Thesis Main Body

1

A Model of the Emission and Dispersion of Pollutants From a Prescribed Forest Fire in a Typical

Eastern Oak Forest

A thesis presented to

the faculty of

the Russ College of Engineering and Technology of Ohio University

In partial fulfillment

of the requirements for the degree

Master of Science

Prafulla S. Rajput

August 2010

© 2010 Prafulla S. Rajput. All Rights Reserved.

2

This thesis titled

A Model of the Emission and Dispersion of Pollutants From a Prescribed Forest Fire in a Typical

Eastern Oak Forest

by

PRAFULLA S. RAJPUT

has been approved for

the Department of Chemical and Biomolecular Engineering

and the Russ College of Engineering and Technology by

Valerie Young

Chair, Dept of Chemical and Biomolecular Engineering

Dennis Irwin

Dean, Russ College of Engineering and Technology

3

ABSTARCT

(RAJPUT, PRAFULLA S.,), M.S., August 2010, Chemical Engineering

A Model of the Emission and Dispersion of Pollutants From a Prescribed Forest Fire in a Typical

Eastern Oak Forest (78 pp.)

Director of Thesis: Valerie Young

A simulation model is completed to study the emission and dispersion of carbon dioxide,

carbon monoxide, particulate matter and the temperature variation caused from the prescribed

forest burning in the typical eastern hardwood forests. The purpose of the present study is an

estimation of the output quantities from the fire and its exposure to the life in the vicinity of the

fire. A FORTRAN code is generated which is furnished as an input to the Fire Dynamic

Simulator (FDS) model to simulate the realistic scenario of prescribed fire occurred at the Arch

Rock forest in south eastern Ohio. This FORTRAN model which provided terrain elevation, heat

release, wind flow, soot yield data for the Arch Rock burning scenario was built using

MATLAB. The heat data was collected by hovering planes over the fire carrying remote sensing

equipment which recorded the Infra Red radiations from the fire. The results show how much the

output quantities from the fire are emitted and how long surrounding life is exposed. The

resultant concentration values obtained give an idea of the extent of the harmful pollutants

released from the fire.

Approved: _____________________________________________________________

Valerie Young

Chair, Dept of Chemical and Biomolecular Engineering

4

ACKNOLEDGEMENT

I express my immense gratitude to my advisor Dr. Valerie Young (Chair, Dept of

Chemical and Biomolecular engineering, Ohio University) for continuous mentoring and support

in pursuing the goals of the present thesis. I would like to thank Dr. Daniel Gulino (Grad chair,

Dept of Chemical and Biomolecular engineering, Ohio University), Dr, Douglas Goetz

(Professor, Chemical and Biomolecular Engineering, Ohio University) for their help in shaping

the present thesis and for their promptness to help me anytime whenever I needed them. I would

also thank Dr. William E. Kaufman (Assistant professor, Mathematics, Ohio University) for

being flexible and supportive for the thesis work.

I also thank to Ohio Supercomputer Center (OSC) for allowing me to run my FORTRAN

codes on their supercomputers. I also appreciate the help I received from Matthew Dickinson

(USDA Forest Services) in Literature search. I am thankful to Dr. Gerardine Botte (Professor,

Dept of Chemical and Biomolecular engineering, Ohio University) for providing me the

computers for my MATLAB code executions.

I really appreciate Loredana G. Suciu (Grad student, Ohio University) for helping me to

get the input data needed for my FORTRAN model in this thesis. I also thanks to my friends who

gave me constructive criticism in my thesis work. In addition, I thank all of those who

were directly or indirectly involved in making this work successful.

5

TABLE OF CONTENTS

Page

ABSTARCT........................................................................................................................3

ACKNOLEDGEMENT.......................................................................................................4

LIST OF TABLES...............................................................................................................6

LIST OF FIGURES.............................................................................................................7

Chapter 1 : INTRODUCTION............................................................................................8

Chapter 2 : RESEARCH OBJECTIVE.............................................................................18

Chapter 3 : LITERATURE REVIEW...............................................................................19

Chapter 4 : APPROACH AND METHODLOGY............................................................26

4.1 Approach.............................................................................................................................26

Specifying Domain................................................................................................................27

Specifying Terrain.................................................................................................................29

Specifying Heat Data.............................................................................................................29

Specifying Emission Factor...................................................................................................30

Final estimation for emission factor for particulate matter....................................................34

4.2 Model Description...............................................................................................................34

4.3 Methodology........................................................................................................................36

Chapter 5 : PRESENTATION AND ANALYSIS OF RESULTS....................................43

Temperature and the Gaseous Emission Profiles Analysis.......................................................53

Maximum values of the output exhaust quantities all over the terrain for a time step..............58

Integrated exposure of the temperature and exhaust gases at different heights all over the terrain.........................................................................................................................................65

Chapter 6 : DISCUSSION.................................................................................................70

Chapter 7 : CONCLUSION...............................................................................................73

REFERENCES..................................................................................................................75

6

LIST OF TABLES

Table 4.1: Literature data of PM10 emission factor from wild fire and prescribed burning in various regions in the U.S..............................................................................................................32Table 5.1: The inputs provided during the execution of the fds2ascii.exe program to extract the text files from the FDS output.......................................................................................................47Table 5.2: Literature values for CO2 mole fractions......................................................................61Table 5.3: Literature values for CO mole fractions.......................................................................62Table 5.4: Literature values for PM concentration........................................................................64

7

LIST OF FIGURES

Figure 4.1: A 3D terrain overview spanning 320× 670 ×270 meters in X, Y and Z directions.. . .39Figure 4.2: 1m×1m heat data was added to obtain 10m×10m to fit with DEM resolution...........40Figure 5.1: A 3D terrain overview of 320× 670 ×270 meters in X, Y and Z directions...............44Figure 5.2: Smoke view after 120 seconds....................................................................................44Figure 5.3: Smoke view after 240 seconds....................................................................................45Figure 5.4: Smoke view after 1080 seconds..................................................................................45Figure 5.5: 2D temperature contours at 1080 seconds..................................................................46Figure 5.6: Wind velocity profile difference between (0-30) and (210-240) second intervals.....52Figure 5.7: Three selected locations used to study the nature of the output quantities.................54Figure 5.8: Temperature trends at three locations........................................................................55Figure 5.9: CO2 trends at three locations.......................................................................................56Figure 5.10: CO trends at three locations......................................................................................56Figure 5.11: Soot (PM) trends at three locations...........................................................................57Figure 5.12: Maximum temperature values at different heights all over the terrain for the entire simulation......................................................................................................................................59Figure 5.13: Maximum CO2 concentration at different heights all over the terrain for the entire simulation......................................................................................................................................60Figure 5.14: Maximum CO concentration at different heights all over the terrain for the entire simulation......................................................................................................................................62Figure 5.15: Maximum particulate matters (soot) concentration at different heights all over the terrain for the entire simulation.....................................................................................................64Figure 5.16: A 3D representation of the temperature exposure all over the terrain at different heights............................................................................................................................................66Figure 5.17: A 3D representation of CO2 exposure all over the terrain at different heights.........67Figure 5.18: A 3D representation of CO exposure all over the terrain at different heights..........68Figure 5.19: A 3D representation of PM/Soot exposure all over the terrain at different heights..68

8

CHAPTER 1 : INTRODUCTION

A forest fire is an unavoidable natural phenomenon which frequently occurs all over the

world. According to the United States Department of Agriculture (USDA), the forest land is

spread over a 296 million hectare area within the U.S. till date, 32% of the total land. This means

significant amount of forest is covered all over the U.S. The records show that fires have been

taking place due to the anthropogenic activities or the naturally occurring lightning in Ohio’s

eastern hardwood forests, (Graham et al., 2006). A forest fire involves the combustion of both

the live and dead vegetation lying on the forest surface as well as the surrounding green

vegetation. This combustion emits several different types of gases and particulate matters which

may prove harmful to nearby life since they are released on the ground level; they may also

spread to distant places. Wild fires are triggered by lightning or anthropogenic activities. Once

ignited, they damage everything in their way. To deal with such uncontrolled fires, the U.S.

Forest Department and some land managers have started burning all kinds of fuel loads lying on

the forest floor so that any accidental or natural fire will not spread wildly and burn the entire

forest. These fuels can be litter, duff, dried twigs and dead logs of big trees. This purposeful

burning process of the woods is called controlled or prescribed fire.

Prescribed fires are implemented in the eastern hardwood forests in the U.S. for many reasons: to

promote regeneration of oak trees, to remove unwanted species from the ecosystem and to

prevent future uncontrolled wildfire (Blankenship et al., 2006). Thus, prescribed burning has

been proven to be a good management tool to restore healthy ecology. However, incomplete

combustion of the vegetation while performing prescribed burning leads to the increased

temperature within the forest and the emission of harmful gaseous components like carbon

monoxide (CO), carbon dioxide (CO2), volatile hydrocarbons (VOCs), semi-volatile

9

hydrocarbons (SVOCs) and particulate matters (PM) (Lemieux et al., 2004). Since these

emissions can create visibility and health problems to the surrounding ecology, it is imperative to

study and control them (Lemieux et al., 2004).

According to Liu (2004), hundreds of thousands of forest fires occur every year in the

U.S., and so it is important to create forest fire emission data collection to study their impact. For

this, large-scale inventories including emission data of pollutants like CO, CO2, VOCs, SVOCs

and PM from both natural and prescribed fires have been developed. For example, the

Environmental Protection Agency (EPA) has developed an inventory for 1985-1995 for

prescribed fires, Grand Canyon Visibility Transport Commission (GCVCT) for wildfires during

1986–1992 and prescribed fires in more than 10 western states between 1990 and 1995. Another

comprehensive inventory of the National Emissions Inventory (NEI) was made for the years of

1996, 1999, and 2002; they contained a spatial distribution of forest fire emissions (Liu, 2004).

Also, the detailed statistical analysis of the various pollutants from prescribed burning has been

provided by Lemieux and coworkers (2004).

Many researchers have been studying the emissions emerging from the prescribed

burnings. These studies are based on the fuel type, fuel loading, weather conditions and the

topography of the area to be burnt (Mell et al., 2007). Since different burning conditions in the

forest can have different amounts of the gaseous emissions, by making observations beforehand

on different kinds of burning conditions, it will be easier to predict the emissions before the

actual burning is implemented. For this, the researchers need to perform experiments on a pilot

scale, imitating all of the burning conditions of the site to be burnt at that particular time. But it is

not feasible to perform the burning experiments directly on the fields because of safety measures.

Furthermore, it is difficult to replicate, expensive, time-consuming and possibly potentially

10

harmful to the ecology (Mell et al., 2007). However, some of the laboratory experiments have

estimated the emission factors for particular types of fuels in different forest types across the

U.S. But these estimations have significant uncertainties since realistic emission values would

evidently always be different from those from laboratory experiments. In the laboratories, fire

flame size and intensity are always restricted (McMahon, 1983).

To overcome these problems, arising in the estimation of the emissions from the forest

fires, some of the researchers started building computational models. With the help of these

models, they could simulate the same realistic environment as in the forest fire site and could

estimate the emissions. The development of these computational methods would be helpful to

reduce the time and the cost, which would have been needed in doing real experiments. Also, the

computational tools give remarkable flexibility in changing the inputs so that the emissions in

different climatic conditions for different fuel types and for different topographies can be easily

calculated. Once this target is achieved, forest departments using these emission predictions

could minimize the fire emissions as much as possible by setting the required prescribed fire

conditions.

Over the past two decades, many computer-based tools and models have been used by

wildland fire management in the U.S. to study the fire behavior, fire planning, smoke

management and fire economics (MacGregor et al., 2004). Also, advanced computer technology

tools which were previously available on higher-end computers are now being used on PC

platforms. Because of such fast developments in computer technology, many computer-based

tools are being used not only by federal fire management agencies but also by universities,

private sectors and state agencies (MacGregor et al., 2004).

11

Different kinds of modeling tools are used for different kinds of needs and purposes of

studies in the forest sector in the U.S. The work of Riebau and coworkers (2001) demonstrates

that forest vegetation keeps on changing due to wildfires, prescribed fires, harvesting, thinning,

road construction and many more reasons. So, few models have been developed to predict forest

vegetation dynamics so that these models can be used to study the fire effectiveness and to create

possible alternatives to fire use in the present vegetation type. For example, there is the

Vegetation Disturbance Dynamics Tool (VDDT), The Forest Vegetation Simulator (FVS). Fire

spread and fire flame intensity are predicted by the models like BEHAVE, The Fire Behavior

Prediction and Fuel Modeling System; the First Order Fire Effects Model, (FOFEM), is used to

predict tree mortality, fuel consumption and smoke production. The model called FARSITE, Fire

Area Simulator, simulates the fire growth. The Emissions Production Model (EPM) is used to

predict the smoke emission from the prescribed burning. Various smoke dispersion models being

used include CALPUFF, the Gaussian puff modeling system used to simulate the long-range

smoke transport. A Gaussian dispersion model, SASEM, Simple Approach Source Emission

Model, is used to predict ground level particulate matter and its visibility impact relative to flat

terrain in the western U.S (Riebau et al., 2001). Many physical assumptions are made for the

simplicity in all of the models described in the current section. There are more forms of air

pollution dispersion models classified according to their way of study. For instance, in the

Eulerian model dispersion is modeled in a particular frame of reference as if an observer is

standing at one place and watching the plume go by. On the other hand, in the Lagrangian model

a frame of reference moves with the pollutants as if somebody is walking along with the plume.

But the purpose of the present thesis is to find out the concentration of pollutants at a fixed place

12

over some amount of time period. So Lagrangian modeling is out of the scope of the present

study.

Because of the recent availability and the development of powerful computer and

software packages, grid generating algorithms and models, for example, Computational Fluid

Dynamics (CFD) which describes complex phenomena like fluid flow, combustion, radiation,

turbulence and pressure have been developed dramatically (McGrattan et al., 1998). As a result,

novel high spatial resolution and efficient fluid flow solving techniques are being developed.

Computational Fluid Dynamics (CFD) is the computation based branch of fluid

mechanics which solves different fluid flow problems using numerical techniques. The CFD

techniques are capable of solving air pollution dispersion simulation problems involving

complex geometries, flow conditions and thermal effects (Anderson, 1995). However, CFD

needs large computer time as fine resolution is used (Baklanov, 2000). The non linear partial

differential equations are very difficult to be solved to find the exact solution since they generate

more unknown variables than the equations. Hence, some assumptions and simplifications are

made to reduce the unknown variables to reach as near as possible to the solution and this

resulting solution is called as a closure equation (Baklanov, 2000). Most of the CFD models are

based on the motion equation known as Navier-Stokes equations. The literature review done by

the author shows that Navier-Stokes equations have been used in most of the fire smoke

dispersion models. According to Anderson, (1995), Navier-Stokes equations are the governing

equations of mass, energy and the momentum balance which describe motion of viscous and

compressible fluids. Turbulence part of Navier-Stokes equations is nonlinear in nature because

all scales of motions in all directions are considered in it. It can be solved directly by Direct

Numerical Solution (DNS), but this method is too expensive for most of the practical flows

13

having high Reynold’s Number (Re) (Patnaik et al., 2003). Therefore, other turbulence models

like Reynolds-Averaged Navier-Stokes (RANS) and Large Eddy Simulation (LES) are mostly

used. The large eddy simulation (LES), the turbulence model, was developed by Dr. Joseph

Smagorinsky to study the large scale circulation of the atmospheric air (McGrattan et al., 1998).

LES is one of the numerical techniques to solve Navier-Stokes equations. According to the

Kolmogorov’s (1941) theory of self similarity of eddies, the large scale eddies formed in the

turbulence carry high energy corresponding to the geometry of the turbulence creating flow;

these larger eddies are unstable and break up into smaller eddies with transfer of energy to them.

This continues till eddy motion is stable and is of the same size. All these small eddies are

universal, i.e. are the same for all kinds of turbulence scales and also known as sub grid scale

motions, thus, Large Eddy Simulation (LES) is class of the turbulence where large scale eddies

formed due to the mixing of gases are simulated on the computational grid and sub grid scale

eddies are filtered out and modeled using sub grid scale model called as Smagorinsky model

(McGrattan et al., 1998). Reynolds-Averaged Navier-Stokes (RANS) further includes turbulence

closure models like k-ɛ models and first and second order closure models which simulate the

averaged equations to model the turbulence. However, LES is a better approach which wards off

the limitations of the both DNS and RANS methods. Besides, LES is more accurate and can

handle those flow features such as large scale unsteady flow which cannot be tackled by RANS.

The LES is capable of resolving those properties which Gaussian plume and RANS model

cannot (Patnaik et al., 2003). DNS and RANS techniques are described in detail in the Literature

Review.

Federal agencies, State agencies and other forest fire managers in the United States have become

aware of forest fire impacts and the necessity of its study. However, prescribed burning and its

14

research have been done less in the eastern hardwood region compared to the Western United

States and the Southeastern Coastal Plain (Waldrop et al., 2006). According Waldrop and co

workers (2006), significant fire related knowledge of pine ecology is available in the Western

United States and the Southeastern Coastal Plain and on the other hand role of fire is not much

recognized for the Central Hardwood region, Southeastern Piedmont and Southern Appalachian

Mountains. As a result, implementation of computational models for prediction and study of fire

spread, smoke dispersion and fire behavior are missing or rarely done. Some of the models

which have been used in eastern hardwood region are BEHAVE, FARSITE and FOFEM.

Another project, The Landscape Fire and Resource Management Planning Tools Project,

(LANDFIRE Project), has been conducted to plan, evaluate and implement the hazardous fuel

treatment and restoration for land managers. In this project comprehensive vegetation maps and

data were collected which was the one of the aim of the National Fire Plan. Also, fuel dynamic

modules to study the forest landscape change have been developed: LANDIS for Missouri

Ozarks region and FORCAT model for the Cumberland Plateau of East Tennessee (Waldrop et

al., 2006).

The purpose of the Author in the present thesis was to find out the degree and the

dispersion of the heat and the concentration of the pollutants especially carbon monoxide (CO),

carbon dioxide (CO2) and particulate matters (PM) emitted in the form of the black soot,

released in prescribed burning in the vicinity of the fire source in eastern hardwood forest region.

This was done by developing an input file in the FORTRAN program to the CFD model called

Fire Dynamics Simulator (FDS) which would satisfy the purpose.

The eastern hardwood forest region has many elevations and slopes (Waldrop et al., 2006). Most

of the trees in this region are deciduous and hardwood leaves are generally flat. So after leaf fall

15

sunlight reaches the forest floor activating micro organisms for decaying process. It becomes

very difficult to burn the fuel lying on the forest floor when decaying going on, which is not the

case in other parts of the United States. In the dry season, fire spreads fast but its intensity is

always low because of moist duff lying on the forest floor. According to Sutherland and co-

workers (2003), the study of prescribed burning shows that the fire is curbed to the forest surface

because of wet duff and the fire flame length approximately progressed up to one meter.

As discussed earlier, eastern hardwood forest region has lacked attention of the land managers in

both prescribed fire implementation and modeling of fire spread. Not a single incidence of

specific smoke dispersion modeling is stated (Waldrop et al., 2006). The literature review done

by the author could hardly find any modeling study showing the estimation of the temperature

and the harmful gaseous pollutants concentrations in the vicinity of the fire source. The

Environmental Protection Agency (EPA) has been developing the simulation system in which it

has been using a few models for fire smoke dispersion like MODELS3, NFSPUFF, SASEM and

VSMOKE (Riebau et al., 2001). MODELS3 model simulates emission of pollutants and

particulate matters other than fire emissions. NASPUFF simulates fire emissions and creates the

trajectories of the resultant smoke for complex terrain in western United States; VSMOKE

simulates steady state smoke dispersion from prescribed forest burning based on the Gaussian

plume model. But the Gaussian plume models have their own limitations like they over predict

the concentrations in the dispersion for low wind conditions and at the site closer than 100

meters from the fire source (Holmes et al., 2006). The literature review also found the

application of the CFD model called as Wildland-urban interface Fire Dynamics Simulator

(WFDS), an extension of Fire Dynamics Simulator (FDS), simulates vegetative fuel fire smoke

dispersion problems on complex terrain (Mell et al., 2006). But, this model is especially used for

16

the modeling of the fire spread and gaseous emission at the interface between forest and urban

area.

As per the authors purpose of study of heat release and the other pollutants transport in nearby

area from the source fire typically in eastern hardwood system, many aspects should be

considered for developing a smoke dispersion model. In the forest fire regime, many wide scales

of ranges are involved. So the model should be able to detect and simulate the complex

phenomenon of many physical and chemical processes of different length scales ranging from

millimeter scale like combustion to the miles scale like convection and even bigger scales like

terrain effects and wind effects (Sun et al., 2006). The model should be able to predict interaction

between fire, fuel and the atmosphere, i.e. combustion, radiation and turbulence and flow of the

pollutants over the heterogeneous vegetation and the complex geometry. By considering all of

the phenomena related to the forest fuel burning can make the development of a model difficult,

but the model can be developed according to the developer’s area of the interest(Sun et al.,

2006). Thus, in concern of all of the aspects related to the present thesis purpose, if the author of

this thesis is able to develop such program, it will generate very important information for the

local fire management and land managers of this region so that they can use this information to

plan more healthy prescribed fires in the future. This modeling work may trigger new ideas to

the users and developers of the other smoke dispersion models in eastern hardwood region as

well as the rest of the U.S.

Taking into account of the prescribed fire simulation requirements for the eastern

hardwood region and the present available modeling tools found out by the author after doing an

intensive literature review, CFD model called as Fire Dynamics Simulator (FDS) was the best

choice. Fire Dynamics Simulator (FDS), simulates vegetative fuel fire smoke dispersion

17

problems on complex terrain (Mell et al., 2006). Various governing equations of fluid flow, the

complex turbulence and the numerical methods are used in FDS to calculate gas phase equations

(Mell et al., 2007). FDS is capable of predicting time dependent fire-atmosphere interactions in

three dimensional forms (Mell et al., 2005). Also, the governing equations are approximated to

low Mach numbers, i.e. speeds lower than that of sound waves in the form of Navier-Stokes

equations which can successfully simulate the combustion model using well established mixture

fraction based approach at wide ranges. Thermal radiation transfer in gas phase is solved using

well known finite volume method (FVM). Large Eddy Simulation (LES) is used to model the

buoyancy driven turbulent flow caused due to vertical temperature difference generated from the

fire flames (Mell et al., 2006). Surprisingly, to date, FDS approach has been used for the

simulation of grassland fires only on flat terrain (Mell et al., 2006). This means FDS approach

for the simulation of smoke transport over complex terrain will be a great help for the evaluation

of this model for future users.

18

CHAPTER 2 : RESEARCH OBJECTIVE

1. Model the emission and dispersion of particulate matter, CO, CO2 and temperature from a

prescribed forest fire in a typical Eastern Hardwood forest to determine the probable

exposure of animals in the immediate vicinity of fire source.

2. To try to generalize the modeling method so that it can be applied to other prescribed

burnings in any type of forest to study smoke dispersion with respect to location and time.

To achieve the above objectives, the following tasks will be performed:

a) A FDS model was built to simulate the prescribed burning occurred in Arch Rock forest

in southern Ohio.

b) Comparison of the model outputs against the literature data available for the increased

temperature and the trace chemical species like CO2, CO and particulate matters that

exist for Eastern oak forests.

19

CHAPTER 3 : LITERATURE REVIEW

As discussed earlier, prescribed burning has been implemented in the US for many

beneficial purposes; on the other hand, it may have negative effects since any kind of burning

emits harmful pollutants. This is why researchers have been building computer models to study

the direction, concentration of smoke compounds, distance travelled by particulate matters and

emissions of pollutants. Over the past two decades, many different modeling tools have been

developed for land managers for smoke modeling (MacGregor et al., 2004). Models are also

built to simulate the real time situation of prescribed burning so that predictions of pollutant

concentrations, transport, and dispersion can be efficiently made.

Many Lagrangian models have been used to estimate the forest fire emissions all over the

world. According to Saarikoski and coworkers (2007), biomass burning in North Europe in

spring 2006 emitted PM2.5 over the Helsinki area forming four strikes of episodes. To study the

impact and to forecast the concentration of these particulate matters, the Finnish emergency and

air quality modeling system SILAM was applied. This model successfully estimated the PM to

be 11 ± 3.3 µg/m3 and 9.7 ± 4.0 µg/m3 in two episodes.

The 3D Lagrangian model called NAME was used in UK to trace the sources and

transport of the smoke emerged from the forest fires and agricultural burning happened in 2002

and 2006 in west Russia. It proved to be adequate to simulate the smoke dispersion from

anthropogenic sources (Witham et al., 2007). But it could not predict the composition of the

smoke near the source.

The Forest Fire Behavior Prediction system (FBS) was applied in 2002 in Canada which

predicted gaseous and particulate emissions from boreal forest in Canada (Lavoue et al., 2007).

This model had used fuel pattern information as well as weather data obtained from Canadian

20

weather forecast model GEM (Global Environmental Multiscale). According to this model, 470

kilo tonnes of fine particulate matter were released.

The Australian Air Quality Forecasting System (AAQFS) and the HYSPLIT model

simulated the transport and dispersion of the smoke emerged from Winchelsea and King Island

bush fires which occurred in 2001 in Australia (Hess et al., 2006). This model correctly

simulated the dispersion of the smoke but could not estimate the concentration of individual

pollutants.

In December 2005, intense Buncefield oil fire occurred near London in UK which was

simulated using a model called ALOFT-FT (A Large Outdoor Fire Plume Trajectory-Flat

Terrain). It used Lagrangian vortex dynamics techniques (Vautard et al., 2007). This model

successfully modeled the fire plume and transported the PM directly into the troposphere at high

altitudes. But it could not estimate the total emissions emerged from the fire.

Some simulation studies have been done using Eulerian models. McGrattan and

coworkers (1998) studied the fire scenario in closed room compartment using the computational

fluid dynamics (CFD) method. This method used Navier-Stokes flow equations assuming

constant viscosity. Turbulence model included the Direct Numerical Simulation (DNS) method

to model large scale eddies and sub-grid scale motion was modeled using Smagorinsky’s model

of Large Eddy Simulation (LES). The mixture fraction based method was used for the transport

of low speed, thermally expandable combustion products. Lagrangian particles were used to

represent the fire which carried heat driven by thermally induced flow. This approach

successfully simulated the fire plume rise with temperature change within the plume and showed

a higher degree of accuracy when compared with the experimental data and other time-averaged

flow equation outputs.

21

Large eddy simulation was performed for atmospheric surface layer (ASL) flow over a

gapped plant canopy strip over a flat terrain (Qiu et al., 2008). Navier-Stokes equations of flow

were used to create LES turbulence model assuming compressible flow. According to Qiu and

coworkers (2008), there were no sufficient experimental data for air flow within and above the

heterogeneous plant canopy till that time. Sub-grid scale motions were parameterized by the

Sagaut mixed length SGS model which is an improved form of the Smagorinsky model. The

model successfully produced the different vortices in the gap between the plant canopies. Only

turbulence structure was studied in this paper, while temperature release pollutant emissions and

smoke transports from forest fire were not studied.

The LES model was applied to study turbulence statistics within and above the sparse

forest canopy for compressible flow (Su et al., 1998). The modeled forest was assumed to be

horizontally uniform. SGS motions were modeled using down-gradient diffusion scheme which

is different from the widely used Smagorinsky method. But SGS models should not affect the

LES simulation since they are applied on small eddies. The LES outputs were compared with the

observational data collected from above and within the canopy of deciduous forest. The LES

outputs were in good agreement with observational data even though the domain size was limited

and grid spacing was coarse. Thus, LES can resolve the most important turbulence

characteristics within and above the forest canopy. LES shows very good agreement of all

turbulence characteristics with field measurements especially within the canopy. In this paper no

forest fire smoke dispersion and fire emissions were studied but turbulence properties were.

Lopes da Costa and coworkers (2006) used Reynolds-Averaged Navier-Stokes (RANS)

technique to simulate the wind flow over forests in Scotland and France. They used a turbulence

closure k-ɛ model. Simulation results of wind flow statistics were in good agreement with the

22

experimental data collected at these two sites. However, the authors of this study found that

RANS has limitations in modeling the flow when it comes to flows through and over the canopy.

No fire emission transport was modeled, only turbulence in general was modeled.

RANS with k-ɛ turbulent model was applied to study turbulence properties for eight

different types of vegetation over complex terrain all over the United States. As discussed in the

Introduction, this model was unable to simulate the turbulence in the canopy (Katul et al., 2004).

An LES simulation of smoke plumes from large oil fires was implemented (McGrattan et

al., 1996). Boussinesq approximation was applied to Navier-Stokes equation and was solved

numerically assuming constant eddy viscosity. But in this study simulation was started at several

meters down the fire source where temperature and radiation effects in the plume are negligible.

The model was two dimensional and terrain effects were ignored for simplicity. No smoke

emission was estimated from the model.

Smoke transport from multiple fire plumes was studied in a stably layered atmosphere using

Navier-Stokes equations in Boussinesq form (Trelles et al., 1999). Turbulence was modeled

using LES in two dimensional structure. Lagrangian particles were used to visualize the fire

plume. No smoke emissions were predicted in this study. Boussinesq approximation precluded

the simulation of flow patterns near the fire source.

The CFD model, Finite Element Model in 3-Dimensions and Massively Parallelized (FEM3MP),

was used to simulate flow of chemical and biological matters released in urban environment

(Chan, 2004). This model used Finite Element method to represent buildings and complex

terrain. A Navier-Stokes equation for incompressible flow was used. Both LES and RANS

approaches were used to tackle turbulence. Results showed that the model outputs, using LES,

23

were in better agreement of field data than that from using RANS. Only velocity components

were modeled in this study.

The CFD model known as CFX4.3 was used to simulate smoke movement from fire in building

(Chow and Yin, 2004). This model was based on RANS equations and used k-ɛ model as a

turbulence closure. Fire Dynamics Simulator (FDS) having LES approach was also used to

simulate the same plume rise from fire and was compared with the CFX results. The outcome

showed that FDS as well as CFX results of flow patterns of smoke and temperature contours

were in good agreement with the source data. However, LES approach could yield more detailed

information.

Another CFD model called WFDS, an extension of Fire Dynamics Simulator (FDS),

simulates vegetative fuel fire smoke dispersion problems on complex terrain (Mell et al., 2006).

The governing equations included and the numerical methods used in WFDS to calculate gas

phase equations are the same as those used in FDS (Mell et al., 2007). WFDS is capable of

predicting time dependent fire-atmosphere interactions in three dimensional forms (Mell et al.,

2005). Also, the governing equations are approximated to low Mach numbers, i.e. speeds lower

than that of sound waves form of Navier-Stokes equations which can successfully simulate the

combustion model using well established mixture fraction based approach at wide ranges.

Thermal radiation transfer in gas phase is solved using well known finite volume method. Large

Eddy Simulation (LES) is used to model the buoyancy driven turbulent flow (Mell et al., 2006).

Surprisingly, to date, WFDS approach has been used for the simulation of grassland fires only on

flat terrain (Mell et al., 2006). However, this model specifically is built for urban and forest

interface. That is, it takes an account of the fire spread and emissions occurring at the boundary

24

of an urban area where forests are adjoined so that the distance between forest and the cities can

be decided to preclude the harm or destruction which can occur from the probable fire.

Thus, several studies have been done on transport and dispersion of smoke in the past.

However, relatively few studies have been done to estimate emissions from eastern forest fires in

the US and very little work specifically on hardwood oak forests is done. No study shows the

effects the temperature and concentrations of harmful gases in the vicinity of the fire. Also, the

literature review by author shows that there has been hardly any work done regarding realistic

vegetation and terrain simulation modeling for smoke transport, dispersion and for estimation of

concentration of pollutants, specifically for eastern hardwood forest. Besides, FDS is available

free of charge for public usage internationally. Author observed that the limitations of FDS do

not affect the purpose of the present study in regard of modeling of smoke dispersion and

emission from forest fires. According to McGrattan and coworkers (2008), as discussed in a

technical reference guide, FDS cannot model the scenario which involves as fast flow of

particles or energy as a speed of sound. In present study, smoke dispersion is evidently slower

than sound speed. FDS cannot produce appropriate results if the boundary layer effects are to be

studied since the numerical rectilinear grids produce sharp edges. But, in present study smoke

dispersion is through atmosphere so no question of boundaries arises. Uncertainty in transport of

heat and exhaust products from fire is higher when the heat release rate is predicted instead of

prescribing it. In present study already collected heat release data for forest fire is fed to FDS.

Combustion model in FDS works poorly when fire is in confined, under-ventilated rooms or

compartments, where lack of oxygen affects the fire growth. But, the present study involves the

forest fires open to atmosphere. Thermal radiation model in FDS cannot distribute thermal

energy to long distances where in present study radiation effects are required within few meters.

25

Thus, application of FDS for Arch Rock forest situated in Ohio, takes an account of the complex

terrain, predicts output temperature and concentrations of harmful pollutants in the vicinity of the

fire source. FDS can add very important information for fire management and land managers and

therefore, they can use this information to plan more healthy prescribed fires in the future. This

modeling work may trigger new ideas to the users and developers of the other smoke dispersion

models in the rest of the U.S. since FDS is not much used for such specific purpose.

26

CHAPTER 4 : APPROACH AND METHODLOGY

4.1 Approach

The variation in hydrocarbon concentrations and particulate matters with time and space

in the Arch Rock prescribed burning was studied by simulating smoke emission and dispersion

using the Fire Dynamics Simulator (FDS) model. The details of the governing equations and

theoretical basis of the model involving the hydrodynamic model, combustion model and

thermal radiation model are described in the FDS manual (McGrattan et al., 2008). FDS assumes

that combustion occurs above the surface fuel, i.e. no heat is released in downward direction and

flames are taller than fuel heights. Fire spread, terrain formation and smoke dispersion were

simulated in three dimensions in a box shaped domain, divided into small 3D grids. The domain

was created in accordance with the span of the input elevation data of the topography of the

burnt site. The terrain was colored using .jpg picture. The fire was started at user defined

locations using the data from an Infra Red (IR) photo series, given as an input to the FDS. FDS

represents Fire flames and smoke particles using Lagrangian particles. Smoke vanishes as it

leaves the domain. Wind was driven from the west side of the domain as an input. Vegetation on

the forest floor was set to vanish after being burnt. Heat release and the emission of gases started

at the upper boundary of the fuel. The animated visualization of the simulation was done using a

companion tool of FDS, called SMOKEVIEW. The latter tool is explained in the subsequent

section.

The first step was to create an input text file for FDS, so that the latter would generate the

similar realistic burning conditions which were present in the Arch Rock forest at the time of the

burning. For this, a MATLAB program was written to automatically build this input file in the

FORTRAN language. The simulation model, an input text file, was built step-by-step from small

27

and simple input parameters, like domain size, topography and color of the terrain. The model

was then expanded to a bigger size to include parameters like heat release, wind flow and other

adjustments. At the beginning, a small domain of a few meters in the X, Y and Z directions was

tested and later it was extended into several hundred meters.

Specifying Domain

The domain is a computational box in which the entire simulation takes place. The

interest of the present study was to estimate the emissions only within 15 meters above the forest

floor. However, the top of the domain was extended even higher to increase the scope of the

study if needed. Also, the domain height was increased beyond 15 meters because the uneven

surface of the terrain. Terrain can have different heights at different places above the ground. So

to achieve this and to disperse the emissions from the fire comfortably through the space above

it, the domain was extended 100 meters above the highest peak of the terrain. The domain was

divided into two horizontal blocks at 15 meters above the highest elevation on the forest floor.

The bottom of the domain was closed though the top and sides were open. The ambient wind at

the speed of 0.7 m/s was driven in from the west side of the domain. This mean-wind speed of

0.7 m/s was based on the weather data from the Arch Rock burning site. The domain was divided

into 3D mesh like structure called grids (small cubes). All of the numerical calculations were

performed in every single grid and the products from that grid were transmitted to the next

adjacent one and so on. The smaller the size of the grid, the greater the accuracy, time

consumption and computational cost. Thus, by taking an account of all of these parameters and

the 15 meters of the space of interest of the domain, the optimum grid size in lower block of it

was set to be 2m × 2m × 2m and that for the upper was computed as 10m×10m×10m. The upper

block was beyond 15 meters of the height of interest so the grids in it were kept coarser so that

28

the computational time needed in the numerical calculations in that area would be less. Different

combinations of the grid sizes were tried for the lower and upper block of the domain like lower

block dimensions as 5m × 5m × 5m versus 10m × 10m × 10m for the upper one and were tested

for the feasibility of the execution of the program and the accuracy of the results. Other

combinations like 4m versus 20m and 3m versus 9m were also tested but the program could not

run due to the error arose because of the alignment mismatch issue. The 1 ×1 × 1 versus 10m ×

10m × 10m grid size took five days to finish an execution of the program for a second period of

burning time on the OSC system. The total simulation period of the present study was supposed

to be of 3210 seconds of burning. So, 1 ×1 × 1 meters resolution could have taken years to finish

a single program-run. So this combination was discarded.

The next possible size, other than 2m × 2m × 2m, was 5m × 5m × 5m for the lower block of the

domain which could run successfully on the OSC system. The numerical results from both 2m ×

2m × 2m and 5m × 5m × 5m were compared statistically for every pollutant, at each height and

at each time. Statistical test was performed using MATLAB. It involved the comparison of the

means of the emission data for each of the pollutant from 2m × 2m × 2m and 5m × 5m × 5m grid

sizes using two-factor paired t-test. First factor was the size of the grid and another one was the

blocking factor, a location, for the emissions. The test showed at more than 90% occasions the

emission data for 2m × 2m × 2m grids was significantly greater than that of 5m × 5m × 5m size

with 95% confidence interval. This means the output data from 2m × 2m × 2m size grids was

more accurate than those from 2m × 2m × 2m size. This was because the numerical output

emission data was an average value for an individual grid cell and was stored at the center of

each grid (Mell et al., 2006). Thus, obviously the average value for the 2m × 2m × 2m volume

would be greater than the one for 5m × 5m × 5m since the decrease in the concentration of the

29

emissions due to dispersion along 2 meters would always be less than that along 5 meters of the

length and so would their averages.

The target output quantities, resulted from the FDS model, such as temperature, wind

velocity, particulate matter, carbon monoxide (CO) and carbon dioxide (CO2) were recorded

every thirty seconds in each grid cell in the domain. The nonpoint measurements were also made

every minute at random places to obtain contours and trends of all the target output quantities all

over the domain so that they can be visualized easily to check their profiles. The simulation was

run for few seconds at the beginning every time when amendments were made in the input text

file to make sure everything was working correctly. After that the program was set to run for the

desired period of time which was 3210 seconds. The last heat release data given as an input to

the FDS was at 3206th second.

Specifying Terrain

The terrain was built so it would imitate the exact topography of Arch Rock forest, where

the prescribed burning took place. The elevation data for the terrain was retrieved by using

Digital Elevation Model (DEM) text file. DEM is a digital representation of the terrain

elevations. This digital data was in the form of raster, i.e. square grids. These square grids or

pixels measured 10m × 10m. Additionally, DEM contains the information of the exact location

of that elevation data in X and Y co-ordinates actually on earth. The MATLAB program was

used to read the DEM data and store it in a matrix form. Terrain was colored using jpg file so

that it can be visualized using SMOKEVIEW software.

Specifying Heat Data

Heat released from the actual fire that occurred in the Arch Rock forest was recorded by

hovering planes over the fire. The planes used remote sensing equipment to record Infra Red (IR)

30

radiations from the fire. A GIS tool extracted the data from IR images to form text files including

heat data for 1m × 1m resolution grid squares and actual X, Y co-ordinates of these grids on

earth. A total of eleven IR images of heat release data were available for different time sequences

starting from zero seconds on a timer at 8, 243, 586, 865, 1228, 1493, 1822, 2057, 2432, 2659

and 2996 seconds as per recorded by the aircraft. The author obtained this IR image information

in the form of a raster text file from his academic advisor Dr.Valerie Young.

Specifying Emission Factor

According to Lawrence and coworkers (2007), “an emission factor is a representative

value that attempts to relate the quantity of a pollutant released to the atmosphere with an activity

associated with the release of that pollutant.” Emission factors are generally expressed as a mass

of pollutant divided by a number of factors such as mass, volume, distance and the duration of

the pollutant emitting activity (Lawrence et al., 2007). An example of this is: kilogram of

particulate matter emitted per kilogram of fuel burned in forest fire.

Thus, the general equation for emission factor estimation would be:

EF= EA

[1]

where,

EF = emission factor

E = emissions of a pollutant

A = fuel consumption rate.

The input file into FDS needed the soot yield, i.e. particulate matter information an input

to produce the emissions of soot and carbon monoxide and carbon dioxide. PM10 were

considered for a soot yield data. Literature emission factors were determined and used for scaling

31

the tracer species to find the amounts of other species of interest. The combustion process

consumes oxygen and releases heat. Different burning materials, for example, different fuels,

consume different amounts of oxygen according to their composition (Hugget, 1980). Therefore,

the amount of energy released per unit mass of oxygen consumed (EPUMO2) for the organic

fuel found in the forest like woody materials, was determined to be 13100 kJ/kg (Hugget, 1980).

For many years, emission factors have been calculated using the carbon balance method

(Koppmann et al., 2005). The basic idea is that the emission factor of a particular pollutant is

expressed by the ratio of the mass concentration of the pollutant itself to the carbon

concentration of the gases emitted from the fire. This is because combustion is basically a

reaction of carbon content in forest fuel with oxygen (Koppmann et al., 2005). Thus, a new

emission factor equation would be:

EF= [ p][C ] A ( g

kg )[2]

where,

[p] = concentration of pollutant under consideration,

[C]A = [C] CO2+ [C] CO + [C] CH4+ [C] Non Methane Hydro Carbons (NMHC) etc.

F = mass fraction of carbon in the fuel.

To convert the emission factor in grams per kilogram fuel burned, EF in the equation [2]

is multiplied by the mass fraction of carbon in the fuel. Literature values for particulate matter

emission factors are summarized in Table 4.1.

32

Table 4.1: Literature data of PM10 emission factor from wild fire and prescribed burning in various regions in the U.S.

No.PM10 (g/kg)

DescriptionWild Fire

Prescribed Fire

1 23.7 -Hardwood Forests(Battye et al., 2002)

2 26Entire US forests(Battye et al., 2002)

3 - 2.5 to 90Entire US forests(McMahon, 1983)

4 75 -Lab experiments on pine litters(McMahon, 1983)

5 21 -Lab experiments on slash type fuel(McMahon, 1983)

6 8.38 -Eastern hardwood forests(AP-42, 1995).

7 - 14In North central and eastern region in all(AP-42, 1995)

8 - 18By fire and fuel configuration(AP-42, 1995).)

9 - 11For state of Oregon(Radke et al., 1990)

10 - 13.3For state of Washington(Radke et al., 1990)

† 11

- 14Average over the period of 1980-2002(Liu, 2004), taken from (AP-42, 1995)

† 12

12.5Temperate broad leaved deciduous forest with closed canopy(Wiedinmyer et al., 2006)

† 13

15Mixed broad leaved or needle leaved with open forest canopy(Wiedinmyer et al., 2006)

Note: The emission values denoted by † are the values for the terrain and forest types similar to that being modeled in the present thesis.

According to Battye and coworkers (2002), total particulate matter emission factor from

the wild fire in hardwood forests was 23.7 (g/kg) and that for the entire US forests as a

whole was 26 (g/kg).

33

According to McMahon (1983), the total suspended particulate matter emitted from

prescribed burning in USA varied from 2.5 to 90 (g/kg) depending upon the local fuel

and fire type. Moreover, laboratory experiments on pine litters showed that the wildfire

particulate matter emission factor was 75 (g/kg) and that for slash type fuel was found to

be 21 (g/kg) (McMahon, 1983).

Particulate matter (PM) emission factor in eastern hardwood forests was 8.38 (g/kg).

According to wildfire fuel consumption in 1971 and further based on the judgment of

forestry experts on prescribed burning, the emission factor for the particulate matter

(PM10) was 14 (g/kg) in North central and the entire Eastern region (AP-42, 1995). By

fire and fuel configuration, hardwood emits particulate matter with an emission factor of

18 (g/kg) for prescribed burning.

Radke and coworkers (1990) showed that the emission factors for total suspended

particulate matters (TSP) for the Oregon prescribed burnings was 11 (g/kg) and was 13.3

(g/kg) for the Washington burnings.

Over the period of 1980-2002 the average emission factor for PM10 from prescribed

burning in North central and Eastern region was 14 (g/kg) (Liu, 2004). But this value,

showed in Liu (2004), was referred from (AP-42, 1995).

According to Wiedinmyer and coworkers (2006), the PM10 emission factor for the

temperate broad leaved deciduous forest with closed canopy was 12.5 (g/kg) and that for

mixed broad leaved or needle leaved with open forest canopy was 15 (g/kg).

34

Final estimation for emission factor for particulate matter

Wildfires emit a comparatively larger amount of particulate matter (PM10) than what

prescribed fires do. This is because wildfires are uncontrolled and burn everything in their way.

On the other hand, prescribed fires yield a comparatively less amount of PM10, since they are

well-controlled and most of the time, only forest floor vegetation is burned. The REAC namelist

group in the input file into FDS required the soot yield value, which is why particulate matter

emission factor especially for prescribed fires was preferred.

McMahon (1983) found that the values of particulate matter emission factor for

prescribed burning usually lie between 2.5 and 90 (g/kg). This means the final value of PM

should lie between these two values. From Table 4.1, for eastern hardwood forest prescribed

burnings, the PM10 emission factor values available in (g/kg) are 14, 12.5, 15 whose average

would be 13.83 (g/kg). Thus, Soot Yield, i.e. PM10 emission value, considered in an input file to

FDS was 0.01383 kg/kg.

4.2 Model Description

FDS is a physics-based 3D model, made up of a FORTRAN program, which uses

computational fluid dynamics, CFD, methods to solve the governing equations of fluid

dynamics, motion and the thermal degradation of biomass fuels in field fires (Mell et al., 2006).

FDS reads the input parameters from a text file and produces user defined outputs in the

numerical form, by solving governing equations. FDS accompanies a software, Smokeview,

which is a visualization program written in C/OpenGL programming language; it produces

images and animations from the numerical results produced by the FDS. Out of two versions, the

fuel element model and the boundary fuel model, the former model has been used in the present

35

simulation study. The fuel element model comprises fuels in a specified volume, e.g. tree crown;

and the boundary fuel model comprises only surface-lying fuels like grass, dried leaves, twigs,

etc. The fuel element model can include any kind of surface fuels such as trees, grass, bushes

with the computational grid with sufficiently fine resolution. Therefore, it was advisable to use it

for the modeling purpose in the present research study.

FDS (Fire Dynamics Simulator) is developed at the National Institute of Standards and

Technology (NIST), Building and Fire Research Laboratory (Mell et al., 2006). The model

solves Navier-Stokes equations of mass, momentum and energy balance for low speed, thermally

driven, compressible flow for simulation of smoke and heat transport from the forest fires. FDS

is also used to model the convective and radiative heat transfer, pyrolysis and fire growth

(McGrattan et al., 2008). The Smagorinsky form of the Large Eddy Simulation (LES) method is

used to simulate the turbulence in buoyancy driven flow allowing a large variation in the

temperature and density. The smagorinsky form solves the equation related to the eddies

produced because of turbulence in the air or smoke. This produces an elliptical character

equation. FDS solves the governing equations on rectilinear grids defined by the user. FDS does

not have Reynolds-Averaged Navier-Stokes (RANS) capability for turbulence solving. RANS

averages the turbulence in fluid motions based on the Navier-Stokes equations (McGrattan et al.,

2008). The combustion of the forest floor fuels is simulated using the combustion model in the

FDS. The combustion model involves using a mixture fraction method. A mixture fraction is the

mass of a gas species in a given volume of the total mixture of the gases. The transportation of

the heat produced during the fire is simulated using the radiative transport equation which is

further solved using the Finite Volume Method (FVM). The heat release rate for the burning,

used in the pyrolysis model of the FDS, is defined by the user as an input. All of these modeling

36

equations are explained in detail in the FDS manual of version 5 in the technical reference guide

(McGrattan et al., 2008).

4.3 Methodology

After reviewing the available models to study the forest fire emission and dispersion, found in

the literatures, it was concluded that the FDS was the best model to serve the purpose of the

present research question. This is discussed in the Literature Review chapter. The FDS and its

companion software, Smokeview, were downloaded from the internet at the website

http://fire.nist.gov/fds. All of the FDS related executable files, source codes, manuals and sample

input files were also downloaded from the same website. At the beginning, a very simple input

file in compliance with the present research scenario was considered and run on the PC to learn

and exercise the functioning of the FDS and Smokeview. This input file was downloaded from

the website again http://fire.nist.gov/fds. Then the parameters like terrain geometry, color of the

terrain were added up later on.

The FDS required lots of computer memory space and higher end processors to solve the

complicated partial differential equations called governing equations in the present study. With

the addition of the complex parameters like combustion and radiation, author experienced more

memory demand for the FDS program which made difficult it to run on a PC. To overcome this

problem, author, with the help of his academic advisor Dr. Valerie Young, established a remote

access connection to the higher end computers at the Ohio Supercomputer Center, Columbus,

Ohio (OSC). According to the official website of OSC, it provides supercomputer services to

Ohio colleges, companies and universities. OSC includes serial as well as parallel processing

using a distributed memory techniques. That means a program is allowed to run on several

37

computers at a time in parallel. The remote connection between OSC computers and author’s PC

was set up which could be accessed using user’s account. The most important thing with the

OSC computers was they could be accessed from any computer. Graphical User Interface (GUI)

called Secure Shell Client (SSH) was employed for a communication between OSC computers

and author’s PC. All FDS source, executables and compilation files needed to run a FDS

program were downloaded from http://fire.nist.gov/fds. These files were installed on the OSC

computer-system with a user account accessed through the SSH file transfer client. The OSC

computers needed UNIX platform for operation. A batch file, called ‘myjob’, (see Appendix A),

including commands for UNIX platform, desired information of the FDS input text and

executable files was used for the execution of the FDS program at the OSC end.

The FDS model is nothing but a FORTRAN program to which geometry, combustion,

time, flow and likes of these input parameters were furnished through the input text file. Such an

input file was created and saved with an extension as .fds. MATLAB program was used to write

this input text file. Also, MATLAB read other input parameters like the elevation and heat data.

The information fed through the input file to FDS consisted of thousands of different FORTRAN

command lines. This was accomplished with the help of MATLAB within a few minutes. As

discussed earlier, the topography information of the Arch Rock forest was reported in a text file

called Digital Elevation Model (DEM). DEM-elevation-data with 10m × 10m resolution was

read by MATLAB and stored as an array along with a reference matrix. The reference matrix

contained the actual reference values of X and Y coordinates of all the data values, square-grid-

points, in a DEM file on earth. The input file called input.fds is attached in an Appendix (B).

Boundaries of the computational domain were set at the periphery of the DEM data

points set. The five meters of the length was added at the external boundaries, peripheral data

38

points of the DEM, of the domain since the elevation data (DEM) was positioned at the centre of

each 10m × 10m grid square else the elevation data grid at the edges of the domain would be out

of the domain boundary by half of its size. The top of the domain was secured 100 meters above

the highest peak of the terrain. All sides of the domain were left open except the bottom one

where the terrain was supposed to be rested on. The mean wind velocity on the day of prescribed

burning was measured at Arch Rock by forest department officials as 0.7m/s. The wind was

blown into the domain from the left. The ambient temperature inside the domain was set as 280C

which was the room temperature at the time of Arch Rock prescribed burning.

The data points in the DEM file were the elevation values of each 10m × 10m grid square

area (pixel) of Arch Rock forest flooring. The elevation value was averaged for each pixel i.e.

the elevation for the particular pixel, 10m × 10m area, was even. The rectilinear solid columns of

the flat tops and 10m × 10m cross-sectional area were built to represent the terrain. These

columns had the heights equal to the respective elevation values from the DEM file as shown in

Figure 4.1. Thus, every column represented a particular DEM data value. All of these columns

were grounded on the bottom side of the domain abutting each other. The top surfaces of these

columns were open where fire could be ignited and the walls were closed and inert. The abutting

tops of all columns formed the realistic ups and downs of forest floor which resembled to Arch

Rock flooring. The sharp corners and edges of all the column tops were made smooth using a

FORTRAN command to avoid the wind vortices when it touches the terrain surface. Terrain

formed was colored using .jpg image file which was visualized with the help of Smokeview as

shown in Figure 4.1.

39

Figure 4.1: A 3D terrain overview spanning 320× 670 ×270 meters in X, Y and Z directions.

Heat release data, watts, was collected over the eleven time sequences varying from 0 to

2996 seconds. It was read by MATLAB from 1m × 1m resolution raster files and was stored in

eleven matrices along with respective reference matrices with their actual X-Y position on earth.

The terrain data was of 10m × 10m resolution and the heat release data was of 1m × 1m over the

Arch Rock forest floor. So the heat data was added up and averaged to construct 10m × 10m grid

square resolutions each to match up with every DEM forest floor data point as shown in Figure

4.2.

40

Figure 4.2: 1m×1m heat data was added to obtain 10m×10m to fit with DEM resolution

The real measured heat release from the actuafl fire was found to be 58% of the heat flux

recorded in IR images so it was multiplied by 0.58 and also converted into units of kilo-watt per

meter square (kW/m2). If the heat release data was available for any location on the terrain

surface that meant the fire existed at that location. All points on the surface of the terrain, with

respect to the DEM-data point-coordinates, were checked to find out if there was any fire at any

time. If fire was present then heat release information was applied at that location i.e. on the top

of the rectilinear column, using the IR heat data. MATLAB was used to confirm the heat release

from eleven heat data files and apply this information on the tops of the respective solid

columns. FDS used this information to produce the exhaust gases emitted from the fire. The heat

data values were ramped on the tops of the columns across the terrain according to the respective

41

eleven time sequences collected by the aircrafts. This was The author observed that when wind

was blown into the domain it took approximately 210 seconds to settle and match the normal

velocity flow pattern which was supposed to be a smooth exponential growth to the velocity of

0.7 m/s. That is why, the ignition was started 210 seconds after i.e. heat release or fire was

triggered and ramped in the domain after 210 seconds had past. As discussed in earlier section,

mass of soot formed per mass of fuel burned i.e. literature soot yield value was estimated to be

0.01383 kg/kg from and conveyed to FDS through the input file. Another input parameter,

amount of energy released per unit mass of oxygen consumed (EPUMO2) for the organic fuel,

was determined to be 13100 kJ/kg from the literature and conveyed through the input text file to

FDS.

From the personal communication with Dr. Valerie Young, the author realized some facts

about the missing heat data in IR images from the Arch Rock forest fire. Forest fire involves

various lengths of times of blazing fire flames. In other words, during the fire, few fuel types

such as thick wooden branches of trees and logs may keep burning for hours and others like dry

grass and dried leaves burn away within few minutes or less. Remote sensing aircrafts were

recording heat release data approximately at every five to six minutes from the fire in Arch Rock

forest. So, there were many chances for aircrafts, which were getting back over the fire to collect

the heat data, of missing those burnings which lasted less than 5 minutes or when aircrafts had

gotten over the fire some of fire incidents might be at dying stage. This missing heat data might

significantly affect the overall heat release from the forest burning. So there was a need to add

this missing heat in current available heat release data. Fire front at the interface close to the

unburned fuels is always at its peak of intensity of heat radiation. If the average of all of these

peaks from eleven-IR image-heat-data was plugged in to the heat release values of fire events

42

occurred at all time sequences matching with the depression of fire could fulfill the missing part

of heat released. For this, eleven Fire Radiative Peak (FRP) data was obtained from Loredana

Suciu, grad student, Ohio University. The average FRP values were inserted into original heat

data using MATLAB and ramped in input file all over the terrain surface for fire ignition. All of

the desired products from the resultant simulated fire were recorded every thirty seconds in entire

domain.

43

CHAPTER 5 : PRESENTATION AND ANALYSIS OF RESULTS

A MATLAB program called input_file.m (see Appendix C) was used to produce an input

text file called input.fds (see Appendix B). The input file was furnished to the FDS and the

simulation was run for 3210 seconds of theoretical burning time. It took almost three days for the

OSC system to run the input file, which included 3210 seconds of simulation. The simulation

domain was a volume of 320× 670 ×270 meters in X, Y and Z directions respectively, including

terrain resembling that of the Arch Rock forest. The simulation outputs included the

concentration data of CO2, CO, temperature and soot (particulate matter) for each grid cell above

the ground surface in the domain, and were recorded every 30 seconds. Vertical planar records

through the half way of the domain from X and Y directions were taken to easily view the

contours of the wind velocity and the emission patterns of temperature, CO2, CO and particulate

matter. All of these numerical results could be visualized using SMOKEVIEW. The requested

output data was stored into output files called slice files with the extension .sf as discussed in the

FDS user guide (McGrattan et al., 2008).

The software SMOKEVIEW produces animated visualizations of the numerical results

generated by the FDS. Figure 5.3 shows the 3D domain of 320× 670 ×270 meters in X, Y and Z

directions respectively, which is a part of the original Arch Rock forest’s terrain. All of the

numerical results from FDS were visualized from the .smv file generated when the input file was

run (McGrattan et al., 2008).

44

Figure 5.3: A 3D terrain overview of 320× 670 ×270 meters in X, Y and Z directions.

Images from randomly chosen times from the SMOKEVIEW animation were used show

examples of smoke formation and the temperature contours as shown in Figure 5.4 through

Figure 5.7.

Figure 5.4: Smoke view after 120 seconds

As the time progresses, the amount of the black soot starts growing in size and amount, as

shown in Figure 5.5 and Figure 5.6.

45

Figure 5.5: Smoke view after 240 seconds

Figure 5.6: Smoke view after 1080 seconds

Because of the heat released from the fire, the air in the vicinity of the flames gets hot.

The 2D temperature contour formed at 1080 seconds at the middle of the X-axis is shown in

Figure 5.7. The red colored area shows the hottest spot and dark blue is the coolest.

46

Figure 5.7: 2D temperature contours at 1080 seconds

As stated earlier, FDS was run on the OSC system. The resultant files were copied back

on the author’s computer for analysis using SSH secure shell file transfer client. The recorded

emission data requested by the user was stored in slice files; refer to FDS user guide (McGrattan

et al., 2008) to know more about slice files. Therefore, the first task was to transform this data

into text format. A small utility FORTRAN program called “fds2ascii.exe” was used to extract

numbers from the FDS output data files. This executable program was downloaded along with

the FDS from the site http://www.fire.nist.gov/fds/downloads.html. It was run from the

Command prompt window where it asked some questions, as shown below, to get the desired

output text files that had the numbers in them. An example of generating a file called

‘210to240’, and typical inputs provided to extract it in the present study, is shown below in Table

5.2. It describes the commands that popped up on the command Prompt screen and the inputs

furnished accordingly. Inputs provided by the author are shown in italic font.

47

Table 5.2: The inputs provided during the execution of the fds2ascii.exe program to extract the text files from the FDS output

Index Command Explanation

1 C:\Users\> file path2 C:\Users\> fds2ascii fds2ascii.exe is invoked3 Enter Job ID string

(CHID):Input

Type the character string ID ‘input’. This was the job name for the input.fds file

4 What type of file to parse? PL3D file? Enter 1 SLCF file? Enter 2 BNDF file? Enter 32

The data requested by the author was stored into slice files (SLCF) so the digit ‘2’ was entered

5 Enter Sampling Factor for Data? (1 for all data, 2 for every other point, etc.)1

All data was extracted by entering ‘1’.

6 Limit the domain size? (y or n)N

The domain size was not altered to obtain the entire data set.

7 Enter starting and ending time for averaging (s)210240

The FDS was set to record data every 30 seconds, and the fire was ignited at the 210th second, so two values at thirty seconds apart were entered.

8 input_01_01.sf TEMPERATURE temp C 1 MESH 1, z= 0.00, TEMPERATURE input_01_02.sf carbon dioxide X_CO2 mol/mol 2 MESH 1, z= 0.00, carbon dioxide input_01_03.sf carbon monoxide X_CO mol/mol 3 MESH 1, z= 0.00, carbon monoxide input_01_04.sf soot density soot mg/m3

The slice files with .sf extensions contained the data requested by the user in the input .fds file. There were 24 types of different data stored in these slice files. In the present thesis, the four quantities of temperature, CO2, CO and PM were supposed to be studied, so the number ‘4’ was entered.

48

4 MESH 1, z= 0.00, soot density input_01_05.sf TEMPERATURE temp C 5 MESH 1, x= 160.00, TEMPERATURE input_01_06.sf VELOCITY vel m/s 6 MESH 1, x= 160.00, VELOCITY input_01_07.sf carbon monoxide X_CO mol/mol 7 MESH 1, x= 160.00, carbon monoxide input_01_08.sf soot density soot mg/m3 8 MESH 1, x= 160.00, soot density input_01_09.sf TEMPERATURE temp C 9 MESH 1, y= 336.00, TEMPERATURE input_01_10.sf VELOCITY vel m/s 10 MESH 1, y= 336.00, VELOCITY input_01_11.sf carbon monoxide X_CO mol/mol 11 MESH 1, y= 336.00, carbon monoxide input_01_12.sf soot density soot

49

mg/m3 12 MESH 1, y= 336.00, soot density input_02_01.sf TEMPERATURE temp C 13 MESH 2, z= 86.00, TEMPERATURE input_02_02.sf carbon dioxide X_CO2 mol/mol 14 MESH 2, z= 86.00, carbon dioxide input_02_03.sf carbon monoxide X_CO mol/mol 15 MESH 2, z= 86.00, carbon monoxide input_02_04.sf soot density soot mg/m3 16 MESH 2, z= 86.00, soot density input_02_05.sf TEMPERATURE temp C 17 MESH 2, x= 160.00, TEMPERATURE input_02_06.sf VELOCITY vel m/s 18 MESH 2, x= 160.00, VELOCITY input_02_07.sf carbon monoxide X_CO mol/mol 19 MESH 2, x= 160.00, carbon monoxide input_02_08.sf soot density

50

soot mg/m3 20 MESH 2, x= 160.00, soot density input_02_09.sf TEMPERATURE temp C 21 MESH 2, y= 340.00, TEMPERATURE input_02_10.sf VELOCITY vel m/s 22 MESH 2, y= 340.00, VELOCITY input_02_11.sf carbon monoxide X_CO mol/mol 23 MESH 2, y= 340.00, carbon monoxide input_02_12.sf soot density soot mg/m3 24 MESH 2, y= 340.00, soot density How many variables to read: (6 max)4

9 Enter index for variable 11 Integral of TEMPERATURE = 0.0000E+00

Out of 24 types of the outputs, the temperature for lower half block i.e. mesh-1 was extracted by feeding the number ‘1’

10 Enter index for variable 22Integral of carbon dioxide = 0.0000E+00Enter index for variable 33Integral of carbon monoxide= 0.0000E+00Enter index for variable

Similarly, ‘2’,’3’ and ‘4’ were entered for CO2, CO and soot density (PM) respectively.

51

44 Integral of soot density = 0.0000E+00

11 Enter output file name:210to240

The out file containing data was named by user.

12 Writing to file... 210to240

The text file then was written according to the path provided.

Using the same set of the inputs shown in Table 5.2, the hundred text files were extracted

for every 30 seconds from the 210th through the 3210th second. All 100 files were converted into

CSV files so that they could be used further. In the present study, only the data for the lower

mesh of 2×2×2 meters grid size was extracted since the purpose of the study was only for within

the height of 15 meters above the ground. The intended area within 15 meters was

accommodated in the lower mesh. The FDS manual called User’s Guide contains the details of

fds2ascii command. Since the exhaust gases information was recorded every 30 seconds in FDS

simulation, the numerical data extracted from the slice files using fds2ascii.exe was for every 30

seconds. The wind, blowing into the domain, acquired expected velocity profile after 210

seconds had past as shown in Figure 5.8.

52

Figure 5.8: Wind velocity profile difference between (0-30) and (210-240) second intervals

A location was randomly chosen on the terrain to check the wind velocity profile. In

Figure 5.8, the average-wind-velocity profile has some noise at the height of 86 meters above the

terrain in 0-30 second interval but was smoother, as was expected, for all heights in the 210-240

interval. So, fire ignition was started at the 210th second all over the terrain to get proper

dispersion of the fire emissions. A total of one hundred data files were produced, starting from

the 210th second and ending at the 3210th second for every thirty second interval, for example, the

first file would be for 210 to 240 seconds, the next from 240 to 270 seconds and so on. Every

output text file contained the output gases information in columns with the header X (meters), Y

(meters), Z (meters), TEMPERATURE (0C), carbon dioxide (mol/mol), carbon monoxide

(mol/mol) and soot density (mg/m3) for each and every cell of the domain. Here, X, Y and Z are

the coordinates of each grid cell.

53

Temperature and the Gaseous Emission Profiles Analysis

At this point, the entire data set was ready for analysis. Naturally, fuel at room

temperature, when ignited, starts burning and attains the highest temperature and emission rates

of the gaseous outputs, and then again comes down to room temperature with the ashes left

behind. The same is true for the gaseous pollutants emitted by the burning fuel. The pollutants

are emitted in accordance with the fuel burning rate. The emissions were zero and the

temperature was 280C before the ignition started at the 210th second. The emission of gases and

the temperature variation during and after the burning was to be studied. The first part was to

verify whether the emissions and temperatures have the expected profiles over the course of the

fire. The temperature and the gases were expected to have an exponentially decaying nature,

since the fire at its peak has higher temperature and emission rates which should drop down after

fuels are burnt over the period. Also there was a need to verify if there is any location difference

in the temperature variation and the gaseous outputs.

Three different locations, i.e. pixels, were chosen on the terrain surface. Out of these

three pixels, one was on the west side slope, one was on the east side and another was on the

ridge as shown in Figure 5.9.

54

Figure 5.9: Three selected locations used to study the nature of the output quantities.

As stated earlier, the burning was started at the 210th second and lasted until 3210

seconds, i.e. 50 minutes. The emission data at every 10 minutes was extracted and plotted to

examine the behavior of the output quantities. This created six time steps starting from the first

minute. A MATLAB program (see Appendix D) generated by the author was used to collect and

plot the data of temperature change, and that of CO2, CO and particulate matter concentrations

from the CSV files at the three locations. The plots generated by the MATLAB code are shown

below in Figure 5.10 through Figure 5.13.

Ridge East side slope

West side slope

55

Figure 5.10: Temperature trends at three locations.

Figure 5.10 shows that the temperatures dropped almost exponentially from their higher

values to the room temperature of 280C. Different locations had different initial peak

temperatures. Also, the temperatures are higher at the ground surface than that at 15 meters

above it. The temperature at 15 meters above the ground at the west side slope was 280C all the

time as shown in the lower right block of Figure 5.10. Maximum temperature attained at the

ridge was 360C at 15 meters above the ground.

56

Figure 5.11: CO2 trends at three locations

The CO2 level drop was also exponential in appearance, as was expected at all locations

and heights. The maximum concentration was found near the surface and it was less at 15 meters

above the ground, as shown in Figure 5.11. The CO2 concentration was almost zero at 15 meters

above the ground at the west side slope of the terrain at that particular location as shown in the

lower right block of Figure 5.11.

Figure 5.12: CO trends at three locations

The same was the case with CO and particulate matter concentration levels as shown in Figure 5.12 and Figure 5.13.

57

Figure 5.13: Soot (PM) trends at three locations

The output temperature and the gaseous quantities data were stored in the CSV files for

every grid cell in the domain. But, the purpose of the present research study was only above the

ground surface. This means there was no temperature or gaseous emission data below the surface

of the ground and so was for all the grid cells below it in the domain. All of the grid cells had

zero value of emissions below the ground and these cells were also included into CSV files

which in turn made them huge in size. So, it was necessary to eliminate the grid cell data which

were below the ground surface. This was accomplished by MATLAB code (see Appendix D) to

save the execution time since it would not have to go through these grids again and again to

reach the desired location. Also, MATLAB looked the data at desired locations with the help of

the X-Y co-ordinates of that location. The emission data was stored at the center of each grid

cell. So, if MATLAB looked for the particular location with particular X and Y co-ordinates

there was the case where the X-Y co-ordinates of the center of the grid cell would not match with

58

that of the location. There were many nearest grid cell centers equidistance from each location in

consideration in 1 meter surrounding in X, Y and Z directions. So, the data found in this area

were collected and averaged to get the emission data for that particular location. This was done

each time whenever any analysis was conducted.

Maximum values of the output exhaust quantities all over the terrain for a time step

As discussed earlier, the CSV file contained all of the grid cells data values including

those below the ground surface where gaseous emissions had zero values. So, the data regarding

these cells was removed from the CSV file to shorten it using MATLAB. This cut down the

execution time for needed for the MATLAB program to run to get the maximum values of

emissions at a time over the terrain.

The maximum concentration values, at each time step, for CO2, CO and soot (particulate

matter) and the temperature were picked from all over the terrain using each CSV file. Thus, 100

values were obtained for 100 CSV files using the MATLAB code (see Appendix E) for 3000

seconds of simulation period. Again, to match the X-Y co-ordinates of the center of each

10m×10m pixel, for entire terrain, with that of the grid cell close to the pixel, MATLAB code

was customized to pick the emission values in +/- 5 meters area on the X-Y plane and +/- 1

meter along the height. Every pixel had 10m×10m X-Y area this is why the emission values in

+/- 5 meters around the center of the pixel were picked up in the domain so that not even a single

value of emission over the terrain could be missed. This code finally generated the plots as

shown below in Figure 5.14 through Figure 5.17.

59

Figure 5.14: Maximum temperature values at different heights all over the terrain for the entire simulation.

The X-axis values are for the 100 time steps at which the emission were recorded at every

30 seconds adding to total of 3000 seconds. Figure 5.14 shows the trends of the maximum

temperature at each time step on the entire terrain surface arose due to the burning. The

maximum temperature at the ground surface varied approximately between 45 to 60 degree

Celsius and those 15 meters above the ground surface was approximately between 36 and 40

degrees. The average maximum value for the entire burning period at the ground surface was

49.30C, at 5 meters above was 41.40C, at 10 meters above was 38.20C and at 15 meters above the

ground was 36.80C.

60

Figure 5.15: Maximum CO2 concentration at different heights all over the terrain for the entire simulation.

Maximum CO2 concentrations were found along the ground surface for entire period of

the simulation and those were found comparatively less as the height above the surface

increased. Figure 5.15 above shows the units of CO2 emissions as moles of CO2 produced per

mole of air.

The average of all the maximum values of each output quantity for the entire period of

simulation, i.e. 3000 seconds, was taken at different heights to compare with literature values.

Table 5.3 to show the literature emission ratios and mole fractions of CO2, CO and particulate

matters emitted by different forests. After literature search, the author found that very less

number of studies has been done to report the mole fractions of CO2, CO and particulate matters

and most of them were done to measure the emission factors instead. According to Waldrop and

co workers (2006), prescribed burning and its research have been done less in the eastern

61

hardwood region compared to the Western United States and the Southeastern Coastal Plain. So,

very few values were available for mole fractions of these forest fire exhaust quantities.

Averages of maximum values of CO2 for all hundred time steps (3000 seconds of

simulation) were calculated from the data obtained from the present simulation study. Those

were 2.2 × 10-3 mole CO2/mole air at ground level, 1.4 × 10-3 mole CO2/mole air at 5 meters above

the ground, 1.0 × 10-3 mole CO2/mole air at 10 meters above the ground and 9.0 × 10-4 mole

CO2/mole air at 15 meters above the ground.

Table 5.3: Literature values for CO2 mole fractions.

Value Units Comments Reference

3.5 × 10-1 mole CO2/mole airComplete combustion of forest fuels in

ideal conditions(Hardy et al., 2001)

4.5 × 10-4 mole CO2/mole air Simulation model output (Mason et al., 2001)

6.7 × 10-4 mole CO2/mole airWestern wild fires, firefighter exposure

at their breathing level(Reinhardt et al., 2000)

The mole fractions of CO2 ranged from 6.7 × 10-4 mole CO2/mole air to 3.5 × 10-1 mole

CO2/mole air in the literature referenced in Table 5.3. The values obtained in the present study lie

in this range. As shown in Figure 5.16 and Figure 5.17, as the height above the ground surface

increases the maximum concentration values decrease for the entire period of simulation. The

CO concentrations at 15 meters above the surface were almost zero.

62

Figure 5.16: Maximum CO concentration at different heights all over the terrain for the entire simulation.

Table 5.4: Literature values for CO mole fractions.

Value Units Comments Reference

2.0 × 10-4 mole CO /mole air Wild fires, near to fire line (McMahon et al., 1983)

1.0 × 10-6 mole CO /mole air Wild fires, within 30 meters (McMahon et al., 1983)

1.8 × 10-6 mole CO /mole air 150 meters above the fire, obtained by

modeling

(Trentmann et al., 2003)

7.2 × 10-6 mole CO /mole air Simulation model output (Mason et al., 2001)

3.9 × 10-5 mole CO /mole air Maximum CO exposure to

firefighters at their breathing level

(Reinhardt et al., 2000)

4.0× 10-7 mole CO /mole air Average CO exposure to firefighters

at their breathing level

(Reinhardt et al., 2000)

0.02 to 0.2 mole CO/mole CO2 Depending upon the type of fuel (Koppmann et al., 2005)

63

0.07 mole CO/mole CO2 Simulation model output (Mason et al., 2001)

The average maximum CO concentrations for 100 time steps for the entire terrain were

calculated. Those were 7.8 × 10-21 mole CO /mole air at ground level, 2.0 × 10-21 mole CO /mole

air at 5 meters above the ground, 3.2 × 10-22 mole CO /mole air at 10 meters above the ground

and 7.7 × 10-23 mole CO /mole air at 15 meters above the ground. shows that the literature values

for CO mole fractions ranged from 4.0× 10-7 mole CO /mole air to 2.0× 10-4 mole CO /mole air.

The values obtained in the present study were considerably low compared to the literature

values. Also, the molar ratio of CO to CO2 obtained in the present study was 3.58 × 10-16 at

ground level, 1.49 × 10-16 at 5 meters above the ground, 3.05 × 10-17 at 10 meters above the

ground and 8.54 × 10-18 at 15 meters above the ground. Those from the literature ranged from

0.02 to 0.2 mole CO/mole CO2 as shown in above. The FDS cannot predict CO production in

the smoldering phase of fire (McGrattan et al., 2008), so there are chances of less amount of CO

formation in the present simulation study.

Figure 5.17: Maximum particulate matters (soot) concentration at different heights all over the terrain for the entire simulation.

64

Soot produced has a unit in terms of milligrams of soot per cubic meter air. The average

values of the maximum concentrations of the soot decrease with increase in the height above the

ground surface as expected.

In the present study, soot values obtained were 29.72 mg/m3 at ground level, 19.29 mg/m3

at 5 meters above the ground, 14.60 mg/m3 at 10 meters above the ground and 12.76 mg/m3 at 15

meters above the ground. The smoke particulates are formed from the mass of the fuel burnt in

the forest fire (McGrattan et al., 2008), which are also be referred as particulate matter. The

literature soot (PM) concentration values ranged from 1.72 to 4.17 mg/m3 at almost 5 meters

above the ground as referenced in .

Table 5.5: Literature values for PM concentration.

Value Units Comments Reference

4.17 mg/m3 Maximum PM exposure to firefighters at their

breathing level

(Reinhardt et al., 2000)

1.72 mg/m3 Average PM exposure to firefighters at their

breathing level

(Reinhardt et al., 2000)

The results obtained in the present study did not match quantitatively with the literature

values because the production of the pollutants from the forest fires depends upon the exact fuel

type present in particular forest. The Arch Rock forest might be having specific soot yield data,

closed canopy confining the fire exhaust gases close to the ground. In the present study the

canopy of the tree branches is not employed. Also, the output concentrations of fire emitted

gases were averaged while performing the analysis in this study which could alter their higher

values.

65

Integrated exposure of the temperature and exhaust gases at different heights all over the

terrain

In the prescribed forest fires, fire starts at a place and then propagates towards unburned

fuel. At a particular spot, the heat and the exhaust gases exposure fluctuates as the fire flame

approaches, burns the fuel at that spot and then leaves it to burn the next. Thus, different places

must have different exposures to these fire emissions in accordance with the amount and type of

the fuel and the topography of that spot. Also, there can be some places where fire never existed

but the exhaust gases from the neighboring fire incidence may travel to that place with the wind.

So it was necessary to study such an integrated exposure of the emissions at all 10m×10m pixels

all over the terrain. The MATLAB code (see Appendix E) was used to achieve this. This is the

same code which was used to get maximum values of the emission with added arrangement to

estimate the integrated exposure as well from the CSV files. This was done to save the total run

time of the program. The MATLAB program generates a text file called

integrated_exposure_for_every_pixel.xls which contains temperature and concentration of the

CO2, CO and soot values for all the pixels and all 100 time-sequences.

A MATLAB code (see Appendix F), collected the output values of each output quantity

at the ground, 5 meters, 10 meters and 15 meters above the ground. This was done for all 100

time steps from the 100 CSV files. Trapezoid rule of integration was applied to get the

integrated exposure of every output quantity for each pixel (see Appendix F). All of these time

steps accrued 3000 seconds starting from 210th second and ending at the 3210 second which was

total burning period of time in area in the consideration at Arch Rock forest. This code also

generated the contour plots of the exposures of all the output quantities as shown below in Figure

5.18 through Figure 5.21

66

The data generated by the FDS model contained some anomalous values of

concentrations of all output quantities. There were some grid cells which possessed these

anomalous values of all output quantities for a particular time step. The ambient temperature was

defined to be 280C but, there were some instances where the ambient temperatures were 00C on

the grid cells. Also, the concentrations of CO, CO2 and soot had extreme values for some grid

cells compared to those for in the nearest neighbors. So, these anomalous values were removed

and replaced by the values equal to the values held by adjacent grid cells using a MATLAB code

(see Appendix F).

Figure 5.18: A 3D representation of the temperature exposure all over the terrain at different heights

As shown in Figure 5.18, heat or temperature exposure had higher values near the ground

surface and was lower above that. The unit of the temperature exposure is degree Celsius times

seconds (0C.s).

67

Figure 5.19: A 3D representation of CO2 exposure all over the terrain at different heights

Figure 5.19 and Figure 5.20 show the CO2 and CO exposure at different heights for the entire

simulation. According to these figures, the exposures of both CO2 and CO were higher at the

center part of the terrain and were very low at the outskirts. These exposures were measured in

terms of (mol/mol).s, (moles of the exhaust gas per unit mole of the air) times seconds.

Figure 5.21 below shows that the maximum exposure of the soot (PM) was found to be at

the ground surface and it decreases with height above the ground. The soot exposure was found

maximum at the center of the terrain. The units were (mg/m3.s) milligram of soot present in

cubic meter of air times second.

68

Figure 5.20: A 3D representation of CO exposure all over the terrain at different heights

Figure 5.21: A 3D representation of PM/Soot exposure all over the terrain at different heights

The integrated exposures of all of the output quantities were found to be the greatest at

the ground surface and were the least at the 15 meters above the ground all over the terrain from

69

the simulation for the entire period of the Arch Rock burning. The exposures at the edges of the

domain were found to be almost zero because the simulation model could have some internal

error in estimating the values at the edges.

70

CHAPTER 6 : DISCUSSION

The present study involved the simulation of a prescribed forest fire that occurred at Arch

Rock forest, including emission dispersion. Also, it included the estimation of the degree of heat

produced and the concentration of CO2, CO and PM within few meters of space above the fire. A

Fire Dynamics Simulator (FDS), a FORTRAN model, was used to serve the purpose.

The flow patterns and output values of the exhaust quantities matched to what was

expected on the basis of the fluid dynamics concepts and knowledge. The interest of the present

study was within 15 meters above the ground surface all over the terrain. The maximum

concentration values of the outputs and their exposure at four different heights such as ground, 5

meters, 10 meters and 15 meters above the surface were found higher at the ground surface and

decreased with the increase in the height above it as expected.

The heat data source used as an input in this study was in the form of raster file of 2m×2m

resolution. The terrain elevation data had 10m×10m resolution in the present study. So, the heat

data cells were added and averaged to make them 10m×10m resolution which could lose the

possible higher heat data values. The detailed small resolution of 2m×2m for the elevation

matched with that of the heat release data can produce more realistic emission concentration of

the output quantities from the fire.

The data from the simulation outputs were extracted into numbers using fds2ascii.exe code into

30 seconds interval. This data was averaged for each 30 seconds of period, which could have

induced the chances of losing higher emission values of the output quantities again. MATLAB

codes were used to generate plots showing the flow patterns and the exposures of different

pollutants from the fire.

71

The grid size in lower mesh in the computational domain was set as a 2m × 2m × 2m.

The smaller is the grid size, the larger is the accuracy and the computational cost. Because of

computational cost and other limitations the author could not try smaller resolution than used in

the present study which could affect the output results.

The soot yield value was provided as an input to FDS which is a fraction of smoke particles

formed from the given amount of the fuel. These values are specific to the specific vegetation to

be burnt, type of burning like wild or prescribed and weather conditions. In the present study an

average soot yield value, for the eastern hardwood region as a whole, was used which was

obtained from the literature data. So, the soot yield value specifically for the Arch Rock burning

was not used which could affect the resultant emissions. The product gases like CO2, CO were

predicted by the FDS using soot yield value (McGrattan et al., 2008).

Turbulence caused due to the density difference between the surrounding air and hot

smoke from the fire can have different pattern in the presence of the vegetation spread over the

forest floor and the canopy of the trees above. In this study, canopy and floor vegetation were not

employed. Thus, the resultant data values for output quantities from the simulation in this study

might differ from the realistic emission values from the actual Arch Rock forest fire. This study

was unable to estimate the total amount of emissions from the entire burning.

However, the data and the plots generated from this study provide fairly good estimation

of the extent of the heat release, concentration of exhaust gases and their exposure to the

surrounding life in the forest. One can get an idea of the maximum values of the emissions

before the actual fire is implemented. The FORTRAN input file developed in this study can be

applied to anywhere by changing the input parameters like forest floor elevation, heat release and

soot yield data pertaining to the location of the fire. This can be very useful for the forest officers

72

who want to know the harmful effects of the prescribed burning beforehand so that they can

modify the fuel load or can have the burning in specific weather conditions to reduce the

pollution.

73

CHAPTER 7 : CONCLUSION

A FORTRAN model was built which could simulate the emission and transport of the

heat, CO2, CO and particulate matters from prescribed fire in a typical eastern hardwood forests.

The pattern of the heat release in terms of temperature and output concentrations of the gaseous

pollutants were tested at three selected locations at ground level and at 5 meters, 10 meters and

15 meters above the ground. These emission concentration plots had exponentially dropping

nature with time. It was found that the maximum temperature and concentrations were greater at

the forest floor and decreased with the increase in height above the surface. The maximum

output values were compared with the available literature data but the comparison is complicated

because different forest fuels have different burning properties. However, CO2 and PM values

were within the range of published values. Also, the exposure of the output quantities was

plotted at different heights for every 10m × 10m area. Again, the maximum exposure was found

to be at floor and there was a decrease with increase in height above the forest floor.

Eastern hardwood region has lacked the smoke dispersion modeling studies compared to

rest of the US forests for both prescribed and wild fires (Waldrop et al., 2006). To date, FDS

approach has been used for the simulation of grassland fires only on flat terrain (Mell et al.

(2006). This study lays the foundation for using FDS to assist land managers and forest officers

to predict an extent of the harm that can be inflicted on the life in a vicinity of the fire. By

changing the inputs such as elevation data of the terrain and emission factor for the soot

formation, this model could be used to predict the emissions and their dispersion for any kind of

forest fire.

Future work should involve the effect of the forest floor vegetation and canopy of the

trees which was not applied in the present studies. This kind of arrangement affects the

74

turbulence created within this area and it might obstruct the dispersion of the smoke emitted

from the fire. This might rectify the deficiency in the concentrations and exposures of the output

quantities considerably. With the help of the powerful computers this can be achieved since it

causes much computational cost since more the details are fed to FDS more the computational

power is needed to simulate that scenario.

A small patch of few meters of the actual Arch Rock prescribed burning area was

considered in the present study. So, the burnt area can be extended to kilometers. This can be

achieved by using bigger elevation data (DEM) file. Computational cost needed is also bigger for

the enlarged area simulation. Also, grid cells size of the computational domain can be made even

smaller than used in this study so that more accurate and detailed resultant data can be obtained.

This can increase the scope of the study and give broad idea of the effect of the burning on

surrounding.

The heat release data fed to FDS in this study was with intervals of around 5 to 6

minutes. The heat release scenario between these intervals could have missed important burning

occurred in that period. If heat data with closer intervals is provided to FDS then it might

produce realistic temperature change and heat exposure around.

The prescribed burning took place in the Arch Rock forest which is the part of the eastern

hardwood region. General and average value of soot yield for the entire region was used, not

specific to Arch Rock forest. If exact soot yield data is used in the present model it might give

more accurate emission concentration and exposure of all the output quantities emitted from the

prescribed burning.

75

REFERENCES

Anderson, J. D. (1995). “Computational Fluid Dynamics: The Basics With Applications”, Science/Engineering/Math, McGraw-Hill Science. ISBN 0070016852.

AP-42, fifth edition, (1995). Enviornmental Protection Agency. “Compilation of air pollutant emission factors, volume I: stationary point and area sources”.

Baklanov, A. (2000). "Application of CFD Methods for Modeling in Air Pollution Problems: Possibilities and Gaps." Environmental Monitoring and Assessment, 65, 181-189.

Battye, W. and Battye,R. (2002). “Development of Emissions Inventory Methods for Wildland Fire”.

Blankenship, B. A. and Arthur, M. A. (2006). “Stand structure over 9 years in burned and fire-excluded oak stands on the Cumberland Plateau, Kentucky”. Forest Ecology and Management, 22(1-3), 134-145.

Chan, S. T. (2004). "Large eddy simulation of turbulent flow and dispersion in urban areas and forest canopies." Proc., UCRL-CONF-203600.

Chow, W. K. and Yin, R. (2004). "A New Model on Simulating Smoke Transport with Computational Fluid Dynamics." Building and Environment, 39, 611-620.

Graham, J. B. (2005). "Forest fuel and fire dynamics in mixed-oak forests of south eastern Ohio". Ohio University.

Hardy, C. C., Ottmar, R. D., Peterson J. L, Core J. E. and Seamon, P. (2001). “ Smoke management guide for prescribed and wildland fire.”

Hess, G. D., Tory, K. J., Lee, S., Wain, A. G. and Cope, M. E. (2006). "Modeling the King Island bushfire smoke." Australian Meteorological Magazine, 55(2), 93-103.

Holmes, N. S. and Morawska, L. (2006). "A Review of Dispersion Modelling and its Application to the Dispersion of Particles: An Overview of Different Dispersion Models Available." Atmospheric Environment, 40, 5902-5928.

Hugget, C. (1980). "Estimation of Rate of Heat Release by Means of Oxygen Consumption Measurement." FIRE AND MATERIALS, 4(2), 61-65.

Katul, G. G., Mahrt, L., Poggi, D. and Sanz, C. (2004). "One-and Two-Equation Models for Canopy Turbulence." Boundary-Layer Meteorology, 113, 81-109.

Koppmann, R., Czaplewski, K. and Reid, J.S. (2005). “A review of biomass burning emissions, part I: gaseous emissions of carbon monoxide, methane, volatile organic compounds, and nitrogen containing compounds”. Atmos. Chem. Phys. Discuss., 5, 10455–10516.

76

Lavoue, D. D., Gong, S. and Stocks, B. J. (2007). "Modeling emissions from Canadian wildfires: A case study of the 2002 Quebec fires." International Journal of Wildland Fire, 16,(6), 649-663.

Lawrence, T. (2007). "Emission Factors." Encyclopedia of Earth.

Lemieux, P. M., Lutes, C. C., and Santoianni, D. A. (2004). "Emissions of organic air toxics from open burning: a comprehensive review". Progress in Energy and Combustion Science, 30, 1-32.

Liu, Y. (2004). "Variability of wildland fire emissions across the contiguous United States". Atmospheric Environment, 38(21), 3489-3499.

Lopes da Costa, J. C., Castro, F. A., Palma, J. M. and Stuart, P. (2006). "Computer Simulation of Atmospheric Flows Over Real Forests for Wind Energy Resource Evaluation." Journal of Wind Engineering and Industrial Aerodynamics, 94, 603-604.

MacGregor, D. G. (2004). "An Inventory of Models, Tools, and Computer Applications for Wildland Fire Management". Wildland Fire Models Inventory.

Mason, S. A., Field, R.J., Yokelson, R. J., Kochivar, M. A., Tinsley, M. R., Ward, D. E. and Hao, W. M. (2001). “Complex effects arising in smoke plume simulations due to direct emissions of oxygenated organic species from biomass combustion.” Journal of Geophysical Research, 106(D12), 12527-12539.

McGrattan, K. B., Baum, H. R. and Rehm, R. G. (1996). "Numerical Simulation of Smoke Plumes from Large Oil Fires." Atmospheric Environment, 30(24), 4125-4136.

McGrattan, K. B., Baum, H. R. and Rehm, R. G. (1998). "Large Eddy Simulations of Smoke Movement." Fire Safety Journal, 30,161-178.

McGrattan, K., Baum, H. and Rehm, R. (2008). "Fire Dynamics Simulator (Version 5) Technical Reference Guide.".

McMahon, C. K.,(1983). “Characteristics of forest fuels, fires and emissions”. For Presentation at the 76th Annual Meeting of the Air Pollution Control Association Atlanta, Georgia.

Mell, W. E., Manzello, S. L. and Maranghides, A. (2006). "Numerical modeling of fire spread through trees and shrubs." Proc., V International Conference on Forest Fire Research.

Mell, W., Charney, J. J., Jenkins, M. A. cheney, P., and Gould, J. (2005). "Numerical simulations of grassland fire behavior from the LANL - FIRETEC and NIST - WFDS models." Proc. EastFIRE Conference, George Mason University, Fairfax, VA, 1-10.

Mell, W.E., Jenkins, M.A., Gould J. and Cheney, P. (2007). “A Physics-Based Approach to Modeling Grassland Fires”. International Journal of Wildland Fire.

77

Patnaik, G., Boris, J. P. and Grinstein, F. F. (2003). "Large scale urban simulations with the miles approach." Proc., 16th AIAA Computational Fluid Dynamics Conference, Orlando, Florida, 1-13.

Qiu, J., Gu, Z. L. and Wang, Z. S. (2008). "Numerical Study of the Response of an Atmospheric Surface Layer to a Spatially Nonuniform Plant Canopy." Boundary-Layer Meteorol, 127, 293-311.

Radke, L. F., Lyons, J. H., Hobbs, P. V., Hegg, D. A., Sandberg, D. V. and Ward, D. E. (1990). “Airborne monitoring and smoke characterization of prescribed fires on forest lands in western Washington and Oregon: Final report”. Gen. Tech. Rep. PNW-GTR-251.

Reinhardt, T. E. and Ottmar, R. D. (2000). “Smoke exposures at western wild-fires.” Res. Pap. PNW-RP-525.

Riebau, A. R. and Fox, D. (2001). "The new smoke management". The International Journal of Wildland Fire, 10(3-4), 415-427.

Saarikoski, S., Sillanpaa, M., Sofiev, M., Timonen, H., Saarnio, K., Teinila, K., Karppinen, A., Kukkonen, J. and Hillamo, R. (2007). "Chemical composition of aerosols during a major biomass burning episode over northern Europe in spring 2006: Experimental and modeling assessments." Atmospheric Environment, 41(17), 3577-3589.

Su, H., Shaw, R. H., U, K., Moeng, C. and Sullivan, P. P. (1998). "Turbulent Statistics of Neutrally Stratified Flow within and Above a Sparse Forest Forest from Large Eddy Simulation and Fiels Observations." Boundary-Layer Meteorology, 88, 363-397.

Sun, R., Jenkins, M. A., Krueger, S. K., Mell, W. and Charney, J. J. (2006). "An Evaluation of Fire-Plume Properties Simulated with the Fire Dynamics Simulator (FDS) and the Clark Coupled Wildfire Model." Canadian Journal of Forest Research, 36, 2894-2908.

Sutherland, E. K. and Hutchinson, T. F. (2003). "Characteristics of mixed-oak forest ecosystems in southern Ohio prior to the reintroduction of fire." Gen. Tech. Rep. NE-299. Newtown Square, PA: U.S. Department of Agriculture, Forest Service, Northeastern Research Station, 1-159.

Trelles, J., McGrattan, K. B. and baum, H. R. (1999). "Smoke Dispersion from Multiple Fire Plumes." American Institute of Aeronautics and Astronautics Journal, 37(12), 1588-1601.

Trentmann, J., Andreae, M. O. and Graf, H. (2003). “Chemical processes in a young biomass-burning plume.” J. Geophys, Res., 108(D22), 4705.

Vautard, R., Ciais, P., Fisher, R., Lowry, D., Breon, F. M., Vogel, F., Levin, I., Miglietta, F. and Nisbet, E. (2007). "The dispersion of the Buncefield oil fire plume: An extreme accident without air quality consequences." Atmospheric Environment, 41(40), 9506-9517.

Waldrop, T. A., Budnak, L., Phillips, R. J. and Patrick, H. B. (2006). "Research efforts on fuels, fuel models, and fire behavior in eastern hardwood forests." Proc., Fire in Eastern Oak Forests:

78

Delivering Science to Land Managers. Proceedings of a Conference, USDA Forest Service, Southern Research Station, 90-103.

Wiedinmyer, C., Quayle, B., Geron, C., Belote, A., McKenzie, D., Zhang, X., O'Neill, S. and Wynne, K. K. (2006). "Estimating emissions from firest in North America for air quality modeling", Atmospheric Environment, 40 3419–3432.

Witham, C. and Manning, A. (2007). "Impacts of Russian biomass burning on UK air quality." Atmospheric Environment, 41,(7), 8075-8090.

Y. Liu (2004). "Variability of wildland fire emissions across the contiguous United States", Atmospheric Environment, 38, 3489-3499.