Supplementing Earth Observation with Twitter data to improve disaster assessments: A case study of 2020 Bobcat fire in Southern California

Abstract

Space-based Earth Observation allows us to simultaneously detect changes on Earth's surfaces over a large area. As a result, it's often used to assist with disaster assessments to understand associated property damages in the affected areas. Geographic Information Systems and Remote Sensing, are very popular for their applications in handling disasters and are being utilized as a key tool to support decision making throughout the disaster management process. And top of it, the use of social media, especially Twitter, has become a popular communication platform that is identified in providing vital information in emergency situations. Twitter users can use the services to work synergistically regardless of physical distance. To demonstrate the benefits of supplementing Earth observations with Twitter data in disaster assessments, we use a recent fire in Southern California, the Bobcat fire, that started on September 6, 2020 and burned for over 3 months until it was finally contained on December 18, 2020 as a case study. The plume from the fire spanned more than 1,000 miles with smoke travelling across the entire North American continent. 116,000 acres of land got affected along with unprecedented wildlife loss. Also, this fire has turned the high sierra granite gorge into bare and ashen sloped. In this study, we integrated Earth observations with data from Twitter, to assess a more comprehensive view of the overall damages including physical and emotional state as an aftermath of a disaster. Remote sensing data lets us understand pre-post-fire conditions of the land as well as temperature variation and soil condition of the ground. Geographical locations are analyzed from tweets which are compared with levels of various pollutants measured from ground instrumentations and the amount of smoke coverage from satellite imagery. With the additions of Twitter data, using machine learning and natural language processing, we are able to derive a more holistic impact of the Bobcat fire on California citizens. Thus, augmenting remote sensing data with socially sensed Twitter data will strengthen the capabilities of experts and staff working to analyze and manage disaster risk by providing them with both spatial and socioeconomic information. Moreover, we can also determine how various factors contribute to the super spreading of messages. A better understanding of social media utilization would allow us to determine a better risk reduction tool, whether it would be for the purposes of early warning of disaster events or reducing mental stresses after a disastrous event.

Full text

72𝑛𝑑 International Astronautical Congress (IAC) - Dubai, 25-29 October 2021.
Copyright 2021 by the authors. Published by the IAF, with permission and released to the IAF to publish in all forms.
IAC-21,B1,5,10,x65420
Supplementing Earth Observation with Twitter data to improve
disaster assessments: A case study of 2020 Bobcat fire in Southern
California
Swarnajyoti Mukherjee𝑎,∗,#, Krittanon Siroratttanakul 𝑏 ,∗,#, Daniela Vargas-Sanabria𝑐, ∗,#,
Samridh Patial𝑑,∗,#, Abinash Silwal𝑒,∗,#, Kristine Jane Atienza𝑓 ,∗,#
𝑎Space System Engineer & Business Professional, GP Advanced Projects, Srl, Brescia, Italy, swarnajyoti.mukherjee@gmail.com
𝑏Division of Geological and Planetary Sciences, California Institute of Technology, Pasadena, CA, USA,
krittanon.sirorattankul@spacegeneration.org
𝑐Laboratorio de Investigación e Innovación Tecnológica, Universidad Estatal a Distancia, San José, Costa Rica,
danielavargas1989@gmail.com
𝑑Department of Aerospace Engineering, Indian Institute of Space Science and Technology, Kerala, India,
samridh.sc21m030@pg.iist.ac.in
𝑒Department of Geomatics Engineering, Kathmandu University, Nepal, afactor.abinash@gmail.com
𝑓World Vision International, Philippines, kristine.atienza@spacegeneration.org
#Space Technology for Earth Applications (STEA) Project Group, Space Generation Advisory Council (SGAC), Vienna, Austria
∗Corresponding Author
Abstract
Space-based Earth Observation allows us to simultaneously detect changes on Earth’s surfaces over a large area. As a
result, it’s often used to assist with disaster assessments to understand associated property damages in the affected
areas. Geographic Information Systems and Remote Sensing, are very popular for their applications in handling
disasters and are being utilized as a key tool to support decision making throughout the disaster management process.
And top of it, the use of social media, especially Twitter, has become a popular communication platform which is
identified in providing vital information in emergency situations. Twitter users can use the services to work
synergistically regardless of physical distance. To demonstrate the benefits of supplementing Earth observations with
Twitter data in disaster assessments, we use a recent fire in Southern California, Bobcat fire, that started on
September 6, 2020 and burned for over 3 months until it was finally contained on December 18, 2020 as a case study.
The plume from the fire spanned more than 1,000 miles with smoke travelling across the entire North American
continent. 116,000 acres of land got affected along with unprecedented wildlife loss. Also this fire has turned the
high sierra granite gorge into bare and ashen sloped. In this study, we integrated Earth observations with data from
Twitter, to assess a more comprehensive view of the overall damages including physical and emotional state as an
aftermath of a disaster. Remote sensing data lets us to understand pre post fire conditions of the land as well as
temperature variation and soil condition of ground. Geographical locations are analyzed from tweets which are
compared with levels of various pollutants measured from ground instrumentations and the amount of smoke
coverage from satellite imagery. With the additions of Twitter data, using machine learning and natural language
processing, we are able to derive a more holistic impact of the Bobcat fire on California citizens. Thus, augmenting
remote sensing data with socially sensed Twitter data will strengthen capabilities of experts and staff working to
analyze and manage disaster risk by providing them both spatial and socio-economic information. Moreover, we can
also determine how various factors contribute to the superspreading of messages. A better understanding of social
media utilization would allow us to determine a better risk reduction tool, whether it would be for the purposes of
early warning of disaster events or reducing mental stresses after a disastrous event.
Keywords: Bobcat Fire, Earth Observation, Twitter, Disaster Assessment, California citizen
IAC-21,B1,5,10,x65420 1

72𝑛𝑑 International Astronautical Congress (IAC) - Dubai, 25-29 October 2021.
Copyright 2021 by the authors. Published by the IAF, with permission and released to the IAF to publish in all forms.
Acronyms
DOI Department of the Interior
EO Earth Observation
EPA Environmental Protection Agency
FIRMS Fire In-formation for Resource Manage-
ment System
FS Forest Service
GDP Gross domestic product
GEE Google Earth Engine
GIS Geographic Information System
LST Land Surface Temperature
MODIS Moderate Resolution Imaging Spectro-
radiometer
NASA National Aeronautics and Space Admin-
istration
NBR Normalized Burn Ratio index
NFS National Forest System
NIFC National Interagency Fire Center
NIR Near Infrared
NLTK Natural Language Toolkit
OLI Operational Land Imager
SWIR Shortwave Infrared
TIRS Thermal Infrared Sensor
UN-SPIDER United Nations Platform for Space-based
Information for Disaster Management
and Emergency Response
USGS United States Geological Survey
1 Introduction
Wildfires are unplanned and unwanted fires, including
lightning-caused fires, unauthorized human-caused fires,
and escaped prescribed fire projects. States are re-
sponsible for responding to wildfires that begin on non-
federal (state, local, and private) lands, except for lands
protected by federal agencies under cooperative agree-
ments. The federal government is responsible for re-
sponding to wildfires that begin on federal lands. The
Forest Service (FS)—within the U.S. Department of Agri-
culture—carries out wildfire management and response
across the 193 million acres of the National Forest System
(NFS). The Department of the Interior (DOI) manages
wildfire response for more than 400 million acres of na-
tional parks, wildlife refuges and preserves, other public
lands, and Indian reservations. Wildfire statistics help
to illustrate past U.S. wildfire activity. Nationwide data
compiled by the National Interagency Fire Center (NIFC)
indicate that the number of annual wildfires is variable
but has decreased slightly over the last 30 years and the
number of acres affected annually, while also variable,
generally has increased. Since 2000, an annual average
of 70,600 wildfires has burned an annual average of 7.0
million acres. This figure is more than double the average
annual acreage burned in the 1990s (3.3 million acres),
although a greater number of fires occurred annually in
the 1990s (78,600 average).[1] From 2011 to 2020, there
were an average of 62,805 wildfires annually and an av-
erage of 7.5 million acres impacted annually. In 2020,
58,950 wildfires burned 10.1 million acres, the second-
most acreage impacted in a year since 1960; nearly 40%
of these acres were in California. Nearly half of the acres
impacted were on NFS lands. These official figures from
NIFC reflect downward revisions from earlier reported
data for 2020. As of September 8, 2021, nearly 44,000
wildfires have impacted over 5.1 million acres. The na-
tionwide preparedness level has been at the maximum
level (5) since July 14, 2021, suggesting a sustained and
significant commitment of shared resources [1].
Usually, large wildfires occur often in dry ecosystems of
the western United States [2], and around 80% of the
world’s fires are caused by humans [3]. The study of
wildfires dynamics can be supported by Earth Observa-
tion (EO) technology. Chuvieco and Congalton (1989)
[4] mentioned the advantage of using Geographic Infor-
mation System (GIS) and satellite images for detecting,
monitoring, and assessing the burn scars caused by fires.
During a fire event, EO is useful for detecting and locating
smoke plumes distribution. For example, multi-spectral
satellite data have been used as a tool to aid in the detec-
tion of changes to ecosystems for burning biomass [5]. In
addition, spectral indexes have been developed to evalu-
ate the frequency, intensity, and localization of a wildfire.
One of the frequent indexes used for change detection is
based on pre-fire and post-fire images from multi-spectral
bands, which can determine biomass loss, smoke produc-
tion, and carbon release [5]. The thermal infrared band in
particular is also very efficient and effective in evaluating
environmental changes that are related to wildfires [6].
The brightness in thermal band can be converted to Land
Surface Temperature (LST) which has shown to increase
significantly during wildfires [7,8]. We used the Normal-
IAC-21,B1,5,10,x65420 2

72𝑛𝑑 International Astronautical Congress (IAC) - Dubai, 25-29 October 2021.
Copyright 2021 by the authors. Published by the IAF, with permission and released to the IAF to publish in all forms.
ized Burn Ratio index (NBR) to highlight burned areas in
large fire zones, as in the case of the Bobcat fire, which
burned 46,861 ha approximately. Its delta-NBR allows it
to estimate the severity of the fire, which means the impact
or degree of environmental changes caused by a wildfire
[9].
Fine particulate matter (PM2.5) is an air pollutant that
when occurs in excessive amounts could pose health con-
cerns such as increased risk of respiratory and cardiovas-
cular diseases [10]. The term refers specifically to parti-
cles that are 2.5 microns in diameter or smaller. It is often
measured in 𝜇𝑔/𝑚3. The United States Environmental
Protection Agency (EPA) established standards for PM2.5
at 35 𝜇𝑔/𝑚3daily average or 12 𝜇𝑔/𝑚3annual average 1.
Wildfires in California could elevate the concentration of
PM2.5 above the standards for a prolonged period of time
[11]. As the wildfires become more frequent and intense
due to climate change [12], better monitoring of PM2.5 is
prudent to avoid health hazards.
The use of social media, especially Twitter, has become a
popular communication platform, and is identified in pro-
viding vital information in emergency situations has be-
come popular. Twitter users can use the services to work
synergistically regardless of physical distance. This paper
is concerned with the use of Twitter data to show the ef-
fectiveness of society’s reaction, awareness, their positive
and negative aspects, relief measures, express gratitude,
complaints and others. By understanding how various
factors contribute to the superspreading of messages, one
can better optimize Twitter as an essential communica-
tions and risk reduction tool with machine learning algo-
rithms. This study introduces which further define the
technological and scientific knowledge base necessary for
developing future competency base curriculum and con-
tent for Twitter assisted disaster management education
and training at the community level.
2 California Wildfires
2.1 Burned Area Statistics
California had more wildfires (See Figure 1) than any
other state in 2019, and by California standards, 2019
was a mild year. State and local resources fought 7,860
1https://www.epa.gov/pm-pollution/
2012-national- ambient-air- quality-standards- naaqs\
-particulate- matter-pm#additional- resources
Fig. 1: California Area Burned from 2016 to 2020 [13]
wildfires that burned more than 259,000 acres [13]. In
August 2020 alone, nearly 585 wildfiresburned nearly one
million acres in a week and 700 homes [13]. California
has perfect conditions for wildfires. Its risk is directly tied
to:
•Dry weather, thanks to the drought.
•Lack of forest management.
•An influx of development in the Wildland Urban In-
terface.
•High winds.
Native trees and plants are dying off, creating more kin-
dling, and being replaced by invasive grasses that burn
easily. The Forest Service tries its best to trim brush and
implement prescribed burns to manage the forests and
curb the risk of fires, but most of its yearly resources go
toward combating massive fires instead.
2.2 Economic Cost
Wildfire destruction is costly. In 2019, more than $12
billion in claims were filed from the 2018 Camp Fire,
Woolsey Fire, and Hills Fire. And that’s just insured
losses – many homeowners most at risk for wildfires may
not have coverage. AccuWeather [14]estimates between
damage to homes and businesses, belongings and cars,
job and wage losses, farm and crop losses, infrastructure
damage, school closures, highway closures, and power
outage costs puts the 2018 wildfires at $400 billion in
IAC-21,B1,5,10,x65420 3

72𝑛𝑑 International Astronautical Congress (IAC) - Dubai, 25-29 October 2021.
Copyright 2021 by the authors. Published by the IAF, with permission and released to the IAF to publish in all forms.
Fig. 2: Economic Cost structure from 2016 to 2019 [14]
Fig. 3: Where major fires have burned in 2020 in relation to
previous ones (2000-2019) Sources: NASA’s Fire Information
for Resource Management System (previous fires), National
Interagency Fire Center (2020 perimeters)
losses (See Figure 2). By 7th December 2020, California’s
economic cost due to wildfire is $120bn - $150bn only in
2020, roughly 0.5% of the United States of America’s
annual Gross domestic product (GDP).
3 The 2020 Bobcat Fire
Bobcat fire, that started on September 6, 2020 and burned
for over 3 months until it was finally contained on De-
cember 18, 2020 as a case study. The plume from the
fire spanned more than 1,000 miles with smoke traveling
across the entire North American continent. This fire has
been one of the most devastating fires in the city’s history
Fig. 4: Aftermath of fires near the foothill of the San Gabriel
Mountains. Picture was taken on September 2021.
Fig. 5: Fire scar on the San Gabriel Mountains taken near Santa
Fe Dam on September 2021. The region above the dotted line
was burned during the 2020 Bobcat Fire.
with 26 casualties and over 180 structures damaged due to
the fire. Over 116,000 acres of land was affected 2due to
the fire with unprecedented loss of wildlife which is still
to be estimated. The fire has turned the high sierra gran-
ite gorge into bare and ashen sloped. As per geologists
examination and report 2, the land is primed for mudslide
condition due to the oncoming winter storm which will
setback all the conservation effort done in the past years
to the area.
4 Methods
4.1 Fire Mapping
With Google Earth Engine (GEE), we delimit the area
affected by fire using an image taken on 2020-11-24
2https://www.cityofmonrovia.org/your-government/
bobcat-fire
IAC-21,B1,5,10,x65420 4

72𝑛𝑑 International Astronautical Congress (IAC) - Dubai, 25-29 October 2021.
Copyright 2021 by the authors. Published by the IAF, with permission and released to the IAF to publish in all forms.
by Landsat 8 OLI/TIRS and atmospherically corrected
Surface Reflectance from the collection of the GEE.
NASA’s Fire In-formation for Resource Management Sys-
tem (FIRMS)3is used for validation. We then analyze the
severity of the fire within the perimeter of the burned
area. This step can be divided into three parts: a general
severity map that includes an analysis of the entire pe-
riod of the wildfire, a weekly severity map, and monthly
severity monitoring since the wildfire started in Septem-
ber 2020 until December 2020. We calculate the Nor-
malized Burn Ratio index (NBR), an index highlighting
burned areas within the fire perimeters. The formula uses
the Near Infrared (NIR) and Shortwave Infrared (SWIR)
wavelengths, given as follows:
𝑁 𝐵𝑅 =
𝑁 𝐼 𝑅 −𝑆𝑊 𝐼 𝑅
𝑁 𝐼 𝑅 +𝑆𝑊 𝐼 𝑅 (1)
The values of the NBR indicate the healthy vegetation
and the low values of the areas that are burned. Then,
we follow the UN-SPIDER recommended practices for
mapping the severity of the burned areas 4. Data down-
loading and processing are done using the GEE platform.
To generate the severity map, we use pre-fire and post-fire
Landsat 8 images. The pre-fire images were from 2020-
01-01 to 2020-08-31, while the post-fire images were from
2020-10-01 to 2020-11-30. We calculate the NBR and the
NBR with these inputs, defined as the differences between
NBRs from pre-fire to post-fire. We use categories de-
fined by the United States Geological Survey (USGS) to
classify delta-NBR into one of the following groups: en-
hanced regrowth high, enhanced regrowth low, unburned,
low severity, moderate-low severity, moderate-high sever-
ity, and high severity.
4.2 Land Surface Temperature (LST)
We have used MODIS satellite time series datasets in this
study to estimate the LST. The MOD11A2, MODIS LST
6th product provides an average 8-day LST in a 1200
km x 1200 km grid. with 1km spatial resolution. Each
pixel value in MOD11A2 is a simple average of all the
corresponding MOD11A1 LST pixels collected within
that 8-day period [15]. MOD11A2 was selected from
3https://firms.modaps.eosdis.nasa.gov/
4https://www.un-spider.org/
advisory-support/recommended- practices/
recommended-practice- burn-severity/in- detail
Fig. 6: Land Surface Temperature (LST) comparison between
2018 and 2020
earth engine’s MODIS collection and ‘𝐿𝑆𝑇𝐷𝑎𝑦 1𝑘𝑚0data
band was used for retrieval of LST. The graph of LST time
series was created from the mean LST data of the study
area from September to December for the years 2018, 2019
and 2020. The reason to study the trend of 3 years LST is to
compare the variation of LST before and during the bobcat
fire. The reason to choose an 8-day compositing period
was because the exact ground track repeat period of the
Terra and Aqua platforms is twice that period [15]. LST
was extracted from the ‘𝐿𝑆𝑇𝐷𝑎𝑦1𝑘𝑚0band and converted
to degree celsius units and represented in a line graph
using google earth engine. It can be observed that there
is about 5-10◦𝐶increase in the LST of 2020 between
September and November than the past trends (year 2018
and 2019) of LST when there was no bobcat fire. This
aberration can be due to the fire in bobcat in the year 2020.
4.3 Air Pollutants
Past records of daily PM2.5 data are available through the
United States Environmental Protection Agency (EPA)
website 5. Data from recent days are taken from the
AirNow portal 6which is updated regularly. As AirNow
data are validated through quality assurance designated by
EPA, it will be released to replace data from the AirNow
portal. Daily data are calculated from the averaging over
a period of one day from midnight to midnight in local
time. In the state of California, there are 165 stations that
5https://www.epa.gov/outdoor-air- quality-data/
download-daily- data
6http://www.airnow.gov/
IAC-21,B1,5,10,x65420 5

72𝑛𝑑 International Astronautical Congress (IAC) - Dubai, 25-29 October 2021.
Copyright 2021 by the authors. Published by the IAF, with permission and released to the IAF to publish in all forms.
recorded PM2.5 data in 2020. Among them include 30
stations that are located within 100 km from the vicinity
of the Bobcat fire. The distance is measured from an as-
sumed reference point within the fire perimeter at latitude
34.2992◦N and longitude 117.9617◦W. PM2.5 data can
be compared with the burned area and land surface tem-
perature measured from Landsat thermal images to assess
the effect of the Bobcat fire on the local climate systems.
4.4 Extract & Process Twitter Data
The Bobcat fire started on 9th September, 2020 which
effectively engulfed 114,926 acres in which more than
6000 structures were threatened with an unprecedented
loss to wildlife 2. One of the primary purpose of studying
the tweet data is to understand the use of twitter by the
general public and to introspect their perception around
the incident. Primary literature study was done around
news covering Bobcat Fire incidents to get an idea which
tweet hashtags cover the most information on Twitter. For
that reason, we have used snscrape7and Tweepy8python
libraries to extract tweets from August, 2020 to Decem-
ber, 2020. Using these libraries, we have specified to
extract only those tweets who has "Bobcatfire" in it as
well as tweet data is compiled by considering hashtags
like, "#Bobcatfire, #Calfire, #CaliforniaFire", to build the
Global database for the Bobcat Fire event. We have an-
alyzed the sentimental analysis of those tweets and by
doing that we had to needed to retrieve the date, time,
place, number of replies & retweets for each and every
tweets.
Later on, we have used Natural Language Toolkit (NLTK),
a Python library, to do Natural Language Processing.
Firstly, we eliminated the links, mentions to get the core
of the tweets and eventually did tokenization to make a list
words and punctuations for each sentence. After that, we
applied lemmatization to extract the exact verb from differ-
ent tense forms of the verbs and removed non-alphabetical
characters. To understand the usage of word frequency, we
have demonstrated the TF-IDF Vectorization algorithm.
TF-IDF (term frequency-inverse document frequency) is
a statistical measure that evaluates how relevant a word is
to a document in a collection of documents. This is done
by multiplying two metrics: how many times a word ap-
7https://github.com/JustAnotherArchivist/snscrape
8https://github.com/tweepy/tweepy
Fig. 7: Examples of Extracted Tweets from Twitter; Due to rules
& regulations, account name can’t be shared publicly
pears in a document, and the inverse document frequency
of the word across a set of documents. Multiplying these
two numbers results in the TF-IDF score of a word in a
document. The higher the score, the more relevant that
word is in that particular document.To put it in more for-
mal mathematical terms, the TF-IDF score for the word t
in the document d from the document set D is calculated
as follows:
𝑡 𝑓 𝑖 𝑑𝑓 (𝑡 , 𝑑, 𝐷 )=𝑡 𝑓 (𝑡, 𝑑) ∗ 𝑖 𝑑𝑓 (𝑡, 𝐷 )(2)
where,
𝑡 𝑓 (𝑡, 𝑑 )=𝑙𝑜𝑔(1+𝑓 𝑟 𝑒𝑞 (𝑡, 𝑑 )) (3)
𝑖𝑑 𝑓 (𝑡 , 𝐷)=𝑙𝑜𝑔 𝑁
𝑐𝑜𝑢𝑛𝑡 (𝑑∈𝐷:𝑡∈𝐷)(4)
For sentimental analysis, we use a python library Textblob
9. This library contains a trained models that could deter-
mine the polarity and subjectivity of a given text. Polarity
ranges between -1and 1 with positive values reflecting
emotionally positive message and negative values reflect-
ing emotionally negative messages. Those that are neutral
would have polarity of 0. Subjectivity ranges between 0
and1 with 1 being subjective and 0 being objective. Using
both polarity and subjectivity would allow us to evaluate
the sentiments of twitter users toward fire issues.
5 Results and Discussions
5.1 Earth Observation (EO) Data Analysis
Using the methodology (see 4.1) applied for the fire map-
ping, the delta–NBR index analysis shows a spatial distri-
9https://textblob.readthedocs.io/en/dev/
IAC-21,B1,5,10,x65420 6

72𝑛𝑑 International Astronautical Congress (IAC) - Dubai, 25-29 October 2021.
Copyright 2021 by the authors. Published by the IAF, with permission and released to the IAF to publish in all forms.
bution of the fire severity (See Figure 9). Overall, there
are areas where fires had caused a lot of damage, most of
them located toward the south of burned area boundary.
Figure 8shows fire monitoring during the first week, the
second week, the first month, and the second month. In
the first week, the fire ranged from moderate-low severity
to high severity distributed toward the southern boundary
of the burned area (See Figure 9). However, in the second
week, the first month, and the second month, the sever-
ity had increased throughout the entire region. Toward
the northern part of the region, there were portions with
reduced severity to low and moderate-low. On the other
hand, in the southern part, the fire remained severe and
was classified as moderate-high and high severity. Fig-
ure 9shows a general severity map indicating the extent of
damages to the ecosystems during the most intense fire in-
terval between September and October. Some areas with
less intense fire were classified as enhanced regrowth-high
and enhanced regrowth-low. Also from MODIS data we
can see (Figure 6) the sudden changes in the Land Surface
Temperature (LST) in between September to November
compare to 2018 & 2019, due to 2020’s Bobcat fire.
Stavros and other NASA fire experts10 have been moni-
toring the blazes using a suite of satellite sensors. One of
them, the Operational Land Imager (OLI) on the Landsat
8 satellite, acquired an image (Figure 10) of the burn scar
(above) on September 21, 2020, while the fire was still
raging in Angeles National Forest. False color makes it
easier to distinguish the burn scar. The image combines
shortwave infrared, near-infrared, and green light (OLI
bands 7-5-2) to show active fires (bright red), scarred land
that has been consumed by the fire (darker red), intact
vegetation (green), and cities and infrastructure (gray).
The advantage of using geographic information systems
and remote sensing tools made it possible to analyze the
effects of fire before, during and after the fire, and generate
information related to severity that will be valuable for post
studies about recovery and ecosystem restoration. The
NBR is a consistent index to map severity, and its results
have been demonstrated in previous California wildfires
studies [5,16].
10https://earthobservatory.nasa.gov/images/147324/
bobcat-fire- scorches-southern- california
Fig. 8: Fire Severity Monitoring, First week to Second Month of
the fire
Fig. 9: Overall Bobcat Fire Severity
IAC-21,B1,5,10,x65420 7

72𝑛𝑑 International Astronautical Congress (IAC) - Dubai, 25-29 October 2021.
Copyright 2021 by the authors. Published by the IAF, with permission and released to the IAF to publish in all forms.
Fig. 10: Landsat 8 satellite image of the burn scar (above) on
September 21, 2020
5.2 Air Pollutants
Readings of PM2.5 from 7 stations within 40 km from the
vicinity of the Bobcat fire is shown in Figure 11. We found
that the time evolution of PM2.5 concentration from these
7 stations follows a similar pattern. The level of PM2.5
concentration is clearly elevated during the time of the
Bobcat fire and three distinct pulses of high pollutants can
be observed. This suggests that the Bobcat fire exhibited
complex evolution. We also found that the closest station
at Azusa, California generally recorded higher pollutants
than further away stations like the one in North Holly-
wood, CA. Further quantitative analysis could provide
insights on the migration of the pollutants.
We also create a spatial map of maximum daily PM2.5
concentrations measured during the Bobcat fire (see Fig-
ure 12). We found that this value reaches up to 100 𝜇𝑔/𝑚3
at a few nearby stations. The pollutants seem to be spread-
ing out over the entire Los Angeles basin. However, as
we move west toward Malibu, CA or south toward San
Diego, CA, the coastal ranges seem to be blocking the
path of PM2.5 and preventing the majority of pollutants
from escaping the Los Angeles basin.
Fig. 11: Time series of daily PM2.5 data recorded from 7
stations within 40 km from the vicinity of the Bobcat fire.
Distances are calculated from a reference point (latitude
34.2992◦N, longitude 117.9617◦W) located within the fire
perimeter.
Fig. 12: Spatial distribution of maximum daily PM2.5
concentration over the duration of the Bobcat fire measured at
each station
IAC-21,B1,5,10,x65420 8

72𝑛𝑑 International Astronautical Congress (IAC) - Dubai, 25-29 October 2021.
Copyright 2021 by the authors. Published by the IAF, with permission and released to the IAF to publish in all forms.
The ground measurements of PM2.5 concentrations used
in this study can provide insights into pollutants that are
produced as a result of wildfires or other anthropogenic
activities. However, their usage can be limited at times.
The measurements obtained are very close to the surface
and the values could change rapidly as we move up the
air column in our atmosphere. Moreover, these measure-
ments are done at discrete points and some areas might not
be as well monitored as others. For example, the Central
Valley of California has significantly much less number
of stations than the Los Angeles basin. Uneven distribu-
tion of ground stations can bias the interpretation of the
measurements.
5.3 Sentimental Analysis using Twitter Data
As defined earlier (Section 4.4) we have presented his-
togram (Figure 13) the magnitude of used those hashtags
during September, 2020 to December, 2020. As you can
see the hike in number tweets in the month of September
is quite accurate due to the fire. Also, clearly we can
see that people were twitting mostly from late evening to
early morning with "#bobcatfire" (Figure 14). With the
processed texts, the sentimental analysis reveals that about
55% shows positive polarity and 40% shows negative po-
larity. In terms of subjectivity of the texts, we found an
average subjectivity of 0.2 (see Figure 15). Since this
value is closer to 0 than 1, this means that the texts are
generally more objective than subjective. Since the con-
tent is generally more factual rather than opinions, it is not
unexpected that the sentimental of the majority of tweets
turns out to be closer to neutral.
6 Conclusions
As such, Twitter is becoming an increasingly important
tool to help people prepare for and recover from disas-
ters. In particular, the mechanisms by which information
is shared across networks during disaster events can have
significant implications for disaster damages and recov-
ery. Our work finds that the time frames in which people
communicate on Twitter varies by the time of event. Fur-
thermore, we find that it is people with “average” sized
Twitter networks that tweet most frequently during the fire
only. Each of these findings provides insight into poten-
tial strategies for disaster communication, based on both
the disaster context and the importance of general mes-
Fig. 13: Tweet Count Comparison between September, 2020 to
December, 2020. We have taken three hashtags; #Bobcatfire,
#Calfire, #CaliforniaFire
Fig. 14: Hourly Distribution of the Tweets consists #bobcatfire
IAC-21,B1,5,10,x65420 9

72𝑛𝑑 International Astronautical Congress (IAC) - Dubai, 25-29 October 2021.
Copyright 2021 by the authors. Published by the IAF, with permission and released to the IAF to publish in all forms.
Fig. 15: Sentimental Analysis of tweets consists #Bobcatfire.
55% Positive, 40% Negative, 5% Neutral words used in 41,971
tweets
saging that is applicable to typical Twitter users. Such
information can be useful for planning for future disasters
and enabling effective recovery following disasters, which
will ideally minimize disaster damages and help increase
resilience in a changing climate.
The NBR helped us to determine the damage caused by fire
in the area, and the risk not only for the ecosystem , but also
for people who lived nearby. This data will be useful to
examine effects of fire in short-or long-term for decision-
making about disaster management. We also found that
pollutants are highly elevated to hazardous levels during
the Bobcat fire and they are concentrated mostly in the
Los Angeles basin where most of the population lives. A
better understanding of the effects of fire on pollutants
could allow us to be more prepared for the next decades
where wildfires are expected to be more frequent and more
intense.
Furthermore, analyzing Twitter data along with remote
sensing data can be a good source to understand the men-
tal state of the community, estimate the number of injured
people, estimate who and what is affected by a natural
disaster and model the prevalence of epidemics. There-
fore, various groups such as politicians, government, non-
governmental organizations, aid workers and the health
system can use this information to plan and implement
interventions.
References
[1] K. Hoover and Laura A. Hanson. Wildfire statistics.
In Focus, (39), 2021.
[2] Penelope Morgan, Guillermo E. Defossé, and Nor-
berto F. Rodríguez. Management implications of fire
and climate changes in the western americas. In Fire
and Climatic Change in Temperate Ecosystems of the
Western Americas, pages 413–440. Springer-Verlag.
[3] FAO Forestry Paper. Fire management-global as-
sessment, 2006.
[4] Emilio Chuvieco and Russell G. Congalton. Appli-
cation of remote sensing and geographic information
systems to forest fire hazard mapping. Remote Sens-
ing of Environment, 29(2):147–159, aug 1989.
[5] Jay D. Miller and Andrea E. Thode. Quantifying
burn severity in a heterogeneous landscape with
a relative version of the delta normalized burn
ratio (dNBR). Remote Sensing of Environment,
109(1):66–80, jul 2007.
[6] Zhao-Liang Li, Bo-Hui Tang, Hua Wu, Huazhong
Ren, Guangjian Yan, Zhengming Wan, Isabel F.
Trigo, and José A.Sobrino. Satellite-derived land
surface temperature: Current status and perspec-
tives. Remote Sensing of Environment, 131:14–37,
2013.
[7] Sander Veraverbeke, Willem W. Verstraeten, Ste-
faan Lhermitte, Ruben Van De Kerchove, and Rudi
Goossens. Assessment of post-fire changes in land
surface temperature and surface albedo, and their
relation with fire–burn severity using multitemporal
modis imagery. International Journal of Wildland
Fire, 21:243–256, 2012.
[8] E.F. Lambin, K. Goyvaerts, and C. Petit. Remotely-
sensed indicators of burning efficiency of savannah
and forest fires. International Journal of Remote
Sensing, 24(15):3105–3118, 2003.
IAC-21,B1,5,10,x65420 10

72𝑛𝑑 International Astronautical Congress (IAC) - Dubai, 25-29 October 2021.
Copyright 2021 by the authors. Published by the IAF, with permission and released to the IAF to publish in all forms.
[9] Jon E. Keeley. Fire intensity, fire severity and burn
severity: a brief review and suggested usage. Inter-
national Journal of Wildland Fire, 18(1):116, 2009.
[10] Jia C. Liu, Gavin Pereira, Sarah A. Uhl, Mercedes A.
Bravo, and Michelle L. Bell. A systematic review of
the physical health impacts from non-occupational
exposure to wildfire smoke. Environmental Re-
search, 136:120–132, jan 2015.
[11] Stephanie E. Cleland, Jason J. West, Yiqin Jia,
Stephen Reid, Sean Raffuse, Susan O’Neill, and
Marc L. Serre. Estimating wildfire smoke concentra-
tions during the october 2017 california fires through
BME space/time data fusion of observed, modeled,
and satellite-derived PM2.5. Environmental Science
and Technology, 54(21):13439–13447, 2020.
[12] Nadin Boegelsack, Jonathan Withey, Gwen
O’Sullivan, and Dena McMartin. A cr itical examina-
tion of the relationship between wildfires and climate
change with consideration of the human impact.
Journal of Environmental Protection, 09(05):461–
467, 2018.
[13] Thomas W. Porter, Wade Crowfoot, and Gavin New-
som. 2020 Wildfire Activity Statistics. California
Department of Forestry and Fire Protection, 2020.
[14] John Roach. California wildfires will cost tens of bil-
lions, AccuWeather estimates. AccuWeather, 2019.
[15] Yuanyuan Wang and Guicai Li. Analysis of
&#x201cfurnace cities&#x0022 in china using
MODIS/LST product (MOD11a2). In 2014 IEEE
Geoscience and Remote Sensing Symposium. IEEE,
jul 2014.
[16] Allison E. Cocke, Peter Z. Fulé, and Joseph E.
Crouse. Comparison of burn severity assessments
using differenced normalized burn ratio and ground
data. International Journal of Wildland Fire,
14(2):189–198, 2005.
IAC-21,B1,5,10,x65420 11