Recurring transboundary haze from Indonesian wildfires in previous
decades significantly elevated particulate matter (PM) concentrations in
Southeast Asia. During that event on 10 to 31 October 2015, we conducted a
real-time observation of non-refractory submicron PM (NR-PM
Transboundary haze caused by wildfires has been a recurring issue in Southeast Asia for the past few decades, causing economic and health problems (Atwood et al., 2013; Engling et al., 2014; Heil and Goldammer, 2001; Nichol, 1997, 1998; Pavagadhi et al., 2013). Haze occurrence has been associated with anomalously low precipitation induced by the El Niño Southern Oscillation and the Indian Ocean Dipole (Field et al., 2009, 2016; Gaveau et al., 2015). Wildfires are especially pronounced in Indonesian tropical rainforests and peatlands due to land clearing activities, including the extensive development of agricultural activities (Dennis et al., 2005; Murdiyarso et al., 2004; Siegert et al., 2001). During the 1997 El Niño period, the wildfires in Indonesia consumed both peat and surface vegetation and caused a severe transboundary haze (Heil and Goldammer, 2001).
A prolonged wildfire event occurred in Indonesia during September and October 2015, coinciding with an intense El Niño period (Field et al., 2016). The scale of this 2015 wildfire is thought to be the largest after the wildfire event in 1997, and was estimated to release 227 Tg of carbon into the atmosphere (Huijnen et al., 2016). These carbon emissions in 2015 were at least on an order of magnitude larger than those emitted by the 2013 wildfires in Indonesia (31 Tg of carbon; Gaveau et al., 2015), although lower than 800–2500 Tg of carbon released from the wildfire in 1997 (Page et al., 2002). By using atmospheric chemistry transport models, previous studies calculated the excess mortality rate associated with the 2015 wildfire haze to be between 11 000 to 100 000 individuals across Indonesia, Malaysia, and Singapore (Crippa et al., 2016; Koplitz et al., 2016). These model estimations already consider emission, atmospheric processing, and removal of gas and particulate matter (PM) emitted by the wildfires. However, atmospheric processing of aerosol particles in haze from Indonesian wildfires has scarcely been investigated. This lack of study inhibits a detailed treatment of atmospheric chemical processes in the models, including aerosol aging and secondary aerosol formation. Laboratory studies have shown that atmospheric processing of biomass burning smoke resulted in organic aerosol (OA) enhancement due to secondary organic aerosol (SOA) formation (Cubison et al., 2011; Ortega et al., 2013). Although SOA has been shown to significantly contribute to OA concentration in other parts of the world (e.g., Huang et al., 2014; Weber et al., 2007), the contribution of SOA formation in haze particles from wildfires in the Southeast Asia has never been quantified.
Online aerosol measurement techniques, such as the aerosol mass spectrometer (AMS) and the aerosol chemical speciation monitor (ACSM) developed by Aerodyne Inc., are useful for investigating atmospheric processing of aerosol particles, due to their high time resolution for chemical characterization of bulk aerosol composition (Jayne et al., 2000; Ng et al., 2011a). These techniques quantify the chemical characteristics of bulk OA, allowing further multivariate factor analysis of the mass spectra matrix by positive matrix factorization (PMF) (Ulbrich et al., 2009; Q. Q. Zhang et al., 2011) or multilinear engine (ME-2) solvers (Canonaco et al., 2013; Crippa et al., 2014). These multivariate factor analyses have been shown to be useful for identifying sources and atmospheric processes of OA, especially when combined with offline analytical techniques providing the molecular-level chemical composition (e.g., Budisulistiorini et al., 2013; Zhang et al., 2005a).
In this study, we conducted an atmospheric observation of
haze particles in Singapore induced by Indonesian wildfires using the
time-of-flight ACSM (ToF-ACSM) from 10 to 31 October 2015. We characterized
subtypes of ambient OA in the non-refractory PM
We conducted the 2015 haze measurement from 10 to 31 October 2015 on the
fourth floor of the North Spine building at Nanyang Technological University
(NTU), Singapore (1.3483
Ambient NR-PM
The time series of OA mass spectra were deconvolved using an ME-2 solver,
Source Finder (SoFi) tool version 6.2 written in Igor Pro (Canonaco et al.,
2013). Details of the ME-2 analysis are described elsewhere (Canonaco et al.,
2013; Crippa et al., 2014; Elser et al., 2016; Fröhlich et al., 2015) as
well as in the Supplement (Sect. B). Briefly, a solution was first explored
without any a priori information, meaning that no reference mass spectra
were provided as constraints. Subsequently, mass spectra which originated
from previous field and laboratory observations were introduced to constrain
one or more solution spectra. The strength of the constraints was adjusted by
varying the degree of freedom (
In this study, mass spectra from ambient measurements, such as
hydrocarbon-like OA (HOA) and biomass burning OA (BBOA) (Ng et al., 2011c)
and laboratory-generated peat burning OA (PBOA; Budisulistiorini et al.,
2017a), were employed to allow each factor to vary within a specific range.
The factor solution variability (
Ambient aerosols were sampled through a PM
The second filter holder was operated at a flow rate of 4.2 L min
EC and OC were quantified using a semicontinuous field type EC/OC analyzer
(Sunset Laboratories Inc.), following the IMPROVE-A protocol (Miyakawa et
al., 2015). The instrument was calibrated using a standard sucrose solution.
The particulate OC was estimated by subtracting the back-filter OC from the
front-filter OC, assuming that (1) particulate OC was collected by the front
filter, and (2) the gas-phase OC was collected on both the front and back
filters. Analysis of WSOC was performed using a total organic
carbon
(TOC) analyzer (Model 800, Sievers, Colorado), following extraction of a
portion (8 mm in diameter) of filter samples by 10 mL of HPLC-grade water
(Kanto Chemical Co. Inc., Tokyo, Japan) and using an orbital shaker for
21 h. The extracted sample solutions were filtered by syringe filters (PTFE;
pore size 0.2
Average concentrations with 1 standard deviation (SD) of chemical species measured during the October 2015 haze episode in Singapore using ToF-ACSM (online) and integrated filter samples (offline) measurements. Wind speed and precipitation data were retrieved from the website of the Meteorological Service Singapore, and other meteorological parameters were from measurement at the sampling location.
For the inorganic ion analysis, a portion of the filter samples
(1 cm
The second set of samples was analyzed by gas and liquid chromatographic
techniques interfaced to mass spectrometry for molecular-level analysis.
Details of the chromatography techniques and some of the results for OA
tracer analysis have been reported in Budisulistiorini et al. (2017a).
Briefly, the liquid chromatography system consists of a ultra-performance liquid chromatograph
coupled in line to both a diode array detector and high-resolution quadrupole time-of-flight
mass spectrometer equipped with an electrospray ionization source (UPLC/DAD-ESI-HR-QTOFMS, hereafter called LC-MS; 6520 Series,
Agilent, California) using a Waters ACQUITY UPLC HSS T3 column
(
The gas chromatograph was coupled in line to a mass spectrometer equipped with electron impact ionization (GC/EI-MS,
hereafter called GC-MS; 5890 Hewlett Packard, HP, Series II gas
chromatograph) and interfaced to an HP 5971A Series mass selective detector. An
Econo-Cap™ EC-5 column (30 m
We measured elemental compositions (carbon, hydrogen, nitrogen, and sulfur)
of peat and vegetation samples from Riau and Central Kalimantan, Indonesia,
using a 2400 CHNS elemental analyzer (Perkin Elmer). Samples were dried at
105
The NOAA HYSPLIT back-trajectory model (Rolph et al., 2017; Stein et al.,
2015) was used for estimating the origins of air masses. The trajectories
were calculated from the observation site at 200 and 500 m above ground
level (a.g.l.) at 06:00 LT of each day during the entire measurement period.
In addition, trajectories were also estimated for four time periods (06:00,
12:00, 18:00, and 00:00 LT) of 19 and 20 October from the observation site
at 500 m a.g.l. Fire hotspots in Indonesia during October 2015 were
retrieved from near-real-time (NRT) Moderate Resolution Imaging
Spectroradiometer (MODIS) Thermal Anomalies/Fire Locations Collection 6
processed by the Land, Atmosphere Near real-time Capability for EOS (Earth
Observing System) Fire Information for Resource Management System
(LANCE-FIRMS, 2015). The air quality monitoring data, including 1 h
PM
The back-trajectories of air masses during the haze episode in Singapore are shown in Fig. S5. Air masses originated from south and southeast of Singapore, likely carried haze from wildfires in Sumatra and Kalimantan, Indonesia. Towards the end of the sampling period (29–31 October) the wind direction shifted, transporting air masses from northeast and east of Singapore. Air masses during this period might have been more dominated by local emissions as well as regional emissions from Malaysia and the South China Sea (e.g., ship emissions and sea spray; Betha et al., 2014). Therefore, we classified the measurement period into P1 (for 10–28 October), and P2 (for 29–31 October) in accordance with the wind directions. The highest concentration of haze particles was observed during 19 and 20 October (hereafter referred to as the peak event). Back-trajectories of air masses arriving at Singapore during this period show that the air mass was influenced by wildfires in Kalimantan and Sumatra, Indonesia (Fig. S6).
Time profiles of the meteorological parameters and aerosol
concentrations. Daily
Figure 1a–c shows a time series of meteorological data. Wind speed and
temperature measured at the sampling site during the P1 period were on
average
Figure 1d shows a time series of mass concentration of aerosol species
measured by the ToF-ACSM as well as PM
Sulfate (
Ammonium (
Nitrate (
Chloride (
Average diurnal profiles of NR-PM
The average diurnal profile of NR-PM
The
Ratios of
Table 2 summarizes the ratios of organic matter (OM, which is OA from
ToF-ACSM) to OC (
Figure S10 shows concentrations of OA tracer compounds and OC measured by the offline analysis, and OA measured by the ToF-ACSM. The OA concentration was averaged over filter sampling time. Brown carbon (BrC) is a sum of light-absorbing constituents identified in Budisulistiorini et al. (2017a). These tracers are used in identifying OA sources resolved by the ME-2 analysis (Table S4).
The average OA mass spectra during the entire campaign (P1 and P2 periods)
are shown in Fig. 3. The mass spectra are generally similar with subtle
differences in the intensity of some ion signals, such as ions at
Average mass spectra of OA during the entire sampling period (top) and the P1 and P2 periods.
Figure 4a and b show the mass spectra and time series of the OA factors,
respectively. To support the identification of the OA factors, we compare the
OA factors with chemical species identified by offline analyses (i.e., LC-MS,
GC-MS, Sunset OC/EC, and IC-MS). Using the offline analyses, we characterized
ambient particle tracers, such as levoglucosan and BrC constituents
(Budisulistiorini et al., 2017a), EC and inorganic cations. Table S4 presents
the correlation between the OA factors and the ambient particle tracers. The
HOA factor, which is identified by a distinctive signal at
The BBOA factor can be attributed to vegetative burning. The MS of BBOA
shows significant ion signals at
We identified the PBOA factor by a good correlation (
Potassium (
The OOA component was identified by the prominent ion signal at
Figure 5 shows the average diurnal variations in the OA factors during the P1 and P2 periods. During the P1 period (Fig. 5a and c), HOA concentration increased in the morning (07:00–09:00 LT) and evening (18:00–23:00 LT) implying that this HOA factor can be associated with morning and evening traffic. A small decrease in the night peak when the peak event (19–20 October) was excluded from diurnal profile calculation (Fig. S12a) proves that the night peak of HOA during the P1 period was mainly influenced by night traffic. Diurnal variations in BBOA and PBOA show significant peaks at nighttime driven by the peak event during the P1 period, although no substantial variations were observed during daytime (Fig. S12b–c). The OOA concentration started to increase around 08:00 LT and peaked around 14:00 LT (Fig. 5a), indicating the contribution of photooxidation processes. The elevated OOA concentration at nighttime was also generated by the peak event on 19–20 October (Fig. S12d).
Average diurnal profiles of OA sources in units of
During the P2 period (Fig. 5b and d), the HOA concentration increased in the evening, while no significant change was observed in the morning. The HOA concentration in the morning of the P2 period was almost half of its concentrations in the afternoon and evening resulting in an insignificant concentration profile in the morning. The HOA enhancement in the evening can be associated with an evening traffic. An increase in the HOA concentration in the late afternoon (14:00–16:00 LT) could be caused by heavy vehicle traffic from construction sites within the NTU campus, as well as regular traffic relating to school activities. It should be noted that the P2 period was shorter (29–31 October) than the P1 period. Hence, the diurnal profile of OA factors might not capture the actual variation in OA sources during non-hazy days. The BBOA concentration increased at noontime, although no significant local source of biomass burning was available during the P2 period. The BBOA enhancement at noontime could be due to the transport of biomass burning plume event from Peninsular Malaysia that might occur during the short P2 period (Miettinen et al., 2017; Smith et al., 2017). The PBOA concentration did not vary significantly during the P2 period, suggesting that it might not be produced locally. The OOA concentration peaked around 14:00 LT and decreased in the evening, indicating that during the P2 period, OOA was likely formed through photooxidation.
The average concentration of OA sources to ambient fine aerosol during the
2015 haze episode in Singapore is illustrated in Fig. 6a. The average
concentrations of three OA components – BBOA (5.7
The changes in average concentration of OA factors during the P1 and P2
periods were not similar to the changes in contributions of OA factors to the
total OA masses. As illustrated in Fig. 6b, the contribution of OOA to the
total OA mass was higher during the P1 period (50.7 %) than during the P2
period (39.9 %). Similarly, the PBOA contribution was also higher during
the P1 period (16.8 %) than the P2 period (7.2 %). The higher
contribution of OOA and PBOA during the P1 period suggests that the factors
were influenced by Indonesian wildfires. The BBOA contribution was relatively
similar (
The oxidation degree of the total OA (and the OA factors) may provide
information of the OOA formation processes. Figure 7a illustrates the
relationship between the mass fraction of
The evolution of OA is also indicated by the diurnal profile of
Van Krevelen diagram of the OA measured during the 2015 haze episode
and the OA factors resolved by the ME-2 analysis. The
Figure 7b shows a relationship between
We estimated the ratios of
Overall, the observation of the 2015 haze episode in Singapore indicated the dominant contribution of OOA to the total OA. OOA formation could be caused by (1) oxidation of POA, and (2) the formation of SOA through oxidation of VOCs emitted from wildfires as well as partitioning of WSOGs onto aerosol liquid water or cloud droplets during atmospheric transport. This present study, however, could not separately quantify the contribution of POA oxidation and SOA formation to the total OOA. Therefore, further studies are needed to investigate the contribution of SOA formation and POA oxidation to the OOA formation in the wildfire haze particles.
We investigated contributions of inorganic and organic particles to the
Indonesian wildfire haze through directly measuring NR-PM
Characterization of OA sources by the ME-2 analysis yielded four components,
i.e., HOA, BBOA, PBOA, and OOA, which can be associated with vehicles and
traffics, wood or biomass burning, wildfires, and atmospheric processing,
respectively. OOA was the most abundant (
The relationships between
The data set for this publication is available upon contacting the corresponding authors.
The supplement related to this article is available online at:
SHB and MK designed the study. SHB performed the ambient and laboratory measurements, analyzed the ToF-ACSM data, and wrote the manuscript with contributions from MK. MR, MW, and JDS conducted the LC-MS and GC-MS analysis. TM conducted the inorganic ions, OC, EC, and WSOC analyses. JC performed the ambient and laboratory measurements. MI provided the biomass samples.
The authors declare that they have no conflict of interest.
We acknowledge Haris Gunawan for supporting our research in Indonesia. We thank
Gisella B. Lebron and Web-Chien Lee for assisting in the particle sampling and data
collection, and Kyle Niezgoda and Shaoneng R. He for the meteorological data. We
gratefully acknowledge the NOAA Air Resources Laboratory (ARL) for the
provision of the HYSPLIT transport and dispersion model and/or READY website
(