Influence of meteorological conditions on PM 2.5 and PM 2.5−10 concentrations during the monsoon season in Hanoi, Vietnam

Atmospheric Environment 36 (2002) 3473–3484

Influence of meteorological conditions on PM2.5 and PM2.5�10

concentrations during the monsoon season in Hanoi, Vietnam

P.D. Hiena,*, V.T. Bacb, H.C. Thamb, D.D. Nhanb, L.D. Vinhc

aVietnam Atomic Energy Agency, 59 Ly Thuong Kiet, Hanoi, Viet Namb Institute for Nuclear Science and Technology, Cau giay, Hanoi 5T-160, Viet Nam

cUpper Air Meteorology Station, Hanoi, Viet Nam

Received 21 January 2002; accepted 10 April 2002

Abstract

Twenty-four hour samples of air particulate matter with aerodynamic diameters from 2 to 10 mm (PM10) and

o2.5mm (PM2.5) were collected in Hanoi throughout 1 year since August 1998. The air sampler was located in a

meteorological garden where routine surface observations and upper air radiosoundings were conducted. Very high

PM2.5 and PM2.5�10 concentrations were observed in conjunction with the occurrence of nocturnal radiation inversions

from October to December and subsidence temperature inversions (STI) from January to March. In the first case, the

PM2.5�10 fraction was much enhanced and particulate pollution was significantly higher at night than in daytime.

During the occurence of STIs particulate mass was almost evenly distributed among the two fractions and no significant

diurnal variations in concentrations were observed. In summer (May–September) particulate pollution was much lower

than in winter.

The multiple regression of 24-h particulate concentrations against meteorological parameters for both the winter and

summer monsoon periods shows that the most important determinants of PM2.5 are wind speed and air temperature,

while rainfall and relative humidity largely control the daily variations of PM2.5�10, indicating the high abundance of

soil dust in this fraction. As to turbulence parameters, among the determinants of 24-h particulate concentrations are

the vertical gradients of potential temperature and wind speed recorded at 06.30 and 18.30, respectively. Meteorological

parameters could explain from 60% to 74% of the day-to-day variations of particulate concentrations.r 2002 Elsevier

Science Ltd. All rights reserved.

Keywords: Coarse and fine particulate matter; Temperature inversions; Meteorology; Diurnal variations; Regression analysis

1. Introduction

The rapid economic development since the introduc-

tion of a market orientation reform in the late 1980s has

dramatically changed the face of the 3.5 million

Vietnam’s capital, Hanoi. In the meantime, uncontrolled

growth of construction works, traffic and small manu-

facturing activities has resulted in an increasing number

of air pollution sources. Dusty atmosphere is visible,

especially during the dry winter season.

A systematic air particulate pollution study has begun

in Hanoi since 1998. The 24-h PM10 (particulate matter

with aerodynamic diameters o10mm) concentration

varies substantially, from as low as 10 mgm�3 in the

rainy monsoon months (June–August) to over

300 mgm�3 in the winter (October–March). The devel-

opment of appropriate pollution abatement measures

requires a thorough understanding of the nature of

major emission sources and atmospheric conditions

governing the variations of air particulate concentra-

tions, particularly those relevant to pollution episodes.

The variations of atmospheric conditions in Hanoi

are governed by large-scale air circulations which are*Corresponding author. Fax: +84-4-9424133.

E-mail address: [email protected] (P.D. Hien).

1352-2310/02/$ - see front matter r 2002 Elsevier Science Ltd. All rights reserved.

PII: S 1 3 5 2 - 2 3 1 0 ( 0 2 ) 0 0 2 9 5 - 9

driving the monsoon regime in East and Southeast Asia.

Relatively few investigations have been conducted on

the relationships between air pollution and the monsoon

regime in this region. These include the works of Chung

et al. (1999), Cheng and Lam (2000) and Wang et al.

(2001) on the responses of pollutant gases to synoptic

conditions in Hong Kong; Hien et al. (1999) on the

seasonal variations of the impact of major emission

sources on a receptor site in Ho Chi Minh City; and

Pochanart et al. (2001) on the long-range transport of

O3 under Asian monsoon regime.

To investigate the impact of monsoon conditions on

air pollution in Hanoi, a year-round monitoring study

was carried out for air particulate matter with aero-

dynamic diameters o2.5mm (PM2.5) and from 2.5 to

10mm (PM2.5�10). Based on daily synoptic weather maps

and meteorological data provided by surface observa-

tions and upper air radiosoundings, atmospheric condi-

tions and air masses relevant to pollution episodes could

be identified. Multiple regression analysis was applied to

derive statistical relationships between air particulate

pollution and meteorological parameters for the winter

and summer monsoon periods. For characterisation of

major emission sources, chemical compositions of

particulate matter were determined by instrumental

neutron activation analysis and ion chromatography

methods. This topic, however, will be reported elsewhere.

2. Meteorological conditions

Hanoi is located in the Red River delta in North

Vietnam (21.021N, 105.851E), about 100 km west of the

South China Sea (Figs. 1a and b). Being influenced by

the Southeast Asia monsoon regime, the climate is

basically tropical and humid. There are two monsoon

seasons, i.e. the northeast monsoon in winter and the

southeast monsoon in summer.

During winter, atmospheric conditions are alternately

affected by air masses from the Highs over Siberia and

East China Sea (Fig. 1a) (Toan and Dac, 1993). Con-

tinental air from the Siberia High yields low temperature

and stable atmospheric conditions. As to air humidity, it

depends on the trajectory (continental or marine) of air

masses from the source origin to North Vietnam.

From October to December, northerly to north-

easterly flow coming from the inland of China brings

dry and cold air. Nocturnal radiation inversions (NRI)

usually occur on clear and calm nights favouring the

accumulation of dust and air particles in the layer just

above the ground. Conversely, from January to March/

April, with the Siberia High system frequently shifted to

the East, air masses have to travel a long way over the

Pacific Ocean before reaching North Vietnam via the

Gulf of Tonkin. Northeasterly flow of moist-laden air

results in smog, low stratus cloudiness and drizzle. Poor

atmospheric dispersion conditions are associated with

anticyclone’s subsidence temperature inversions (STI) in

the near ground layer.

Maritime air from the High over East China Sea

prevailing during the transition period between cycles of

continental air intrusion brings warm, humid and better

dispersion conditions. Toward the winter end, north-

easterly flow of continental air is getting less frequent

and southeasterly flow of maritime air becomes domi-

nant in March–April.

In summer, high-pressure systems are expanded

northward from the Southern Hemisphere (Fig. 1b).

Atmospheric conditions in North Vietnam are governed

by air masses coming from the Highs over Indian Ocean

and the subtropical High over the South China Sea. The

two systems bring moist air and monsoon rains.

However, heavy rains mainly occur in July and August

in association with tropical depressions, highly unstable

conditions around the Intertropical Convergence Zone

and cyclones, which frequently appear in the South

China Sea and move westward striking the West Pacific

coast. The mean annual rainfall in Hanoi is 1800mm,

80% of which are recorded from May to September.

3. Sampling

Coarse (PM2.5�10) and fine (PM2.5) particulates were

collected on two separate 47mm diameter Nuclepore

polycarbonate filters using a Gent stack filter unit (SFU)

following the instruction manual by Maenhaut et al.

(1992). The air sampler head was mounted at 1.6m

above the ground. The flow rate was kept between 18

and 14 lmin�1. To mitigate filter clogging an on–off

regime was set for the SFU.

The air sampler is located in the Hanoi meteorological

garden. The expansion of the city over the last two

decades had completely transformed this originally

semi-rural area into a new urban residential setting.

Surface observation parameters were recorded at every

3-hour interval. Balloon radiosondes using a RS80-15G

DigiCORA system were launched twice per day at 06.30

(23:30 GMT) and 18.30 (11:30 GMT), providing

information on vertical profiles of air temperature,

relative humidity (RH), dew point, wind speed (WS)

and wind direction. Balloon ascent rate was approxi-

mately 5m s�1 with data retrieved every 2 s, yielding a

vertical resolution of about 10m.

From September 1998 to August 1999, 24-h samples

were taken continuously for the study of the pollution–

meteorology relationships. Daytime and nighttime

samples were taken on more than 100 days in winter

1998–99, providing information on diurnal variations of

particulate pollution. Since September 1999, 24-h

samples were taken twice a week. These samples are

not included in this study.

P.D. Hien et al. / Atmospheric Environment 36 (2002) 3473–34843474

Nuclepore filters were pre- and post-weighed to

determine the gravimetric masses of collected materials

using a Mettler balance placed in a dedicated room with

controlled temperature and humidity. The filters were

acclimatised in the room condition for 3–4 days prior to

weighing. The readability of the balance is 1mg. A 210Po

electrostatic charge eliminator was used to neutralise

charges accumulated on the filters before weighing.

4. Meteorological data treatment

4.1. Surface observation parameters

Surface observation meteorological parameters re-

corded at every 3-h interval include temperature,

pressure, RH, WS, wind direction and sunshine dura-

tion. Rainfall was recorded as a 24-h total value. The

seasonal averages of meteorological parameters are

given in Table 1. For simplicity, the 1 October and 1

May are assigned to the beginning of the winter and

summer, respectively. For the summer, only a period

from May to July 1999 was analysed. The two summer

months of 1998 (August and September) were not

included in the regression analysis. WS and RH show

little seasonal variations, while temperature and rainfall

were much higher in summer than in winter.

4.2. Radiosoundings

Vertical profiles of temperature and WS in the

boundary layer were analysed to derive atmospheric

Fig. 1. High-pressure systems in East Asia in winter (a) and summer (b).

P.D. Hien et al. / Atmospheric Environment 36 (2002) 3473–3484 3475

turbulence parameters. For this purpose, the actual

temperature (T) was converted into potential tempera-

ture (y), thus enabling us to easily distinguish three typesof buoyancy according to the gradient Dy=Dz; namelystable (Dy=Dz > 0), neutral (Dy=Dz ¼ 0) and unstable

(Dy=Dzo0). It was found that high particulate levels

were recorded in conjunction with the occurrence of

NRIs in the first winter period (October–December) and

STIs in the second winter period (January–March). The

criteria adopted for these temperature inversions were

taken as in Heffter (1983), Marsik et al. (1995) and

Calori and Carmichael (1999), namely

Dy=DzX0:005 Km�1;

yT � yBX2K; ð1Þ

where Dy=Dz is the potential temperature gradient in the

inversion layer and yT and yB refer to the potential

temperatures at the top and the base of the inversion

layer.

NRI appears at dawn, extending from the ground to

about 100–150m. Over the night the NRI layer moves

up to a higher and higher elevation, reaching a few

hundred meters in the early morning (Fig. 2a), and

presumably disappears thereafter as the sunlight warms

the ground. STI usually persists for days during which it

can be observed in both the morning and evening

soundings with the inversion layer height varying within

several hundred meters above the ground (Fig. 2c).

The WS profiles are illustrated in Figs. 2b and d. WS

increases with height reaching the first maximum at

some elevation below 1000m. No significant relation-

ship exists between the height of this maximum and the

top or bottom of temperature inversions. The profiles of

RH and dew point have good correlations with that of

potential temperature. For this reason upper air data on

these parameters were not included in the regression

analysis.

4.3. Atmospheric turbulence

To characterise atmospheric turbulence, the mixing

depth is usually derived from the vertical profile of

temperature. A literature survey, however, did not find

an overall acceptable definition and criteria for the

practical determination of the mixing depth that could

encompass a wide range of atmospheric stability and a

variety of its governing physical processes (Beyrich,

1997). For daytime convective conditions the mixing

depth was estimated by using a temperature profile

intersection scheme developed by Holzworth (1967). For

nighttime stable conditions, several profile-derived

heights have been proposed, e.g. the height of the NRI

or the first WS maximum. (Mahrt et al., 1982; Baxter,

1991; Berman et al., 1999; Lena and Desiato, 1998;

Seibert et al., 2000). However, several researchers e.g.

Aron (1983), Lena and Desiato (1998) and Seibert et al.

(2000) noted that the mixing depth estimated by the

above methods in general poorly correlate with air

pollutant concentrations.

In our work, the gradients of potential temperature

Dy=Dz and WS Du=Dz between two elevations z1 and z2in the surface layer were used for characterising atmo-

spheric turbulence. Such a simple empirical method is

Table 1

Summary statistics of 24-h average particulate mass concentrations and meteorological parameters

Notation October 98–March 99 May 99–July 99

Mean S.D. Mean S.D.

Surface observations

Coarse mass (mgm�3) CO 69.8 52.3 27.6 15.0

Fine mass (mgm�3) FI 51.5 28.5 18.9 8.0

Wind speed (m s�1) WS 1.6 0.7 1.8 0.6

Air temperature (1C) T 21.5 3.5 28.5 2.3

Air pressure (mb) P 1014.6 5.5 1004.2 3.9

Relative humidity (%) RH 74.5 10.3 78.9 5.7

Sunshiness (h) SUN 3.4 3.5 4.9 3.2

Rainfall (mm) RAIN 0.9 3.8 8.0 20.3

Upper air observationsa

(Dy=Dz), 06.30 ðDy=DzÞm 0.61 0.39 0.55 0.26

(Du=Dz), 06.30 ðDu=DzÞm 1.06 0.64 0.92 0.52

(Dy=Dz), 18.30 ðDy=DzÞe 0.30 0.20 0.32 0.32

(Du=Dz), 18.30 ðDu=DzÞe 0.86 0.54 1.10 0.58

aThe values in (Km�1) for Dy=Dz and (s�1) for Du=Dz are multiplied by 100.

P.D. Hien et al. / Atmospheric Environment 36 (2002) 3473–34843476

based on the physical consideration that the gradient

Richardson number is expressed in terms of these

gradients, namely

Ri ¼ ðg=TÞðDy=DzÞðDu=DzÞ�2; ð2Þ

where g is the constant of gravity and Dz ¼ z1 � z2:Other turbulence parameters, such as the Monin–

Obukhov length L; the sensible heat flux H and the

friction velocity u* ;could also be estimated from these

gradients (see e.g. Berkowicz and Prahm, 1982) based on

Monin–Obukhov similarity theory for the surface layer

(Monin and Obukhov, 1954). The gradients were

computed for the layer between z1 ¼ 5m (the lowest

elevation recorded in radiosoundings) and z2 ¼ 500m.

By varying z2 from 200 to 800m, we found that the

elevation of 500m had yielded highest correlations

between the gradients and the observed 24-h PM2.5�10

and PM2.5 concentrations. Both the morning and

evening soundings were analysed to provide four

turbulence parameters, the seasonal variations of which

are given in Table 1.

5. Experimental results

5.1. 24-h particulate concentrations

Figs. 3a–c display the time series of 24-h PM2.5 and

PM10 concentrations. The annual mean concentrations

(7standard errors) from August 1998 to July 1999 were

(87.173.1) mgm�3 for PM10 and (36.171.3) mgm�3 for

PM2.5. The PM10 US NAAQS (150mgm�3) was

exceeded on 52 days. Also, PM2.5 concentrations exceed

50 mgm�3 on 77 days. Most of these pollution episodes

were observed from October 1998 to February 1999. The

time series of 24-h PM2.5�10 and PM2.5 concentrations

and some relevant meteorological parameters for that

period are displayed in Figs. 4a and h.

5.2. Winter pollution episodes

From October to early January, most pollution

episodes were associated with the occurrence of NRIs,

which are marked by full height columns in Figs. 4a–h.

0

100

200

300

400

500

600

290 292 294 296 298

theta, K

heig

ht, m

11/13/98 6:3011/12/98 18:30

(a) (b)

(c) (d)

0 1 2 3

wind speed, m s-1

0

200

400

600

287 290 293 296

theta, K

heig

ht, m

01/22 18:30

01/22 06:30

0

200

400

600

0 2 4 6 8 10 12 14

w ind speed, m s-1

Fig. 2. Vertical profiles of potential temperature (left) and wind speed (right). NRI (a,b), STI (c,d).

P.D. Hien et al. / Atmospheric Environment 36 (2002) 3473–3484 3477

NRI usually appeared several days after the passage of a

cold front, which can be recognised by a peak in the WS

chart (Fig. 4f) and a sudden drop of temperature

(not shown). The weather conditions during these

events were light wind (Fig. 4f), long sunshine duration

(Fig. 4g) and low RH (Fig. 4h). From late December to

March, the prevailing humid and overcast conditions

did not favour NRIs. Instead, most pollution episodes

were observed in conjunction with STIs located in

the layer below, say, 500m, which are marked by half

height columns in Figs. 4a–h. Temperature inversions

above this level show little influence on particulate

concentrations.

Figs. 4b–d show the diurnal change (the nighttime-to-

daytime concentration ratio) of particulate pollution in

response to different dispersion conditions. The inter-

play of emissions and atmospheric dispersion processes

operating on a diurnal cycle results in generally lower

particulate loadings at night in non-episode and STI

cases. The NRI, however, entails much higher loadings

at night, especially for coarse particulates. As a result,

the diurnal change was much larger for coarse particu-

lates (Fig. 4b) and the coarse mode was significantly

enhanced (Fig. 4e). These trends are summarised in

Table 2, where the averages of particulate concentra-

tions, diurnal changes and coarse mode enhancement

are calculated for the three turbulence categories. The

average WS and RH are also given as relevant surface

meteorological parameters.

To gain insight into the above findings, Figs. 5a–e

provide an example that details the diurnal variations

of particulate pollution during a period covering the

NRI episodes on 13 and 14 November followed by a

non-inversion case on 15 November. Potential tempera-

ture profiles are shown in Fig. 5a. The pattern of

surface WS (Fig. 5b) shows calm to light wind at

night on 13 and 14 November when NRIs occurred

(Fig. 5a). Particulate concentrations markedly

increased after 18.00 (Figs. 5c and d), which is about

evening rush hours (from 17.30 to 19.00) and time for

street sweeping (from 18.30 to 20.00). Significant

amounts of road dust and car exhaust particulates were

trapped in a shallow layer just above the ground,

causing the surges of particulate loadings (Figs. 5c and

d) and a significant enhancement of the coarse mode

until midnight or later (Fig. 5e). Particulate loadings

dropped in daytime and returned to a non-inversion

level on 15 November. As dry soil favours the

resuspension of soil dust by traffic and street sweeping,

both the coarse mode enhancement and the diurnal

change of coarse particulates tend to inversely correlate

with RH, as seen in Table 2.

5.3. Summer

The mean concentrations of PM2.5 and PM2.5�10

were about 2.5 times lower in summer than in

winter (Table 1). The daily variations of particulate

concentrations in summer were also lower. The

lowest monthly averaged concentrations were recorded

in July–August, during which prolonged rains

occurred. Details of daily variations of particulate

concentrations and weather characteristics in summer

are not shown.

PM10

0

50

100

150

0

300

200

100

400

7/23/98 9/6/98 10/21/98 12/5/98 1/19/99 3/5/99 4/19/99 6/3/99 7/18/99Date

7/23/98 9/6/98 10/21/98 12/5/98 1/19/99 3/5/99 4/19/99 6/3/99 7/18/99Date

Co

nce

ntr

atio

n (µ

g/m

3)C

on

cen

trat

ion

(µg

/m3)

PM2.5

Fig. 3. Time series of PM2.5 and PM10. The bold horizontal line in the PM10 chart represents the US NAAQS standard (150mgm�3).

P.D. Hien et al. / Atmospheric Environment 36 (2002) 3473–34843478

Fig. 4. Time series of particulate concentrations during a winter period from October 1998 to February 1999 (a: coarse, c: fine),

nighttime-to-daytime concentration ratios (b: coarse, d: fine), coarse-to-fine concentration ratio (e), surface wind speed (f), sunshine

duration (g) and RH (h). The occurrence of NRIs and STIs is marked by full- and half-height columns, respectively.

P.D. Hien et al. / Atmospheric Environment 36 (2002) 3473–3484 3479

Table 2

Particulate pollution in NRI, STI and non-inversion categories in the two winter periods

1st winter period (October–December) 2nd winter period (January–March)

NRI (43 days) Non-inversion (43 days) STI (37 days) Non-inversion (89 days)

RH (%) 65.570.7 75.771.7 77.571.3 78.671.2

WS (m s�1) 1.170.1 2.070.1 1.270.1 2.070.1

Coarse (mgm�3) 13978 3973 7975 4572

Fine (mgm�3) 6974 3472 8074 4472

Coarse mode enhancement 2.170.1 1.1870.08 1.0370.06 1.0570.05

Diurnal change, coarse 3.270.3 0.9070.06 1.0170.06 0.7270.02

Diurnal change, fine 1.570.1 0.8870.03 1.0470.06 0.8970.04

The values are given as means7standard errors.

0

100

200

300

11/13 6:00 11/13 18:00 11/14 6:00 11/14 18:00 11/15 6:00 11/15 18:00 11/16 6:00

0200400

800600

11/13 6:00 11/13 18:00 11/14 6:00 11/14 18:00 11/15 6:00 11/15 18:00 11/16 6:00

Fin

e, u

g m

-3C

oars

e, u

g m

-3

0

1

2

3

11/13 6:00 11/13 18:00 11/14 6:00 11/14 18:00 11/15 6:00 11/15 18:00 11/16 6:00

w. s

p.,

m s

-1

0

100

200

300

400

500

291 293

Hei

gh

t, m

6:3011/13

(a)

(b)

(c)

(d)

(e)

295 298

18:3011/13

290 292 294

6:3011/14

293 295

6:3011/15

295 298 301

18:3011/15

295 298 301

6:3011/16

Theta (K)

0123456

11/13 6:00 11/13 18:00 11/14 6:00 11/14 18:00 11/15 6:00 11/15 18:00 11/16 6:00

Co

ars

e/f

ine

296 298

18:3014/11

Fig. 5. A scenario of a NRI high pollution episode on 13 and 14 November followed by a non-inversion case on 15 November. (a)

Vertical profiles of potential temperature, (b) surface wind speed, (c) coarse concentration, (d) fine concentration, (e) coarse-to-fine

concentration ratio.

P.D. Hien et al. / Atmospheric Environment 36 (2002) 3473–34843480

6. Relationships between particulate pollution and

meteorological parameters

6.1. Multiple regression analysis method

Meteorological parameters governing the day-to-day

variations of PM2.5 and PM2.5�10 for the two monsoon

periods were studied by multiple regression analysis. In

the regression analysis dependent and independent

variables were constructed by logarithmically transform-

ing observation values. The logarithm transformation

was necessary because atmospheric dispersion equations

suggest that the relationship between particulate con-

centration (C) and meteorological parameters (Pi) is

multiplicative rather than additive (Elsom and Chand-

ler, 1978), i.e.,

C ¼ kY

i

ðPiÞai : ð3Þ

The exponent ai measures the response of particulate

concentration C to the rate of change in meteor-

ological parameter Pi; other meteorological parametersbeing constant. Eq. (3) leads to the multi-linear

regression model for logarithmically transformed

variables, i.e.

ln C ¼ ln k þX

i

ailn Pi þ e; ð4Þ

where ln k; ai and e are the intercept, regression

coefficients and the error term, respectively. Note that

the weather condition has a delayed impact on

particulate pollution, e.g. the today’s drop in fugitive

soil dust is a consequence of the previous day’s rain

that made the surface soil humid. To take into account

this effect, the previous day’s meteorological parameters

are also included in the data set as independent

variables.

Thus, 20 independent variables were used in the

regression analysis including 2� 6 surface observation

and 2� 4 turbulence parameters. To reveal the pre-

dictors (determinants) of 24-h particulate concentrations

among these descriptors, a stepwise multiple regression

method (SPSS, version 7.5) was applied and a statistical

significance p ¼ 0:01 was set for the regression coeffi-

cients.

6.2. Regression models

The determinants Pi; coefficients ai and the intercept

ln k of the regression models (5) for the winter and

summer periods are presented in Table 3. The determi-

nant acronyms are listed in Table 1. In addition, a suffix

p is used to denote the previous day’s meteorology. The

standardised regression coefficients b are shown in

column 8. The determinants of 24-h particulate con-

centrations are listed in decreasing order of relative

importance according to the standardised regression

coefficients b: The statistical significance of the determi-nants is po0:01:Among surface observation parameters, rainfall

(RAIN) and RH largely control the daily variations of

PM2.5�10, while WS and air temperature (T) are

most important determinants of PM2.5. As can be

expected, these parameters are inversely related to

particulate concentrations (a; bo0). Rain and moisture

remove atmospheric particulates and diminish the

amount of resuspended soil dust by making the

soil humid. The governing role of rain and humidity

for PM2.5�10 confirms the high abundance of resus-

pended soil dust in the coarse fraction, as suggested in

Section 4.5.

WS plays a leading role in cleansing fine particulates.

In winter both the prompt (WS) and delayed response

(WSp) of WS are observed making it a most important

meteorological parameter in explaining the variations in

fine particulate concentrations. For PM2.5�10, the

relationships with WS are not as significant as with

rainfall. Moreover, PM2.5�10 positively correlates with

WS in summer. Harrison et al. (1997) found a positive

relationship between coarse particulates and WS for the

summer in Birmingham (UK), pointing to the evidence

of resuspension of soil particles.

Air temperature controls PM2.5 but PM2.5�10 and

more significant in winter than in summer. In early

works on air pollution in European cities, the inverse

relationship between air temperature and air pollutant

concentration was attributed to the fuel burning for

space heating of buildings in winter (see Elsom and

Chandler, 1978; and references therein). Space heating is

not a practice in our tropical conditions. The control of

air temperature on PM2.5 in our case reflects a trend that

more favourable atmospheric dispersion conditions are

observed under warm air than cold air masses. In winter,

for example, dispersion conditions are better under

warm maritime air than cold continental air.

The gradients of potential temperature and WS also

contribute to explaining the variations of particulate

concentrations. As expected, particulate concentrations

have positive relationships with the gradient of potential

temperature (Dy=Dz) and negative relationships with the

gradient of WS (Du=Dz). However, only ðDy=DzÞm and

ðDu=DzÞe appear as determinants in the regression

models. The two remaining gradients ðDy=DzÞe and

ðDu=DzÞm are much less sensitive to the 24-h averaged

particulate concentrations. This finding can be illu-

strated in Fig. 6, where the gradients of potential

temperature and WS are plotted along with the

occurrence of NRIs (full height columns) for the first

winter period. NRIs occurred in conjunction with high

ðDy=DzÞm (Fig. 6a) and low ðDu=DzÞe (Fig. 6d), but notin correlation with ðDy=DzÞe (Fig. 6b) and ðDu=DzÞm(Fig. 6d).

P.D. Hien et al. / Atmospheric Environment 36 (2002) 3473–3484 3481

Among determinants in the regression models, 9 out

of 25 (7 in winter and 2 in summer) are the previous

day’s parameters, showing the importance of the

delayed response of air particulate matter to meteorol-

ogy, especially in winter. There are cases with only the

previous day’s parameters present, e.g. Tp and RAINp in

the PM2.5 winter model. Meanwhile, in the PM2.5

summer model both WS and WSp appear, making WS

a leading determinant of fine particulates.

Therefore, only 5–7 among 20 meteorological para-

meters appear as determinants in each regression model,

that could explain from 60% to 74% of the day-to-day

variations of particulate concentrations in the winter

and summer. The remaining variances should be

attributed to the variabilities of emission strengths and

long-range transported (LRT) pollutants. Potential

sources of these LRT pollutants are areas under fast

growing urbanisation and industrialisation in South-

eastern China (Wang et al., 2001).

7. Summary and conclusion

Continental air masses from the Siberia High take the

way to North Vietnam either through the inland of

China or on the Pacific Ocean via the Gulf of Tonkin.

Inland trajectories, dominating in the first winter period,

supply dry air and facilitate the NRIs. In the second

winter period cold air parcels mostly travel through the

Pacific Ocean, supplying moist air and under antic-

yclonic conditions near ground STIs occur. During the

winter 1998–99, NRIs occurred on 43 days from

October to early January, while STIs were observed on

37 days (not including those occurring along with NRIs

in the first winter period). NRIs and STIs may persist for

days, yielding prolonged particulate pollution episodes.

The averaged PM10 concentrations were 84, 159 and

208 mgm�3 for the non-episode, STI and NRI cate-

gories, respectively. The corresponding figures for the

PM2.5�10 and PM2.5 concentrations were 43, 79 and

Table 3

Results of regression analysis (p ¼ 0:01)

PM Period R2 ln k (7std. err.) Determinant (loge) a Std. err. of a b

PM2.5�10

October 98–March 99 0.64 8.071.1

RAIN �0.22 0.03 �0.33ðDu=DzÞe �0.20 0.04 �0.26ðDy=DzÞm 0.25 0.05 0.24

RHp �0.95 0.25 �0.20WS �0.30 0.08 �0.20WSp �0.22 0.07 �0.15RAINp �0.09 0.03 �0.14

May 99–July 99 0.74 18.671.7

RH �5.08 0.60 �0.83RAIN �0.07 0.01 �0.32SUN 0.07 0.02 0.23

RHp 1.56 0.54 0.29

WS 0.24 0.09 0.18

ðDu=DzÞe �0.11 0.04 �0.16

PM2.5

October 98–March 99 0.60 7.770.5

Tp �1.25 0.15 �0.45WS �0.33 0.06 �0.33RAINp �0.16 0.02 �0.33WSp �0.17 0.06 �0.16ðDu=DzÞep �0.08 0.03 �0.15ðDy=DzÞm 0.12 0.04 0.15

ðDu=DzÞe �0.08 0.03 �0.14May 99–July 99 0.60 13.972.0

WS �0.44 0.07 �0.49RH �1.67 0.33 �0.42T �0.97 0.31 �0.26ðDu=DzÞep �0.10 0.04 0.21

ðDu=DzÞe �0.07 0.04 �0.17

P.D. Hien et al. / Atmospheric Environment 36 (2002) 3473–34843482

139mgm�3 and 41, 80 and 69 mgm�3, respectively,

showing a significant enhancement of the coarse mode

during NRI episodes. Soil particles thrown into the

atmosphere by street sweeping and traffic shaking in

evening rush hours are found to be a main cause giving

rise to the coarse mode enhancement as well as the very

high levels of PM2.5�10 during NRI episodes. As the

abundance of this soil dust component is inversely

related to the humidity of surface soil, the PM2.5�10

concentrations are much suppressed in humid condi-

tions during non-episode and STI periods. The above

findings provide guidance not only for forecasting

pollution episodes but also for developing abatement

measures.

Multiple regression analysis was applied to reveal

atmospheric parameters controlling the day-to-day

variations of particulate concentrations. PM2.5 is gov-

erned mainly by WS and air temperature, while rainfall

and RH largely control the daily variations of PM2.5�10,

indicating the high abundance of soil dust in the

PM2.5�10 fraction. Dusty air resulting from uncontrolled

construction works and unpaved roads and sidewalks is

common in urban areas of Vietnam (Hien et al., 2001).

As to parameters characterising atmospheric turbulence,

among the predictors of 24-h particulate concentrations

are the vertical gradients of potential temperature and

WS recorded at 06.30 and 18.30, respectively. These

controlling meteorological parameters dominate the

regression models for both the winter and summer

periods. Regression models could explain from 60% to

74% of the variances of 24-h particulate concentrations.

The remaining unexplained parts are associated mainly

with the variabilities of emission strengths and LRT air

pollutants.

Acknowledgements

This research was funded by the Ministry for Science,

Technology and Environment and was supported by

UNDP/IAEA/RCA Co-ordinated Project for Asia and

the Pacific on Air Pollution and its Trends. The authors

are grateful to the Hanoi Meteorological Station for the

kind assistance in providing routine surface observation

data. We also gratefully acknowledge contributions of

N.H. Quang and N.Q. Long (Institute of Nuclear

Fig. 6. The gradients of potential temperature and wind speed at 06.30 (a,b) and 18.30 (c,d) during the first winter period October–

December 1998 (see footnote a in Table 1 for the units of the gradients). The graphs show the association of ðDy=DzÞm and ðDu=DzÞewith the occurrence of NRIs, which is marked by full-height columns.

P.D. Hien et al. / Atmospheric Environment 36 (2002) 3473–3484 3483

Science and Technology) during the implementation of

this research project.

References

Aron, R., 1983. Mixing height—an inconsistent indicator of

potential air pollution concentrations. Atmospheric Envir-

onment 17, 2193–2197.

Baxter, R., 1991. Determination of mixing heights from data

collected during the 1985 SCCCAMP field program.

Journal of Applied Meteorology 30, 598–606.

Berkowicz, R., Prahm, L.P., 1982. Evaluation of the profile

method for estimation of surface fluxes of momentum and

heat. Atmospheric Environment 16, 2809–2819.

Berman, S., Ku, J.Y., Rao, S.T., 1999. Spatial and temporal

variation in the mixing depth over the northeastern United

States during the summer of 1995. Journal of Applied

Meteorology 38, 1661–1673.

Beyrich, F., 1997. Mixing height estimation from sodar data—a

critical discussion. Atmospheric Environment 31, 3941–

3953.

Calori, G., Carmichael, G.R., 1999. An urban trajectory model

for sulfur in Asian megacities: model concepts and

preliminary application. Atmospheric Environment 33,

3109–3117.

Cheng, S., Lam, K.C., 2000. Synoptic typing and its application

to the assessment of climatic impact on concentrations of

sulphur dioxide and nitrogen oxides in Hong Kong.

Atmospheric Environment 34, 585–594.

Chung, K.K., Chan, J.C.L., Ng, C.N., Lam, K.S., Wang, T.,

1999. Synoptic conditions associated with high carbon

monoxide episodes at a coastal station in Hong Kong.

Atmospheric Environment 33, 3087–3095.

Elsom, D.M., Chandler, T.J., 1978. Meteorological controls

upon ground level concentrations of smoke and sulfur

dioxide in two urban areas of the United Kingdom.

Atmospheric Environment 12, 1543–1554.

Harrison, R.M., Deacon, A.R., Jones, M.R., Appleby, R.S.,

1997. Sources and processes affecting concentrations of

PM10 and PM2.5 particulate matter in Birmingham (UK).

Atmospheric Environment 31, 4103–4117.

Heffter, J.L., 1983. Branching atmospheric trajectory (BAT)

model. NOAA Technical Memorandum ERL ARL-121.

Hien, P.D., Binh, N.T., Truong, Y., Ngo, N.T., 1999. Temporal

variations of source impacts at the receptor as derived from

air particulate monitoring data in Ho Chi Minh City,

Vietnam. Atmospheric Environment 31, 1073–1076.

Hien, P.D., Binh, N.T., Truong, Y., Ngo, N.T., Sieu, L.N.,

2001. Comparative receptor modelling study of TSP, PM2

and PM2�10 in Ho Chi Minh City. Atmospheric Environ-

ment 35, 2669–2678.

Holzworth, C.G., 1967. Mixing depths, wind speed and air

pollution potential for selected locations in the United

States. Journal of Applied Meteorology 6, 1039–1044.

Lena, F., Desiato, F., 1998. Intercomparison of nocturnal

mixing height estimate methods for urban air pollution

modelling. Atmospheric Environment 33, 2385–2393.

Mahrt, L., Andre, J.C., Heald, R.C., 1982. On the depth of the

nocturnal boundary layer. Journal of Applied Meteorology

21, 90–97.

Maenhaut, W., Francois, F., Calmayer, J., 1993. The GENT

stacked filter unit sampler for collection of atmospheric

aerosols in two size tractions, IAEA NAHRES-19, Vienna,

pp. 249–263.

Marsik, F.J., Fischer, K.W., McDonald, T.D., Samson, P.J.,

1995. Comparison of methods for estimating mixing height

used during the 1992 Atlanta field intensive. Journal of

Applied Meteorology 34, 1802–1847.

Monin, A.S., Obukhov, A.M., 1954. Dimensionless character-

istics of turbulence in the atmospheric surface layer.

Doklady Academii Nauk Uzbekskoi SSR 93, 223–226.

Pochanart, P., Kreasuwun, J., Sukasem, P., Geeratithada-

niyom, W., Tabukanon, M.S., Hirokawa, J., Kajii, Y.,

Akimoto, H., 2001. Tropical tropospheric ozone observed

in Thailand. Atmospheric Environment 35, 2657–2668.

Seibert, P., Beyrich, F., Gryning, S.E., Joffre, S., Rasmussen,

A., Tercier, P., 2000. Review and intercomparison of

operational methods for the determination of the mixing

height. Atmospheric Environment 34, 1001–1027.

Toan, P.N., Dac, P.T., 1993. The climate of Vietnam. Science

and Technology Publisher, Hanoi (in Vietnamese).

Wang, T., Cheung, V.T.F., Lam, K.S., Kok, G.L., Harris, J.M.,

2001. The characteristics of ozone and related compounds in

the boundary layer of the South China coast: temporal and

vertical variations during autumn season. Atmospheric

Environment 35, 2735–2746.

P.D. Hien et al. / Atmospheric Environment 36 (2002) 3473–34843484