. .ux.

$

. t.
w... . :,
is? A... 39:35». sum,

. s I
kgfiifi
i‘t.’

I
\
.

.. ”5mm?
..

.
.«v

.FIIJ. . 3

. .. a. 3:23
2m. 3 L.

{Wag SW“!
..
§ ‘

“unadwfiwm:
. «

 

 

 

 

 

m LIBRARY

' 2 Michigan State
W University

This is to certify that the
thesis entitled

INVESTIGATION INTO THE ABILITY OF THE BLUESKY
SMOKE MODELING FRAMEWORK IN SIMULATING SMOKE
IMPACTS FROM WILDFIRES

presented by

Lesley Adele Fusina

has been accepted towards fulfillment
of the requirements for the

MS. degree in Geolaphy

 

 

49%“ 942'

/ Major Professor’s Signature

Mal/‘1 Q; , 3’02?

Date

 

MSU is an affirmative-action. equal-opportunity employer

-.-c-l-I-O-t-I-I-n-c-I-I-o-I-o-I-I-o-1--n-u-l---o-o-o-o-n-I-n-

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

5/08 KzlProyAchreyCIRC/Dateoue.indd

INVESTIGATION INTO THE ABILITY OF THE BLUESKY SMOKE MODELING
FRAMEWORK IN SIMULTAING SMOKE IMPACTS FROM WILDFIRES

By

Lesley Adele Fusina

A THESIS
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
MASTER OF SCIENCE
Department of Geography

2008

ABSTRACT

INVESTIGATION INTO THE ABILITY OF THE BLUESKY SMOKE MODELING
FRAMEWORK IN SIMULTAING SMOKE IMPACTS FROM WILDFIRES

By

Lesley Adele Fusina

Wildland fires are necessary for regeneration and stimulation of soils and plant
growth as well as for forest management practices. In the United States, increased fire
suppression and prescribed burn activities have decreased the number of total fires over
time but, instead of seeing a reduction in fire size as well, fires are becoming larger.
These large fires have the potential to release substantial amounts of smoke which can
lead to poor air quality and reduce visibility. The BlueSky Smoke Modeling Framework
is designed to assist fire and air quality managers in predicting the timing, location and
magnitude of smoke impacts from wildland fires. In this study, large wildfire episodes in
California were used to examine BlueSky’s ability to accurately predict timing and
location of wildfire smoke impacts. This analysis is necessary as BlueSky is being used
more and more as a one-stop shopping tool for predicting smoke concentration and
emissions across the United States, but the accuracy and uncertainties in BlueSky
predictions are still largely unknown. This study found BlueSky able to highlight areas
of potential smoke impact by reliably predicting long-range transport of wildfire smoke
plumes. The magnitudes of potential smoke impacts were less reliable with accurate
predicted surface PM2,5 concentrations occurring infrequently and only when observed

magnitudes were greater than 10 pg m'3.

Copyright by
Lesley Adele Fusina
2008

ACKNOWLEDGEIVIENT S

The completion of this thesis would not be possible without the guidance, support
and help from some very important people. First and foremost, to my advisor, Dr.
Shiyuan Zhong, thank you for the opportunity to pursue this research and the constant
editing and support along the way. To Xindi Bian, without whom, I would still be crying
in my office trying to get the model to run! Your help will forever be appreciated. I
would also like to thank Wenqing Yao, your expertise in MATLAB, willingness to help,
and analysis suggestions helped to make this project a success. Importantly, much thanks
is given to the USDA Forest Service Pacific Northwest AirFire Team, especially Robert
Soloman, Tara Strand and Sim Larkin, without your expertise and vision, none of what I
achieved would have been possible. Additionally, I’d like to thank Dr. Jay Charney for
his revisions, knowledge and challenging questions, you certainly kept me on my toes.
To the rest of my committee members, Dr. Julie Winkler and Dr. Jeff Andresen, you’ve
been with me six years now and have always looked out for my best interest. Without
your help I would not be where I am today. Finally and most importantly, to my family
and friends. Your support and words of encouragement are the reason this thesis is
completed. I will never be able to repay you for your constant energy, patience and

ability to say just the right thing at just the right time.

iv

TABLE OF CONTENTS

List of Tables ........................................................................................ vi

List of Figures ..................................................................................... viii

PART I. INTRODUCTION ........................................................................ 1

Thesis Objective ............................................................................ 3

PART I]. BACKGROUND ......................................................................... 6

Literature Review ........................................................................... 6

Smoke Dispersion Modeling ...................................................... 6

The BlueSky Smoke Modeling Framework .................................... 17

Model Description ............................................................................................... 23

PM; 5 Concentrations and Smoke Trajectory Simulations 23

Summary .................................................................................... 27

PART III. NORTHERN CALIFORNIA, AUGUST 2006 .................................... 28

Introduction ................................................................................. 28

Methods ..................................................................................... 30

Study Area ......................................................................... 30

Observational Data ............................................................... 32

Satellite Observations ............................................................ 33

Model Simulations ............................................................... 36

Results and Discussion ................................................................... 37
Wildland Fires and Synoptic Weather Conditions during the Study

Period .............................................................................. 37

Meteorological Fields ........................................................... 40

Smoke Plume and PM 2, 5 Prediction ............................................ 49

Summary and Conclusions ............................................................... 55

PART IV. SOUTHERN CALIFORNIA, OCTOBER 2007 .................................. 58

Introduction ................................................................................. 58

Methods ..................................................................................... 61

Study Area .......................................................................... 61

Model Setup and Simulations .................................................... 63

Surface and Satellite Observations ............................................. 65

Results and Discussion ................................................................... 75
Wildland Fires and Synoptic Weather Conditions during the Study

Period .............................................................................. 75

Meteorological Fields ............................................................ 77

Smoke Plume andPszPredictions 86

Summary and Conclusions ............................................................... 97

PART V. CONCLUSIONS ..................................................................... 100

Summary .................................................................................. 100
Study Limitations ........................................................................ 103
Future Work .............................................................................. 105
REFERNCES ...................................................................................... 108

vi

LIST OF TABLES

2.1 Temporal allocation of acres burned in a day for wildfires. Data developed

by the Western Regional Air Partnership (from WGA/W RAP, 2005) 26
3.1 Geographic station information for meteorological and PM25 observations 35
3.2 BlueSky Input Information .................................................................. 37
3.3 Meteorological comparison statistics for surface temperature, mixing ratio

and wind speed ................................................................................ 46
4.1 Meteorological observation station geographic information ............................ 68
4.2 PM25 observation station geographic information ........................................ 72

4.3 Meteorological comparison statistics for surface temperature, mixing ratio
and wind speed ................................................................................ 82

vii

LIST OF FIGURES

2.1 Fire modeling includes fire environment, fire characteristics, first-order fire
effects and second-order fire effects. Examples are given for each category
(modified from Andrews et al, 2001; pg. 345) ............................................... 8

2.2 Elements of smoke dispersion models (from Breyfogle and Ferguson, 1996;
pg 2) ............................................................................................... 9

2.3 Image analysis (left) of smoke plume on night of March 20, 1997 and PB-
Piedmont simulated smoke (right) at (a) 2150 LST, (b) 2215 LST, (c) 2255
LST and (d) 2354 LST (modified from Achtemeier, 2006; pg. 92) .................... 12

2.4 Schematic of modeling framework. Dotted arrows indicate interactions which
are turned off or for which default values are assumed (from McKenzie et al,
2006; pg 281) .................................................................................. 14

2.5 Observed and predicted PM25 concentrations (left) a monitoring site in Idaho
and the corresponding predicted plume concentration contours (right) during
August 25, 2006 (from Jain et al, 2007; pg. 6756) ....................................... 16

2.6 Maximum average and ground concentrations for the Rex Creek wildfire for
the period of August 19-26, 2006 when the fire is (a) treated as a single fire,
(b) split into 5 fires and (0) split into 10 fires. In all cases the total area burned
is the same (from Larkin et al, 2007) ....................................................... 21

2.7 Components of the BlueSky modeling system ............................................ 24

3.1 Terrain map of northern California and southern Oregon, including locations
of observational meteorological and PM” stations. The black box indicates
the simulated 4km domain ................................................................... 31

3.2 Plots of 500 hPa height (left) and wind speed (right) synoptic plots at 12Z (0400
PST). Light shading in the wind speed panels (right) indicates weaker 500
hPa wind speeds while darker shading indicates stronger wind speeds.
Contours in the left panel represent 500 hPa isoheights ................................. 39

3.3 Simulated and observed hourly temperature (top), dewpoint temperature
(center), and wind speed (bottom) at all fifteen surface meteorological stations
used within this study. Gray dots are the day time values while the black dots
are night time values .......................................................................... 41

 

l Note that some images in this thesis are presented in color

viii

3.4 Time series plots of surface temperature, wind speed and wind direction at
three locations which show (a) the typical MM5 simulated pattern, (b) the
pattern observed at stations located along the west coast, and (c) the pattern
seen in the northern stations located in diverse terrain. Black dotted lines and
upside down triangles (in wind direction time series) represent the observations
while gray lines and x’s (in wind direction) represent MM5 simulated surface

meteorological conditions ..................................................................... 43

3.5 Upper level patterns at 002 (1600 PST) Medford, Oregon station at three
different dates (a) August 21, 2006, (b) August 29, 2006, and (c) August 31,
2006. Variables examined are theta, mixing ratio, wind speed and wind
direction; all plotted against height. Black dotted lines and upside down
triangles (wind direction) represent observations while gray lines and x’s

(wind direction) represent MM5 simulated meteorological conditions

3.6 The images across the top show BlueSky visual output of surface PM25
concentrations for three days (a) August 21, (b) August 29 and (c) August 31.
The middle images show the corresponding satellite data taken from MODIS
hazard mapping system (HMS), where gray shading indicates significant smoke
plumes. Darker shades of gray represent higher concentrations of smoke. The
bottom images are the aerosol optical depth data for each day, also taken from
MODIS. Warmer colors indicate a higher concentration of aerosol within the

atmosphere .....................................................................................

3.7 Hourly PM2_5 concentrations at the 8 monitoring stations located in northern
California and southern Oregon. Solid black lines represent the observed
adjusted concentrations of PM2_5 while the gray dotted lines indicate the

BlueSky estimated PM2_5 concentrations ....................................................

4.1 Modeling domains used within this study. D01 is the 36km domain, D02
is the 12km domain and D03 is the 4km domain. The 12km and 4km domains

are one-way nested domains ................................................................

4.2 Hourly data from 75 surface meteorological stations and twice daily soundings
from the 6 upper air stations were used to validate MM5 predictions during

October 15 — 30, 2007. The black box indicates the 4km one-way nested domain
used for the simulations. Black circles indicate surface stations while red

triangles indicate an upper air station .......................................................

4.3 Surface particulate matter (PM2_5) monitoring stations where hourly PM25 data
was obtained from the AIRN ow monitoring sensors (circles) and the
California Air Resources Board (ARB) monitoring sensors (stars). The

black box indicates the 4km one-way nested domain used for the simulations

ix

48

.50

.53

.64

.67

71

4.4 Plots of 500 hPa height (left) and wind speed (right) synoptic plots at OOZ
( 1600 PST). Light shading in the wind speed panels (right) indicates weaker
500 hPa wind speeds while darker shading indicates stronger wind speeds.
Contours in the left panel represent 500 hPa isoheights ................................. 76

4.5 (a) Simulated and observed hourly temperature (top), mixing ratios (center),
and wind speed (bottom) at all surface meteorological stations. Each circle
indicates an hourly daytime average (0700 PST to 1900 PST) ......................... 78

4.5 (b) Simulated and observed hourly temperature (top), mixing ratios (center),
and wind speed (bottom) at all surface meteorological stations. Each circle
indicates an hourly nighttime average (2000 PST to 0600 PST) ....................... 79

4.4 Comparison of upper level profiles at 1600 PST (OOZ) at the San Diego,
CA rawinsonde site. Black circles represent observations. Light gray solid
lines and circles (wind direction) represent MM5 simulated meteorological
conditions initialized with Eta output. Darker gray dashed lines and circles
(wind direction) represent MM5 simulations initialized with NARR output .......... 85

4.7 (a) Satellite images taken from the MODIS Hazard Mapping System (HMS)
for October 24 — 27, 2007. Gray shading indicates significant smoke plumes.
Darker shades of gray represent higher concentrations of smoke.
Corresponding BlueSky images are total PM2,5 concentrations for the same
days for the hours of 1300 PST to 2300 PST for (b) Eta MM5 initialization
data and (c) NARR MM5 initialization data. Warmer colors indicate higher
concentrations of PM” ........................................................................ 87

4.8 (a) MODIS/GASP Aerosol Optical Depth (AOD) images at 1500 PST
(23002) for October 24 - 27, 2007 with corresponding BlueSky simulated
smoke plumes for (b) Eta MM5 initialization data and (c) NARR MM5
initialization data. Warmer colors in the AOD images indicate areas of higher
aerosol concentrations. Warmer colors within the BlueSky images indicate
higher levels of surface PM25 concentration ............................................... 89

4.9 (Left) Eta simulated and observed local time rate change of PM25
concentrations at four monitoring sites within the domain. Black lines
represent observed PM2_5 and colored lines represent BlueSky simulated
PM25 concentrations. (Right) Corresponding surface meteorological
stations. Black triangles represent observations while gray lines and stars
represent MM5 simulated temperature (Temp), wind speed (W spd), and
wind direction (W dir) .......................................................................... 92

4.10 (Left) Eta simulated and observed local time rate change of PM”
concentrations at four monitoring sites within the domain. Black lines
represent observed PM25 and colored lines represent BlueSky simulated
PM2_5 concentrations. (Right) Corresponding surface meteorological
stations. Black triangles represent observations while gray lines and stars
represent MM5 simulated temperature (Temp), wind speed (Wspd), and
wind direction (Wdir) ............................................................................ 94

xi

PART I

INTRODUCTION

Fire is an important ecological process that dramatically affects both forested and
non-forested ecosystems around the world. Naturally occurring fires are an integral
component of the carbon cycle. They promote regeneration of threatened/endangered
habitats, stimulate growth of many plant species, and also aid state and local agencies in
management efforts of forest undergrowth (i.e. wildfires reduce dead and downed
material to make room for new growth). Recently, humans have influenced these natural
fire regimes through management strategies, such as prescribed burning. These strategies
decrease the number of large catastrophic wildfires, but increase the overall size of
wildfires (Minnich 1987; Odion et al. 2004). Wildfires also present a major threat to
human populations who live near forested regions where the loss of home, property and
life can be a direct result of uncontained wildfires. Secondary threats, such as decreased
visibility and poor air quality also arise due to smoke emissions from the burning of
biomass. The nature of wildland fires and their impacts on ecosystems and the people
living in or near ecosystems complicates many aspects of fire management decisions.

Tools available for fire managers to help balance often conflicting management
goals are prescribed burns and Wildland Fire Use (W FU) fires. Prescribed fires are those
set for a specific management purpose, most often to remove down and dead
undergrowth to reduce the risk of catastrophic wildfires or to develop or maintain habitat
for threatened and endangered species. WFU fires are wildfires that ignite naturally and
are allowed to burn under strict monitoring without suppression in locations where a fire

is an important component of the ecosystem. If a WFU fire becomes a threat to a

community or begins to grow out of control, fire suppression activities begin. While the
prescribed and WFU fires are tools that fire managers can use to mitigate the direct
threats of wildfires, secondary threats due to smoke release must also be addressed.

Most of the wildfires in the United States occur in western states, which also
include some of the worst air quality regions in the country (Association 2007). 16 of the
25 most polluted counties and over half of the US. cities cited on the worst ozone and
particulate matter (PM) concentration lists are located in California. Considering
southern California’s synoptic circulation patterns, large population, and air quality
degradation as a result of smoke contributions, California is often at risk of exceeding the
national air quality standards set by the EPA.

Smoke constituents include gases, such as carbon monoxide, carbon dioxide,
methane, and particulate matter (PM). The PM component of smoke is responsible for
most of the smoke-induced undesirable physiological effects that occur within our body
(Rapp 2006). PM” is particulate matter that is less than or equal to 2.5 microns in
diameter. Due to its small size, PM25 passes readily into the human respiratory system.
Those most affected by increased levels of PM2_5 are children, the elderly and those with
weakened immune systems. Health effects associated with increased levels of PM25
include eye and throat irritation, coughing, reduced lung function, blocked and runny
noses, and in extreme cases, mortality (EPA 2003). In addition to the health impacts, PM
can also greatly decrease visibility both locally and regionally (Malm, Day, and
Kreidenweis 2000). Smoke impacts vary depending on fire location, types of fuel, and

current and past weather conditions.

The BlueSky Smoke Modeling Framework is a tool that is used to aid smoke
management of wildland fires (prescribed fires and wildfires) (Ferguson, Peterson, and
Acheson 2001). Developed by the USDA Forest Service AirFire Team under the
National Fire Plan, BlueSky links together five component models (fuel load, fire
behavior, emissions, smoke dispersion and meteorology) to simulate the cumulative
impacts of smoke in the form of PM concentration. BlueSky output is employed by
operational fire managers and air quality regulators to help make ‘go’ and ‘no-go’
decisions related to prescribed burns, to assess the status of a WFU fire, and to support
wildfire suppression strategies. While BlueSky is widely used by fire managers and air
quality regulators, quantitative studies of BlueSky predictions are limited. The goal of
this study is to investigate BlueSky’s predictive capacity by simulating the smoke
impacts from two extreme wildfire events that occurred in California in August, 2006 and
October, 2007 and compare the BlueSky PM predictions with in-situ and remotely sensed

observational data.

Thesis Obiective

This thesis is composed of two case studies. The first case study is concerned
with an outbreak of fires that occurred in northern California during late summer of 2006.
The first goal of this case study is to evaluate the ability of the BlueSky Smoke Modeling
Framework to accurately simulate smoke plume trajectories and PM25 concentrations
during this outbreak. Secondly, since northern California is characterized by varying and
rough topography, the evaluation results will contribute to our knowledge of BlueSky

ability to deal with varying terrain when simulating smoke movement, as BlueSky’s

performance under those conditions is not well known. The second case study simulates
wildfires that occurred in southern California in October of 2007. This case study is a
first step at evaluating individual model components to assess how changes in these
components affect the smoke plume trajectories and surface concentrations. The
meteorological model used in BlueSky is evaluated using two different initialization data
sets to see how different meteorological conditions affect the simulated smoke plumes.
This evaluation will help assess the sensitivity of the BlueSky predictions to inputted
meteorological data. The first dataset is the National Center for Environmental
Prediction (NCEP) 40 km Eta data series and the second is the NCEP’s North American
Regional Reanalysis (NARR) 32 km data set. The BlueSky Smoke Modeling Framework
is only as good as its individual components and the subsequent input information each
provides. For example, the meteorological model provides predictions of surface and
upper level wind speeds and directions, which are then inputted into the dispersion and
trajectory models. If the wind speeds and/or directions are not accurate then one cannot
expect the BlueSky predicted smoke plume trajectories and locations to be accurate.
Analysis of each component’s reliability and accuracy is necessary so that improvements
can be made to each component, and subsequently to BlueSky predictions as well. In this

study the following research questions will be addressed:

0 Is the BlueSky Smoke Modeling Framework able to accurately predict wildfire

smoke trajectories and timing of smoke impacts?

0 Does BlueSky accurately predict surface PM2,5 concentrations?

0 Are BlueSky smoke plume and concentration predictions sensitive to the input

data used within the meteorological model?

As stated above, this project helps to fill an information void that exists in the
accuracy and sensitivity of the BlueSky Smoke Modeling Framework. In depth studies
of model performance are missing from the literature. BlueSky has not been validated
for different regions of the United States, since most existing literature focuses mainly on
the Pacific Northwest region where BlueSky was initially developed. This project will
provide results of BlueSky’s predictive capabilities in regions of varying topography
(northern California) and dense population (southern California).

Background information including a literature review of smoke dispersion
modeling and the BlueSky framework and model descriptions are provided in Part H.
Parts III and IV describe the case studies outlined previously. Part V summarizes the
overall research conclusions and implications, study limitations and suggestions for

future work within this research area.

PART II
BACKGROUND
Literature Review
The following section reviews the published literature on smoke dispersion
modeling. The first part focuses on research and issues associated with smoke dispersion
modeling. Next, literature and evaluation efforts of the BlueSky Smoke Modeling

Framework will be presented in the second part.

Smoke Dispersion Modeling

Fire effects, defined by Reinhardt et al (2001), are the results of the combustion
process. Fuel consumption, plant mortality, soil heating, erosion, nutrient cycling,
vegetation succession and smoke production and dispersion are some important fire
effects. Smoke dispersion is considered a second-order fire effect, one that occurs over
longer time frames (days to months) and is located greater distances from the fire
(Reinhardt, Keane, and Brown 2001; Andrews and Queen 2001). Secondary fire effects
are modeled from, and dependent upon, inputs from the fire environment, fire
characteristics and first order fire effects models (those effects measured within a few
days and appear in a close proximity to the fire) (Figure 2.1). Because of this dependent
nature, secondary fire effect’s are difficult to model (Reinhardt, Keane, and Brown 2001).

Fire effects and behavior models are useful when assessing the risks from wildfire
smoke to short and long-term air quality, ecosystem health and risks to surrounding
communities. Although the terms are sometimes used interchangeably, fire behavior

models and fire effects models are different with respect to their data inputs. Fire

behavior models provide a description of qualities of fuels that contributes to fire spread
near the flaming front of the fire (Reinhardt, Keane, and Brown 2001). These fuels
dictate the direction, rate and area of fire spread. While fire effects models also use
knowledge of how the fire is spreading, the quantitative nature of these models also
considers what remains burning after a fire front has passed (Reinhardt, Keane, and
Brown 2001). In both types of modeling, an evaluation of the way fuel burns (Andrews
and Queen 2001) and the associated impacts of each burn pattern is analyzed.

To predict smoke emissions and dispersion, estimates of smoke production
(considered a first-order effect), terrain data and meteorology are necessary (Reinhardt,
Keane, and Brown 2001). Of these, McGrattan (2003) indicates that terrain height and
the mixing layer depth are most important for predicting wildfire smoke impacts. While
this may be true, it is also essential that dispersion models integrate fire progression, fire
emissions, atmospheric flow, smoke dispersion and chemical reactions (Miranda 2004).
Figure 2.2 illustrates the basic components of a smoke model. Smoke dispersion models
created specifically for wildland biomass burning include: SASEM (Sestak and Riebau
1988), VSMOKE (Lavdas 1996), CALPUFF (Scire, Strimaitis, and Yamartino 2000),
and TSARS plus (Hummel and Rafsnider 1995). These models vary in complexity and
have differing advantages and disadvantages. Breyfogle and Ferguson (1996) conducted
a detailed evaluation of these smoke dispersion models. The following is a brief

overview of their findings, highlighting the limitations in smoke dispersion modeling.

 

FIRE ENVIRONMENT

Pre-fire conditions
Fuel type, description
Fuel condition, moisture content
Weather - wind, temperature, etc.
Terrain- slope, elevation, aspect

1

FIRE CHARACTERISTICS

Processes during the fire
Ignition
Extinction
Fire state - flaming/smouldering
Intensity
Rate of spread
Fuel consumption
Emissions - gas and particulates

 

 

 

 

 

 

 

 

 

 

1*"t ORDER FIRE EFFECTS

Local - measurable within

a few days after the fire,

restricted to burn area
Reduction in fuel loading
Exposure of mineral soil
Mortality/injury to vegetation
Local air quality
Chemical and phsyical response
of fire-heated soils

~11

2nd ORDER FIRE EFFECTS

Removed from the fire area
and/or resulting after a

longer time delay

Erosion

Smoke transport/dispersion
Wildlife habitat change
Economic impact

Visual change of landscape

 

 

 

 

 

 

 

 

 

Figure 2.1 Fire modeling includes fire environment, fire characteristics, first—order fire
effects and second-order fire effects. Examples are given for each category (modified
from Andrews et al, 2001; pg. 345).

 

 

   

  

Ignition

 

 

 

 

  

 

 

 

 

 

 
   

 

 

 

 

 

 

 

 

 

 

 

 

 

Pattern
/°
' Loading
Species
Components
Source L I
Strength W mar
Plume Rise Mixing Height
Plume
Species _ .
Concentrations V's'bimll

 

 

 

Figure 2.2 Elements of smoke dispersion models (modified from Breyfogle and
Ferguson, 1996; pg 2).

CALPUFF, a multilayer, multispecies, non-steady state Gaussian puff dispersion

model, and TSARS plus, a complex terrain wind-field model driving gaussian puffs for
pollutant dispersion, account for complex terrain by interpolating observations to a 3-D

grid (Breyfogle and Ferguson 1996). CALPUFF performed well in Breyfogle and

 

Ferguson’s (1996) evaluation for a variety of factors: it has many different alternatives
for model input and it can simulate an unlimited number of burns over an unlimited area
(ideal for large-scale planning activities). CALPUFF, however, has only two options for
calculating dispersion, both of which are based on Gaussian approximations, which may
make it difficult to research all types of biomass burning. TSARS plus has similar model
components as CALPUFF, but unlike CALPUFF, TSARS plus includes five different
dispersion calculations that enable modeling of different types of biomass burning.
Unfortunately, TSARS plus is domain dependent, meaning it can be applied to Wyoming
only. Also, the usefulness of TSARS plus in large-planning efforts is hampered by the
limit to the number of fires allowed to burn at one time. The major disadvantage and
limitation of both of these models is that the difficult configuration and tedious input
requirements often deter those with limited computer proficiency.

VSMOKE, a dispersion model only that was originally created to estimate
potential visibility reduction for motorists, and SASEM, a screening tool for prescribed
burns, utilize a simpler approach than CALPUFF and TSARS plus, by facilitating their
use to anyone with limited to no computer experience (Breyfogle and Ferguson 1996).
Both models can simulate one burn at a time, which is useful in screening individual
burns. The simplistic nature of VSMOKE and SASEM, however, is also a limitation, as
both models are restricted to areas where the terrain is relatively flat and they rely upon
weather-observation data input directly by the user, compared to meteorology inputs
estimated from a numerical model. These models are continually being modified so that

improvement is made in both the inputs requirements from the user, as well as, in the

10

outputs of smoke plume and concentration predictions themselves (Breyfogle and
Ferguson 1996).

In efforts to improve smoke dispersion models, many different frameworks have
been tested. One such effort was to create a model that is more regional or location
specific. Achtemeier (2005) describes a fine scale, smoke tracking model called Planned
Bum-Piedmont (PB-Piedmont) valid for regions within the Piedmont plateau in the
south-east US from Maryland to Alabama. The Piedmont is a low plateau roughly 100-
300 km wide and 1500 km that separates the Atlantic Coastal Plain and Gulf Coastal
Plain from the Appalachian Mountains. PB-Piedmont outputs high resolution space and
time predictions of smoke movement within shallow layers at the ground. While PB-
Piedmont is designed to work only when weather conditions are conducive to smoke
entrapment (ie. at night with clear skies and light winds) and over certain geographical
regions, it is capable of accurately capturing and simulating the movement of smoke
plumes (Figure 2.3). This model however, is only concerned with simulating the
movement of smoke plumes and not the smoke concentrations (Achtemeier 2005). PB-
Piedmont exemplifies the usefulness and limitations of creating a model that is location
specific. The location specific nature of the model also allows for the use of more
detailed terrain, weather and fuels input data, producing small errors in smoke movement

predictions and thus increased reliability in impact assessments.

11

—1.0 —0.5
km

 

Figure 2.3 Image analysis (left) of smoke plume on night of March 20, 1997 and PB-
Piedmont simulated smoke (right) at (a) 2150 LST, (b) 2215 LST, (c) 2255 LST and (d)
2354 LST (modified from Achtemeier, 2006; pg. 92).

While progress is continually made in smoke dispersion and fire effect modeling,
there are still short-comings that need to be addressed. First and foremost, many smoke
dispersion models were created based on typical industrial point sources, such as

smokestacks (McGrattan 2003). Because the energy output and plume rise are smaller

for industrial sources than for wildland fires, the governing equations need to be modified
for modeling high intensity wildfires. Also, as Miranda (2004) points out, smoke
dispersion depends on the emission of pollutants to the atmosphere, which is highly
dependent upon fire progression and characteristics. To account for this dependency
there needs to be more emphasis on integration between smoke and fire progression
models, taking into account wind field distribution (both horizontal and vertical), the
advance of the fire line and the interaction between the fire and wind (Miranda 2004). To
try to account for these interactions, several attempts have been made to create modeling
systems that incorporate different aspects of the fire and the fire environment in attempts
to better predict aerosol concentrations and reduced visibility.

The integration of four simulation models was discussed by McKenzie et al
(2006). In their study, a Fire Scenario Builder (FSB), which creates scenarios of fire
starts, sizes and locations, was combined with a meteorological model (in this case the
PSU/NCAR Mesoscale Model, MM5), an emissions production model (EPM) and a
dispersion model (CALPUFF) to simulate the contribution of wildfires to elevated PM2_5
concentrations and reduced visibility (Figure 2.4) (McKenzie et a1. 2006). 24 hr mean
PM” concentrations and light extinction coefficients were calculated. Results of this
study showed that while the system did accurately predict the 20 worst days of reduced
visibility, it underestimated those daily averages of PM25 concentrations. The system _
was also not able to predict the most extreme event, largely due to the fact that wildfires
were most likely not the only source of regional haze during this time. These limitations
within the modeling system need to be addressed and improvements need to be

implemented in the FSB, to better simulate the total number and size of simulated fires.

l3

Changes in vegetation cover over time need to be incorporated into the system as well,
allowing for modification of the fuel layers, which influences potential and actual

biomass consumed and smoke produced from the emissions simulator (McKenzie et a1.

 

 

 

 

 

 

 

 

 

 

 

 
 
  
  

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

2006).
Climate Prescribed fire and
g management
5 (Fire) weather 5
an--.” Wildfire starts 9 i
--------- :73 Fire severity :_ fl. Consumptior>
IT‘“ Fire regimes Emissions
II I 3 Fuels (I' /d d) . ,
ii i we ea <DisperSIon>
Vegetation

 

 

 

l' ire s cenario builder

 

Visibility/Haze

Figure 2.4 Schematic of modeling framework. Dotted arrows indicate interactions which
are turned off or for which default values are assumed (from McKenzie et al, 2006; pg
28 1).

Valente et al (2007) describes another integration system that combines

DISPERFIRE (Miranda, Borrego, and Viegas 1994) and FireStation (Lopes, Cruz, and

14

Viegas 2002). DISPERFIRE is a real-time system developed to simulate the atmospheric
dispersion of pollutants emitted during a forest fire while FireStation is a software system
aimed at the simulation of fire spread over complex topography (Valente et a1. 2007).
Input data for DISPERFIRE (topography, wind speed, wind direction, ignition time, fire
rate of spread and heat released) were estimated by the FireStation software system.
Measurements of aerosol pollutants such as PM“), NO, N02 and CO were collected at the
burn locations to validate the model and compare observed and simulated of aerosol
concentrations and smoke dispersion. Results of this study showed that although
estimated and measured aerosol concentrations compared relatively well, emission
estimates are a large area of uncertainty in this model (as well as other dispersion
models).

Another integrated system designed for smoke impact prediction is ClearSky, a
numerical smoke dispersion forecast system for agricultural field burning created for
smoke management purposes in the Inland Pacific Northwest (Jain et a1. 2007). Three
components, a web—based field burning scenario generator, the CALPUFF dispersion
model, and a web-based application for reviewing CALPUFF animations, are combined
with an atmospheric model (PSU/NCAR mesoscale model, MM5 (Grell, Dudhia, and
Stauffer 1994)) to create the modeling system. Input information for the simulations is
provided by data collected directly by the fire manager. Default parameters allow
managers to estimate smoke dispersion and aerosol concentrations when they are unable
to submit their specific burn information in enough time for estimates to be generated.
While there are four different types of agricultural field burning (head fire, mass ignition,

strip head fire and backing fire) ClearSky models the field burns as head fires (Jain et al.

15

2007). An evaluation of the model by Jain et al (2007) shows that ClearSky is able to
successfully predict where and by how much agricultural burning increased aerosol
concentrations (Figure 2.5) on days when the MM5 predicted meteorology is in

agreement with observed meteorology.

 

    

 

 

 

 

 

 

 

 

100
so ,etbméizwi 90
OPredicted
I. / conserved 8°
"1‘ col / . 70
E. j ,1 60
3 4o~ /
a I ./ 5°
5 201‘ l, . 1L _, :3
.1 _-‘i __,_o-'
0: ma. - - . , “"fi , 20
9200 11:00 13200 15200 17200 A
Local Time :10 f
uglma

Figure 2.5 Observed and predicted PM” concentrations (left) a monitoring site in Idaho
and the corresponding predicted plume concentration contours (right) during August 25,
2006 (modified from Jain et al, 2007; pg. 6756).

On these days, the maximum modeled concentrations at the observing sites were
approximately the same as observed maximum concentrations, allowing the fire manager
to have confidence in the model results and in their ‘bum’l’no-bum’ decision. While
ClearSky simulations were in good agreement when the simulated meteorological
conditions were satisfactory, large discrepancies occurred in PM concentrations and

timing when the simulated meteorological conditions were not in good agreement with

the observations. This suggests that the modeling framework is highly dependent on the
meteorological forecasts. Jain et al (2007) suggest using ensemble meteorological
forecasts or nested, higher resolution domains, particularly in areas with complex terrain.
While more efforts are needed to assess ClearSky in varying situations, it shows promise
in evaluating and estimating smoke and aerosol impacts. ClearSky’s usefulness in
simulating smoke impacts from agricultural fires initiated the creation of BlueSky, a
framework designed to predict smoke impacts from not only agricultural burns, but from

forest and range fires as well.

The BlueSky Smoke Modeling Framework

The BlueSky Smoke Modeling framework is a smoke forecasting system that provides
daily predictions of surface smoke concentrations from prescribed fire, wildfire and
agricultural burn activities (O'Neill, Ferguson, and Peterson 2003). BlueSky is a modular
component system comprised of the following basic elements: source characteristics,
emissions models, weather models, and smoke dispersion models. The modular system
of BlueSky eliminates the need for large amounts of user inputs and provides immediate
web-based feedback, showing cumulative impacts of smoke from biomass burning
(Ferguson, Peterson, and Acheson 2001; O'Neill et al. 2005).

As discussed above, smoke dispersion models, though available for years, have
been hindered by being too complicated for operational use or too simple to be
considered realistic. The USDA Forest Service with cooperation from the US
Environmental Protection Agency (EPA) worked together with fire managers and air

quality regulators to create the BlueSky modeling system that is both realistic and user

17

friendly. Since its inception for operational modeling of wildfire smokes in 2002 (Berg
et al. 2003), written documentation regarding the BlueSky Smoke Modeling Framework
has been limited to mainly conference proceedings and short technical notes.

There are many advantages to the BlueSky’s modular framework. The system
can be utilized in a wide array of applications as unnecessary model components for a
particular application can be easily turned on and off (Larkin et al. 2007). Sestak et al
(2002) notes that because BlueSky relies strongly on real-time fuels and weather
information, it can provide realistic and accurate estimations of smoke dispersion
throughout a region. The ability to use information concerning all fire activity in a region
allows BlueSky to not only predict smoke impacts from a single fire but also the
cumulative impacts from multiple fires, further increasing its applicability to air resource
management (BSRW 2006).

In 2003, BlueSky was integrated with the EPA’s Rapid Access INformation
System (RAIN S) which is a Geographical Information System (GIS) useful for visual
displays of geographical data. The combined system is called BlueSkyRAIN S. The
integration of these two products allows users to zoom in on areas of interest, step
through time, and overlay other GIS layers (geopolitical boundaries, meteorological
fields, topography, etc) in order to interactively view mapped geographic information and
analyze the potential smoke impacts more completely (O'Neill, Ferguson, and Peterson
2003; O'Neill et al. 2005).

BlueSky’s deterministic approach of using physical laws to determine the
direction and speed of smoke movement across a landscape allows it to be used for

multiple applications (Potter, Larkin, and Nikolov 2006). It is most often used by smoke

18

and fire managers to assess smoke impacts due to prescribed burn activities, wildfires and
wildland fire use fires (BSRW 2006; Adkins et al. 2003). The goal of BlueSky is to help
minimize smoke impacts on local and regional environments. BlueSky has proven to be
a valuable communication tool. It provides regional forecasts of smoke impacts that can
be used to issue public safety notifications and inform the public in those regions to be
cautious when driving or being outdoors as smoke may be reducing visibility and/or air
quality (Berg et al. 2003; Evers 2006). If a community was “smoked out” or heavily
impacted by high amounts of smoke, BlueSky can be a useful post-smoke event research
tool to assess the situation and evaluate if that situation could have been predicted and/or
prevented (O'Neill et al. 2005).

The ability of BlueSky to predict timing and location of smoke impacts as well as
surface concentrations of PM as a result of wildfires has been tested in a small number of
case studies. Wildfires used as test cases include: the Quartz Complex Fire of 2002
(O'Neill, Ferguson, and Peterson 2003; Evers et al. 2006), the Hayman Fire and Bitterroot
Complex of 2002 (Adkins et al. 2003; Rapp 2006), the Rex Creek wildfire of 2001 (Berg
et al. 2003), the Frank Church Complex of 2005 (Larkin et a1. 2007), the Black Range
Complex of 2005 (BSRW 2006) and a series of wildfires across northern California in
2006 (Fusina et al. 2007). BlueSky was consistently found to accurately predict the
smoke plume trajectory and timing of smoke impacts seen in the observations. These
studies also showed that while simulated surface PM25 concentrations were significantly
lower than observed concentrations, BlueSky was able to predict the maximum,

cumulative concentration over an entire domain. The previous case studies suggest that

19

while BlueSky is a useful tool that integrates fire occurrence, fuels and meteorological
data form multiple fires, there is still a need to improve its performance.

Uncertainty and limitations within the BlueSky framework include: errors in real-
time fire characteristics data due to lack of consistent and reliable reporting systems and
the migration of wildfires through varying complex fuel load and fuel conditions
(Ferguson, Peterson, and Acheson 2001); lack of operational monitoring networks to
compare with simulated data (BSRW 2006); and high sensitivity of the model to
predicted wind directions that may lead to small errors in timing and location of
simulated smoke plumes to severe underestimates in PM concentrations at a particular
location (Berg et al. 2003). Most importantly, the accuracy of the input data is vital to the
accuracy of BlueSky predictions of smoke impacts as inaccuracies within the input data
will propagate through the entire system (Larkin et al. 2007). Larkin et al (2007) also
suggests that while BlueSky simulates wildfires as a single, large convective plume, in
reality wildfires are comprised of multiple convective plumes and this could be reason for
the underestimates in surface emissions. In the study, the Rex Creek fire was simulated
as one single convective plume and also as 5 and 10 smaller convective plumes. Results
showed that while increasing the number of convective plumes substantially increases the
near-surface PM2_5 concentration estimates (Figure 2.6), it also shows less variability and
higher than observed averages. Liu et al (2006) found similar results, splitting the fire
into several component fires increased the surface concentrations of PM”, which
suggests that knowledge of the fire behavior, and thus accurately simulating wildfires, is

important to accurately estimate smoke impacts (Larkin et al. 2007).

20

PM” Concentration (uglm3)

 

 

 

 

 

 

 

Figure 2.6 Maximum average and ground concentrations for the Rex Creek wildfire for
the period of August 19-26, 2006 when the fire is (a) treated as a single fire, (b) split into
5 fires and (c) split into 10 fires. In all cases the total area burned is the same (modified
from Larkin et al, 2007).

With some success of the BlueSky system, researchers have focused on the
integration and expansion of BlueSky with other operational systems to further increase
the smoke impact assessment potential. Pouliot et al (2005) integrated BlueSky and the
Sparse Matrix Operator Kernel Emission (SMOKE) processing system to produce higher
resolution smoke estimates for use in regional-scale chemical transport models such as
the Community Multi-scale Air Quality (CMAQ) model. SMOKE is a tool that creates
gridded, speciated, and temporally allocated emissions estimates for use in atmospheric

chemistry models (Pouliot et al. 2005), and can convert the resolution of the emission

21

data to that needed by an air quality model (in this case CMAQ). For example, SMOKE
can convert wildfire emission data, which are available daily, to hourly emission data for
each grid cell and for each model layer. This system (called BlueSky-EM) was tested for
wildfires that occurred in Florida in May of 2001. Preliminary results show a good
correlation between BlueSky estimated PM2_5 concentrations and those same predictions
by the CMAQ simulations. The agreement between models is encouraging as CMAQ is
a widely used air quality dispersion model. The graphic predictions from both of these
models also closely mimic the satellite images of the wildfires as well as plume
movement during the study period (Pouliot et al. 2005).

At the 6th Annual CMAS Conference in 2007, Craig et al (2007) presented
preliminary findings on the expansion of BlueSkyRAINS from its limited coverage over
the western US. to national coverage over the contiguous United States (called BlueSky-
N). According to Craig et al (2007), the expansion of BlueSkyRAlN S can be used for
preparation of a national emissions inventory of PM2,5 due to anthropogenic, biogenic and
fire emissions. Existing satellite technologies and remote sensing are being combined to
help expand the applications of BlueSky-N and create a uniform method of determining
fire sizes and locations, so that the fire contribution can be more accurately determined.
This national emissions inventory will provide a 36 km resolution national forecast of
ground level PM25 concentrations (Craig et al. 2007). Consistent with previous findings,
preliminary results presented showed that the BlueSky-N system was able to predict the
timing and transient nature of the smoke plume but under—predicts the peak magnitude of

surface PM2_5 concentrations.

22

As BSRW (2006) concluded, the BlueSky system is reliably the “best modeling
framework available for wildland fire smoke predictions at this time, but it is still is a
research tool and needs further development”. As indicated previously, attempts have
been made at model evaluation and improvement, but research is still necessary as model
enhancements become integrated and operational in the BlueSky system. Studies
evaluating BlueSky predictions have not been done on large wildfires and currently there
is still uncertainty concerning all the system components. Rather than just looking at the
entire BlueSky framework as a whole, studies concerning individual system components

are necessary for understanding the error sources for BlueSky predictions.

Model Description
PM 2_ 5 Concentrations and Smoke Trajectory Simulations

The BlueSky Smoke Modeling framework produces predictions of smoke plume
trajectories and PM25 concentrations. The BlueSky framework is a system that combines
meteorology and emissions to produce a regional-scale analysis of aerosol concentrations
and smoke dispersion. The major components of BlueSky are fire characteristics,
meteorology, emissions and smoke dispersion (Figure 2.7). Of those, only fire
characteristics and meteorology are required as inputs for the running of BlueSky, while
all other variables are derived within the framework.

Hourly, four-dimensional (x,y,z,t) gridded data is needed for the dispersion and

trajectory models within BlueSky (Larkin et al. 2007). Although different meteorological

23

The BlueSky Smoke Modelifl Framework

 

\(

 

Figure 2.7 Components of the BlueSky modeling system.

24

models can be used to estimate meteorological inputs for BlueSky, the meteorological
predictions presented in this thesis were simulated using the Pennsylvania State
University/National Center for Atmospheric Research Mesoscale Model (MM5) (Grell,
Dudhia, and Stauffer 1994) as it is the most widely used meteorological model in the
BlueSky forecasting community. MM5 is a non-hydrostatic, primitive equation
mesoscale model that uses a terrain-following coordinate system. Initialization and
boundary conditions were obtained using National Center for Environmental Prediction
(NCEP) 6-hourly Eta model output (Part III, IV) as well as the 3-hourly North American
Regional Reanalysis (NARR; Part IV only).

Fire characteristics such as fire location, acres burned, date and time of ignition,
fuel loading and fuel moisture information downloaded from local and state reporting
systems and inputted into BlueSky. Currently for wildfires, there are two reporting
systems being used: the Incident Command System (ICS-209), a US. national wildfire
and wildland fire use reporting system, and the Satellite Mapping Automatic Reanalysis
Tool for Fire Event Reconciliation (SMARTFIRE), a system that aggregates data on
wildland fire size and location from various sources into one unified data set. When fuel
loading data is not provided by fire managers or reporting systems, BlueSky utilizes
lookup tables and classification systems imbedded within the system to get the data.
There are three options: the Fuel Characteristics Classification System (FCCS)
(McKenzie et al. 2007), which is the default map used, the US. National Fire Danger
Rating System (NFDRS) (Cohen and Deeming 1985) and the revision to the NFDRS by
Hardy et al (1998) over the 11 western US. states. All three classification systems have

a 1 km grid spacing.

25

Fire characteristics, are then inputted into the Emission Production Model (EPM;
(Sandberg and Peterson 1984) to calculate total fuel consumption and emissions. Total
fuel consumption is first calculated and then a time profile is used to allocate the
emissions diumally (Larkin et al. 2007). Wildfire time profiles are based upon the
Western Regional Air Partnerships profile (WRAP and WGA 2005), the percent of total
area burned per hour is given in Table 2.1.

Table 2.1 Temporal allocation of acres burned in a day for wildfires. Data developed by
the Western Regional Air Partnership (from WGA/W RAP, 2005).

 

 

 

 

 

 

 

 

 

 

 

 

 

flog; Percent
0-9 0.57
10 2.00
l 1 4.00
12 7.00
13 10.00
14 13.00
15 16.00
16 17.00
17 12.00
18 7.00
19 4.00
20-23 0.57

 

 

 

 

Emission dispersion is calculated using CALPUFF (Scire, Strimaitis, and
Yamartino 2000). CALPUFF is a puff dispersion model that simulates point, volume or
area sources (assuming that the plume dispersion is Gaussian). A pre-processing
program, EPM2BAEM, is used to convert EPM output to the proper format for use in

CALPUFF. CALMET, a diagnostic meteorological model that calculates the three

26

dimensional winds and temperatures, as well as microphysical parameters such as surface
characteristics, dispersion parameters and mixing heights (Scire et al. 2000), is used to
convert the MM5 estimations into a mass-conserving field for use in CALPUFF. Smoke
trajectories are simulated using the Lagrangian particle model, HYSPLIT (Draxler and
Hess 1997). Trajectories are initiated at a height of 10 m and thus do not account for
initial buoyancy effects due to the heat given off by a fire. The final products created by
BlueSky are visual images of smoke plume trajectories and shape, as well as surface

aerosol concentrations such as PM”.

Summag

Part II provided background information on smoke dispersion modeling as well as
the BlueSky smoke modeling framework. A literature review of each area was presented
highlighting major limitations of smoke dispersion modeling and evaluation studies
conducted using the BlueSky framework. A discussion of the BlueSky model setup was
also given. The following chapters present the results of two case studies used to
evaluate BlueSky’s ability in predicting smoke plume trajectories and surface

concentration distributions of PM25.

27

PART III
NORTHERN CALIFORNIA, AUGUST 2006

Introduction

The total number of acres burned in the 2006 and 2007 wildland fire seasons were
over 200% of the ten year average (NIFC 2008). Although the number of wildfires has
been decreasing since the 1980’s, the sizes of wildfires have been steadily increasing,
with nearly 10 million acres burned in just over 96,000 individual wildfires in 2006 and
9.3 million acres burned in over 85,000 wildfires in 2007 (N IFC). The number of acres
treated by prescribed fires also increased in 2006 and 2007 to more than 9.8 and 9.3
million acres, respectively (up from 8.6 million acres in 2005) (NCDC 2008). Wildland
fires are a necessary component of the forest ecosystem as well as an important tool for
fuels management practices. But as wildland fires continue to increase in size, increased
smoke emissions can cause problems related to air quality standards and visibility.

While numerous atmospheric dispersion models have been developed for
predicting the transport and dispersion of atmospheric pollutants (Scire, Strimaitis, and
Yamartino 2000; Binkowski and Roselle 2003; Bacon et al. 2000), few have been
designed specifically to simulate smoke from wildland fires. To provide a tool for
simulating smoke transport and dispersion from wildland fires, a smoke modeling
framework called BlueSky was developed by the USDA Forest Service AirFire Team in
early 2000 and became operational in parts of the Pacific Northwest by 2002 (Berg et al.
2003). This modeling framework, initially designed for prescribed burn decision support,
integrates consumption, emissions, meteorology, and dispersion models to predict smoke

trajectories and concentrations of particulate matter. BlueSky smoke predictions are

28

available daily across the contiguous US. and the output has been used by air regulators,
burn bosses and smoke managers as a guide to help make ‘go’ and ‘no-go’ decisions
about prescribed fires and the subsequent burn plans. Since its inception, BlueSky has
evolved to be a useful tool in wildfire monitoring as well, especially in the West where
the risk of large wildfires is greater. BlueSky can help to track day-to-day emissions and
plan suppression strategies. BlueSky is becoming a “one-stop shop” for regional smoke
concentration and emissions tracking across all land ownership. While used widely,
BlueSky’s output has not been rigorously validated against field observations, and the
accuracy of its predictions are uncertain under different meteorological conditions. To
date, validation efforts of BlueSky’s ability to simulate wildfires have been limited to
isolated cases in the Northwest (Adkins et al. 2003; Berg et al. 2003; Larkin et al. 2007;
BSRW 2006) and Southeast (Pouliot et al. 2005; Craig et al. 2007).

Although none of the existing validation efforts have been subjected to peer
review, the studies listed above have produced similar results. In a study focused on the
2002 Hayman Fire in Colorado, Adkins et al. (2003) found the location and timing of
predicted smoke impacts to agree with observations while the predicted PM2,5
concentrations were orders of magnitudes smaller than observations. Larkin et al. (2006)
conducted a similar study using data collected during the 2001 Rex Creek Wildfire in
Washington and the 2005 Frank Church Wildfire in Idaho. The comparisons between
smoke plume trajectories and shapes were found to agree quite well with satellite images
of significant observed smoke concentrations, but, BlueSky failed to accurately predict

the observed magnitudes of PM25, concentrations.

29

Additional case studies in different geographical regions are needed to further
assess the accuracy of BlueSky and to document the strength and weakness of the
modeling system. In an effort to provide better information pertaining to fire weather and
smoke dispersion/transport, the California and Nevada Smoke and Air Committee
(CAN SAC) has been running a BlueSky smoke prediction system in real-time for the
state of California and Nevada. In this paper, we describe results from an effort to
validate the CAN SAC BlueSky smoke predictions as part of a project funded by the Joint
Fire Science Program to develop tools for estimating contributions of wildland and
prescribed fires to air quality in the Sierra Nevada mountains. Surface in-situ
observations and remotely sensed imagery are used in comparison with model output to
assess the accuracy of the BlueSky modeling framework in predicting PM”

concentrations and smoke plume trajectories.

Methods
Study Area

Northern California is a region of contrasting landscapes and diverse topography
(Figure 3.1). Northern California is bordered to the west by the Cascade Range and the
Klamath and Siskiyou Mountains. The Klamath and Siskiyou mountains range in
elevation from around 6,000 ft (1829 m) to 8,000 ft (2438 m) above the mean sea level
(MSL). These mountainous regions are forest covered and their small ranges are
separated by deep canyons. The Cascade Mountains were formed by volcanoes and, in

California, elevations range from about 4,500 ft (1372 m) to 5,000 ft (1524 m) MSL.

30

Eoom

E009. .

Eoomr

EOOON

E comm

E ooom

E comm

E ooov

 

 

 

 

.525“. Ed? BEE—=5 2: $8065. gen :83 EC. .2236 33E Ea 30%208808

3:23:88 Co 28:32 mE—éofi Eowouo E0558 Ea EEoEmU fiasco: Co map: 59:3. ._.m 053%

 

30w: >>oo~w >>0NNF govuw

         

Es. 8% SEE: Seam.
“Seam SEE: 8&3.
293% Sn. 835m 0
225m 52 8e. 3..
Snow E2 6&3 o

......................

h . ”p >to.<m_:>
t .1]:
93! 3......

Mu. uuDJmDmm
is... .

4‘ .. l . . s "t .
. (Wat. E<om<
.11}? u
w L. 293.5 5 {Eu .Emowmmc
:1 $05202 . .
4 . .‘. . P. . or . . 1. \mg.
I. . HI Swims... IL 0‘. a

        

{
r

 

I

 

 

Zoom

203

209.

zowv

z o:

 

 

31

To the east, northern California is defined by a basin and range province, which extends
into Oregon and Nevada. California experiences a variable climate with two main
seasons, one rainy and one dry. In northern California, the rainy season occurs during the
months of October through April, leaving the summer months comparatively hot and dry.
Foliage throughout the western area of this region is a conglomerate of mixed evergreen
forests, including Douglas, White and Red Fir, and Ponderosa Pine. To the east the
vegetation becomes a mixture of Chaparral, California prairie, mixed Chaparral and a
valley needlegrass series. Dry conditions, available fuels and high air temperatures
during the dry season make this area ideal for wildfire outbreaks. During the period of
study, August 20 — 31, 2006, a series of major wildfires ignited across northern California

and southern Oregon.

Observational Data

Hourly surface meteorological data were collected for fifteen stations within the
region to evaluate how well the simulated meteorology matches observations. The
locations of these stations are shown in Figure 3.1, while Table 3.1 lists each station’s
geographic information. Observational data were collected from the N CDC Climate Data
Online website, http://cdo.ncdc.noaa.gov/CDO/dataproduct. Upper air sounding data

were collected for two stations, Medford, OR and Reno, NV, from the upper air data

 

archives at the University of Wyoming (http://weather.uwvo.edu/) (Figure 3.1).
PM2.5 (particulate matter with diameter less than 2.5 pm) observational data were
collected for eight stations within the region (Figure 3.1). Four stations are situated in

Oregon near the California border and are useful in capturing the north-northeasterly

32

transport of fire smoke. The PM2_5 data for these stations was collected from the Oregon
Department of Environmental Quality (DEQ). To obtain hourly visibility data, the
Oregon DEQ collects Beta Scattering Units (BSCAT) which are collocated with 24 hour
average PM” filters. A linear regression is used with the data from the collocated
monitors and the “correlation” equation is applied to the hourly visibility data to obtain
hourly PM” estimates. The other four PM25 monitoring stations are found within the
northern California region. Data for those stations came from the California Air
Resources Board (CARB) Aerometric Data Analysis Management (ADAM) system
database. Hourly PM25 values for these stations were sampled by Met One Beta

Attenuation Monitors (BAM).

Satellite Observations

BlueSky not only outputs hourly surface concentrations of PM2_5 but also
graphical images showing smoke plume shape and trajectory. To qualitatively assess the
accuracy of smoke plume shape and trajectory, images from the Moderate Resolution
Imaging Spectrometer (MODIS) Satellite Fire Detection website Hazard Mapping
System (HMS) are used. These images are quality controlled displays of fires and
significant smoke plumes as detected by meteorological satellites (McNamara et al.
2004). While HMS is a useful and advanced tool for identifying smoke plumes, there are
limitations that should be considered when using this imagery. First, there is no way to
discriminate between agricultural or controlled burns and wildfires. This discrimination
can be done if the duration of a fire is known as wildfires usually last longer than

agricultural or controlled fires. Second, small fires that are not producing large amounts

33

of heat or fires burning below a thick canopy overgrowth may go undetected because
they cannot be seen or they blend in with the hot background due to high surface
temperatures and reflectivity (this is especially prominent in the western US summers).
Also, the satellites themselves can be a limitation. Satellites (using the 3.9 micron band)
are not able to see through clouds other than cirrus and in the middle of the day and when
the sun angle is high, smoke is hard to detect because satellites have a hard time
distinguishing the differences between high levels of reflectivity and smoke (McNamara
et al. 2004). Despite these limitations, the HMS is still a very useful imagery database
for qualitative smoke plume comparisons. MODIS aerosol optical depth data images are
used to quantitatively compare the concentrations of aerosols within BlueSky predicted

smoke plumes to observed. concentrations.

34

2:;
0mm mo
0mm mo
0mm «0
0mm mo
oouz
0mm mo
2:5
0mm mo
oouz
mm<anc
oouz
mm<xnc
mm<anc
oooz
uooz
oouz
oooz
oooz
oooz
oooz
uooz
mm<xmc
oouz
oooz
uooz

mungmumafiw msflnflm agauaea ayamm “saw

as

c_m_
wmm
mmv
ovM
mVNM
mVNM
Mva
ham
com
com
mM
QM
wM
oom_
h—
max
No
mmM
00M
owM
we
occ—
mm
mm
MVMM
who_

8093.830 22M 98 Bong—80808 8M :anESME c233 oEQEwooO .~.m ~35.

08983 05:5 33.5:885 mambo»? Mo 3633.5 05 8.865 x5250: 15>? 0F?
mcocaiomno 32M 8236.: .M. .mcocmtomno 323—2082: he comm: can 8me 8365 .D. 93 .m. 093 202..

553 T
magma-
wchT
mwméfi-
OSANM
5:42-
WONT
$.37
awaNT
www.mmT
GANT
$.57
SNNM-
Smog-
$2.va
SRNNT
3&7
:mNNT
mNdNM-
N53-
woods-
CthT
342.
mwANM-
mmmme
mmmdNT

omdm
wmdv
flaw
mwoNv
wwfimv
2 .NV
5— .Nv
wmdv
E mdv
mwmfiv
mfiom
dem
Ndm
5.3
m5;
mwhgv
mwodv
2 Wow
2.0V
m2 .3
:odv
mdm
wdm
wdm
W;
mmm. IV

>Z
MO
M0
M0
M0
M0
MO
M0
M0
M0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0

OZmM

QM<20MO Qm—Mm HAO>OMM
>m§1~<> mMOZHAIM
HMOMMM< OZUMOOMM
QZ<AmM Mqu—ME - mat—<M EH<2<AM
M< AHZH mus/E =H<2<AM
AmO>><v BE>mM<A
ZOmM0<QQMOMQmE
HZO—qum QZ< HZ<M0 .QMOMQm—E
M< 12.7: >m13<> mDOOM QMOMQME
Hm QZOEA< - >.:0 <§

O0 <MD>

<mDAO0

Hm <Q<>mZ BM - <x—OHMOM
>.E0 Hzm0mm—M0
M< >HZDO0 DO>MMmMm mDO<HZOE
HMOMMM< <H<0M<
EMM< A<MMOMZD§ 07MQDm—M
EM< A<n=0MZDE ”EDI—m Qm—M
M< A<MMOMZDE ZED
m>O0 Mmfiummm
M< ZO>Z<0 mem
m>< <MMZ<NZ<§ O0=M0
A<EOMZDE O0Mm0
m<MDHA<
<Hm<flm HZDO—Z

 

oEaz

mwvah
vawM
wco_m
omoommomomh
ONMOM
ommvammwmmh
oomooenommh
woman
wvvom
nNNVNONQmNh
moooMoMQ
omaoawmwvmh
moo—MMoo
ace—meow
oommmmMovao
mmmvmmmommh
mwmvmmvmmmh
hmmvmomommh
cMNVNOMmmNh
mummmmommmh
omamowwwmmb
mmmmmmvwmmh
N00Ohooc
amoQQMhavmh
aomaowmmmmh
mMN¢thmmNh
A: 3%

 

cm
mm

~Nm¢m©bw¢O~NmV
i—t—u—uc—tv—r—n—ru—rt—rNNNNN

v-‘NMVWCI‘OOGO
q—u‘

35

Model Simulations

The data described above are used to evaluate the real-time smoke predictions
from the operational CAN SAC BlueSky runs at the Desert Research Institute. BlueSky
has five components that work together to predict surface concentrations and smoke
plume trajectories from wildfires and prescribed burns. The five components are
meteorology, fuel load, fire behavior, emissions, and smoke dispersion (O’Neill 2003).
The meteorological fields are predicted using the Penn State/NCAR fifth generation
mesoscale model (Grell, Dudhia, and Stauffer 1994). The model was configured with
four nested domains with the outermost domain of 36-km grid spacing covering western
United States and eastern Pacific Ocean and the innermost domain of 4-km grid spacing
overlaying the state of California and Nevada. Initial and boundary conditions were
obtained from the 6 hourly 40-km ETA forecasts from the National Centers for
Environmental Prediction (NCEP). To simulate the transport and dispersion of aerosols
from smoke emissions, CALPUFF, a multi-layer, multi-species non-steady state
Lagrangian puff dispersion model that can simulate the time and space varying pollutant
transport, transformation and removal (Scire, Strimaitis, and Yamartino 2000). The
necessary fire input, such as fire location, size, date of burn, and ignition time, was
obtained from the National Wildfire 209 Situation Reports (ICS-209). Secondary input
information (that information which does not need to be entered manually) includes:
length of ignition, type of fire (crown, grass, brush), vegetation type, 1hr, 10hr, and 100hr
fuel loading, snow melt month, duff depth, and burn site slope (Table 3.2). Hourly PM2,5
emissions were computed using Consume/EPM v1.03, and fuel characteristics input to

BlueSky were derived from the Fuel Characteristic Classification System (FCCS).

36

Table 3.2 BlueSky Input Information

 

 

Primagy Information

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Location of fire 0 Number of acres burned
Date of burn
Secondau Information
0 Ignition duration 0 Slope of burn site
0 Type of fire 0 Fuel harvest date
0 Fuel loading: 0—0.25, 025-], 1-3 in 0 Number of days since last rain
0 Fuel loading: 3-9, 9-20, 20 + in 0 Duff depth
0 1000 hr fuel moisture 0 10 hr fuel moisture
Tertiag Information
0 Grass and herbs fuel loading 0 Rottenwood
0 Litter depth 0 Shrub fuel loading
0 Terrain elevation 0 Total PM2.5 emissions (in tons)
0 Total PMlO emissions (in tons) 0 Total PM emissions (in tons)
0 Total CO emissions in (in tons) 4» Total C02 emissions (in tons)
0 Total CH4 emissions in (in tons) 0 Total non-methane hydrocarbon
emissions (in tons)

 

cNote the total emissions are calculated internally in BlueSky

Results and Discussion

Wildland Fires and Synoptic Weather Conditions during the Study Period

Four main fire complexes were burning during August 20 — 31, 2006. The
Orleans complex consisted of two fires (Somes and Buck fires) and was located east of
Orleans, CA near the Arcata surface meteorological station (Figure 3.1). Bar Complex,
comprised of four separate fires, was located northwest of Weaverville, CA, located in
Shasta-Trinity National Forest, near Mt Shasta meteorological station (Figure 3.1). The
three fires: Uncles, Rush, and Hancock were part of Uncles Complex located

approximately 20 miles northeast of Orleans, CA. Finally, Happy Camp was a large

37

 

complex that included 15 separate fires located near Happy Camp, CA (Crescent City is
the meteorological station located closest to Happy Camp, Figure 3.1). Containment
efforts for these fires were hampered by steep and rugged terrain as well as two main
active fire behavior periods on August 24—25 and August 30-31. By August 23 Happy
Camp was 100% contained, and the extreme fire behavior during this study period
occurred only in the Orleans, Bar and Uncles complexes.

Synoptically, at the beginning of the study period, the West was dominated by an
500 hPa upper level ridge with 12-15 m s'1 south-southwesterly winds at the 500 hPa
level over northern California and southern Oregon (Figure 3.2a). These wind speeds
increased steadily to 18-24 m s'l over the next few days as the ridge weakened and a
trough developed over the Pacific Northwest. By August 24, a day when many of the
fires exhibited active fire behavior the cut-off low over the Pacific Northwest intensified
and within 24 hours (Figure 3.2b), northerly 500 hPa winds, around 8-15 m s'1 prevailed
over northern California. Wind speeds weakened through the next 36 hours as an upper
level ridge moved back into the area. 500 hPa southwesterly winds allowed fire fighters
to regain control of the fires and substantial containment progress was made. By the end
of the study period, as the ridge moved off to the east and a trough moved in from the
Northwest, active fire behavior was again observed as 500 hPa wind speeds increased to
21 m s'1 and westerly flow was observed over northern California while northwesterly
flow was observed over southern Oregon (Figure 3.2c). By the beginning of September,
the 500 hPa winds had weakened again and containment efforts were continued. The
Orleans complex was 100% contained as of September 3, but the other two complexes

continued to smolder for the next few weeks.

38

(3) Au ust 20,2006

f BOX 1;- "

 
 
 

    
 
  
   

 

 

 

    

5900 ‘. ,

c) Au ust 31,2006

 

 

 

 

 

 

 

Figure 3.2. Plots of 500 hPa height (left) and wind speed (right) synoptic plots at 12Z
(0400 PST). Light shading in the wind speed panels (right) indicates weaker 500 hPa
wind speeds while darker shading indicates stronger wind speeds. Contours in the left
panel represent 500 hPa isoheights.

39

Meteorological fields

The accuracy of smoke prediction depends to a large degree on how well the
meteorological conditions are simulated by the meteorological model (MM5).
Comparison of simulated and observed near surface temperature, dewpoint temperature,
and wind speed are shown in Figure 3.3. Each point represents an hourly value across the
entire study period. The data points are separated into daytime values (0700 to 1900
PST) and nighttime values (2000 to 0600 PST). The model appears to have a cold bias
during daytime and warm bias during nighttime. Generally the model seems to be less
accurate when predicting nighttime temperatures as the warm bias at night appears to be .
larger than the cold bias during the day. Also evident is a higher predicted dewpoint
temperatures regardless of the time of day, an indication of a moisture bias in the model
prediction. The meteorological model has a difficult time in estimating near surface wind
speeds as the comparison with these wind speeds show a large scatter with no in both day
and night estimations. While it appears that the model has trouble representing near calm
conditions when the observed wind speeds were 0-2 m s“, it under predicts surface winds
when the observed surface winds are stronger (6-10 m 5"). Within the context of the
BlueSky modeling framework, weaker simulated wind speeds would allow pollutants
(specifically PM2.5) to remain in an area longer than what is actually observed and
simulated aerosol concentrations would be higher at the stations affected by a particular
smoke plume. Stronger simulated wind speeds would inaccurately suggest the pollutant
has cleared from a particular region, and those stations would show lower aerosol

concentrations.

40

45

 

 

 

 

 

 

A Nightime
<3: 3 5- Daytime
2
a
E
a) 25» _ .
O. O _, ,4
I— sfirii‘Ig' w
‘8 15_ _"'-f’_-«-W
Eu .0. °
3
a?) 5 -
-5 . . L . .
—5 5 15 25 35 45
Observed Temperature (0C)
8 35’
9..
93
3 25
S
a:
E
15
.9.’
.‘é‘
8
3 5 ’
a:
D
'D
2 '5r
3
:3
.2
(I) -15 4 . m . .
-15 -5 5 15 25 35
Observed Dewpoint Temperature (°C)
14”

Simulated Wind Speed (m s")
on

Figure 3.3 Simulated and

 

 

# * «or
_ fl sin n a ’F if
H *: at
r'. 2'.- .. .‘l‘ A
i‘ a I :+ .3 '1 ’7
rat {"5- .H. ‘ '9‘:
:5 ”"7" -‘ 3U."
I: ' i 9(-
_ 33:;l .5.
4 a _ atr-
‘ ii 1: “R R
w '1' * ¥ 1 u 1 i r J
O 2 4 6 8 1O 12 14

Observed Wind Speed (m 5'1)

observed hourly temperature (t0p), dewpoint temperature (center), and

wind speed (bottom) at all fifteen surface meteorological stations used within this study. Gray

dots are the day time valu

es while the black dots are night time values.

41

To evaluate how well the model simulated the diurnal variation of the
meteorological fields, time series plots of predicted and observed meteorological
variables were produced for all surface meteorological stations. Figure 3.4 shows three
stations representing (a) the general pattern represented by the MM5 simulations, (b) the
pattern seen along the west coast, and (c) the pattern seen in the northern stations, which
are located in steep topography (Figure 3.4c). As Figure 3.4 shows, MM5 was able to
capture the day-to-day variations in meteorological conditions due primarily to changes
in synoptic conditions. For example, the simulation captured the observed temperature
trend with high temperatures at the beginning of the simulation, decreasing temperatures
towards the end of the first week, and warming again in the middle of the second week.
The model also estimated the transitions between moderate, weak, and strong wind
regimes during the study period. Further, the simulation predicted the differences in the
diurnal temperature cycle between the inland (Figs. 3.4a and c) and coastal (Figure 3.4b)
sites with the latter exhibiting a weaker diurnal signal due to the influence of the ocean,
this is true of all the coastal stations used in this study. However the simulated
amplitudes of the diurnal temperature cycle are much smaller than observed. It is
interesting to note the strong diurnal variation in the observed wind speed with higher
winds in the afternoon and lower winds at night. This is due to daytime coupling of the
surface winds aloft by turbulent mixing, allowing stronger winds aloft to mix downwards

into the surface layer. At night the formation of a surface-based inversion decouples the

42

------ Observed

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

(a) Alturas Simulated
A I I I I I I l I I I I
8, 40
m
.5
E
8
E
,9
‘70)
E
ii;
13
.E
3
’0‘ I
é a,
8 . X )2 .. yr. ,( } Y
t 180 %X K x v X V; V
0 WV v ._ v X
D 90"V V g I??? fl
,5 v x v V V v 7 x v xv
; 0 X L 1V 1 l 6 it; % 1“ V L r x I
8/21 8/23 8/25 8/27 8/29 8/31
Date
------ Observed
(b) Arcata __.... Simulated
A 0 l T I l l I r r I l I
g ZOWW
E ._ r. .. _.N. r‘, - °. :-
8. 10 I ..‘Ji-filf ‘ ‘r T
E
'9 0 l 1 l l l l 1 1 1 l L
8/21 8/23 8/25 8/27 8/29 8/31

 

10

 

  

Wind Speed (ms-1)

   

 

  

 

 

 

8/21 8/23 8/25 8/27 8/29 8/31
E 360 r l l I ., J; . l l I I 1 $3..
C X X at” Y! ,.,-..imgt§‘¢ (Wye/I‘m _ xiv 558 i. #1] {X
o . " w. , 2i. ‘7
.8270 W’s fix, 3, visit 5% 9% fv’
.3 180 V. V v. 3:: v
D V V V
g 90 v I I; V V v VVV
3 O l l I I X I r I r l l 1
8/21 8/23 8/25 8/27 8/29 8/31
Date

43

Observed

 

 

 

 

 

 

 

 

 

 

 

 

 

(c) Montague Simulated
A 40 1 1 1 1 T I 1 1 1 r
9 A ' .A .
e I" f' P. A :' .’ ° ' . h .1
E 20 P '1' '. o O : ' ‘y' I.“
0 g.‘ . \° I, . I 0.0 .0 ..: \ .
g. 5 o \ g o ‘ \:
'2 O 1 1 1 1 1 1 1 1 1 1 1
‘3) 10 1 I i 1 1 i 1 I i I 1
E - '
E 5 . ° ° . °. °° . ‘ ' ° 4
(I) .° 0 .° ' ' ' o 'o
'D l o o ‘ o . o o no
.5 o :0 .0 I . o o o o o o o .0 . o 1
g 0 1 1 1 1 1 1 1 1 1 1 1
i: a WW
.9 X ' "‘
T5 x x viii
e c
5 v; I”
'o x v ‘
.E 9‘ . vV
3 a??? 113W $137M

8/21 8/23 8/25 8/27 8/29 8/31

Date

Figure 3.4. Time series plots of surface temperature, wind speed and wind direction at
three locations which show (a) the typical MM5 simulated pattern, (b) the pattern
observed at stations located along the west coast, and (c) the pattern seen in the northern
stations located in diverse terrain. Black dotted lines and upside down triangles (in wind
direction time series) represent the observations while gray lines and x’s (in wind
direction) represent MM5 simulated surface meteorological conditions.

surface layer from the layer above. The simulation reproduced this diurnal variation of
wind speed at all locations, although the agreement between simulated and observed
wind speed is better at coastal sites (represented by Figure 3.4b) than at inland and higher
elevation sites (Figure 3.4a and 3.4c). The difference between stations is most likely due
to the less complex topography in the coastal region. The predicted wind direction agrees

quite well with the in-situ observations.

Table 3.3 summarizes statistics (mean, RMSE, bias, correlation and error standard
deviation) comparing the model with observations to help quantify the errors of the
simulation. Simulated errors can be either systematic or random. Systematic errors are
errors due to biases within the estimations and if known can be reduced through the use
of calibration. Examples of systematic errors are misrepresentation of topography, land
use properties and radiation. Random errors on the other hand are due to uncertainty.
These types of errors can be reduced through statistical analysis. Examples of random
errors are uncertainty within the observations themselves or the initial and boundary
conditions used by a model.

Simulated standard deviations (STD) of temperature and deWpoint are smaller
than observed, indicating (as mentioned previously) that the model does not estimate the
full extent of the temporal variations as seen in the observations. However, the model
does do a better job at capturing the variations in wind speed as the STD between
simulations and observations are comparable. Linear correlation coefficients (r) show
strong correlations between simulated and observed temperatures while weak correlations
are apparent between dewpoint temperatures and wind speed. Bias is calculated by
subtracting MM5 simulated temperatures, mixing ratios and wind speeds from
observations. These values are then summed and then divided by the total number of
observations to obtain an average. Day and nighttime temperature bias further illustrate
what was seen in Figure 4.5, a warm bias at night, 2.819 °C and a daytime cool bias of -
1.492°C. Mixing ratio bias indicates a large moist bias, ranging from 8.340 g kg"1 at
night and 6.110 g kg’1 during the day. Wind speed bias is negative during the entire day,

in agreement with scatter plots results, MM5 simulated near-surface wind speeds are

45

hobo 03:? SSE

88 fl mmZM £2333. E853 Sta 2: fl Ohm Sum .EoEEooo corn—8.50 32:. 05 2 h coho 88? SEE .2: 2: fl ”mm—>2 302U

 

 

.5 8; 3.. 8e meg mod :e mom >3. 8%
8; o? 5.. 33. 8e 5e 8e Be and Em 5am
need 88 mad See coed 8nd e86 owed mood .
:3- 8.0- 3.0- e E to and :3 me. P- Ed «.8.
mm; mm; 84 8; mm: >5 mum Rd 8; Em

8d mod Baa. > 5: EN? 8.? 8.2 8. a 8.9 :85. 8:2
mm; mm; 8; o 3 88 «Na 03 med > 8 Em

8e 8e mam 8e 8e med 8.2 8.8 5.3 :85. wzo:<>mmmmo

ommm o 5 F CNN m 50 0mm _. mmm P mmom th F mmw _. 958 San *0 55:52
.2 >8 £92 .2 >8 2%,. .2. >8 £92 855m

3%...

 

 

 

m 53 :a 23.. mix—E dash—82.53 3a a .8 8.63.3 near—2.53 393295392 m. «Ema.
u u . v . . . .t u . n

46

under predicted. For temperature and wind speed, the bias is significantly less than the
error standard deviation (Error STD). This suggests that the random error component of
the model is contributing more significantly to the simulations overall error than
systematic errors.

Upper level atmospheric patterns were compared at OOZ (1600 PST) for each day.
Figure 3.5 shows the comparison on three different days which represent the general
vertical profiles seen in MM5 predicted temperatures, mixing ratio, wind speed and wind
direction. On most days the vertical temperature and wind speed profiles agree with
observed profiles in both the vertical structure and in magnitude. Mixing ratio and wind
speed both show that although the model does seem to produce the general vertical
structure as the observations, it does not do nearly as good of a job of capturing correct
magnitudes as it did with temperature. MM5 estimates a moisture bias within the lower 3
km of the atmosphere. Comparisons between MM5 vertical estimations and observations
on August 31 show more disagreement than is seen on previous days. Simulated vertical
temperatures are slightly warmer than the observations and wind speeds are
overestimated by about 2-3 m s". The disagreements between observations and
simulations seen on August 31 were unusual and are most likely due to the northerly
direction of near surface and upper level winds. As these moves in from the complex
terrain of the north it is harder for the model to simulate the meteorological conditions
compared to previous days when the surface winds were coming from the west, over the
smooth surface of the ocean. Simulated wind directions on all three days agree well with
observations. While there are some clear disagreements between the predicted and the

observed meteorological fields both at the surface and aloft, the overall performance of

47

8:03:58 Ro_wo_o>oo~oE noun—=86 £22 Bowman: €28.86 23¢ Px can

8:: bum 223 83820on E8958 92826 255 mad—want“ :38 023: can 3:: wagon gum—m .Ewfi: awn—Ewe veto—m
=a 52826 9:3 98 30% 9:3 6:8 wEfiE .305 0.3 85:35 838:3 doom 4m ~m=w=< 3 use doom .8 .m=w=<
so .88 .a “2&3. 3 some 228% as: a Sea... cameo .2882 CE o8: N8 3 mega E2 can: 2 28E

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Acycofigo uc_>> $-95 Beam as; 3-9. 9 ozmm 9.2.2 O: 99F
own EN 8' cm 9 o v N o w .o w w n 03 new con mam
. q 4 o d o a q o
afikm b b b no 1
m... s 1 F . F . . _. H
s . Pu.
5
: x . E N
a... x x. N u. m . 1 N M
4 b X o
.9 .
. x m m t . 1 m
. meow .3 3:913
08 SN 2: 8 o . 1m in ms NF, 2m 2m
1 1 4 o o 4 q 4 m o
m .__
a . F F P H
.... a.
.M. W
a... N 1 m * N 1W).
a . p
i :ii 1.... .m m t . l m
33 .3 «2631.:
con cum ow? om o v 2m «5 5m 3m mom
. 4 . o o 1 4 a o
o\.&> L
a
. as . P m P . P H
p... 3
l. . . W.
a. .H
x N N . .N M
w... . 1
- - m m i . .- .. m
33385 moon .9" 3:92 3

32030

48

the model in this region is good especially considering the highly complex topography

and land cover.

Smoke Plume and PM 2_ 5 Prediction

Aerosol optical depth (AOD) data and BlueSky estimations (Figure 3.6) indicate
high correlation in aerosol concentration peaks and plume locations. Figure 3.6a shows a
peak in AOD values off the coast of northern California which is represented in the
BlueSky images as well. MODIS HMS imagery displays this plume as well, showing the
located in similar south-northwest orientation — all of which are consistent with the
synoptic conditions and the southerly winds on August 21. On August 29 (Figure 3.6b),
BlueSky detects the smoke plume over northern California and its southwest-northeast
orientation — which agrees with the HMS imagery. BlueSky also shows slightly higher
aerosol concentrations over central—northem California, in comparison to central
California, again in agreement with the AOD values shown. MODIS HMS does not
show the higher aerosol concentrations over central Nevada, which is evident in both
BlueSky and the AOD images. This is most likely due to the low intensity (thus low
heat) of the fire or the existence of low, thick clouds during the observation times, which
prevents the satellite from picking out the fire from its surroundings. The AOD image is
able to pick up this increase because its not trying to distinguish the fire from the
background, it is measuring the amount of light that is able to pass through the
atmosphere. A similar situation is found in Figure 3.6c, where BlueSky, HMS, and AOD

identify a plume off the coast of northern CA, but HMS does not pick up the smoke

49

s, 2006 c) August 31. zoos
' . ' ‘ BLUESKY‘

 

 

        

‘12?
R 1'1

(1) Au ust 21 . 2006 [b] August 2

  

65 _,—I— 1 '3 - irk— —4—
60 t1 1' T {,4- 5,1—
50 1 ~.» ' ' . h'mu.

'.V‘ ' I .y A? "k
\L ‘.Q" “.1 V

l — '- 7 ‘-..'.‘_
L W" ,. 1 I
10 ~ t . v. !
‘1 l '>"‘ \ \

 

PM” Concentratlon (ug m3)

 

 

 

 

 

 

Depth at 0.55 microns

 

Figure 3.6. The images across the top show BlueSky visual output of surface PM2_5
concentrations for three days (3) August 21, (b) August 29 and (c) August 31. The middle
images show the corresponding satellite data taken from MODIS hazard mapping system
(HMS), where gray shading indicates significant smoke plumes. Darker shades of gray
represent higher concentrations of smoke. The bottom images are the aerosol optical
depth data for each day, also taken from MODIS. Warmer colors indicate a higher
concentration of aerosol within the atmosphere.

50

plume evident in Nevada. In most instances, BlueSky is able to simulate the existence,
orientation and shape of observed smoke plumes and areas of higher aerosol
concentration.

A quantitative evaluation of the BlueSky-predicted PM2.5 concentration is not as
straight forward as the evaluation of predicted meteorological fields. PM2_5 observations
include both a background value and a fire contribution, whereas BlueSky predicted PM
concentrations only consider the fire emissions. Because of this, a direct comparison of
the modeled and observed concentrations is not possible. The observed background
value must first be estimated and subtracted out so that we are only comparing the
observed PM2.5 concentration due to fire with the BlueSky PM2.5 concentrations. While
several methods may be used to estimate the background concentration, including
running a photochemical model, one simple approach is to use PM observations from the
same sites for a time period under similar meteorological conditions but with no fire
activities in the region as background reference values. Here we use observations from
July 16-24 which had synoptic conditions similar to the study period. Mean values were
calculated for each hour of the day for the 9 days of the control period. At each station
during the study period these mean values were subtracted from the observed hourly
values. The subtraction may yield a negative value in some cases, which means that the
observed PM2.5 value at that particular hour is lower than average values for similar
conditions. The observed values are non-zero and vary with time. This is because that the
observed PM2.5 concentrations are from both primary and secondary sources and some of
these sources may differ from the period used for the calculation of the ‘background

values’. In addition, the meteorological conditions during the simulations period were

51

not exactly the same as the period used for background calculation. Although this is a
relatively crude way to take into account the background PM concentration, the
comparison between observed and estimated surface PM2.5 concentrations should reveal
how well the BlueSky predictions estimate the increase of PM concentration due to the
fires at a particular location.

Comparisons of BlueSky-predicted hourly PM2_5 concentrations with the adjusted
observations are shown in Figure 3.7 for the eight monitoring sites within the domain.
Since BlueSky only considers PM2.5 emissions from fires, the predicted PM2.5
concentrations at each site is zero except for times when a smoke plume from the fires
were advected to the location. Despite these complications, the comparisons show that
the BlueSky predicted PM” concentrations appear to emulate the observed peaks
although the timing of the peaks may not be the same in some instances. The magnitudes
of large increases in PM2.5 are comparable to observations, an important distinction not
seen in previous studies (Adkins et a1. 2003; Berg et al. 2003; Larkin et al. 2007). At all
stations, it appears that BlueSky does a better job capturing predicted surface
concentrations of PM2.5, in terms of both magnitude and timing, when the observed
concentrations are higher than 10 ug m'3. The passage of the smoke plume that was
traveling to the northeast on August 29 (Figure 3.6b) is visible in the surface
concentrations of the stations located in the northern section of the study area (Figures
3.7a, 3.7b, 3.7c, and 3.7d) as an increase in PM2_5 concentrations is evident in both the
model results and the observations. At all four of these stations the simulated PM2.5
increase occurred approximately 24 hours before the increase is seen in the observations.

This is most likely due to the overestimated model wind speeds during previous days

52

— Observed
(a) *Provoltv ' . . - ' ' . . (b) Hodford . . . . g ‘1 - Simulated

 

10

g 11
ll ill

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

o _\-"\I..
-10 . . . .
(c) Illinois Valley
20 T ' ' '
N
10 ml.
. l

.. . .W
‘1, )l;\

E o‘dL \‘\_

a

3

1':

23-10 . .

g (e)Chico

£20 T '

o

0

“210
N u.
2 1"
a. . ‘ \ i
' ln ‘.1. I 'I
I \ ‘ r‘ “ ,
J . ‘11", .1. i _ 1' I ‘1... v.— d"'| d"
' t (I 'l 'l

 

 

 

(g) Colusa

 

 

10»

 

 

 

 

 

 

 

 

   

 

 

-10

 

am 3/23 8/25 8/27 8/29 8/31 8/21 8/23 8/25 8/27 8/29 8/31
Date

Figure 3.7. Hourly PM2.5 concentrations at the 8 monitoring stations located in northern
California and southern Oregon. Solid black lines represent the observed adjusted
concentrations of PM2.5 while the gray dotted lines indicate the BlueSky estimated PM2.5

concentrations.

53

(Figure 3.4b and 3.4c), causing the smoke to be advected into these regions prematurely.
BlueSky was not able to predict the large increases seen in the observed concentration in
Figures 3.7c, 3.7f and 3.7h near the beginning and the middle of the study period. All of
these stations (except for Portola, Figure 3.71) are found at elevations lower than 100 m
and are located in the northern section of the Central Valley. A disadvantage of using
single station time series in comparing PM2.5 concentrations is the dependence of each
station to the location of the simulated smoke plume. For instance, a simulated smoke
plume may miss a station’s location by only one model grid point and the expected
increase in PM2.5 is then not produced by the simulation, giving an impression that the
BlueSky simulations to severely under-predict surface PM2_5 concentrations. While this
may be true, the under-prediction may also just be due to the “miss” of the smoke plume
due to differences in the simulated and observed upper level meteorological patterns in
wind speed and direction. Other possible explanations in BlueSky concentration errors
are uncertainties within the BlueSky framework itself. According to Larkin et al (2007),
fuel loadings (which are largely unknown for the majority of fires) and the consumption
calculations (which are dependent upon the fuel loadings) are two large sources of
uncertainty. The above results indicate that while some large increases of surface PM2.5
concentrations are accurately estimated by BlueSky, there is a need to improve these
predictions so that all of the observed increases in PM2_5 are estimated, regardless of their

magnitude.

54

Summag and Conclusion

BlueSky is a smoke modeling framework that integrates consumption, emissions,
meteorology, and dispersion models to predict surface concentrations of air pollutants,
such as PM2.5. BlueSky also provides the users with visual images of projected smoke
plume trajectories and shape. This framework is being used by smoke managers and
burn bosses to make ‘go’ and ‘no-go’ decisions concerning prescribed burns. Lately,
BlueSky is also being used in management decisions concerning wildfires and wildland
fire use fires. Despite its popularity, limited validation has been performed and the
uncertainties of the model under conditions different than those common in the Pacific
Northwest are largely unknown. The purpose of this study is to compare BlueSky
predicted surface PM2.5 concentrations with in-situ observations and to compare smoke
plume shape and trajectories with remotely sensed images of significant smoke plumes
and aerosol optical depth to qualitatively assess the accuracy of BlueSky predicted
results.

The performance of the MM5 model in estimating meteorological fields were
evaluated first as accurate meteorological conditions are essential to the transport and
dispersion of PM from fires. The simulations captured the trend and day-to-day variation
in temperature and wind speeds due to changes in synoptic conditions. In most cases, the
simulated temperatures were found to have a large warm bias during the nighttime and a
cold bias during the day, leading to an overall smaller diurnal range than is actually
observed. The simulations tended to overestimate wind speed for near calm conditions
while the underpredicted wind speeds for strong wind conditions. On most days, the

simulated wind directions, however, agree quite well with the observations. The

55

simulated vertical profiles of wind speed, direction, potential temperature, and mixing
ratio are also in good agreement with the rawinsonde observations with the exception of
August 31 which was under different synoptic conditions than the rest of the study
period. On all days, there is a wet bias in the lower atmosphere up to 3 km, and a cold
bias in the lower atmosphere. The model reproduced the observed wind direction and
vertical shear of the wind direction, which is very important for transport simulations.

Next, PM2.5 surface concentrations and trajectories simulated from BlueSky were
compared with observations. It is difficult to quantitatively compare BlueSky predicted
PM25 concentrations with observations as the latter contain PM concentrations from all
sources, not just from fire emissions. At each station, a ‘background value’ was estimated
for each hour of the day using hourly averages from a fire-free period with similar
weather conditions. The comparison of the simulated and observed hourly PM2.5
concentrations at eight stations indicate that BlueSky is in general able to simulate the
observed large increases in PM2_5, although there are some discrepancies in the simulated
and observed timing of the peaks in some cases. Comparisons of the BlueSky predicted
particle plumes with the satellite images show reasonable agreement in the orientation
and aerial extension of the fire plumes.

This study shows that the BlueSky smoke modeling system can serve as a tool for
estimating long range transport of smoke plumes. BlueSky can be used to assess smoke
plume shape and orientation, but it needs improvement in capturing the magnitudes of
increases in PM2.5 concentrations. More accurate reporting of wildfire size and locations
within the national [CS-209 reports, provided by fire managers, will help to improve

emissions estimations within BlueSky. Also, sensitivity studies of the emissions and

56

consumption models are useful to help identify areas of uncertainty within these models.
Removing these uncertainties and improving each models estimations will lead to
increased accuracy of the BlueSky predicted surface concentrations as well as the overall

model performance.

57

PART IV
SOUTHERN CALIFORNIA, OCTOBER 2007
Introduction

On October 20, 2007, a series of wildfires broke out across southern California.
For the following 19 days, aided by unusually strong Santa Ana Winds (gusting as high
as 140 km/hr) and hot, dry conditions, these wildfires destroyed 1500 homes, burned over
500,000 forested acres, caused 9 deaths and 85 injuries, and forced over 900,000 people
to be evacuated (the largest evacuation in the history of California) (Flaccus 2007). A
state of emergency was declared in seven counties (Archibold 2007) and extensive power
outages occurred throughout the region. At the height of the extreme, wind-driven fire
activity, 17 separate fires were burning and PMIO and PM2.5 concentrations reached and
maintained unhealthy levels for several hours. The extensive destruction and publicity
caused local, state and federal agencies to closely examine the protocols used in the
suppression efforts to aid in prevention and suppression of future outbreaks.

In addition to direct fire threats of loss of life, property and wildlife, secondary
threats such as reduced visibility and increased air pollution also exist, making it unsafe
to be outdoors. Air quality and wildfire managers need to be able to make fast and
accurate decisions about impending risks current fires and their accompanying smoke
release may cause, and they must be able to convey that information to the public. The
EPA has been instrumental in establishing air quality initiatives, setting air pollution
standards and aiding in the design and implementation of numerous air quality models

(Agency 2006).

58

Modeling of aerosol dispersion over the western United States is complicated by
the mountainous terrain which greatly affects this region’s meteorology (Kim and
Stockwell 2007). Furthermore, the long-term effects of smoke are dependent upon the
fire environment and the fires characteristics (Reinhardt, Keane, and Brown 2001),
further increasing the complexity of modeling smoke dispersion in the West. Several
models, such as CALPUFF (Scire, Strimaitis, and Yamartino 2000), SASEM (Sestak and
Riebau 1988), VSMOKE (Lavdas 1996), and TSARS Plus (Hummel and Rafsnider 1995)
have been designed to predict fire emissions, but these models are either too simplistic to
represent the complexity of smoke dispersion in complex terrain, or too complex to be
implemented by users with limited computer experience and/or scientific background. In
order to simulate smoke dispersion and the subsequent effects on air quality, dispersion
models must integrate fire spread, fire emissions, atmospheric flow, smoke dispersion
and chemical reactions (Miranda 2004). Clearly, a balance must be maintained between
the components needed to accurately simulate smoke dispersion and a user-friendly
interface that allows the model to be utilized for different applications.

In early 2000, under the National Fire Plan, the USDA Forest Service AirFire
team launched a modeling system that predicts smoke trajectories and ground
concentrations from wildland fires (Ferguson, Peterson, and Acheson 2001). This
framework, coined BlueSky, integrates meteorology, emissions, dispersion and trajectory
models to produce these estimations and the results are made available to fire managers
and air resource regulators. BlueSky is an interactive tool where users can choose which
components they need, and when used along with GIS, users can zoom in on areas of

interest, step through time, and overlay other GIS layers (geopolitical boundaries,

59

meteorological fields, topography, etc) in order to interactively view mapped information
and analyze the underlying data (O'Neill et al. 2005).

BlueSky was designed with the users in mind, with necessary input requirements
limited to only fire information such as date and location. After the information is
entered BlueSky employs a mesoscale model, to provide feedback on wildland fire
smoke impacts. Other secondary information can be provided, but if additional
information is not available, default values are obtained from specified databases and
look-up tables. Originally designed as a tool for making ‘go’ and ‘no-go’ decisions for
prescribed burns, BlueSky’s applications have been expanded to include smoke impact
assessments concerning wildland fire use (W FU) fires and wildfires. Evaluation of the
framework has been focused mainly on fires in the Pacific Northwest (Adkins et al. 2003;
Berg et al. 2003; Larkin et al. 2007; BSRW 2006) and Southeast (Pouliot et al. 2005;
Craig et al. 2007).

Although none of the existing validation efforts have been subjected to peer
review, many of the studies listed above have produced similar results. In a study
focused on the 2002 Hayman Fire in Colorado, Adkins et al (2003) found the location
and timing of predicted smoke impacts to agree with observations while the predicted
PM2.5 concentrations were orders of magnitudes smaller than observations. Larkin et al
(2007) conducted a similar study using data collected during the 2001 Rex Creek
Wildfire in Washington and the 2005 Frank Church Wildfire in Idaho. Similar to Adkins
et al (2003), the comparisons between smoke plume trajectories and shapes were found to
agree quite well with satellite images, but BlueSky again failed to accurately predict the

observed magnitudes of PM2.5 concentrations. Results from these studies show that while

60

BlueSky can accurately predict smoke trajectories and the timing of smoke impacts,
predictions of ground concentrations are poor. This paper presents results of the
performance of the BlueSky modeling framework in simulating smoke dispersion and
surface PM2_5 impacts from the October 2007 southern California wildfire outbreaks by
using two different meteorological initialization data sets. The use of two different sets
of initialization data will help to identify the sensitivity accurate meteorological

estimations are to the overall performance of BlueSky.

Methods
Study Area

California’s large area (almost 10° longitude and more than 30° latitude) leads to
vastly different climate and vegetation types between the northern and southern areas of
the state. Like the mountainous basin and range provinces of the north, southern
California has varying topographical features such as hills, mountains, plains and mesas.
The Transverse Range, which occupies an area approximately 482 km wide beginning
near Point Arguello (southern Califomia’s western most point) to the east at the Little
San Bemardino mountains, is the only east — west trending mountain and valley system
in California (Durrenberger 1968). The Transverse Ranges western portion is occupied
by the Los Angeles basin, an area of relatively flat topography, broken up by gentle hills
and mesas and divided by the Newport — Inglewood fault zone. The eastern extent of
southern California is separated from the rest of the United States by the Sierra Nevada
mountain range. The highest peak is Mt Whitney at an elevation of 4,421 m (14,505 ft).

The Sierra Nevada range aids in the creation of strong, dry, downslope adiabatic winds

61

known as the Santa Ana’s. These winds occur when a high pressure system is located
over the Great Basin and a low forms over the coast, causing warm and dry (due to their
interior origination) winds to move from the east to the west accelerating through
mountain passes.

Western California is bordered by the Pacific Ocean, which creates drastically
different climate and vegetation patterns in the western, coastal regions than in the
eastern, inland regions. The moderating effects of the ocean produce large differences
between inland and coastal temperature and precipitation. Generally, near the coasts
lower annual temperatures are seen, with yearly maximums ~20°C while approximately
250 km inland temperatures are ~28°C (WRCC 2008). Yearly total precipitation near the
coast averages approximately 33 cm, while only one third of that amount falls over the
inland area, as average rainfall is closer to 11 cm. Precipitation occurs predominately
during the winter season, beginning in late October or November and persisting through
May or June (Durrenberger 1968).

Vegetation patterns shift from thicker, dense forests at higher elevations to sparse
oak forests, shrubs and Chaparral at lower elevations, valley bottoms and near the coast.
Ponderosa pine forests, composed of ponderosa pine, white fir, sugar pine, douglas fir
and incense cedar, are found between 1,524 m and 2,438 m (MSL) and are the most
extensive forests found within southern California (Durrenberger 1959). Near the upper
reaches of the ponderosa pine forests, white fir and sugar pine intermix with red fir as the
ponderosa pine forests transition to the red fir forests. Red fir forests are scattered and
sparse over the landscape. Generally, the lower elevations of southern California are

dominated by oak woodland and Chaparral. In the oak woodland zones, oak trees occur

62

erratically across grass covered hills and valleys. Foothill woodlands, one of two types of
oakland woodlands, are a transition zone between valley bottom prairie grasses and the
forested slopes above (Durrenberger 1959). Chaparral, one of the most characteristic
vegetation types of southern California, is composed of many different types of shrubs,
but is dominated by evergreen plants. These shrubs have a long root system that allows
them to survive the long, hot and dry summers of southern California. The ability to
survive harsh conditions and a location on drier slopes makes Chaparral an ideal fuel

source for wildfires.

Model Setup and Simulation

There are two main components necessary for running the BlueSky Smoke
Modeling framework: meteorology and fire input information. Predictions of the
meteorological fields during October of 2007 were generated using the Pennsylvania
State University/NCAR mesoscale model (MM5) (Grell, Dudhia, and Stauffer 1994).
Terrain and domain information was provided by California and Nevada Smoke and Air
Committee (CAN SAC). Three nested domains were used, with the innermost 4 km
domain covering the California and Nevada region, while the outermost 36 km domain
encompassed the western US and eastern edge of the Pacific Ocean (Figure 4.1). To
evaluate how BlueSky performs under different meteorological simulations, two different
meteorological inputs were used. Initial and boundary conditions for the first simulation
were provided by National Center for Environmental Prediction (NCEP) 40 km Eta
forecasts and for the second simulation, initial and boundary conditions came from the

NCEP 32 km North American Regional Reanalysis (NARR) output. BlueSky aerosol

63

transport and dispersion is simulated using CALPUFF, a 2 dimensional, non-steady state
Lagrangian puff dispersion model which simulates time and space varying pollutant
transport, transformation and removal (Scire, Strimaitis, and Yamartino 2000). The
Satellite Mapping Automatic Reanalysis Tool for Fire Incident Reconciliation
(SMARTFIRE) was used to obtain information on fire location, size and emissions. In
order to create a comprehensive and spatially accurate data set, SMARTFIRE combines
both wildfire incident data from the national wildfire (ICS-209) reports along with
satellite-detected fire data (Craig et a1. 2007). PM2.5 emissions were calculated within
BlueSky using Consume/EPM v1.03 while the Fuel Characteristic Classification System

(FCCS) was used to obtain fuel characteristics information.

 

 

 

 

Figure 4.1. Modeling domains used within this study. D01 is the 36km domain, D02 is
the 12km domain and D03 is the 4km domain. The 12km and 4km domains are one-way
nested domains.

64

Surface and Satellite Observations

Surface meteorological parameters simulated by the PSU/NCAR Mesoscale
Model (MM5) were validated against hourly surface data from the National Climatic
Data Center website for 75 stations (Figure 4.2, Table 4.1) within the region. NCDC
combines surface meteorological data from the National Weather Service (NWS) first-
order, ASOS and COOP networks. Upper air observational data were obtained for six
stations at OOZ (16 PST) and 12Z (0400 PST) from the University of Wyoming Weather
Web website. The names of the upper air stations are listed on Figure 4.2. The output
from the BlueSky Smoke Modeling Framework simulations includes hourly PM2.5
concentrations and a visual image of the smoke plume shape and trajectory. To compare
simulated surface concentrations of PM2.5 to observed concentrations, hourly PM2.5
observations were obtained from the AIRNow database and the California Air Resources
Board (ARB) database. Sixty-one stations cover southern and central California, and five
stations are located in nearby southeastern Nevada (Figure 4.3, Table 4.2). Satellite
observations for comparison with BlueSky images were obtained from the Moderate
Resolution Imaging Spectrometer (MODIS) Satellite Fire Detection, Hazard Mapping
System (HMS). These images allow a qualitative comparison between simulated and
observed smoke plume shapes and orientations. Each image is a composite of all
detected smoke plumes for a given 24 hour period. HMS imagery is at times not in
agreement with aerosol optical depth data due to limitations in HMS fire detection.
When fires are too small and/or are of weak intensity, the HMS imagery has a difficult
time in distinguishing the fire from the background, especially in the western United

States summer season when surface temperatures are very hot. Also, satellites (using the

65

3.9 micron band) are not able to see through clouds other than cirrus, so fires occurring in
those areas will go undetected (McNamara et al. 2004). To gain a more quantitative
comparison with the concentrations of PM2.5 within the BlueSky predicted smoke
plume, GOES Aerosol/Smoke Product (GASP) images are employed. These images are
hourly displays of the concentration of aerosol in the atmosphere. GASP images
represent the thickness of air columns due to aerosols (or other airborne particles).
Higher AOD values indicate more aerosols in the column and so less light is able to pass

through.

66

doufiw 2d 88: as 3862 88.2.: 62 263 8283. 8th 8862 8820 82m 6:22:88 06 e2 68: 2286
688: 83.28 ~56 06 88062 Koo. 82a 2: .88 dm 1 2 8880 826 808693 m2: 866; 2 68: 083
8283. 2a 8&5 o 2: Bob 32258 >66 02.5 23 828% 32368088 08.2; mm Sod «86 3.5on N6 95me

. . 3 res: . . 3 hour
iburp 3.0m: . 3.08.. 25mg? 2.0.8

 

      

 

80.88 825m 0
Scream .__< 825 ‘
z .3”

 

z .36

 

 

z .09.

 

 

 

Z adv

67

8282030 ha 3&5 mocEwfi .D. 233 mcocmtomno Rama—8020:. 885m 85ch .m...

 

wmmmmmmmmmmmmmmmmmmmmmmmmmm

a

Own
QmOB
WE
5v
cm
@—
md:
2
mm
wda
_.mv
Wm
Nd
mm
5.8
fimmm
m3
snow
v
v
md—o—
m4:
odm
on
5h
w.mm
m.m%

”Eng

mcdeT
mm”: T
03.: T
SAINT
mmchT
www.mmT
55.3 T
mmmfi. T
53ng
_o~.mNT
NSNNT
wamNNT
SUN—NT
mwod— T
and. T
omm.w_ T
0.: T
3.9 T
VNNT
CONNT
wofiomT
_N_.NNT
mmame
mw._NT
v8.3 T
mmH—T
52.3 T

03.3“
@va
wodm
mafim
Smém
mafim
39mm
mmw.~m
mam
cm _ .om
ooflwm
mofim
vowSm
Swim
3.0m
Smém
Qmm
homém
$0.0m
mhdm
Nomdm
$ch
cmofim
mwmdm
amodm
m.mm
mém

<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0

gag

m< 03mm?— <HM<E <FZ<m
mm< mgaam
m< mZ><>> ZEOH <Z< <HZ<m
m< A<E0~ZDE Z~5m5<
OAAmDmum/x Hag
<me< EDA
m< O0 memmmE m=P>Am
damn—2H
mAH>OMO
m< A<n=0~752 ZED
AmO>><v <m0m <E<m
m< HHZH 009072?» Z<m
m< gag ZOHv—0OHm
AmO>><V Ogdm<2< 0
m< 40.7: mbEmmO> 02mm??—
m< <ZmQ<mm<m-mA<DZmAO-MZ<mMDm
m< 230952 <ZOMO0
m< >b0 m<mm .05
800v >ODm nay—OOZ
moccv >ODm Our—MODE
m< ZO>Z< 0 mqu
Sam—H. Mn< QM<>>><$
m< :ZDOU->.E0 OHmm—QOE
<1—DmZHme >mmmPZOE
m< min—22m
m<02 20.550me m2<0

m5§<m Egg
oEaz

 

$039885 ofimfiwoow :ocflm :ocmtompo 30%298802 Ad ozah

ovommh
o_wmmh
whoNNh
bemoan
o_mvam
om_voo
ww_hv>
nw_th
wvomVB
moomwh
hmovmh
owavmh
omovmh
omommh
oawmmh
owwmmb
mmmomh
mcficmh
chmhoo
mbmhoa
mvwmmh
mmavmh
omovmh
m_avmh
omwmmh
omammh
0m_ooc
minamw

 

o—‘NMVWOI‘OOQOF‘NMVWOF
u—n—nv—Iv—nu—v—tv—t—nv—tv—tNNNNNNNN

v-‘NMVI'WOFOOO‘

68

0000030000 0.0 0000: 000.0me .3. 0.0.3 2.0090030 30.00.8082: 000050 80.00% .06

 

git/20:1)mmmmmmmmmmmmwmmmmmmmmmmmm

0m .
000
0.00m
.N
0.000
00.

0N0

Now
0%
0.000
m .
N .0 .
00
NE...0.|:0.0|0>.0|.M—

000.0. T
000.0..-
00.0. T
0000. T
0.0.0. T
0.0. T
08me
0.0.0. T
N.NN.-
mm0..NT
0. ..0NT
00..NT
000..NT
000.0. T
0.0. T
000.0. T
0N0.w. T
000.0. T
0.NNT
0000. T
0.0. T
000.0NT
000.0. T
000.0. T
000.0. T
000..NT
000.0. T
0.0.0NT

India. :3 locaufi llssm

000.00
000.00
000.00
0.0.00
000.00
000.00
00.00
000.00
0.00
000.00
000.00
000.00
mm. .00
000.00
0.0.00
000.00
000.00
000.Nm
000.00
0.0.00
00
00.00
0.0.N0
00.00
000.00
00000
000.00
03.00

<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0

.<0

<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0

803$ 000500.000

n.< ..r..m.00<Q->>O.Hmm<m. tm00<fl

0< 000002
0 000 00000002 005
0< <ZE<H<0 $00,020 5.25
0< 0.70 0020000 2120
005000 0000
<202<0
0000 0003000
000>20m0<3
<000<2
0< 00002 0E02<00<m
010300.020”? 00,000
0< 000.000
803$ 0752 Sim;
0000 030902 0000000040
00000720
00 23000000000 25
02020000
0000072 mou—
<00202 5.2%
m0: 720 0000
603$ 00520
0< 00.70 0005.20
EB 0005 $0000
000200000 93.00
000$ 0005 00000

Our. >Q<mm0<25m00m2
2:07.

 

“00250.50 .3 2.3

00wmm0
0.00N0
0000N0
000NNO
0NONNO
w0wmm0
000000
000000
.00000
000000
000000
0mm0~0
0mw0N0
0000N0
00wm~0
00wm~0
0mwmm0
000NNO
000000
000000
0N0000
0N0000
0w.000
000000
0N.000
00.000
00w0~0
0.00N0
a 03m

 

00
00
00
N0
.0
00
00
00
00
00
00

00
N0
.0
00
00
mm
0m
0m
0m
00
mm
mm
.m
00
0N
mm

69

0000000000(13000000000330.3023

gHH

00.000
00.000
00.000
00.00
00.00
00..
000
0.0. 0.
0.000 .
000.
0.00. .
..00. .
0.000.
0.000.
.00
0.000.
000
0.000
000.
000
00
0. .0
..0.
000
0.00

.0
.33

00.0..-
00.0..-
00.0..-
0.0..-
00.00.-
00.00.-

000.0. .-
000.0..-
000.0. .-
000.0. .-
0000. .-

0.0..-

0000. .-
000.0. .-

0.0..-

000.0..-
000.0..-
00..0. .-

000. .-

00.0..-

0.0..-
00.0..-
00.00.-

Rods-
802 _-
3.0.0. _-
00.3% gala...- gum.

00.00
00.00
00.00
00.00
00.00
00.00
000.00
000.00
000.00
0. 0.00
000.00
0.0.00
000.00
00.00
0.0.00
.0000
000.00
000.00
0.0.00
000.00
000.00
000.00
0.0.00
000.00
.0000
000.00

>7.
>7.
>Z
<0
<0
<0
>7.
>7.
>7.
>7.
>7.
>Z
>7.
>7.
>7.
>7.
>7.
>7.
>7.
>7.
<0
<0
<0
<0
<0
<0

004m
07.0.”.

0.00% Hmmmmg>g0mm2
M.<.>.<~...>.\00m..0 7.<0
mm< 0mmm2002<>
.07.. 07.<-.v.<0
mm< .00-”.000
n.< -.<n..0.75.>. <00DEmZ7E$
m< -.<n..0.ZD.>. 0v.-.m.
.00.).<M.v <v.m.~.0m.
040...”. >m~.m.n. v.00-.m.>0-.
0<<Z 204-.<n.

DAM-Fm 07.<-.-.m.> >-.m.
Hm§< ...<n.07.0.0
m<0m.> m<-. 7.
n.< v.00y. .50.me >~.DU~.m.S.
mn.< 0.4-0.7.
n.< H02. Z<~.~.<00$. 0<0m.> 0<-.
000 0.07.: I<n.020...
m.>.....D0m.Xm. 7.00m002m...
n.< -.<.>.~.m.:-.. 007..~.n.0 Sin.
0<7. mm>mmm 05.00qu
n.< .00 <n.< 7.
n.< -.<n..0.75.>. mmAmOM 00<n.
m< QM<ZXO

n.< 20.0.7504 <~.<m.~.<m <..-Z<0
0:37.

 

F052950 .0... 2%...

000005030 0.0 0000: 0.0.-0.00.0 .0. 0.0.3 20.003530 000.00.800.00. 0000.03. 000.:00 .0...

00000
00000
00000
00000
00000
00000
.0.000
000000
000000
000000
000000
000000
000000
000000
000000
000000
000000
000000
000000
000000
00.000
000000
000000
000000
000000
000000
AH—sz

 

.0
00
00
00
00
00
00
00
00
00
.0
00
00
00
00
00
00

00
00
.0

00
00
00
00

7O

00000.20? 2.0 00.. .003 0.0800 0060: 003-000 00.0 05 3.8.0:. 02. 0.00.0
2.... 0.8.3 88000 0.0.-000.008 .2. 080m $00083. 0.< 0.500.000 0... .000 90.00.00 80008 00.00.0008 3ozy..<
2.. 800.. 000.800 003 80.0 325. 0.500. 002.3 000.00% 0509.008 .32.”: 080.20 80.00.50 000.000 0.0 050.".

           

 

 

 

 

 

. . .5 as: . . .5 0.00..
3.0 0: 3.00: . 3.0000 3.0000 . 2.0.8
0:03.02 05.53.32 m¢< <0 5
0:252 0:08.82 .3202 o 2.3“
0008.00
80. , 0050020 ., .
. 009:3 z 0.00
22.9930 .
88- - .
0:00.“.
Sma- -
2.0.00
. , i . _ . ,. 00300
.. 00.50 2.0.8

 

 

 

Z .000

71

 

302MH<
30E<
30§<
BOZE<
302E<
BOZMH<
302MH<
Boza<
Boza<
30é<
30é<
302MH<
302MH<
BOZE<
30é<
30E<
BOZE<
30E<
302MH<
302MH<
30§<
Boza<

0.00.50 Z

own
3».
av
.c
vm
0w
3.
No
5..
o.
w.
2.
mm
b.
0N
No
mv.

«mm
050

.5. 00.056...

mwmmvfi. .-
mmmuwd. .-
wmmom..m.-
c.0306. .-
awmaNmT
SGNw. .-
mwome. .-
mmmmmd. .-
wmovod. .-
whowmw. .-
mmmwvfi. .-
wmoo..mm.-
Nmmpd. .-
#0393..-
mmwmb..m.-
mmmvw..m.-
oomwh..m.-
VEN. .m. .-
$2.6. .-
voomcw. .-
omcooé. .-

3.8.2 .-
lulaeé =3

whNOSéM
wbtmmdm
mmwmvfiwm
Nw.mm
3&8. .wm
3008.3”
3me 6m
mmdm
. . .cmm.mm
whhhnodm
...o$.mm
m... mdm
owmgfiom
enemomdm
mmmwOmdm
mhmhdm
3.00%.?”
.. . .0. 0m
03$. 0m
mmmmmodm
ommoamdm
whamvmdm
0.0.5.3..

<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
>2
>7.
>Z
>Z
>7.
35m

 

0500.0... - 3.9.85.
Hm ><B<E<m :58 I 02.22%.
300000.322022 - qu>mmom

Em2<z<
mmwmm 92.0..

Hm 22.2 .2 - $1.002... m0.
.955 - 3.00230
2<200200
m>< <0 wmmm - 01.823520
5%. - 00.236
Hm 002.5 I 00.08.20
pm .555 m I 950...;
pm a. I 0282”.
05m. ”.5220... - $0.00
5.0.20
<22<Nz<2 - 00.20
2002.2 - mmOmeZ.
2.56 0..
mmmu< $220.0,
>m...<> 20$...

20......

9. 0220
EGmuflu—m any 0562

00.38.80. 0.....8m00w 00...“; 00.29.0800 2.)... .N... 0.00...

.oowmcoo
N.o0mcoc
coco.©oc
hooommoc
mooo.voo
mo..hmoc
c.oo>moc
vooo.moo
v.ooawow
©OCOmNoc
mOOOmNoc
Nooo.moc
wooom.oc
moo...oo
.oovnooo
NOOOhooo
booo.ooo
moommoomm
.cWOmoomm
meOmoomm
mmo0moomm
ONOOmoomm
A: 3%

 

Ov-‘NMV'IOOhooao—‘N
c—n—cv—qc—tu—tv—tu—no—d—qc—INNN

v-‘NMVV‘JOFOOO\

72

 

302M..<
302M..<
30 ZM.<
302m.<
302~=<
3ozyn<
302~=<
30727..
302M..<
302M..<
3ozyn<
302M..<
302E<
302M=<
302E<
302M..<
302m.<
3022....
3025.2
302~=<
302m.<
302M..<
3022.2.

2203......7...

o.
vm
cm

.5
on
5m
. .
M...
m
cm
05
o. .
m
cm
.
om
m5m
3.
mm .
cw
m
Nov
.5. 20.3.6...

.omm..o. .-
N5vwo.w. .-
mwmvmd. .-
.w.w. .-
3032. .-
m5.©..m.-
$.80me
55..N.-
50.02..N.-
N5vow. . N.-
500513.-
3232me
wmoood. .-
NNNONNNT
033.2-
w55om.mm.-
5.22.5. .-
m50. .-
5..V5o.5. .-
50320. .-
w5~©...m.-
22.9% . N.-

mwmmm.5. .-
0.05.9204

w5~momém
m55~ém
5000.?va
vvvvovém
3.2m
owwwfldm
mmmwmc5m
mmmwgwm
mammdm
mmmme.5m
w555moém
omowvoém
m5mvém
w55mwv.5m
mmwo.5m
mmwmc5.5m
w5mm . 5.Nm
mmmmmw.mm
w555m..mm
owm .a5Nm
mwwmwcdm
mmmwmm.wm
$5250.?”

25.. 2...... slam

<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0

2. 200202 2223. 0.2 - 0.2 22
.2 2222000 - >222> .22
2>< 2.0 - 2.0
0.2.022 -22...

02 22222002 - 2220 0222202..
220 2222
.2 E: - 0.22002
2.2; 0.2
222.20.;

,2 202222. - 220. 222
.2 2 002202
22302022 - 2.22.). 22.222
2222222 22722
220 2003022
220222 20222
.2 22222222 .. 002.0222". 222
2 222222222 - 002.0 722
22.22
00.020022
2>< 0003022 - 2020 22
20220...

2 2 - 0222.22022-

..2 22.2... .3 - 2202.222 222.
:Avmasm MO OEQZ

825.80 .3. 2.3

.ooM...©
Noom...c
voo....e
coco...o
5ooo...c
mooo.o.c
mooomooo
amammooc
mOOmmooc
mOOOmwoo
voommwoo
woo.mwoo
..oonoo
.oo..woc
mOOm55oo
moo0m5oc
o.o.m5oc
coo.m5oc
Noo.m5oo
m0OOm5oo
N.OO5©O©
c.005ooo
.ooomooc
A: 35

 

mv

mv
NV
3.
ow
mm
mm
5m
0m
mm
cm
mm
mm
.m
om
mm
mm
5N
0N
mm
vm
mm

73

 

mm< <0
mm< <0
mM< <0
md< <0
mm< <0
mm< <0
mM< <0
mm< <0
mm< <0
mm< <0
mm< <0
mm< <0
302M.<
3OZM=<
302M.<

302M;
0.. 203.0. . 7.. I

.m.
EN
9%
N
Om
Nam
5.
0
cm
5
N5m
0mm.
mm.
m
.N

w .
.5. 00.0030...

.05. .-
0.0.5..-
92w..-
NV..N.-
and..-
5.x. .-
3.me
5m..~.-
N56. .-
mNNT
and. .-
No.0..-
cm©m..mm.-
5ccmN.NN.-
w55m5..m.-

m55..~.-
063mm. 0?—

mmwm.mm
300.3
mmm5.mm
mm0m.wm
mNmm.©m
3m.0m
Nwwv.5m
5.m0.5m
.3502“
v. .m.wm
£05.?
mm00.5m
ownwcowm
NNNNO..wm
300.09%
mmmmmmdm

«03:02. 2 I332

82580 .3. 2...;

<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0
<0

>3... Hm 759400 I Qqammmmm<m
Zm—MDm Z<> I <EOA <53.
m0.>~.m.m HmmEOn. I m5d>ZMmM
QM. qu>m0Dmm I m>OM0 vim
Hm :0MDIO 7. - <....<m.>
m0.>~.m.m Hmmmon. I qu>0§mm
Hm Hmm<7mz m - MOOAMDH
Hm. ZOHAMN<I I ZOHMOOHm
m.>< <41=> Z - m.>Ox.0
m>< ZOmmmnEma I <n.<Z
MEZm0 MObm.>I 0<Ad> mEEMmO>
mat/”Em HmmEOn. I HmMDImZE

.30.me

Hm mZEDAODH 3m - Onmqq<>
QZ<AQOO>9

m.><Q

Isaasm .o «'52

mv.m
voov
m5Nm
550m
Nmom
m5~m
coca
voom
oNCm
mmcm
mmmm
V5Nm
OCOOmmoc
VOOOmmcc
moo.m..0
VOOOM..©
n=.Q=m

 

.0
co
0m
mm
5m
cm
mm
vm
mm
mm
.m
om
av
wv
5v
0%

74

Results and discussion
Wildland Fire and Synoptic Conditions During the Study Period

During the ten days prior to the extreme fire outbreaks, fire activity in southern
California was minimal with. Major changes occurred on October 22“d as nineteen new
fires were reported, ten of which were large and uncontained. Firefighters in the southern
California counties of Los Angeles, Ventura, Orange, Santa Barbara, Riverside, San
Diego, and San Bemardino battled extreme, wind-driven fire behavior for. the next five
days. While the number of large fires did not increase again during this period, many
smaller fires ignited, at times as many as forty a day. October 27th finally brought relief
to the fire fighters as only moderate, minimal and smoldering fire activities were
reported. By October 30th fire fighters were able to get five large fires contained and by
November that number had increased to eight. The Poomacha and Santiago fires were
the longest lasting fires, with full containment not achieved until November 9‘“.

At the beginning of the study period, when fire activity was minimal, the west
cost was under the influence of a strong upper level ridge. Beginning on October 15, a
strong trough off the southwest coast of Canada and an upper level ridge over Mexico
caused moderate (12 -15 m s") 500 hPa west-southwesterly winds over the US. west
coast. As this trough over Canada continued to move east, the upper level ridge over
Mexico moved north, and by October 18 a strong 500 hPa contour gradient was created
between the two regions (Figure 4.4a). Wind flow during this time was generally zonal
with strong (35 - 40 m s '1) westerly 500 hPa winds over the western US. By the 21St the
ridge had moved off to the east, and the west coast was dominated by a strong upper level

trough, causing 500 hPa northwesterly winds over most of California and Nevada,

75

 

 

 

a October 18,2007

 

 

 

 

 

 

 

Figure 4.4 Plots of 500 hPa height (left) and wind speed (right) synoptic plots at OOZ
(1600 PST). Light shading in the wind speed panels (right) indicates weaker 500 hPa
wind speeds while darker shading indicates stronger wind speeds. Contours in the left
panel represent 500 hPa isoheights.

76

contributing to the extreme wildfire activity. This system was quickly replaced with a
upper level ridge, building off the coast of southern California, creating moderate, north-
northeasterly 500 hPa winds over California. As this air descended and moved to the
west of the Sierra Nevadas, they increased in intensity and speed. This upper level ridge
dominated the west coast for the next three days, creating moderate (5-15 m s") 500 hPa
easterly winds over southern California (Figure 4.4b). During the morning of October
26th another upper level rige began to form over the Baha peninsula while a trough began
to form over the Pacific Northwest. These two systems created 500 hPa southwesterly
winds ranging in speed from 10 m 5" near central and southern California to 25 m s'1 in
northern California. By October 27‘h the cut-off low over the Pacific Northwest was
developed and had moved south so it was just off the coast of central California (Figure
4.4c), and light to moderate (5-10 m s'l)southwesterly winds dominated southern

California.

Meteorological Fields

The smoke and aerosol predictions simulated by the BlueSky framework
are dependent upon the accuracy of the input information. Figure 4.5 is a comparison of
simulated and observed surface temperature, mixing ratio, and wind speed for both the
Eta and NARR runs. The circles represent hourly average values at all surface
meteorological stations throughout the entire study period. Gray circles indicate daytime
values (0700 PST to 1900 PST) while the black circles indicate nighttime values (2000

PST to 0600 PST).

77

(:3) Daytime
4

    
 

 

 

 

 

A 45 MM5-NARR
5') 3 35
v ‘ O
2 0%
3
E 2 25
g. o
1 15
I‘l-J 8 °
O
E 5 5 (b o
E 0
‘1 7'15 4
-15 -5 5 15 2 35 45 -15 -5 5 15 25 35 45
Observed Temperature (00)
FA 16 1st 0
'3’ 14 14 036 8
Q 0 0:80 0 0g
12 12 o 00 2% o
.9
a geogooo . 9 asgwso
DE 10- cm) ° 10- o o
e 0“” ° °
._ 3. o 8‘»
.5 I
2 g 3% 0 Q) g5)
13 6’ o 6 o
2 O 0 Q %8 0
m _ .
3 4 a I 4 o 0 g
.g 0 0 09 go o
co 2* o 2» o 0 08
0 . . m; 0 . . . . .
0 2 4 6 8 10 12 14 16 0 8 10 12 14 16

Observed Mixing Ratio (9 kg 56

 

Simulated Wind Speed (m s'1)

 

 

 

_._ . I“ 03.0 A.———I——A_ ‘1 00
0 3 6 9 12 15 18 21 24
Observed Wind oSpee3d (m6 s 1)

Figure 4.5 (a) Simulated and observed hourly temperature (top), mixing ratios (center),

and wind speed (bottom) at all surface meteorological stations. Each circle indicates an
hourly daytime average (0700 PST to 1900 PST).

78

(b) Nighttime
45

  
   
  
    

co
9'" .

N
9"

 
   

0'1

Simulated Temperature (OC)
5'” En‘

 

J.
01

. ,W, W i .—1 ,, W . , ,Wv,
15 25 35 45 —15 —5 5 15 25 35 45
Observed Temperature (00)

 

 

 

 

 
  
  
 
   

 

 

  
   

161
p
31 14-
93 12
.9 g '
g 10 ' 0:. e
U) 0
I: .
:2 8 .0, :°
2 O
‘o 6’ . O
o
E 4» '-
3 o 6
2% 2’ 0"."
0 . W . _L_ . s O ,W. . . r i . . .
0 2 4 6 8 0 12 14 16 0 2 4 6 8 10 12 14 16
Observed Mixing Ratio (9 kg'1)
24’ 24'
p
I; 21 . . 21
v O 0 . O
'8 18 . 03:. . z. . 18
g 15- 15
(n
'o 12 0 o 12
.E
0 o
E . I e . .
2
L“
:1
.E
(I)

owmcp

 

1'2 1‘5 18 212.3% 3 .
Observed Wind Speed (m 5'1)

Figure 4.5 (b) Simulated and observed hourly temperature (top), mixing ratios (center),

and wind speed (bottom) at all surface meteorological stations. Each circle indicates an
hourly nighttime average (2000 PST to 0600 PST).

79

Simulated surface daytime temperatures in both models appear to have no distinct
cold or warm bias when temperatures are below 25°C (77°F). As surface temperatures
increase, a slight cold bias becomes apparent, with most discrepancies being less than
five degrees. Comparisons between simulated and observed nighttime temperatures,
however, show a warm bias in both simulations, which is consistent with previous studies
(Zhong, In, and Clements 2007). Similar to what was found by Fusina et al (2007), a
simulated nighttime warm bias along with a daytime cold bias, results in a smaller
simulated diurnal temperature range. Consequently, the observed extreme (maximum
and minimum) temperatures are not reached within the model.

The mixing ratio plots in Figure 4.5 show a small moisture bias evident within the
model, especially when the mixing ratios are less than 5 g kg'l. This is consistent with
warmer nighttime temperatures and the resulting smaller diurnal cycle. The moist bias
appears to be larger in the NARR simulations. As the mixing ratio values increase, the
moist bias is not as pronounced as the plots show an increased scatter in both directions
suggesting that the MM5 model has a difficult time estimating the increased amounts of
moisture in the atmosphere. There also does not appear to be a difference between
daytime and nighttime moisture biases, with both showing a large (moist) bias at the
lower end (1-5 g kg") of the mixing ratio scale. Daytime and nighttime wind speeds
show an extremely large scatter which suggests that wind speeds, whether weak or
strong, are poorly simulated by MM5.

Table 4.3 summarizes statistics (mean, RMSE, bias, correlation and error standard
deviation) comparing the model with observations to help quantify the errors of the

simulation. Simulated errors can be either systematic or random. Systematic errors

80

occur due to biases within the estimations and if known can be reduced through the use of
calibration. Examples of systematic errors are misrepresentation of topography, land use
properties and radiation. Random errors are due to uncertainty, whether it be in
instrumentation and data collection or because of the grid spacing used within MM5.
These types of errors can be reduced through statistical analysis. Examples of random
errors are uncertainty within the observations themselves or the initial and boundary
conditions used by a model.

Simulated standard deviations (STD), with the exception of NARR mixing
ratios, are smaller than observed, indicating (as mentioned previously) that the model
does not estimate the full extent of the temporal variations as seen in the observations.
However, the model does do a better job at capturing the variations in wind speed and
NARR mixing ratios as the STD between simulations and observations are comparable.
Linear correlation coefficients (r) show strong correlations between simulated and
observed temperatures and mixing ratios. NARR simulations indicate slightly higher
temperature correlations ranging from 0.748 to 0.849 (Eta ranges from 0.747 to 0.861)
while Eta simulations indicate slightly higher mixing ratio correlations, 0.810 to 0.825
(NARR from 0.775 to 0.805). Wind speed correlations also show moderately strong
positive correlations, though slightly lower in both simulations than the other two
variables. Wind speed coefficients range from 0.508 during NARR daytime to 0.668
during Eta nighttime. Bias is calculated by subtracting MM5 simulated temperatures,
mixing ratios and wind speeds from observations. These values are then summed and
then divided by the total number of observations to obtain an average. Day and nighttime

temperature biases further illustrate what was seen in Figure 4.5, a slight warm bias at

81

.55 89:8 :88

88 fl mmSE 62330—5 Eat—33. hobo 2: fl Ohm Sum £56586 core—280 8:: 2: fl ._ :28 883 :88 88 2.: fl mmEM 802...

 

 

88 mod SN 8; 8; 8; Sn :3 K...” 85?.
8d 5m Now mm; mm. P mm; one 8d 8d Em 58
«moo ommd 89o 53 89o mkd 98.0 coed mid .
F to 8.9 93 $5 $5 85 mad- 2. F- mod 8m
8; m3 8; mm; mm: 5; and 5m New Em
2......” 9% mud and Em mam 8.9 8.: 3.2 :85. mm<z
New :8 and o: m3 8; one «0m 5m 8.2m
New. omd mad 8; o: m? 9% Be Be Em 58
>85 88.0 83 m 5.0 88 o 5.0 85 $3 $3 .
86 o to- So a to 86 N to Fe- 9.. F- m to 85
mm; 85 a: one 8; :1 EN 52.. v 3 05
Ed 5m o 3 Re 85 m B 3.2 5.: Name :82 mm
05 m: x: mm; 84 :3 mad 5m 3w Em
Se «in New 86. m 3 mod m E: 8.2 53 :82 80.22880
wmvvw mom? mar F F m 50m 5.. _. F #95 Ohoem comm: Swap _. 958 San .0 59:32
.2 >8 £92 .2 >8 £92 .2 >8 292 88:8

 

 

3° 8 «Bo Eek

 

068% 153 ES 33.— wfixfl: 63:22—53 8853 he 85353 :825955 .3322332 .mé ~35.

82

night, 015°C in the Eta simulation and 074°C in the NARR simulation. The Eta
daytime cool bias is larger than NARR, -1.458°C compared to -1.195°C. Mixing ratio
shows a weak bias, with the exception being the NARR daytime, which indicates a moist
bias of 0.588 g kg]. A positive nighttime bias of ~0.5 m s'1 and negative daytime bias of

0.1 ms"

is apparent in both simulations. For all variables in both simulations, the bias is
less than the error standard deviation (Error STD). This suggests that the random error
component of the model is contributing more to the simulations overall error than
systematic errors.

Vertical profile comparisons of meteorological data at upper levels (up to 3
km) at the six different monitoring stations showed similar patterns. The vertical profiles
shown in Figure 4.6 are from the San Diego, CA monitoring site, located at an elevation
of less than 50 m. Compared to the other five stations, San Diego was the most impacted
by the fires during the study period, especially during October 24-27, when the highest
amount of fire activity occurred. Potential temperature comparisons at this station
indicate that while both MM5 simulations are able to estimate the same profile shape as
the observations, a warm bias of ~l-4°K exists that tends to increase with height. Mixing
ratio comparisons for both simulations show that consistently, simulated mixing ratios are
higher than observed. This moist bias continues with height in the Eta simulations while
with the NARR simulations, the bias is largest at the lower levels and decreases with
height. Upper level wind speeds for both simulations behave in the same fashion as the
scatter plots of surface wind speeds. There is a large scatter between simulated and

observed wind speeds and no clear bias in either direction. Generally, the simulations

tend to predict vertical shape and change in speed as height increases. There is however,

83

no clear bias in either direction, again suggesting MM5 has an equally difficult time
estimating upper air wind speeds as it did in estimating surface wind speeds. Wind
directions for both simulations show good agreement with the discrepancies being less
than 90°. NARR simulations appear to have more disagreement on October 24 and 25,
but on a whole the simulations are able to predict the observed wind directions. Wind
speed and direction have implications for the BlueSky framework as they control how
fast (or slow) the smoke plumes and aerosols move in and out of specific areas and where
they go. The agreement between the observed and the simulated surface and upper level
wind directions allows the emissions from fires and smoke plumes be transported along
the observed trajectories. Large scattering in the wind speeds will either speed up or
delay the transport of smoke, resulting in errors in the timing of peak PM2.5

concentrations at specific locations downwind.

84

.5900 ”:22 53> “BE—«£5 2.032086 242 008808

30.88% 055 8.0.5 use 3:: 00:30 3% .50—ED .5900 Sm FEB 8:335 2.02950 EQEEBBEE
022206 342 2.8052 30:00.00 0:33 3750 use was: 0:8 anew Em: 8.569830 2.3202 8.86
gum—m .05 0.008538 <0 0on 00m 05 3 Q9: Ema— 82 3 8:35 _0>o_ .500: m0 comtaqfiou 9v 05w?—

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Ecozoea 05.5 At» .5 83» 35> .79. 9 28¢ 9.3.2 c: 82:
08 Eu 2.: 8 o m... . s I o m. I n - o 08 mom can
WOhO’IHLI OHJO‘I .011\ oilOIIIII-gi.” I. 0 w e‘h.“‘.0’ O
a 0000” v “I.“ I; 00 00‘ _F W
4.. o. x .5 \o 5
w... . ... a. m . x-.. N 0
40 l - 0 H m. r \...0 M
m I. . m .. . a x. mu 828.9...
can CNN 60-. on o m n 0 mr . h 0 ONM non CON 1
Ian I I OI O rulle — .0 . 0 I 01W.-
W.-... 1.. 1». Ilitq » . \kt. 9 W
”~ I. O 0 A Q I“ 6
9. ._0 O b 0 I\O “-
. . O . _ 1..
e“ n . O x e )
o I . o I do I \. O N ‘-
/ w
on 0.. I .- ./q 0 me \I 0 l‘
. .. . . . z a a I. e 8.80808 n
00” CNN 03. om o Ov m o a v O Gun Own DOM
... 09M “I .I tau... firm... .a . 5.1 (3.1.1... I.../. o _. o. I I \ Iii. O
0. on m .... o l .-.I l.I . o slut.
. 00 on. u. . 00 Pt...“ W/ N w P m
9 o . [1.11.8.0 0 \e 6
.2. . a. I . , x . x. m
r . . O H O /./O. . / O I 0 N )
I. . 000 00 O . -/./. / N0 \ ..\O~ m
. W m . m ., Iv . , f \1 IE.”
08 emu cm 0.0 o 3 m o a . w . a cum . 2.» I can
-....\ w W .. 1.....- (111.171 .. 1....
O u 0.0.. .. Q 51...: . 0 ¥
. . I I 0A 1“"! f / ~ x F W.
o O I! s? / o \\ 0 6
e H e .A.“ to: V 00 \\H e m
. d 9 II o l} I a 0 .\..e N ‘-
. W W o .. e _ . x . MI
w u I. . . .1... x . x . eweonouoOAgn

 

 

 

 

 

 

 

 

 

“COafl-asmm t“: I . I . I n I -
285:5...“ 5 III. I-..
”comgzvnno I I I I I I I I

85

Smtfie Plume and PMLsPredictions

Comparison of BlueSky predicted smoke plume orientation and shape with
observations is shown in Figure 4.7 for October 24-27 for both the Eta and NARR
simulations. In this figure, MODIS HMS images are compared with the BlueSky PM2.5
hourly totals for 1300 PST to 2300 PST. This 11 hour period is the period of MODIS
HMS detection where significant smoke plumes are recorded and analyzed (McNamara
et a1. 2004). Overall, BlueSky is able to predict both the shape and orientation of the
observed smoke plumes. On October 24th and 25th, the Eta-BlueSky simulation appears
to do a slightly better job of predicting the aerial extent of the smoke plume than the
NARR-BlueSky simulation. Both simulations on this day predict the tongue of higher
aerosol concentration along the coast of California, but compared to NARR-BlueSky, the
Eta-BlueSky simulation better predicts the shape and magnitude of PM2.5 concentration
within the plume, indicated by the darker shading within the HMS image and warmer
colors in the BlueSky image. The NARR upper level meteorological profile on this day
(Figure 4.6a) indicates weaker estimations of wind speeds above 1.5km, most likely the
cause of the inconsistencies between PM2.5 concentrations and shape for the NARR-
BlueSky simulation as weaker winds will delay smoke impacts. On October 25, both the
Eta and N ARR BlueSky simulations fail to predict the northern extent of the observed
smoke plume, but the areas of highest PM” concentration again are consistent with those
in MODIS HMS images. The NARR-BlueSky smoke estimations appear to be slightly
more spatially condensed as the Eta-BlueSky run. Both runs predict an area in northern

California of moderate (5-15 pg m'3) PM2.5 concentrations not seen in the HMS imagery.

86

 

a MODIS HMS b Eta-Blues c NARR-Blues

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 4.7 (a) Satellite images taken from the MODIS Hazard Mapping System (HMS)
for October 24 — 27, 2007. Gray shading indicates significant smoke plumes. Darker
shades of gray represent higher concentrations of smoke. Corresponding BlueSky images
are total PM2.5 concentrations for the same days for the hours of 1300 PST to 2300 PST
for (b) Eta MM5 initialization data and (c) NARR MMS initialization data. Warmer
colors indicate higher concentrations of PM2.5.

87

BlueSky images on October 26 again indicate the highest PM2.5 peaks in similar locations
to the satellite images, but the NARR-BlueSky run predicts the shape and extent of the
smoke plume more completely, capturing the elevated concentrations of PM2.5 in
southern Nevada. Disagreements between observations and both Eta and NARR MM5
estimated wind directions above 1 km (Figure 4.6c) are apparent on October 26.
However below 1 km, the NARR simulated wind directions agree with the observed wind
directions, most likely causing this difference between the Eta and N ARR-BlueSky
smoke plume orientations. Both BlueSky runs simulate for October 27 PM2.5
concentrations <5 pg 111’3 over most of the region. Although the HMS image does not
detect any aerosols in the area on this day, MODIS is known to have difficulty
differentiating small fires and subsequent smoke emissions from the background
(McNamara et a1. 2004). Winds on this day agree with observations for both Eta and
NARR simulations, and most likely the moderate southerly winds (up to 3 km) helped
remove the large aerosol concentrations from the region.

Figure 4.8 compares MODIS/GASP aerosol optical depth (AOD) images with
BlueSky images at 1500 PST (2300 Z) on each of the four days. AOD images are used to
establish a quantitative understanding of the aerosol concentrations within the air column
to surface BlueSky predicted PM2.5 concentrations. These images show that while
BlueSky predicts similar increases of aerosols (specifically PM2.5) as the AOD images, in
reality there is much more variability in aerosol concentrations than BlueSky indicates.
On October 24th, the NARR-BlueSky run simulates an increase in PM2.5 off the coast of

northern California, which is not seen in the Eta-BlueSky simulation. On October 25, the

88

 

aM

._ID

 

ODISIGASP

 

(b)Eta-BlueSky .

    

 

(c)NARR-B|ueS
,' w
I r

I

 

Figure 4.8 (a) MODIS/GASP Aerosol Optical Depth (AOD) images at 1500 PST
(23002) for October 24 - 27, 2007 with corresponding BlueSky simulated smoke plumes
for (b) Eta MM5 initialization data and (c) NARR MMS initialization data. Warmer

colors in the AOD images indicate areas of higher aerosol concentrations. Warmer colors

within the BlueSky images indicate higher levels of surface PM2.5 concentration.

 

Eta-BlueSky simulation is able to predict the larger increases in PM2.5 off the coast of
southern California. Both BlueSky simulations estimate large PM2_5 increases over
southern California on October 26, whereas the AOD images shows these increases
further to the north, perhaps due to disagreements between simulated and observed wind
speeds and directions (Figure 4.6c).

A quantitative evaluation of the BlueSky-predicted PM2.5 concentrations is not as
straightforward as the evaluation of predicted meteorological fields. PM2.5 observations
include not only the contribution of PM2.5 due to wildland fires, but also a “background
value” due to anthropogenic and natural sources. BlueSky PM2.5 predictions however,
only include the amount of PM2.5 generated by wildland fires. Thus, a direct comparison
of the modeled and observed concentrations is not possible because in theory, without the
background value included, BlueSky predictions will always be smaller than the
observations. While several methods may be used to estimate the background
concentration, including running a photochemical model, Fusina et al (2007) calculated
an average observed PM concentration from the same sites under similar synoptic
conditions, but without the influence of fire. These averages were then subtracted from
the observed PM concentrations during the period with fire activity. While this method is
reasonable, it is difficult to find a time period experiencing identical synoptic conditions,
and the results also can yield negative values which are difficult to interpret. In this
study, instead of trying to calculate the background value, the time-rate changes of the
simulated and observed surface PM2.5 concentrations were compared. The rationale
behind this approach is that a rapid rate of increase in concentration would indicate the

influence of smoke plumes on a specific location.

90

Comparisons of time-rate change of BlueSky predicted surface PM2_5
concentrations with observations for the Eta-BlueSky simulation are presented in Figure
4.9 and comparisons for the NARR-BlueSky simulation are in Figure 4.10 at four
monitoring sites within the domain. Along with each station’s time-rate change time
series, Figures 4.9 and 4.10 show the surface meteorological comparisons and the station
locations. A disadvantage of using single station time series in comparing PM2.5
concentrations is the dependence of each station to the location of the simulated smoke
plume. For instance, a simulated smoke plume may miss a station’s location by only one
model grid point and the expected increase in PM2.5 is then not produced by the
simulation, giving an impression that the BlueSky simulations to severely under-predict
surface PM2.5 concentrations. While this may be true, the under-prediction may also just
be due to the “miss” of the smoke plume due to differences in the simulated and observed
upper level meteorological patterns in wind speed and direction. For this reason, the
comparison was made at the grid point closest to the observational site as well as at the
four surrounding grid points (to the north, south, east and west).

Both Figure 4.9a and Figure 4.10a illustrate a natural variability of about a: 10 pg
m"3 that exists within the observed PM2_5 concentrations, since no smoke plumes passed
this monitoring site during the seven days presented. This variability due to factors other
than wildland fires makes it difficult to assess the ability of BlueSky to accurately predict
the magnitude of smoke impacts, although BlueSky did correctly not simulate fire
impacts during this period at the station located in Figures 4.9a and 4.10a. Using Figure

4.7 as a reference for smoke plume trajectories during October 24-27, Figures 4.9b and

91

.0630 852:. use new .636 88” ES .3505 232852 Base; 282

E8202 25.“. Eu 8:: bum 2:? 8283800 E8202 8285.: gum—m 8:038 E28288... 00228 win—88.2.00
Cnmmmv 8:068:85”. 23E 092088 .0685 E8202 was: 0228 Ed ESE 38830 E8202 8:: x85 22:00
2: E53, 828 wctoEoE .58 E 82022505022.”— 20 omega 22 0E: :80— BEQBO Ea Bum—0E8 Sm and 04.. 20mE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

200 20a
«NB w wNB w VNB F NNB v «Nae wNB _. vNB _. NNB w
.‘e . 000C. _ 9’0 41 ‘0I . O I .1015 4"} o w . . u 1 . . CV!
...: .sE . ... _ ..... .. .... :../p ..... Y : ft... . .13.}
I ......... 22.2.5971 ......... I::.Io... ............................... OQF H. I on:
W:: c ........ ..X... ....... e1. .. .......................... \W
. ........ 4...}... ..... .. .......... ...... ..... .L..,....:..Lm a... ....... . .......... .oom ( - 0N-
.. ... .r 71...... 1...“... ......p. :.:...I. . . o w - 2-
V I. 90 e
. _. v4.50... “0.. W ,. o
NJ.OeIo\vo ON w. r ...
. . . . . . It a 2
1 u . . . . OF I—
n \‘I It a 3 ON
.I 0‘ .0 II 0 arr . 9’ I
.. are... I. I... :... ...s/ . .... . .x (on .w a
i. as .... I. .. .. ....)xrk I ) on
.\ t .(. I. ow % , 3..
<0 ..osfiéoucso 3.-.
Z ... V. “a. 38,. t...» . ..o . . . . . . 0v.
15’5101. 1.50.0015... ..... b. IV!J:Z\N¢‘H’ ..... 9001...}... ........ or... ............... O”? m
a ...... e. ....... H.123...- .......... . ......... u ........... . ......... 00100” (0\
'0 0.0 e A4 0 4001 A 0 O . . 9 c w
. e 0 Q v.0 ”(0 O o tetv one... O ...-o
.o “(3 .. co... ......v. .. .....qu .... . .XI. . .. ... . . .... n W
O H ... w
. . . or (1..
. . - I . < 0 ad—
I “all. a), It)!» . . .713. snail/(skill ON m
\. l\ /\ ...». ...K C )
. r . . . :0 1mm . o¢

 

 

 

 

 

 

 

2.....on . >2 ....zEm .....2. E

«co—8.:Efi

92

(Pu €_UJ 6n) uoImuaouoo S'chl to efiueuo 9188 euII_j_ leoo1

2.5
0.06 _.

VNBF

uwB v

 

QNB _.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

¥ . . Its .1.
”coaflzgo IIIIIIIIII
2.050395“

 

 

o _ on . . a I . .. C
.3. . :..? .. ....u..v... .vghmn.»$&¢w§.im
. o
.. ........... ....2......... ..... ......... . om?
.1 . 9 .03.. t? ..\
I;Q.OI.OO‘- ......... ‘ ........ “ .................... ..
O . Q
I .......... . ..... o..b....11.1... ............ . ............ n ............ . ........... COM
.. ......s .3. . .. . c
r... ..fl. 3 . ...... It shtm‘ n:
. . ... . ,2
I ...\. ‘.fl
...sl 3. .. .
nos,
0 o
. . . . . . ON
- u - I - — c
’9- I Q... .
. a. low\ .... nu ...)! ) ON
.\ c of... » ...? . IL
.4. . ... (a
. . . .S .0. . at
u u C — u d — o
............. g 1.1.1.1 .0111.
‘ O w. (I o
.............. o. . if s .4va
O . I .‘r o
I ...O.. .. ......... (4 .0. ..... 1a Q , .9
I‘JQ‘HO 0‘0 '8‘ :00 3”. 0’0... ‘\oooo
I .......... . ............ ............... ........... . ............ . .......... r .......... com
— u - d — — c
o Qoovooo‘.‘ 04 0 $0.. 9 0. § .1} 90...! O 0%
o t . .1 x. o. o. v .-- o o o t
o 0 iv. .. t
9 ii 90 v 00 floflfi‘éo o to 00¢. ... o.
. . .m
.
. . . . . . 0..
— — — 1

. .... «.... ...: .... .. . ...om
m (s .21..» (. .ufl..m.)r/\}€(\

(00) dwei (kw!) DdSM (o) J!DN\

(00) dwaJ. (ksw) DdSM (0)1!DM

.82....8 a... 2&5

33

as. 8.2 «Na...

 

 

tNBF

_ .. . cw-
on-
om-
o T

 

 

 

 

u . .

 

 

 

 

 

 

 

 

Bol .

 

 

 

 

. . . . . . 0 ¢
<0 am >uBu5u: £30m . aficcam Au.-.

(Pu 8_u1 5n) uogmuaouoo 92Wd ;0 efiueuo emu emu |eoo1

93

 

.335 55026 :53 98 5.08:5 U808 :53 A0805 28200an Baa—=88 $.42 E8202 2:8 :5. 8::

8% 223 8:28:88 E8202 835...: :85 8:288 3822028. 080.58 88508200 AEwEV 8:253:85”.
220 Bum—2:6 Ewes—m E8202 8:: @228 :.:: 220 083890 E8202 8:: gum—m .5880: 05 55:5

8.8 w:to::oE .58 E 82.2.8050 320 00 85:0 22 2:: Eoc— cotomno Ea @2225... $2 885 o: .0. 2:30

 

 

 

 

 

 

 

 

 

 

 

 

 

 

oEo
«N3 —. wNBF VN‘OP NNB we
9...‘ u 11 .tv u d . t o c . ...-{gig
..us...£¥.. ..... ...3...&................ ........... A ....... .. m
.9 .4. I 0 o Lv a
:.. .......... 0...UO..0.0.0....¢ .......... Wm... ...... 0.0.; o ............................... ow? “-
a. ....... .33.. ...... 5.3 ....... Wu. .......... .... . . . . ................................ . \OI
1 ......... 432.9: ................ 0.... ...... ...... ...... ..94 50b... .................. .08 (\
AN filo Olliw+.; iii-3Q gig 1.10%. 00 q l w
r .- .. 0 00%.”..0‘ .1 .m-
3 0 . A
I r\...|.| (ft. ... (......‘Wtij w,
L p p 8 L F on [L5
i q . q n J u 0F l
.x ‘8? ..6 a
. O O . ..-: r.
f . ”0?..‘(Uvo ’.v ‘6 (”or moo)“. ...] o k 3.... a \ InmN m
.. as. ... ... . .... w... .3... . )
Av o .v . 0
..t. I m
b p n p h h 0* (
t r k ..va -
................................................ h . w
o I . . .
'21.... I". 03.0801...“ ..... \. ’.Vlfq OQF “-
I .......... - ........................................................................ W
a ........ Cw." ......... n0. .0.- ........ .’c ....... V. y’ ....... t“ ................... .00 OS
. . W
.700 x. ..6.009 ... ’9‘ 0
o 00’} o o 000* 000....“va 0 m
t o o )
m.
D b F g
: - W
..olr’o. o. ...)I o 8’ “(loo . ... . .8. so. ’4‘ ’1 ON d
4“ 1‘. car\ 0“ .....V (u w
p - n n 6* (

 

 

 

meagzg IIOOIIIOII
8055.553

«NB ..

 

oNB F

250
VNB —.

«NB —.

F
I

P

<0 ._0..:flu_uflhOU..—0_mv n—

 

 

 

 

 

 

 

 

 

 

 

. . . . . ov.

. . on-

. . om-

- . o..-

it? a ..1 .14: .l- 3.4. .....r‘. —:._ :.:. ......r. 2 .. . o
:83 . or
5:31]! Tom
39' . . 2.
>2 ..:.Em d... :...

(L-u €411 6n) uogmuaouoo (3°sz 10 95.,er alga emu leoo1

94

0E0

QNBP 0&2.

 

was —.

 

 

 

 

 

 

L I
(elixir. o :(u. .. f 0 .. q a J D
€00. 0 W000i00t€¢dfi00§0 3 .0 ...10 ”0" ‘w’ G
rft ... _ .: -..: ~ (0 0 .3, V;
r H MOE; .. s 0 01°F
00 :0 00000000 ‘0
0000
L p . b r . ON
q . . q . q o

o
N
(00) dune; 95w) DdSM (o) J!DM

 

O
V

 

 

 

CO
0
r

oom

 

 

or

 

y\0 v.0
h b b D‘ P! P
0:. d d . - .0 d ..V I u
,9
T ........... O ........... I'd ........ O ......................... O .................... ..l
.‘ 0. $ ‘0 8’0
r! ............ :.:-.......... ............ 0.‘ ............. .P ........ 3 ....................... J
.79 ...! 0r.......,........ 0...“! .....D £5 ‘ ...? v.00
1......0‘00030..- I... ............. [fin-u}; éjff 0:. 00.0.“?01.
02.00.. 0.00 00? z. .3 I. mm
v::::.0 ............................ . ................................ I000 ..... 3‘...
b P F b b b
q i . fi . q
01r0a 0 .00 0 .0 . f. 0.. A (0 0 00 0 0
0 0000 00 (0‘0 ’0 0.0.0 00k. 0 0 0:30 0050000033000 0 000% «0 0
.0 fl 00 0 0 0 to 0 000¢I800 O ...0.. 0 .00 0
0 h 0. 0 . 0.-. 9 0 ..f 0
_l \ , «s ‘.
. 0
a... A
n D b b b .F
u q q u u -
' s ...u r
0 0 ’ . z
I Q . I.
.. 0.; a? 0 .’0 01.400! 30000 \uvtxux . 0 ”(a
. .. 0 or... ¢ . ‘7 .0 0
0 . .0 . O . .
.. . x ‘ .. Q .. ’ .M‘
'k\ . c r L _ ’18. .l. I
x 0 0 0 <
A‘ {
P I n r h h

0..
mm

 

 

 

”coawtano IIIIIIIIII
82535

av

(00) dwal (l5w) DdSM (o) J!PN\

60:50:00 2 .v 25?:

350

$39. VNBF

«NB _.

 

<0 .o><

532+!
“mmmlll

rl

Ecum 0

 

<0 .8 $55“: 5:8 -

«N5

(ru 9w 6n) uoymueouoo ”and :o afiueuo 9193 emu .9901

P
I

F

95

4.10b should show an increase in PM2.5 on October 24"‘-26th and a slight increase on the
27‘“. At this station, both the Eta-BlueSky and NARR-BlueSky simulations capture the
increases in PM2.5 just before October 26th and again before October 27‘“, but the
predicted magnitude is less than the observed on both days. On October 24*, however,
while the Eta-BlueSky simulation is able to predict the timing and magnitude of the
observed smoke impact, the NARR-BlueSky simulation does not. Simulated surface
meteorology in both runs agree with observations, suggesting discrepancies may be due
to disagreements in upper level wind speeds and directions (Figure 4.6). At the station in
Figures 9c and 10c, smoke impacts were observed on October 26th and 27th as the smoke
plume moved to the east. Both Eta and N ARR BlueSky predictions again correctly
simulate the timing and magnitude of the increase in PM2.5 on the 26", but fail to capture
the larger increases found before and after this event. While both simulations do not
predict the increase observed on October 25, the Eta-BlueSky run shows an increase in
simulated PM2.5 around midday on the 26th. The Eta-BlueSky run is able to predict the
observed increase on October 27th however, the simulations predict this increase in PM2.5
to be ~eight hours earlier. Discrepancies within MM5 upper level (3 km) wind speeds
cause the simulated smoke plume to reach the station location earlier than the
observations indicate. The station shown in Figures 9d and 10d indicates BlueSky
correctly simulates an increase in PM2_5, in magnitude and timing, near October 24th and
26‘“. On the 24th, the simulated smoke impact is delayed compared to observations,
which is most likely a result of weaker simulated surface and upper level wind speeds.

Clearly, the timing and magnitude of BlueSky predictions were comparable to

96

observations when BlueSky did predict a smoke impact. However, BlueSky was not able
to predict all of the large increases the observations indicated. Possible explanations
include the randomness within the observations of PM2_5 concentration such as the natural
variability seen in Figures 9a and 10a. Other random factors are, sudden, unrealistic
increases in PM2.5 concentrations. These sudden increases could be due a number of
different things, an example being, a large bus passing by a monitor. The exhaust from
the bus releases large amounts of pollutants into the air, increasing the aerosol
concentration at that site at a particular time. While this does increase the amount of
PM2_5 at that particular time, it is not a true indication of the PM2.5 levels during that
day/hour. Other errors within the BlueSky simulations may be due to uncertainties
within the emissions and dispersion models incorporated into the framework. Larkin et
al (2007) explain that there is a high level of uncertainty with the fuel loadings used
because they are usually unknown in the majority of wildfires. Consumption calculations

also contain uncertainty as they are dependent upon fuel loadings.

Summ_arv and conclusions

In early 2000, the USDA Forest Service AirFire team launched the BlueSky
Smoke Modeling Framework to better predict wildland fire smoke plume dispersion and
subsequent impacts. This system integrates meteorology, emissions, dispersion and
trajectory models, and the simulations are made available to fire managers and air
resource regulators. Originally designed as a tool for making ‘go’ and ‘no-go’ decisions
for prescribed burns, BlueSky’s applications have been expanded to include smoke

impact assessments concerning wildland fire use (WFU) fires and wildfires. The purpose

97

of this study was two-fold: 1) evaluate BlueSky’s ability to simulate smoke dispersion
and surface PM2.5 impacts from the October 2007 wildfires and 2) to assess BlueSky’s
sensitivity to different meteorological (MM5) initialization data sets.

The performance of the meteorological model (MM5) in simulating both surface
and upper air variables was first evaluated. The simulations for both data sets predicted
the trend and day-to-day variation that occurred due to changes in synoptic conditions.
NARR surface temperatures were slightly warmer than Eta and both simulations created
a warm bias at night and a cool bias during the day. These warm and cool biases create a
smaller overall diurnal temperature range than what was actually observed. Mixing ratio
between the two data sets are comparable, both exhibiting a moist bias throughout the
day. Both Eta and NARR data sets tended to under predict surface wind speeds, but were
able to predict accurate surface wind directions. At the upper levels, the simulations
using the Eta initialization fields simulated better predictions of potential temperature,
mixing ratio and wind speeds than the NARR initialization fields. Both data sets,
however, show a warm, moist bias throughout the atmosphere. Important for transport
simulations, both model runs were able to produce the observed change in wind
directions with height.

After the meteorological comparisons were completed, comparisons of smoke
plume shape and trajectory were analyzed. Results indicate the predicted smoke plumes
using both MM5-Eta and MM5-NARR meteorological output to be in agreement with
HMS satellite images, in both shape and orientation. For days that the simulated smoke
plume did not match the satellite image, upper level observed and simulated wind

directions were in disagreement. Comparisons between BlueSky and MODIS/GASP

98

Aerosol Optical Depth (AOD) images show BlueSky captures the peaks and locations of
large aerosol concentrations but tends to estimate smaller aerial coverage and spatial
variation when compared to the satellite observations.

Next, single station time series comparisons were analyzed. These time series
comparisons were difficult to evaluate because smoke impacts in BlueSky are a result
only of wildfires, while those in observations can be due to anthropogenic sources as
well. To overcome this, the local time rate change of the hourly PM2.5 concentration was
used for comparison. A natural variability on the magnitude of about :10 pg rn‘3 is
apparent in the observed PM2.5 concentrations. Timing and magnitude of BlueSky
predicted surface PM2_5 concentrations were comparable to observations when BlueSky
did predict an increase in PM2.5. However BlueSky was not able to predict all of the
large increases in PM2.5 the observations indicated. The prediction with MM5-Eta
meteorology seemed to perform slightly better than NARR simulations, which may be
due to slightly more surface and upper air disagreements within wind speed and wind
direction in the MM5-NARR model run.

While BlueSky is useful for predicting smoke plume shape and transport, much
more work needs to be done at improving the surface PM2.5 concentration predictions. It
is also clear that BlueSky is also dependent upon the accuracy of the meteorological
forecasts, but because both runs still created differences between observed and simulated
smoke impacts; neither initialization data set can be considered better than the other. Eta
initialization data however, is available for real-time forecasting this data set could be

more beneficial to fire managers.

99

PART V
CONCLUSIONS
Summam

Since the 1980’s, the annual number of wildfires has decreased substantially
(NIFC 2008). While the number of wildfires has steadily decreased, the average size of
fires has increased, and the total numbers of acreage burned in the 2007 fire season, 9.3
million acres, was the second highest on record, behind only the 2006 totals (NCDC
2008). Wildfires are a beneficial component within the environment. Not only are they a
necessary component of the carbon cycle, wildfires also promote regeneration and
stimulation of the growth of many plant species, and aid in management of forest
undergrowth. However, the smoke released from these fires can have adverse effects on
surrounding communities. Decreased regional and local visibility, and potential impaired
health conditions due to the air we breathe is cause for concern. To balance the positive
and negative effects of wildfires, the Environmental Protection Agency (EPA) created air
quality regulations and established limits for acceptable levels of any particular pollutant
(EPA 2003).

In early 2000, the BlueSky Smoke Modeling Framework was developed (Berg et
al. 2003). This framework links together meteorological, emissions, dispersion and
trajectory models to provide fire managers with estimated locations and timing of
wildland fire smoke impacts (Berg et al. 2003). Initially, BlueSky was designed with
prescribed burning in mind and aimed assisting fire managers to make ‘go’ and ‘no-go’
decisions quickly and confidently. Recently, however, BlueSky has been modified and

its use extended to the suppression and management of wildfires. This research project

100

was undertaken to validate BlueSky’s ability to predict the timing and impacts of wildfire

smoke. Observed PM2.5 concentrations were collected from two separate case studies,

northern California in August of 2006 and southern California in October of 2007, to

compare with BlueSky estimated PM2_5 concentrations. Both case studies represent large

wildfire outbreaks, providing strong surface and satellite smoke footprints. Not only was

this research focused on evaluating the accuracy of BlueSky predicted trajectory, timing

and strength of smoke impacts but, it also assessed the sensitivity of BlueSky predictions

on accurate meteorological inputs. Two sets of MM5 initialization data were used:

NCEP 40 km Eta data and 32 km North American Regional Reanalysis (NARR) data.

Major findings of this research:

MM5 is able to accurately simulate the observed temporal variations in
surface temperatures, wind speeds and wind directions due to changes in
synoptic conditions.

MM5 is able to correctly estimate wind directions both at the surface as
well as the observed changes as height increased.

MM5 is able to simulate the upper level (up to 3 km) vertical profiles of
wind speeds, but is not able to simulate the observed magnitudes of upper
level or near-surface wind speeds.

BlueSky is able to accurately predict regional patterns of wildfire smoke
in both smoke plume orientation and aerial extent.

Peak aerosol maximums within MODIS aerosol optical depth imagery

align with peak BlueSky predicted PM2_5 concentrations.

101

o Magnitudes of BlueSky predicted increases in surface PM2.5
concentrations agree with the observed increases in PM2_5, especially when
the observed concentrations are greater than 10 pg m'3. However,
BlueSky is unable to predict all of the observed increases in surface PM2.5
concentrations.

0 BlueSky smoke prediction is proven to be sensitive to the meteorological
inputs, but only small differences are produced in the BlueSky predictions
driven by meteorological output produced with two different popular

large-scale fields (Eta and NARR).

Overall, BlueSky proved to be a good tool for predicting long-range location of
wildfire smoke plumes and their subsequent increases of surface PM2_5 concentration, but
had difficulty in predicting the overall magnitude of PM2.5 increases. BlueSky was found
to be sensitive to the input meteorology used, with slight differences in upper level wind
patterns creating larger differences in the predicted and observed smoke impacts. While
improvements to the emissions model and fire characteristic inputs are necessary for
accurate predictions of the magnitudes of smoke impacts, BlueSky is still a useful tool for
obtaining the overall location, size and path of a smoke plume and indicating the areas of
possible large impacts. With this information, warnings can be issued to those regions
within the smoke plume’s path and cautionary steps can be taken to ensure the safety of

the community.

102

Study Limitations

This study, like any research project, contains limitations that are summarized
here. First, fire and climate effects are difficult to model because of the fine spatial scale
necessary for useful, detailed results. Downscaling and/or upscaling model predictions
and input data create assumptions that may not be accurate at the new scale. These
assumptions are then propagated throughout the model and leads to uncertainties within
the model results. These uncertainties affect the BlueSky system because as a whole,
accurate smoke trajectory and concentration predictions are dependent upon the accuracy
and reliability of all necessary inputs to each of the modeling system components. Error
propagation due to model scaling is a limitation because, without confidence in model
results, the evaluation efforts and usefulness of the study diminishes. Another limitation
concerning fire and climate modeling is the need to balance simplicity in data input
requirements for users with the detailed data input requirements needed for the complex
framework components found within the BlueSky system. Complex computer models
that are not user friendly are not likely to be employed by fire or land managers and air
quality engineers, or if the models are indeed used, the input data may be incorrectly
entered into the system, creating erroneous results. This balance between input simplicity
and computer complexity is limiting because creating simpler models with less strict data
input requirements may not be suitable for providing the needed output information for
use in planned burn activities, air quality management or fire suppression strategies.

There were also limitations when trying to compare modeled and observed smoke
trajectories and ground concentrations. One of the main limitations was the lack of an

extensive PM2_5 monitoring network in northern California. The available monitoring

103

network in northern California was restricted by the varying terrain that characterizes the
region, and in turn created a sparse observing network. The sparse monitoring network in
northern California made it difficult to accurately compare surface simulated PM2.5
concentrations with observed concentrations. For instance, a simulated smoke plume
may miss a station’s location by only one model grid point and the expected increase in
PM2.5 is then not produced by the simulation, giving an impression that the BlueSky
simulations to severely under-predict surface PM2_5 concentrations. While this may be
true, the under-prediction may also just be due to the “miss” of the smoke plume due to
differences in the simulated and observed upper level meteorological patterns in wind
speed and direction. A denser network would allow for a more comprehensive surface
concentration analysis.

Comparison between satellite smoke plumes trajectory and concentration with
BlueSky output was also limited by the available data. The MODIS Hazard Mapping
System (HMS) archives only the 24 hour composites of the smoke evident each day
within the full air column. BlueSky however, outputs are surface hourly images, thus, a
direct comparison between the two is difficult. Composite images of BlueSky output can
be created, however this increases the computer time needed to provide the smoke plume
estimations to the users. In addition, the HMS satellite images do not give a quantitative
measure of the aerosol concentrations within the plume, so a second database, in this case
MODIS aerosol optical depth (AOD) images, needed to be employed.

Another source of limitations is the observed PM2_5 concentrations contain not
only PM2.5 produced by fire emissions, but also contain PM2_5 contributions from other

natural and anthropogenic sources. BlueSky predicted concentrations of particulate

104

matter only consider the contributions from fire and not those due to natural or
anthropogenic sources such as cars and industry. Because of this, direct comparison
between BlueSky derived concentrations and observed concentrations was difficult as
BlueSky concentrations will, theoretically, always be less than the observed.

Finally, there are limitations within the individual BlueSky model parameters
themselves. For example, there is uncertainty associated with initialization data sets and
the physical parameterizations used within the meteorological mesoscale model. These
uncertainties are due to smoothing and estimating that occurs when trying to simulate
real-world phenomena using physical equations. Also, there is uncertainty within the
emissions and dispersion models used within the framework. Wildfires burn through
multiple fuel loadings (ie. grasses, Chaparral, trees, etc) and often times these fuel
loadings are unknown. To obtain the fuel loadings for wildfires, BlueSky uses fuel
classification reference databases (ie. FCCS, NFDRS). While these databases are based
on reality, smoothing and estimation is used to create complete spatial coverage. Error
and uncertainty within fuel loadings propagate into the consumption calculations, as these

calculations are dependent upon fuel loadings.

Future Work

Although the BlueSky smoke modeling framework is operational in many parts of
the country, more work needs to be done in order to improve the accuracy of its
predictions. One major improvement would be the creation and implementation of a
nationally-consistent wildland fire (including both wildfires and prescribed fires)

reporting system. This reporting system would allow for increased reliability and

105

timeliness of wildland fire input data for all regions within the United States. Currently,
because no consistent reporting system exists, BlueSky inputs are different depending on
the region, making it difficult to validate the models accuracy.

Equally important is an expansive and geographically representative PM2.5
monitoring network. This network would need to cover the entire United States, with an
emphasis on rural areas were prescribed burning typically occurs, and mountainous
regions where wildfires often start. Currently, PM2.5 monitoring networks are regionally
diverse, such that some areas have expansive networks (southern California) while other
areas have sparse networks (northern California). These monitoring networks will not
only help to better monitor aerosol and pollutant levels in the atmosphere, but they will
allow for easier and better comparison with surface BlueSky PM2_5 estimations, and
ultimately help to improve the accuracy and reliability of these estimations.

The BlueSky smoke modeling system as a whole needs further validation and
evaluation of its individual components. This study found that mesoscale meteorology
forecasting using the PSU/NCAR Mesoscale Model (MM5) provided accurate three-
dimensional meteorological fields as well as their temporal variations for use in BlueSky
dispersion modeling. However, a study by BSRW (2006), found that the model was
causing unrealistic collapses of the mixing height. These findings imply that further
improvement to the mesoscale models boundary layer scheme is necessary.
Experimenting with different model parameterizations, whether in the meteorological
model or the emissions model, makes clear the importance and sensitivity the accuracy of
that particular model has on the overall smoke concentration estimations. J ain et al

(2007) also suggests ensemble techniques should be employed to see whether the models

106

themselves increase the relative success or failure of the BlueSky framework (Jain et al.
2007). To date, no BlueSky studies on the remaining framework components have been
evaluated (fire emissions, dispersion and trajectory models) to compare their estimations
with reality. Also, an evaluation switching BlueSky’s emission model from EPM, a
model originally designed with prescribed burn in mind, to one that includes smoldering
and the consumption of live fuels, such as the fire production emission simulator (FEPS)
has also been suggested (Berg et al. 2003). Errors that occur in any of BlueSky’s
components are propagated throughout the rest of the framework and appear in the smoke
estimations. Only accurate results in each component can produce reliable and accurate

results from the framework as a whole.

107

REFERENCES

Achtemeier, G. L. 2005. Planned Bum-Piedmont. A local operational numerical
meteorological model for tracking smoke on the ground at night: model
development and sensitivity tests. International Journal of Wildland Fire 14
(1)285-98.

Adkins, J. W., S. M. O'Neill, M. Rorig, S. A. Ferguson, C. M. Berg, and J. L. Hoadley.
2003. Assessing the Accuracy of the BlueSky Smoke Modeling Framework
During Wildfire Events. Paper read at 5th Symposium on Fire and Forest
Meteorology and the 2nd International Wildland Fire Ecology and Fire
Management Congress, at Orlando, FL.

Agency, U. S. E. P. 2006. 2006-2011 Strategic Plan: Charting Our Course, 184: US
Environmental Protection Agency.

Andrews, P. L., and L. P. Queen. 2001. Fire modeling and information system
technology. International Journal of Wildland Fire 10 (3-4):343-352.

Archibold, R. C. 2007. California Fires Destroy Scores of Homes. The New York Times,
Oct 22, 2007.

Association, A. L. 2007. American Lung Association State of the Air Report: 2007:
American Lung Association.

Bacon, D. P., N. N. Ahmad, Z. Boybeyi, T. J. Dunn, M. S. Hall, P. C. S. Lee, R. A.
Sarrna, M. D. Turner, K. T. Waight, S. H. Young, and J. W. Zack. 2000. A
dynamically adapting weather and dispersion model: The Operational Multiscale
Environment Model with Grid Adaptivity (OMEGA). Monthly Weather Review
128 (7):2044-2076.

Berg, C. M., S. M. O'Neill, S. A. Ferguson, and J. W. Adkins. 2003. Application of the
BlueSky Smoke Modeling Framework to the Rex Creek Wildfire. Paper read at
5th Symposium on Fire and Forest Meteorology and the 2nd International
Wildland Fire Ecology and Fire Management Congress, at Orlando, FL.

Binkowski, F. S., and S. J. Roselle. 2003. Models-3 community multiscale air quality
(CMAQ) model aerosol component - 1. Model description. Journal of
Geophysical Research-Atmospheres 108 (D6):-.

Breyfogle, S., and S. A. Ferguson. 1996. User Assessment of Smoke-Dispersion Models

for Wildland Biomass Burning, 30: US Department of Agriculture: General
Technical Report PNW-GTR-379.

108

BSRW. 2006. BlueSkyRAIN S West (BSRW) Demonstration Project: Final Report 2006,
38: US Department of Agriculture.

Cohen, J. D., and J. E. Deeming. 1985. The National Fire-Danger Rating System: Basic
Equations, 16: PSW GTR PSW 82.

Craig, K. J ., N. J. M. Wheeler, S. B. Reid, E. K. Gilliland, and D. C. Sullivan. 2007.
Development and Operation of National CMAG-based PM2.5 Forecast System
for Fire Management. Paper read at 6th Annual CMAS Conference, October 1-3,
at Chapel Hill, NC.

Draxler, R. R., and G. D. Hess. 1997. Description of the HYSPLIT-4 modeling system.

Durrenberger, R. W. 1959. The geography of California in essays and readings: Brewster
Pub. Co.

 

. 1968. Elements of Califomia geography: National Press Books.

EPA. 2003. Air Quality Criteria for Particulate Matter: Fourth External Review:
Environmental Protection Agency.

Evers, L. 2006. BlueSky Proves Its Value in Predicting Smoke. USDA Fire Management
Today 66 (3):9-11.

Evers, L., S. Ferguson, S. O'Neill, and J. Hoadley. 2006. Applying BlueSky Smoke
Modeling Framework on Wildland Fires. USDA Fire Management Today 66
(3):5-8.

Ferguson, S. A., J. Peterson, and A. Acheson. 2001. Automated, Real-Time Predictions
of Cumulative Smoke Impacts From Prescribed Forest and Agricultural Fires.
Paper read at 4th Symposium on Fire and Forest Meteorology, 13-15 November,
at Reno, NV.

Flaccus, G. 2007. 1,500 homes lost; $1B loss in San Diego area. The Associated Press;
Seattle Times, Oct 24, 2007.

Fusina, L., S. Zhong, J. Koracin, T. Brown, A. Esperanza, L. Tarney, and H. Preisler.
2007. Validation of BlueSky Smoke Prediction System Using Surface and
Satellite Observations During Major Wildland Fire Events in Northern California.
Paper read at The fire environment - innovations, management, and policy;
conference proceedings, at Destin, FL.

Grell, G. A., J. Dudhia, and D. R. Stauffer. 1994. A Description of the Fifth-Generation

Penn State/NCAR Mesoscale Model (MM5). NCAR Technical Note TN-398,
June,l994

109

Hardy, C., J. P. Menakis, D. G. Long, and J. L. Garner. 1998. FMI/WESTAR Emissions
Inventory and Spatial Data for the Western United States. Missoula, Montana:

USDA Forest Service, Rocky Mountain Research Station, Fire Sciences
Laboratory.

Hummel, J ., and J. Rafsnider. 1995. TSARS plus Smoke Production and Dispersion
model User's Guide. Ft Collins, CO: Unpublished report on file with Environment
Science and Technology Center.

Jain, R., J. Vaughan, H. Kyle, C. Rarnosa, C. Clalbom, S. Maarten, M. Schaaf, and B.
Lamb. 2007. Development of the ClearSky smoke dispersion forecast system for

agricultural field burning in the Pacific Northwest. Atmospheric Environment 41
(32):6745-6761.

Kim, D., and W. R. Stockwell. 2007. An online coupled meteorological and air quality
modeling study of the effect of complex terrain on the regional transport and
transformation of air pollutants over the Western United States. Atmospheric
Environment 41 (l 1):2319—2334.

Larkin, N. K., S. M. O'Neill, R. Soloman, C. Krull, S. Raffuse, M. Rorig, J. Peterson, and
S. A. Ferguson. 2007. The BlueSky Smoke Modeling Framework. (Submitted).

Lavdas, L. 1996. Program VSMOKE -— User's Manual. Asheville, NC: USDA Forest
Service General Technical Report SRS-6.

Liu, Y. Q., G. L. Achtemeier, and S. Goodrick. 2006. Modeling Air Quality Effects of
Prescribed Burn in Georgia with CMAQ-Daysmoke. Paper read at Proceedings of
the Workshop on Agricultural Air Quality, ESA.

Lopes, A. G., M. G. Cruz, and D. X. Viegas. 2002. FireStation - An Integrated Software
System for the Numerical Simulation of Fire Spread on Complex Topography.
Environmental Modeling & Software 17:269-285.

Malm, W. C., D. E. Day, and S. M. Kreidenweis. 2000. Light scattering characteristics of
aerosols as a function of relative humidity: Part I - A comparison of measured
scattering and aerosol concentrations using the theoretical models. Journal of the
Air & Waste Management Association 50 (5)2686-700.

McGrattan, K. B. 2003. Smoke plume trajectory modeling. Spill Science & Technology
Bulletin 8 (4):367-372.

McKenzie, D., S. M. O'Neill, N. K. Larkin, and R. A. Norheim. 2006. Integrating models

to predict regional haze from wildland fire. Ecological Modelling 199 (3):278-
288.

110

McKenzie, D., C. L. Raymond, L. K. B. Kellogg, R. A. Norheim, A. G. Andreu, A. C.
Bayard, K. E. Kopper, and E. Elman. 2007. Mapping fuels at multiple scales:
landscape application of the Fuel Characteristic Classification System. Canadian

Journal of Forest Research-Revue Canadienne De Recherche F orestiere 37
(12):2421-2437.

McNamara, D., G. Stephens, M. Ruminski, and T. Kasheta. 2004. The Hazard Mapping
System (HMS) — NOAA's Multi-Sensor Fire and Smoke Detection Program Using

Environmental Satellites. In 13th Conference on Satellite Meteorology and
Oceanograhy. Norfolk, VA.

Minnich, R. A. 1987. Fire Behavior In Southern California Chaparral Before Fire
Control: The Mount Wilson Burns at the Turn of the Century. Annals of the
Association of American Geographers 77 (4)1599-618.

Miranda, A. I. 2004. An integrated numerical system to estimate air quality effects of
forest fires. International Journal of Wildland Fire 13 (2):217-226.

Miranda, A. I., C. Borrego, and D. X. Viegas. 1994. Forest Fire Effects On The Air
Quality. Air Pollution II, computer simulation: 191-199.

NCDC. 2008. Climate of 2007: Wildfire Season Summary at
http://www.ncdc.noaa.gov/oa/climate/research/2007/fire07.html

NIFC. 2008. Fire Information: Available at: http://www.nifc.gov/.

 

O'Neill, S. M., S. A. Ferguson, and J. Peterson. 2003. The BlueSky Smoke Modeling
Framework (wwalueSkvRAINSorg). Paper read at 5th Symposium on Fire
and Forest Meteorology and the 2nd International Wildland Fire Ecology and Fire
Management Congress, at Orlando, FL.

O'Neill, S. M., J. L. Hoadley, S. A. Ferguson, R. Soloman, J. Peterson, N. K. Larkin, R.
Wilson, D. Matheny, and R. S. Peterson. 2005. Applications of the
BlueSkyRAINS Smoke Prediction System. In EM Magazine, 20-23.

Odion, D. C., E. J. Frost, J. R. Strittholt, H. Jiang, D. A. Dellasala, and M. A. Moritz.
2004. Patterns of fire severity and forest conditions in the western Klamath
Mountains, California. Conservation Biology 18 (4):927-936.

Potter, B. E., N. Larkin, and N. Nikolov. 2006. Smoke, Fire and Weather: What Forest
Service Research is Doing To Help. USDA Fire Management Today 66 (3): 12—16.

Pouliot, G., T. Pierce, W. Benjey, S. M. O'Neill, and S. A. Ferguson. 2005. WIldfire

Emission Modeling: Integrating BlueSky and SMOKE. Paper read at 14th
International Emission Inventory Conference, at Las Vegas, NV.

111

Rapp, V. 2006. A Clear Picture of Smoke: BlueSky Smoke Forecasting. In Science
Update 14, 12. Portland, OR: US. Department of Agriculture, Forest Service,
Pacific Northwest Research Station.

Reinhardt, E. D., R. E. Keane, and J. K. Brown. 2001. Modeling fire effects.
International Journal of Wildland Fire 10 (3-4):373-380.

Reza, H. G., J. Leovy, and A. Pham. 2007. Scale of the fires' disruption on display at San
Diego Stadium. In Los Anglese Times. San Diego.

Sandberg, D. V., and J. Peterson. 1984. A source-strength model for prescribed fires in
coniferous logging slash. Paper read at Proceedings: 21st Annual Meeting of the
Air Pollution Control Association, Pacific Northwest International Section, at
Pittsburgh, PA: Air Pollution Control Association.

Scire, J. S., F. R. Robe, M. E. Femau, and R. J. Yamartino. 2000. A User's Guide for the
CALMET Meteorological Model (version 5). Concord, MA: Earth Tech Inc.

Scire, J. S., D. G. Strimaitis, and R. J. Yamartino. 2000. A User's Guide for the
CALPUFF dispersion model, version 5. Earth Tech. Inc. Concord, MA.

Sestak, M., S. O'Neill, S. Ferguson, J. Ching, and D. G. Fox. 2002. Integration of
Wildfire Emissions into Models - 3/CMAQ with the Prototypes: Commnity
Smoke Emissions Modeling System (CSEM) and BlueSKy. Paper read at lst
CMAS Model-3 User's Conference, at Research Triangle Park, NC.

Sestak, M. L., and A. R. Riebau. 1988. SASEM Simple Approach Smoke Estimation
Model: USDI Bureau of Land Management Technical Note 382.

Valente, J ., A. I. Miranda, A. G. Lopes, C. Borrego, D. X. Viegas, and M. Lopes. 2007.
Local-scale modelling system to simulate smoke dispersion. International Journal
of Wildland Fire 16 (2): 196-203.

WRAP, and WGA. 2005. 2002 Fire Emission Inventory for the WRAP Region - Phase 11:
Western Regional Air Partnership and Western Governors Association.

WRAP, W. R. A. P. a. W. G. A. 2005. 2002 Fire Emission Inventory for the WRAP
Region - Phase H.

WRCC. 2008. Southern California Climate Summaries (Western Regional Climate
Center) available at: http://www.wrcc.dri.edu.

 

Zhong, S. Y., H. J. In, and C. Clements. 2007. Impact of turbulence, land surface, and
radiation parameterizations on simulated boundary layer properties in a coastal
environment. Journal of Geophysical Research-Atmospheres 112 (D13):-.

112

 

 

 

 

 
 

I

3 1293 0297

I
I
ll
1’.
I
I!
I

l I

 

 

 

. a
.
. .