Science of the Total Environment 573 (2016) 110

Contents lists available at ScienceDirect

Science of the Total Environment

journal homepage: www.elsevier.com/locate/scitotenv

An AIS-based high-resolution ship emission inventory and its uncertainty

in Pearl River Delta region, China
Cheng Li a, Zibing Yuan a, Jiamin Ou a, Xiaoli Fan a, Siqi Ye b, Teng Xiao a, Yuqi Shi a, Zhijiong Huang a,
Simon K.W. Ng c, Zhuangmin Zhong a, Junyu Zheng a,
a
b
c

School of Environment and Energy, South China University of Technology, Guangzhou 510006, China
Guangdong Environmental Monitoring Center, Modiesha Street, Haizhu District, Guangzhou, 510308, China
Civic Exchange, 23/F, Chun Wo Commercial Centre, 23-29 Wing Wo Street, Central, Hong Kong

H I G H L I G H T S

G R A P H I C A L

A B S T R A C T

A comprehensive database with

N 12,000 ships information was
established.
A yearlong AIS dataset was used to develop time-in-mode and spatiotemporal surrogates.
A bottom-up approach was used to estimate emissions from major ship source
sectors.
The ECA region covered over 80% of the
total ship emissions in the PRD region.
Application of AIS data reduced uncertainty in ship emission estimates.

a r t i c l e

i n f o

Article history:
Received 26 June 2016
Received in revised form 30 July 2016
Accepted 30 July 2016
Available online xxxx
Editor: D. Barcelo
Keywords:
Ship emission inventory
Automatic identication system
Spatiotemporal characteristics
Pearl River Delta
Emission control area

Corresponding author.
E-mail address: zheng.junyu@gmail.com (J. Zheng).

http://dx.doi.org/10.1016/j.scitotenv.2016.07.219
0048-9697/ 2016 Elsevier B.V. All rights reserved.

a b s t r a c t
Ship emissions contribute signicantly to air pollution and impose health risks to residents along the coastal area.
By using the rened data from the Automatic Identication System (AIS), this study developed a highly resolved
ship emission inventory for the Pearl River Delta (PRD) region, China, home to three of ten busiest ports in the
world. The region-wide SO2, NOX, CO, PM10, PM2.5, and VOC emissions in 2013 were estimated to be 61,484,
103,717, 10,599, 7155, 6605, and 4195 t, respectively. Ocean going vessels were the largest contributors of the
total emissions, followed by coastal vessels and river vessels. In terms of ship type, container ship was the leading
contributor, followed by conventional cargo ship, dry bulk carrier, shing ship, and oil tanker. These ve ship
types accounted for N90% of total emissions. The spatial distributions of emissions revealed that the key emission
hot spots all concentrated within the newly proposed emission control area (ECA) and ship emissions within ECA
covered N80% of total ship emissions in the PRD, highlighting the importance of ECA in emissions reduction in the
PRD. The uncertainties of emission estimates of pollutants were quantied, with lower bounds of 24.5% to
21.2% and upper bounds of 28.6% to 33.3% at 95% condence intervals. The lower uncertainties in this study
highlighted the powerfulness of AIS data in improving ship emission estimates. The AIS-based bottom-up methodology can be used for developing and upgrading ship emission inventory and formulating effective control
measures on ship emissions in other port regions wherever possible.
2016 Elsevier B.V. All rights reserved.

C. Li et al. / Science of the Total Environment 573 (2016) 110

1. Introduction
Rapid economic growth has fueled signicant progress of shipping
industry in China, resulting in great energy demand and signicant
emissions of air pollutants (Endresen et al., 2003, 2007; Corbett et al.,
2007; Dalsren et al., 2007). Since 1980s, control measures have been
focused on land-based emission sources and their emissions have
been greatly reduced (Zhong et al., 2013; Zhao et al., 2013). Ship emissions, a signicant contributor to ambient pollutants in coastal areas
(Ye, 2014), were however largely ignored (Lu et al., 2010; Zhao et al.,
2013). In order to understand the impact of ship emissions on air quality
and to formulate effective control measures for ship emissions, a detailed understanding on the emission characteristics and the development of a high-resolution emission inventory for ship sources are
needed.
As an international manufacturing hub with a dense waterway network, the Pearl River Delta (PRD) region in southern China shows
booming shipping industry. According to Global Port Development Report (Shanghai International Shipping Institute, 2015), Shenzhen,
Hong Kong and Guangzhou were among the top ten ports worldwide
in terms of container throughput in 2015. The PRD as a port cluster
was bound to contribute the largest throughput in the world. In 2015,
the PRD was designated as a vessel emission control area (ECA) by the
Ministry of Transport of China (2015). A series of stringent control measures have been or will be implemented according to a prescribed
schedule. To better describe emission characteristics, efforts have been
paid to develop ship emission inventories around some ports in this region, e.g. Hong Kong (Yau et al., 2012; Ng et al., 2013) and Shenzhen (Ye
et al., 2014; Yang et al., 2015). However, this kind of port-level emission
inventory cannot support regional air quality modeling to quantify their
impacts because of limited coverage and low spatial and temporal resolutions. Development of a high-resolution region-wide emission inventory is therefore needed in the PRD.
In recent years, considerable efforts have been paid and technological advancements have been achieved to improve the accuracy of ship
emission estimates, such as on-board survey of fuel consumption and
action (Cooper, 2003; Hulskotte and Denier, 2010; Song, 2014) and
eld-testing and reviewing of ship emission factors (Todd and
Andrew, 2007; Zhang et al., 2015; Song, 2015). The Waterway Network
Ship Trafc, Energy, and Environmental Model (STEEM) was also developed which combined ship characteristics and activity-based emission
factors to establish ship emission inventory in North America (Wang
et al., 2007; ICF international, 2009; Corbett, 2010; Bandemehr et al.,
2015). Automatic Identication System (AIS) data was increasingly applied to improve activity-based estimation such as load factor (Winther
et al., 2014) and to develop more accurate spatiotemporal surrogates to
replace empirical waterway network (Wang et al., 2008; MARIN, 2010,
2011; Yang et al., 2015). Despite improved techniques in information
collection, current studies on ship emissions are still subject to a number
of limitations. First, most studies targeted on ocean-going vessels
(OGVs), while local ship emissions from rivers and coastal areas were
ignored (Agrawal et al., 2008; Yau et al., 2012; Merk, 2014; Tan et al.,
2014). Secondly, although the representativeness of spatiotemporal
surrogates had been improved by AIS density distribution data
(Jalkanen et al., 2009, 2011; Gutierrez et al., 2012), the inuence from
trafc routes of ship types and activity modes on ship emissions were
lack of consideration. Furthermore, the uncertainties in ship emission
estimates were generally analyzed in qualitative ways (Fu et al., 2012;
Ng et al., 2013) and only the uncertainties from emission factors
(Todd and Andrew, 2007; Entec UK Limited, 2010) were considered. It
calls for a quantitative assessment of the uncertainties in ship emissions
arising from both emission factor and activity level data.
In this study, we aimed to establish a high-resolution ship emission
inventory for the PRD region by using a bottom-up approach for the
base year of 2013. Main engine power, load factor, time-in-mode, fuel
consumption and other ship activities data were collected from the

rened data from AIS, eld survey and local statistics. The estimated
emissions covered all kinds of OGVs, coastal vessels (CVs) and river vessels (RVs) within PRD waters, except the transit ships which sailed
through without hotelling in ports of the region (excluding Hong
Kong). Emissions distribution maps and temporal surrogates with ship
types and activity modes were developed. The uncertainties of ship
emission inventory were quantitatively estimated, and implications
for control policies were discussed.

2. Data and methods

2.1. Study domain and ship categorization
The study domain was set within the latitude of 21.54 to 24.53N
and the longitude of 111.55 to 115.61E, covering all ports with N2100
berths, major fairway network and the proposed ECA in the PRD region,
as shown in Fig. 1. Nine port groups in this study domain are administrated by local Maritime Departments (MDs), namely Guangzhou,
Shenzhen, Huizhou, Zhuhai, Dongguan, Foshan, Zhongshan, Zhaoqing
and Jiangmen.
This study estimated ship emissions from OGVs as well as CVs and
RVs in PRD waters. According to the characteristics of the ship administration system in China, we developed a new approach in identifying
the ship type. OGVs are certied by both International Maritime Organization (IMO) and Maritime Mobile Service Identify (MMSI), therefore
IMO number and MMSI code were used to dene OGVs, while CVs
only have MMSI codes for identication. As all kinds of local ships are
also registered in MD, we used MD registration number to identify
RVs. Under these three categories, vessel calls were further divided
into 84 sub-categories by ship type and dead weight tonnage (DWT).
Details for source classication are listed in Table 1. It was noted that
the ship types in port statistics in MD were not classied based on
DWT. To deal with that, an adaptive sampling period statistic approach
based on real-time AIS data was adopted by considering port sizes and
wharf structure to remove errors in individual sampling period.
To investigate and evaluate the impact of cruising speed on emissions, ships were alternatively categorized according to their operational modes, i.e. fairway cruise, slow cruise, maneuvering and hotelling
(Entec UK Limited, 2010; Ng et al., 2013), as shown in Table SI-1.

Fig. 1. Waterway and port distribution map of the research domain in the PRD region.

C. Li et al. / Science of the Total Environment 573 (2016) 110

Table 1
DWT classication of different ship types in the PRD.
OGV

CV

RV

OGV

CV

Container

Container

Container

Chemical tanker

Chemical tanker

Chemical tanker

DWT b 10,000
DWT 10,00019,999
DWT 20,00029,999
DWT 30,00039,999
DWT 40,00049,999
DWT 50,00074,999
DWT 7500099,999
DWT 100,000
Gas tanker
DWT b 5000
DWT 50009999
DWT 10,00019,999
DWT 20,00039,999
DWT 40,000
Oil tanker
DWT b 10,000
DWT 10,00029,999
DWT 30,00059,999
DWT 60,000119,999
DWT 120,000

DWT b 3000
DWT 30004999
DWT 50009999
DWT N 10,000

DWT b 500
DWT 5001000
DWT N 1000

DWT b 5000
DWT 50009999
DWT 10,00019,999
DWT 20,00039,999
DWT 40,000
Conventional cargo ship
DWT b 2000
DWT 20004999
DWT 50009999
DWT 10,00029,999
DWT 30,000
Dry bulk carrier
DWT b 10,000
DWT 10,00029,999
DWT 30,00059,999
DWT 60,00099,999
DWT 100,000
Tug
Passenger ship
Fishing ship
Others

DWT b 3000
DWT 30005000
DWT 50009999
DWT 10,000

DWT b 500
DWT N 500

Conventional cargo ship

DWT b 5000
DWT 50009999
DWT 10,00029,999
DWT 30,000

Conventional cargo ship

DWT b 500
DWT 5001000
DWT N 1000

Dry bulk carrier

DWT b 3000
DWT 30004999
DWT 50009999
DWT 10,000

Dry bulk carrier

DWT b 500
DWT 5001000
DWT N 1000

Tug
Passenger ship
Fishing ship
Others

Tug
Passenger ship
Fishing ship
Others

Gas tanker
DWT b 3000
DWT 30004999
DWT 50009999
DWT 10,000

Gas tanker
DWT b 500
DWT N 500

Oil tanker
DWT b 3000
DWT 30004999
DWT 50009999
DWT 10,000

Oil tanker
DWT b 500
DWT N 500

Emissions from main engine and auxiliary engine were considered, but
those from auxiliary boilers were not included.
2.2. Data sources and characteristics
Vessel calls from six ports (Shenzhen, Guangzhou, Dongguan,
Zhongshan, Foshan and Zhaoqing) were directly obtained from MD,
while those from other ports were collected from information open to
the public. In 2013, OGVs, CVs and RVs accounted for 5.8%, 15.9% and
78.3% of the total 1,808,906 vessel calls received at all PRD ports, respectively. In terms of ship type, 32.3%, 23.6%, 15.0% and 5.3% were contributed by conventional cargo ship, dry bulk carrier, container and oil
tanker, respectively, as detailed in Table 2. As MD ofcial statistics did
not include shing ships which presented a large quantity in PRD waters, we made use of the statistics about their operation types from
Guangdong Agriculture Statistic Yearbook, activity time and energy
consumption data from survey by Guangdong Ocean and Fishery Administration, and other empirical parameters such as load factor from
literatures (Bao et al., 2014).
The activity information of different types of ships was collected
from various sources. We developed a database to integrate and process
the data, as illustrated in Fig. 2. Information of N 5000 OGVs and CVs
were acquired from Lloyd's Register of ships (LRS). Registration information of nearly 7600 RVs were acquired from local MDs. The LRS and
MD registration databases provide registration number, ship type and

Table 2
Vessel arrivals for different ship types in the PRD region in 2013.
Ship type

Chemical tanker
Conventional cargo ship
Dry bulk carrier
Container
Gas tanker
Oil tanker
Tug
Passenger ship
Others
Total

Vessel arrived number (VAN)

OGVs

CVs

RVs

Total

4207
14,145
14,317
32,326
1439
8950
4770
19,718
5230
105,102

19,138
76,582
34,175
20,284
5755
40,159
20,636
55,797
14,883
287,409

17,845
493,456
378,968
219,070
6378
46,535
71,518
111,847
70,778
1,416,395

41,190
584,183
427,460
271,680
13,572
95,644
96,924
187,362
90,891
1,808,906

RV

tonnage, major sea routes, fuel type, engine information, and other relevant information for emission estimation.
Specically, about 700 AIS-based navigation trajectories for 2013
were collected, covering OGVs, CVs and RVs and major ship types. The
AIS data from LRS agent (www.shipxy.com) offered a continuous and
complete set of ship trajectories that were used to estimate time-inmodes, load factors and spatiotemporal surrogates, e.g. real time, longitude, latitude, speed, navigation condition. The AIS data include shorebased and satellite-based data. The shore-based data is featured by
high temporal resolution (every 30 s), but only covers ships b50 nautical miles from the shore. For the areas beyond 50 nautical miles, satellite-based data in 2-h interval was used.
2.3. Emission estimation
For ship emissions except shing ship, a bottom-up approach was
used, as shown by Eq. (1) (U.S.EPA, 2000, 2008; Ng et al., 2012):
Ek

n
X
VANi  P lj  L F ljm  T ljm  E F ljk  106

i1

where i, j, k, l, m, and n represents one single voyage, engine type, pollutant, DWT class in ship types, activity mode, and total vessel arrived
number (VAN), respectively. E is the emission (t), P is the average
installed engine power (kW), LF is the average engine load factor, T is
the average operation time in four activity modes (h), and EF is emission
factor of engine types and fuel types (g/kWh).
Emissions from shing ship were estimated by Eq. (2) based on fuel
consumption rate and activity time.
Ek

n
X
N l  P l  LF l  T l  F l  E F k  106

l1

where l and k is ship types and pollutant, respectively. P is average

installed engine power (kW), LF is average engine load factor, T is average ship activity time (h) in one year, F is fuel consumption rate (kg of
fuel/kWh), and EF is emission factor (g/kg of fuel).
In both formulae, engine power and load factors of main engine
were estimated by Propeller Law which was widely used in ship emission estimation (ICF international, 2009). Propeller Law calculates the
ratio between instantaneous and the maximum designed speed at

C. Li et al. / Science of the Total Environment 573 (2016) 110

Fig. 2. Emissions estimation methods and data sources in this study.

time t. The actual speed was determined from the AIS data, while the
maximum cruising speed was either obtained from LRS directly or replaced by the same ship type and DWT subsector. The load factors at
all points along the track for all ships in 2013 were calculated. The averaged load factors for the four operational modes were identied according to the speed, and further categorized according to pollutant sources
(ICF international, 2009). Due to the lack of information and the similar
propulsion ratios for auxiliary engine (Ng et al., 2012), the propulsion
ratios and load factors for different ship types and operational modes
were obtained from a technical report (U.S.EPA, 2008).
AIS data was also used to dene time-in-mode for OGVs and CVs (Ng
et al., 2013; Ye et al., 2014). AIS ship track data was rstly screened from
raw data based on geographic coordinate of the study domain. Each data
point along the track was then assigned to an operation mode according
to the corresponding ship speed. For continuous data with the same
mode, data was only retained with a 510 min interval to reduce dimension. Data fragments in different modes were then grouped into a
complete call duration to calculate time-in-mode. By excluding the invalid duration segments, time-in-mode was then developed according
to different ship types and tonnages. For RVs, the activity hours were obtained from onboard survey and local technical report (Ng et al., 2012).
Emission factors were selected from literature (ICF International,
2009; Starcrest Consulting Group, 2009) according to local emission
characteristics, as detailed in Table SI-2. Oil with sulfur content of
2.7%, 1.0%, and 0.5% in mass was regarded as residual oil (RO), marine
diesel oil (MDO) and marine gas oil (MGO), respectively (Ng et al.,
2012). Emission factors for different combinations of activity type, engine type, and fuel type were listed in Table SI-3. Low-load adjustment
multipliers were applied when load factors were below 20% to account
for lower combustion efciency during low main engine loading condition (Table SI-4).
2.4. Temporal and spatial allocation of emissions
The bulk emissions were temporally and spatially allocated with surrogates from the AIS data as well as other ofcial statistics. Since the pollutant emissions are signicantly correlated with port throughput, the
average monthly throughput data in 20112015 were used to depict
the monthly variations of emissions from container ships, conventional
cargo ships and passenger ships. Diurnal proles of emissions were
developed from AIS ship track data. To minimize the bias raised

from seasonal transition and locations, sampling was conducted

one day every month with the sampling grid cells covering different
water areas.
A dot density weighted algorithm was applied for spatial allocation
of emissions. This algorithm used density of data dots to calculate the
spatial surrogates of six major ship types, i.e. conventional cargo ship,
bulk carrier, container ship, passenger ship, tanker, and others, by
weighting emissions from different ship types, pollutant types and navigating modes. The emissions in different ports and water areas were
dened by navigating mode. According to the weights of the above spatial surrogates in every grid cell, the bulk emissions were distributed to
the 3 km 3 km grid cells covering the PRD waters.
2.5. Uncertainty analysis
Previous qualitative studies indicated that uncertainties of ship
emissions were mainly introduced by time-in-mode, load factor and
emission factor (Yang et al., 2007; Ng et al., 2013). In this study, AIS
data were used to quantitatively characterize uncertainties of time-inmode and load factor by using bootstrap simulation. Statistical methods
and expert judgment were both used to estimate uncertainties of emission factor. Monte Carlo simulation was applied to propagate uncertainties in the above inputs to the estimated results (Corbett, 2002;
NARSTO, 2005; He et al., 2011). Key sources of uncertainties were
then identied for further improvement.
3. Results and discussion
3.1. Emission inventory and sources characteristics
As shown in Table 3, total emissions of SO2, NOX, CO, PM10, PM2.5,
and VOC from all kinds of ships were 61,484, 103,717, 10,599, 7155,
6605, and 4195 t, respectively, for the PRD region in 2013. Ship emission
accounted for 14.1% of total SO2, 11.6% of total NOX and 1.5% of total
PM10 emissions in the PRD. It accounted for 75% of SO2, 22.9% of NOX
and 21.2% of PM10 emissions from transport sector. To have an indepth understanding of pollutant emission characteristics in the PRD region, emissions in terms of ports, ship types, engine types, navigation
routes, and operational modes were discussed below and compared
with other studies.

C. Li et al. / Science of the Total Environment 573 (2016) 110

Table 3
Ship emission in the PRD region in 2013 and its percentages in the transport sector and all
emissions in the PRD (unit: ton/year).
Ship types

SO2

NOX

CO

PM10

PM2.5

VOC

Chemical tanker
Conventional cargo
ship
Dry bulk carrier
Container
Gas tanker
Oil tanker
Tug
Passenger ship
Fishing ship
Others
Sum of ship sector
Transport sectora
% of transport
sector
All sourcesa
% of all sources

2415
12,431

4078
20,173

394
2102

267
1368

247
1267

161
832

6791
30,659
348
5195
1230
1041
877
498
61,484
81,952
75.0%

11,035
43,272
578
7799
2193
1761
11,666
1161
103,717
452,216
22.9%

1122
3907
65
759
229
163
1737
120
10,599
1,894,567
0.6%

757
3430
38
581
82
67
522
43
7155
33,702
21.2%

701
3167
35
537
76
61
474
40
6605
31,831
20.8%

408
1380
26
304
88
87
862
47
4195
271,411
1.5%

434,827 893,196 4,077,285 493,383 182,693 775,587

14.1%
11.6%
0.3%
1.5%
3.6%
0.5%

From Zheng et al., 2016 (in preparation). Based year was 2012.

As shown in Fig. 3(a), Guangzhou and Shenzhen ports together contributed to over 60% of total pollutant emissions from nine port groups.
This signicant fraction can be attributed to their higher throughput and
better infrastructure. Emissions from the other ports in a descending
order were Huizhou, Dongguan, Foshan, Zhuhai, Jiangmen, Zhongshan
and Zhaoqing.
Regarding different ship types, domestic container was the largest
contributor to all kinds of pollutants, accounting for 49.9%, 41.7%,
36.9%, 47.9%, 47.9% and 32.9% of the region-wide SO2, NOX, CO, PM10,
PM2.5 and VOC emissions, respectively, followed by conventional

cargo ship, shing ship, dry bulk carrier and oil tanker (Fig. 3(b)). Significant emissions from container ship and conventional cargo ship were
associated with large number of total ship calls and high engine
power. Fishing ship also contributed notably because of its large population (N28,000) sailing and working every day except the shing moratorium season from June to August. In comparison, only b2% of total
emissions were from passenger ships, as these passenger ships were
generally to or from Hong Kong and cleaner fuels were required under
the restriction of the Fair Winds Charter (Ng et al., 2013). The emission
characteristics in terms of ship types in top three ports, i.e., Guangzhou,
Shenzhen and Huizhou port, were slightly different from the regionwide pattern, with bulk carrier, container ships and tankers the major
contributing ship types.
Main engines contributed 59.7%, 52.1%, 48.6%, 57.5%, 57.4%, and
63.9% to the SO2, NOX, CO, PM10, PM2.5, and VOC emissions, respectively,
as shown in Fig. 3(c). Emissions from main engines were higher than
those from auxiliary ones for all pollutants except CO. Higher contributions of main engines were resulted from their large fuel consumption.
For CO emission, the combustion efciencies of auxiliary engines were
generally lower due to their lower load factor and therefore more favorable for the production of CO.
Fig. 3(d) showed the contributions by navigating areas. It was obvious that OGVs were the largest contributor to all six kinds of pollutants.
In particular, OGVs accounted for nearly 70% of SO2 and PM10, due much
to their high engine power and low oil quality with high sulfur and aromatic contents. Among OGVs, containers were the largest contributing
group. Installed with medium-speed diesel engines, CVs were associated with higher engine power and emission factors therefore contributed greater emissions of NOX, PM10, PM2.5 and VOC than RVs installed
with high-speed diesel engines. It should be noted that emissions
from RVs in the PRD internal waterways accounted for 10.8%, 19.2%,
30.7%, 9.4%, 9.8% and 15.2% of total emissions of SO2, NOX, CO, PM10,
PM2.5, and VOC, respectively, which were signicantly higher than

a) Ports

c) Engine types

b) Ship types

d) Navigating area

e) Modes

Fig. 3. Emission contributions from different sources for (a) ports, (b) ship types, (c) engine types, (d) navigating area, and (e) modes.

C. Li et al. / Science of the Total Environment 573 (2016) 110

Table 4
Comparisons of pollutant emissions estimates in the Pearl River Delta with those in Los
Angeles and Hong Kong (unit: ton/year).
Domain/year Los Angeles/2013

Hong Kong/2007 PRD/2013

Objects

OGVs

OGVs

Call number
SO2
NOX
CO
PM10
PM2.5
VOC
Reference

2033
2033
535
542
3515
6967
423
2056
105
193
79
178
486a
231a
Starcrest Consulting
Group, 2014

Total

Total

37,152 145,930
12,438 15,719
14,462 32,744
1421
10,535
1447
2139

635
3489
Ng et al., 2012

OGVs

Total

105,960 1,890,359
41,073
61,484
53,243
103,717
4239
10,599
4614
7155
4245
6605
1871
4195
This study

a
Report given HC, which used to estimate VOC by equation [VOC: HC = 1.053]
(U.S.EPA, 2008).

those around the Shanghai port in the Eastern China. This can be attributed to the larger number of RVs and their intense activities due to the
prosperous regional trades at the Pearl River Estuary (Yang et al., 2007).
Emission contributions from various sailing modes were shown in
Fig. 3(e). The hotelling mode was associated with the largest emissions,
responsible for 30.9%, 35.9%, 38.8%, 32.7%, 32.7%, and 27.0% for SO2, NOX,
CO, PM10, PM2.5, and VOC emissions, respectively. Fairway cruise and
slow cruise also contributed signicantly while emissions from maneuvering were the lowest. This was different from most port emission inventories worldwide in which emissions from cruising and slow
cruising were lower than maneuvering (Yau et al., 2012; Yang et al.,
2015). The lower contribution of maneuvering in this region might be
partly explained by longer periods of cruising and slow cruising in the
PRD waters than those around other ports, such as Shanghai, Hong
Kong and Los Angeles (Yang et al., 2007; Yau et al., 2012; Starcrest
Consulting Group, 2014).
Results of this study were compared with those in Los Angeles and
Hong Kong, the two ports with abundant studies on ship emissions
(Ng et al., 2012; Starcrest Consulting Group, 2014) (Table 4). Due to
the variations in ship activities and the scope of study domains, exhaust
emissions were normalized based on vessel call number. Emission
strengths of most pollutants from ships in the PRD region were comparable to those in Hong Kong but less than those in Los Angeles. This may
be ascribed to differences in hotelling times, fuel quality, fairway distance, and navigation condition. Ship emissions in the PRD also showed

a) Monthly

higher SO2/NOX ratio than Hong Kong and Los Angeles. This was probably resulted from more stringent regulation on sulfur content in shipping fuels under the North American ECA and the Hong Kong Fair
Winds Charter (IMO, 2010; H.K.EPD, 2015). Furthermore, the regulations of ECA in the PRD region have not been fully implemented yet.
3.2. Temporal characteristics
Fig. 4 showed the monthly and diurnal variations of emissions from
different ship types in the PRD region. The monthly variations of container and cargo ship emissions were similar, slightly higher in August
and December and lower in February. This was mainly due to greater
human activities in summer and autumn while the relatively inactive
cargo transportations during the long public holiday of Spring Festival
in February. Passenger ships showed a bimodal monthly variation pattern with two peaks in August and December. It was interesting that
port passenger volume in February increased quickly in recent 5 years,
which reected increased travels in the Spring Festival in the PRD
region.
Passenger ship was the only type that displayed signicant diurnal
patterns. With hourly percentage of b 1% at the midnight and early
morning, emissions from passenger ship increased dramatically from
8 am, reached the emission peak at 10-11 am, slightly declined afterwards and reached another peak at 5 pm. Emissions from other ship
types remained constant in a day, since they were generally for longdistance transport and sailed at all times under the 24 h rotation system.
3.3. Spatial characteristics
The spatial distributions of four pollutants, SO2, NOX, PM10 and VOC,
were illustrated in Fig. 5. Their spatial distribution roughly followed the
waterway along the main ports of the PRD, e.g. Guangzhou, Humen of
Dongguan, and Chiwan and Mawan of Shenzhen. Groups of grid cells
along the coast where the ports are located had higher emission intensities since nearly half of the total emissions were from ships under
hotelling conditions. Emissions in other grid cells were scarce and exhibited a relatively uniform distribution over the PRD given the open
and wide coastal water areas, small mooring point and discrete channels. In terms of the emissions by pollutants, the emission strengths
(emissions per grid cell) of SO2 and NOX were particularly higher,
which were also found in other studies (Ye et al., 2014; Yang et al.,
2015).

b) Diurnal

Fig. 4. (a) Monthly and (b) diurnal variations of percentage of vessel types.

C. Li et al. / Science of the Total Environment 573 (2016) 110

Fig. 5. Spatial distributions of SO2, NOX, PM10 and VOC.

To illustrate spatial distribution of emissions by ship types, NOX

emissions from different ship types were shown in Fig. 6 as examples. Regarding emissions from container ships, the hot spots were
around Yantian port of Shenzhen and Nansha port of Guangzhou,
both of which were world-class deep water ports. Emissions from
bulk cargo ships and conventional cargo ships were concentrated
around Guangzhou port, the largest distribution hub port of bulk
cargo and groceries. In addition, Guangzhou port in connection
with many other ports in the PRD forms the Three-latitudinal &
three-longitudinal ship transport network. Emissions from tankers
were distributed in relatively wide areas, including the coast of
Huizhou, the harbor waterway of Guangzhou and the longitudinal
waterway of Shenzhen. This was due to the strong demand of energy
and chemical products transported by tankers in all PRD cities to

satisfy the rapid economic development. Emissions from passenger

ships were along the waterway connecting the cities of Shenzhen
and Hong Kong in the east coast of the PRD and Zhongshan, Zhuhai
and Macau in the west.
According to the weights of spatial surrogates within the proposed ECA areas, emissions inside the ECA were estimated. Result
showed that emission inside the ECA area accounted for 84.3%,
94.9%, 89.6%, 96.8%, 87.9% and 94.3% of the region-wide emissions
from conventional cargo ship, bulk carrier, container ship, passenger
ship, tanker, and others in the PRD, respectively. By covering all ship
emission hot spots inside the PRD and responsible for N80% of the region-wide emissions, the delineation of ECA and the proposed stringent control measures have great potential in controlling ship
emissions in the PRD.

C. Li et al. / Science of the Total Environment 573 (2016) 110

Fig. 6. Spatial distributions of NOX emissions from different ship types.

C. Li et al. / Science of the Total Environment 573 (2016) 110

3.4. Uncertainties in ship emission estimates

Uncertainties of the ship emission inventory were quantitatively estimated in this study. The uncertainties distribution for time-in-mode,
load factors and emission factors were given in Tables SI-5 & SI-6. Uncertainties in emission estimates were derived for each combination
of pollutant type, ship type and navigating route. As summarized in
Table 5, uncertainties in six pollutants were found with lower bounds
of 24.5% to 21.2% and upper bound of 28.6% to 33.3% at 95% condence intervals. These uncertainties were generally lower than those reported in other inventories (about 40% to 50% at 95% condence
intervals) (Ng et al., 2012; Ye, 2014). This can be attributed to the significant efforts in improving the accuracy of the ship activity data on the
regional scale by AIS data. The activity-based methods are regarded as
the best practices for ship inventories currently. Nevertheless, reliable
local emission factors, auxiliary engine power rating and load factors
were still lacking in the PRD and contributed greatly to the uncertainties
in this study.
Factors contributed to the emission uncertainties were further
analyzed. Among the six ship types, conventional cargo ship was associated with the highest NOX emission uncertainty ( 43.8% to 81.3% at
95% condence interval), while uncertainties in others ship types
were relatively small ( 23.8% to 38.0% at 95% condence interval)
(Table SI-7). High uncertainties in conventional cargo ship estimates
can be attributed to the fact that both MD and LRS relied on human discretion when dealing with some ambiguities in conventional cargo ship
registered information. For example, cargo ships with container capacity, vehicle carrier, roll on/roll off or other multipurpose ship might be
counted as conventional cargo ships. In addition, the uncertainties distribution for conventional cargo ship in model input was also high.
Uncertainties from different ship activities types were given in Table
SI-8. It can be found that emission from RVs had higher uncertainties
than CVs and OGVs, especially for SO2 and PM10 emissions with about
50% to 80% at 95% condence interval. These high uncertainties
were introduced by the emission factor of PM, the sulfate content of
oil, and the big knowledge gap on oil quality and ship activity of RVs.
We assumed there were about 2060% relative error for the activity
input and 30%50% relative error for the emission factors.
Although the uncertainties of ship emission inventory had been reduced to some extent by using AIS data, uncertainties were still high
that may pose signicant impact to the air quality modeling. Reliable
local emission factors, auxiliary engine power rating and load factors
were still lacking in the PRD and contributed greatly to the uncertainties
in this study. Further efforts in reducing uncertainties may include measurement of onboard emission and fuel consumption, onboard survey to
validate activity data estimated by AIS data, and incorporation of emissions from transit ships and other equipment into the inventory.
3.5. Implications for future work and policy-making
This study provides solid scientic support for ship emissions control in the PRD region. With the powerful AIS-based data, emissions by
different ports, ship types, engine types, navigation routes, and

operational modes were identied. OGVs and hotelling mode each contributed to more than half of the ship emissions, suggesting the importance of cleaner fuels and shore power in the port areas. More
importantly, this study provides justications on the establishment of
ECA in the PRD region, which is virtually the rst concrete action to address air pollution from ship activities in China. The ECA covered over
80% of the ship associated emissions within the region, including all
the ship emission hot spots along the PRD estuary. The cleaner marine
fuel regulation and strict emission standards inside the ECA are therefore expected to signicantly reduce the ship emissions by 2020.
The AIS-based methodology used in this study is also expected to
shed light on the development and renement of marine emission inventory in China. As home to seven out of the world's top ten container
ports, the ship emissions in China are huge but not comprehensively
studied due to the data gap in activity level data and lack of public
awareness. The data gap in ship emission estimation hinders the implementation and assessment of control schemes in association with the
ECA. Given its high temporal and spatial resolutions, AIS-based data is
demonstrated to be a reliable data source to develop highly resolved
marine emission inventory, which can be extended to other port cities
in China and other areas.
Further renement of ship emission inventories should be targeted
on introducing local emission factors and improving completeness and
accuracy in ship registration information. In addition, as coastal areas
in China are also densely populated, health risk assessment should be
conducted based upon reliable emission inventory in order to quantify
the potential health impact and socioeconomic cost borne by local residents from ship emissions.
4. Summary and conclusions
We demonstrated that AIS data is useful in identifying ship characteristics, time-in-modes, load factor and ship type-specic emission factors that are essential in the bottom-up ship emission inventory
development. The results showed that the total ship emissions for SO2,
NOX, CO, PM10, PM2.5 and VOC were 61,484, 103,717, 10,599, 7155,
6605 and 4195 t, respectively, in the PRD in 2013. OGVs were the largest
contributors of the total emissions, followed by CVs and RVs. Container
ship was the top emitting ship type, followed by conventional cargo
ship, dry bulk carrier, shing ship, and oil tanker. These ve ship types
accounted for N90% of total emissions.
The monthly variations of container and cargo ship emissions were
slightly higher in August and December and lower in February. Passenger ship was the only type associated with signicant diurnal pattern,
with the hourly percentage from almost zero at midnight to over 8%
in the morning and 7% in the late afternoon. Emission hot spots for container ships were around Yantian port of Shenzhen and Nansha port of
Guangzhou, and emissions from bulk cargo ships and cargo ships were
concentrated around Guangzhou port.
The uncertainties of total emission estimation on all kinds of pollutants ranged from 24.5% to 33.3% at 95% condence intervals. Although varied by ship type and tonnage, the uncertainties of PRD ship
emission inventory were generally lower than those from previous

Table 5
Uncertainties in emission estimates.
Species

Emission estimatea (ton)

Mean (ton)

95%CI

Uncertainty

SO2
NOX
CO
PM10
PM2.5
VOC

60,607
92,051
8862
6633
6132
3333

69,116
86,226
9340
6989
6740
2785

(54,443, 88,847)
(67,184, 112,566)
(7231, 12,170)
(5402, 9133)
(5202, 8859)
(2104, 3713)

(21.2%, 28.6%)
(22.1%, 30.6%)
(22.6%, 30.3%)
(22.7%, 30.7%)
(22.8%, 31.5%)
(24.5%, 33.3%)

Only emissions from commercial ships given here (except shing ships).

10

C. Li et al. / Science of the Total Environment 573 (2016) 110

studies which can be largely attributed to the usage of AIS data. The
emission inventory supported the establishment of ECA as N 80% of the
region-wide ship emissions were covered by the ECA and the strengthened control measures implemented in ECA are expected to reduce the
ship emission signicantly. This AIS-based methodology can shed light
on ship emission inventory development in other port areas with significant emissions from ship sources.
Acknowledgment
This research was supported by the Public Environmental Service
Project of the Ministry of Environmental Protection of People's Republic
of China (204109012), National Distinguished Young Scholar Science
Fund of the National Natural Science Foundation of China (No.
41325020), and Ministry of Science and Technology of the People's Republic of China (2014BAC21B02).
Appendix A. Supplementary data
Supplementary data to this article can be found online at http://dx.
doi.org/10.1016/j.scitotenv.2016.07.219.
References
Agrawal, H., Malloy, Q.G.J., Welch, W.A., Miller, J.W., Cocker, D.R., 2008. In-use gaseous
and particulate matter emissions from a modern ocean going container vessel.
Atmos. Environ. 42, 55045510.
Bandemehr, A., Muehling, B., Corbett, J., Comer, B., Boyle, J., May 2015. U.S.-Mexico cooperation on reducing emissions from ships through a Mexican emission control area:
development of the rst national Mexican emission inventories for ships using the
waterway network ship trafc, energy, and environmental model (STEEM). Ofce
of International and Tribal Affairs EPA-160-R-15-001. Available from https://www.
epa.gov/nscep.
Bao, X.F., Ding, Y., Yin, H., Huang, Z.H., Wang, J.F., Ma, D., 2014. Guideline of method for
emissions estimation from non-road mobile source. Available from http://www.
qhepb.gov.cn/xwzx/zqyj/201407/W020140711608259681099.pdf (in Chinese).
Cooper, D.A., 2003. Exhaust emission from ships at berth. Atmos. Environ. 37, 38173830.
Corbett, J.J., 2002. Emissions from ships in the northwestern United States. Environ. Sci.
Technol. 36, 12991306.
Corbett, J.J., March 2010. Improved geospatial scenarios for commercial marine vessels.
Prepared for the California Air Resources Board and the California Environmental Protection Agency (CN07-320).
Corbett, J.J., Windebrake, J.J., Green, E.H., Kasibhatla, P., Eyring, V., Lauer, A., 2007. Mortality from ship emissions: a global assessment. Environ. Sci. Technol. 24, 85128518.
Dalsren, S.B., Endresen, ., Isaksen, I.S.A., Gravir, G., Srgard, E., 2007. Environmental impacts of the expected increase in sea transportation, with a particular focus on oil and
gas scenarios for Norway and northwest Russia. J. Geophys. Res. 112, D02310. http://
dx.doi.org/10.1029/2005JD006927.
Endresen, O., Srgard, E., Behrens, H.L., Brett, P.O., 2007. A historical reconstruction of
ships' fuel consumption and emissions. J. Geophys. Res. 112, D12301. http://dx.doi.
org/10.1029/2006JD007630.
Endresen, O., Sorgard, E., Sundet, J.K., Dalsoren, S.B., Isaksen, I.S.A., Berglen, T.F., Gravir, G.,
2003. Emission from international sea transportation and environmental impact.
J. Geophys. Res. 108 (D20), 46504665. http://dx.doi.org/10.1029/2003JD003751.
Entec UK Limited, November 2010. UK Ship Emissions Inventory Final Report.
Fu, Q.Y., Shen, Y., Zhang, J., 2012. Study on ship emission inventory in shanghai port. J. Saf.
Environ. 12 (5), 5763 in Chinese with abstract in English.
Gutierrez, J.M., Grados, V.D., Uriondo, Z., Llamas, J.A., 2012. Emission-factor uncertainties
in maritime transport in the Strait of Gibraltar, Spain. Atmos. Meas. Tech. Discuss. 5,
59535991.
He, M., Zheng, J.Y., Yin, S.S., Zhang, Y.Y., 2011. Trends, temporal and spatial characteristics,
and uncertainties in biomass burning emissions in the Pearl River Delta, China.
Atmos. Environ. 45, 40514059.
Hong Kong Environment Protection Department (H.K.EPD), July 2015. Ocean going vessels fuel switch at berth. English version can available online at http://www.epd.
gov.hk/epd/sites/default/les/epd/english/environmentinhk/air/guide_ref/les/
Leaet%20for%20Fuel%20Switch%20at%20Berth%20Regulation%20English.pdf.
Hulskotte, J.H.J., Denier, H.A.C., 2010. Fuel consumption and associated emissions from
seagoing ships at berth derived from an on-board survey. Atmos. Environ. 44,
12291236.
ICF international, April 2009. Current Methodologies in Preparing Mobile Source Port-related Emission Inventories Final Report.
International Maritime Organization (IMO), May 2010. Information on North American
Emission Control Area under MARPOL ANNEX VI. English version can available online at https://www3.epa.gov/nonroad/marine/ci/mepc1-circ-re-na-eca.pdf.
Jalkanen, J.P., Brink, A., Kalli, J., Pettersson, H., Kukkonen, J., Stipa, T., 2009. A modelling
system for the exhaust emissions of marine trafc and its application in the Baltic
Sea area. Atmos. Chem. Phys. 9, 92099223.

Jalkanen, J.P., Johansson, L., Kukkonen, J., Brink, A., Kalli, J., Stipa, T., 2011. Extension of an
assessment model of ship trafc exhaust emissions for particulate matter and carbon
monoxide. Atmos. Chem. Phys. 11, 2212922172.
Lu, Z., Streets, D.G., Zhang, Q., Wang, S., Carmichael, G.R., Cheng, Y.F., Wei, C., Chin, M.,
Diehl, T., Tan, Q., 2010. Sulfur dioxide emissions in China and sulfur trends in East
Asia since 2000. Atmos. Chem. Phys. 10, 63116331.
MARIN, June 2010. Emissions2008: Netherlands Continental Shelf, Port Areas and OSPAR
Region II.
MARIN, June 2011. Emissions2009: Netherlands Continental Shelf, Port Areas and OSPAR
Region II.
Merk, O., December 2014. Shipping emissions in ports. Discussion Paper no. 2014-20
http://dx.doi.org/10.1787/5jrw1ktc83r1-en.
Ministry of Transport of China [internet], 2015. Implementation Plan on Domestic Emission Control Areas in Waters of the Pearl River Delta, the Yangtze River Delta and
Bohai Rim (Beijing, Tianjin, Hebei), 2015-01183. English version can available online
at http://zizhan.mot.gov.cn/zfxxgk/bzsdw/bhsj/201512/P020160215528398838765.
pdf.
NARSTO, 2005. Improving emission inventories for effective air quality management
across North America: a NARSTO assessment. Available from www.narsto.org.
Ng, S.K.W., Lin, C., Chan, J.W.M., Yip, A.C.K., Lau, A.K.H., Fung, J.C.H., February 2012. Study
on Marine Vessels Emission Inventory Final Report.
Ng, S.K.W., Loh, C., Lin, C., Booth, V., Chan, J.W.M., Yip, A.C.K., Li, Y., Lau, A.K.H., 2013. Policy
change driven by an AIS-assisted marine emission inventory in Hong Kong and the
Pearl River Delta. Atmos. Environ. 76, 102112.
Shanghai International Shipping Institute, December 2015. Report on Global Ports Development.
Available from http://sisi-smu.org/viewarticle.asp?ArticleID=
43756&ClassID=174 (in Chinese).
Song, S., 2014. Ship emissions inventory, social cost and eco-efciency in Shanghai
Yangshan port. Atmos. Environ. 82, 288297.
Song, Y.N., 2015. Research of Emission Inventory and Emission Character of Inland and
Offshore Ships. Beijing Institute of Technology, Beijing, China A thesis for the degree
of master. (in Chinese with abstract in English).
Starcrest Consulting Group, December 2009. The Port of Los Angeles Inventory of Air
Emissions for Calendar Year 2008. Technical Report. ADP# 050520-525. Available
from https://www.portoosangeles.org.
Starcrest Consulting Group, July 2014. Port of Los Angeles Inventory of Air Emissions2013. Technical Report. ADP# 131016-541. Available from https://www.
portoosangeles.org.
Tan, J.W., Song, Y.N., Ge, Y.S., Li, J.Q., Li, L., 2014. Emission inventory of ocean-going vessels
in Dalian Coastal area. Research of Environment Sciences 27 (12), 14261431 in Chinese with abstract in English.
Todd, S., Andrew, A., 2007. A Critical Review of Ocean-Going Vessel Particulate Matter
Emission Factors. California Air Resources Board.
U.S. Environmental Protection Agency (U.S.EPA), February 2000. Analysis of commercial
marine vessels emissions and fuel consumption data (EPA420-R-00-002).
U.S. Environmental Protection Agency (U.S.EPA), September 2008. Control of Emissions
from Marine SI and Small SI Engines, Vessels, and Equipment (EPA420-R-08-014).
Wang, C.F., Corbett, J.J., Firestone, J., 2007. Modeling energy use and emissions from North
American shipping: application of the ship trafc, energy, and environment model.
Environ. Sci. Technol. 41, 32263232. http://dx.doi.org/10.1021/es060752e.
Wang, C.F., Corbett, J.J., Firestone, J., 2008. Improving spatial representation of global ship
emissions inventories. Environ. Sci. Technol. 42, 193199. http://dx.doi.org/10.1021/
es0700799.
Winther, M., Christensen, J.H., Plejdrup, M.S., Ravn, E.S., Eriksson, O.F., Kristensen, H.O.,
2014. Emission inventories for ships in the arctic based on satellite sampled AIS
data. Atmos. Environ. 91, 114.
Yang, D.Q., Kwan, S.H., Lu, T., Fu, Q.Y., Cheng, J.M., Streets, D.G., Wu, Y.M., Li, J.J., 2007. An
emission inventory of marine vessels in Shanghai in 2003. Environ. Sci. Technol. 41,
51835190.
Yang, J., Yin, P.L., Ye, S.Q., Wang, S.S., Zheng, J.Y., Ou, J.M., 2015. Marine emission inventory
and its temporal and spatial characteristics in the city of Shenzhen. Environ. Sci. 36
(4), 12171226 in Chinese with abstract in English.
Yau, P.S., Lee, S.C., Corbett, J.J., Wang, C.F., Cheng, Y., Ho, K.F., 2012. Estimation of exhaust
emission from ocean-going vessels in Hong Kong. Sci. Total Environ. 431, 299306.
Ye S.Q., 2014. Study of Characteristics of Marine Vessel Emission and its Impact on Regional Air Quality of the Pearl River Delta Region. South China University of Technology, Guangzhou, China. (A thesis for the degree of master). (in Chinese with abstract
in English).
Ye, S.Q., Zheng, J.Y., Pan, Y.Y., Wang, S.S., Lu, Q., Zhong, L.J., 2014. Marine emission inventory and its temporal and spatial characteristics in Guangdong Province. Acta Sci.
Circumst. 34 (3), 537547 in Chinese with abstract in English.
Zhang, F., Chen, Y., Tian, C., Li, J., Zhang, G., Matthias, V., 2015. Emissions factors for gaseous and particulate pollutants from offshore diesel engine vessels in China. Atmos.
Chem. Phys. 15, 135.
Zhao, Y., Zhang, J., Nielsen, C.P., 2013. The effects of recent control policies on trends in
emissions of anthropogenic atmospheric pollutants and CO2 in China. Atmos. Chem.
Phys. 13, 487508.
Zheng, J.Y., Zhong, Z.M., et al., 2016. Recent Development of Air Pollutant Emission Inventory in Guangdong Province In preparation.
Zhong, L.J., P.K.K, L., Zheng, J.Y., Yuan, Z.B., Yue, D.L., Ho, J.W.K., Lau, A.K.H., 2013. Sciencepolicy interplay: air quality management in the Pearl River Delta region and Hong
Kong. Atmos. Environ. 76, 310.