Impacts of Biomass Burning Emission Inventories and Atmospheric Reanalyses on Simulated PM10 over Indochina

Biomass burning (BB) is a major source of atmospheric particles over Indochina during the dry season. Moreover, Indochina has convoluted meteorological scales, and regional meteorological conditions dominate the transport patterns of pollutants. This study focused on the impacts of BB emission inventories and atmospheric reanalyses on simulated PM10 over Indochina in 2014 using the Community Multiscale Air Quality (CMAQ) model. Meteorological fields to input to CMAQ were produced by using the Weather Research and Forecasting (WRF) model simulation with the United States National Centers for Environmental Prediction Final (NCEP FNL) Operational Global Analysis or European Centre for Medium Range Weather Forecasts Interim Reanalysis (ERA-interim). The Fire INventory from NCAR (FINN) v1.5 or the Global Fire Emissions Database including small fires (GFED v4.1s) was selected for BB emissions for the air quality simulation. The simulation case with NCEP FNL and FINN v1.5 (FNL + FINN) performed best throughout 2014, including the season when BB activities were intensified. The normalized percentage difference for maximum daily mean PM10 concentrations at Chiang Mai for FNL + FINN and the two simulation cases applying GFED v4.1s for BB emissions (−53% to −27%) was much larger than that between the FNL + FINN and ERA + FINN cases (10%). BB emission inventories more strongly impacted PM10 simulation than atmospheric reanalyses in highly polluted areas by BB over Indochina in 2014.


Introduction
Indochina, the peninsular region that includes Cambodia, Laos, Myanmar, Thailand, and Vietnam, has a monsoonal climate with a dry and wet season; the prevailing wind direction and the pattern of precipitation changes seasonally. During the dry season, particulate pollution degrades air quality over Indochina [1]. Outdoor exposure to particulate matter (PM) contributes to ill health, such as cardio-and cerebrovascular disease, respiratory illnesses, lung cancer, and possibly other diseases [2]. A major source of atmospheric particles over Indochina is biomass burning (BB), such as forest fires and agricultural burning. Thailand is surrounded by mountainous areas where BB frequently occurs in the dry season. Chiang Mai, the largest city in northern Thailand, has experienced severe air pollution caused by BB [3][4][5]. Furthermore, measurements of aerosol properties on Dongsha Island in the northeastern South China Sea revealed that smoke originating from BB in Indochina rises and is trapped within the free troposphere (3-4 km above the earth's surface) in some situations [5][6][7].
As part of NASA's 2006 Biomass Burning Aerosols in Southeast Asia: Smoke Impact Assessment (BASE-ASIA), regional modeling studies were conducted to assess the impacts of BB in Southeast

Meteorological Model
Meteorological fields to input to CMAQ were produced by using the Weather Research and Forecasting (WRF) model [22] v3.4. The WRF simulations were conducted using the high-resolution, real-time, global sea surface temperature analysis (RTG_SST_HR) of NCEP, and two atmospheric reanalyses: NCEP FNL or ERA-Interim. The physics options were as follows: Rapid Radiative Transfer Model for General Circulation Models Shortwave and Longwave Schemes [23], planetary boundary layer physics of Asymmetric Convection Model 2 Scheme [24], cumulus parameterization of the Kain-Fritsch Scheme [25], micro physics of Morrison 2-moment Scheme [26], and Pleim-Xiu Land Surface Model [27]. Grid nudging was applied to horizontal wind components at all the vertical layers with a coefficient of 3.0 × 10 −4 s −1 in D1 and 1.0 × 10 −4 s −1 in D2 for the entire period.

Meteorological Model
Meteorological fields to input to CMAQ were produced by using the Weather Research and Forecasting (WRF) model [22] v3.4. The WRF simulations were conducted using the high-resolution, real-time, global sea surface temperature analysis (RTG_SST_HR) of NCEP, and two atmospheric reanalyses: NCEP FNL or ERA-Interim. The physics options were as follows: Rapid Radiative Transfer Model for General Circulation Models Shortwave and Longwave Schemes [23], planetary boundary layer physics of Asymmetric Convection Model 2 Scheme [24], cumulus parameterization of the Kain-Fritsch Scheme [25], micro physics of Morrison 2-moment Scheme [26], and Pleim-Xiu Land Surface Model [27]. Grid nudging was applied to horizontal wind components at all the vertical layers with a coefficient of 3.0 × 10 −4 s −1 in D1 and 1.0 × 10 −4 s −1 in D2 for the entire period.  Meteorological model Chemical transport model Figure 2. Procedure of air quality simulation by meteorological model and chemical transport model.

Meteorological Model
Meteorological fields to input to CMAQ were produced by using the Weather Research and Forecasting (WRF) model [22] v3.4. The WRF simulations were conducted using the high-resolution, real-time, global sea surface temperature analysis (RTG_SST_HR) of NCEP, and two atmospheric reanalyses: NCEP FNL or ERA-Interim. The physics options were as follows: Rapid Radiative Transfer Model for General Circulation Models Shortwave and Longwave Schemes [23], planetary boundary layer physics of Asymmetric Convection Model 2 Scheme [24], cumulus parameterization of the Kain-Fritsch Scheme [25], micro physics of Morrison 2-moment Scheme [26], and Pleim-Xiu Land Surface Model [27]. Grid nudging was applied to horizontal wind components at all the vertical layers with a coefficient of 3.0 × 10 −4 s −1 in D1 and 1.0 × 10 −4 s −1 in D2 for the entire period.

Chemical Transport Model
CMAQ was configured with the Carbon Bond chemical mechanism (CB05) [28] for gas-phase chemistry and the sixth generation CMAQ aerosol module for the aerosol process. The initial and lateral boundary conditions for D1 were obtained from the Model for Ozone and Related Chemical Tracers v4 (MOZART-4) [29]. Several datasets were used to produce emission data. In particular, the Fire INventory from NCAR (FINN) [11] v1.5 or the Global Fire Emissions Database including small fires (GFED v4.1s) [30] was selected for BB emissions. FINN v1.5 provided 1 km-resolution BB emissions estimated from the active fire observations from the Moderate Resolution Imaging Spectroradiometer (MODIS) Terra and Aqua satellite. GFED v4.1s provided 0.25 • -resolution BB emissions estimated from the 500 m Collection 5.1 MODIS direct broadcast burned area product plus additional burned area from small fires based on a revised version of the Randerson et al. [31] small-fire estimation approach. Anthropogenic emissions were obtained from the Task Force Hemispheric Transport of Air Pollution (HTAP) v2.2 inventory [32]. The other natural emissions were derived from the Model of Emissions of Gases and Aerosols from Nature (MEGAN) [33] v2.04 for biogenic emissions and the Aerosol Comparisons between Observations and Models (AEROCOM) data [34] for baseline volcanic SO 2 emissions. There were differences between the periods of simulations and the reference years in the anthropogenic emission data available for the simulations, and the uncertainties associated with the differences may have affected model performance.
In order to evaluate the impacts of BB emission inventories and atmospheric reanalyses on simulated PM 10 concentrations, four cases of simulations (FNL + FINN; FNL + GFED; ERA + FINN; and ERA + GFED) were executed. The first parts of cases' names indicate the applied reanalysis: FNL and ERA stand for NCEP FNL and ERA-Interim, respectively. The second parts indicate the selected BB emission inventories. Furthermore, simulations without BB emission inventories (FNL + exBB; ERA + exBB) were conducted. The difference between cases with BB emissions and those without BB emissions was calculated to estimate BB contributions to PM 10 concentrations in the four simulation cases with BB emissions. Figure 3 shows spatial distributions of PM 10 emissions from FINN v1.5 and GFED v4.1s in March and April 2014. During these two months, PM 10 emissions in both BB emission inventories were concentrated in Indochina, especially the mountainous areas in Laos, Myanmar, and Thailand. FINN v1.5 estimated larger emissions in a wider area than GFED v4.1s. Figure 4 shows the variation of the amount of PM 10 emissions over D2 and the relative contributions from BB. PM 10 emissions from BB produced by FINN v1.5 and GFED v4.1s accounted for 244 mg s −1 km −2 and 91 mg s −1 km −2 , respectively, which contributed 93% and 83%, respectively, of the total amount of PM 10 emissions in March when burning activities were intensified.
The WRF performance was evaluated through comparison with ground-level air temperature, relative humidity, and wind speed data in Bangkok and Chiang Mai distributed by the University of Wyoming [35]. The CMAQ performance was evaluated through comparison with ground-level PM 10 concentration data in Bangkok and Chiang Mai distributed by the Acid Deposition Monitoring Network in East Asia (EANET) [36].

Daily Meteorological Factors at Observation Sites
The WRF performance was evaluated by using the normalized mean bias (NMB) and the index of agreement (IA) for ground-level air temperature, relative humidity, and wind speed for Bangkok and Chiang Mai (Table 1). More statistics were shown in supplementary data (Table S1). IA is a statistical measure to present the model performance and varies between 0 and 1, where 1 means model performance is perfect [37]. IA values for air temperature, relative humidity, and wind speed were relatively high in both WRF simulations: the models applying both NCEP FNL and ERA-interim simulated spatial and temporal variations of these parameters in the same level.

Daily Concentrations of PM 10 at Observation Sites
The performance of daily mean PM 10 concentration simulations in the four simulation cases with BB emissions was evaluated by using NMB and IA for the periods of both the entire year and the two month period from March to April 2014 (Table 2). More statistics were shown in supplementary data (Table S2). The highest PM 10 Figure 5 shows simulated and observed daily concentrations at Bangkok and Chiang Mai. All of the four simulation cases simulated the day-to-day variation patterns well at both sites for the year 2014. Cases applying FINN v1.5 for BB emissions (FNL + FINN; ERA + FINN) showed higher trends of simulated PM 10 concentrations at both observation sites compared to those applying GFED v4.1s (FNL + GFED; ERA + GFED), regardless of the atmospheric reanalysis. In particular, the former two cases clearly presented higher trends of PM 10 concentrations than the latter cases at Chiang Mai from the end of February to early April, when high PM 10 concentrations were observed. Table 3 shows the normalized percentage difference for FNL + FINN and the other three cases for annual mean and maximum daily mean PM 10 concentrations in 2014 at the two observation sites. The percentage difference at Bangkok (−9% to −4% for annual mean; −15% to −2% for maximum) was less than that at Chiang Mai (−19%-8% for annual mean; −53%-10% for maximum). At both of the two sites, the percentage difference between the FNL + FINN and ERA + FINN cases was the smallest for annual mean and maximum daily mean PM 10 concentrations in 2014. In particular, the difference for maximum daily mean PM 10 concentrations at Chiang Mai for FNL + FINN and the two cases applying GFED v4.1s for BB emissions (−53% to −27%) was much larger than that between the FNL + FINN and ERA + FINN cases (10%). Consequently, BB emission inventories more strongly impacted the PM 10 simulation than atmospheric reanalyses at the two sites in Indochina.
Atmosphere 2020, 11, x FOR PEER REVIEW 7 of 13  Figure 5 shows simulated and observed daily concentrations at Bangkok and Chiang Mai. All of the four simulation cases simulated the day-to-day variation patterns well at both sites for the year 2014. Cases applying FINN v1.5 for BB emissions (FNL + FINN; ERA + FINN) showed higher trends of simulated PM10 concentrations at both observation sites compared to those applying GFED v4.1s (FNL + GFED; ERA + GFED), regardless of the atmospheric reanalysis. In particular, the former two cases clearly presented higher trends of PM10 concentrations than the latter cases at Chiang Mai from the end of February to early April, when high PM10 concentrations were observed. Table 3 shows the normalized percentage difference for FNL + FINN and the other three cases for annual mean and maximum daily mean PM10 concentrations in 2014 at the two observation sites. The percentage difference at Bangkok (−9% to −4% for annual mean; −15% to −2% for maximum) was less than that at Chiang Mai (−19%-8% for annual mean; −53%-10% for maximum). At both of the two sites, the percentage difference between the FNL + FINN and ERA + FINN cases was the smallest for annual mean and maximum daily mean PM10 concentrations in 2014. In particular, the difference for maximum daily mean PM10 concentrations at Chiang Mai for FNL + FINN and the two cases applying GFED v4.1s for BB emissions (−53% to −27%) was much larger than that between the FNL + FINN and ERA + FINN cases (10%). Consequently, BB emission inventories more strongly impacted the PM10 simulation than atmospheric reanalyses at the two sites in Indochina.   BB contributions were estimated by the difference between cases with and without BB emissions ( Figure 5). The BB contributions to PM 10 concentrations of the four simulation cases were 2%-12% at Bangkok and 75%-89% at Chiang Mai for each day that the maximum concentrations were observed. Li et al. [12] reported that BB contributions to simulated PM 2.5 was 70%-80% in the BB source regions during March-April 2013 and the contributions were close to 75%-89% at Chiang Mai on 21 March 2014. Figure 6 shows simulated and observed wind speed, planetary boundary layer (PBL) height, and daily concentrations at Bangkok and Chiang Mai from March to April. At Bangkok, these parameters were on the similar trends among the four simulation cases. At Chiang Mai, the maximum PM 10 concentration was observed in weak-wind condition. On the other hand, there was little relationship between PBL height and PM 10 concentrations. The simulated spatial distributions on January 3 and March 21 are analyzed in detail in the next subsection.
Atmosphere 2020, 11, x FOR PEER REVIEW 8 of 13 Note: Mean-annual mean value, Maximum-maximum daily mean value.
BB contributions were estimated by the difference between cases with and without BB emissions ( Figure 5). The BB contributions to PM10 concentrations of the four simulation cases were 2%-12% at Bangkok and 75%-89% at Chiang Mai for each day that the maximum concentrations were observed. Li et al. [12] reported that BB contributions to simulated PM2.5 was 70%-80% in the BB source regions during March-April 2013 and the contributions were close to 75%-89% at Chiang Mai on 21 March 2014. Figure 6 shows simulated and observed wind speed, planetary boundary layer (PBL) height, and daily concentrations at Bangkok and Chiang Mai from March to April. At Bangkok, these parameters were on the similar trends among the four simulation cases. At Chiang Mai, the maximum PM10 concentration was observed in weak-wind condition. On the other hand, there was little relationship between PBL height and PM10 concentrations. The simulated spatial distributions on January 3 and March 21 are analyzed in detail in the next subsection.  Figure 7 shows the spatial distributions of daily PM10 concentrations and wind fields at ground level on January 3, when maximum PM10 concentration was observed in Bangkok. The large BB contribution area was spread out over Southern China and Cambodia. On January 3, ground-level winds were weak over Indochina. The area where simulated PM10 concentrations were around 80 µ g m −3 was spread across Thailand, and the BB contributions were relatively small in the four cases; discrepancy among the four cases was small across Thailand.

Spatial Distributions for Daily Concentrations of PM10
In March 2014, the largest number of hotspots were detected in Thailand and Myanmar from March 18 to 21 [5]. A large number of fire spots occurred in the mountainous areas of Indochina on  Figure 7 shows the spatial distributions of daily PM 10 concentrations and wind fields at ground level on January 3, when maximum PM 10 concentration was observed in Bangkok. The large BB contribution area was spread out over Southern China and Cambodia. On January 3, ground-level winds were weak over Indochina. The area where simulated PM 10 concentrations were around 80 µg m −3 was spread across Thailand, and the BB contributions were relatively small in the four cases; discrepancy among the four cases was small across Thailand.

Spatial Distributions for Daily Concentrations of PM 10
converged at the Myanmar and Thailand border in all cases. Concentrations of 160 µ g m −3 of PM10 were spread over the area along the Myanmar and Thailand border in the four cases with BB emissions; the two cases that FINN v1.5 selected for BB emissions were predicted to have higher PM10 concentrations in and around Northern Laos. In highly polluted areas caused by BB over Indochina, the discrepancy of simulated PM10 concentrations resulting from different BB emission inventories was larger than that resulting from different atmospheric reanalyses. In March 2014, the largest number of hotspots were detected in Thailand and Myanmar from March 18 to 21 [5]. A large number of fire spots occurred in the mountainous areas of Indochina on March 21 ( Figure 1). Figure 8 shows spatial distributions of daily PM 10 concentrations and wind fields at ground level on March 21. Ground-level winds blew easterly in the eastern part of Indochina and converged at the Myanmar and Thailand border in all cases. Concentrations of 160 µg m −3 of PM 10 were spread over the area along the Myanmar and Thailand border in the four cases with BB emissions; the two cases that FINN v1.5 selected for BB emissions were predicted to have higher PM 10 concentrations in and around Northern Laos. In highly polluted areas caused by BB over Indochina, the discrepancy of simulated PM 10 concentrations resulting from different BB emission inventories was larger than that resulting from different atmospheric reanalyses. 1

Conclusions
This study focused on the impacts of BB emission inventories and atmospheric reanalyses on the simulation of PM 10 over Indochina for the year 2014. Meteorological fields to input to CMAQ were produced by using the WRF model simulation with NCEP FNL or ERA-interim. FINN v1.5 or GFED v4.1s were selected for BB emissions. In order to evaluate the impacts, four cases of simulations (FNL + FINN; FNL + GFED; ERA + FINN; ERA + GFED) were executed.
The performance of CMAQ simulations for PM 10 concentrations at ground-level was evaluated at Bangkok and Chiang Mai in Thailand. The FNL + FINN case performed best for both the full year and the two-month period when BB activities were intensified in 2014. To compare the impacts of input data (BB emission inventories and atmospheric reanalyses) on simulated PM 10 , we introduced the normalized percentage difference for FNL + FINN and the other three cases (FNL + GFED; ERA + FINN; ERA + GFED) for annual mean and maximum daily mean PM 10 concentrations in 2014 at the two observation sites. At both of the two sites, the normalized percentage difference for annual mean and maximum daily mean PM 10 concentrations in 2014 for FNL + FINN and the two cases applying GFED v4.1s for BB emissions (FNL + GFED; ERA + GFED) was larger than that between FNL + FINN and ERA + FINN cases. In particular, the difference for maximum daily mean PM 10 concentrations at Chiang Mai for FNL + FINN and the two cases applying GFED v4.1s (−53% to −27%) was much larger than that between the FNL + FINN and ERA + FINN cases (10%). BB contributions were estimated by the difference between the cases with BB emissions and those without BB emissions. BB contributions to PM 10 concentrations in the four simulation cases at Chiang Mai (75%-89%) were larger than those at Bangkok (2%-12%) on the day when the highest PM 10 concentrations were observed in 2014 at each observation site. BB highly contributed on the particulate pollution over Southern china and Cambodia on January 3, and over the area along the Myanmar and Thailand border on March 21 in weak-wind condition. In polluted areas caused by BB, the cases applying FINN v1.5 for BB emission inventories showed higher trends of PM 10 concentrations. Selected BB emission inventories more strongly impacted the PM 10 simulation than selected atmospheric reanalyses in highly polluted areas by BB over Indochina in 2014.
This approach is applicable to other years and atmospheric pollutants to evaluate the impacts of selected input datasets on air quality simulations in polluted areas caused by BB. For example, GFAS can be used for one of the other BB emission inventories. However, emission factors of global BB emission inventories, such as FINN, GFAS, and GFED, were determined based on generic vegetation types of each fire pixel. BB-emission data focusing on the unique characteristics of BB in the specific region would reduce the uncertainties of BB emissions. We hope that this study will help to improve understanding about the mechanisms that degrade air quality over Indochina.
Supplementary Materials: The following are available online at http://www.mdpi.com/2073-4433/11/2/160/s1, Table S1. Model performance for daily mean meteorological factors in the two simulation cases for the periods of one year and two months (from March to April) in 2014. Table S2. Model performance for daily mean PM10 concentrations in the four simulation cases for the periods of one full year (2014) and two months (from March to April 2014).