

157 cities. But the representative errors and differences should not be large enough to affect

158 the results and conclusions of the daily mean time series analysis in this paper. The

10
 ACCEPTED MANUSCRIPT

159 meteorological data used in this study are from the European Center for Medium Range

160 Weather Forecast (ECMWF) Re-Analysis (ERA) (Dee et al., 2011). The ERA-Interim

161 data structured in a resolution of 0.75°×0.75° are available at a website

PT
162 (https://rda.ucar.edu/datasets/ds627.0/). Daily ERA-Interim data at 0600 UTC (i.e., 1400

RI
163 LST) on model grids ECF, ECG, and ECD (see Fig. 1) are selected to represent the

SC
164 collocated meteorological data for air quality at FS, GZ, and DG, respectively.

165
U
Since the cities in GHMGBA are influenced by the East Asian Monsoon (EAM)
AN
166 and have obvious alternating wet and dry season of precipitation, different sampling
M

167 periods of full year (FUL), winter half year (i.e., dry season) (WIN, from November to
D

168 April), and summer half year (i.e., wet season) (SUM, from May to October) are
TE

169 considered.
EP

170
C

171 2.2 Methods

AC

172

173 The idea of empirical mode decomposition (EMD) originated from the

174 Hilbert−Huang transform (Huang et al., 1996). This decomposition method is adaptive

11
 ACCEPTED MANUSCRIPT

175 to locality since it utilizes local minima and maxima to extract the IMFs from raw data.

176 Sequentially, N oscillatory IMFs from higher to lower frequency are extracted and the

177 residual is a non-oscillatory trend mode that consists of the background state and a trend.

PT
178 Each IMF represents an oscillation on a band of comparable scales, and the frequencies

RI
179 or periods can be estimated by the numbers of local minima and maxima. The EMD

SC
180 method is useful for data analysis and has been widely applied in many fields including

181
U
meteorology, but it has never been used for multi-scale correlations analysis of air
AN
182 quality data. Issues of mode mixing or frequency aliasing in EMD sometimes present,
M

183 where signals of disparate scales are mixed in one mode or a signal of similar scales
D

184 presents in different modes. As indicated in Huang et al. (1998), these issues result from
TE

185 signal intermittency and could lead to indistinct physical meaning of individual modes.
EP

186 The EEMD introduced by Wu and Huang (2009) is a noise assisted EMD, where the
C

187 EMD is performed on a noise-added ensemble of datasets. The EEMD method reduces
AC

188 the chance of frequency aliasing and is suitable for processing series that are

189 non-stationary and non-linear. The newest open version of NCAR (National Center for

190 Atmospheric Research) Command Language (NCL, http://www.ncl.ucar.edu) v6.4.0

12
 ACCEPTED MANUSCRIPT

191 provides the EEMD code.

192 A multi-scale correlation analysis is introduced in this paper with correlation

193 coefficients calculated using the corresponding daily time series of decomposed IMFs.

PT
194 For a comparison, the traditional correlation analysis is also performed using daily time

RI
195 series of raw datasets.

SC
196 Resembling the noise-to-signal ratio (NSR) concept in Fang and Kuo (2015), we

197
U
quantify the relative importance of IMFs with NSRIMF defined as
AN
∑
198 =∑ , (1)
M

199 where is a raw time series with = 1,2 … !, is the temporal mean,
D

200 is a decomposed IMF time series with = 1,2 … !, and M is the total length
TE

201 of time series (either monthly mean or daily mean series); " # is a spatial field with
EP

202 $ = 1,2 … %, " is the regional mean of the spatial field, and P is the total number of
C

203 points.
AC

204 By definition, the NSRIMF values of different modes for different variables at

205 different locations for different sampling periods are all comparable. In the following

206 analysis, the oscillation of an individual IMF is thought to be significant if the

13
 ACCEPTED MANUSCRIPT

207 corresponding NSRIMF value is larger than a criterion & '

≅ 0.0557, where & =

208 -1 + √50/2.

209

PT
210 3 Multi-scale correlation analysis of air quality and meteorology

RI
211

SC
212 3.1 Decomposed time series of air quality and meteorology in FGD

213
U
AN
214 Nine oscillatory IMFs (noted as IMFs 1-9) and one trend mode (noted as IMF10)
M

215 are extracted for each daily dataset of air quality variables and meteorological
D

216 parameters in FGD using the EEMD method. Figures 2-4 show daily series of EEMD
TE

217 IMFs of PM10, PM2.5, and O3, respectively, during 2015−2017. From the numbers of
EP

218 local maxima and minima, the periods of oscillations can be estimated as follows. IMFs
C

219 1-4 represent for sub-monthly oscillations of 2−3 days, 5−7 days, quasi two weeks, and
AC

220 20−30 days, respectively. IMFs 5 and 6 are for inter-seasonal oscillations of 40−60 days

221 and 80−90 days, respectively. IMF7 is for the seasonal oscillation. IMFs 8 and 9 are for

222 inter-annual fluctuations resolved by available data samples. IMF10 includes the

14
 ACCEPTED MANUSCRIPT

223 background state and a trend. As a good example, the zoomed segmental series of IMFs

224 1-7 of PM2.5 from 1 October 2016 to 31 March 2017 in Fig. 5a depict that the coherent

225 multi-scale developments on modes IMFs 1-7 produce an extreme PM2.5 air pollution

PT
226 episode at FS in January 2017. To present the higher frequencies in more details, Fig. 5b

RI
227 shows a zoomed-in segmental series of IMFs 1 and 2 of PM2.5 from 1 December 2016

SC
228 to 1 February 2017. Figure 6 shows daily time series of EEMD IMFs of U, V, WS, RH,

229 T, Q, and SP during 2015−2017.

U
AN
230 Following Eq.1, the NSRIMF values of modes IMFs 1-10 of each dataset are
M

231 calculated (see Table 1 for values of IMFs 1-7). It is found that the air quality in FGD
D

232 has very significant seasonal oscillation, which suggests the important impact of EAM
TE

233 on air quality. All PM fractions and O3 have significant common sub-monthly
EP

234 oscillation periods of 2−3 days, 5−7 days, quasi two weeks, and 20−30 days although
C

235 the significance is variable and has inter-city diversity. O3 also has significant
AC

236 inter-seasonal low frequency oscillations of about 40−60 days and 80−90 days in winter

237 half year. In addition, air quality in FGD has obvious inter-annual anomalies during

238 2015−2017. It is also found that V, U, SP, T, Q, and RH in FGD have significant

15
 ACCEPTED MANUSCRIPT

239 seasonal oscillations which reflect the EAM signals; but WS do not have significant

240 seasonal oscillation. U, V, WS, and RH in FGD have significant oscillations of 2−3 days,

241 5−7 days, and quasi two weeks. U and RH also have significant 20−30 days’ oscillations.

PT
242 T (except for T at ECD) has significant quasi two weeks’ oscillation. RH and V (except

RI
243 for V at ECG) have significant 40−60 days’ oscillations. T, Q, and SP in FGD are

SC
244 generally much more stable than other parameters on higher frequency modes but and

245
U
their seasonal oscillations are extremely significant. The significance of oscillations also
AN
246 has obvious seasonal dependence.
M

247 It should be noted that the amplitude evolution of 5−7 days’ period in Fig. 5b does
D

248 not present too much “weekend-weekday” difference. That is, the so-called “weekend
TE

249 effect” is not obvious in the studied air quality time series. One reason for this fact is
EP

250 that the city-wide averaging has removed the possible “weekend effect” associated with
C

251 small scale phase-locked emissions. And besides, day-to-day changes of emissions are
AC

252 general very small. Actually, the 5−7 days’ period is a typical meteorological period

253 which is usually controlled by the synoptic scale Rossby waves. As will be shown in the

254 following text, the impacts from meteorological conditions on the 5−7 days’ oscillation

16
 ACCEPTED MANUSCRIPT

255 of city mean daily air quality are very significant. Therefore, compared with the impacts

256 from meteorological conditions, the impacts from changes of emissions on changes of

257 city mean daily air quality are small.

PT
258

RI
259 3.2 Multi-scale correlations versus whole-scale correlations

SC
260

261
U
The FUL, WIN, and SUM sampling multi-scale correlation coefficients on modes
AN
262 IMFs 1-7 at FS are listed in Tables 2-4 for correlations with PM10, PM2.5, and O3
M

263 respectively. The corresponding multi-scale correlation coefficients at GZ and DG are

D

264 omitted for simplicity. The correlations on modes IMFs 8-10 are not considered here
TE

265 since the oscillation energies on these modes are very small. The FUL, WIN, and SUM
EP

266 sampling whole-scale correlation coefficients of air quality variables including PM10, O3,
C

267 and PM2.5, and various meteorological parameters including U, V, WS, T, RH, Q, and SP
AC

268 calculated using the daily raw time series during 2015−2017 at FS, GZ, and DG are

269 listed in Table 5.

270 As shown in Tables 2-4, air quality-meteorology correlation coefficients on

17
 ACCEPTED MANUSCRIPT

271 different modes are usually different, and the significance of correlations on some

272 modes has large seasonal dependence. The comparison with results in Table 5 indicates

273 that the apparent whole scale correlation coefficients can reflect real correlations well if

PT
274 the correlations on various modes are consistent, like for correlations with WS; but

RI
275 cannot reflect real correlation details if the correlations on various frequency bands are

SC
276 very different even contrary, like correlations with T or Q. For example, as shown in the

277
U
FUL sampling results in Tables 2-4, PM10 and PM2.5 have good positive correlations
AN
278 with V on higher frequency oscillations but good negative correlations on 80−90 days’
M

279 oscillation and seasonal oscillation; PM2.5 has good positive correlations with T on
D

280 sub-monthly oscillations but good negative correlation on 80−90 days’ oscillation and
TE

281 seasonal oscillation; PM2.5 has good positive correlation with Q on 2−3 days’ oscillation
EP

282 but good negative correlations on 20−30 days’ oscillation and inter-seasonal oscillations;
C

283 O3 has good negative correlations with Q on oscillations of quasi two weeks, 20−30
AC

284 days, and 40−60 days but good positive correlations on 80−90 days’ oscillation and

285 seasonal oscillation. Obviously, it is impossible to separate synoptic scale correlations

286 from lower frequency correlations simply with the whole-scale correlation analysis. In

18
 ACCEPTED MANUSCRIPT

287 order to focus on synoptic scales, Tai et al. (2010) tried to use the deseasonalized and

288 detrended meteorological data for the correlation and regression analysis by subtracting

289 the 30-day moving averages from raw data. However, they still could not separate

PT
290 correlations on different synoptic scales. By applying the correlation analysis on

RI
291 decomposed modes, we can separate correlations on different scales without any other

SC
292 filtering or smoothing. In addition, the coherent multi-scale correlation analysis makes it

293
U
possible to investigate correlation mechanisms between air quality and meteorology on
AN
294 individual scales.
M

295
D

296 3.3 Correlation mechanisms between PM and meteorology on lower frequency

TE

297 oscillations
EP

298
C

299 As shown by the FUL sampling correlation coefficients of PM on mode IMF7 in

AC

300 Tables 2 and 3, PM in FGD has consistent significant negative correlations with U, V, T,

301 RH, and Q on mode IMF7. Especially, PM2.5 has quite good negative correlations with

302 U (r=-0.60~-0.84), V (r=-0.71~-0.81), T(r=-0.63~-0.80), RH (r=-0.65~-0.86), and Q

19
 ACCEPTED MANUSCRIPT

303 (r=-0.78~-0.79). As a comparison, the apparent whole-scale correlation coefficients (in

304 Table 5) of PM2.5 at FS, GZ, and DG with U (r=-0.14, 0.03, and -0.18), V (r=-0.12,

305 -0.03, -0.16), T (r=-0.11, -0.09, and -0.15), RH (r=-0.42, -0.38, and -0.48), Q (r=-0.35,

PT
306 -0.31, and -0.41) are either much less significant or mixing for interpretation.

RI
307 As shown by IMF7 in Figs. 2 and 3, there are large positive (negative) anomalies

SC
308 of PM under cold dry northeasterly (warm moist southwesterly) anomalies during

309
U
winter (summer) half year, which reflects the significant impact of EAM on PM. The
AN
310 significant PM-meteorology correlations on mode IMF7 can be clearly interpreted as
M

311 follows. On the one hand, the ABL height is much lower during winter half year than
D

312 during summer half year; on the other hand, more scavenging of PM is produced by
TE

313 summer monsoon precipitation. As for WIN (SUM) sampling, stronger cold
EP

314 northeasterly (warm southwesterly) brings more polluted inland air (clean marine air) to
C

315 GHMGBA, makes more accumulations (better dispersion) of PM, thus produces higher
AC

316 (lower) level of PM concentrations during winter (summer) half year. In addition, more

317 precipitation is produced by stronger warm and moist southwesterly during summer half

318 year. Therefore, the impact mechanisms of EAM on PM is the combined effects of

20
 ACCEPTED MANUSCRIPT

319 vertical ventilation associated with T, horizontal dilution associated with U, V, and the

320 large-scale PM gradient, as well as the wet deposition associated with RH and Q.

321 The correlations on mode IMF6 (80−90 days’ oscillation) is a little complicated

PT
322 since several meteorological parameters like T, Q, and SP have frequency aliasing on

RI
323 modes IMF6 and IMF7. The apparent correlations with these parameters on mode IMF6

SC
324 are not considered for the following discussion focusing on the 80−90 days’ oscillation.

325
U
As shown by the SUM sampling correlation coefficients in Tables 2 and 3, PM fractions
AN
326 in FGD generally have good negative correlations with U, V, WS, and RH on mode IMF
M

327 6 in summer half year. As shown by IMF6 in Figs. 2 and 3, large negative anomalies in
D

328 PM are related to the large southerly and westerly anomalies (i.e., stronger summer
TE

329 monsoon) during the summer half year of 2015, which are mostly a response to the
EP

330 ENSO signal during the strong El Nino year 2015-2016, via scavenging mechanisms.
C

331 As shown by the WIN sampling correlation coefficients in Tables 2 and 3, PM2.5 in FGD
AC

332 has good negative (positive) correlations with V and RH (U) on mode IMF6 in winter

333 half year. We also checked the correlations with PM2.5-10 (the difference between PM10

334 and PM2.5) in FGD (figure and table omitted) and found that PM2.5-10 has good positive

21
 ACCEPTED MANUSCRIPT

335 correlations with U, V, WS, and RH on mode IMF6 in winter half year. It is obvious that

336 the PM2.5-RH correlation coefficients and the PM2.5-10-RH correlation coefficients have

337 different sign on mode IMF6 by WIN sampling. As shown by IMF6 in Fig. 6, there are

PT
338 large northerly anomalies during spring of 2015, winter of 2016, and autumn in 2017,

RI
339 which are probably impacted by the ENSO. Together with these negative V anomalies,

SC
340 there are large negative RH anomalies. Anomalies in U have sharp seasonal transition

341
U
during spring of 2015 and autumn in 2017 and are generally out of phase with those V
AN
342 anomalies. WS has large anomalies in 2017 only. It seems that the positive correlations
M

343 of PM2.5-10 with U, V, WS, and RH on mode IMF6 in winter half year might be related
D

344 to the variations in dust emission and advection which are dominantly impacted by WS.
TE

345 While the negative (positive) correlations of PM2.5 with V and RH (U) on mode IMF 6
EP

346 in winter half year might be due to scavenging of precipitation induced by the
C

347 southeasterly anomalies during transition seasons.

AC

348

349 3.4 Correlation mechanisms between PM and meteorology on higher frequency

350 oscillations

22
 ACCEPTED MANUSCRIPT

351

352 As shown in Tables 2 and 3, PM has large positive (negative) correlations with V, T,

353 and Q (WS and SP) on modes IMFs 1-3 and has large negative correlation with RH on

PT
354 mode IMF3 by WIN sampling. Under the impact of Nanling mountains (refer to the

RI
355 north part of Fig. 1 for the topography), there is usually a quasi-stationary front

SC
356 hovering over South China in winter half year. Before a new cold front comes, there is a

357
U
period of several days when the FGD region is located at the warm sector of a
AN
358 quasi-stationary front. The front inversion suppresses the vertical dispersion, and the
M

359 accumulation of high level of PM is accompanied with local warm advection, moist
D

360 convergence, and updrafts at meso-low. Stagnation with low wind and high moisture
TE

361 (but no precipitation) also induces hygroscopic growth of PM. After a long period of
EP

362 seesaw battle, once a new cold front comes and the wet deposition is triggered, the PM
C

363 level reduces rapidly. This is a very typical synoptic process in FGD (also in
AC

364 GHMGBA). The significant higher frequency energy reflects the close competitions of

365 warm mass and cold mass. Modes IMFs 1 and 2 (i.e., 2−3 days’ and 5−7 days’

366 oscillations) in winter half year usually do not involve precipitation, since PM (except

23
 ACCEPTED MANUSCRIPT

367 for PM2.5-10 on mode IMF2 at DG) has quite good correlations with Q on modes IMFs 1

368 and 2 due to hygroscopic effect but poor and mixed correlations with RH. For example,

369 the significant positive correlation with Q on mode IMF1 by WIN sampling is up to

PT
370 0.29~0.33 for PM10, 0.29~0.30 for PM2.5, and 0.13~0.34 for PM2.5-10, while the

RI
371 correlations with RH (either positive of negative) range only from 0.01 to 0.14. On the

SC
372 contrary, on mode IMF3 (i.e., quasi two weeks’ oscillation), the negative correlations

373
U
with RH by WIN sampling turn to be as good as 0.46~0.55 for PM10, 0.37~0.49 for
AN
374 PM2.5, and 0.48~0.55 for PM2.5-10, which reflects the scavenging effect on PM from
M

375 precipitation. Taking the above analyses together, the correlation mechanisms between
D

376 PM and meteorology on modes IMFs 1-3 in winter half year are the combined effects of
TE

377 two sequential asymmetric processes: 1) a long no-precipitation stagnation period with
EP

378 alternating weak advantages of warm mass or cold mass when ventilation, dilution, and
C

379 hygroscopic growth make effects; and 2) a short precipitation period associated with
AC

380 strong northerly, high RH, and cold T when ventilation, dilution, and wet deposition

381 make effects. The accumulation of PM during the first period also contributes to the

382 triggering of precipitation, which is a negative feedback mechanism.

24
 ACCEPTED MANUSCRIPT

383 As shown in Tables 2 and 3, PM has large negative (positive) correlations with WS,

384 RH, and Q (T and SP) on modes IMFs 4 and 5 by WIN sampling. On the one hand, the

385 dry downdraft anomalies accompanied with higher pressure and warm anomalies

PT
386 depresses the vertical dispersion of PM; on the other hand, the stagnation with lower

RI
387 wind favors the accumulations of PM. Therefore, the correlation mechanisms are the

SC
388 combined effects of enhanced dry downdraft suppression associated with SP, T, RH, and

389
U
Q and the horizontal dispersion capability associated with WS.
AN
390 As shown in Tables 2 and 3, PM has large negative (positive) correlations with WS
M

391 and RH (T) on modes IMFs 1 and 2 by SUM sampling. On the one hand, the dry
D

392 weather and subsidence inversion suppress the vertical ventilation of PM and the low
TE

393 wind situations reduce the horizontal dilution. On the other hand, warmer and drier
EP

394 surface is related to more ground dust emission in summer half year. The fact that
C

395 PM2.5-10 has larger correlations with T and RH than PM2.5 on modes IMFs 1 and 2 at FS
AC

396 and GZ supports this point, since ground dust is a major contribution to PM2.5-10 in these

397 two cities. PM2.5-10 has smaller correlation with T than PM2.5 at DG on modes IMFs 1

398 and 2 mostly because sea salt rather than ground dust contributes more to coarse PM at

25
 ACCEPTED MANUSCRIPT

399 DG. The dry and hot weather under high pressure is typical in summer half year in FGD

400 (also in GHMGBA), and the PM level is generally not high due to good ventilation and

401 dilution. However, the persistent strong dry downdraft suppression induced by enhanced

PT
402 subtropical high or distant cyclones may make very high level of PM accumulations.

RI
403 The length of PM oscillation period depends on the duration of subsidence inversion

SC
404 which is usually related to duration of suppression and the moving speed of remote

405
U
cyclones. Note that although PM fractions have good positive correlation with T on
AN
406 mode IMF2 (i.e., 5−7 days’ oscillation) by SUM sampling: 0.49~0.52 for PM10,
M

407 0.46~0.48 for PM2.5, and 0.40~0.51 for PM2.5-10, it is not the high surface temperature
D

408 but the subsidence inversion that induces the PM accumulations.

TE

409 Taken together, the correlation mechanisms of PM with WS, RH, and T on modes
EP

410 IMFs 1 and 2 in summer half year are the combined effects of dry downdraft
C

411 suppression mostly driven by the subtropical high or remote cyclones, the enhanced
AC

412 upward surface flux of ground dust associated with low RH and hot T, and the

413 horizontal dispersion capability associated with low WS.

414 As shown in Tables 2 and 3, PM has quite good negative correlations with V, WS,

26
 ACCEPTED MANUSCRIPT

415 RH, and Q on modes IMFs 3-5 by SUM sampling. During summer half year, stronger

416 southerly flow brings more moisture and produces more scavenging of PM by

417 precipitation. The correlation mechanisms are the combined effects of the enhanced wet

PT
418 deposition associated with both RH and Q, dilution associated with WS, and the

RI
419 horizontal advection associated with V and regional scale PM gradient. The large

SC
420 positive correlation of PM with T on mode IMF3 is a coexisting phenomenon since

421
U
precipitation in summer half year is usually accompanied with surface cooling.
AN
422
M

423 3.5 Correlation mechanisms between O3 and meteorology on lower frequency

D

424 oscillation
TE

425
EP

426 As shown by the FUL WIN, and SUM sampling correlation coefficients of O3 with
C

427 various meteorological parameters on modes IMF7 in Table 4, O3 has large positive
AC

428 correlation with U, V, T, RH and Q on mode IMF7 by FUL sampling, and the

429 correlations are generally much more significant in winter half year. As shown by IMF7

430 in Figs. 4 and 6, there are large negative (positive) O3 anomalies under the cold dry

27
 ACCEPTED MANUSCRIPT

431 northeasterly (warm moist southwesterly) anomalies during winter (summer) half year,

432 which reflects the significant impact of EAM on O3. Basically, the shortwave radiation

433 is much weaker and produces less O3 during winter half year than during summer half

PT
434 year. In addition, the cold temperature in winter half year also depresses the production

RI
435 of volatile organic compounds which are important precursors of O3. The main impact

SC
436 mechanisms of EAM on O3 are the combined effects of seasonal oscillation of

437
U
shortwave radiation associated with T and Q and the volatility variation associated with
AN
438 T. The positive correlations of O3 with U, V, and RH on mode IMF7 are actually the
M

439 coexisting phenomena. The reason for the less significant correlations by SUM
D

440 sampling than by WIN sampling might be that the production of O3 in summer half year
TE

441 is usually upper bounded by the availability of various precursors.

EP

442 As shown by the FUL WIN, and SUM sampling correlation coefficients of O3 on
C

443 modes IMFs 5 and 6 in Table 4, O3 have large negative correlations with U, V, WS, RH,
AC

444 and Q on modes IMFs 5 and 6 by SUM sampling. As shown by IMFs 5 and 6 in Figs. 4

445 and 6, large negative anomalies of O3 during summer half year of 2015 are correlated

446 with large positive anomalies in U and V (i.e., stronger summer monsoon). Basically,

28
 ACCEPTED MANUSCRIPT

447 stronger (weaker) southwesterly brings more (less) moisture from ocean, induces more

448 (less) cloud and lower (higher) level of shortwave radiation, and thus produce lower

449 (higher) level of O3. This reflects the impact of ENSO on O3 during summer half year

PT
450 through modulations on oscillations of 40−60 days and 80−90 days, and the impact

RI
451 mechanisms are the combined effects of shortwave radiation variation in moist

SC
452 microphysics process associated with RH and Q and the local horizontal dilution

453 associated with U, V, and WS.

U
AN
454
M

455 3.6 Correlation mechanisms between O3 and meteorology on higher frequency

D

456 oscillations
TE

457
EP

458 As shown in Table 4, O3 has large negative (positive) correlations with WS and RH
C

459 (T) on modes IMFs 1 and 2 by SUM sampling and has large negative (positive)
AC

460 correlations with WS, RH and Q (T) on modes IMFs 1 and 2 by WIN sampling. These

461 large correlations are represented well in the FUL sampling results. During both

462 summer half year and winter half year, higher level of shortwave radiation associated

29
 ACCEPTED MANUSCRIPT

463 with warmer and drier environments favors the production of more O3, and lower WS

464 induces the local accumulation of O3. Therefore, on modes IMFs 1 and 2, the

465 meteorological impact mechanisms on O3 are generally consistent along the whole year,

PT
466 that is, the combined effects of shortwave radiation fluctuations associated with T, RH,

RI
467 and Q and the horizontal dispersion capability associated with WS.

SC
468 As shown in Table 4, O3 has large negative (positive) correlations with V, WS, RH,

469
U
and Q (T) on mode IMF3 by SUM sampling and has large negative (positive)
AN
470 correlations with WS, RH, and Q (T and V) on mode IMF3 by WIN sampling. Basically,
M

471 stronger southerly during summer half year brings more warm moist air, induces more
D

472 warm clouds and less shortwave radiations, and produces less O3, where the warm
TE

473 advection dominates over the radiation cooling. Weaker northerly during winter half
EP

474 year brings less cold dry air, induces less cold front clouds and more shortwave
C

475 radiation, and produces more O3. The impact mechanisms are the combined effects of
AC

476 shortwave radiation fluctuations associated with T, RH, and Q and the horizontal

477 dispersion capability associated with WS and V.

478 As shown in Table 4, O3 has large negative (positive) correlations with WS, RH

30
 ACCEPTED MANUSCRIPT

479 and Q (T and SP) on mode IMFs 4 and 5 by WIN sampling. On the one hand, more

480 shortwave radiation associated with high pressure and clear sky produces more O3 in

481 column; on the other hand, the subsidence accompanied with high pressure and dry

PT
482 warm air usually produces net downward transportation of O3 and increases the level of

RI
483 surface O3. Therefore, the impact mechanisms are the combined effects of shortwave

SC
484 radiation fluctuations, vertical transportation of O3, and the horizontal dispersion

485
U
capability associated with WS. Obviously, the less significant correlations between O3
AN
486 and meteorology on mode IMF4 by SUM sampling have similar mechanisms.
M

487
D

488 4 Conclusions and discussions

TE

489
EP

490 Philosophically, the response of air quality to weather and climate variabilities
C

491 should be explored with a coherent multi-scale continuum of atmosphere in mind. In

AC

492 this paper, a multi-scale correlation analysis method is introduced to separate

493 correlations on different scales for mechanism analysis by applying the correlation

494 analysis on the intrinsic modes decomposed with the EEMD method. A combined

31
 ACCEPTED MANUSCRIPT

495 analysis of the multi-scale correlations between air quality and meteorology in

496 GHMGBA during 2015−2017 is performed . Some findings are summarized as follows:

497 (1) The air quality in GHMGBA has very significant high and low frequency

PT
498 oscillations that are typical oscillations in meteorology although the significance is

RI
499 variable.

SC
500 (2) The apparent whole scale correlation coefficients cannot reflect real correlation

501
U
details if the correlations on various frequency bands are very different even contrary.
AN
502 (3) PM in GHMGBA has consistent significant negative correlations with V, T, RH,
M

503 and Q on seasonal oscillation, which reflects the significant impact of EAM. The impact
D

504 mechanisms are the combined effects of vertical ventilation associated with T,
TE

505 horizontal dilution associated with U, V, and the large-scale PM gradient, as well as the
EP

506 wet deposition associated with RH and Q. PM generally has good negative correlations
C

507 with U, V, WS, and RH on 80−90 days’ oscillation in summer half year mostly via
AC

508 scavenging mechanisms. PM2.5-10 generally has good positive correlations with U, V,

509 WS, and RH on 80−90 days’ oscillation in winter half year. PM2.5 has good negative

510 (positive) correlations with V and RH (U) on 80−90 days’ oscillation in winter half year

32
 ACCEPTED MANUSCRIPT

511 due to the scavenging of precipitation induced by southeasterly anomalies during

512 transition seasons.

513 (4) PM in GHMGBA has large positive (negative) correlations with V, T, and Q

PT
514 (WS and SP) on oscillations of 2−3 days, 5−7 days, and quasi two weeks and has large

RI
515 negative correlation with RH on quasi two weeks’ oscillation in winter half year. These

SC
516 correlations reflect the impact of the typical quasi-stationary front and cold front

517
U
activities in winter half year. The correlation mechanisms are the combined effects of
AN
518 two sequential asymmetric processes: a long no-precipitation stagnation period with
M

519 alternating weak advantages of warm mass or cold mass when ventilation, dilution, and
D

520 hygroscopic growth make effects; and a short precipitation period associated with
TE

521 strong northerly, high RH, and cold T when ventilation, dilution, and wet deposition
EP

522 make effects. The accumulation of PM during the first period also contributes to the
C

523 triggering of precipitation, which is a negative feedback mechanism. PM has large

AC

524 negative (positive) correlations with WS, RH, and Q (T and SP) on oscillations of

525 20−30 days and 40−60 days in winter half year. The correlation mechanisms are the

526 combined effects of enhanced dry downdraft suppression associated with SP, T, RH, and

33
 ACCEPTED MANUSCRIPT

527 Q and the horizontal dispersion capability associated with WS.

528 PM has large negative (positive) correlations with WS and RH (T) on oscillations

529 of 2−3 days and 5−7 days in summer half year. The correlation mechanisms are the

PT
530 combined effects of dry downdraft suppression mostly driven by the subtropical high or

RI
531 remote cyclones, the enhanced upward surface flux of ground dust associated with low

SC
532 RH and hot T, and the horizontal dispersion capability associated with low WS. PM has

533
U
quite good negative correlations with V, WS, RH, and Q on oscillations of quasi two
AN
534 weeks, 20−30 days, and 40−60 days in summer half year. The correlation mechanisms
M

535 are the combined effects of the enhanced wet deposition associated with both RH and Q,
D

536 the dilution associated with WS, and the horizontal advection associated with V and
TE

537 regional scale PM gradient. The large positive correlation of PM with T on quasi two
EP

538 weeks’ oscillation is a coexisting phenomenon.

C

539 (5) O3 in GHMGBA has large positive correlation with U, V, T, RH and Q on

AC

540 seasonal oscillation, which reflects the significant impact of EAM on O3. The main

541 impact mechanisms are the combined effects of seasonal oscillation of shortwave

542 radiation associated with T and Q and the volatility variation associated with T. The

34
 ACCEPTED MANUSCRIPT

543 positive correlations of O3 with U, V, and RH are actually the coexisting phenomena.

544 The reason for the less significant correlations by SUM sampling than by WIN

545 sampling might be that the production of O3 in summer half year is usually upper

PT
546 bounded by the availability of various precursors. O3 has large negative correlations

RI
547 with U, V, WS, RH, and Q on the inter-seasonal modes in summer half year. The impact

SC
548 mechanisms are the combined effects of shortwave radiation variation in moist

549
U
microphysics process associated with RH and Q and the local horizontal dilution
AN
550 associated with U, V, and WS.
M

551 (6) O3 in GHMGBA has large negative (positive) correlations with WS and RH (T)
D

552 on oscillations of 2−3 days and 5−7 days in summer half year; and have large negative
TE

553 (positive) correlations with WS, RH and Q (T) on oscillations of 2−3 days and 5−7 days
EP

554 in winter half year. The meteorological impact mechanisms on O3 are generally
C

555 consistent along the whole year, that is, the combined effects of shortwave radiation
AC

556 fluctuations associated with T, RH, and Q and the horizontal dispersion capability

557 associated with WS.

558 O3 has large negative (positive) correlations with V, WS, RH, and Q (T) on quasi

35
 ACCEPTED MANUSCRIPT

559 two weeks’ oscillation in summer half year, and has large negative (positive)

560 correlations with WS, RH, and Q (T and V) on quasi two weeks’ oscillation in winter

561 half year. The impact mechanisms are the combined effects of shortwave radiation

PT
562 fluctuations associated with T, RH, and Q, and the horizontal dispersion capability

RI
563 associated with WS and V.

SC
564 O3 has large negative (positive) correlations with WS, RH and Q (T and SP) on

565
U
oscillations of 20−30 days and 40−60 days in winter half year. The impact mechanisms
AN
566 are the combined effects of shortwave radiation fluctuations, vertical transportation of
M

567 O3, and the horizontal dispersion capability associated with WS. The less significant
D

568 correlations between O3 and meteorology on oscillation of 40−60 days in summer half
TE

569 year have similar mechanisms.

EP

570
C

571 Overall, an important contribution of this paper is that we introduced a multi-scale

AC

572 correlation analysis method that is useful to identify the respective main impact

573 mechanisms of meteorology on air quality on different scales. Using this method, we

574 can separate correlations on different scales without any other filtering or smoothing.

36
 ACCEPTED MANUSCRIPT

575 The coherent multivariate multi-scale correlation analysis makes it possible to interpret

576 correlation mechanisms on individual scale.

577 Multi-scale correlations were not considered even in a recent novel study by Shen

PT
578 et al. (2017), which was somehow as the extension and update of some previous studies

RI
579 (Tai et al., 2010; Tai et al., 2012a; Tai et al., 2012b). They regressed monthly mean

SC
580 PM2.5 concentrations with both local meteorology and surrounding patterns diagnosed

581
U
by singular value decomposition (SVD) algorithm and underscored the importance of
AN
582 evaluating the sensitivity of PM2.5 to its key controlling meteorological variables in
M

583 climate-chemistry models on multiple timescales before they are applied to project
D

584 future air quality. Our results confirm the different scale sensitivity of PM to
TE

585 meteorology and alert careful extension of synoptic sensitivity for estimating the
EP

586 response of air quality to climate change projections.

C

587 In addition, the multi-scale correlation analysis method introduced in this paper
AC

588 might lead to a seamless multi-scale regression. Our future plan is to extend our work to

589 build up a meteorology-based multi-scale multiple linear regression (MLR) statistic

590 model for air quality study. Our idea should be not only good for day-to-day air quality

37
 ACCEPTED MANUSCRIPT

591 forecasts but also useful for deriving more probability details of climate projection.

592 We do not consider the diurnal cycle in this paper, neither the relative abundance of

593 PM components. More improvements can also be achieved in the future by considering

PT
594 some more relevant meteorological parameters, for example, the rainfall and radiation

RI
595 data, and on-site meteorological observations. Moreover, we do not include emissions in

SC
596 this study; further study on emission variabilities with relevant data available should

597
U
benefit better understanding of air quality variabilities.
AN
598
M

599
D

600 Acknowledgments
TE

601
EP

602 This work is supported by National Key R&D Program of China

C

603 (No.2017YFC0209606, No.2016YFC0203305) and National Science Foundation Key

AC

604 Program of China (No.41630422).

605

606 Figure Captions

38
 ACCEPTED MANUSCRIPT

607 Fig. 1 Topography map of the Guangdong−Hong Kong−Macao Greater Bay Area

608 (GHMGBA, the pinked area) and its surroundings (see the blue box in the

609 superimposed administrative map for reference). The GHMGBA includes 11 cities:

PT
610 Guangzhou (GZ), Zhaoqing (ZQ), Foshan (FS), Dongguan (DG), Huizhou (HZ),

RI
611 Jiangmen (JM), Zhongshan (ZS), Shenzhen (SZ), Zhuhai (ZH), Hong Kong (HK),

SC
612 and Macao (MC). The four dashed boxes centered at (22.807°N, 113.203°E),

613
U
(23.509°N, 113.203°E), (22.807°N, 113.906°E), and (23.5 09°N, 113.906°E),
AN
614 respectively, show the closest model grids where the ERA-Interim reanalysis data
M

615 (in a resolution of 0.75°×0.75°) are extracted as the collocated meteorological data
D

616 for the air quality data at cities FS, GZ, DG, and HZ, respectively; and they are
TE

617 denoted as ECF, ECG, ECD, and ECH, respectively. The point (23.158°N,
EP

618 113.555°E) marked by a red star is the common vertex of the four model grids and
C

619 for convenience it is selected as the representative point to anchor the geographical
AC

620 location for GHMGBA in the larger scale administrative map. The green circles are

621 the loci of observation stations.

622 Fig. 2 Daily time series of EEMD intrinsic mode functions (IMFs) (unit: µg/m3) of

39
 ACCEPTED MANUSCRIPT

623 PM10 during 2015−2017 at FS, GZ, and DG. The corresponding square root of

624 variance (unit: µg/m3) and NSRIMF are listed at top of plots for IMFs 1-10. The

625 square root of variance (unit: µg/m3) and mean (unit: µg/m3) for raw time series are

PT
626 listed at bottom of plot for IMF10.

RI
627 Fig. 3 The same as Fig. 2 but for PM2.5.

SC
628 Fig. 4 The same as Fig. 2 but for O3.

629
U
Fig. 5 Highlighted daily time series of IMFs of PM2.5 (unit: µg/m3) at FS, GZ, and DG:
AN
630 (a) IMFs 1-7 from October 2016 to March 2017; (b) IMFs 1 and 2 from 1
M

631 December 1 2016 to 1 February 2017. The X axis in (b) is labelled in weekly cycle
D

632 with “1” represents “Monday” and so on.

TE

633 Fig. 6 Daily time series of EEMD intrinsic mode functions (IMFs) of (a) U (unit: m/s),
EP

634 (b) V (unit: m/s), (c) WS (unit: m/s), (d) RH (unit: %), (e) T (unit: °C), (f) Q (unit:
C

635 mg/kg3), and (g) SP (unit: hPa) during 2015−2017 at model grids ECF, ECG, and
AC

636 ECD representing for FS, GZ, and DG, respectively. The corresponding square root

637 of variance and NSRIMF are listed at top of plots for IMFs 1-10. The square root of

638 variance and mean for raw time series are listed at bottom of plot for IMF10.

40
 ACCEPTED MANUSCRIPT

639

PT
RI
U SC
AN
M
D
TE
EP
C
AC

41
 ACCEPTED MANUSCRIPT

640

641 Table lists:

642

PT
643 Table 1 FUL, WIN, and SUM sampling NSRIMF of modes IMFs 1-7 of air quality (PM10,

RI
644 PM2.5, and O3) and meteorological parameters (U, V, WS, RH, T, Q, and SP) in FGD

SC
645 during 2015−2017*

646
U
Table 2 FUL, WIN, and SUM sampling multi-scale correlation coefficients of PM10 with
AN
647 various meteorological parameters on modes IMFs 1-7 at FS*
M

648 Table 3 FUL, WIN, and SUM sampling multi-scale correlation coefficients of PM2.5 with
D

649 various meteorological parameters on modes IMFs 1-7 at FS*

TE

650 Table 4 FUL, WIN, and SUM sampling multi-scale correlation coefficients of O3 with
EP

651 various meteorological parameters on modes IMFs 1-7 at FS*

C

652 Table 5 FUL, WIN, and SUM sampling whole-scale correlation coefficients between
AC

653 PM10, PM2.5, and O3, respectively, and various meteorological parameters at FS, GZ,

654 and DG*

655

42
 ACCEPTED MANUSCRIPT

656 Reference

657

658 Dawson, P.J., Adams, P.J., Pandis, S.N., 2007. Sensitivity of PM2.5 to climate in the
659 Eastern US: a modeling case study. Atmos. Chem. Phys. 7, 4295-4309.

PT
660 Dee, D.P., Uppala, S.M., Simmons, A.J., Berrisford, P., Poli, P., Kobayashi, S., Andrae,
661 U., Balmaseda, M.A., Balsamo, G., Bauer, P., Bechtold, P., Beljaars, A.C.M., Berg,

RI
662 L.v.d., Bidlot, J., Bormann, N., Delsol, C., Dragani, R., Fuentes, M., Geer, A.J.,
663 Haimberger, L., Healy, S.B., Hersbach, H., Hólm, E.V., Isaksen, L., Kållberg, P.,

SC
664 Köhler, M., Matricardi, M., McNally, A.P., Monge‐Sanz, B.M., Morcrette, J.J.,
665 Park, B.K., Peubey, C., Rosnay, P.d., Tavolato, C., Thdrae, U., Balmaseda, M.A.,
666

U
Balsamo, G.‐ Interim reanalysis: configuration and performance of the data
667 assimilation system. Quarterly Journal of the Royal Meteorological Society 137,
AN
668 553-597.
669 Fan, S., Wang, B., Tesche, M., Engelmann, R., Althausen, A., Liu, J., Zhu, W., Fan, Q.,
M

670 Li, M., Ta, N., 2008. Meteorological conditions and structures of atmospheric
671 boundary layer in October 2004 over Pearl River Delta area. Atmos Environ 42,
D

672 6174-6186.
TE

673 Fan, S.J., Fan, Q., Yu, W., Luo, X.Y., Wang, B.M., Song, L.L., Leong, K.L., 2011.
674 Atmospheric boundary layer characteristics over the Pearl River Delta, China,
675 during the summer of 2006: measurement and model results. Atmospheric
EP

676 Chemistry and Physics 11, 6297-6310.

677 Fang, X., Kuo, Y.H., 2015. A New Generic Method for Quantifying the Scale
C

678 Predictability of the Fractal Atmosphere: Applications to Model Verification. J

AC

679 Atmos Sci 72, 1667-1688.

680 Galindo, N., Varea, M., Gil-Moltó, J., Yubero, E., Nicolás, J., 2010. The Influence of
681 Meteorology on Particulate Matter Concentrations at an Urban Mediterranean
682 Location. Water, Air, & Soil Pollution 215, 365-372.
683 Huang, N.E., Long, S.R., Shen, Z., 1996. The Mechanism for Frequency Downshift in
684 Nonlinear Wave Evolution, in: Hutchinson, J.W., Wu, T.Y. (Eds.), Advances in
685 Applied Mechanics. Elsevier, pp. 59-117C.
686 Huang, N.E., Shen, Z., Long, S.R., Wu, M.C., Shih, H.H., Zheng, Q., Yen, N.-C., Tung,

43
 ACCEPTED MANUSCRIPT
687 C.C., Liu, H.H., 1998. The empirical mode decomposition and the Hilbert
688 spectrum for nonlinear and non-stationary time series analysis. Proceedings of the
689 Royal Society of London. Series A: Mathematical, Physical and Engineering
690 Sciences 454, 903-995.
691 Jacobson, M.Z., 2001. Global direct radiative forcing due to multicomponent
692 anthropogenic and natural aerosols. Journal of Geophysical Research: Atmospheres

PT
693 106, 1551-1568.
694 Ji, Z.P., Luo, S.B., Kuang, J.X., Luo, Q.H., Chen, M.S., Liang, G.X., 2006. Variation

RI
695 characteristics and prediction of air pollution in Guangzhou (in Chinese). Journal
696 of Tropical Meteorology 22, 574-581.

SC
697 Leung, D.M., Tai, A.P.K., Mickley, L.J., Moch, J.M., van Donkelaar, A., Shen, L.,
698 Martin, R.V., 2017. Synoptic meteorological modes of variability for fine
699

U
particulate matter (PM2.5) air quality in major metropolitan regions of China.
700 Atmospheric Chemistry and Physics Discussions, 1-29.
AN
701 Li, H., Wang, B., Fang, X., Zhu, W., Fan, Q., Liao, Z., Liu, J., Zhang, A., Fan, S., 2018.
702 Combined effect of boundary layer recirculation factor and stable energy on local
M

703 air quality in pearl river delta over southern China. Journal of the Air & Waste
704 Management Association, 1-15.
D

705 Li, X., Ma, Y., Wang, Y., Liu, N., Hong, Y., 2017. Temporal and spatial analyses of
TE

706 particulate matter (PM 10 and PM 2.5 ) and its relationship with meteorological
707 parameters over an urban city in northeast China. Atmos Res 198, 185-193.
708 Liao, Z., Sun, J., Yao, J., Liu, L., Li, H., Liu, J., Xie, J., Wu, D., Fan, S., 2018.
EP

709 Self-organized classification of boundary layer meteorology and associated

710 characteristics of air quality in Beijing. Atmospheric Chemistry and Physics 18,
C

711 6771-6783.
AC

712 Rosenfeld, D., Andreae, M.O., Asmi, A., Chin, M., de Leeuw, G., Donovan, D.P., Kahn,
713 R., Kinne, S., Kivekäs, N., Kulmala, M., Lau, W., Schmidt, K.S., Suni, T., Wagner,
714 T., Wild, M., Quaas, J., 2014. Global observations of
715 aerosol-cloud-precipitation-climate interactions. Reviews of Geophysics 52,
716 750-808.
717 Shen, L., Mickley, L.J., Murray, L.T., 2017. Influence of 2000–2050 climate change on
718 particulate matter in the United States: results from a new statistical model.
719 Atmospheric Chemistry and Physics 17, 4355-4367.

44
 ACCEPTED MANUSCRIPT
720 Tai, A.P.K., Mickley, L.J., Jacob, D.J., 2010. Correlations between fine particulate
721 matter (PM2.5) and meteorological variables in the United States: Implications for
722 the sensitivity of PM2.5 to climate change. Atmos Environ 44, 3976-3984.
723 Tai, A.P.K., Mickley, L.J., Jacob, D.J., 2012a. Impact of 2000–2050 climate change on
724 fine particulate matter (PM2.5) air quality inferred from a multi-model analysis of
725 meteorological modes. Atmospheric Chemistry and Physics 12, 11329-11337.

PT
726 Tai, A.P.K., Mickley, L.J., Jacob, D.J., Leibensperger, E.M., Zhang, L., Fisher, J.A., Pye,
727 H.O.T., 2012b. Meteorological modes of variability for fine particulate matter

RI
728 (PM2.5) air quality in the United States: implications for PM2.5 sensitivity to
729 climate change. Atmospheric Chemistry and Physics 12, 3131-3145.

SC
730 Wu, D., Liao, B.T., Chen, H.Z., Wu, C., 2014a. Advances in studies of haze weather
731 over Peral River Delta (in Chinese). Climatic and Environmental Research 19,
732

U
248-264.
733 Wu, M., Wu, D., Fan, Q., Wang, B.M., Li, H.W., Fan, S.J., 2013. Observational studies
AN
734 of the meteorological characteristics associated with poor air quality over the Pearl
735 River Delta in China. Atmospheric Chemistry and Physics 13, 10755-10766.
M

736 Wu, M., Wu, D., Fan, S.J., Chen, H.Z., Pan, H.M., 2014b. Research Progress in the
737 study of air pollution and atmospheric boundary layer over the pearl river delta city
D

738 cluster (in Chinese). Advances in Meteorological Science and Technology, 22-28.
TE

739 Wu, Z., Feng, J., Qiao, F., Tan, Z.M., 2016. Fast multidimensional ensemble empirical
740 mode decomposition for the analysis of big spatio-temporal datasets. Philos Trans
741 A Math Phys Eng Sci 374, 20150197.
EP

742 Wu, Z.H., Huang, N.E., 2009. Ensemble Empirical Mode Decomposition: A
743 Noise-Assisted Data Analysis Method. Advances in Adaptive Data Analysis 01,
C

744 1-41.
AC

745 Yadav, R., Beig, G., Jaaffrey, S.N.A., 2014. The linkages of anthropogenic emissions
746 and meteorology in the rapid increase of particulate matter at a foothill city in the
747 Arawali range of India. Atmos Environ 85, 147-151.
748 Zhang, H., Wang, Y., Hu, J., Ying, Q., Hu, X.M., 2015. Relationships between
749 meteorological parameters and criteria air pollutants in three megacities in China.
750 Environ Res 140, 242-254.
751 Zhang, J.P., Zhu, T., Zhang, Q.H., Li, C.C., Shu, H.L., Ying, Y., Dai, Z.P., Wang, X., Liu,
752 X.Y., Liang, A.M., Shen, H.X., Yi, B.Q., 2012. The impact of circulation patterns

45
 ACCEPTED MANUSCRIPT
753 on regional transport pathways and air quality over Beijing and its surroundings.
754 Atmospheric Chemistry and Physics 12, 5031-5053.
755 Zhang, Y., Ding, A., Mao, H., Nie, W., Zhou, D., Liu, L., Huang, X., Fu, C., 2016.
756 Impact of synoptic weather patterns and inter-decadal climate variability on air
757 quality in the North China Plain during 1980–2013. Atmos Environ 124, 119-128.
758 Zhou, D., Ding, A., Mao, H., Fu, C., Wang, T., Chan, L.Y., Ding, K., Zhang, Y., Liu, J.,

PT
759 Lu, A., Hao, N., 2013. Impacts of the East Asian monsoon on lower tropospheric
760 ozone over coastal South China. Environmental Research Letters 8, 044011.

RI
761

U SC
AN
M
D
TE
C EP
AC

46
 ACCEPTED MANUSCRIPT
1 Table 1
2 FUL, WIN, and SUM sampling NSRIMF of modes IMFs 1-7 of air quality (PM10, PM2.5, and O3)
3 at FS and meteorological parameters (U, V, WS, RH, T, Q, and SP) at ECF during 2015−2017*
Sampling Mode PM10 PM2.5 O3 U V WS RH T Q SP
IMF1 0.14 0.15 0.16 0.21 0.16 0.31 0.15 0.04 0.04 0.03
IMF2 0.13 0.12 0.11 0.14 0.13 0.20 0.13 0.05 0.05 0.05
IMF3 0.15 0.13 0.16 0.13 0.12 0.19 0.13 0.06 0.04 0.05
FUL IMF4 0.10 0.12 0.08 0.08 0.05 0.05 0.12 0.04 0.03 0.04

PT
IMF5 0.05 0.05 0.07 0.04 0.02 0.04 0.08 0.01 0.02 0.01
IMF6 0.03 0.03 0.07 0.02 0.07 0.02 0.03 0.49 0.25 0.65
IMF7 0.19 0.22 0.11 0.12 0.18 0.02 0.08 0.06 0.15 0.05
IMF1 0.16 0.18 0.11 0.32 0.19 0.34 0.14 0.04 0.04 0.03

RI
IMF2 0.15 0.14 0.12 0.15 0.15 0.22 0.13 0.05 0.06 0.06
IMF3 0.17 0.15 0.11 0.10 0.13 0.17 0.12 0.07 0.06 0.06
WIN IMF4 0.11 0.15 0.09 0.04 0.05 0.05 0.15 0.04 0.04 0.04
IMF5 0.05 0.05 0.10 0.03 0.02 0.04 0.10 0.01 0.02 0.01

SC
IMF6 0.02 0.02 0.08 0.03 0.04 0.01 0.03 0.31 0.14 0.46
IMF7 0.08 0.11 0.12 0.18 0.15 0.02 0.06 0.05 0.12 0.02
IMF1 0.11 0.10 0.20 0.16 0.13 0.27 0.16 0.03 0.02 0.02
IMF2 0.08 0.07 0.10 0.14 0.11 0.18 0.12 0.03 0.03 0.03

U
IMF3 0.10 0.11 0.20 0.15 0.11 0.20 0.14 0.03 0.02 0.04
SUM IMF4 0.08 0.08 0.07 0.10 0.06 0.05 0.08 0.02 0.02 0.04
AN
IMF5 0.05 0.05 0.05 0.04 0.03 0.05 0.05 0.01 0.02 0.02
IMF6 0.05 0.04 0.05 0.01 0.12 0.02 0.02 0.79 0.40 0.87
IMF7 0.46 0.45 0.09 0.09 0.23 0.02 0.12 0.07 0.19 0.09
4 * Values larger than ≅ 0.0557 are in bold, where = 1 + √5 /2.
M
D
TE
C EP
AC
 ACCEPTED MANUSCRIPT
1 Table 2
2 FUL, WIN, and SUM sampling multi-scale correlation coefficients of PM10 with various
3 meteorological parameters on modes IMFs 1-7 at FS*
Sampling Mode U V WS RH T Q SP
IMF1 0.07 0.19 -0.30 -0.02 0.34 0.27 -0.16
IMF2 0.09 0.26 -0.41 -0.20 0.43 0.18 -0.19
IMF3 -0.02 0.16 -0.46 -0.48 0.51 -0.02 -0.10
FUL IMF4 0.08 -0.13 -0.51 -0.61 0.27 -0.43 0.14

PT
IMF5 0.07 0.04 -0.65 -0.30 0.09 -0.30 0.27
IMF6 -0.10 -0.27 -0.13 -0.22 -0.12 -0.20 0.16
IMF7 -0.71 -0.74 -0.09 -0.85 -0.68 -0.78 0.32
IMF1 0.13 0.29 -0.30 0.03 0.37 0.33 -0.24

RI
IMF2 0.10 0.39 -0.43 -0.13 0.41 0.23 -0.24
IMF3 0.01 0.37 -0.51 -0.46 0.52 0.08 -0.13
WIN IMF4 0.03 -0.03 -0.54 -0.64 0.29 -0.42 0.29
IMF5 0.07 0.28 -0.65 -0.27 0.22 -0.24 0.29

SC
IMF6 0.29 0.02 0.25 -0.01 0.31 0.24 -0.28
IMF7 -0.70 -0.64 -0.54 -0.79 -0.26 -0.64 0.09
IMF1 -0.03 -0.06 -0.31 -0.15 0.28 0.09 0.04
IMF2 0.12 -0.10 -0.40 -0.44 0.49 -0.02 -0.01

U
IMF3 -0.05 -0.36 -0.44 -0.57 0.44 -0.35 -0.04
SUM IMF4 0.16 -0.34 -0.49 -0.50 0.23 -0.48 -0.15
AN
IMF5 0.07 -0.29 -0.66 -0.42 -0.10 -0.45 0.28
IMF6 -0.44 -0.41 -0.43 -0.44 -0.19 -0.37 0.24
IMF7 -0.52 -0.73 -0.42 -0.78 -0.68 -0.61 0.29
4 * Coefficients with abstract value larger than 0.0993, 0.141, and 0.140 are significant on 0.001 level for FUL, WIN, and SUM,
M

5 respectively; and coefficients with abstract value larger than 0.3 (0.35) are highlighted in bold for FUL (WIN & SUM).
D
TE
C EP
AC
 ACCEPTED MANUSCRIPT
1 Table 3
2 FUL, WIN, and SUM sampling multi-scale correlation coefficients of PM2.5 with various
3 meteorological parameters on modes IMFs 1-7 at FS*
Sampling Mode U V WS RH T Q SP
IMF1 0.05 0.20 -0.29 0.02 0.28 0.25 -0.14
IMF2 0.09 0.19 -0.41 -0.18 0.35 0.13 -0.14
IMF3 -0.04 0.12 -0.49 -0.43 0.46 0.00 -0.09
FUL IMF4 0.06 -0.10 -0.46 -0.62 0.24 -0.43 0.12

PT
IMF5 0.08 0.05 -0.64 -0.27 0.06 -0.25 0.25
IMF6 -0.12 -0.37 -0.18 -0.43 -0.21 -0.32 0.28
IMF7 -0.77 -0.81 -0.14 -0.86 -0.71 -0.78 0.41
IMF1 0.10 0.28 -0.29 0.07 0.29 0.30 -0.22

RI
IMF2 0.10 0.30 -0.43 -0.10 0.32 0.18 -0.18
IMF3 0.02 0.34 -0.54 -0.37 0.47 0.10 -0.10
WIN IMF4 0.02 0.01 -0.48 -0.66 0.26 -0.42 0.27
IMF5 0.11 0.34 -0.64 -0.22 0.18 -0.14 0.26

SC
IMF6 0.23 -0.19 0.05 -0.34 0.10 -0.03 -0.03
IMF7 -0.77 -0.72 -0.52 -0.75 -0.39 -0.67 0.20
IMF1 -0.05 -0.03 -0.31 -0.10 0.25 0.12 0.04
IMF2 0.11 -0.11 -0.40 -0.41 0.46 -0.04 -0.03

U
IMF3 -0.11 -0.38 -0.46 -0.56 0.43 -0.35 -0.07
SUM IMF4 0.14 -0.39 -0.49 -0.46 0.19 -0.46 -0.19
AN
IMF5 0.07 -0.34 -0.66 -0.40 -0.15 -0.49 0.27
IMF6 -0.47 -0.43 -0.41 -0.48 -0.17 -0.42 0.30
IMF7 -0.61 -0.83 -0.47 -0.86 -0.74 -0.69 0.43
4 * Coefficients with abstract value larger than 0.0993, 0.141, and 0.140 are significant on 0.001 level for FUL, WIN, and SUM,
M

5 respectively; and coefficients with abstract value larger than 0.3 (0.35) are highlighted in bold for FUL (WIN & SUM).
D
TE
C EP
AC
 ACCEPTED MANUSCRIPT
1 Table 4
2 FUL, WIN, and SUM sampling multi-scale correlation coefficients of O3, respectively, with
3 various meteorological parameters on modes IMFs 1-7 at FS*
Sampling Mode U V WS RH T Q SP
IMF1 -0.01 0.02 -0.26 -0.41 0.39 -0.10 0.08
IMF2 -0.05 -0.03 -0.35 -0.53 0.40 -0.12 0.05
IMF3 -0.15 -0.16 -0.46 -0.67 0.40 -0.25 -0.01
FUL IMF4 0.07 0.06 -0.47 -0.65 0.48 -0.33 0.17

PT
IMF5 -0.14 -0.31 -0.46 -0.55 0.28 -0.47 0.14
IMF6 -0.23 -0.12 -0.30 0.00 0.35 0.23 -0.22
IMF7 0.60 0.36 -0.30 0.42 0.70 0.79 -0.14
IMF1 0.01 0.02 -0.18 -0.51 0.38 -0.24 0.05

RI
IMF2 -0.11 0.07 -0.26 -0.55 0.29 -0.20 0.03
IMF3 -0.01 0.17 -0.36 -0.67 0.40 -0.21 0.02
WIN IMF4 -0.03 0.22 -0.53 -0.68 0.53 -0.29 0.22
IMF5 0.10 -0.15 -0.34 -0.65 0.37 -0.56 0.39

SC
IMF6 0.13 -0.01 -0.14 0.19 0.14 0.03 0.02
IMF7 0.66 0.42 0.17 0.23 0.87 0.74 -0.21
IMF1 -0.02 0.01 -0.35 -0.37 0.46 0.01 0.11
IMF2 -0.02 -0.15 -0.45 -0.52 0.61 -0.01 0.07

U
IMF3 -0.20 -0.46 -0.55 -0.74 0.58 -0.37 -0.04
SUM IMF4 0.11 -0.14 -0.42 -0.68 0.43 -0.41 0.11
AN
IMF5 -0.32 -0.51 -0.61 -0.36 0.14 -0.31 -0.06
IMF6 -0.64 -0.48 -0.45 -0.51 -0.06 -0.40 0.38
IMF7 0.12 -0.31 -0.16 -0.29 -0.13 0.15 0.22
4 * Coefficients with abstract value larger than 0.0993, 0.141, and 0.140 are significant on 0.001 level for FUL, WIN, and SUM,
M

5 respectively; and coefficients with abstract value larger than 0.3 (0.35) are highlighted in bold for FUL (WIN & SUM).
D
TE
C EP
AC
 ACCEPTED MANUSCRIPT
1 Table 5
2 FUL, WIN, and SUM sampling whole-scale correlation coefficients between PM10, PM2.5, and
3 O3, respectively, and various meteorological parameters at FS, GZ, and DG*
FUL WIN SUM
PM10 PM2.5 O3 PM10 PM2.5 O3 PM10 PM2.5 O3
U -0.12 -0.14 -0.01 -0.02 -0.04 0.02 -0.11 -0.13 -0.15
V -0.09 -0.12 -0.01 0.21 0.17 0.10 -0.42 -0.44 -0.34
WS -0.44 -0.45 -0.45 -0.52 -0.53 -0.36 -0.45 -0.45 -0.56

PT
FS T -0.06 -0.11 0.53 0.33 0.24 0.45 0.12 0.11 0.45
RH -0.45 -0.42 -0.49 -0.35 -0.32 -0.67 -0.53 -0.50 -0.63
Q -0.33 -0.35 0.14 -0.02 -0.06 -0.18 -0.46 -0.45 -0.27
SP 0.28 0.30 -0.24 -0.01 0.04 0.02 0.22 0.18 0.06

RI
U 0.07 0.03 0.14 0.22 0.20 0.07 0.05 0.01 0.08
V 0.01 -0.03 0.07 0.30 0.27 0.08 -0.28 -0.32 -0.15
WS -0.43 -0.41 -0.37 -0.50 -0.49 -0.26 -0.44 -0.44 -0.49
GZ T -0.01 -0.09 0.56 0.39 0.30 0.46 0.22 0.18 0.59

SC
RH -0.43 -0.38 -0.54 -0.37 -0.32 -0.71 -0.51 -0.45 -0.63
Q -0.28 -0.31 0.18 0.05 0.01 -0.19 -0.35 -0.33 -0.14
SP 0.25 0.28 -0.25 -0.05 -0.01 0.07 0.22 0.16 0.00
U -0.14 -0.18 0.07 -0.04 -0.05 0.01 -0.09 -0.13 -0.02

U
V -0.11 -0.16 0.05 0.20 0.16 0.16 -0.41 -0.44 -0.26
WS -0.45 -0.44 -0.48 -0.52 -0.52 -0.43 -0.46 -0.46 -0.56
AN
DG T -0.09 -0.15 0.50 0.28 0.24 0.45 0.17 0.11 0.51
RH -0.49 -0.48 -0.48 -0.39 -0.36 -0.64 -0.53 -0.52 -0.61
Q -0.38 -0.41 0.10 -0.10 -0.10 -0.17 -0.44 -0.48 -0.22
SP 0.30 0.34 -0.18 0.03 0.06 0.03 0.21 0.22 0.06
4
M

* Coefficients with abstract value larger than 0.0993, 0.141, 0.140 are significant on 0.001 level for FUL, WIN, and SUM,

5 respectively; and coefficients with abstract value larger than 0.3 (0.35) are highlighted in bold for FUL (WIN & SUM).
D
TE
C EP
AC
 ACCEPTED MANUSCRIPT
1

PT
RI
U SC
AN
M

2
D

3 Fig. 1. Topography map of the Guangdong−Hong Kong−Macao Greater Bay Area (GHMGBA,
TE

4 the pinked area) and its surroundings (see the blue box in the superimposed administrative map
5 for reference). The GHMGBA includes 11 cities: Guangzhou (GZ), Zhaoqing (ZQ), Foshan (FS),
6 Dongguan (DG), Huizhou (HZ), Jiangmen (JM), Zhongshan (ZS), Shenzhen (SZ), Zhuhai (ZH),
EP

7 Hong Kong (HK), and Macao (MC). The four dashed boxes centered at (22.807°N, 113.203°E),
8 (23.509°N, 113.203°E), (22.807°N, 113.906°E), and (23.5 09°N, 113.906°E), respectively, show
C

9 the closest model grids where the ERA-Interim reanalysis data (in a resolution of 0.75°×0.75°)
AC

10 are extracted as the collocated meteorological data for the air quality data at cities FS, GZ, DG,
11 and HZ, respectively; and they are denoted as ECF, ECG, ECD, and ECH, respectively. The
12 point (23.158°N, 113.555°E) marked by a red star is the common vertex of the four model grids
13 and for convenience it is selected as the representative point to anchor the geographical location
14 for GHMGBA in the larger scale administrative map. The green circles are the loci of
15 observation stations.
16
 ACCEPTED MANUSCRIPT

PT
RI
U SC
AN
M
D
TE
C EP

1
AC

2
3 Fig. 2. Daily time series of EEMD intrinsic mode functions (IMFs) (unit: µg/m3) of PM10 during
4 2015−2017 at FS, GZ, and DG. The corresponding square root of variance (unit: µg/m3) and
5 NSRIMF are listed at top of plots for IMFs 1-10. The square root of variance (unit: µg/m3) and
6 mean (unit: µg/m3) for raw time series are listed at bottom of plot for IMF10.
7
 ACCEPTED MANUSCRIPT
1
2

PT
RI
SC
U
AN
M
D
TE
C EP
AC

3
4 Fig. 3. The same as Fig.2 but for PM2.5.
 ACCEPTED MANUSCRIPT
1

PT
RI
U SC
AN
M
D
TE
C EP
AC

3 Fig. 4. The same as Fig.2 but for O3.

 ACCEPTED MANUSCRIPT
1 (a)
2
3
4
5
6

PT
7
8

RI
9
10

SC
11
12

U
13
14
AN
15
16
M

17
18
D

19 (b)
TE

20
21
22
EP

23
24
C

25
AC

26
27
28
29
30
31 Fig. 5. Highlighted daily time series of IMFs of PM2.5 (unit: µg/m3) at FS, GZ, and DG: (a) IMFs
32 1-7 from October 2016 to March 2017; (b) IMFs 1 and 2 from 1 December 1 2016 to 1 February
33 2017. The X axis in (b) is labelled in weekly cycle with “1” represents “Monday” and so on.
 ACCEPTED MANUSCRIPT
1
2 (a) U (b) V

PT
RI
U SC
3
AN
4 (c) WS (d) RH
M
D
TE
C EP
AC

5
6 Fig. 6. Daily time series of EEMD intrinsic mode functions (IMFs) of (a) U (unit: m/s), (b) V
7 (unit: m/s), (c) WS (unit: m/s), (d) RH (unit: %), (e) T (unit: °C), (f) Q (unit: mg/kg3), and (g) SP
8 (unit: hPa) during 2015−2017 at model grids ECF, ECG, and ECD representing for FS, GZ, and
9 DG, respectively. The corresponding square root of variance and NSRIMF are listed at top of plots
10 for IMFs 1-10. The square root of variance and mean for raw time series are listed at bottom of
11 plot for IMF10.
 ACCEPTED MANUSCRIPT
12 (e) T (f) Q

PT
RI
U SC
13
14 (g) SP
AN
M
D
TE
C EP
AC

15
16 Fig. 6. (Continued)
17
 ACCEPTED MANUSCRIPT

Highlights:

We offered a method for combined multi-scale air quality-meteorology

correlations.

PT
We confirmed the different scale sensitivity of PM to meteorology in

RI
GHMGBA.

SC
Synoptic sensitivity should be carefully extended to climate change impacts.

U
AN
M
D
TE
C EP
AC