62
W AGENINGEN U NIVERSITY AND R ESEARCH C ENTER METEOROLOGY AND AIR QUALITY (MAQ) C HAIR GROUP Simulating Air Pollution in the Severe Fires Event during 2015 El-Niño in Indonesia using WRF-Chem Author: Dyah Ayu P UTRININGRUM Reg. No.: 900612-206-130 Supervisor: Folkert BOERSMA Michiel Van Der MOLEN Maarten KROL March 29, 2017

Simulating Air Pollution in the Severe Fires Event during

  • Upload
    others

  • View
    4

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Simulating Air Pollution in the Severe Fires Event during

WAGENINGEN UNIVERSITY AND RESEARCH CENTER

METEOROLOGY AND AIR QUALITY (MAQ) CHAIR GROUP

Simulating Air Pollution in the Severe FiresEvent during 2015 El-Niño in Indonesia using

WRF-Chem

Author:Dyah Ayu PUTRININGRUMReg. No.:900612-206-130

Supervisor:Folkert BOERSMA

Michiel Van Der MOLENMaarten KROL

March 29, 2017

Page 2: Simulating Air Pollution in the Severe Fires Event during

Abstract

During October 2015 El-Niño episode, massive forest fires occurred in Indonesia causing an air pollu-tion catastrophe. A global composite image of satellite shows a high level (>3) of observed aerosol opticaldepth (AOD) over Indonesia. We investigate the temporal variability of aerosol concentration and CO mix-ing ratio using Weather Research Forecasting coupled with Chemistry (WRF-Chem v3.2) model. We runthe simulation without and with coupling the chemistry-aerosol mechanism, only the biomass burning inputused in the experiments is different. We employed FINN and GFED inventory in separate runs. Resultsshow that the meteorological and chemical variables underestimate the observations. The underestimationof the biomass burning inventory is due to the thick haze that hamper the burned area detection. We, there-fore, need to boost the emission by 30% of emission factor. The simulation with increased biomass burningemissions improve the results significantly, although an underestimation of simulated chemical variablesoccurred during a dense haze event (21-29 October 2015). To overcome this underestimation, we thenmodified the GFED emissions during 26-30 October 2015 with (1) repeating the emissions on 21-25 Octo-ber 2015 and (2) keeping the high emissions of 20 October until 30 October 2015. Modifying the biomassburning emissions improved the model quite well. In sensitivity test where we included biomass burningaerosols, the skill of the model in reproducing the observed spatial variation of precipitation increases.Moreover, the run with biomass burning aerosol shows the reduction in precipitation over the area of highbiomass burning emissions. Finally, we suggest investigating the biomass burning inventories quality inseveral future types of research for gaining a better simulation for an extreme event such as 2015 Indonesiaforest fires.

Keywords: WRF-Chem, FINN, GFED, aerosols, precipitation.

i

Page 3: Simulating Air Pollution in the Severe Fires Event during

AcknowledgementThe supervision by Folkert Boersma, Michiel van der Molen, and Maarten Krol helped greatly in the

execution of this master thesis. They also contribute greatly to the improvement of my scientifict writingby giving constructive feedback. Thanks to MAQ Thesis Ring participants for the suggestions each of mysubmission. Further, the author acknowledges Indonesian Weather Service at Kototabang for providing themeteorology and chemistry data. Also, I would like to thank Indonesia Endowment Fund for Educationfor financially supporting my master thesis in Wageningen University. My parents and families here inWageningen that are endlessly giving support so this master thesis can be finished.

ii

Page 4: Simulating Air Pollution in the Severe Fires Event during

ContentsList of Figures v

List of Tables vi

1 Introduction 11.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2 Hypothesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

2 Getting WRF-Chem ready 42.1 Study Area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42.2 Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

2.2.1 Model Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52.2.2 Gas-phase Chemical Mechanisms and Aerosol Modules . . . . . . . . . . . . . . . 52.2.3 Model Pre-Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52.2.4 Model Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2.3 Meteorological and Emission Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72.3.1 Data for Model Set-up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72.3.2 Air Quality Data for Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.4 Emissions Input . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72.4.1 Model of Emissions of Gases and Aerosols from Nature (MEGAN) . . . . . . . . . 82.4.2 Fire INventory from NCAR (FINN) . . . . . . . . . . . . . . . . . . . . . . . . . . 82.4.3 Global Fire Emissions Database (GFED) . . . . . . . . . . . . . . . . . . . . . . . 9

3 Meteorology and Atmospheric Chemistry Evaluation of WRF-Chem 103.1 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103.2 Evaluation Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103.3 Meteorological Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113.4 Aerosol Optical Depth (AOD) Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . 143.5 Atmospheric Chemistry Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153.6 Effects of the haze cover on biomass burning emissions detection . . . . . . . . . . . . . . . 163.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

4 Sensitivity to Biomass Burning Emissions 204.1 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204.2 Model Improvement on Atmospheric Chemistry Variables . . . . . . . . . . . . . . . . . . 204.3 Model Improvement on Meteorology Variables . . . . . . . . . . . . . . . . . . . . . . . . 224.4 Summary and Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

5 Effects of biomass burning aerosols on precipitation in Indonesia 255.1 Theoretical background and previous studies . . . . . . . . . . . . . . . . . . . . . . . . . . 255.2 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255.3 Comparison with GPCP satellite data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265.4 The influence of aerosol scheme on the modeled precipitation . . . . . . . . . . . . . . . . . 285.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

iii

Page 5: Simulating Air Pollution in the Severe Fires Event during

6 Summary and Outlook 316.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 316.2 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

References 33

A Modifications in WRF-Chem code 36A.1 Adding CBMZ-MOSAIC routine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36A.2 Adding plume rise module . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

B Biomass Burning Emissions Input 39B.1 FINN Biomass Burning Emissions Preparation . . . . . . . . . . . . . . . . . . . . . . . . 39B.2 GFED Biomass Burning Emissions Preparation . . . . . . . . . . . . . . . . . . . . . . . . 40

C Formula for Model Evaluation 53

D Vertical Profile Comparison of Temperature and Relative humidity 54

E HYSPLIT Back Trajectory Model 55

iv

Page 6: Simulating Air Pollution in the Severe Fires Event during

List of Figures1 Mean Atmospheric Optical Depth (AOD) in October 2015 derived from MODIS satellite at

a wavelength of 550nm. AOD is the measure for light extinction by aerosol particles in theatmosphere. Indonesia is located between 95-141◦E and 11◦S-6◦N. . . . . . . . . . . . . . 1

2 Domain setup for the model runs. Indonesia is located at 11 ◦S - 6 ◦N and 95 ◦E - 141 ◦E. Theblue dots are ground stations that observed various meteorology and chemistry observationswhich are described in section 2.3. Sumatra, Kalimantan, and Papua are the largest islands inIndonesian archipelago . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

3 A comparison of modeled (a) daily mean temperature, (b) daily mean relative humidity, (c)daily mean wind direction, and (d)daily mean wind speed with the observations in Kototabang. 12

4 A comparison of meteorological variables at (a and d) Padang, (b and e) Singapore, and (cand f) Cilacap at 21 October 2015 0700 local time. Temperature vertical profile is shown onthe upper panel. Meanwhile, relative humidity vertical profile is shown on the lower panel. . 13

5 A comparison of modeled AOD with FINN input(a) and with GFED input (b). AOD fromMODIS retrieval is presented from wavelength of 550 nm (c). AOD from FINN and GFEDare compared with AOD retrieved from AERONET (Kuching station, Malaysia) at wave-length of 500nm (d). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

6 A comparison of PM10 concentration (a) and CO mixing ration (b) in Kototabang, PM10concentrationand PM25 concentration in Pekanbaru (c), and PM25 concentration in Singa-pore (d) from WRF-Chem model using FINN (black line), GFED (green line) with groundobservation in each station (red line). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

7 Correlation of AOD and PM25 emissions from GFED biomass burning inventory at Sumatra(left panel) and at Kalimantan (right panel) . . . . . . . . . . . . . . . . . . . . . . . . . . 17

8 A comparison of AOD from MODIS satellite retrieval (left panel) and PM25 emissions fromGFED biomass burning inventory (right panel). . . . . . . . . . . . . . . . . . . . . . . . . 18

9 Daily mean PM25 emissions in October 2015 over Indonesia estimated by GFED inventories(blue line). We modified the last 5 days of October 2015 with the same emission on 20-25October 2015 (green dashed line). We also modified emission on 21-30 October 2015 withflat emission equal to emission on 20 October 2015 (red line). . . . . . . . . . . . . . . . . 20

10 Same as fig. 6 but with boosted emissions by 30% for FINN and GFED . . . . . . . . . . . 2111 Same as fig. 6 but with boosted emissions by 30% for FINN and GFED . . . . . . . . . . . 2212 A comparison of PM10 A comparison of modeled daily mean (a) temperature, (b) relative

humidity, (c) wind direction, and (d) wind speed with the observations in Kototabang. . . . . 2313 Linear regression of simulated precipitation and GPCP precipitation. . . . . . . . . . . . . . 2614 Spatial variability of the modeled precipitation . . . . . . . . . . . . . . . . . . . . . . . . . 2715 Monthly average of (a) precipitation difference and (b) the PM25 concentration over Indonesia 2816 Domain-averaged of temperature difference between WRF-GFED_flat and WRF-METEO . 2917 A one month average of shortwave radiation difference from WRF-GFED_flat and WRF-

METEO . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2918 The distribution of aerosol particles in four different bins size used in MOSAIC aerosol scheme. 3619 Backward airmass trajectory at Kototabang at 24-26 October 2015 . . . . . . . . . . . . . . 55

v

Page 7: Simulating Air Pollution in the Severe Fires Event during

List of Tables1 Parameterization summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62 Observation Locations and data availability at each location . . . . . . . . . . . . . . . . . . 83 A summary of biomass burning (BB) inventories comparison) . . . . . . . . . . . . . . . . 94 Summary of experiment simulations. Checkmark means the variables are included. The dash

line means the variables are not included . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105 Summary of model skill in reproducing daily mean temperature, relative humidity, wind

speed, and wind direction at Kototabang. This evaluation is done for 5-30 October 2015. . . 116 Summary statistics of model skill in reproducing daily mean concentrations of PM25 at Sin-

gapore, PM10 at Pekanbaru, and PM10 at Kototabang are reported in terms of coefficient ofdetermination (r2) and Normalized Mean Bias Factor (NMBF) (Yu, Eder, Dennis, Chu, &Schwartz, 2006). NMBF units for PM25 and PM10 are µgm−3 and CO is ppm. . . . . . . . . 15

7 Summary statistics of r2 from model simulations . . . . . . . . . . . . . . . . . . . . . . . 218 Summary of model skill in reproducing daily mean temperature, relative humidity, wind

speed, and wind direction at Kototabang. This evaluation is done for 5-30 October 2015. . . 239 Emission Factors used for GFED input (Akagi et al., 2011) . . . . . . . . . . . . . . . . . . 53

vi

Page 8: Simulating Air Pollution in the Severe Fires Event during

1 Introduction

1.1 BackgroundMassive forest fires from August to November 2015 have led to the environmental catastrophe in Indone-

sia. The fires occurred at the two biggest islands, Kalimantan and Sumatra, home to the largest forests inIndonesia. At the same time, a strong El-Niño was detected since July 2015 and developed in October 2015(cpcp.noaa.gov). Historically, Indonesian fires were ignited due to deforestation (Page et al., 2002), peopleopened the forest area by slashing and burning to build shelters and industrial plantations. Siegert (2010)found that fires in Indonesia are intensified during El-Niño episodes. Until now, people are still opening theforests by setting fires in the open area and let fires be aggravated by a dry condition. The 2015 El-Niño lastedabout six months during the drought. Consequently, Indonesia and its vicinity countries were blanketed bya thick haze produced by fires. In this study, we focused on the effect of 2015 El-Niño on the extreme airpollution levels in Indonesia.

We analyzed the degree of air pollution levels over Indonesia using the measurement of Aerosols OpticalDepth (AOD). Figure 1 shows a global composite image of satellite observed AOD with extremely high (>3)levels over Indonesia, larger than anywhere else in the world. AOD tells to what degree aerosols are present inthe atmosphere preventing the transmission of light by absorbing and scattering. 2015 Indonesian forest firescontributed to ∼80-85% of the aerosol emissions at Palangkaraya, Kalimantan (Stockwell et al., 2016). Theamount of aerosol emission is similar to the aerosol emission at Kalimantan Island during 1997 Indonesianforest fires which were considered as the largest fires event in the last two centuries of recorded history (Pageet al., 2002). The high amount of aerosol concentration indicates that Indonesia had a serious air quality issuein October 2015.

Figure 1: Mean Atmospheric Optical Depth (AOD) in October 2015 derived from MODIS satellite at a wave-length of 550nm. AOD is the measure for light extinction by aerosol particles in the atmosphere. Indonesiais located between 95-141◦E and 11◦S-6◦N.

Aerosol particles are causing visibility reduction, environmental damage, and adversely impact the humanhealth (Mauderly and Chow, 2008). There are 69 million people exposed to the poor air quality with anestimated excess 11,880 mortality in September-October 2015 which was during the period of large firesand haze (Crippa et al., 2015). The thick haze also causes environmental and economic damage (WorldBank, 2016) not only for Indonesia but also the neighboring countries (Singapore, Malaysia, Philippine,

1

Page 9: Simulating Air Pollution in the Severe Fires Event during

Thailand, Vietnam, and Brunei). Indonesia itself spent a 16 billion USD from 2015 forest fires that includethe emergency response for fire extinguished, health aid and environment recovery (World Bank Group,2016).

As mentioned previously, the air pollution issue is not only affecting human health and national economybut also the environment. Aerosol particles play a major role in driving the atmosphere’s stability. Aerosolparticles have direct effects on the radiative balance of the Earth and the atmosphere. They scatter and absorbsunlight (depend primarily on the type of aerosol particles). Black carbon, the major pollutant emitted bybiomass burning, can absorb radiation and stabilize the atmosphere below. Meanwhile, aerosol particles alsohave indirect effects. One of the aerosol indirect effects is changing the cloud optical properties so that cloudsbecome brighter and longer-lived (Lohmann et al., 2000).

Beside the aerosol emissions, 2015 Indonesia forest fires also released 11.3Tg carbon per day to theatmosphere, double the emissions on days without fires (Huijnen et al., 2016). The carbon emissions areprimarily in the form of CO2, CO, and CH4. Among these carbon forms, CO is of concern because it causesheart and respiratory health problems. CO resides for ∼0.1 year in the atmosphere (Weinstock, 1969) and itpromotes the formation of ozone (O3) (Fishman et al., 1979). CO and aerosol particles are harmful both tothe environment and people’s health. Hence we focus on how they distributed temporally and spatially duringthe haze episodes in October 2015.

Many studies have investigated the gas phase and aerosol particles distribution over Indonesia and evenSouth East Asia using fully coupled meteorology-chemistry model, such as the Weather Research and Fore-casting with Chemistry (WRF-Chem) model (Aouizerats et al., 2015; Nuryanto, 2015; Crippa et al., 2015).By using WRF-Chem, they succeeded to perform simulations of trace gasses and aerosols transport in theatmosphere with several types of biomass burning emission inventories. Two main biomass burning emissioninventories are Fire INventory from NCAR (FINN) and Global Fire Emissions Database (GFED) which usedifferent methods to estimate biomass burning emissions. Little is known about similarities and differencesof FINN and GFED quality in WRF-Chem especially in Indonesia where land use and land cover are difficultto estimate. We, therefore, aim to (1) simulate CO and aerosol particles distribution produced by Indonesiaforest fires by using WRF-Chem, (2) evaluate the influence of (errors in) biomass burning emissions inven-tory on the simulations, and (3) investigate the influence of biomass burning emission on the cloud formationand precipitation. To achieve the objectives of this study, we address these research questions.

1. To what extent can we reproduce the observed temporal variation in aerosol concentrations and COmixing ratio measured at ground stations?

2. How do biomass burning inventories affect the model results? (What are the weaknesses in biomassburning inventories?)

3. How do biomass burning emission influence precipitation over the Indonesian archipelago?

1.2 HypothesisThe WRF-Chem model has been used by other groups, and within the MAQ department to simulate air

pollution. For example, Visser (2016) successfully simulated ozone concentrations over Europe. However,WRF-Chem with aerosol scheme coupling has not been tested yet in the MAQ group. In other group, WRF-Chem has been tested to simulate a severe Indonesian forest fires event (Aouizerats et al., 2015; Crippaet al., 2015; Nuryanto, 2015). The model has been shown to capture the spatial variability of the tracegasses and aerosol particles from Indonesian forest fires quite well. However, the modeled concentrationsare not capturing the temporal variability. In Crippa et al. (2015), the simulated PM25 concentration inSingapore during 2015 Indonesian fires event lags behind the ground observation although the average andpeak of the observed PM25 concentration are well reproduced by the model. Simulating air pollution using

2

Page 10: Simulating Air Pollution in the Severe Fires Event during

WRF-Chem model in a severe forest fires event is challenging. Aouizerats et al. (2015) underestimated thePM25 concentrations in Singapore (from 2006 Indonesian forest fires event). He then suggested boosting theemission factor of biomass burning inventory by 28%. In this study, we will follow Visser(2016) in usingcarbon bond mechanism version Z (CBM-Z) as the gas-phase mechanism in WRF-Chem. We will extend theWRF-Chem model by including MOdel for Simulating Aerosol Interactions and Chemistry (MOSAIC) asthe aerosol scheme. This has not been done yet in MAQ group. We, therefore, need to do a pre-configurationin the WRF-Chem model to make this inclusion work. The detail of pre-configuration will be explained insection 2.

In this study, we will compare the WRF-Chem performance using two biomass burning inventories. Thefirst inventory is the GFED biomass burning inventory (as used by Aouizerats et al.(2015)) and the second isthe FINN biomass burning inventory. The use of different biomass burning emission inventory will affect theresult of the model. There are several differences between FINN and GFED, including on how they estimatethe occurrence of fires, the spatial and temporal resolution, vegetation types, and the emission factors. Forexample, FINN calculates the burned area and takes the fraction of biomass loading that is burned in thefire to estimate the emissions. Meanwhile, GFED used dry matter combustion rate that depends on the landcover type to estimate the emissions. Each method has a potential weakness in estimating biomass burningemissions especially if the satellite detection of fire is hampered by a very thick haze; this could underestimatethe burned area and the emission estimate. Detail of FINN and GFED will be explained in Section 2.4. Thecomparison of two biomass burning emissions in WRF-Chem thus for not received much attention.

Further, the inclusion of biomass burning emissions in WRF-Chem model has an impact on clouds as wellas cloud microphysics with decreased precipitation coverage and amounts (G. Grell et al., 2011). Page et al.(2002) revealed that fine aerosol particles, PM25, are emitted in the high amount in 1997 Indonesian forestfires. Due to the high concentration of PM25, precipitation can be suppressed (Rosenfeld, 1999). However, thedelay of early precipitation in deep convective clouds with warm bases, such as those that occur in Indonesia,can lead to their invigoration and overall additional precipitation (Andreae & Merlet, 2001). On the otherhand, a layer of fine aerosol particles in the lower troposphere influences the radiative process that leads tochanges in atmospheric stability. Because of precipitation as the sink for aerosol particles through washoutprocess, less precipitation leads to an increase in the concentration of aerosol particles. This possible positivefeedback cycle may lead to an even stronger reduction of precipitation (Zhao et al., 2006). Here we willinvestigate the impact of biomass burning aerosol concentrations on precipitation in Indonesia.

3

Page 11: Simulating Air Pollution in the Severe Fires Event during

2 Getting WRF-Chem readyIn this chapter, we present the WRF-Chem model description, configuration, and data needed to run the

simulation. WRF-Chem has been set-up in Wageningen to simulate nitrogen oxides chemistry in Europeusing anthropogenic and biogenic emissions input (Visser, 2016). In this study, we develop the WRF-Chemmodel further to run air quality simulations with the same gas phase chemistry mechanism (CBM-Z), explic-itly include biomass burning emissions, and simulate aerosol emissions and transformations.

2.1 Study AreaIndonesia is an archipelago country that spans along the equatorial belt with latitude from 11 ◦S - 6 ◦N

and longitude from 95 ◦E - 141 ◦E. Indonesia is geographically located between the Pacific Ocean and theIndian Ocean. The weather pattern in Indonesia is regulated by the regional and local circulation. The majorweather pattern in Indonesia is linked with the monsoon circulation which is a large-scale seasonal reversal ofthe wind regime (Serreze and Barry, 2010). Due to the monsoon, Indonesia receives maximum precipitationduring December-January-February and minimum precipitation during June-July-August.

The Indonesian weather pattern is also influenced by a large-scale circulation from the Pacific Ocean,called El-Niño and La-Niña. The movement of warm sea water from the east part of Indonesia to the westcoast of America continent causes the easterly winds flow toward the eastern Pacific. Consequently, lessprecipitation and dry air cause drought in large part of Indonesia (Quinn et al.,1978) which plays a rolein igniting fires and making fires last longer. This episode is called El-Niño. On the other hand, La-Niñaepisodes occur when the wind flows in the opposite direction, it will bring moisture and precipitation overIndonesia.

Figure 2 shows the domain of this study. It covers Indonesia, Singapore, Malaysia, and Timor-Leste.Fires were ignited at Sumatra and Kalimantan. Sumatra is located at 6.2 ◦S - 4.5 ◦N and 95 ◦E - 105 ◦E.Kalimantan is the biggest island located at 3.7 ◦S - 4◦N and 106 ◦E - 119 ◦E. Small fires are also detectednear Papua, located at 6.2 ◦S and 136.5 ◦E, but burning causes remain unknown. Fires in Sumatra andKalimantan occurred on plantations, tropical forests, and peatlands.

Figure 2: Domain setup for the model runs. Indonesia is located at 11 ◦S - 6 ◦N and 95 ◦E - 141 ◦E. The bluedots are ground stations that observed various meteorology and chemistry observations which are describedin section 2.3. Sumatra, Kalimantan, and Papua are the largest islands in Indonesian archipelago

4

Page 12: Simulating Air Pollution in the Severe Fires Event during

2.2 ModelIn this section, we describe the WRF-Chem model. We also present the WRF-Chem set-up which is

including model pre-configuration and model configuration.

2.2.1 Model Description

This study uses the version 3.2 of the fully compressible and non-hydrostatic Weather Research andForecast (WRF) model coupled with Chemistry (WRF-Chem). WRF-Chem is an on-line weather-chemistrymodel that considers coupled physical and chemical processes in the atmosphere like transport, deposition,emission, chemical transformation, aerosol interactions, photolysis, and radiation. Based on Eulerian for-mulation of fluid transport processes, WRF-Chem approaches transport processes by calculating the massconcentration of fluid elements as a function of space and time instead of following its trajectory.

Advection and diffusion within the WRF-Chem are done by the WRF. Sub-grid scale transport is gov-erned by WRF parameterizations, planetary boundary layer (PBL), and convection. WRF-Chem’s capabilityof feedback from chemistry to meteorology is done by meteorological radiation and microphysics parame-terizations. Photolysis packages are coupled to aerosols and hydrometeors. There are three available aerosolmodules which are modal, sectional, and bulk. In the ’modal’ mode, aerosol particles are divided into threemodes (Aitken, accumulation, and coarse). Sectional puts the aerosol particles into bins according to particlediameter. Bulk calculates the total mass of aerosol components. However, not all aerosol modules can be cou-pled with any microphysics parameterizations. In this study, we used four-size bins sectional aerosol modulewhich is processed in the MOdel for Simulating Aerosol Interactions and Chemistry (MOSAIC) algorithmwithin WRF-Chem model; this algorithm only works by coupling with the Lin et al. michropyshic scheme.The partitioning of the four-size bins aerosol module will be explained later.

For the dry deposition mechanism, WRF-Chem uses Wesely parameterization (Wesely, 1989). This pa-rameterization uses bulk surface resistances, which is divided into three distinct pathways (upper canopies,lower portions, and the ground/water surface) to allow calculating the flux of trace gasses and particles withspatially and temporally varying deposition velocity. The wet scavenging in WRF-Chem model is regulatedusing G. A. Grell et al. (2005) mechanism.

2.2.2 Gas-phase Chemical Mechanisms and Aerosol Modules

WRF-Chem 3.2 has several choices of gas-phase chemical mechanisms and aerosol modules. In theMAQ group, the Carbon Bond Mechanism version Z (CBM-Z) is commonly used for the gas-phase chemistry(Zaveri & Peters, 1999). This mechanism uses 67 prognostic species and 164 reactions in a lumped structureapproach. We choose Model for Simulating Aerosol Interactions and Chemistry (MOSAIC) for the aerosolscheme (Zaveri et al., 2008). The MOSAIC simulation calculates particle evolution via the major aerosolprocesses, including binary nucleation, coagulation, condensation and scavenging by cloud droplets, as wellas wet and dry deposition. It also includes inorganic aerosol thermodynamic equilibrium and particulateformation via aqueous-phase chemistry. Secondary organic aerosols (SOA) are not included in our modelconfiguration.

2.2.3 Model Pre-Configuration

To extend the experience with the CBM-Z chemistry mechanism in our group, we used CBM-Z but nowapply it with biomass burning emissions. Several additional routines were added in the chemical setting ofthe WRF-Chem model. We included CBM-Z case routine in the module_add_emiss.F (see Appendix A.1).By adding this routine, we can prescribe emissions of each chemical species for CBM-Z. We also added a

5

Page 13: Simulating Air Pollution in the Severe Fires Event during

routine in the module_plumerise1.F (see Appendix A.2). In the plume rise module, we added the biomassburning emissions from the inventories to be processed in the module_add_emiss.F.

2.2.4 Model Configuration

WRF-Chem was set to perform simulations over one single domain covering Indonesia with a horizontalresolution of 30km × 30km. The model has 30 vertical levels from ground level up to 23 km height with astretching resolution from 60 m to 1.6km for the bottom and top level, respectively. The domain was selectedin this way to capture the aerosol distribution over the region.

The simulation was run for one months’ integration time from 1 October 2015 00.00 UTC to 31 October2015 18.00 UTC. This chosen simulation period is corresponding to the strongest El-Niño episode in 2015according to National Oceanic and Atmosphere Administration (NOAA), more information can be found athttp://www.cpc.noaa.gov/products/analysis_monitoring/ensostuff/ensoyears.shtml.

CBM-Z gas phase chemical scheme is coupled with MOSAIC aerosol scheme as this coupling is the stan-dard configuration to run an explicit aerosol model using CBM-Z. The RRTMG scheme for both longwaveand shortwave radiation scheme (Iacono et al., 2008) were chosen to simulate the aerosol direct effects. Forthe microphysics scheme, we selected the Lin et al. (Y.-L. Lin et al., 1983) which is a double microphysicsscheme to run the aerosol indirect effect. We included aqueous phase chemical reaction within the clouddroplet. We turned the wet scavenging and cloud chemistry options on to fully simulate aerosol processesand life cycle. As we interested to investigate the effect of biomass burning aerosol on the precipitation, weused Grell-3D to resolve the cumulus cloud physic (G. a. Grell, 2002).

We only used organic aerosol (OC) and black carbon (BC) as the main aerosol species which are dis-tributed into four-size bins, which are: 0.039–0.10 µm, 0.10–1.0 µm, 1.0–2.5 µm and 2.5– 10 µm (thepartition portion is shown in fig. 3). The size bins are defined by their lower and upper dry particle diam-eters, so water uptake or loss does not transfer particles between bins. Furthermore, each bin is assumed tobe internally mixed so that all particles within a bin have the same chemical composition, while particles indifferent bins are externally mixed.

We conducted several simulations, including the experiment with two biomass burning emission inven-tories. We used FINN and GFED inventories product in separate simulations. We used MEGAN productfor the biogenic emission. Detail of the emission input is described later in this section. Table 1 lists theconfiguration options used in this study.

Configuration Scheme Namelist Value ReferencesLongwave radiation RRTMG ra_lw_physics = 4 Iacono et al., 2008Shortwave radiation RRTMG ra_sw_physics = 4 Iacono et al., 2008

Microphysics The Lin et. al mp_physics= 2 Lin et al., 1983Cumulus physics Grell-3D cu_physics = 3 Grell and Devenyi, 2002

Planetary boundary layer YSU bl_pbl_physics = 1 Hong et al., 2006Fire emissions I FINN biomass_burn_opt = 2 Wiedinmyer et al., 2011Fire emissions II GFED biomass_burn_opt = 2 Van Der Werf et al., 2010

Biogenic emissions MEGAN bio_emiss_opt = 2 Guenther et al., 2012Aerosol process MOSAIC chem_opt = 9* Zaveri et al., 2008

Gas-phase chemistry CBM-Z chem_opt = 9* Zaveri et al., 2010

Table 1: Parameterization summary

6

Page 14: Simulating Air Pollution in the Severe Fires Event during

2.3 Meteorological and Emission DataHere we present the data used to run and evaluate the model. Data used in the model consists of initial-

boundary data and emission input. Data used to evaluate model are obtained from ground and soundingobservations.

2.3.1 Data for Model Set-up

This study used meteorological boundary condition from National Center for Environmental PredictionFiNal reanalysis (NCEP-FNL) for the initialization of meteorological variables (NCEP-FNL, 2000). TheNCEP-FNL has a horizontal resolution of 1-degree by 1-degree prepared operationally every six hours. Thedataset is available on the surface, at 26 mandatory (and other pressure) levels from 1000 millibars to 10millibars, in the surface boundary layer and at some sigma layers, the tropopause and a few others. Parametersinclude surface pressure, sea level pressure, geopotential height, temperature, sea surface temperature, soilvalues, ice cover, relative humidity, u- and v- winds, vertical motion, vorticity, and ozone.

For the surface emissions, the model uses surface emission data from two sources: (1) biogenic emissionsand (2) biomass burning emissions. The biogenic emissions data is provided by the Model of Emissionsof Gases and Aerosols from Nature version 2.1 (MEGAN2.1). MEGAN2.1 is a model framework for aresolution of ∼1km. This model estimates natural emissions that occur in a terrestrial ecosystem (Guentheret al., 2012). Meanwhile, the biomass burning emissions data is provided by the Fire INventory from NCARversion 1.0 (FINNv1). The FINNv1 provides a spatial resolution of ∼1km (Wiedinmyer et al., 2011). Thesecond biomass burning emission inventory used in this study is Global Fire Emission Dataset version 4(GFEDv4). GFEDv4 has coarser horizontal resolution than FINNv1, that is 0.25◦× 0.25◦(van Der Werf etal., 2010). More detail on emissions input description is on chapter 2.4.

2.3.2 Air Quality Data for Validation

To validate the model output, we will use ground measurement, sounding data, and satellite data. Groundobservations are located in 6 locations (see fig. 2) which are chosen based on their location relative to thefires. We included three stations which are influenced directly by fires (downwind) and three stations whichare located far from the fires. Data availability can be seen in Table 2.

The vertical profile from the model is validated using sounding data from Wyoming University. We canaccess the sounding data at http://weather.uwyo.edu/upperair/sounding.html. We used the same location asthe ground measurement. The ground measurement is beneficial for the high time resolution, meanwhilesounding data gives a good vertical coverage. Because ground observation data availability in Indonesia islimited, we also used satellite data to validate model results. We used satellite data to verify the WRF-ChemAtmospheric Optical Depth (AOD) using the MODIS satellite on board Terra spacecraft.

2.4 Emissions InputWRF-Chem needs three emissions input inventories: anthropogenic, biogenic, and biomass burning emis-

sions to simulate the air pollution. In this study, we ignored the anthropogenic emissions because we assumedthat anthropogenic emissions in October 2015 did not contribute to the air pollution as much as biomassburning emissions. As stated by Gaveau et al. (2014), in June 2013 Indonesia forest fire event when fires lessintense than in October 2015, a one week period with intense fires in Sumatra amounts to 10% of the yearlyaccumulated emissions over the whole country.

7

Page 15: Simulating Air Pollution in the Severe Fires Event during

Location Latitude Longitude Species MeasuredPenang 5.358 100.30 ground AOD

Singapore 1.22 103.59 Daily PM25, vertical T and RHKototabang -0.20 100.3013 Daily PM10, Daily CO, T, RH, WSPD, WDIRPekanbaru 0.28 101.27 Daily PM10Kuching 1.49 110.35 vertical T and RH

Pontianak 0.08 109.19 ground AOD, vertical T and RHMakassar -5 119.57 ground AODCilacap -7.73 109.01 vertical RH and T

Table 2: Observation Locations and data availability at each location

2.4.1 Model of Emissions of Gases and Aerosols from Nature (MEGAN)

MEGAN version 2.1 (MEGANv2.1) is a modeling framework that estimates fluxes of biogenic com-pounds between terrestrial ecosystem and the atmosphere (Guenther et al., 2012). Although it estimates thesum of all emission sources that can naturally occur in a terrestrial ecosystem, it doesn’t account for biomassburning emission. Hence, we used biomass burning inventories from other inventories to complete the emis-sion input in WRF-Chem. The biomass burning emission inventories will be described afterwards (subsection2.4.2 and 2.4.3).

To run MEGANv2.1, we needed to give meteorological variables input such as: solar radiation, tem-perature, and moisture. We obtained these features from NCEP-FNL reanalysis data set. Before employ-ing these features into the MEGAN model, we re-gridded the data set to match with the desire resolu-tion which done by executing the real program in WRF-Chem. The monthly leaf area index (LAI) infor-mations are provided within the model with a resolution of 30 second for 2003. MEGANv2.1 estimatesemission of chemical species according to the emission factor at standard conditions for vegetation typewith fractional grid box areal coverage. We run MEGANv2.1 using processor which can be downloaded athttp://lar.wsu.edu/megan/guides.html at no cost.

2.4.2 Fire INventory from NCAR (FINN)

In this study, we use FINN version 1 (FINNv1). FINNv1 provides daily high horizontal resolution (∼1km2) global estimates of biomass burning emission. It includes trace gases and particulate matter that isreleased by open fires. There are six land cover classes in FINN, which are: boreal forest, tropical forest,temperate forest, woody savanna/shrub-land, savanna/grassland, and croplands. Emission factors are deter-mined for each species and land cover classes.

FINNv1 estimates emission (E) by using four components, which are: area burned (A), biomass burningloading (B), the fraction of biomass that is burned in the fire (FB), and the emission factor of the species (ef )which is the fraction of total emissions that were emitted in the different days of the month. Burned area isobtained from the standard MODIS burned area product which are characterized by deposits of charcoal andash, removal of vegetation, and alteration of the vegetation structure . Biomass burning emission is calculatedas follow:

E = A×B×FB× e f (1)

According to (Wiedinmyer et al., 2010), there are considerable uncertainties of FINNv1 biomass burningemission. FINNv1 tends to underestimate number of fires due to cloud cover that inhibit satellite detection.However, it also potentially overestimates the size of small fires detected. Uncertainty is also caused by

8

Page 16: Simulating Air Pollution in the Severe Fires Event during

misidentification of land cover. Consequence of land cover misidentification led to false estimate of fuelloading consumption and emission factors.

Chemical speciation profiles in FINNv1 are provided for three used chemical mechanism, which are:SAPRC99, GEOS-CHEM, and MOZART-4. In this study, we use chemical speciation for MOZART-4. Weuse fire_emis to do a pre-pocessing FINNv1 data before using them as biomass burning input in WRF-Chem. The fire_emis utility is re-griding the FINNv4 data to WRF-Chem spatial resolution. It also maps thechemical speciation to WRF-Chem chemical speciation.

2.4.3 Global Fire Emissions Database (GFED)

Similar to FINNv1, GFED version 4 (GFEDv4) is also a biomass burning inventory that provides amonthly global biomass burning emission data. GFEDv4 has a coarser resolution which is 0.25◦ × 0.25◦

(with 0.25◦ in the equator equal to ∼25km). Besides the resolutions, GFEDv4 used different methods inestimating biomass burning emissions compared to FINNv1.

In GFEDv4, Along Track Scanning Radiometer (ATSR) is used to generate the monthly gridded burnedarea in Equatorial Asia (?, ?). MODIS direct broadcast area product derives the daily surface reflectance inputto locate the daily fraction of the monthly burned area from ATSR (Giglio et al., 2013). The burned area isnot used directly in the emission estimate. Nevertheless, the burned area is used to estimate the area of eachvegetation type that contribute as the dry matter emission, the mass of vegetation when completely dried dueto fires. Hence, to estimate each species emission, we need to calculate dry matter (DM) emission with dailyfraction ( fdaily), contribution of each land vegetation type (C), and Akagi emission factor (e f ) (Akagi et al.,2011). GFEDv4 estimates emission by using equation as follow:

E = DMemission × fdaily ×C× e f (2)

Species emissions are calculated using dry matter estimates and emission factors developed for GFED(see table 6 in the appendix). This inventory also distinguished emission sources according to the conributiontype. In this study, we use 6 vegetation type contributions which are: savana, boreal forest, temperate forest,deforestation, peat, and agriculture.

Criteria Biomass burning inventoryFINN GFED

Spatial resolution 1×1 km 0.25◦ × 0.25◦

Temporal resolution daily monthlyEmissions estimate E = A×B×FB× e f E = DMemission × fdaily ×C× e f

Fire detection method MODIS active fire combination of MODIS and ATSR satellite

Table 3: A summary of biomass burning (BB) inventories comparison)

9

Page 17: Simulating Air Pollution in the Severe Fires Event during

3 Meteorology and Atmospheric Chemistry Evaluation of WRF-ChemIn this section, we present the results from model simulations and comparison with the observation data.

We use observation from six meteorology stations in Indonesia (detail can be seen on section 2.). Satellitedata is also used to evaluate atmospheric optical depth (AOD) from the model results.

3.1 ExperimentsTable 4 summarizes the experiments done in this chapter. WRF-METEO run is done in the WRF-Chem

by switching off the chemistry option. The second and third experiment are ran by switching the chemistryoption on. We distinct the biomass burning emissions input on both experiment. We used MEGAN biogenicemissions input for the second and third experiments.

Meteo Chem Biomass Burning Inventory Biogenic EmissionWRF-METEO — — —

WRF-FINN FINNv1 MEGANWRF-GFED GFEDv4 MEGAN

Table 4: Summary of experiment simulations. Checkmark means the variables are included. The dash linemeans the variables are not included

3.2 Evaluation MethodsIn this evaluation, we only use evaluation from 6 Oct 2015 onwards to avoid a large bias. The spin-

up time of this model is five days. We evaluated the model results using two statistical metrics which are acoefficient of determination (r2) and Normalized Mean Bias Factor (NMBF). These statistical metrics indicatethe proportion of the variance in the modeled variables from the observed variables and reduce the biasproblem when some values (denomination) are trivially low or high (Yu et al., 2006). Detail of these metricscan be seen in Appendix C.

For the meteorology evaluation, we used the ground observation at Kototabang (0.20◦S, 100.32◦E, 865m MSL), which is located in the mountainous region of West Sumatra and is roughly 17 km north of thenearest town (Bukittinggi) and 40 km of the western coastline. The meteorology variables at Kototabangthat will be evaluated include surface temperature (T), relative humidity (RH), wind speed (WSP), and winddirection (WD). To validate each variable, we pick a cell that is covering the Kototabang station. We also usedvertical profile of relative humidity and temperature from sounding data provided by Wyoming University atfour stations which are Singapore, Kuching, and Cilacap (see section 2). Singapore and Kuching stations arelocated on the downwind of the air pollutions. Meanwhile, Cilacap is located at the southern fire plumes thatare not affected by the fires emissions. Sounding data is available daily only at 0000 UTC and 1200 UTC. Wechose sounding measurement at 0000 UTC or 0700 Western Indonesia Time when the atmosphere conditionis relatively stable. We calculated the monthly averaged bias of the model at each station. Then, we averagethe monthly averaged biases at all station.

To assess the model performance in aerosol-related simulations, we compared the Aerosol Optical Depth(AOD) result from the model with the corresponding Moderate Resolution Imaging Spectroradiometer (MODIS;Terra) level 2 products. The MODIS level 2 values are derived from the level 1 aerosol and cloud productsby averaging the level 1 retrievals (10 km spatial resolution) across each 1 by 1 grid box of the level 3product. Aerosol optical properties in WRF-Chem are available in 4 bins which are TAUAER1, TAUAER2,

10

Page 18: Simulating Air Pollution in the Severe Fires Event during

TAUAER3, TAUAER4 in the wavelength of 300, 400, 600 and 999nm, respectively. Meanwhile, AOD fromMODIS satellite data and AERONET is available at a wavelength of 550 and 500nm, respectively. To derivethe model AOD at 500 and 550 nm, the Ångström power law is used:

W (λ ) =W (400)× (λ

400)−α (3)

where W(λ ) is the WRF-Chem AOD at wavelength λ (550) nm and α is the Ångström exponent calculatedfrom WRF-Chem AOD at 300 and 999 nm using the following relation:

α =lnW (300)

W (999)

ln 999300

(4)

The atmospheric chemistry evaluation is done for carbon monoxide (CO) and particulate matter (PM).We obtained ground observation of CO and PM10 at Kototabang, PM10 at Pekanbaru, and PM25 at Singapore.The observations are only given as 24-hr averaging; therefore, we need to calculate the daily mean of modeledCO, PM10 and PM25 from the WRF-Chem simulations.

3.3 Meteorological EvaluationTable 5 shows the r2 and NMBF of WRF-METEO, WRF-FINN, and WRF-GFED for meteorological

variables (2m-temperature (T), relative humidity (RH), wind direction (WDIR), and wind speed (WSPD))calculated at Kototabang station. Figure 3 shows time series of the meteorological variables from the WRFsimulations and the observation. In general, the WRF-Chem simulations overestimate the observation. Theoverestimation is indicated by the positive NMBF of all experiment schemes of each meteorological variables.However, the inclusion of chemistry scheme in WRF-Chem reduces the degree of the overestimation. Forexample, the NMBF of WRF-FINN wind direction is 17% lower than WRF-METEO, and the NMBF ofWRF-FINN wind direction is 21% lower than WRF-METEO. The WRF-GFED always produces the lowestNMBF for all meteorological variables. The WRF-GFED r2 shows the strongest correlation among the modelsimulations. Although, the degree of correlation for all meteorology variables of WRF-GFED are very low,however, this is indicating that including the biomass burning emissions in WRF-Chem is indeed improve themodel results.

Variables T RH WDIR WSPDr2 NMBF r2 NMBF r2 NMBF r2 NMBF

WRF-METEO 0.13 0.088 0.26 0.12 0.14 0.47 0.07 0.43WRF-FINN 0.21 0.084 0.50 0.11 0.19 0.39 0.16 0.41WRF-GFED 0.22 0.082 0.51 0.11 0.29 0.37 0.24 0.31

Table 5: Summary of model skill in reproducing daily mean temperature, relative humidity, wind speed, andwind direction at Kototabang. This evaluation is done for 5-30 October 2015.

As shown in figure 3, the models poorly capture the temporal variability of 2m-temperature measured atKototabang. However, the models capture the magnitude of 2m-temperature quite well. The mean observed2m-temperature in Kototabang during 6-30 October 2015 was 22.4◦C, well reproduced by the model (23.4◦C,23.1◦C, and 22.7◦C for WRF-METEO, WRF-FINN, and WRF-GFED, respectively). The warm bias occurson 16-21 October 2015 when the biomass burning emissions are maximum during October 2015. WRF-Chem

11

Page 19: Simulating Air Pollution in the Severe Fires Event during

consistently overestimates relative humidity. The consistent overestimation is also shown by the modeledwind speed, however, the wind speed bias is less important since the wind speed is generally low (with meanof 1.25 ms−1), 1.5 ms−1, 1.5 ms−1, and 2.5 ms−1 for the observation, WRF-GFED, WRF-FINN, and WRF-METEO, respectively). The modeled wind directions show less variability than the daily observed winddirections.

(a) daily mean temperature (b) daily mean relative humidity

(c) daily mean wind direction (d) daily mean wind speed

Figure 3: A comparison of modeled (a) daily mean temperature, (b) daily mean relative humidity, (c) dailymean wind direction, and (d)daily mean wind speed with the observations in Kototabang.

Several factors influence the model results: model representativeness and the presence of atmosphericchemistry feedback mechanisms. As mentioned previously, the ground observation in Kototabang is locatedon the varying landscape and altitude of the surrounding that is probably underestimated by the model. Thecoarse resolution of the model underestimates the surface roughness hence the model representativeness forthe specific site of Kototabang can be questioned. This underestimation of surface roughness causes themodel to assume the flat surface. The low friction causes modeled wind speed blow faster than reality inKototabang. This is causing the overestimation of modeled wind speed.

The absence of chemical and physical feedback in the model may also have caused the discrepancies. Forinstance, the shading effect caused by aerosol particles in the atmosphere may prevents the solar radiationto reach the ground. However, the shading effect is weak as will be shown in fig. 6. The absence ofaerosols in WRF-METEO model also leads the model to overestimate RH. Although aerosol present in theWRF-FINN and WRF-GFED model, the effect of including aerosol on RH is weak. The presence of aerosolin the model is affecting the modeled RH as the fine particle mass comprises the mass of water vapor inthe atmosphere (Xiao et al., 2011). We therefore need more aerosol particles with smaller diameter in the

12

Page 20: Simulating Air Pollution in the Severe Fires Event during

(a) T at Padang (b) T at Singapore (c) T at Cilacap

(d) RH at Padang (e) RH at Singapore (f) RH at Cilacap

Figure 4: A comparison of meteorological variables at (a and d) Padang, (b and e) Singapore, and (c andf) Cilacap at 21 October 2015 0700 local time. Temperature vertical profile is shown on the upper panel.Meanwhile, relative humidity vertical profile is shown on the lower panel.

13

Page 21: Simulating Air Pollution in the Severe Fires Event during

modeled atmosphere to show the comprising effect on the water vapor volume in the modeled atmosphere.To further examine the ability of the model to reproduce meteorological variables we compared WRF-

Chem outputs against three ground-base sites measuring vertical profile of temperature and relative humidity.The models capture well both the magnitude and the temperature vertical profile variability measured overPadang, Singapore, and Cilacap at 21 October at 0700 local time (see fig. 4a, b, and c). In general, temper-ature biases among the simulations are very low (mean bias is 1.7◦C, 0.9 ◦C, and 0.7◦C for WRF-METEO,WRF-FINN, and WRF-GFED, respectively). However, the relative humidity profiles are poorly simulated(see fig. 4d, e, and f). The relative humidity biases are varying within the measured altitude. Models arenot able to capture several spikes. The bias is possibly caused by the presence of particulate matters in theatmosphere which are underestimated by the model.

The cold bias on the top level of the modeled atmosphere is a result of the radiation parameterizationemployed in the model. The Rapid Radiative Transfer Model (RRTM) is used as short- and long-waveradiation scheme (Mlawer et al., 1997). The RRTM scheme assumes that the temperature between the modeltop and the top of atmosphere (TOA) is isothermal and the mixing ratio is constant to calculate the radiativefluxes at the model upper boundary. To do this assumption, they add a new layer between the model top andTOA in such condition. This assumption leads to a negative bias at the top of modeled atmosphere layerbecause the actual temperature on that level varies significantly (Powers et al., 2010).

3.4 Aerosol Optical Depth (AOD) Evaluation

(a) AOD from WRF-Chem with FINN input (b) AOD from WRF-Chem with GFED input

(c) AOD from from MODIS retrieval (d) AOD from WRF-Chem with GFED input

Figure 5: A comparison of modeled AOD with FINN input(a) and with GFED input (b). AOD from MODISretrieval is presented from wavelength of 550 nm (c). AOD from FINN and GFED are compared with AODretrieved from AERONET (Kuching station, Malaysia) at wavelength of 500nm (d).

We evaluate the model skill in reproducing spatiotemporal variability of aerosol optical properties againstan AOD retrieval from 500nm MODIS and a ground observation from 550nm AERONET located in Kuching,Malaysia. Figure 5 shows that WRF-FINN and WRF-GFED can simulate the spatial distribution of observed

14

Page 22: Simulating Air Pollution in the Severe Fires Event during

AOD. However, the underestimation in regions impacted by fires is clearly seen. The monthly mean AODover Indonesia is 0.29, 0.36, and 0.86 for WRF-FINN, WRF-GFED and MODIS retrieval, respectively.

The optical properties in WRF-Chem depend strongly on particle size and the 4-bins partition we as-sumed, thereby contributing to the uncertainty in the predicted AOD. Within these constraints, the agreementof AOD spatial distribution from both models is reasonable. From our previous finding, we need more aerosolparticles with a smaller diameter in the model. The underestimation of simulated AOD by WRF-Chem modelin regions impacted by fires has been reported by numerous previous studies (Crippa et al., 2015; Aouizeratset al., 2015; Kim et al., 2015; Reddington et al., 2014; Marlier et al., 2012).

In contrast to satellite observation, AERONET provides ground measurement with daily temporal res-olution over a given location where satellite cover might not always be available. As shown in fig. 5d, ahigh level of AOD (∼3.4) from AERONET observed on 23 October 2015. AOD from WRF-GFED agreeswith the AERONET. High level of modeled AOD with GFED input occurred on 24 October 2015. Mod-eled AOD from WRF-GFED shows the similar pattern as indicated by AERONET. Meanwhile, modeledAOD from WRF-FINN underestimates AERONET with a factor of ∼6. The underestimation indicates thatWRF-FINN produces less particulate matters than WRF-GFED. The sensor resolution possibly causes themismatch between MODIS retrieval and the ground observation.

3.5 Atmospheric Chemistry EvaluationWe used observation of PM10, CO, and PM25 from different locations to evaluate the model results. First

location is Kototabang (0.08◦N - 100.33◦E) which provides daily measurement of PM10 concentration andCO mixing ratio. Second location is Pekanbaru (0.88◦S - 100.35◦E) which only provides daily measure-ment of PM10 concentration. The last station is Singapore which also only has daily measurement of PM25concentration.

Variables PM25 at Singapore PM10 at Pekanbaru PM10 Kototabang CO Kototabangr2 NMBF r2 NMBF r2 NMBF r2 NMBF

WRF-FINN 0.016 -0.90 0.35 -0.96 0.26 -0.93 0.26 -0.93WRF-GFED 0.035 -0.11 0.45 -0.52 0.46 -0.09 0.39 -0.09

Table 6: Summary statistics of model skill in reproducing daily mean concentrations of PM25 at Singapore,PM10 at Pekanbaru, and PM10 at Kototabang are reported in terms of coefficient of determination (r2) andNormalized Mean Bias Factor (NMBF) (Yu et al., 2006). NMBF units for PM25 and PM10 are µgm−3 andCO is ppm.

The statistical metrics for the model simulations against the observation are shown in Table 6. This tableshows that both simulations always underestimate the observation (indicated by negative sign of NMBF).However, the degree of underestimation on WRF-GFED is lower than WRF-FINN. Meaning that WRF-GFED performs better than WRF-FINN. This also proved by the better r2 value of WRF-GFED compared toWRF-FINN.

Figure 6 shows the comparison of simulated PM10 and PM25 concentrations and CO mixing ratio withthe observation. We can clearly see that both WRF-FINN and WRF-GFED underestimate the observations.WRF-FINN simulates around 50% lower surface concentration than simulated by WRF-GFED. In Pekanbaruand Singapore, the modeled PM10 and PM25 concentrations are also underestimating the observation. Ingeneral, model results capture similar pattern with the observation, except in Singapore. The peak of airpollution event started at 18 October 2015 and it continues to 26 October 2015. The model results indicate

15

Page 23: Simulating Air Pollution in the Severe Fires Event during

that both WRF-FINN and WRF-GFED fail to simulate the peak of polluted event. This evolution of PM10concentration at Pekanbaru is comparable with Crippa et al. (2015).

This underestimation is possibly due to the underestimation of emissions which also indicated by ourprevious finding. The underestimation at maximum hazy period of (21-30 October 2015) could be caused byfalse emissions estimation by both biomass burning inventories. Since both inventories make use the satelliteretrieval to estimate burning areas, the inaccuracy could be caused by the haze cover effect. During the peakperiod of fires, a very thick haze reduces the visibility of satellite to see the burned area. The relationshipbetween fire detections and area burned is highly uncertain. Especially for emissions produced by peat firesthat tend to be smoldered which is hardly detected by the satellite. In addition, in fig. 6 also indicated thatpeak concentrations are shifted relative to the enhancement in the observations. This may be related to thefalse simulated of wind direction that shifts the air mass and the underestimation of wind speed that slowsdown the transport.

(a) daily PM10 in Kototabang (b) daily CO in Kototabang

(c) daily PM10 in Pekanbaru (d) daily PM25 in Singapore

Figure 6: A comparison of PM10 concentration (a) and CO mixing ration (b) in Kototabang, PM10 concen-trationand PM25 concentration in Pekanbaru (c), and PM25 concentration in Singapore (d) from WRF-Chemmodel using FINN (black line), GFED (green line) with ground observation in each station (red line).

3.6 Effects of the haze cover on biomass burning emissions detectionHere we investigated the relation of AOD levels and the estimate of PM25 emission in Indonesia. We

compared the AOD level with the PM25 emission in Sumatra and Kalimantan separately. We used the 5-daysaverage of AOD and PM25 at a period of 6-30 October 2015. Figure 7 shows the negative correlation (r= -0.631) between PM25 emission and AOD level over Sumatra. A high AOD level indicates a low PM25

16

Page 24: Simulating Air Pollution in the Severe Fires Event during

emission over Sumatra. The reverse correlation (r = 0.754) is presented over Kalimantan. AOD is a unitlessmeasure of the attenuation of light due to the presence of particulate matters that prevent light transmissionvia absorption or scattering. This fundamental property of AOD is clearly seen in Kalimantan. There is anindication that WRF-GFED failed to estimate the biomass burning emission over Sumatra.

(a) Correlation at Sumatra (b) Correlation at Kalimantan

Figure 7: Correlation of AOD and PM25 emissions from GFED biomass burning inventory at Sumatra (leftpanel) and at Kalimantan (right panel).

As shown in fig. 8, the AOD distribution of 11-15 October 2015 is less confined than of 16-20 October2015. Nevertheless, the average PM25 emissions at 11-15 October 2015 is higher than at 16-20 October2015. This result is remarkable since higher AOD should indicate higher emissions on the surface. Wealso compared the AOD composite of 21-25 October 2015 which is even more widely spread than of 16-20October 2015, the PM25 emissions of 21-25 October 2015 is less spread than of 16-20 October 2015. Duringthick haze of air pollution event at 21-25 October 2015, fewer PM25 emissions were estimated by the GFEDinventory over Sumatra. On the other hand, emissions over Kalimantan is 20% higher than over Sumatra.The mean PM25 emission during 21-25 October 2015 over Indonesia is 15% lower than during 16-20 October2015. And again, this is remarkable given the extremely AOD levels.

During 26-30 October 2015, the reduction of AOD level was observed over Kalimantan. The decreasein emission in Kalimantan over 26-30 October 2015 is consistent with the observed AOD level. The PM25emission reduction was caused by precipitation which occurred over Kalimantan on 26 October 2015, andthis ends the fires episode in Kalimantan (http://www.globalfiredata.org/updates.html#2015_indonesia).

In Sumatra, the AOD level at 26-30 October 2015 was stronger than in Kalimantan. The relation of PM25emission and AOD over Kalimantan, unfortunately, is opposite to the PM25 emission estimate at Sumatra.The decrease of PM25 emission at Sumatra does not reflect the massive pollution detected by the MODISsatellite. The discrepancy of the PM25 estimate leads to the underestimation of WRF-GFED simulationduring the thick haze of air pollution event (21-30 October 2015).

17

Page 25: Simulating Air Pollution in the Severe Fires Event during

(a) AOD composite of 11-15 October 2015(b) PM25 emission averaged over 11-15 October2015

(c) AOD composite of 16-20 October 2015(d) PM25 emission averaged over 16-20 October2015

(e) AOD composite of 21-25 October 2015(f) PM25 emission averaged over 21-25 October2015

(g) AOD composite of 26-30 October 2015(h) PM25 emission averaged over 26-30 October2015

Figure 8: A comparison of AOD from MODIS satellite retrieval (left panel) and PM25 emissions from GFEDbiomass burning inventory (right panel).

3.7 ConclusionsGround measurement in Kototabang shows that WRF-Chem model could not capture the evolution of

daily observed 2m temperature, relative humidity, wind speed and wind direction. The poor model perfor-mance is indicated by the low r2 for all variables. The poor results are possibly caused by the unrepresentative

18

Page 26: Simulating Air Pollution in the Severe Fires Event during

of the model resolution. Kototabang has a diverse landscape; it is located in a mountainous area which is as-sumed flat by the model with a coarse horizontal resolution of 30 by 30 km.

However, WRF-Chem with biomass burning emissions does improve the model. The presence of at-mospheric chemistry in the model also plays a role. This can lower the simulated temperature and relativehumidity. The presence of biomass burning aerosols in the atmosphere can also slow down the simulatedwind speed due to the "cooling" effect on the surface. The wind direction is also slightly shifted. WRF-GFED performs better in simulating the meteorological variables at Kototabang than WRF-FINN.

To further examine the performance of WRF-GFED and WRF-FINN, we evaluated the model with sound-ing profiles located at three separated locations. In general, models can capture the observed vertical profileof temperature profile quite well. Unfortunately, none of these simulations can capture several peaks of theobserved relative humidity. Still and all, the simulated vertical profile of relative humidity succeed to capturethe general pattern of the observed relative humidity’s vertical profile. Moreover, WRF-GFED produced theleast mean bias among other simulations.

Atmospheric chemistry evaluation for WRF-FINN and WRF-GFED results underestimate the observa-tions. In Kototabang, PM10 concentration and CO mixing ratio are reasonably well simulated. Both WRF-FINN and WRF-GFED can capture some peaks. However, they failed to capture the maximum pollutantconcentrations during 21-30 October 2015. WRF-FINN yields around 75% underestimation of PM10 con-centration and CO mixing ratio observed in Kototabang while WRF-GFED "only" yields around 40% un-derestimation. The small underestimation indicates that GFED biomass burning input gives more realisticemissions than FINN. However, the uncertainties of burned area estimate in GFED inventory involved in theunderestimation of simulation results.

We investigated the relation of AOD level from MODIS retrieval and PM25 emission from GFED inven-tory. The positive correlation of these variables occurred in Kalimantan. Surprisingly, Sumatra exhibits thenegative correlation. This indicates that GFED inventories underestimate the emission whenever the AODlevel being high. This unrealistic estimate of biomass burning emission is crucial and needs some adjustmentto improve the model results, especially in a very polluted area such as Sumatra.

To cope with the emission issue, several studies reveal that GFED biomass burning inventory needs tobe boosted to realistically simulate AOD and atmospheric chemistry concentrations. Aouizerats et al.(2015)boosted the emission by 28% of emission factors. We suggested boosting the emissions slightly higher (to30% of emissions) than suggested. From our previous finding, GFED inventory misinterpreted the emissionsat a period of thick haze (26-30 October 2015). Investigations on the biomass burning emissions inventoryare valuable to improve the WRF-Chem model. The next questions we will address is: to what extent can weimprove the WRF-Chem model by modifying the biomass burning emissions input?

19

Page 27: Simulating Air Pollution in the Severe Fires Event during

4 Sensitivity to Biomass Burning EmissionsHere we investigate the sensitivity of WRF-Chem to the biomass burning emissions. We conduct two

experiments which are (1) boosting GFED inventory by 30% of emission factor and (2) modifying GFEDinventory during the thick haze event (21 - 29 October 2015).

4.1 MethodologyIn chapter 3, we conclude that WRF-Chem simulations using FINN and GFED underestimated obser-

vations. Previous simulations also denoted the underestimation during the thick haze event (26-30 October2015). To improve the model, we tried to investigate the sensitivity of biomass burning emissions by con-ducting two experiments. First, we boosted both biomass burning emissions by 30% of total emission of eachspecies. We recalculate the biomass burning emission input in module_plumerise1.F. Complete routine is inAppendix A2.

Second, we will try to capture high pollutant concentrations on 25-29 October 2015 using boosted GFEDinventory and modify the emission input on 25-29 October 2015. We set two scenarios. Figure 9 illustrates thesimulation scenarios compared to the boosted PM25 emission (WRF-GFED as the baseline). First scenariois by imitating the emission pattern on 20-24 October 2015 as the emission pattern on 26-29 October 2015(hereafter, WRF-GFED_rep). Third scenario is by keeping constant emission on 20 October 2015 to 29October 2015 (hereafter, WRF-GFED_flat). These scenarios apply over whole Indonesia and for all chemicalspecies in GFED inventory. Afterwards, we will compare the r2 and NMBF before and after adding chemistrymechanism on WRF-Chem model to quantify the model improvement.

Figure 9: Daily mean PM25 emissions in October 2015 over Indonesia estimated by GFED inventories (blueline). We modified the last 5 days of October 2015 with the same emission on 20-25 October 2015 (greendashed line). We also modified emission on 21-30 October 2015 with flat emission equal to emission on 20October 2015 (red line).

4.2 Model Improvement on Atmospheric Chemistry VariablesBoosting biomass burning emissions by 30% increased the simulated concentration of gas and aerosol.

Figure 10 shows the temporal evolution of CO and aerosol concentrations compared to the observations.The simulations with boosted emission improve CO and aerosol concentration relative to observations. TheNMBF of both model decreases 30% from result without emission boosting. Nevertheless, WRF-FINN still

20

Page 28: Simulating Air Pollution in the Severe Fires Event during

underestimates the observation, but WRF-GFED succeed in producing the observed concentration. Still, bothmodels lagged 2-3 days the observation and not able to capture high concentration in 25-29 October 2015.

(a) daily PM10 in Kototabang (b) daily CO in Kototabang

(c) daily PM10 in Pekanbaru (d) daily PM25 in Singapore

Figure 10: Same as fig. 6 but with boosted emissions by 30% for FINN and GFED

As indicated in section 2, emissions from GFED are always higher than those from FINN. In the emissionfactor calculation, GFED takes into account emissions produced by burning on peatland (van Der Werf et al.,2010). Fires on peatland tend to smolder rather than flame. Smolder is a flameless form of combustion,sustained by the heat evolved when oxygen directly attacks the surface of a condensed-phase fuel. Thealgorithm of FINN inventory does not include peatland fires (Wiedinmyer et al., 2010). In Indonesia, peatlandfires have contributed to biomass burning emissions up to 30% of total emissions on September - December2015 Indonesia emission flux (Huijnen et al., 2016). According to this finding, we suggest to use GFEDinventory rather than FINN inventory to simulate the severe air pollutaion event.

WRF-GFED WRF-GFED_rep WRF-GFED_flatPM10 concentration Kototabang 0.46 0.43 0.60

CO mixing ratio Kototabang 0.39 0.35 0.45PM10 concentration Pekanbaru 0.32 0.44 0.45PM25 concentration Singapore 0.035 0.016 0.0096

Table 7: Summary statistics of r2 from model simulations

21

Page 29: Simulating Air Pollution in the Severe Fires Event during

Table 7 summarize the r2 of PM10 concentration and CO mixing ratio in Kototabang, PM10 concentra-tion in Pekanbaru and PM25 concentration in Singapore. The WRF-GFED_flat simulation improves the airpollutant concentration, except for PM25 concentration in Singapore. We are also aware that the simulatedconcentration of CO and aerosol are likely 1-2 days ahead of the observations. We have an impression thatthe fire onset in GFED inventory occurs earlier than in reality. We do not investigate this further because wedo not compare GFED either with the observed emission nor other inventory, except FINN.

(a) daily PM10 in Kototabang (b) daily PM10 in Pekanbaru

(c) daily CO in Kototabang (d) daily PM25 in Singapore

Figure 11: Same as fig. 6 but with boosted emissions by 30% for FINN and GFED

Figure 11 shows that WRF-GFED_flat performs better than WRF-GFED_rep, except for simulated PM25concentration in Singapore. In Kototabang, we succeed to enhance simulated concentrations PM10 and COat 23-27 October 2015. Despite underestimation of PM10 concentration, we succeed to reproduce CO mixingratio at 26 October 2015. The modification, however, failed to capture the temporal evolution of aerosolconcentration and CO mixing ratio. Generally, the simulated concentrations are 1-3 days earlier than theobservations, probably due to the early fire onset of GFED inventory.

4.3 Model Improvement on Meteorology VariablesBoosting the biomass burning emissions improves the model ability to simulate the atmospheric chemistry

variables as shown in the previous subsection. The presence of aerosols will affect the simulated meteorologyvariables. Here we present the role of atmospheric chemistry in the modeled atmosphere. Figure 12 showsthe comparison of modeled meteorology variables with observations. We can clearly seen that the modeledmeteorology variables (temperature, relative humidity, wind direction, and wind speed) are improved quitewell for both model (WRF-FINN and WRF-GFED).

22

Page 30: Simulating Air Pollution in the Severe Fires Event during

(a) daily mean temperature (b) daily mean relative humidity

(c) daily mean wind direction (d) daily mean wind speed

Figure 12: A comparison of PM10 A comparison of modeled daily mean (a) temperature, (b) relative humid-ity, (c) wind direction, and (d) wind speed with the observations in Kototabang.

Table 8 summarizes the statistical metric of all meteorology variables from WRF-FINN and WRF-GFED.Compared to statistical metric in table 5 (without boosting and modification of biomass burning emissions),the statistical metric of simulated meteorological variables are improved. The r2 of modeled temperatureimproved +55% for WRF-FINN and +72% for WRF-GFED. Meanwhile, the NMBF of the modeled tem-perature from WRF-FINN and WRF-GFED reduced to 0.066 and 0.057, respectively. This means that thebiomass burning boosting is reducing the surface temperature. The presence of aerosol produced by forestfires plays a significant role in reducing the ground temperature. The reduction in NMBF of relative humidityto 0.07 and 0.02 for WRF-FINN and WRF-GFED, respectively, shows the elevation of aerosol concentrationsin the modeled atmosphere succeed to suppress the relative humidity. Aerosol components may absorb thewater vapor and therefore reduce the relative humidity.

Variables T RH WDIR WSPDr2 NMBF r2 NMBF r2 NMBF r2 NMBF

WRF-METEO 0.13 0.088 0.26 0.12 0.14 0.47 0.07 0.43WRF-FINN 0.24 0.082 0.53 0.11 0.21 0.37 0.27 0.38WRF-GFED 0.36 0.074 0.61 0.09 0.31 0.35 0.39 0.26

Table 8: Summary of model skill in reproducing daily mean temperature, relative humidity, wind speed, andwind direction at Kototabang. This evaluation is done for 5-30 October 2015.

23

Page 31: Simulating Air Pollution in the Severe Fires Event during

4.4 Summary and ConclusionsModel simulations with boosted emission and enhanced emissions in last 10 days have a superior per-

formance for meteo and chemistry. This simulation likely corrects to some degree for emission otherwisemissed, such as those from smoldering peat fires. However, WRF-Chem using GFED is still shifted 3 daysahead that is possibly due to earlier fire onset. Underestimation is caused by thick haze during these periodso that the satellite could not detect fires on the ground.

Modifying GFED biomass burning emissions input on 26-30 October 2015 is done in two ways. First, wemodified emissions on these period using the same emissions on 21-25 October 2015. Second, we modifiedemissions on these period using flat emissions equal to emissions on 21 October 2015. As we can see, thisexperiment succeed to improve the pollutant concentrations on 26-30 October 2015. However, second modi-fication shows significant improvement to the model. It is able to capture high CO and PM10 concentrationon 26-30 October 2015 in Kototabang. This important finding leads to a conclusion that thick haze can coverthe satellite from seeing fires on the ground hence the emissions estimate are underestimated.

Modifying emissions inventory in this case is very important because we always underestimate the obser-vations. We see the significant improvement by boosting the emissions to 30% in the simulated meteorologyvariables and the chemistry variables as well. This shows the importance of chemistry scheme coupled inmeteorology model. As this biomass burning events related to drought caused by El Nino, the investigationon modeling precipitation is interesting. The next question we try to answer is how is the feedback of biomassburning emissions to the precipitation?

24

Page 32: Simulating Air Pollution in the Severe Fires Event during

5 Effects of biomass burning aerosols on precipitation in IndonesiaIn this section, we present the feedback mechanism of biomass burning aerosols and precipitation over

Indonesia. Results from previous sections indicate that chemistry scheme is necessary to improve the modeledmeteorology variables in WRF-Chem model. We hypothesize that the inclusion of biomass burning emissionwill also influence on the precipitation over Indonesia.

5.1 Theoretical background and previous studiesAerosols are among the most important influences on precipitation. The physical mechanism suggested

that a large number of small cloud droplet is formed by the biomass burning aerosols that act as cloud con-densation nuclei. Thus, warm rain suppression occurs in polluted environments. The mechanisms for theseinfluences are complicated, but several studies have shown that precipitation is significantly influenced byatmospheric aerosols (Mashayekhi & Sloan, 2014; Gonçalves et al., 2015; Vendrasco et al., 2009; J. C. Linet al., 2006). Huang et al. (2016) shows the suppression of precipitation due to aerosol concentration en-hancement occurred in China during an intensive biomass burning event in 2012. In Indonesia itself, biomassburning aerosol particles from 2013 forest fires suppressed the precipitation as the water droplet size reduced.(Kusumaningtyas & Aldrian, 2016; L’Ecuyer et al., 2009).

High biomass burning aerosol concentrations led to an increase in the cloud water mixing ratio, a decreasein the rainwater mixing ratio, and an increase in droplet numbers (G. Grell et al., 2011). On the other hand, theinfluences of aerosol concentrations on a precipitating system are also regulated by the atmospheric degreeof instability (Gonçalves et al., 2015). Precipitation increases with the increase of aerosol concentrations inan unstable atmosphere.

In this section we will compare WRF model without Chemistry (WRF-METEO) and with GFED inven-tory (WRF-GFED) with the observed precipitation. We neglect results from WRF-FINN because this modelshows less improvement compared to WRF-GFED. The aim of this study is to focus on the feedback mech-anism of aerosols from biomass burning emissions to precipitation over Indonesia. The similar studies havebeen conducted in several types of research except for Indonesia, especially in El-Niño episodes.

5.2 MethodologyIn this section, we investigated the influence of biomass burning aerosols on precipitation in the WRF-

Chem model. In the WRF-Chem model, total precipitation is calculated by the sum of convective and non-convective precipitation. The convective precipitation involved processes such as vertical transport of watervapor and the presence of resolved clouds affects convection. Also, aerosols have indirect effects on cumulusformation by thermal mechanisms that include direct radiative transfer and their formation of resolved clouds.It is not possible to make a clear distinction between aerosol thermal effects and aerosol microphysics on theformation of resolved clouds, but the largest effects of aerosols on convective rain are thermal in origin, andof course, their largest effects on resolved clouds are microphysical. A simple way to estimate the magnitudeof the effects caused by the addition of chemistry to the simulation is to compare the results from WRF-Chemwith those with the same WRF configuration, with the chemistry turned off.

In this study, we used two simulation scheme from the previous (see section 3 and 4) experiments. Asshown in the section 4., WRF-Chem model with the inclusion of GFED biomass burning inventory that hasbeen boosted by 30% of emission factor and modified with flat emissions on 21-30 October 2015 (WRF-GFED_flat) is the best model for four simulations. We, therefore, use WRF-METEO and WRF-GFED_flatto investigate the effect of biomass burning aerosols on precipitation.

Due to the absence of ground observation for precipitation, we used a monthly mean precipitation datafrom Global Precipitation Climatology Project (GPCP) data to evaluate the model results. GPCP data com-

25

Page 33: Simulating Air Pollution in the Severe Fires Event during

bines observations (rain gauge station and sounding observations) and satellite precipitation data into 2.5◦×2.5◦global grids. GPCP data is provided by the NOAA/OAR/ESRL PSD, Boulder, Colorado, USA, fromtheir website at http://www.esrl.noaa.gov/psd/.

The drawback of using the GPCP data is we have to include five days spin-up time in the calculation ofmonthly mean precipitation. Consequently, we allow some expected biases in the simulated precipitations tooccur.

5.3 Comparison with GPCP satellite dataFigure 13 shows the linear regression of (a) WRF-METEO and GPCP and (b) WRF-GFED_flat with

GPCP. In general, both simulations perform quite well with r2 of 0.96 and 0.98 for WRF-METEO andWRF-GFED_flat, respectively. However, the graph is also indicating the overestimation of the simulatedprecipitation.

In general, both simulations produce the same pattern with the GPCP data (fig. 14). However, an un-derestimation of precipitation on the northern part of Indonesia is apparent. This region is the downwindarea of forest fire emissions. Meanwhile, the several locations in the southern part of Indonesia exhibit aslight overestimation (mean bias is 0.14mm/day and 0.026mm/day for WRF-METEO and WRF-GFED_flat,respectively). Despite the biases, WRF-GFED_flat shows the best performance in modeling precipitation.Mean precipitation is 1.395mm/day and 1.281mm/day for WRF-METEO and WRF-GFED_flat, respectively.These mean modeled precipitations are slightly higher than mean precipitation of GPCP which is 1.255mm/-day. This discrepancy is likely caused by the inclusion of spin-up time in the precipitation, the coarse hor-izontal resolution of the domain which is unable to resolve the cloud convective in the grid cell, and thelimitations in the Grell 3-D convective scheme, which could be a matter for future examination.

Changes in precipitation patterns occur when GFED biomass burning aerosols are simulated, though theseeffects occur over a small region. Specifically, there is a decrease in modeled precipitation over Kalimantanand Sumatra, but an increase in the ocean at the biomass burning pollution downwind area as shown in fig.14. This result is in line with the results of Ott et al. (2010). Precipitation over the ocean is higher when thebiomass burning aerosol is incorporated in the model. However, they simulated without the indirect effectin the model. The presence of aerosol concentrations in the modeled atmosphere can improve the modeledprecipitation over Indonesia during a server forest fires event. The presence of biomass burning aerosolsuppressed the monthly mean precipitation by 8%.

(a) Simulated total daily precipitation averagedfor one-month using WRF-METEO

(b) Simulated total daily precipitation averagedfor one-month using WRF-GFED

Figure 13: Linear regression of simulated precipitation and GPCP precipitation.

26

Page 34: Simulating Air Pollution in the Severe Fires Event during

(a) Simulated total daily precipitation averaged for one-month using WRF-METEO

(b) Simulated total daily precipitation averaged for one-month using WRF-GFED

(c) GPCP precipitation

Figure 14: Spatial variability of the modeled precipitation

27

Page 35: Simulating Air Pollution in the Severe Fires Event during

5.4 The influence of aerosol scheme on the modeled precipitationFigure 14 shows the difference of simulated precipitation from WRF-METEO and WRF-GFED_flat. The

precipitation suppression occurs when we incorporate the GFED inventory in the model. Over the locationsof high biomass burning aerosol, the suppression is more pronounced (figure 15). Biomass burning aerosolsare thought to suppress precipitation because the aerosol particles decrease the size of water droplets in cloudsand, thus, increases the lifetime of the cloud and delays the onset of precipitation (Andreae & Merlet, 2001)

(a) WRF-GFED_flat - WRF-METEO

(b) PM25 concentration

Figure 15: Monthly average of (a) precipitation difference and (b) the PM25 concentration over Indonesia

Figure 15 denotes that the reductions of the monthly simulated precipitation are in the area where thehigh PM25 concentrations are distributed. However, the maximum precipitation suppression occurred in thesouthern part of Kalimantan where the PM25 concentration is 50% lower than the PM25 concentration atthe south part of Sumatra. The nonlinear influence of biomass burning aerosol on reducing precipitationcorrespondence to the regional weather pattern. From the HYSPLIT trajectory model (see Appendix F),we can see that the air mass is originated from the the Java strait and brings moistures to Sumatra. Themountainous landscape of Sumatra causes the upward deflection of the horizontal air mass flow. The coolingadiabatic upon ascent allowing for condensation, and hence precipitation.

28

Page 36: Simulating Air Pollution in the Severe Fires Event during

Figure 16: Domain-averaged of temperature difference between WRF-GFED_flat and WRF-METEO

The presence of biomass burning aerosol concentrations is also causing the "dome effect," a layer ofbiomass burning aerosol particles that absorbed the sunlight prevents the solar light reaching the ground. Thiseffect is due to light-absorbing carbonaceous aerosols emitted by the biomass burning emissions (Reid et al.,1998). Figure 16 shows the temperature difference between WRF-GFED_flat and WRF-METEO runs showsthe biomass burning aerosols cool the surface and while warm the planetary boundary layer. A substantialcooling is shown on the surface. During the thick haze of air pollution event (21-29 October 2015), a strongwarming with a maximum of 0.27◦C is produced around 900 hPa, located near the aerosol layer.

Figure 17: A one month average of shortwave radiation difference from WRF-GFED_flat and WRF-METEO

For the radiative fluxes, the biomass burning aerosols reduce downward shortwave fluxes reaching thesurface with the maximum reduction occur over the location of fires (figure 17). A modest change in theradiative fluxes is produced when biomass burning aerosol is included in the model. Absorbing aerosols,such as black carbon heats the local atmosphere and possibly reduce cloud formation.

29

Page 37: Simulating Air Pollution in the Severe Fires Event during

5.5 ConclusionsThe inclusion of biomass burning emissions in the WRF-Chem model affecting several changes which

are suppressing precipitation, cooling down the surface temperature, and reducing downward solar radiationon the surface. The changes are due to injection of biomass burning aerosol particles in the atmospherewhich absorbing the sunlight and preventing the sunlight from reaching the surface. These changes obviouslyoccurred in the region where PM2.5 concentrations were exceeding 50µgm−3.

30

Page 38: Simulating Air Pollution in the Severe Fires Event during

6 Summary and Outlook

6.1 SummaryBetween September and November 2015, severe forest fires occurred during high signal of El-Niño in

Indonesia. The high level of aerosol optical depth (AOD) over Indonesia is indicative of the high levels ofair pollution. Motivated by this event, we conducted research on simulating the air pollution over Indone-sia in October 2015. We run three separate WRF-Chem simulations using no-chemistry input and biomassburning emissions input (FINN and GFED biomass burning inventories). Each of the simulations is run for1-month fire episode, with a horizontal resolution of 30×30 km2. Meteorology, AOD, and chemistry obser-vations used to evaluate the model performance. We utilized point, sounding, and satellite data obtained fromground stations, Wyoming University, Moderate Resolution Imaging Spectroradiometer (MODIS) satellite,and Global Precipitation Climatology Project (GPCP) satellite for mean monthly precipitation.

WRF-Chem simulations in Kototabang shows that the model poorly captures the evolution of meteorol-ogy variables (temperature, relative humidity, wind speed and wind direction). However, the model resultsimproved the model results significantly once we include chemistry and aerosol scheme. The improvementis indicated by an increase of r2 in all variables. We also evaluated the model with sounding profiles oftemperature and relative humidity from three separated measurement locations (Padang, Pekanbaru, and Sin-gapore). The model results can capture modeled temperature and relative humidity vertical profile quite well.Even though we saw large biases for all modeled variables, we conclude that the inclusion of chemistry andaerosol mechanism indeed improved the model results. The choice of biomass burning inventory also influ-ences the results by producing different amount of chemistry species concentrations. Overall, WRF-Chemusing GFED inventory performs better than WRF-Chem with FINN inventory. The biases are possibly causedby the unrepresentative of the model resolution (too coarse horizontal resolution).

Atmospheric chemistry evaluation was done for PM10 and PM2.5 concentrations and CO mixing ratioat Kototabang, Pekanbaru, and Singapore. Simulated concentrations underestimate the observations. InKototabang, both simulation results can capture some peaks. However, they failed to capture peak on 26-31October 2015. However, WRF-Chem with FINN input has a larger underestimation than underestimationof PM10 concentration and CO mixing ratio, than with GFED input. The evaluation indicates that GFEDbiomass burning input gives more realistic emissions than FINN at least over Indonesia in October 2015.Uncertainties of emission factor estimate and burned area contribute to the underestimation of simulationresults. In Aouizerats et al.(2015), emissions from GFED inventory need to be boosted by 28% of emissionfactors. We suggested boosting the emissions slightly higher (to 30% of emissions) than suggested becausethe 2015 forest fires event is more extreme than 2006 forest fires event. After boosting the emission factor,the simulated CO and aerosol concentrations increased. Nonetheless, simulation with increased emission stillunderestimates CO and aerosol concentration during the thick haze of air pollution period (21-29 October2015). We attribute this to misinterpretation of satellite in detecting burned area being obscured by the verythick haze.

We extended our investigation to improve the simulation during the very thick haze of air pollution pe-riod by modifying the biomass burning emissions in that period. We conducted two experiments: repeatingemissions at 21-25 October 2015 (hereafter WRF-GFED_rep) and extending the high emission of 20 Octo-ber to 30 October 2015 (hereafter WRF-GFED_flat). The WRF-GFED_flat succeed to improve the pollutantconcentrations on 26-30 October 2015. The extended emission shows a significant improvement in the modelresults. This finding leads to a conclusion that thick haze can cover the satellite and underestimate burnedarea on the ground, leading to substantial underestimation in the biomass burning inventory. Modifying thebiomass burning inventory in this case (an extreme fires episode) is critical.

Further, we investigated the influence of biomass burning emissions inclusion on simulated precipitationin Indonesia during fires episode. We used WRF-GFED_flat as this is the best model compared to other

31

Page 39: Simulating Air Pollution in the Severe Fires Event during

models. We used a monthly precipitation dataset obtained from GPCP with a coarse horizontal resolutionof 2.5◦× 2.5◦global grids. Based on WRF-Chem simulations without and with biomass burning aerosol in-clusion (WRF-METEO and WRF-GFED, respectively), we conclude that the inclusion of biomass burningemission improves the model performance in simulating precipitation over Indonesia. However, both simula-tions are showing a slight overestimation of monthly averaged precipitation systematically. The discrepanciesare possibly caused by the difference resolution between the model and the observation and the small portionof fine particles in the finest size bins in the WRF-Chem aerosol scheme.

6.2 OutlookSuggestion for future study related to the extreme forest fires event is employing ground measurement

of biomass burning emissions as the emissions input in WRF-Chem to overcome the undetected emissionsby the satellite retrieval. We also overestimated precipitation because we only put 10% of aerosol particlesin the finest size bin in MOSAIC aerosol scheme. We, therefore, suggest to put a larger portion of aerosolsin the finest size bin or use the 8-bins instead of 4-bins. Additionally, a discrepancy is also caused by theunderrepresentation of the model resolutions. Suggestion for future study is to use a finer horizontal resolutionto resolve the small-scale circulation and the cloud convective.

32

Page 40: Simulating Air Pollution in the Severe Fires Event during

ReferencesAkagi, S. K., Yokelson, R. J., Wiedinmyer, C., Alvarado, M. J., Reid, J. S., Karl, T., . . . Wennberg, P. O.

(2011). Emission factors for open and domestic biomass burning for use in atmospheric models. At-mospheric Chemistry and Physics, 11(9), 4039–4072. doi: 10.5194/acp-11-4039-2011

Andreae, M. O., & Merlet, P. (2001). Emission of trace gases and aerosols from biomass burning. GlobalBiogeochemical Cycles, 15(4), 955–966. doi: 10.1029/2000GB001382

Aouizerats, B., van Der Werf, G. R., Balasubramanian, R., & Betha, R. (2015). Importance of transboundarytransport of biomass burning emissions to regional air quality in Southeast Asia during a high fire event.Atmospheric Chemistry and Physics. doi: 10.5194/acp-15-363-2015

Crippa, P., Castruccio, S., Lebron, G. B., Kuwata, M., & Thota, A. (2015). Population exposure to hazardousair quality due to the 2015 fires in Equatorial Asia. Nature Publishing Group, 1–9. Retrieved fromhttp://dx.doi.org/10.1038/srep37074 doi: 10.1038/srep37074

Gaveau, D. L. a., Salim, M. a., Hergoualc’h, K., Locatelli, B., Sloan, S., Wooster, M., . . . Sheil, D. (2014).Major atmospheric emissions from peat fires in Southeast Asia during non-drought years: evidencefrom the 2013 Sumatran fires. Scientific reports, 4, 1–7. doi: 10.1038/srep06112

Giglio, L., Randerson, J. T., & Van Der Werf, G. R. (2013). Analysis of daily, monthly, and annual burnedarea using the fourth-generation global fire emissions database (GFED4). Journal of GeophysicalResearch: Biogeosciences, 118(1), 317–328. doi: 10.1002/jgrg.20042

Gonçalves, W. A., Machado, L. A. T., & Kirstetter, P.-E. (2015). Influence of biomass aerosol on precipitationover the Central Amazon: an observational study. Atmospheric Chemistry and Physics, 15(12), 6789–6800. Retrieved from http://www.atmos-chem-phys.net/15/6789/2015/ doi: 10.5194/acp-15-6789-2015

Grell, G., Freitas, S. R., Stuefer, M., & Fast, J. (2011). and Physics Inclusion of biomass burning in WRF-Chem : impact of wildfires on weather forecasts. , 5289–5303. doi: 10.5194/acp-11-5289-2011

Grell, G. a. (2002). A generalized approach to parameterizing convection combining ensemble and dataassimilation techniques. Geophysical Research Letters, 29(14), 10–13. doi: 10.1029/2002GL015311

Grell, G. A., Peckham, S. E., Schmitz, R., McKeen, S. A., Frost, G., Skamarock, W. C., & Eder, B.(2005). Fully coupled "online" chemistry within the WRF model. Atmospheric Environment. doi:10.1016/j.atmosenv.2005.04.027

Guenther, A. B., Jiang, X., Heald, C. L., Sakulyanontvittaya, T., Duhl, T., Emmons, L. K., & Wang, X.(2012). The model of emissions of gases and aerosols from nature version 2.1 (MEGAN2.1): Anextended and updated framework for modeling biogenic emissions. Geoscientific Model Development,5(6), 1471–1492. doi: 10.5194/gmd-5-1471-2012

Huang, X., Ding, A., Liu, L., Liu, Q., Ding, K., Niu, X., . . . Fu, C. (2016). Effects of aerosol-radiationinteraction on precipitation during biomass-burning season in East China. Atmospheric Chemistry andPhysics, 16(15), 10063–10082. doi: 10.5194/acp-16-10063-2016

Huijnen, V., Wooster, M. J., Kaiser, J. W., Gaveau, D. L. A., Flemming, J., Parrington, M., . . . Van Weele,M. (2016). Fire carbon emissions over maritime southeast Asia in 2015 largest since 1997. scientificreports, 6:26886(31 May 2016). doi: 10.1038/srep26886

Iacono, M. J., Delamere, J. S., Mlawer, E. J., Shephard, M. W., Clough, S. A., & Collins, W. D. (2008).Radiative forcing by long-lived greenhouse gases: Calculations with the AER radiative transfer models.Journal of Geophysical Research Atmospheres. doi: 10.1029/2008JD009944

Kim, P. S., Jacob, D. J., Mickley, L. J., Koplitz, S. N., Marlier, M. E., DeFries, R. S., . . . Mao, Y. H.(2015). Sensitivity of population smoke exposure to fire locations in Equatorial Asia. AtmosphericEnvironment, 102(June 2013), 11–17. doi: 10.1016/j.atmosenv.2014.09.045

Kusumaningtyas, S. D. A., & Aldrian, E. (2016). Impact of the June 2013 Riau province Sumatera smokehaze event on regional air pollution. Environmental Research Letters, 11(7), 75007. Retrieved from

33

Page 41: Simulating Air Pollution in the Severe Fires Event during

http://stacks.iop.org/1748-9326/11/i=7/a=075007 doi: 10.1088/1748-9326/11/7/075007L’Ecuyer, T. S., Berg, W., Haynes, J., Lebsock, M., & Takemura, T. (2009). Global observations of aerosol

impacts on precipitation occurrence in warm maritime clouds. Journal of Geophysical Research Atmo-spheres, 114(9), 1–15. doi: 10.1029/2008JD011273

Lin, J. C., Matsui, T., Pielke, S. A., & Kummerow, C. (2006). Effects of biomass-burning-derived aerosolson precipitations and clouds in the Amazon Basin: A satellite-based empirical study. Journal of Geo-physical Research Atmospheres, 111(19), 1–14. doi: 10.1029/2005JD006884

Lin, Y.-L., Farley, R. D., & Orville, H. D. (1983). Bulk Parameterization of the Snow Field in a Cloud Model(Vol. 22) (No. 6). doi: 10.1175/1520-0450(1983)022<1065:BPOTSF>2.0.CO;2

Marlier, M., DeFries, R., Voulgarakis, A., Kinney, P., Randerson, J., Shindell, D., . . . Faluvegi, G. (2012). ElNiño and health risks from landscape fire emissions in southeast Asia. Nature Climate Change, 2(8),1–6. Retrieved from http://dx.doi.org/10.1038/nclimate1658 doi: 10.1038/nclimate1658

Mashayekhi, R., & Sloan, J. J. (2014). Effects of aerosols on precipitation in north-eastern North America.Atmospheric Chemistry and Physics, 14(10), 5111–5125. doi: 10.5194/acp-14-5111-2014

Mlawer, E., Taubman, S., Brown, P., Iacono, M., & Clough, S. (1997). Radiative transfer for inhomogeneousatmosphers: RRTM, a validated correlated0k model for the long-wave. J. Geophys. Res., 102, 16663–16682.

Nuryanto, D. E. (2015). Simulation of Forest Fires Smoke Using WRF-Chem Model withFINN Fire Emissions in Sumatera. Procedia Environmental Sciences, 24, 65–69. Retrievedfrom http://www.sciencedirect.com/science/article/pii/S187802961500078X doi:10.1016/j.proenv.2015.03.010

Ott, L., Duncan, B., Pawson, S., Colarco, P., Chin, M., Randles, C., . . . Nielsen, E. (2010). Influence ofthe 2006 Indonesian biomass burning aerosols on tropical dynamics studied with the GEOS-5 AGCM.Journal of Geophysical Research Atmospheres, 115(14), 1–16. doi: 10.1029/2009JD013181

Page, S. E., Siegert, F., Rieley, J. O., Boehm, H.-D. V., Jaya, A., & Limin, S. (2002). The amount of carbonreleased from peat and forest fires in Indonesia during 1997. Nature, 420(6911), 61–65. Retrievedfrom http://www.nature.com/doifinder/10.1038/nature01131 doi: 10.1038/nature01131

Powers, J. G., Cavallo, S. M., & Manning, K. W. (2010). Improvements to radiation on upper-levelWRF performance over the Antarctic. WRF Workshop: 2010 proceedings, Suppl 7. Retrieved fromhttp://www.ncbi.nlm.nih.gov/pubmed/22213910

Reddington, C. L., Yoshioka, M., Balasubramanian, R., Ridley, D., Toh, Y. Y., Arnold, S. R., &Spracklen, D. V. (2014). Contribution of vegetation and peat fires to particulate air pol-lution in Southeast Asia. Environmental Research Letters, 9(9), 094006. Retrieved fromhttp://stacks.iop.org/1748-9326/9/i=9/a=094006 doi: 10.1088/1748-9326/9/9/094006

Reid, S., Hobbs, V., Vanderlei, J., Weiss, E., & Eck, F. (1998). absorption and black carbon content ofaerosols from biomass burning in Brazil independent of wavelength and has a mean value of _ osextinction - Os + o •. , 103(98), 31–32.

Rosenfeld, D. (1999). TRMM observed first direct evidence of smoke from forest fires inhibiting rainfall.Geophysical Research Letters, 26(20), 3105–3108. doi: 10.1029/1999GL006066

Stockwell, C. E., Jayarathne, T., Cochrane, M. A., Ryan, K. C., Putra, E. I., & Yokelson, R. J. (2016). Fieldmeasurements of trace gases and aerosols emitted by peat fires in Central Kalimantan , Indonesia ,during the 2015 El Niño. , 11711–11732. doi: 10.5194/acp-16-11711-2016

van Der Werf, G. R., Randerson, J. T., Giglio, L., Collatz, G. J., Mu, M., Kasibhatla, P. S., . . . Van Leeuwen,T. T. (2010). Global fire emissions and the contribution of deforestation, savanna, forest, agricul-tural, and peat fires (1997-2009). Atmospheric Chemistry and Physics, 10(23), 11707–11735. doi:10.5194/acp-10-11707-2010

Vendrasco, E., Silva Dias, P., & Freitas, E. (2009, 11). A case study of the direct radiative effect of biomassburning aerosols on precipitation in the Eastern Amazon. Atmospheric Research, 94(3), 409–421.

34

Page 42: Simulating Air Pollution in the Severe Fires Event during

Retrieved from http://linkinghub.elsevier.com/retrieve/pii/S0169809509001951 doi:10.1016/j.atmosres.2009.06.016

Visser, A. (2016). Simulating nitrogen oxides chemistry in Europe using a complex chemistry scheme inWRF-Chem.

Weinstock, B. (1969). Carbon Monoxide: Residence Time in the Atmosphere. Science, 166(3902), 224–225.Retrieved from http://www.sciencemag.org/cgi/doi/10.1126/science.166.3902.224 doi:10.1126/science.166.3902.224

Wesely, M. L. (1989). Parameterization of Surface Resistances To Gaseous Dry Deposition in Regional-ScaleNumerical Models. Atmospheric Chemistry and Physics, 23(6), 1293–1304.

Wiedinmyer, C., Akagi, S. K., Yokelson, R. J., Emmons, L. K., Al-Saadi, J. a., Orlando, J. J., & Soja, a. J.(2010). The Fire INventory from NCAR (FINN) – a high resolution global model to estimate theemissions from open burning. Geoscientific Model Development Discussions, 3(4), 2439–2476. doi:10.5194/gmdd-3-2439-2010

World Bank Group. (2016). The Cost of Fire. Indonesia Sustainable Landscapes Knowledge Note:1(February).

Xiao, Z. M., Zhang, Y. F., Hong, S. M., Bi, X. H., Jiao, L., Feng, Y. C., & Wang, Y. Q. (2011). Estimationof the main factors influencing haze, based on a long-term monitoring campaign in Hangzhou, China.Aerosol and Air Quality Research, 11(7), 873–882. doi: 10.4209/aaqr.2011.04.0052

Yu, S., Eder, B., Dennis, R., Chu, S.-H., & Schwartz, S. E. (2006). New unbiased symmetric metricsfor evaluation of air quality models. Atmospheric Science Letters, 7(1), 26–34. Retrieved fromhttp://doi.wiley.com/10.1002/asl.125 doi: 10.1002/asl.125

Zaveri, R. A., Easter, R. C., Fast, J. D., & Peters, L. K. (2008). Model for Simulating Aerosol Interac-tions and Chemistry (MOSAIC). Journal of Geophysical Research Atmospheres, 113(13), 1–29. doi:10.1029/2007JD008782

Zaveri, R. A., & Peters, L. K. (1999). A new lumped structure photochemical mechanism for large-scaleapplications. Journal of Geophysical Research, 104(D23), 30387. doi: 10.1029/1999JD900876

Zhao, C., Tie, X., & Lin, Y. (2006). A possible positive feedback of reduction of precipitation and in-crease in aerosols over eastern central China. Geophysical Research Letters, 33(11), 2–5. doi:10.1029/2006GL025959

35

Page 43: Simulating Air Pollution in the Severe Fires Event during

A Modifications in WRF-Chem codeThis appendix describes the code modifications that were performed for this MSc-thesis research. Those

modifications were performed for WRF-Chem version 3.2.1.

A.1 Adding CBMZ-MOSAIC routineCBMZ-MOSAIC chemistry scheme is not available in the module_add_emiss_burn.F. We therefore need

to add a routine to define CBMZ-MOSAIC chemistry scheme in this module. In this study, we used 4size-bins with aqueous chemistry in the MOSAIC aerosol module. We used 28 gas-phase and two aerosolschemical species. We calculated the chemical species concentration as formula below:

Xconcentration = Xemission ∗4.828e−4

ρ ∗∆t(5)

With X is a gas-phase chemical species, ρ is a density of the chemical species and ∆ t is the model timesteps. We only used organic carbon (OC) and black carbon (BC) as the aerosol-phase chemical species. Wedivided the OC and BC into 4size-bins with distribution of 20% for 0.039-0.10 µm bin, 20% for 0.10-1.0 µmbin, 50% for 1.0-2.5 µm bin, and 10% for 2.5-10 µm bin (fig. 17 illustrates the partitioning of the aerosolparticles in bins size)

Figure 18: The distribution of aerosol particles in four different bins size used in MOSAIC aerosol scheme.

CASE (CBMZ_MOSAIC_4BIN, CBMZ_MOSAIC_4BIN_AQ)if( biomass_burn_opt == BIOMASSB_MOZC ) thendo j=jts,jtedo k=kts,ktedo i=its,ite

!unit of conv_rho = mol kg-1! 4.828 --> molar mass of air (28.966e-3 kg/mol) / 60! rho_phy --> density (kg/m3)! dt --> seconds --> check the time steps units in the namelist.! dz8w --> m

conv_rho=r_q*4.828e-4/rho_phy(i,k,j)*dtstep/60./dz8w(i,k,j)

chem(i,k,j,p_so2) = chem(i,k,j,p_so2) + ebu(i,k,j,p_ebu_so2)*conv_rho

36

Page 44: Simulating Air Pollution in the Severe Fires Event during

chem(i,k,j,p_no2) = chem(i,k,j,p_no2) + ebu(i,k,j,p_ebu_no2)*conv_rhochem(i,k,j,p_no) = chem(i,k,j,p_no) + ebu(i,k,j,p_ebu_no)*conv_rhochem(i,k,j,p_hcho) = chem(i,k,j,p_hcho) + ebu(i,k,j,p_ebu_ch2o)*conv_rhochem(i,k,j,p_nh3) = chem(i,k,j,p_nh3) + ebu(i,k,j,p_ebu_nh3)*conv_rhochem(i,k,j,p_co) = chem(i,k,j,p_co) + ebu(i,k,j,p_ebu_co)*conv_rhochem(i,k,j,p_tol) = chem(i,k,j,p_tol) + ebu(i,k,j,p_ebu_toluene)*conv_rhochem(i,k,j,p_ch3oh) = chem(i,k,j,p_ch3oh) + ebu(i,k,j,p_ebu_ch3oh)*conv_rhochem(i,k,j,p_c2h5oh) = chem(i,k,j,p_c2h5oh)+ ebu(i,k,j,p_ebu_c2h5oh)*conv_rhochem(i,k,j,p_open) = chem(i,k,j,p_open) + ebu(i,k,j,p_ebu_open)*conv_rhochem(i,k,j,p_bigalk) = chem(i,k,j,p_bigalk) + ebu(i,k,j,p_ebu_bigalk)*conv_rhochem(i,k,j,p_bigene) = chem(i,k,j,p_bigene) + ebu(i,k,j,p_ebu_bigene)*conv_rhochem(i,k,j,p_c10h16) = chem(i,k,j,p_c10h16) +ebu(i,k,j,p_ebu_c10h16)*conv_rhochem(i,k,j,p_c2h4) = chem(i,k,j,p_c2h4) +ebu(i,k,j,p_ebu_c2h4)*conv_rhochem(i,k,j,p_c2h6) = chem(i,k,j,p_c2h6) +ebu(i,k,j,p_ebu_c2h6)*conv_rhochem(i,k,j,p_c3h6) = chem(i,k,j,p_c3h6) +ebu(i,k,j,p_ebu_c3h6)*conv_rhochem(i,k,j,p_c3h8) = chem(i,k,j,p_c3h8) +ebu(i,k,j,p_ebu_c3h8)*conv_rhochem(i,k,j,p_ald) = chem(i,k,j,p_ald) +ebu(i,k,j,p_ebu_ch3cho)*conv_rhochem(i,k,j,p_acet) = chem(i,k,j,p_acet) +ebu(i,k,j,p_ebu_ch3coch3)*conv_rhochem(i,k,j,p_mgly) = chem(i,k,j,p_mgly) +ebu(i,k,j,p_ebu_mgly)*conv_rhochem(i,k,j,p_gly) = chem(i,k,j,p_gly) +ebu(i,k,j,p_ebu_gly)*conv_rhochem(i,k,j,p_ch3cooh) = chem(i,k,j,p_ch3cooh) +ebu(i,k,j,p_ebu_ch3cooh)*conv_rhochem(i,k,j,p_cres) = chem(i,k,j,p_cres) +ebu(i,k,j,p_ebu_cres)*conv_rhochem(i,k,j,p_glyald) = chem(i,k,j,p_glyald) +ebu(i,k,j,p_ebu_glyald)*conv_rhochem(i,k,j,p_macr) = chem(i,k,j,p_macr) +ebu(i,k,j,p_ebu_macr)*conv_rhochem(i,k,j,p_mek) = chem(i,k,j,p_mek) +ebu(i,k,j,p_ebu_mek)*conv_rhochem(i,k,j,p_mvk) = chem(i,k,j,p_mvk) +ebu(i,k,j,p_ebu_mvk)*conv_rhochem(i,k,j,p_oc_a01) = chem(i,k,j,p_oc_a01) &

+r_q*0.2*ebu(i,k,j,p_ebu_oc)/rho_phy(i,k,j)*dtstep/dz8w(i,k,j)chem(i,k,j,p_oc_a02) = chem(i,k,j,p_oc_a02) &

+r_q*0.2*ebu(i,k,j,p_ebu_oc)/rho_phy(i,k,j)*dtstep/dz8w(i,k,j)chem(i,k,j,p_oc_a03) = chem(i,k,j,p_oc_a03) &

+r_q*0.5*ebu(i,k,j,p_ebu_oc)/rho_phy(i,k,j)*dtstep/dz8w(i,k,j)chem(i,k,j,p_oc_a04) = chem(i,k,j,p_oc_a04) &

+r_q*0.1*(ebu(i,k,j,p_ebu_oc))/rho_phy(i,k,j)*dtstep/dz8w(i,k,j)chem(i,k,j,p_bc_a01) = chem(i,k,j,p_bc_a01) &

+r_q*0.2*ebu(i,k,j,p_ebu_bc)/rho_phy(i,k,j)*dtstep/dz8w(i,k,j)chem(i,k,j,p_bc_a02) = chem(i,k,j,p_bc_a02) &

+r_q*0.2*ebu(i,k,j,p_ebu_bc)/rho_phy(i,k,j)*dtstep/dz8w(i,k,j)chem(i,k,j,p_bc_a03) = chem(i,k,j,p_bc_a03) &

+r_q*0.5*ebu(i,k,j,p_ebu_bc)/rho_phy(i,k,j)*dtstep/dz8w(i,k,j)chem(i,k,j,p_bc_a04) = chem(i,k,j,p_bc_a04) &

+r_q*0.1*ebu(i,k,j,p_ebu_bc)/rho_phy(i,k,j)*dtstep/dz8w(i,k,j)chem(i,k,j,p_oc_cw01) = chem(i,k,j,p_oc_cw01) &

+r_q*0.2*ebu(i,k,j,p_ebu_oc)/rho_phy(i,k,j)*dtstep/dz8w(i,k,j)chem(i,k,j,p_oc_cw02) = chem(i,k,j,p_oc_cw02) &

+r_q*0.2*ebu(i,k,j,p_ebu_oc)/rho_phy(i,k,j)*dtstep/dz8w(i,k,j)chem(i,k,j,p_oc_cw03) = chem(i,k,j,p_oc_cw03) &

+r_q*0.5*ebu(i,k,j,p_ebu_oc)/rho_phy(i,k,j)*dtstep/dz8w(i,k,j)chem(i,k,j,p_oc_cw04) = chem(i,k,j,p_oc_cw04) &

+r_q*0.1*ebu(i,k,j,p_ebu_oc)/rho_phy(i,k,j)*dtstep/dz8w(i,k,j)chem(i,k,j,p_bc_cw01) = chem(i,k,j,p_bc_cw01) &

+r_q*0.2*ebu(i,k,j,p_ebu_bc)/rho_phy(i,k,j)*dtstep/dz8w(i,k,j)

37

Page 45: Simulating Air Pollution in the Severe Fires Event during

chem(i,k,j,p_bc_cw02) = chem(i,k,j,p_bc_cw02) &+r_q*0.2*ebu(i,k,j,p_ebu_bc)/rho_phy(i,k,j)*dtstep/dz8w(i,k,j)

chem(i,k,j,p_bc_cw03) = chem(i,k,j,p_bc_cw03) &+r_q*0.5*ebu(i,k,j,p_ebu_bc)/rho_phy(i,k,j)*dtstep/dz8w(i,k,j)

chem(i,k,j,p_bc_cw04) = chem(i,k,j,p_bc_cw04) &+r_q*0.1*ebu(i,k,j,p_ebu_bc)/rho_phy(i,k,j)*dtstep/dz8w(i,k,j)

A.2 Adding plume rise moduleIn the WRF-Chem’s namelist.input, a routine for biomass burning emissions with MOZART model base

(this is for FINN) need to be activated. We add routine as follows:

if ( config_flags%biomass_burn_opt == BIOMASSB ) thendo j=jts,jte

do i=its,iteebu(i,kts,j,p_ebu_no) = ebu_in(i,1,j,p_ebu_in_no)ebu(i,kts,j,p_ebu_co) = ebu_in(i,1,j,p_ebu_in_co)ebu(i,kts,j,p_ebu_co2) = ebu_in(i,1,j,p_ebu_in_co2)ebu(i,kts,j,p_ebu_eth) = ebu_in(i,1,j,p_ebu_in_eth)ebu(i,kts,j,p_ebu_hc3) = ebu_in(i,1,j,p_ebu_in_hc3)ebu(i,kts,j,p_ebu_hc5) = ebu_in(i,1,j,p_ebu_in_hc5)ebu(i,kts,j,p_ebu_hc8) = ebu_in(i,1,j,p_ebu_in_hc8)ebu(i,kts,j,p_ebu_ete) = ebu_in(i,1,j,p_ebu_in_ete)ebu(i,kts,j,p_ebu_olt) = ebu_in(i,1,j,p_ebu_in_olt)ebu(i,kts,j,p_ebu_oli) = ebu_in(i,1,j,p_ebu_in_oli)ebu(i,kts,j,p_ebu_pm25) = ebu_in(i,1,j,p_ebu_in_pm25)ebu(i,kts,j,p_ebu_pm10) = ebu_in(i,1,j,p_ebu_in_pm10)ebu(i,kts,j,p_ebu_dien) = ebu_in(i,1,j,p_ebu_in_dien)ebu(i,kts,j,p_ebu_iso) = ebu_in(i,1,j,p_ebu_in_iso)ebu(i,kts,j,p_ebu_api) = ebu_in(i,1,j,p_ebu_in_api)ebu(i,kts,j,p_ebu_lim) = ebu_in(i,1,j,p_ebu_in_lim)ebu(i,kts,j,p_ebu_tol) = ebu_in(i,1,j,p_ebu_in_tol)ebu(i,kts,j,p_ebu_xyl) = ebu_in(i,1,j,p_ebu_in_xyl)ebu(i,kts,j,p_ebu_csl) = ebu_in(i,1,j,p_ebu_in_csl)ebu(i,kts,j,p_ebu_hcho) = ebu_in(i,1,j,p_ebu_in_hcho)ebu(i,kts,j,p_ebu_ald) = ebu_in(i,1,j,p_ebu_in_ald)ebu(i,kts,j,p_ebu_ket) = ebu_in(i,1,j,p_ebu_in_ket)ebu(i,kts,j,p_ebu_macr) = ebu_in(i,1,j,p_ebu_in_macr)ebu(i,kts,j,p_ebu_ora1) = ebu_in(i,1,j,p_ebu_in_ora1)ebu(i,kts,j,p_ebu_ora2) = ebu_in(i,1,j,p_ebu_in_ora2)ebu(i,kts,j,p_ebu_sulf) = ebu_in(i,1,j,p_ebu_in_sulf)ebu(i,kts,j,p_ebu_bc) = ebu_in(i,1,j,p_ebu_in_bc)ebu(i,kts,j,p_ebu_oc) = ebu_in(i,1,j,p_ebu_in_oc)ebu(i,kts,j,p_ebu_so2) = ebu_in(i,1,j,p_ebu_in_so2)ebu(i,kts,j,p_ebu_dms) = ebu_in(i,1,j,p_ebu_in_dms)

enddoenddo

elseif ( config_flags%biomass_burn_opt == BIOMASSB_MOZC ) thendo j=jts,jte

do i=its,iteebu(i,kts,j,p_ebu_so2) = ebu_in(i,1,j,p_ebu_in_so2)*1.3

38

Page 46: Simulating Air Pollution in the Severe Fires Event during

ebu(i,kts,j,p_ebu_no2) = ebu_in(i,1,j,p_ebu_in_no2)*1.3ebu(i,kts,j,p_ebu_no) = ebu_in(i,1,j,p_ebu_in_no)*1.3ebu(i,kts,j,p_ebu_ch2o) = ebu_in(i,1,j,p_ebu_in_ch2o)*1.3ebu(i,kts,j,p_ebu_nh3) = ebu_in(i,1,j,p_ebu_in_nh3)*1.3ebu(i,kts,j,p_ebu_co) = ebu_in(i,1,j,p_ebu_in_co)*1.3ebu(i,kts,j,p_ebu_toluene)= ebu_in(i,1,j,p_ebu_in_toluene)*1.3ebu(i,kts,j,p_ebu_ch3oh) = ebu_in(i,1,j,p_ebu_ch3oh)*1.3ebu(i,kts,j,p_ebu_c2h5oh) = ebu_in(i,1,j,p_ebu_c2h5oh)*1.3ebu(i,kts,j,p_ebu_open) = ebu_in(i,1,j,p_ebu_in_open)*1.3ebu(i,kts,j,p_ebu_bigalk) = ebu_in(i,1,j,p_ebu_bigalk)*1.3ebu(i,kts,j,p_ebu_bigene) = ebu_in(i,1,j,p_ebu_bigene)*1.3ebu(i,kts,j,p_ebu_c10h16) = ebu_in(i,1,j,p_ebu_in_c10h16)*1.3ebu(i,kts,j,p_ebu_c2h4) = ebu_in(i,1,j,p_ebu_c2h4)*1.3ebu(i,kts,j,p_ebu_c2h6) = ebu_in(i,1,j,p_ebu_c2h6)*1.3ebu(i,kts,j,p_ebu_c3h6) = ebu_in(i,1,j,p_ebu_c3h6)*1.3ebu(i,kts,j,p_ebu_c3h8) = ebu_in(i,1,j,p_ebu_c3h8)*1.3ebu(i,kts,j,p_ebu_ch3cho) = ebu_in(i,1,j,p_ebu_in_ch3cho)*1.3ebu(i,kts,j,p_ebu_ch3coch3) = ebu_in(i,1,j,p_ebu_ch3coch3)*1.3ebu(i,kts,j,p_ebu_mgly) = ebu_in(i,1,j,p_ebu_in_mgly)*1.3ebu(i,kts,j,p_ebu_gly) = ebu_in(i,1,j,p_ebu_in_gly)*1.3ebu(i,kts,j,p_ebu_ch3cooh) = ebu_in(i,1,j,p_ebu_ch3cooh)*1.3ebu(i,kts,j,p_ebu_cres) = ebu_in(i,1,j,p_ebu_in_cres)*1.3ebu(i,kts,j,p_ebu_glyald) = ebu_in(i,1,j,p_ebu_in_glyald)*1.3ebu(i,kts,j,p_ebu_isop) = ebu_in(i,1,j,p_ebu_in_isop)*1.3ebu(i,kts,j,p_ebu_macr) = ebu_in(i,1,j,p_ebu_in_macr)*1.3ebu(i,kts,j,p_ebu_mek) = ebu_in(i,1,j,p_ebu_in_mek)*1.3ebu(i,kts,j,p_ebu_mvk) = ebu_in(i,1,j,p_ebu_in_mvk)*1.3ebu(i,kts,j,p_ebu_oc) = ebu_in(i,1,j,p_ebu_in_oc)*1.3ebu(i,kts,j,p_ebu_bc) = ebu_in(i,1,j,p_ebu_in_bc)*1.3ebu(i,kts,j,p_ebu_acetol) = ebu_in(i,1,j,p_ebu_in_acetol)*1.3ebu(i,kts,j,p_ebu_pm25) = ebu_in(i,1,j,p_ebu_in_pm25)*1.3ebu(i,kts,j,p_ebu_pm10) = ebu_in(i,1,j,p_ebu_in_pm10)*1.3

enddoenddo

endif

B Biomass Burning Emissions InputAs mentioned on the section 2 (Getting WRF-Chem ready), we used two different biomass burning

emission inventories, which are: Fire INventory from NCAR (FINN) and Global Fire Emissions Database(GFED). Before we can employ these biomass burning emission inventories in WRF-Chem model, we needto prepare these emission. In this appendix, we explained step-by-step of biomass burning emission prepara-tion.

B.1 FINN Biomass Burning Emissions PreparationIn this section, the biomass burning emissions input in WRF-Chem from FINN biomass burning inventory

is explained.

1. download the processor named fire_emis at http://bai.acom.ucar.edu/Data/fire/.

39

Page 47: Simulating Air Pollution in the Severe Fires Event during

2. The server only provides a year data set contains the daily emissions from 28 species gas-phase andtwo aerosol chemical species, fire size and fire faction from four different sources (extra tropical for-est,tropical forest, grassland and savanna).

3. we need information about the grid resolution, simulation’s domain, and the matching time in the filewith in the model run. We used wrf_inputd<0X>.

4. map each chemical species from MOZART database to match with the exact variable for WRF-Chemas fire_emis.mozc.namelist.input as follows:

&controldomains = 1,fire_directory = ’/home/mmolen/data/fire_emis/data_files/’,fire_filename = ’GLOBAL_FINNv15_2015_MOZ4_5252016.txt’,wrf_directory = ’/home/mmolen/data/WRFdata/’,start_date = ’2015-10-01’,end_date = ’2015-10-31’,

wrf2fire_map = ’co -> CO’, ’no -> NO’, ’so2 -> SO2’, ’bigalk -> BIGALK’,’bigene -> BIGENE’, ’c2h4 -> C2H4’, ’c2h5oh -> C2H5OH’,’c2h6 -> C2H6’, ’c3h8 -> C3H8’, ’c3h6 -> C3H6’, ’ch2o -> CH2O’,’ch3cho -> CH3CHO’, ’ch3coch3 -> CH3COCH3’, ’ch3oh -> CH3OH’,’mek -> MEK’, ’toluene -> TOLUENE’, ’nh3 -> NH3’, ’no2 -> NO2’,’open -> BIGALD’, ’c10h16 -> C10H16’, ’ch3cooh -> CH3COOH’,’cres -> CRESOL’, ’glyald -> GLYALD’,’mgly -> CH3COCHO’, ’gly -> CH3COCHO’,’acetol -> HYAC’, ’isop -> ISOP’,’macr -> MACR’, ’mvk -> MVK’,’oc -> OC;aerosol’, ’pm10 -> PM10;aerosol’, ’pm25 -> PM25;aerosol’,’bc -> BC;aerosol’

/

B.2 GFED Biomass Burning Emissions PreparationGFED biomass burning inventory available. We skipped cresol, glyald, mgly, gly, acetol, isop, pm10,

ch3cho and ch3coch3 because the emission factor are not defined in Akagi et al. (2011). The GFED biomassburning preparation is following step-by-step as below:

1. Download a one year GFED biomass burning inventory package at: http://www.falw.vu/∼gwerf/GFED/GFED4/.

2. In this package, the biomass burning emission is represented by the dry matter emissions. We there-fore need to calculate each chemical species emissions according to emission factor, vegetation typecontribution, and daily fraction. (See section 2 for detail).

3. Re-grid the GFED biomass burning emission dataset to match the WRF-Chem model horizontal reso-lution

4. Save each biomass burning emission files as wrffirechemi_d<nn>_<yyyy-mm-dd_hh:mm:ss> with nnis the domain used, and the rest is date of the file.

40

Page 48: Simulating Air Pollution in the Severe Fires Event during

"""emissions unit from GFED --> kgDM m-2 month-1emissions factor unit --> gr<species> kg-1DM

to convert: DM emissions * emissions factorunit: gr<species> m-2 month-1

unit for WRFChem: microgram <species> m-2 s-1 (for aerosols)mole <species> km-2 h-1 (for gas)

1 gr(CO) m-2 month-1 = CO (gr)* mrCO (mol/gr) /(1 m2 / 1000000 km2/m2)/(1month* 31 days/month)

aerosol --> PM25 gr m-2 month-1 = 1000000 (ug/gr) /(1 (m2/m2) *31*24*60*60 (second/month))

"""

import numpy as npimport h5pyimport matplotlib.pyplot as pltimport netCDF4 as ncfrom netCDF4 import Datasetfrom mpl_toolkits import basemap

sources = ’SAVA’,’BORF’,’TEMF’,’DEFO’,’PEAT’,’AGRI’

emiss_factor = ’GFED4_Emission_Factors.txt’geodata = ’geo_em.d01.nc’GFED_data = ’GFED4.1s_2015.hdf5’mf = nc.Dataset(geodata)

"""#Read in emission factors"""

EFs = np.zeros((41, 6)) # 41 species, 6 sources

k = 0f = open(emiss_factor)while 1:

line = f.readline()if line == "":

break

if line[0] != ’#’:contents = line.split()species.append(contents[0])EFs[k,:] = contents[1:]k += 1

41

Page 49: Simulating Air Pollution in the Severe Fires Event during

f.close()

EF_CO = EFs[3,:]EF_NOx = EFs[7,:]EF_PM25 = EFs[9,:]EF_OC = EFs[12,:]EF_BC = EFs[13,:]EF_SO2 = EFs[14,:]EF_C2H6 = EFs[15,:]EF_CH3OH = EFs[16,:]EF_C2H5OH = EFs[17,:]EF_C3H8 = EFs[18,:]EF_C2H4 = EFs[20,:]EF_C3H6 = EFs[21,:]EF_C10H16 = EFs[23,:]EF_toluene = EFs[27,:]EF_bigene = EFs[28,:]EF_bigalk = EFs[29,:]EF_CH2O = EFs[30,:]EF_NH3 = EFs[33,:]EF_CH3COOH = EFs[37,:]EF_MEK = EFs[38,:]

"""#read in GFED data"""

g = h5py.File(GFED_data, ’r’)xlat = g[’lat’][:]xlon = g[’lon’][:]

##CO emission globalCO_emissions = np.zeros((720, 1440))NOx_emissions = np.zeros((720, 1440))PM25_emissions = np.zeros((720, 1440))OC_emissions = np.zeros((720, 1440))BC_emissions = np.zeros((720, 1440))SO2_emissions = np.zeros((720, 1440))C2H6_emissions = np.zeros((720, 1440))CH3OH_emissions = np.zeros((720, 1440))C2H5OH_emissions = np.zeros((720, 1440))C3H8_emissions = np.zeros((720, 1440))C2H4_emissions = np.zeros((720, 1440))C3H6_emissions = np.zeros((720, 1440))C10H16_emissions = np.zeros((720, 1440))toluene_emissions = np.zeros((720, 1440))bigene_emissions = np.zeros((720, 1440))bigalk_emissions = np.zeros((720, 1440))CH2O_emissions = np.zeros((720, 1440))NH3_emissions = np.zeros((720, 1440))CH3COOH_emissions = np.zeros((720, 1440))

42

Page 50: Simulating Air Pollution in the Severe Fires Event during

MEK_emissions = np.zeros((720, 1440))

string = ’/emissions/10/DM’ # read in DM emissions for October onlyDM_emissions = g[string][:]

# read in the day fraction

d = ’/emissions/10/daily_fraction/day_15’day_frac = g[d][:]

for source in range(6):# read in the fractional contribution of each sourcestring = ’/emissions/10/partitioning/DM_’+sources[source]contribution = g[string][:]for k in range(len(specs)):

if specs[k]==’CO’:CO_emissions += DM_emissions * day_frac * contribution *

EF_CO[source]/28.01*10e6/(31*24) #mole CO per km2 per hourselif specs[k]==’NOx’:

NOx_emissions += DM_emissions* day_frac*contribution*EF_NOx[source]/30.01*10e6/(31*24) #NOx is represented as NO (moleNO per km2 per hours)

elif specs[k]==’PM25’:PM25_emissions += DM_emissions* day_frac*

contribution*EF_PM25[source]*1000000/(31*24*60*60) #microgr PM25 per m2 persecond

elif specs[k]==’OC’:OC_emissions += DM_emissions* day_frac*

contribution*EF_OC[source]*1000000/(31*24*60*60) #microgr OC per m2 per secondelif specs[k]==’BC’:

BC_emissions += DM_emissions* day_frac*contribution*EF_BC[source]*1000000/(31*24*60*60) #microgr BC per m2 per second

elif specs[k]==’SO2’:SO2_emissions += DM_emissions* day_frac*

contribution*EF_SO2[source]/64.02*10e6/(31*24) #mole SO2 per km2 per hourselif specs[k]==’C2H6’:

C2H6_emissions += DM_emissions* day_frac*contribution*EF_C2H6[source]/30.07*10e6/(31*24) #mole C2H6 per km2 per hours

elif specs[k]==’CH3OH’:CH3OH_emissions += DM_emissions* day_frac*

contribution*EF_CH3OH[source]/32.04*10e6/(31*24) #mole CH3OH per km2 per hourselif specs[k]==’C2H5OH’:

C2H5OH_emissions += DM_emissions* day_frac*contribution*EF_C2H5OH[source]*46.07/10e6/(31*24) #mole C2H5OH per km2 perhours

elif specs[k]==’C3H8’:C3H8_emissions += DM_emissions* day_frac*

contribution*EF_C3H8[source]/44.1*10e6/(31*24) #mole C3H8 per km2 per hourselif specs[k]==’C2H4’:

C2H4_emissions += DM_emissions* day_frac*

43

Page 51: Simulating Air Pollution in the Severe Fires Event during

contribution*EF_C2H4[source]/28.05*10e6/(31*24) #mole C2H4 per km2 per hourselif specs[k]==’C3H6’:

C3H6_emissions += DM_emissions* day_frac*contribution*EF_C3H6[source]/42.08*10e6/(31*24) #mole C3H6 per km2 per hours

elif specs[k]==’C10H16’:C10H16_emissions += DM_emissions* day_frac*

contribution*EF_C10H16[source]/136.24*10e6/(31*24) #mole C10H16 per km2 perhours

elif specs[k]==’toluene’:toluene_emissions += DM_emissions* day_frac*

contribution*EF_toluene[source]/92.14*10e6/(31*24) #toluene is C7H8(mole perkm2 per hours)

elif specs[k]==’bigene’:bigene_emissions += DM_emissions* day_frac*

contribution*EF_bigene[source]/56*10e6/(31*24) #mole butene per km2 per hourselif specs[k]==’bigalk’:

bigalk_emissions += DM_emissions* day_frac*contribution*EF_bigalk[source]/58*10e6/(31*24) #mole butane per km2 per hours

elif specs[k]==’CH2O’:CH2O_emissions += DM_emissions* day_frac*

contribution*EF_CH2O[source]/30.03*10e6/(31*24) #CH2O is formaldehydeelif specs[k]==’NH3’:

NH3_emissions += DM_emissions* day_frac*contribution*EF_NH3[source]/17.03*10e6/(31*24) #mole NH3 per km2 per hours

elif specs[k]==’CH3COOH’:CH3COOH_emissions += DM_emissions* day_frac*

contribution*EF_CH3COOH[source]/60.05*10e6/(31*24) #mole CH3COOH per km2 perhours

else:MEK_emissions += DM_emissions* day_frac*

contribution*EF_MEK[source]/72.11*10e6/(31*24) #mole MEK per km2 per hours

"""regrid"""#global arrayxlat_globe = np.flipud(xlat[:,0]) #because the lat is not exceeding, it needs to be flippedxlon_globe = xlon[0,:]

##define indonesia arraylat_indo = mf.variables[’XLAT_M’][0,:,:]lon_indo = mf.variables[’XLONG_M’][0,:]

CO_emissions_indonesia = basemap.interp(np.flipud(CO_emissions), xlon_globe, xlat_globe,lon_indo, lat_indo, order=1).reshape((1,1,74,169))

NOx_emissions_indonesia = basemap.interp(np.flipud(NOx_emissions), xlon_globe,xlat_globe, lon_indo, lat_indo, order=1).reshape((1,1,74,169))

PM25_emissions_indonesia = basemap.interp(np.flipud(PM25_emissions), xlon_globe,xlat_globe, lon_indo, lat_indo, order=1).reshape((1,1,74,169))

OC_emissions_indonesia = basemap.interp(np.flipud(OC_emissions), xlon_globe, xlat_globe,lon_indo, lat_indo, order=1).reshape((1,1,74,169))

BC_emissions_indonesia = basemap.interp(np.flipud(BC_emissions), xlon_globe, xlat_globe,

44

Page 52: Simulating Air Pollution in the Severe Fires Event during

lon_indo, lat_indo, order=1).reshape((1,1,74,169))SO2_emissions_indonesia = basemap.interp(np.flipud(SO2_emissions), xlon_globe, xlat_globe,

lon_indo, lat_indo, order=1).reshape((1,1,74,169))C2H6_emissions_indonesia = basemap.interp(np.flipud(C2H6_emissions), xlon_globe,

xlat_globe, lon_indo, lat_indo,order=1).reshape((1,1,74,169))CH3OH_emissions_indonesia = basemap.interp(np.flipud(CH3OH_emissions), xlon_globe,

xlat_globe, lon_indo, lat_indo,order=1).reshape((1,1,74,169))C2H5OH_emissions_indonesia = basemap.interp(np.flipud(C2H5OH_emissions),

xlon_globe,xlat_globe, lon_indo, lat_indo, order=1).reshape((1,1,74,169))C3H8_emissions_indonesia = basemap.interp(np.flipud(C3H8_emissions), xlon_globe,

xlat_globe, lon_indo, lat_indo, order=1).reshape((1,1,74,169))C2H4_emissions_indonesia = basemap.interp(np.flipud(C2H4_emissions), xlon_globe,

xlat_globe, lon_indo, lat_indo, order=1).reshape((1,1,74,169))C3H6_emissions_indonesia = basemap.interp(np.flipud(C3H6_emissions), xlon_globe,

xlat_globe, lon_indo, lat_indo, order=1).reshape((1,1,74,169))C10H16_emissions_indonesia = basemap.interp(np.flipud(C10H16_emissions), xlon_globe,

xlat_globe, lon_indo, lat_indo, order=1).reshape((1,1,74,169))toluene_emissions_indonesia = basemap.interp(np.flipud(toluene_emissions), xlon_globe,

xlat_globe, lon_indo, lat_indo, order=1).reshape((1,1,74,169))bigene_emissions_indonesia = basemap.interp(np.flipud(bigene_emissions), xlon_globe,

xlat_globe, lon_indo, lat_indo, order=1).reshape((1,1,74,169))bigalk_emissions_indonesia = basemap.interp(np.flipud(bigalk_emissions), xlon_globe,

xlat_globe,lon_indo,lat_indo,order=1).reshape((1,1,74,169))CH2O_emissions_indonesia = basemap.interp(np.flipud(CH2O_emissions), xlon_globe,

xlat_globe, lon_indo, lat_indo,order=1).reshape((1,1,74,169))NH3_emissions_indonesia = basemap.interp(np.flipud(NH3_emissions), xlon_globe,

xlat_globe, lon_indo, lat_indo, order=1).reshape((1,1,74,169))CH3COOH_emissions_indonesia = basemap.interp(np.flipud(CH3COOH_emissions), xlon_globe,

xlat_globe, lon_indo, lat_indo, order=1).reshape((1,1,74,169))MEK_emissions_indonesia = basemap.interp(np.flipud(MEK_emissions), xlon_globe, xlat_globe,

lon_indo, lat_indo, order=1).reshape((1,1,74,169))

"""write a netCDF file"""

FINN = Dataset(’wrffirechemi_d01_2015-10-29_23:00:00.nc’,’r’)#we need FINN data to have fire size and faction of land type. These data will be used on

plume rise module

GFED = Dataset(’wrffirechemi_d01_2015-10-29_23:00:00’,’w’,format=’NETCDF3_CLASSIC’)#WRF-Chem version 3.2 only allows netcdf version 3 file format with classic type.

##global attributesGFED.Title = "GFED emissions for WRF"GFED.History = "Created on 2016-12-25"GFED.Author = "Dyah Ayu Putriningrum"GFED.START_DATE = "2015-10-01_00:00:00"GFED.SIMULATION_START_DATE = "2015-10-01_00:00:00"GFED.WEST_EAST_GRID_DIMENSION = int(170)GFED.SOUTH_NORTH_GRID_DIMENSION = int(75)

45

Page 53: Simulating Air Pollution in the Severe Fires Event during

GFED.BOTTOM_TOP_GRID_DIMENSION = int(28)GFED.DX = float(30000.0)GFED.DY = float(30000.0)GFED.GRIDTYPE = "C"GFED.DIFF_OPT = int(1)GFED.KM_OPT = int(4)GFED.DAMP_OPT = int(3)GFED.DAMPCOEF = float(0.2)GFED.KHDIF = float(0.0)GFED.KVDIF = float(0.0)GFED.MP_PHYSICS = int(2)GFED.RA_LW_PHYSICS = int(4)GFED.RA_SW_PHYSICS = int(4)GFED.SF_SFCLAY_PHYSICS = int(1)GFED.SF_SURFACE_PHYSICS = int(2)GFED.BL_PBL_PHYSICS = int(1)GFED.CU_PHYSICS = int(3)GFED.SURFACE_INPUT_SOURCE = int(1)GFED.SST_UPDATE = int(0)GFED.GRID_FDDA = int(0)GFED.GFDDA_INTERVAL_M = int(0)GFED.GFDDA_END_H = int(0)GFED.GRID_SFDDA = int(0)GFED.SGFDDA_INTERVAL_M = int(0)GFED.SGFDDA_END_H = int(0)GFED.WEST_EAST_PATCH_START_UNSTAG = int(1)GFED.WEST_EAST_PATCH_END_UNSTAG = int(169)GFED.WEST_EAST_PATCH_START_STAG = int(1)GFED.WEST_EAST_PATCH_END_STAG = int(170)GFED.SOUTH_NORTH_PATCH_START_UNSTAG = int(1)GFED.SOUTH_NORTH_PATCH_END_UNSTAG = int(74)GFED.SOUTH_NORTH_PATCH_START_STAG = int(1)GFED.SOUTH_NORTH_PATCH_END_STAG = int(75)GFED.BOTTOM_TOP_PATCH_START_UNSTAG = int(1)GFED.BOTTOM_TOP_PATCH_END_UNSTAG = int(27)GFED.BOTTOM_TOP_PATCH_START_STAG = int(1)GFED.BOTTOM_TOP_PATCH_END_STAG = int(28)GFED.GRID_ID = int(1)GFED.PARENT_ID = int(0)GFED.I_PARENT_START = int(1)GFED.J_PARENT_START = int(1)GFED.PARENT_GRID_RATIO = int(1)GFED.DT = float(60.0)GFED.CEN_LAT = FINN.getncattr(’CEN_LAT’)GFED.CEN_LON = FINN.getncattr(’CEN_LON’)GFED.TRUELAT1 = FINN.getncattr(’TRUELAT1’)GFED.TRUELAT2 = FINN.getncattr(’TRUELAT2’)GFED.MOAD_CEN_LAT = FINN.getncattr(’MOAD_CEN_LAT’)GFED.STAND_LON = FINN.getncattr(’STAND_LON’)GFED.POLE_LAT = FINN.getncattr(’POLE_LAT’)GFED.POLE_LON = FINN.getncattr(’POLE_LAT’)GFED.GMT = FINN.getncattr(’GMT’)

46

Page 54: Simulating Air Pollution in the Severe Fires Event during

GFED.JULYR = int(2015)GFED.JULDAY = int(274)GFED.MAP_PROJ = int(3)GFED.MMINLU = "USGS"GFED.NUM_LAND_CAT = int(24)GFED.ISWATER = int(16)GFED.ISLAKE = int(-1)GFED.ISICE = int(24)GFED.ISURBAN = int(1)GFED.ISOILWATER = int(14)

GFED.createDimension(’west_east’,169)GFED.createDimension(’south_north’,74)GFED.createDimension(’DateStrLen’,19)GFED.createDimension(’emissions_zdim_stag’,1)GFED.createDimension(’Time’, 1)

Times = GFED.createVariable(’Times’,’S1’,(’Time’,’DateStrLen’))Times.units = "secs since 1970-01-01 00:00:00"Times.long_name = "synthesized time coordinate from Times(time)"Times._CoordinateAxisType = "Times"

Times [:] = [’2’,’0’,’1’,’5’,’-’,’1’,’0’,’-’,’2’,’9’,’_’,’2’,’3’,’:’,’0’,’0’,’:’,’0’,’0’]

co_emiss = GFED.createVariable(’ebu_in_co’, ’f4’, (’Time’,’emissions_zdim_stag’, ’south_north’, ’west_east’))

co_emiss.MemoryOrder = ’XYZ’co_emiss.description = ’Carbon Monoxide Emissions’co_emiss.units = ’mole km-2 hr-1’co_emiss.stagger = ’Z’co_emiss.FieldType = 104

#NOx is defined as NO since NO and NO2 are rapidly interconverted in the atmosphere (Akagiet al, 2011)

no_emiss = GFED.createVariable(’ebu_in_no’, ’f4’, (’Time’,’emissions_zdim_stag’, ’south_north’, ’west_east’))

no_emiss.MemoryOrder = ’XYZ’no_emiss.description = ’Nitrogen Monoxide Emissions’no_emiss.units = ’mole km-2 hr-1’no_emiss.stagger = ’Z’no_emiss.FieldType = 104

so2_emiss = GFED.createVariable(’ebu_in_so2’, ’f4’,(’Time’,’emissions_zdim_stag’, ’south_north’, ’west_east’))

so2_emiss.MemoryOrder = ’XYZ’so2_emiss.description = ’Sulphur Dioxide Emissions’so2_emiss.units = ’mole km-2 hr-1’so2_emiss.stagger = ’Z’so2_emiss.FieldType = 104

47

Page 55: Simulating Air Pollution in the Severe Fires Event during

bigalk_emiss = GFED.createVariable(’ebu_in_bigalk’,’f4’, (’Time’,’emissions_zdim_stag’, ’south_north’, ’west_east’))

bigalk_emiss.MemoryOrder = ’XYZ’bigalk_emiss.description = ’Alkanes with >4 carbon Emissions’bigalk_emiss.units = ’mole km-2 hr-1’bigalk_emiss.stagger = ’Z’bigalk_emiss.FieldType = 104

bigene_emiss = GFED.createVariable(’ebu_in_bigene’, ’f4’,(’Time’,’emissions_zdim_stag’, ’south_north’, ’west_east’))

bigene_emiss.MemoryOrder = ’XYZ’bigene_emiss.description = ’Alkenes with >4 carbon Emissions’bigene_emiss.units = ’mole km-2 hr-1’bigene_emiss.stagger = ’Z’bigene_emiss.FieldType = 104

c2h4_emiss = GFED.createVariable(’ebu_in_c2h4’, ’f4’, (’Time’,’emissions_zdim_stag’, ’south_north’, ’west_east’))

c2h4_emiss.MemoryOrder = ’XYZ’c2h4_emiss.description = ’Ethylen Emissions’c2h4_emiss.units = ’mole km-2 hr-1’c2h4_emiss.stagger = ’Z’c2h4_emiss.FieldType = 104

c2h5oh_emiss = GFED.createVariable(’ebu_in_c2h5oh’, ’f4’, (’Time’,’emissions_zdim_stag’, ’south_north’, ’west_east’))

c2h5oh_emiss.MemoryOrder = ’XYZ’c2h5oh_emiss.description = ’Ethanol Emissions’c2h5oh_emiss.units = ’mole km-2 hr-1’c2h5oh_emiss.stagger = ’Z’c2h5oh_emiss.FieldType = 104

c2h6_emiss = GFED.createVariable(’ebu_in_c2h6’, ’f4’, (’Time’,’emissions_zdim_stag’, ’south_north’, ’west_east’))

c2h6_emiss.MemoryOrder = ’XYZ’c2h6_emiss.description = ’Ethane Emissions’c2h6_emiss.units = ’mole km-2 hr-1’c2h6_emiss.stagger = ’Z’c2h6_emiss.FieldType = 104

c3h8_emiss = GFED.createVariable(’ebu_in_c3h8’, ’f4’, (’Time’,’emissions_zdim_stag’, ’south_north’, ’west_east’))

c3h8_emiss.MemoryOrder = ’XYZ’c3h8_emiss.description = ’Propane Emissions’c3h8_emiss.units = ’mole km-2 hr-1’c3h8_emiss.stagger = ’Z’c3h8_emiss.FieldType = 104

c3h6_emiss = GFED.createVariable(’ebu_in_c3h6’, ’f4’, (’Time’,’emissions_zdim_stag’, ’south_north’, ’west_east’))

c3h6_emiss.MemoryOrder = ’XYZ’c3h6_emiss.description = ’Propene Emissions’

48

Page 56: Simulating Air Pollution in the Severe Fires Event during

c3h6_emiss.units = ’mole km-2 hr-1’c3h6_emiss.stagger = ’Z’c3h6_emiss.FieldType = 104

ch2o_emiss = GFED.createVariable(’ebu_in_ch2o’, ’f4’, (’Time’,’emissions_zdim_stag’, ’south_north’, ’west_east’))

ch2o_emiss.MemoryOrder = ’XYZ’ch2o_emiss.description = ’Formaldehyde Emissions’ch2o_emiss.units = ’mole km-2 hr-1’ch2o_emiss.stagger = ’Z’ch2o_emiss.FieldType = 104

ch3oh_emiss = GFED.createVariable(’ebu_in_ch3oh’, ’f4’, (’Time’,’emissions_zdim_stag’, ’south_north’, ’west_east’))

ch3oh_emiss.MemoryOrder = ’XYZ’ch3oh_emiss.description = ’Methanol Emissions’ch3oh_emiss.units = ’mole km-2 hr-1’ch3oh_emiss.stagger = ’Z’ch3oh_emiss.FieldType = 104

mek_emiss = GFED.createVariable(’ebu_in_mek’, ’f4’, (’Time’,’emissions_zdim_stag’, ’south_north’, ’west_east’))

mek_emiss.MemoryOrder = ’XYZ’mek_emiss.description = ’Methyl Ethyl Keton Emissions’mek_emiss.units = ’mole km-2 hr-1’mek_emiss.stagger = ’Z’mek_emiss.FieldType = 104

toluene_emiss = GFED.createVariable(’ebu_in_toluene’, ’f4’, (’Time’,’emissions_zdim_stag’, ’south_north’, ’west_east’))

toluene_emiss.MemoryOrder = ’XYZ’toluene_emiss.description = ’Lumped Aromatics Emissions’toluene_emiss.units = ’mole km-2 hr-1’toluene_emiss.stagger = ’Z’toluene_emiss.FieldType = 104

nh3_emiss = GFED.createVariable(’ebu_in_nh3’, ’f4’, (’Time’,’emissions_zdim_stag’, ’south_north’, ’west_east’))

nh3_emiss.MemoryOrder = ’XYZ’nh3_emiss.description = ’Ammonia Emissions’nh3_emiss.units = ’mole km-2 hr-1’nh3_emiss.stagger = ’Z’nh3_emiss.FieldType = 104

c10h16_emiss = GFED.createVariable(’ebu_in_c10h16’, ’f4’, (’Time’,’emissions_zdim_stag’, ’south_north’, ’west_east’))

c10h16_emiss.MemoryOrder = ’XYZ’c10h16_emiss.description = ’Terpenes Emissions’c10h16_emiss.units = ’mole km-2 hr-1’c10h16_emiss.stagger = ’Z’c10h16_emiss.FieldType = 104

49

Page 57: Simulating Air Pollution in the Severe Fires Event during

ch3cooh_emiss = GFED.createVariable(’ebu_in_ch3cooh’, ’f4’, (’Time’,’emissions_zdim_stag’, ’south_north’, ’west_east’))

ch3cooh_emiss.MemoryOrder = ’XYZ’ch3cooh_emiss.description = ’Acetic Acid Emissions’ch3cooh_emiss.units = ’mole km-2 hr-1’ch3cooh_emiss.stagger = ’Z’ch3cooh_emiss.FieldType = 104

oc_emiss = GFED.createVariable(’ebu_in_oc’, ’f4’, (’Time’,’emissions_zdim_stag’, ’south_north’, ’west_east’))

oc_emiss.MemoryOrder = ’XYZ’oc_emiss.description = ’Organic Carbon Emissions’oc_emiss.units = ’ug m-2 s-1’oc_emiss.stagger = ’Z’oc_emiss.FieldType = 104

pm25_emiss = GFED.createVariable(’ebu_in_pm25’, ’f4’, (’Time’,’emissions_zdim_stag’, ’south_north’, ’west_east’))

pm25_emiss.MemoryOrder = ’XYZ’pm25_emiss.description = ’PM25 Emissions’pm25_emiss.units = ’ug m-2 s-1’pm25_emiss.stagger = ’Z’pm25_emiss.FieldType = 104

bc_emiss = GFED.createVariable(’ebu_in_bc’, ’f4’, (’Time’,’emissions_zdim_stag’, ’south_north’, ’west_east’))

bc_emiss.MemoryOrder = ’XYZ’bc_emiss.description = ’Black Carbon Emissions’bc_emiss.units = ’ug m-2 s-1’bc_emiss.stagger = ’Z’bc_emiss.FieldType = 104

FIRESIZE_AGEF = GFED.createVariable(’FIRESIZE_AGEF’, ’f4’, (’Time’,’south_north’, ’west_east’))

FIRESIZE_AGEF.MemoryOrder = ’XY’FIRESIZE_AGEF.description = ’mean firesize of extra tropical forest from FINN’FIRESIZE_AGEF.units = ’m2’FIRESIZE_AGEF.stagger = ’ ’FIRESIZE_AGEF.FieldType = 104

FIRESIZE_AGGR = GFED.createVariable(’FIRESIZE_AGGR’, ’f4’, (’Time’,’south_north’, ’west_east’))

FIRESIZE_AGGR.MemoryOrder = ’XY’FIRESIZE_AGGR.description = ’mean firesize of grassland from FINN’FIRESIZE_AGGR.units = ’m2’FIRESIZE_AGGR.stagger = ’ ’FIRESIZE_AGGR.FieldType = 104

FIRESIZE_AGSV = GFED.createVariable(’FIRESIZE_AGSV’, ’f4’, (’Time’,’south_north’, ’west_east’))

FIRESIZE_AGSV.MemoryOrder = ’XY’FIRESIZE_AGSV.description = ’mean firesize of savana from FINN’

50

Page 58: Simulating Air Pollution in the Severe Fires Event during

FIRESIZE_AGSV.units = ’m2’FIRESIZE_AGSV.stagger = ’ ’FIRESIZE_AGSV.FieldType = 104

FIRESIZE_AGTF = GFED.createVariable(’FIRESIZE_AGTF’, ’f4’, (’Time’,’south_north’, ’west_east’))

FIRESIZE_AGTF.MemoryOrder = ’XY’FIRESIZE_AGTF.description = ’mean firesize of tropical forest from FINN’FIRESIZE_AGTF.units = ’m2’FIRESIZE_AGTF.stagger = ’ ’FIRESIZE_AGTF.FieldType = 104

MEAN_FCT_AGEF = GFED.createVariable(’MEAN_FCT_AGEF’, ’f4’, (’Time’,’south_north’, ’west_east’))

MEAN_FCT_AGEF.MemoryOrder = ’XY’MEAN_FCT_AGEF.description = ’mean fraction of extra tropical forest from FINN’MEAN_FCT_AGEF.units = ’m2’MEAN_FCT_AGEF.stagger = ’ ’MEAN_FCT_AGEF.FieldType = 104

MEAN_FCT_AGGR = GFED.createVariable(’MEAN_FCT_AGGR’, ’f4’, (’Time’,’south_north’, ’west_east’))

MEAN_FCT_AGGR.MemoryOrder = ’XY’MEAN_FCT_AGGR.description = ’mean fraction of grassland from FINN’MEAN_FCT_AGGR.units = ’m2’MEAN_FCT_AGGR.stagger = ’ ’MEAN_FCT_AGGR.FieldType = 104

MEAN_FCT_AGSV = GFED.createVariable(’MEAN_FCT_AGSV’,’ f4’, (’Time’,’south_north’, ’west_east’))

MEAN_FCT_AGSV.MemoryOrder = ’XY’MEAN_FCT_AGSV.description = ’mean fraction of savana from FINN’MEAN_FCT_AGSV.units = ’m2’MEAN_FCT_AGSV.stagger = ’ ’MEAN_FCT_AGSV.FieldType = 104

MEAN_FCT_AGTF = GFED.createVariable(’MEAN_FCT_AGTF’, ’f4’, (’Time’,’south_north’, ’west_east’))

MEAN_FCT_AGTF.MemoryOrder = ’XY’MEAN_FCT_AGTF.description = ’mean fraction of tropical forest from FINN’MEAN_FCT_AGTF.units = ’m2’MEAN_FCT_AGTF.stagger = ’ ’MEAN_FCT_AGTF.FieldType = 104

co_emiss[:,:,:,:] = CO_emissions_indonesiano_emiss[:,:,:,:] = NOx_emissions_indonesiapm25_emiss[:,:,:,:] = PM25_emissions_indonesiaoc_emiss[:,:,:,:] = OC_emissions_indonesiabc_emiss[:,:,:,:] = BC_emissions_indonesiaso2_emiss[:,:,:,:] = SO2_emissions_indonesia

51

Page 59: Simulating Air Pollution in the Severe Fires Event during

c2h6_emiss[:,:,:,:] = C2H6_emissions_indonesiach3oh_emiss[:,:,:,:] = CH3OH_emissions_indonesiac2h5oh_emiss[:,:,:,:] = C2H5OH_emissions_indonesiac3h8_emiss[:,:,:,:] = C3H8_emissions_indonesiac2h4_emiss[:,:,:,:] = C2H4_emissions_indonesiac3h6_emiss[:,:,:,:] = C3H6_emissions_indonesiac10h16_emiss[:,:,:,:] = C10H16_emissions_indonesiatoluene_emiss[:,:,:,:] = toluene_emissions_indonesiabigene_emiss[:,:,:,:] = bigene_emissions_indonesiabigalk_emiss[:,:,:,:] = bigalk_emissions_indonesiach2o_emiss[:,:,:,:] = CH2O_emissions_indonesianh3_emiss[:,:,:,:] = NH3_emissions_indonesiach3cooh_emiss[:,:,:,:] = CH3COOH_emissions_indonesiamek_emiss[:,:,:,:] = MEK_emissions_indonesiaFIRESIZE_AGEF[:,:,:] = FINN.variables[’FIRESIZE_AGEF’]FIRESIZE_AGGR[:,:,:] = FINN.variables[’FIRESIZE_AGGR’]FIRESIZE_AGSV[:,:,:] = FINN.variables[’FIRESIZE_AGSV’]FIRESIZE_AGTF[:,:,:] = FINN.variables[’FIRESIZE_AGTF’]MEAN_FCT_AGEF[:,:,:] = FINN.variables[’MEAN_FCT_AGEF’]MEAN_FCT_AGGR[:,:,:] = FINN.variables[’MEAN_FCT_AGGR’]MEAN_FCT_AGSV[:,:,:] = FINN.variables[’MEAN_FCT_AGSV’]MEAN_FCT_AGTF[:,:,:] = FINN.variables[’MEAN_FCT_AGTF’]

GFED.close()

52

Page 60: Simulating Air Pollution in the Severe Fires Event during

Species SAVA BORF TEMF DEFO PEAT AGRIDM 1000 1000 1000 1000 1000 1000CO 63 127 88 93 210 102NOx 3.9 0.9 1.92 2.55 1 3.11

PM25 7.17 15.3 12.9 9.1 9.1 6.26OC 2.62 9.6 9.6 4.71 6.02 2.3BC 0.37 0.5 0.5 0.52 0.04 0.75SO2 0.48 1.1 1.1 0.4 0.4 0.4

C2H6 0.66 1.79 0.63 0.71 0.71 0.91CH3OH 1.18 2.82 1.74 2.43 8.46 3.29

C2H5OH 0.024 0.055 0.1 0.037 0.037 0.035C3H8 0.1 0.44 0.22 0.126 0.126 0.28C2H4 0.82 1.42 1.17 1.06 2.57 1.46C3H6 0.79 1.13 0.61 0.64 3.05 0.68

C10H16 0.081 2.003 2.003 0.15 0.15 0.005toluene 0.270 1.626 0.540 0.697 4.360 0.415bigene 0.133 0.385 0.369 0.267 0.267 0.333bigalk 0.055 0.349 0.225 0.072 0.072 0.340CH2O 0.73 1.86 2.09 1.73 1.4 2.08

NH32H6 0.52 2.72 0.84 1.33 1.33 2.17CH3COOH 3.55 4.41 2.13 3.05 8.97 5.59

MEK 0.181 0.22 0.13 0.5 0.5 0.9

Table 9: Emission Factors used for GFED input (Akagi et al., 2011)

C Formula for Model EvaluationBelow is the Normalized Mean Bias Factor (NMBF) equation:if M ≥ O

NMBF =∑

Ni=1 Mi

∑Ni=1 Oi

−1 (6)

if M < O

NMBF = 1− ∑Ni=1 Mi

∑Ni=1 Oi

(7)

A positive NMBF indicates the model results overestimate the observations by a factor of 1+NMBF. A nega-tive NMBF indicates the model results underestimate the observations by a factor of 1-NMBF. The coefficientof determination r2 is the proportion of variability in a data set that is accounted for by a statistical model.

r2 = 1− ∑(yi − y)2

∑(yi − yi)2 (8)

The correlation coefficient measures the strength and direction of a linear relationship between modeledvariable to the observed variable on a scatter plot. r is calculated as follow:

r =∑

Ni=0(O−O)(Mi −M)√

∑Ni=0(O−O)2 ∑

Ni=0(Mi −M)2

(9)

53

Page 61: Simulating Air Pollution in the Severe Fires Event during

D Vertical Profile Comparison of Temperature and Relative humidity

54

Page 62: Simulating Air Pollution in the Severe Fires Event during

E HYSPLIT Back Trajectory ModelThe HYSPLIT model is a system to compute simple air mass trajectory. In this study, we used the

back trajectory analysis to determine the origin of air masses in three locations: Kototabang, Pekanbaru, andSingapore.

Figure 19: Backward airmass trajectory at Kototabang at 24-26 October 2015

55