ArticlePDF Available

Early Wildfire Detection Technologies in Practice—A Review


Abstract and Figures

As fires grow in intensity and frequency each year, so has the resistance from their anthropic victims in the form of firefighting technology and research. Although it is impossible to completely prevent wildfires, the potential devastation can be minimized if fires are detected and precisely geolocated while still in their nascent phases. Furthermore, automated approaches without human involvement are comparatively more efficient, accurate and capable of monitoring extremely remote and vast areas. With this specific intention, many research groups have proposed numerous approaches in the last several years, which can be grouped broadly into these four distinct categories: sensor nodes, unmanned aerial vehicles, camera networks and satellite surveillance. This review paper discusses notable advancements and trends in these categories, with subsequent shortcomings and challenges. We also describe a technical overview of common prototypes and several analysis models used to diagnose a fire from the raw input data. By writing this paper, we hoped to create a synopsis of the current state of technology in this emergent research area and provide a reference for further developments to other interested researchers.
Content may be subject to copyright.
Citation: Mohapatra, A.; Trinh, T.
Early Wildfire Detection Technologies
in Practice—A Review. Sustainability
2022,14, 12270.
Academic Editor: Diamando
Received: 31 July 2022
Accepted: 16 September 2022
Published: 27 September 2022
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
Copyright: © 2022 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
Early Wildfire Detection Technologies in Practice—A Review
Ankita Mohapatra * and Timothy Trinh
Computer Engineering Program, California State University, Fullerton, CA 92831, USA
As fires grow in intensity and frequency each year, so has the resistance from their an-
thropic victims in the form of firefighting technology and research. Although it is impossible to
completely prevent wildfires, the potential devastation can be minimized if fires are detected and
precisely geolocated while still in their nascent phases. Furthermore, automated approaches without
human involvement are comparatively more efficient, accurate and capable of monitoring extremely
remote and vast areas. With this specific intention, many research groups have proposed numerous
approaches in the last several years, which can be grouped broadly into these four distinct categories:
sensor nodes, unmanned aerial vehicles, camera networks and satellite surveillance. This review
paper discusses notable advancements and trends in these categories, with subsequent shortcomings
and challenges. We also describe a technical overview of common prototypes and several analysis
models used to diagnose a fire from the raw input data. By writing this paper, we hoped to create a
synopsis of the current state of technology in this emergent research area and provide a reference for
further developments to other interested researchers.
wildfire detection; unmanned aerial vehicle; satellite; infrared; wireless sensor networks
1. Introduction
Wildfires are a necessary natural phenomenon and have been a critical part of many
ecosystems for years [
]. However, the last decade has witnessed a higher-than-normal
frequency and magnitude of wildfires affecting thousands of people in various continents
(Figure 1). Besides causing damage to life and property, wildfires negatively impact the
climate and air quality of the surrounding region. The European Space Agency estimates
that approximately four million square kilometers of land are affected globally due to
wildfires [
]. In the US, the National Interagency Fire Center reports 50 major wildfire
events in 2022 recorded between January 1 and June 29—a number that surpasses the
10-year average—with about 192,016 acres burnt [3].
In the Northern Hemisphere, the fire season was typically marked as July to September.
However, due to shifting climate patterns and expanding drought areas, the fire season
has lengthened by several months in many countries. CalFire reports that the fire season
in the Sierras in California has increased by 75 days each year [
]. Furthermore, 75% of
the deadliest wildfires in the state have occurred since 2000 [
]. A scaled extrapolation for
similar impacts on other states and countries places wildfires as one of the most urgent
climate crises of this century.
The global impacts of wildfires are discussed in the following four categories.
Sustainability 2022,14, 12270.
Sustainability 2022,14, 12270 2 of 21
Figure 1.
Plot of acres burned from 1984–2020 in the United States shows an overall increase in the
burn severity and impact over the years. Data was obtained from the Monitoring Trends in Burn
Severity (MTBS) program by US Geological Survey Center for Earth Resources Observation and
Science (EROS) and USDA Forest Service Geospatial Technology and Applications Center (GTAC).
1.1. Environmental Impact
Although pyrodiversity is essential for maintaining biodiversity and biome distri-
bution, the balance is delicate and dependent on the intensity, size and season of the
wildfire [
]. The recent trends in climate changes and severity of wildfires have tipped
the scales negatively against the sustenance of floral and faunal diversity. Studies have
documented the adverse effects of wildfires on critically endangered species like Bornean
orangutans in Indonesia [
] and Palethorpe’s pinwheel snails, koalas and native vegetation
groups in Australia [
], among countless other species that have been pushed to the
brink of extinction by such fire events. Additionally, wildfires release copious amounts of
greenhouse gases, contributing to global warming and drought conditions that trigger new
fires—a never-ending positive feedback loop. To put the wildfire emission into perspective,
the California Air Resources Board estimated that the state’s 2020 wildfires emitted about
112 million metric tons of carbon dioxide, the highest recorded in the past decade [
]. This
number is about three times higher than the second-highest emission value recorded in
2008 and is equivalent to the amount of greenhouse gases released by all the passenger
vehicles in California annually. The European Union’s Copernicus Atmosphere Monitoring
Service estimated that 1.76 billion tons of carbon were generated by wildfires globally in
2021, with the emissions projected to increase significantly in the future.
1.2. Health Impact
The destructive nature of fires may cause immediate bodily harm, but its emissions,
some largely unseen, also bring danger to those outside of the surrounding area. For
example, Oregon’s Bootleg Fire had its smoke carried over three thousand miles from its
origin, bringing health effects to people in distant cities who may not have been aware of
the fire. It is a well-established fact that wildfires increase the concentration of particulate
matter and harmful gases in the air [
], which are directly linked to various diseases,
such as asthma, bronchitis, reduced lung volume, etc. [
]. Although the effect on
cardiovascular events are mixed, sufficient studies indicate that exposure to fire smoke
leads to respiratory morbidity [
]. Emerging research also hints at a possible correlation
between exposure to particulate matter and higher neurophysiological irregularities like
ADHD, autism, decline in memory, etc., in children [
]. Prenatal exposure to polycyclic
aromatic hydrocarbons, one of the constituents of fire, has been associated with childhood
obesity [
], while exposure to PM2.5 in the prenatal stage could possibly lead to higher
blood pressure in children [
]. Johnston et al., estimate that between 1997 to 2006, the
Sustainability 2022,14, 12270 3 of 21
average annual mortality attributed to fire smoke was 339,000, with Southeast Asia and
Africa experiencing the maximum brunt of the crisis [18]. Even short exposure to wildfire
smoke increases risk of premature death through lasting damages to the respiratory and
cardiovascular system.
1.3. Sociological Impact
Disadvantaged, low-income communities, people of color and elderly people face an
increased risk to wildfire. They are also disproportionately affected by the aftermath of
wildfire events, like smoke-related morbidity and mortality, homelessness, displacement,
psychological stress, etc. [
]. The 2017 and 2018 wildfires in northern California killed
more than a hundred people, displaced thousands more and annihilated an entire city,
exacerbating the homelessness crisis in the area [
]. Wildfires also cause power outages [
and deplete long-term water supplies for the community [23].
1.4. Economic Impact
Firefighting efforts and subsequent rehabilitation are excessively costly, and the costs
of mitigating wildfires are increasing exponentially each year. At a total loss of $12.5 billion,
California’s 2018 Camp Fire was the most expensive disaster globally in that year. Wildfires
incurred similar financial burdens on other countries, with state expenditures increasing
by millions of dollars [
]. Along with the immediate property damage and recovery
expenses, each fire disrupts the surrounding economy and workforce, causing a ripple
effect throughout the communities [25].
There are several ongoing global efforts to mitigate wildfires. As highlighted in
Figure 1, the frequency and impact of wildfires have escalated significantly in the past few
years, and this has prompted numerous research groups to propose their unique solutions
to combat this issue. An ideal wildfire monitoring system should be able to survey a large
area (spatial coverage) with reasonable frequency (temporal coverage).
Two popular approaches are predicting the likelihood of a fire in an area and detecting
wildfires still in early stages from the associated emission of gases, aerosols, temperature
rise, etc. Fires at young stages are simpler to control and mitigate. This paper discusses
many such efforts, with their merits and shortcomings. We hope to create a substantial
resource for the research community about the state of current technology and motivate
advancements needed to further reduce the losses from this disaster.
2. Literature Review
2.1. Review Methods
For this review, we chose published conferences and journal articles that are peer
reviewed. We have also presented a summary of the pivotal accomplishments in this area
of research, including those by companies that have contributed to the wildfire detection
technologies focused on in this review. A comprehensive review of all articles published in
this area is impossible, but we believe the selection of work in this review is an adequate
representation of the fundamental research trends. Figure 2shows a distribution of the
papers and articles we selected for this paper, in 5-year incremental periods. Satellites were
one of the first to be utilized for wildfire detection, with a notable surge in other methods
observed in the last couple of decades. This pattern correlates with the rising frequency of
fires shown in Figure 1.
Sustainability 2022,14, 12270 4 of 21
Figure 2. Frequency of papers and articles selected for review from each 5-year span.
2.2. Discussion
Many methods have been proposed for wildfire detection in the past. The first or-
ganized fire alert networks in the US came in the form of fire lookout towers built in the
early 1900s, after the Great Fire of 1910. These towers were built on top of high moun-
tains and staffed by “fire spotters”. The spotters would inspect the surroundings for
signs of fire and communicated using heliographs. As better technologies were made
available, fire towers were retired from service by the end of the 20th century. In this
paper, we focus on the advanced mechanisms currently utilized for early wildfire detec-
tion, which can be broadly classified into the following main groups:
Section 2.2.1
. Sensor
Section 2.2.2.
Unmanned Aerial Vehicles (UAV);
Section 2.2.3
. Camera Networks;
Section 2.2.4. Satellite Surveillance.
2.2.1. Sensor Nodes
Sensor nodes typically consist of low-power sensors like humidity, temperature and
gases to monitor the surrounding area for fire and generate alerts (Figure 3). When multiple
sensor nodes are arranged as a network distributed throughout an area of interest and capa-
ble of communicating wirelessly, they are referred to as a wireless sensor network (WSN).
Figure 3.
Basic sequence of operation for the majority of WSNs. Many may not share the exact
steps but will follow these core functions. (Images sourced from Creative Commons, following the
guidelines on Attribution 3.0 Unported, CC By 3.0 [
]. The only modifications to the images are
cropping and resizing).
Sustainability 2022,14, 12270 5 of 21
In Table 1, we have summarized the salient features and capabilities of several
such kinds of nodes designed by different research groups. Each node is operated
by a microcontroller, typically powered by solar energy and rechargeable batteries
for long-term operability. In the event of a fire, the rate of change in environmental
variables is considerably different from their normal daily variation. For example, the
daily temperature is characterized by a gradual change within a predictable range,
whereas the temperature differential during a fire can quickly surpass several hundred
degrees Celsius. Several algorithms have been developed to distinguish between wildfire
and non-wildfire conditions from sensor data behavior. Varela et al., proposed a low-
complexity algorithm using only temperature and humidity [
]. They created two
base functions using regression analysis with temperature and humidity as dependent
variables, and time as an independent variable. The base functions were modeled to
represent the behavior of the dependent variables in a fire event. The data received
from the sensor nodes can then be compared with these base functions to determine the
occurrence of an actual fire. Other researchers have also proposed more sophisticated
algorithms like cellular automata and machine learning to detect wildfire. Khalid et al.,
logged flame, smoke, temperature, humidity and light intensity sensors at a central
station, and used a Bayesian classifier to categorize fire events from non-fire events
with an accuracy of 97.2% [
]. Ammar and Souissi fed temperature, humidity and
wind speed to a fuzzy logic controller. In the first step, the data was subjected to
trapezoidal and triangular fuzzification functions, then compared to presets to infer
estimated fire risk levels as Very Low, Low, Medium, High and Very High for each
month [
]. Bolourchi and Uysal used a similar approach with temperature, smoke,
light, humidity and distance as inputs to a five-membership function-based fuzzy logic
system to determine the probability of fire. They also studied the dependency of fire on
each of these inputs as isolated variables [
]. Dampage et al., collected temperature,
relative humidity, light intensity and CO level from sensors under different climate
conditions and at various times of the day, and then calculated the threshold ratio R
between current data values and the values recorded 30 s ago. This further strengthened
the model with a machine learning algorithm to reduce false alarms. A dataset of
7000 sensor data samples was collected during no-fire and fire conditions to train the
algorithm, which was then used to corroborate the presence or absence of fire after
threshold analysis [
]. Yan et al., used an artificial neural network (ANN) to identify
the phase of combustion in real-time using CO2, air temperature and smoke sensors [
Their model considered three combustion phases: no fire, smoldering combustion and
flaming combustion, and was able to obtain a consistent accuracy of >82% when multiple
sensor input data were considered.
Abbassi et al.,
implemented a multiple-level alert
using a combination of KNN classifier and fuzzy logic [
]. In the first level, a cluster of
neighboring nodes generates an alert using K-means clustering. The second level adopts
a fuzzy inference system to evaluate the scale of fire and its direction of propagation.
This two-layer system was demonstrated to be more efficient than simple fuzzy logic
systems and generated fewer incorrect alarms.
These sensor nodes can also be used to predict the likelihood of a future wildfire
using algorithms such as the 30-30-30 rule of thumb [
] or the fire weather index (FWI),
besides the machine learning or regression methods described above. The 30-30-30 rule
states that a temperature above 30
C, humidity less than 30% and wind speed above
30 kmph is indicative of extreme fire risk. The FWI is a more elaborate indexing system
that predicts fire behavior from moisture content in the organic matter on the forest floor,
availability of combustible fuel, fire intensity and rate of fire spread. The fuel moisture
content is categorized into three numeric fuel moisture codes, which are determined from
temperature, relative humidity, wind and precipitation readings [35].
To analyze and log sensor data at a central node, numerous wireless transmission
methods such as long-range radio (LoRa), Bluetooth, Zigbee, XBee, cellular networks, etc.,
have been explored. Cellular networks provide higher bandwidth and lower latencies
Sustainability 2022,14, 12270 6 of 21
but are power hungry and extremely complicated to setup in remote areas without prior
LTE connectivity. Bluetooth range extends to only a few meters; therefore, it is unsuitable
for supporting data transmission in vast networks. LoRa and Zigbee have been the most
popular choices for communication, as they can be easily used in remote areas without any
cellular connectivity or Wi-Fi. Both are capable of transmitting data over longer ranges
compared to Bluetooth, with LoRa surpassing Zigbee’s typical transmission range for two
devices connected directly. LoRa network architectures are arranged in a star pattern, with
a gateway node at the center that must be in line of sight of the sensor nodes to exchange
data. In contrast, Zigbee networks that are set up in a mesh pattern can bypass the need for
line of sight, as intermediary sensor nodes with Zigbee can be used for relaying data. The
Zigbee network can also be set up to be self-healing and re-route the data pathway through
healthy nodes if any node in the normal pathway is damaged. Although Zigbee has higher
bandwidth than LoRa, LoRa is more suited for outdoor low-power IoT applications that
also require longer communication range with fewer nodes and simple encryption methods
at a lower cost. Researchers have also proposed direct communication of sensor nodes with
geostationary and low orbit satellite networks using a novel collaborative beamforming
(CB) technique [
]. In CB, neighboring nodes form a virtual antenna array and transmit the
shared data synchronously. Doing so increases the transmission range significantly without
the need of a signal booster or gateway, while still operating under low power consumption
restrictions on the sensor nodes. In a hybrid approach, unmanned aerial vehicles (UAVs)
sweep over the sensor node networks to facilitate data collection from sensor nodes, feasibly
solving the transmission issues faced by WSNs spread over large areas, data latency or
interruption due to damaged nodes. Such methods have been thoroughly reviewed by
Nguyen et al. [37].
Among the four technologies considered in this paper, sensor nodes are the simplest
and least expensive to design and implement. The resources required to build such
a node are widely available and supported by an extensive collection of open-source
codes, drivers, libraries, etc. These factors allow for a high degree of personalization
and customization, hence attracting relatively more users from diverse backgrounds and
promoting discoveries. Sensor nodes also consume significantly lesser amount of energy,
compared to their counterparts, for the same surveillance duration, therefore decreasing
the demands on design complexity for power supply and storage solutions. A well-
planned sensor node network can easily achieve the finest spatial coverage and temporal
frequency for monitoring an area for wildfires. In the last few years, this category has
received the most attention, with many large-scale projects funded by governments. For
example, in 2021 USA’s Department of Homeland Security (DHS) Science and Technology
directorate collaborated with four industry partners—Ai4 Technologies (San Francisco,
CA, USA), Breeze Technologies (Hamburg, Germany), N5 Technologies (Rockville, MD,
USA) and Valor Fire Safety of Londonderry (Londonderry, NH, USA)—to find innovative
solutions for early wildfire detection [38].
Despite its advantages, a sensor node network faces the crippling challenge of needing
human power or elaborate methods like autonomous helicopters for initial deployment.
This challenge can quickly devolve into an insurmountable hurdle when the target terrain
is remote, vast and hostile. However, once deployed, sensor nodes can function indepen-
dently for several years. The obvious physical proximity of the nodes to fires presents
another hurdle. Some nodes are likely to be damaged in the intense heat generated by
fires. Retrieving and replacing these non-functional nodes are additional maintenance
issues that will need to be accounted for during planning and resource allocation. Addi-
tionally, many wilderness zones are protected by laws or ordinances against introduction
of unnatural/non-native objects, which can complicate the installation of these nodes.
Sustainability 2022,14, 12270 7 of 21
Table 1. Summary of sensor nodes designed for early wildfire detection.
Name of
Prototype/Reference Location Sensor Types Range Communication
Type Processing Power Source Year
FireWxNet [39] USA Temperature, relative humidity, wind speed and
direction 138–393 m 900–930 MHz radio ATmega128
Solar (two panel,
24 V and 12 V) and
four batteries (12 V)
Bayo et al. [40] Spain
Temperature (NTC), relative humidity (H25K5A,
SHT11), pressure (MS5540B), soil moisture (Decagon
EC5), light intensity (S8265)
100 m (comm) XBee/LR-WPAN ATmega1281 Two AA batteries 2010
Firoxio [41] Lebanon Relative humidity and temperature (SHT10), smoke,
carbon monoxide (MQ-5) Unknown Zigbee PIC16F877A
Solar (17.26 V),
700 mAh
lithium-ion battery
Yan et al. [32] China
Relative humidity and temperature (SHT11), smoke
(MS5100), carbon monoxide (EC805-CO), carbon
dioxide (S-100)
20 m Zigbee 8051 (included in
CC2430) Solar (12 V 7 W) 2016
Molina-Pico et al. [42] Spain Relative humidity and temperature (SHT75), gas
(carbon monoxide, carbon dioxide)
25 m (SN),
1.6 km (CN)
433 MHz ISM
between SN and GW,
868–870 MHz and
GSM/GPRS between
GW and CN
PIC24FJ256GB110 for
CN, MSP430 for SN
600 mAh Lithium
coin battery 2016
Lutakamale and
Kaijage [43]Tanzania Temperature (LM35DZ), smoke (MQ-2), relative
humidity and temperature (DHT22)
100–120 m (SN
to GW)
Zigbee between SN
between GW and CN
Arduino Uno
Two 3.7 V
SISVIA Vigilancia y
Ambiental [44]
Spain Waspmote gas board (temperature, humidity, light
intensity, carbon monoxide, carbon dioxide) 70 m ZigBee ATmega1281 Rechargeable AA
and solar panel 2017
Smart Forests [45] Brazil Temperature, relative humidity 100 m WPAN, Bluetooth
Low Energy N/A Batteries 2018
Kadir [46] Indonesia Temperature, humidity, smoke, carbon dioxide Unknown ZigBee Unknown Direct power
supply 2018
LADSensors [47] Portugal Temperature, humidity, air pressure, carbon dioxide 300 m (SN) LoRa Unknown Solar 2018
Silvanet (Dryad) [48] Germany Temperature, humidity, air pressure, gases (hydrogen,
carbon monoxide, etc.) (BME 688) 100 m LoRaWAN STM microcontroller
Solar and
supercapacitors for
energy storage
Sustainability 2022,14, 12270 8 of 21
Table 1. Cont.
Name of
Prototype/Reference Location Sensor Types Range Communication
Type Processing Power Source Year
Khalid [28] Turkey IR flame (760–1100 nm), smoke (MQ-2), light,
temperature and humidity (DHT-22) 250 m
NRF24L01+ (2.4 GHz
RF) ATmega328p Two Iithium-ion
cells (3.7 V) 2019
Knotifire [49] Canada Unknown Surface fire Internet Unknown Energy harvested
from fire 2020
BurnMonitor [50]France
and US Humidity, temperature 50 m 3G Unknown Unknown 2020
Benzekri et al. [51] Morocco
Temperature, humidity, air pressure (BME280),
particulates (Nova SDS011), carbon dioxide
(MH-Z14A-CO2), carbon monoxide (ZE07-CO)
Unknown LoRa Lora32u4
and lithium-ion
U. Dampage et al. [31] Sri Lanka
Temperature, humidity (DHT22), light intensity (LDR),
carbon monoxide (MQ9) 5 m 2.4–2.5 GHz ISM Arduino Nano
Solar panel and
lithium-ion cell
N5 sensors [52] USA Proprietary nanowire-based gas sensor array, IR
camera, particulate matter detector Unknown LoRa Unknown
Solar panel
and rechargeable
30,000 mAh battery
SN: sensor node; GW: gateway, CN: control node.
Sustainability 2022,14, 12270 9 of 21
2.2.2. Unmanned Aerial Vehicles (UAV)
UAVs commonly refer to vehicles or systems that are remotely operated and travel
by flight. The data gathered by UAVs are often in real-time, accurate and provide unique
vantage points that would otherwise be inaccessible, dangerous or time-consuming to
obtain by emergency responders. This is attributed to the mobility of the UAVs, which
allows rapid and continuous visual monitoring throughout the fire’s progression and
movement. While UAVs may define any unmanned aerial vehicle such as military drones,
the majority of them that were first designed for wildfire fighting purposes were comparable
to consumer or hobby drones, operating within 1000–4000 feet, as shown in Figure 4[
This range has increased to 30,000 ft as drone technology has commendably progressed
in the past years [
]. The data collected commonly take the forms of GPS location,
images, video feeds or readings from sensor nodes such as those described in the previous
section [
]. These systems may be remotely controlled by humans or automated systems.
Automated systems are easily adaptable by users with little to no experience compared to a
system that requires human pilots. For example, Fotokite Sigma can follow a programmed
flight path set through an application on a mobile device [
]. This is particularly useful
for applications such as detecting wildfires, where the autonomy is a necessity to support
persistent vigilance. A few years ago, autonomous UAV were limited to low-complexity
surveillance tasks but have now been vastly improved by researchers. Chi Yuan et al.
discuss a system where fleets of different types of semi-autonomous drones are deployed
in stages to search, confirm and observe fires [
]. Although each drone may not handle
dynamic and complex tasks such as those with dedicated pilots, the autonomy allows for
quick commanding and organization within a fleet regardless of its size. With general
usages laid, a UAV system’s main limitation is summarized by its relatively low-capacity
power source. This is due to the structural constraints limiting usage of heavy energy-dense
batteries. UAVs are usually outfitted with various cameras (videos, IR and imaging) but
may contain many other features depending on the designer [
]. For example, some
systems are capable of controlling a fleet of UAVs. Others may have additional devices
onboard, such as a specialized fire sensor or a tank of water [
]. The IGNIS system
by Drone Amplified is an example of an advanced drone that utilizes several of these
features [
]. It combats wildfires by initializing backfires through chemical payloads, and
can be actively controlled by both remote controller and/or by preset instructions sent by a
mobile app.
Figure 4.
Possible roles of UAVs. (Images sourced from Creative Commons, following the guidelines
on Attribution 3.0 Unported, CC By 3.0 [
]. The only modifications to the images are cropping
and resizing).
Several data processing algorithms have been proposed to process the video feed from
the UAV and swiftly detect the presence of active wildfires, intensity of the fire and rate
Sustainability 2022,14, 12270 10 of 21
of spread. Lin et al., employed a Kalman filter-based approach that uses only the UAV
position and temperature recorded at each sampling point to identify a wildfire and predict
the fire-spread behavior [
]. The performance of this model with multiple vehicles was
observed to be comparable to the results obtained with the FARSITE benchmarks, with less
computational cost. Bushnaq et al., proposed a combined UAV–IoT system as a more cost-
effective solution than satellite surveillance. In a fire event, the ground-based IoT devices
in the vicinity of the wildfire generate alarms to the UAVs nearby, which then verify the fire
probability to avoid misdetections [
]. The authors designed an algorithm to optimize the
IoT density and number of UAVs to minimize the cost of resources while also maximizing
the fire detection probability. Lewicki and Liu designed a Fire Perception Box (FPB) with
RGB/IR cameras and an ARM microcontroller that can be installed on UAVs in a plug-and-
play manner [
]. At a suspected fire scene, the RGB image is first fed to a convolutional
neural network (CNN) classifier to calculate an RGB fire score, followed by the IR image
being fed to its corresponding pipeline to calculate an IR fire score. Both scores are combined
to establish the presence of fire, and subsequently find the fire localization hotspots and
RGB + IR heatmaps. Researchers have also demonstrated a large-scale YOLOv3- and
YOLOv5-based deep-learning network for fire identification from images and video feeds
captured by UAVs [
]. After training the YOLO network on annotated fire images, the
researchers were able to obtain high fire detection accuracies in real-time analysis of UAV
video feeds. The only disadvantage of this method is that the YOLO model needs to be
implemented on a ground-based high-performance computer, therefore the performance
of this system is contingent on an uninterrupted data transmission between base station
and UAVs. Besides these methods, other deep-learning methods such as recurrent neural
networks (RNN), long short-term memory neural network (LSTM), generative adversarial
network (GAN), deep belief network (DBN), etc., have also been shown to have good
accuracy with identifying wildfires [65,66].
Another active avenue of research is building efficient algorithms for coordinating
the communication between multiple UAVs and controlling their flight patterns, with the
aim of optimizing coverage of the area within the expected flight endurance period. The
methods vary from classic linear controllers such as proportional–integral–derivative (PID),
and linear quadratic regulators to complicated non-linear implementations, such as
neural networks, genetic algorithms, etc. [
]. La et al., designed a control framework
by incorporating the following multiple objectives into the cost function: deployment of
UAVs to the first location of fire detection, avoiding in-flight collision with other UAVs
by maintaining a minimum separation distance, maximizing coverage of the wildfire and
tracking progression of the fire front [
]. Mawanza et al., discussed a real-time cooperative
and adaptive fire monitoring method for multiple UAVs [
]. They trained a radial basis
function neural network (RBFNN) to counteract system uncertainties, aerodynamic drag,
etc. and a non-singular fast terminal sliding mode control (NFTSMC) function to dynami-
cally track the fire. A similar evolutionary multi-objective algorithm was also formulated
for dynamically tracking fire boundary with minimal resources, and the algorithm was
able to achieve 100% fire coverage with 15 UAVs in the FARSITE-simulated model [
There have been numerous examples of detailed control strategies for quadrotors, but the
detailed review of these methods is beyond the scope of this review paper. The reader is
encouraged to review the papers by Kim et al. [
] and Kangunde et al. [
] for a study of
such techniques.
UAVs have limited flight endurance, as the size of the power source on board is
constrained by the payload capacity of the UAV. To extend the flight time of UAVs, solar
panels have been carefully integrated into the design. Airbus’ Zephyr and Titan Aerospace’s
Solara set the record for longest flight times, with several days of autonomous flight. Both
are equipped with solar panels and are intended to be used as high-altitude long-endurance
(HALE) UAVs for atmospheric monitoring [
]. HALEs fly at altitudes of about
30,000 feet
and cost millions of dollars. In contrast, solar-powered low-altitude low-endurance (LALE)
UAVs have been able to achieve maximum flight times of a few days at an altitude of
Sustainability 2022,14, 12270 11 of 21
only <10,000 ft, with the record set by ETH Zurich’s AtlantikSolar at 81 h of continuous
flight [
]. In theory, the flight time increases proportionally with flight altitude, explaining
the stark difference in maximum flight times achieved in HALEs vs LALEs. However,
HALEs have expansive wingspans and require a runway, which make them impractical
for many applications. Using an ATmega328 microcontroller-based photovoltaic power
management system, researchers were able to achieve a continuous surveillance of about
6 h using a swarm of six LALE UAVs [
]. XSun’s SolarXOne claims to be able to fly
continuously for 12 h, powered by solar cells [
]. It can support a payload of 5 kg, and
provides customizable options including RGB or IR cameras, and other sensors which
can be used to detect wildfires. Zhao et al., proposed a thoroughly detailed method for
designing a solar-powered, hand-launched UAV considering payload capacity, expected
solar irradiance, aerodynamic efficiency considerations at proposed flight altitude and
flight control [
]. Using their model, they were able to achieve flight endurance durations
of several hours at an altitude of 15,000 ft.
The biggest advantage of UAVs is flexibility in repositioning, which allows dynamic
readjustment of areas surveyed as deemed necessary. UAVs can also be programmed
to provide high temporal and spatial resolution over areas determined to be at high
risk. Unlike static nodes or camera networks, UAVs’ usage often extends past the initial
detection, as they can follow the spreading patterns of fires from a safe distance. Using a
limited number of UAVs to scan an area through multiple passes can be more efficient than
installing a vast sensor node network. They are less expensive to implement than satellites
and may also be more capable of identifying smaller wildfires than the latter technology.
Most UAVs still need some form of human involvement in and throughout their
operation. The major hurdle for this technology is distance and duration of flight. Despite
the advancements in assimilating solar panels into the device, the flight times are practically
limited to a few days for hand-launched LALEs. Therefore, this technology is likely to
be unsuitable when uninterrupted surveillance is needed in wildfire-prone areas. The
technology is also relatively new; therefore, the costs of operating and maintaining a UAV
fleet can add up exponentially. UAVs are also mired in frequent software and hardware
issues that have impeded their adoption as a widespread firefighting solution.
2.2.3. Stationary Camera Networks
Camera networks consist of advanced and feature-rich, interconnected cameras that
monitor a vast area for fires (Figure 5). Early camera networks consisted of live camera
images or videos streamed to a control room, where an operator would manually scan the
feeds for signs of fire. In modern camera networks, the cameras are still the main drivers
of the system, but are often partnered with other systems, such as AI and communication
servers, to fully optimize the camera to the desired area. Some may also begin as a single,
standalone camera with the option of expansion through additional cameras. They can
be stationed both in urban and remote areas and combat fires through early detection
and prediction. While stationary, the cameras are usually free to pan in wide or complete
angles, significantly increasing coverage zone per camera and limiting blind spots.
In addition to video feed, the cameras may also include different types of IR and thermal
to enhance detection ability by broadening the features captured. The integration of AI
enables them to be autonomous, allowing them to succeed the bulkier, man-powered and
more intrusive watch towers [
]. This allows the cameras to not only capture images of
the surroundings, but to independently identify if the captures are indicative of a fire. The
inclusion of internet access grants the system the ability to communicate, allowing them to
broadcast their feeds to a network which directly notifies specified users of notable events.
For example, a camera may capture a column of smoke an abnormality that its AI program-
ming classifies as a probable fire. It then relays the information to a central processor, the
Sustainability 2022,14, 12270 12 of 21
component that notifies authorities through an app [
]. A widely implemented example
of camera networks is ALERTWildfire, a collaborative effort between The University of
Nevada (Reno), University of California San Diego and the University of Oregon [
Hundreds of ALERTWildfire cameras have been installed in the southwest and west United
States and assisted with real-time monitoring of fires. In 2021, Pacific Gas and Electric
(PG&E) installed AI-equipped ALERTWildfire cameras in northern and central California,
in collaboration with the AI company Alchera [
]. A San Francisco company founded in
2019, Pano AI, also uses AI on HD camera video feeds to automatically detect wildfires and
reduce the response time to fire events [
]. IQ-Firewatch, a Portuguese company, employs
AI as well, but also provides customization and scalability to serve a variety of settings [
Their cameras, commercially sold, are optionally attached with sensors like monochrome,
RGB, near-IR, or thermal IR and can smoothly work with cameras added in the future. Its
method of wildfire prevention revolves around capturing high-quality raw images with
cameras and processing them in real-time using smoke detecting algorithms [
]. Detecting
fires at the earliest signs of smoke may be sufficient in many cases, but IQ-Firewatch and
similar systems directly compete with ground level technologies such as sensor nodes,
which may detect fires even before they emit visible amounts of smoke due to their sen-
sors and proximity to the fire [
]. Furthermore, while camera systems are not directly
constrained by the need to conserve power, they are tethered to the nearest power source,
requiring users to consider optimal placements. By the same token, they benefit from stable
power, eliminating problems with power sources such as batteries, allowing them to utilize
more power-intensive tasks such as real-time data processing.
Figure 5.
Typical application of camera networks paired with AI. (Images sourced from Creative Com-
mons, following the guidelines on Attribution 3.0 Unported, CC By 3.0 [
]. The only modifications
to the images are cropping and resizing).
Although each designer’s techniques are trade secrets and often undisclosed, popular
methods often involve smoke detection and machine learning, including the usual feature
extraction from raw captures and applying various classifiers. Jie Shi et al., have reviewed
such algorithms and listed their common extracted features: color, motion, texture, shape
and energy [
]. Each provides unique attributes of a smoke image that enhances the
classification process. The next steps involve choosing classifiers, which can be of many typ-
ical sorts such as: k-nearest-neighbor, neural networks, support vector machines, etc. [
The models programmed in the cameras are trained and tested before deployment. The
decoupling of the capturing and processing provides the advantage of continuous improve-
ment and integration as researchers can update the models and analyze past performance
without having to alter any of the physical components or interrupt its ongoing operation.
For example, by basing their project on cloud computation [
], Darko Stipaniˇcev et al.,
have continued to release improvements to their iForestFire semi-automatic monitoring
system after its commercial deployment and operations in various regions.
Many techniques have successfully utilized smoke detection in early wildfire detec-
tion prototypes; yet, researchers continue to refine and improve the process. A common
Sustainability 2022,14, 12270 13 of 21
weakness in smoke detection and image-based detection is deterioration of image quality
at night and by the presence of haze. Haze is defined as an “an obscurity occurring due to
smoke, dust, or other particles suspended in the atmosphere” [
]. This causes images and
colors to become fuzzy or blurry, decreasing contrast between elements of the image and
therefore increasing the difficulty of identifying smoke. Some examples of haze are fog and
air pollution. To address this, Darko Stipaniˇcev et al., utilized a de-hazing technique known
as “dark channel prior”. It reconstructs haze-reduced images with equations generated,
with the knowledge that (1) most haze-free non-sky patches of an image have low or
zero intensity in at least one color channel, and (2), that the intensity of these patches is
increased by the presence of haze [
]. Like haze, the lack of light significantly diminishes
the usability of captures taken at night. Inadequate lighting prevents distinguishing of
colors and behaves like noise, often causing fire misclassification with other light sources
such as car lights or lanterns. To combat this, Ahmet Agirman et al., extract smoke temporal
features and motion in addition to spatial features, considering factors such as flickering
and flaring [
]. With the extra information, their process uses two machine-learning
stages: the GoogLeNet CNN first extracts spatial features to then pass onto a bidirectional
long short-term memory network (BLSTM), which learns the temporal relationships in
the data [
]. The models’ decisions are then congregated and ultimately decided through
majority voting [85].
Camera networks are largely autonomous and positioned farther away from the
site of potential fires, unlike sensor networks and some UAVs. Therefore, there are little
opportunities for them to receive any sort of damage from fires. Although considerations
still need to be taken in the camera’s placement, it is simplified by the longer range and
more straightforward operation when compared to sensor nodes, whose low coverage and
various short-range sensors necessitate more planning.
The relatively large coverage of a stationary camera partially alleviates, but does
not eliminate, the constraint of a required continuous power source, as video feeds or
frequent captures are power consuming. Consequently, it may be impossible to use these
systems in an extremely remote area. The setup of the camera towers also needs an initial
investment of resources and manpower for building the towers, and, therefore, might not
be economically feasible in many geographical locations. Due to this positioning of the
camera, any sensors considered for inclusion in the camera system may also need to have
similar range or function as the cameras, limiting possible additions. For example, it would
be ineffective to outfit cameras with temperature or gas sensors, as cameras are designed to
be distant from the target area.
2.2.4. Satellite Surveillance
NASA and NOAA were two of the first organizations to observe wildfires using an
extensive network of polar orbiting (Terra, Aqua) and geostationary (GOES) satellites.
Polar satellites scan the entire Earth a few times each day and can monitor the entire planet
for fires (Figure 6). However, each consecutive scan over the same geolocation is spaced
several hours apart and, therefore, the temporal sampling rate for such satellites is low. On
the other hand, geostationary satellites provide much higher temporal data for a specific
area but cannot monitor other global environmental developments. The instruments on
board these satellites monitor several different kinds of data, some of which are actively
analyzed to detect/observe wildfires, global transport of pollutants and long-term climate
impacts of fire (Table 1, [
]). Processing this data for wildfire anomalies, especially small
fires, poses a challenge due to the lower spatial resolution of satellite images. In addition,
smoke can easily appear identical to clouds, as shown in Figure 6of the Camp Fire, imaged
by moderate resolution imaging spectroradiometer (MODIS) on Terra.
Sustainability 2022,14, 12270 14 of 21
Figure 6.
MODIS image of the Camp Fire in northern California on Nov 14, 2018 [
]. Image by
NASA Earth Observatory.
Most environment observational satellites have visible and infrared detectors on board.
Among the infrared bands, the 3.8/4
m (middle infrared) and 10.8/11
m (long-wave
infrared) wavelengths are used to identify active fires (Table 2). The relative brightness
temperatures of fire pixels on these two bands are different from each other, with the 4
channel (T
) registering a significantly increased radiance for high-temperature sources
than the 11
m channel (T
). The difference in brightness (
) of pixels between
these two channels is compared with a pre-determined threshold to detect fires and map
affected areas. This simple and intuitive approach was first proposed by Matson and
Dozier in 1981 on AVHRR data [
]. Additional IR channels are often used to improve
detection, by helping reject false alarms caused by clouds, bright surfaces, sun glints,
water vapor, etc. [
]. A similar approach was applied to data from MODIS, ASTER
and GOES, with many iterations proposing improved sensitivity to smaller and lower-
temperature fires [
]. One of the major challenges of the earlier algorithms was that
the threshold levels for comparing
T were empirically determined, and therefore varied
across different kinds of biomes on Earth. In a modified approach (version 4),
Giglio et al.,
proposed subsequent steps after the basic absolute threshold test has identified a fire pixel
by comparing T
or T
to a certain numeric value [
]. In the following step, the pixels
neighboring the pixel under consideration are used to estimate the radiance of the fire pixel
in the absence of a fire. The pixel window is started at 3
3 and gradually increased until
sufficient ground-based, non-fire pixels without clouds have been registered to compute
the background signature. If this step is successful, a third step uses this background
signature to calculate threshold values that hold contextual relevance to the specific area
being investigated.
T is now compared to this threshold for identifying a tentative fire. In
the last step, further tests using another set of different threshold formulae help in rejecting
Sustainability 2022,14, 12270 15 of 21
false alarms due to sun glints, deserts and coastal boundaries, water bodies, clouds and
other sources of abrupt radiance transition.
In a radically alternate approach compared to the above method, deep-learning tech-
niques were proposed for active fire identification. To facilitate this, Pereira et al., generated
a publicly available dataset of image patches from Landsat-8 satellite covering global fire
events in 2021 [
]. The database also included annotated fire pixels and is available on
GitHub. The Landsat-8 was launched in 2013 and provides data in visible, SWIR, NIR
and thermal IR spectrums (Table 2). The Operational Land Imagers (OLI) channels 1, 5, 6
and 7 are typically used for fire detection [
]. Using the Pereira dataset [
] for training,
Rostami et al., developed a convolutional neural network (CNN)-based architecture called
“MultiScale-Net” for finding fires [
]. To accommodate for fires of different sizes and
patterns, the CNN used multiple sized kernels for feature extraction and changing dilation
rates for the dilated convolution layer. SWIR bands and blue channel data were used as
inputs to the CNN.
As an attempt to make satellite technology accessible to the general academic commu-
nity, the CubeSat project was started in 1999 by a collaborative effort between California
Polytechnic state University (San Luis Obispo) and Stanford University. The objective of
this project was to boost research exploration and invention at affordable budgets. Cube-
Sats have a standardized design and are launched as part of the payload on other satellite
launches [
]. Gangestad et al., used the images from 1U AeroCube-4 equipped with
three 2 MP cameras to demonstrate CubeSats as a viable option for wildfire detection [
However, downloading the images to a ground station usually takes hours. To mitigate
this challenge, Azami et al., proposed k-nearest and CNN deep-learning (DL) models for
image classification implemented on a Raspberry Pi [
]. The runtime for this RPi unit
was optimized to minimize the power consumption and was integrated on to a KITSUNE
6U CubeSAT. The KITSUNE uses a Sony IMX342 color sensor that generates the images
fed into the DL model for classification. Using this setup, the group was able to achieve
classification accuracy of above 95% using networks like ShallowNet and LeNet [
Another recent example is Orora Technologies, which specifically designs 3U CubeSats
furnished with IR cameras for finding and monitoring wildfires from space [
]. They
claim to be currently protecting 302,588,843 hectares of forests spread over six continents.
The Orora CubeSats have onboard GPUs to assist with wildfire detection without needing
to download the data to the ground station for analysis.
Satellites can survey large expanses of land in a single sweep and provide an expansive
perspective of fire behavior. They are also built to last for several decades and serve multi-
ple purposes beyond fire detection. For example, the NASA and NOAA meteorological
satellites also track oceanic currents, upper atmospheric climate changes, drought risks
and patterns, etc. They are adept at generating a comprehensive view of the biosphere
and environmental processes, of which wildfires are only a contributing factor. The exten-
sive data generated are available for public use and have spurred many discoveries and
breakthroughs in many diverse disciplines.
Aside from the cost aspects, the obvious bottleneck of wildfire observation using
satellite technology is optimizing the area covered with the frequency of data collection.
The high-flying altitudes limit the resolving power of fires smaller than a pixel in the images,
which can make identification of young fires extremely difficult. Dense clouds and smoke
cover can severely reduce ground visibility to even IR and, therefore, adversely impact their
utility for tracking active fires under such conditions. Until the end of the 20th century, this
technology was available only to a few niche organizations, due to high mission cost and
limited capacity for onboard systems. Although CubeSats project have largely addressed
this issue, launching a CubeSat network is still expensive compared to other methods
discussed in this paper. Currently, the cheapest option is Interorbital Systems (IOS) TubeSat
Sustainability 2022,14, 12270 16 of 21
satellite kits, at about USD 16,000 (academic pricing) and USD 32,000 (commercial pricing)
for a 1U kit, including launch. The price for a single 3U kit with inclusive launch services is
USD 36,480 (academic pricing) and USD 72,960 (commercial pricing) [99].
Table 2.
NOAA and NASA satellites that help in wildfire detection, tracking and for studying its
global impacts on climate and pollution.
Instrument Notes Launch Date
Airborne Visible/Infrared
Spectrometer (AVIRIS) [100]
Optical sensor images 224 spectral bands between 400–2500 nm
Assesses fuel availability and condition, and stage of the fire
Carried by ER-2 jet, WB-57, Twin Otter International’s
Turboprop, etc.
Advanced Very High-Resolution
Radiometer (AVHRR)
(NOAA) [101]
Measures data in six channels in 0.6 µm–12 µm range
Records land surface temperature, fire
temperature/expanse/radiative power 1978–1994
Advanced Spaceborne Thermal
Emission and Reflection
Radiometer (ASTER) [102]
Records images in 14 wavelengths between visible to thermal
IR, including VNIR, SWIR and TIR bands
Captures high resolution images and surface temperature
of Earth
Helps track smokes and active fires
Carried by Terra
18 December 1999
Moderate Resolution Imaging
Spectroradiometer (MODIS) [103]
Records 36 spectral bands data between 0.4 µm–14.4 µm
Monitors fires and hot spots through visible and IR imaging
Carried by Terra and Aqua
Dec 1999 (Terra);
May 2002 (Aqua)
Multi-angle Imaging
SpectroRadiometer (MISR) [104]
Nine cameras gather RGB-IR data from different angles
tracks smoke plume and concentration of smoke particulates
in air
Tracks wildfires
Carried by Terra
18 December 1999
Measurement of Pollution in the
Troposphere (MOPITT) [105]
Measures IR radiation between 4.7 µm–2.4 µm wavelengths
tracks emission and movement of carbon monoxide globally,
and emission contributions from fires
Carried by Terra
18 December 1999
Atmospheric Infrared Sounder
(AIRS) [106]
Tracks global movement of greenhouse gases, including
emission from fires, and IR mapping of Earth’s surface
Helps understand impact on weather patterns, climate change
and forecast droughts
Carried by Aqua
4 May 2002
Cloud Aerosol Lidar with
Orthogonal Polarization
(CALIOP) [107]
Uses 532 nm and 1064 nm LIDAR
Tracks smoke plumes and aerosol concentrations, focusing on
vertical profiles
Carried on CALIPSO
CALIPSO also has a 645 nm widefield camera, infrared
radiometers (8.65
m, 10.6
m, 12
m) to assist in data collection
28 April 2006
Sustainability 2022,14, 12270 17 of 21
Table 2. Cont.
Instrument Notes Launch Date
Visible Infrared Imaging
Radiometer Suite (VIIRS) [108]
Spans visible and infrared wavelengths, covers 22 channels
between 0.41 µm–12.01 µm
Similar to MODIS but gathers data at higher resolution. Also
measures aerosol, among other parameters
Carried by Suomi National Polar-Orbiting Partnership
(NPP) satellite
28 October 2011
Hyperspectual Thermal Emission
Spectrometer (HyTES) [109]
Covers 256 spectral channels between 7.5 µm–12 µm
Determines fire temperature, tracks gas plumes of methane,
hydrogen sulfide, ammonia, sulphur dioxide, nitrogen dioxide
Intended to support the HyspIRI mission that will track
natural disasters
July 2012
Landsat 8 [110]
The first Landsat was launched in 1978
The Landsat-8 has an Operational Land Imager (OLI) and
thermal infrared sensors
Provides data in visible, near-IR, SW-IR and thermal IR
ECOsystem Spaceborne Thermal
Radiometer Experiment on Space
Station (ECOSTRESS) [111]
Measures six spectral bands data in 8.29 µm–12.09 µm range
Helps estimate drought levels by measuring transpiration and
water stress in critical areas
A potential fire identification product is currently being tested
29 June 2018
3. Conclusions
In this review paper, we emphasized the impact of wildfires on human life and
global ecosystems, and the current climate research indicates worsening fire seasons in
the future. Advanced technologies like satellites, drones, ground-based sensor nodes and
camera systems have been proposed to supplement traditional firefighting techniques.
We discussed the stages and current state of research in these four main methodologies,
specifically focusing on detecting wildfires at early stages, predicting hot spots at high risk
and monitoring the spread. The paper highlights the technical aspects of the front end of
the device, along with the algorithms used in the back end for processing the data and
identifying an actual fire. Each category has its own shortcomings and limitations, but also
excels at its own niche applications, and under different circumstances one technology will
outperform the other. For example, satellites have the capability of scanning a considerable
area in a short period of time, which is either impossible or time-consuming using the other
methods. However, satellites have very low resolving power and typically sweep the same
area once in several hours, making them unsuitable for real-time fire detection. Stationary
camera networks address these issues but are immobile and need to be tethered to a power
source. UAVs are characterized by flexible flight paths that can be used to adapt to the
changes in fire patterns and survey different areas as required. However, they are also
power hungry, and are inadequate for long-term continuous monitoring. In comparison
to the above, sensor nodes are the most low-power units that can be placed to monitor
very remote locations continuously, are relatively inexpensive to produce on a large scale
and do not need intensive data processing like those for analyzing images from UAVs,
satellites or cameras. On the other hand, nodes face the highest risk of damage due to their
proximity to fires on the ground. Ultimately, none of these methods can fully substitute
the usefulness of the others, but all four can be unified for a robust firewatch system that
together can exhaustively detect wildfires around the clock. The shortcomings and gaps
in research in each area were underlined, which we hope will be addressed in the future
through rigorous research.
Sustainability 2022,14, 12270 18 of 21
Funding: This research received no external funding.
Conflicts of Interest: The authors declare no conflict of interest.
1. Pausas, J.G.; Keeley, J.E. Wildfires as an ecosystem service. Front. Ecol. Environ. 2019,17, 289–295. [CrossRef]
The European Space Agency. New Long-Term Dataset to Analyse Global Fire Trends; The European Space Agency: Paris,
France, 2021
3. National Fire News: National Preparedness Level 2; National Interagency Fire Center: Boise, ID, USA, 2022.
4. Available online: (accessed on 25 July 2022).
California Department of Fish and Wildlife. 2022. Available online:
(accessed on 29 June 2022).
Steel, Z.L.; Collins, B.M.; Sapsis, D.B.; Stephens, S.L. Quantifying pyrodiversity and its drivers. R. Soc. Publ.
,288, 20203202.
[CrossRef] [PubMed]
Erb, W.M.; Barrow, E.J.; Hofner, A.N.; Utami-Atmoko, S.S.; Vogel, E.R. Wildfire smoke impacts activity and energetics of wild
Bornean orangutans. Sci. Rep. 2018,8, 7606. [CrossRef] [PubMed]
Wintle, B.A.; Legge, S.; Woinarski, J.C.Z. After the Megafires: What Next for Australian Wildlife? Trends Ecol. Evol.
753–757. [CrossRef] [PubMed]
Godfree, R.C.; Knerr, N.; Encinas-Viso, F.; Albrecht, D.; Bush, D.; Cargill, D.C.; Clements, M.; Gueidan, C.; Guja, L.K.;
Harwood, T.; et al
. Implications of the 2019–2020 megafires for the biogeography and conservation of Australian vegetation.
Nat. Commun. 2021,1023, 12. [CrossRef] [PubMed]
California Air Resources Board. 2020. Available online: (accessed on 29 July 2022).
Geng, G.; Murray, N.L.; Tong, D.; Fu, J.S.; Hu, X.; Lee, P.; Meng, X.; Chang, H.H.; Liu, Y. Satellite-Based Daily PM2.5 Estimates
During Fire Seasons in Colorado. JGR Atmos. 2018,123, 8159–8171. [CrossRef]
Stowell, J.D.; Yang, C.-E.; Fu, J.S.; Scovronick, N.C.; Strickland, M.J.; Liu, Y. Asthma exacerbation due to climate change-induced
wildfire smoke in the Western US. Environ. Res. Lett. 2021,17, 014023. [CrossRef]
Stowell, J.D.; Geng, G.; Saikawa, E.; Chang, H.H.; Fu, J.; Yang, C.-E.; Zhu, Q.; Liu, Y.; Strickland, M.J. Associations of wildfire
smoke PM2.5 exposure with cardiorespiratory events in Colorado 2011–2014. Environ. Int. 2019,133, 105151. [CrossRef]
Reid, C.E.; Brauer, M.; Johnston, F.H.; Jerrett, M.; Balmes, J.R.; Elliott, C.T. Critical Review of Health Impacts of Wildfire Smoke
Exposure. Env. Health Perspect 2016,124, 1334–1343. [CrossRef]
Holm, S.M.; Miller, M.D.; Balmes, J.R. Health effects of wildfire smoke in children and public health tools: A narrative review.
J. Expo. Sci. Environ. Epidemiol. 2021,31, 1–20. [CrossRef]
Rundle, A.; Hoepner, L.; Hassoun, A.; Oberfield, S.; Freyer, G.; Holmes, D.; Reyes, M.; Quinn, J.; Camann, D.; Perera, F.; et al.
Association of Childhood Obesity With Maternal Exposure to Ambient Air Polycyclic Aromatic Hydrocarbons During Pregnancy.
Am. J. Epidemiol. 2012,175, 1163–1172. [CrossRef]
Rosa, M.J.; Hair, G.M.; Just, A.C.; Kloog, I.; Svensson, K.; Pizano-Zárate, M.L.; Pantic, I.; Schnaas, L.; Tamayo-Ortiz, M.;
Baccarelli, A.A.; et al.
Identifying critical windows of prenatal particulate matter (PM 2.5) exposure and early childhood blood
pressure. Environ. Res. 2020,182, 109073.
Johnston, F.H.; Henderson, S.B.; Chen, Y.; Randerson, J.T.; Marlier, M.; Defries, R.S.; Kinney, P.; Bowman, D.M.J.S.; Brauer, M.
Estimated Global Mortality Attributable to Smoke from Landscape Fires. Environ. Health Perspect
,120, 695–701. [CrossRef]
US Fire Administration. Administration, Socioeconomic Factors and the Incidence of Fire; National Fire Data Center: Washington, DC,
USA, 2017.
Masri, S.; Scaduto, E.; Jin, Y.; Wu, J. Disproportionate Impacts of Wildfires among Elderly and Low-Income Communities in
California from 2000–2020. Int. J. Environ. Res. Public Health 2021,18, 3921. [CrossRef]
Gin, J.L.; Balut, M.D.; Der-Martirosian, C.; Dobalian, A. Managing the unexpected: The role of homeless service providers during
the 2017–2018 California wildfires. J. Community Psychol. 2021,49, 2532–2547. [CrossRef]
Ma, A.L.; Loughland, M.E.D.; Lapcharoensap, W.; Dukhovny, D.; Lee, H.C. California and Oregon NICU Wildfire Disaster
Preparedness Tools. Children 2021,8, 465. [CrossRef]
Robinne, F.-N.; Hallem, D.W.; Bladon, K.D.; Flannigan, M.D.; Boisramé, G.; Bréthaut, C.M.; Doerr, S.H.; Baldassarre, G.D.;
Gallagher, L.A.; Hohner, A.K.; et al. Scientists’ warning on extreme wildfire risks to water supply. Hydrol. Process.
35, e14086
Barbosa, J.V.; Nunes, R.A.O.; Alvim-Ferraz, M.C.M.; Martins, F.G.; Sousa, S.I.V. Health and Economic Burden of the 2017
Portuguese Extreme Wildland Fires on Children. Int. J. Environ. Res. Public Health 2022,1, 19. [CrossRef]
Reiff, N. How Fire Season Affects the Economy, Investopedia, 28 February 2022. Available online: https://www.investopedia.
20over%20two%20years (accessed on 5 July 2022).
Creative Commons Legal Code. Available online: (accessed on
21 July 2022).
Varela, N.; Diaz-Martinez, L.J.; Ospino, A.; Zelaya, N.A.L. Wireless Sensor Network for Forest Fire Detection.
Procedia Comput. Sci.
2020,175, 435–440. [CrossRef]
Sustainability 2022,14, 12270 19 of 21
Khalid, W.; Sattar, A.; Qureshi, M.; Amin, A.; Malik, M.A.; Memon, K.H. A smart wireless sensor network mode for fire detection.
Turk. J. Electr. Eng. Comput. Sci. 2019,27, 2541–2556. [CrossRef]
Ammar, M.B.; Souissi, R. A New Approach based on Wireless Sensor Network and Fuzzy Logic for Forest Fire Detection. Int. J.
Comput. Appl. 2014,89, 0975–8887.
Bolourchi, P.; Uysal, S. Forest Fire Detection in Wireless Sensor Network Using Fuzzy Logic. In Proceedings of the Fifth
International Conference on Computational Intelligence, Communication Systems and Networks, Riga, Latvia, 3–5 June 2013.
Dampage, U.; Bandaranayake; Wanasinghe, R.; Kottahachchi, K.; Jayasanka, B. Fire detection system using wireless sensor
networks and machine learning. Sci. Rep. 2022,12, 46. [CrossRef]
Yan, X.; Cheng, H.; Zhao, Y.; Yu, W.; Huang, H.; Zheng, X. Real-Time Identification of Smoldering and Flaming Combustion
Phases in Forest Using a Wireless Sensor Network-Based Multi-Sensor System and Artificial Neural Network. Sensors
16, 1228. [CrossRef]
Abbassi, M.A.E.; Jilbab, A.; Bourouhou, A. Efficient Forest Fire Detection System Based on Data Fusion Applied in Wireless
Sensor Networks. Int. J. Electr. Eng. Inform. 2020,12, 1–18. [CrossRef]
Cruz, M.G.; Alexander, M.E. The 10% wind speed rule of thumb for estimating a wildfire’s forward rate of spread in forests and
shrublands. Ann. For. Sci. 2019,76, 44. [CrossRef]
Natural Resources Canada. Canadian Forest Fire Weather Index (FWI) System. Available online:
background/summary/fwi (accessed on 9 July 2022).
Poulakis, M.I.; Vassaki, S.; Panagopoulos, A.D. Satellite-Based Wireless Sensor Networks: Radio Communication Link Design. In
Proceedings of the 2013 7th European Conference on Antennas and Propagation (EuCAP), Gothenburg, Sweden, 8–12 April 2013.
Nguyen, M.T.; Nguyen, C.V.; Do, H.T.; Hua, H.T.; Tran, T.A.; Nguyen, A.D.; Ala, G.; Viola, F. UAV-Assisted Data Collection in
Wireless Sensor Networks: A Comprehensive Survey. Electronics 2021,10, 2603. [CrossRef]
News Release: DHS S&T Successfully Evaluates Wildfire Sensors with California Emergency Responders, 10 Jun 2021. Available
sensors (accessed on 28 July 2022).
Hartung, C.; Han, R.; Seielstad, C.; Holbrook, S. FireWxNet: A multi-tiered portable wireless system for monitoring weather
conditions in wildland fire environments. In Proceedings of the 4th International Conference on Mobile Systems, Applications,
and Services (MobiSys 2006), Uppsala, Sweden, 19–22 June 2006.
Bayo, A.; Antolin, D.; Medrano, N.; Calvo, B.; Celma, S. Development of a Wireless Sensor Network System for Early Forest Fire
Detection. In Proceedings of the ITG-Fachbericht 224-RFID Systech, Ciudad, Spain, 15–16 June 2010.
Owayjan, M.; Freiha, G.; Achkar, R.; Abdo, E.; Mallah, S. Firoxio: Forest Fire Detection and Alerting System. In Proceedings of the
17th IEEE Mediterranean Electrotechnical Conference, Beirut, Lebanon, 13–16 April 2014.
Molina-Pico, A.; Cuesta-Frau, D.; Araujo, A.; Alejandre, J.; Rozas, A. Forest Monitoring and Wildland Early Fire Detection by a
Heirarchical Wireless Sensor Network. J. Sens. 2016,2016, 8325845. [CrossRef]
Lutakamale, A.S.; Kaijage, S. Wildfire Monitoring and Detection System Using Wireless Sensor Network: A Case Study of
Tanzania. Wirel. Sens. Netw. 2017,9, 274–289. [CrossRef]
44. Teixeira, J. Wireless Sensor Network for Forest Fire Detection; Universidade Do Porto: Porto, Portugal, 2017.
Neumann, G.B.; de Almeida, V.P.; Endler, M. Smart Forests: Fire Detection Service. In Proceedings of the IEEE Symposium on
Computers and Communications, Natal, Brazil, 25–28 June 2018.
Kadir, E.A.; Rosa, S.L.; Yulianti, A. Application of WSNs for Detection Land and Forest Fire in Riau Province Indonesia. In Pro-
ceedings of the International Conference on Electrical Engineering and Computer Science, Pangkal, Indonesia,
2–4 October 2018
LADSensors—Firest Supervisor Early Wildfire Detection System. 2018. Available online: (accessed
on 23 July 2022).
48. Silvanet Wildfire Detection. 2019. Available online: (accessed on 28 July 2022).
49. Knotifire. 2020. Available online: (accessed on 28 July 2022).
BurnMonitor: An Early Wildfire Detection IoT Solution, Inria. 2020. Available online:
/05/04/burnmonitor-an-early-wildfire-detection-iot-solution/ (accessed on 28 July 2022).
Benzekri, W.; Moussati, A.E.; Moussaoui, O.; Berrajaa, M. Early Forest Fire Detection System using Wireless Sensor Network and
Deep Learning. Int. J. Adv. Comput. Sci. Appl. 2020,11, 5. [CrossRef]
ChemNode—N5 Sensors. 2022. Available online:
content/uploads/2022/02/N5-ChemNode-Data-Sheet.pdf (accessed on 28 July 2022).
National Wildfire Coordinating Group. NWCG Standards for Fire Unmanned Aircraft Systems Operations; National Wildfire
Coordinating Group: Washington, DC, USA, 2019.
Zhao, X.; Zhou, Z.; Zhu, X.; Guo, A. Design of a Hand-Launched Solar-Powered Unmanned Aerial Vehicle (UAV) System for
Plateau. Appl. Sci. 2020,10, 1300. [CrossRef]
Allison, R.S.; Johnston, J.M.; Craig, G.; Jennings, S. Airborne Optical and Thermal Remote Sensing for Wildfire Detection and
Monitoring. Sensors 2016,16, 1310.
FOTOKITE, The Fotokite Sigma. A Situational Awareness System for First Responders, Perspective Robotics AG. 2022. Available
online: (accessed on 9 July 2022).
Sustainability 2022,14, 12270 20 of 21
Yuan, C.; Liu, Z.; Zhang, Y. UAV-based forest fire detection and tracking using image processing techniques. In Proceedings of
the 2015 International Conference on Unmanned Aircraft Systems (ICUAS), Denver, CO, USA, 9–12 June 2015; p. 640.
Ball, M. Forest Fire Monitoring with Drones and Wind Sensors, Unmanned Systems Technology, 25 January 2021. Available online: (accessed on
9 July 2022).
Drone Amplified, Fight Fire with Fire, Drone Amplified. Available online:
(accessed on 9 July 2022).
Lin, Z.; Liu, H.H.T.; Wotton, M. Kalman Filter-Based Large-Scale Wildfire Monitoring With a System of UAVs. IEEE Trans.
Ind. Electron. 2019,66, 606–615. [CrossRef]
Bushnaq, O.M.; Chaaban, A.; Al-Naffouri, T.Y. The Role of UAV-IoT Networks in Future Wildfire Detection. IEEE Internet Things J.
2021,8, 16984–16999. [CrossRef]
Lewicki, T.; Liu, K. Multimodal Wildfire Surveillance with UAV. In Proceedings of the IEEE Global Communications Conference,
Madrid, Spain, 7–11 December 2021.
Jiao, Z.; Zhang, Y.; Mu, L.; Xin, J.; Jiao, S.; Liu, H.; Liu, D. A YOLOv3-based Learning Strategy for Real-time UAV-based Forest
Fire Detection. In Proceedings of the IEEE Chinese Control and Decision Conference, Changsha, China, 22–24 August 2020.
Tahir, H.U.A.; Waqar, A.; Khalid, S.; Usman, S.M. Wildfire Detection in Aerial Images Using Deep Learning. In Proceedings of the
2nd International Conference on Digital Futures and Transformative Technlogies, Rawalpindi, Pakistan, 24–26 May 2022.
Bouguettaya, A.; Zarzour, H.; Taberkit, A.M.; Kechida, A. A review on early wildfire detection from unmanned aerial vehicles
using deep learning-based computer vision algorithms. Signal Processing 2020,190, 108309. [CrossRef]
Barmputis, P.; Papaioanou, P.; Dimitropoulos, K.; Grammalidis, N. A Review on Early Forest Fire Detection Systems Using
Optical Remote Sensing. Sensors 2020,20, 6442. [CrossRef]
Han, B.; Zhou, Y.; Deveerasetty, K.K.; Hu, C. A Review of Control Algorithms for Quadrotor. In Proceedings of the International
Conference on Information and Automation, Wuyishan, China, 11–13 August 2018.
Pham, H.X.; La, H.M.; Feil-Seifer, D.; Deans, M.C. A Distributed Control Framework of Multiple Unmanned Aerial Vehicles for
Dynamic Wildfire Tracking. IEEE Trans. Syst. Man Cybern. 2020,50, 1537–1548. [CrossRef]
69. Mawanza, J.T.; Agee, J.T.; Bhero, E. Adaptive Finite-Time Time-Varying Elliptical Formation Control for a Group of Quadrotors
UAVs for Cooperative Wildfire Monitoring. In Proceedings of the 30th Southern African Universities Power Engineering, Durban,
South Africa, 25–27 January 2022.
Shrestha, K.; Dubey, R.; Singandhupe, A.; Louis, S.; La, H. Multi Objective UAV Network Deployment for Dynamic Fire Coverage.
In Proceedings of the IEEE Congress on Evolutionary Computation, Kraków, Poland, 28 June–1 July 2021.
Kim, J.; Gadsden, A.A.; Wilkerson, S.A. A Comprehensive Survey of Control Strategies for Autonomous Quadrotors. Can. J.
Electr. Comput. Eng. 2020,43, 3–16. [CrossRef]
Kangunde, V.; Jamisola, R.S., Jr.; Theophilus, E.K. A review on drones controlled in real-time. Int. J. Dyn. Control.
1832–1846. [CrossRef]
Oettershagen, P.; Melzer, A.; Mantel, T.A.; Rudin, K.; Lotz, R.; Siebenmann, D.; Leutenegger, S.; Alexis, K.; Siegwart, R. A
solar-powered hand-launchable UAV for low-altitude multi-day continuous flight. In Proceedings of the IEEE International
Conference on Robotics and Automation, Seattle, WA, USA, 26 May 2015; pp. 3986–3993.
Jung, S.; Jo, Y.; Kim, Y. Aerial Surveillance with Low-Altitude Long-Endurance Tethered Multirotor UAVs Using Photovoltaic
Power Management System. Energies 2019,12, 1323. [CrossRef]
SolarXOne: Autonomous, Long Range and Solar Drone, XSun. 2019. Available online:
(accessed on 22 July 2022).
Shi, J.; Wang, W.; Gao, Y.; Yu, N. Optimal Placement and Intelligent Smoke Detection Algorithm for Wildfire-Monitoring Cameras; IEEE:
Piscataway, NJ, USA, 2020; Volume 8, pp. 3–4.
SmokeD, Automatic Fire and Smoke Detection System, SmokeD. Available online: (accessed on
10 July 2022).
78. Alert Wildfire. 2014. Available online: (accessed on 27 July 2022).
News Releases, Pacific Gas and Electric, 18 November 2021. Available online:
newsroom/ (accessed on 27 July 2022).
80. Pano AI. 2019. Available online: (accessed on 27 July 2022).
IQ FireWatch, IQ FireWatch Technology, IQ Firewatch. Available online:
(accessed on 10 July 2022).
Planet, Linking Ground and Space Systems to Autonomously Assess Wildfires, Planet, 25 August 2020. Available online: https:
// (accessed on
10 July 2022)
Stipaniˇcev, D.; Šeri´c, L.; Braovi´c, M.; Krstini´c, D.; Jakovˇcevi´c, T.; Štula, M.; Bugari´c, M.; Maras, J. Vision Based Wildfire and Natural
Risk Observers; IEEE: Piscataway, NJ, USA, 2012; p. 6.
84. He, K. Single Image Haze Removal Using Dark Channel Prior; IEEE: Piscataway, NJ, USA, 2009.
85. Agirman, K.T.A.K. BLSTM based night-time wildfire detection from video. PLoS ONE 2022,17, e0269161. [CrossRef]
Gutro, R. NASA Covers Wildfires Using Many Sources, NASA Goddard Space Flight Center, 9 December 2021. Available online: (accessed on 10 July 2022).
Sustainability 2022,14, 12270 21 of 21
Stevens, J. Smoke from Camp Fire Billows Across California, 14 November 2018. Available online: https://earthobservatory.nasa.
gov/images/144252/smoke-from-camp-fire-billows-across-california (accessed on 13 July 2022).
Matson, M.; Dozier, J. Identification of Subresolution High Temperature Sources Using a Thermal IR Sensor. Photogramm. Eng.
Remote Sens. 1981,47, 1311–1318.
89. Giglio, L.; Descloitres, J.; Justice, C.O.; Kaufman, Y. An Enhanced Contextual Fire Detection Algorithm for MODIS. Remote Sens.
Environ. 2003,87, 273–282.
Giglio, L.; Schroeder, W.; Justice, C.O. The collection 6 MODIS active fire detection algorithm and fire products.
Remote Sens. Environ.
2016,1789, 31–41. [CrossRef]
Pereira, G.H.A.; Fusioka, A.M.; Nassu, B.T.; Minetto, R. Active Fire Detection in Landsat-8 Imagery: A Large-Scale Dataset and a
Deep-Learning Study. J. Photogramm. Remote Sens. 2021,178, 171–186. [CrossRef]
Schroeder, W.; Oliva, P.; Giglio, L.; Quayle, B.; Lorenz, E.; Morelli, F. Active fire detection using Landsat-8/OLI data.
Remote Sens. Environ. 2016,185, 210–220. [CrossRef]
Rostami, A.; Shah-Hosseini, R.; Asgari, S.; Zarei, A.; Adhdami-Nia, M.; Homayouni, S. Acrive Fire Detection from Landsat-8
Imagery Using Multiple Kernel Learning. Remote Sens. 2022,14, 992. [CrossRef]
Johnstone, A. CubeSat Design Specification (1U-12U) rev 14.1, February 2022. Available online:
static/5418c831e4b0fa4ecac1bacd/t/5f24997b6deea10cc52bb016/1596234122437/CDS+REV14+2020-07-31+DRAFT.pdf (accessed
on 24 July 2022).
Gangestad, J.W.; Rowen, D.W.; Hardy, B.S. Forest fires, sunglint, and a solar eclipse: Responsive remote sensing with AeroCube-4.
In Proceedings of the IEEE Geoscience and Remote Sensing Symposium, Kuala Lumpur, Malaysia, 17–22 July 2014; pp. 3622–3625.
Azami, M.H.; Orger, N.C.; Schulz, V.H.; Cho, M. Demonstration of Wildfire Detection Using Image Classification Onboard
CubeSat. In Proceedings of the International Symposium on Geoscience and Remote Sensing, Brussels, Belgium, 11–16 July 2021.
Azami, M.H.; Orger, N.C.; Schulz, V.H.; Oshiro, T.; Cho, M. Earth Observation Mission of a 6U CubeSat with a 5-meter Resolution
for Wildfire Image Classification Using Convolutional Neural Network Approach. Remote Sens. 2022,14, 1874. [CrossRef]
98. OroraTech Home Page. 2018. Available online: (accessed on 24 July 2022).
99. Interorbital Systems. 2022. Available online: (accessed on 24 July 2022).
AVIRIS Airborne Visible/Infrared Imaging Spectrometer. Available online: (accessed on
28 July 2022
101. Advanced Very High Resolution Radiometer. Available online: (accessed on 28 July 2022).
102. NASA. ASTER|Terra. Available online: (accessed on 28 July 2022).
103. NASA. MODIS Web. Available online: (accessed on 28 July 2022).
104. NASA. Multi-Angle Imaging SpectroRadiometer. Available online:
spectroradiometer-misr (accessed on 28 July 2022).
105. NASA. MOPITT|Terra. Available online: (accessed on 28 July 2022).
106. NASA. AIRS. Available online: (accessed on 28 July 2022).
NASA. CALIPSO. Available online: (accessed on
28 June 2022).
NOAA. Visible Infrared Imaging Radiometer Suite (VIIRS). Available online:
missions/currently-flying/joint-polar-satellite-system/jpss-mission-and-2 (accessed on 28 June 2022).
NASA. Welcome to Hyperspectral Thermal Emission Spectrometer Website. Available online:
(accessed on 28 June 2022).
NASA. Landsat 8|Landsat Science. Available online: (accessed on
28 June 2022).
111. NASA. Welcome to ECOSTRESS. Available online: (accessed on 28 June 2022).
... They have become an increasingly popular tool for detecting wildfires due to their ability to cover large areas quickly and provide real-time data to ground crews [37,46]. Despite their advantages, UAVs are unsuitable for continuous, uninterrupted monitoring in wildfire-prone areas owing to their limited flight times [31]. Geostationary satellites address the lack of continuous availability of UAVs, as they can monitor a specific area continuously for fires [31]. ...
... Despite their advantages, UAVs are unsuitable for continuous, uninterrupted monitoring in wildfire-prone areas owing to their limited flight times [31]. Geostationary satellites address the lack of continuous availability of UAVs, as they can monitor a specific area continuously for fires [31]. With more accessible access to satellites through projects such as the CubeSat [6,20], recent research has focused on wildfire detection algorithms using the low-resolution images taken by the satellites. ...
... Despite these efforts, the limited spatial and temporal nature of the data makes timely detection challenging. For example, detecting young fires smaller than a pixel is extremely difficult [31]. In addition, dense clouds, smoke, and other atmospheric effects may lead to false alarms or missed detections. ...
Full-text available
Continuous monitoring of areas nearby the electric grid is critical for preventing and early detection of devastating wildfires. Existing wildfire monitoring systems are intermittent and oblivious to local ambient risk factors, resulting in poor wildfire awareness. Ambient sensor suites deployed near the gridlines can increase the monitoring granularity and detection accuracy. However, these sensors must address two challenging and competing objectives at the same time. First, they must remain powered for years without manual maintenance due to their remote locations. Second, they must provide and transmit reliable information if and when a wildfire starts. The first objective requires aggressive energy savings and ambient energy harvesting, while the second requires continuous operation of a range of sensors. To the best of our knowledge, this paper presents the first self-sustained cyber-physical system that dynamically co-optimizes the wildfire detection accuracy and active time of sensors. The proposed approach employs reinforcement learning to train a policy that controls the sensor operations as a function of the environment (i.e., current sensor readings), harvested energy, and battery level. The proposed cyber-physical system is evaluated extensively using real-life temperature, wind, and solar energy harvesting datasets and an open-source wildfire simulator. In long-term (5 years) evaluations, the proposed framework achieves 89% uptime, which is 46% higher than a carefully tuned heuristic approach. At the same time, it averages a 2-minute initial response time, which is at least 2.5× faster than the same heuristic approach. Furthermore, the policy network consumes 0.6 mJ per day on the TI CC2652R microcontroller using TensorFlow Lite for Micro, which is negligible compared to the daily sensor suite energy consumption.
... Recently, due to climate change, natural disasters such as fires, earthquakes, and floods have occurred frequently, and their intensity is gradually increasing. These natural disasters have caused substantial economic losses and threatened human life, animals, and plants [1]. Early detection of forest fires or floods for proper emergency response is essential to reduce the damage from natural disasters. ...
Full-text available
In recent years, there has been a significant increase in the demand for unmanned aerial vehicle (UAV)-based monitoring systems to ensure proper emergency response during natural disasters such as wildfires, hurricanes, floods, and earthquakes. This paper proposes a real-time UAV monitoring system for responding to forest fires or floods. The proposed system consists of a hardware part and a software part. The hardware configuration is an embedded camera board mounted on the UAV, a Qualcomm QCS610 SoC with cores suitable for running deep learning-based algorithms. The software configuration is a deep learning-based semantic segmentation model for detecting fires or floods. To execute the model in real time on edge devices with limited resources, we used a network slimming technique which generates a lightweight model with reduced model size, number of parameters, and computational complexity. The performance of the proposed system was evaluated on the FLAME dataset consisting of forest fire images and the FloodNet dataset consisting of flood images. The experimental results showed that the mIoU of slimmed DeepLabV3+ for FLAME is 88.29%, and the inference speed is 10.92 fps. For FloodNet, the mIoU of the slimmed DeepLabV3+ is 94.15%, and the inference speed is 13.26 fps. These experimental results confirm that the proposed system is appropriate for accurate, low-power, real-time monitoring of forest fires and floods using UAVs.
... This review focused on sensor nodes because they are the least expensive to implement and design compared to other methods for wildfire detection [12]. Sensor nodes are placed throughout a wildfire risk area to form a wireless sensor network (WSN). ...
Full-text available
Wildfires have played an increasing role in wreaking havoc on communities, livelihoods, and ecosystems globally, often starting in remote regions and rapidly spreading into inhabited areas where they become difficult to suppress due to their size and unpredictability. In sparsely populated remote regions where freshly ignited fires can propagate unimpeded, the need for distributed fire detection capabilities has become increasingly urgent. In this work, we evaluate the potential of a multitude of different sensing modalities for integration into a distributed downwind fire detection system, something which does not exist today. We deployed custom sensor-rich data logging units over a multi-day-controlled burn event hosted by the Marin County Fire Department in Marin County, CA. Under the experimental conditions, nearly all sensing modalities exhibited signature behaviors of a nearby active fire, but with varying degrees of sensitivity. We present promising preliminary findings from these field tests but also note that future work is needed to assess more prosaic concerns. Larger scale trials will be needed to determine the practicality of specific sensing modalities in outdoor settings, and additional environmental data and testing will be needed to determine the sensor system lifetime, data delivery performance, and other technical considerations. Crucially, this work provides the preliminary justification underscoring that future work is potentially valuable and worth pursuit.
... The early wildfire detection technologies were described in various works [28]. Several studies focused on four categories: sensor nodes, UAVs, camera networks, and satellite surveillance. ...
Full-text available
Open burning is the main factor contributing to the occurrence of wildfires in Thailand, which every year result in forest fires and air pollution. Open burning has become the natural disaster that threatens wildlands and forest resources the most. Traditional firefighting systems, which are based on ground crew inspection, have several limits and dangerous risks. Aerial imagery technologies have become one of the most important tools to prevent wildfires, especially drone real-time monitoring for wildfire surveillance. This paper presents an accuracy assessment of drone real-time open burning imagery detection (Dr-TOBID) to detect smoke and burning as a framework for a deep learning-based object detection method using a combination of the YOLOv5 detector and a lightweight version of the long short-term memory (LSTM) classifier. The Dr-TOBID framework was designed using OpenCV, YOLOv5, TensorFlow, LebelImg, and Pycharm and wirelessly connected via live stream on open broadcaster software (OBS). The datasets were separated by 80% for training and 20% for testing. The resulting assessment considered the conditions of the drone’s altitudes, ranges, and red-green-black (RGB) mode in daytime and nighttime. The accuracy, precision, recall, and F1-Score are shown for the evaluation metrics. The quantitative results show that the accuracy of Dr-TOBID successfully detected open burning monitoring, smoke, and burning characteristics, where the average F1-score was 80.6% for smoke detection in the daytime, 82.5% for burning detection in the daytime, 77.9% for smoke detection at nighttime, and 81.9% for burning detection at nighttime.
... The objective of the review article [7] is to strike a balance between the positive and negative effects of wildfires on the environment and the economy. Another review paper [8] discusses trends and advancements in four categories of wildfire detection technologies, including unmanned aerial vehicles, satellite surveillance, sensor nodes, camera networks, etc., along with their respective drawbacks and difficulties. The authors emphasize the impact of wildfires on the world's ecosystems. ...
... Unfortunately, these primary techniques are inefficient due to the unreliability of human observation towers and difficult living conditions. Advanced technologies such as satellites, UAV, UGV, ground-based sensor nodes, and camera systems supplement traditional firefighting techniques, focusing on detecting wildfires at early stages, and predicting hot spots by combining all the methods for a robust fire monitoring system [22]. Cyber-Physical Systems (CPSs), which integrate computing and control technologies into physical systems aim to contribute to real-time wildfire detection through the Internet of Things (IoT) wireless sensor [23] and the use of a collaborative unmanned system [11]. ...
Full-text available
The increasing frequency and severity of forest fires necessitate early detection and rapid response to mitigate their impact. This project aims to design a cyber-physical system for early detection and rapid response to forest fires using advanced technologies. The system incorporates Internet of Things sensors and autonomous unmanned aerial and ground vehicles controlled by the robot operating system. An IoT-based wildfire detection node continuously monitors environmental conditions, enabling early fire detection. Upon fire detection, a UAV autonomously surveys the area to precisely locate the fire and can deploy an extinguishing payload or provide data for decision-making. The UAV communicates the fire’s precise location to a collaborative UGV, which autonomously reaches the designated area to support ground-based firefighters. The CPS includes a ground control station with web-based dashboards for real-time monitoring of system parameters and telemetry data from UAVs and UGVs. The article demonstrates the real-time fire detection capabilities of the proposed system using simulated forest fire scenarios. The objective is to provide a practical approach using open-source technologies for early detection and extinguishing of forest fires, with potential applications in various industries, surveillance, and precision agriculture
... They also presented popularly used datasets such as CorsicanFire, FLAME, DeepFire, and the FD-dataset, as well as major challenges related to these techniques, such as data collection and labeling. Mohapatra and Trinh [22] provided a review of the recent trend and advancements in technologies (ground sensors, cameras, drones, and satellites) proposed for wildfire monitoring and firefighting tasks. Akhloufi et al. [23] reviewed the development of unmanned aerial vehicles (UAVs) for wildfires, highlighting wildland fire datasets, fire detection, segmentation, geolocation, and modeling methods, as well as cooperative autonomous systems for wildland fires. ...
Full-text available
Wildland fires are one of the most dangerous natural risks, causing significant economic damage and loss of lives worldwide. Every year, millions of hectares are lost, and experts warn that the frequency and severity of wildfires will increase in the coming years due to climate change. To mitigate these hazards, numerous deep learning models were developed to detect and map wildland fires, estimate their severity, and predict their spread. In this paper, we provide a comprehensive review of recent deep learning techniques for detecting, mapping, and predicting wildland fires using satellite remote sensing data. We begin by introducing remote sensing satellite systems and their use in wildfire monitoring. Next, we review the deep learning methods employed for these tasks, including fire detection and mapping, severity estimation, and spread prediction. We further present the popular datasets used in these studies. Finally, we address the challenges faced by these models to accurately predict wildfire behaviors, and suggest future directions for developing reliable and robust wildland fire models.
... Weather is one obvious factor controlling fire activity and has been the main focus of recent discussions of future fire regimes, pointing to the effect of climate change [1][2][3]. However, human factors such as fuel management, fire prevention, detection and suppression capability are also critical [4][5][6][7]. Notably, anthropogenic ignitions dominate heavily over lightning ignitions in most regions [8,9]. At regional levels, fire occurrence therefore tends to correlate positively with anthropogenic parameters such as population- [10][11][12] and road density [13][14][15]. ...
Full-text available
Organization of successful wildfire prevention and suppression require detailed information on ignition causes, size distributions and relations to weather. From a large and highly detailed dataset of Swedish wildfire incidents (n = 124 000) we assess temporal, geographical and human-related patterns over a 25 year-period (1996–2020). We find strong positive correlations between population density and wildfire occurrence, primarily caused by a wide spectrum of human activities. However, fires >10 ha mostly occurred in sparsely populated regions and were more often ignited by lightning or heavy machinery. Further, large fires had a history of long response times and insufficient mop-up, in turn intimately linked to low population density. We detect no trend over the period 25-year-period in either fire weather, number of ignitions or burned area, but a dramatic decline in wildfire caused by children's play as well as by springtime burning of dead grass, a traditional fire use in rural areas. Our results indicate that irrespective of climate change, societal changes such as rural depopulation and cultural shifts are imminently important for the future fire regime in this intensely managed part of the boreal, and may warrant more attention worldwide.
Full-text available
Distinguishing fire from non-fire objects in night videos is problematic if only spatial features are to be used. Those features are highly disrupted under low-lit environments because of several factors, such as the dynamic range limitations of the cameras. This makes the analysis of temporal behavior of night-time fire indispensable for classification. To this end, a BLSTM based night-time wildfire event detection from a video algorithm is proposed. It is shown in the experiments that the proposed algorithm attains 95.15% of accuracy when tested against a wide variety of actual recordings of night-time wildfire incidents and 23.7 ms per frame detection time. Moreover, to pave the way for more targeted solutions to this challenging problem, experiment-based thorough investigations of possible sources of incorrect predictions and discussion of the unique nature of night-time wildfire videos are presented in the paper.
Full-text available
The KITSUNE satellite is a 6-unit CubeSat platform with the main mission of 5-m-class Earth observation in low Earth orbit (LEO), and the payload is developed with a 31.4 MP commercial off-the-shelf sensor, customized optics, and a camera controller board. Even though the payload is designed for Earth observation and to capture man-made patterns on the ground as the main mission, a secondary mission is planned for the classification of wildfire images by the convolution neural network (CNN) approach. Therefore, KITSUNE will be the first CubeSat to employ CNN to classify wildfire images in LEO. In this study, a deep-learning approach is utilized onboard the satellite in order to reduce the downlink data by pre-processing instead of the traditional method of performing the image processing at the ground station. The pre-trained CNN models generated in Colab are saved in RPi CM3+, in which, an uplink command will execute the image classification algorithm and append the results on the captured image data. The on-ground testing indicated that it could achieve an overall accuracy of 98% and an F1 score of a 97% success rate in classifying the wildfire events running on the satellite system using the MiniVGGNet network. Meanwhile, the LeNet and ShallowNet models were also compared and implemented on the CubeSat with 95% and 92% F1 scores, respectively. Overall, this study demonstrated the capability of small satellites to perform CNN onboard in orbit. Finally, the KITSUNE satellite is deployed from ISS on March 2022.
Full-text available
Active fires are devastating natural disasters that cause socio-economical damage across the globe. The detection and mapping of these disasters require efficient tools, scientific methods, and reliable observations. Satellite images have been widely used for active fire detection (AFD) during the past years due to their nearly global coverage. However, accurate AFD and mapping in satellite imagery is still a challenging task in the remote sensing community, which mainly uses traditional methods. Deep learning (DL) methods have recently yielded outstanding results in remote sensing applications. Nevertheless, less attention has been given to them for AFD in satellite imagery. This study presented a deep convolutional neural network (CNN) “MultiScale-Net” for AFD in Landsat-8 datasets at the pixel level. The proposed network had two main characteristics: (1) several convolution kernels with multiple sizes, and (2) dilated convolution layers (DCLs) with various dilation rates. Moreover, this paper suggested an innovative Active Fire Index (AFI) for AFD. AFI was added to the network inputs consisting of the SWIR2, SWIR1, and Blue bands to improve the performance of the MultiScale-Net. In an ablation analysis, three different scenarios were designed for multi-size kernels, dilation rates, and input variables individually, resulting in 27 distinct models. The quantitative results indicated that the model with AFI-SWIR2 SWIR1-Blue as the input variables, using multiple kernels of sizes 3 × 3, 5 × 5, and 7 × 7 simultaneously, and a dilation rate of 2, achieved the highest F1-score and IoU of 91.62% and 84.54%, respectively. Stacking AFI with the three Landsat-8 bands led to fewer false negative (FN) pixels. Furthermore, our qualitative assessment revealed that these models could detect single fire pixels detached from the large fire zones by taking advantage of multi-size kernels. Overall, the MultiScale-Net met expectations in detecting fires of varying sizes and shapes over challenging test samples.
Full-text available
Forest fires have become a major threat around the world, causing many negative impacts on human habitats and forest ecosystems. Climatic changes and the greenhouse effect are some of the consequences of such destruction. Interestingly, a higher percentage of forest fires occur due to human activities. Therefore, to minimize the destruction caused by forest fires, there is a need to detect forest fires at their initial stage. This paper proposes a system and methodology that can be used to detect forest fires at the initial stage using a wireless sensor network. Furthermore, to acquire more accurate fire detection, a machine learning regression model is proposed. Because of the primary power supply provided by rechargeable batteries with a secondary solar power supply, a solution is readily implementable as a standalone system for prolonged periods. Moreover, in-depth attention is given to sensor node design and node placement requirements in harsh forest environments and to minimize the damage and harmful effects caused by wild animals, weather conditions, etc. to the system. Numerous trials conducted in real tropical forest sites found that the proposed system is effective in alerting forest fires with lower latency than the existing systems.
Full-text available
Wildland fires release substantial amounts of hazardous contaminants, contributing to a decline in air quality and leading to serious health risks. Thus, this study aimed to understand the contributions of the 2017 extreme wildland fires in Portugal on children health, compared to 2016 (with burned area, in accordance with the average of the previous 15 years). The impact of long-term exposure to PM10 and NO2 concentrations, associated with wildland fires, on postneonatal mortality, bronchitis prevalence, and bronchitis symptoms in asthmatic children was estimated, as well as the associated costs. The excess health burden in children attributable to exposure to PM10 and NO2, was calculated based on WHO HRAPIE relative risks. Fire emissions were obtained from the Fire INventory from NCAR (FINN). The results obtained indicate that the smoke from wildfires negatively impacts children’s lung function (PM10 exposure: increase of 320 and 648 cases of bronchitis in 2016 and 2017; NO2 exposure: 24 and 40 cases of bronchitis symptoms in asthmatic children in 2016 and 2017) and postneonatal mortality (PM10 exposure: 0.2 and 0.4 deaths in 2016 and 2017). Associated costs were increased in 2017 by around 1 million € for all the evaluated health endpoints, compared to 2016.
Full-text available
Climate change and human activities have drastically altered the natural wildfire balance in the Western US and increased population health risks due to exposure to pollutants from fire smoke. Using dynamically downscaled climate model projections, we estimated additional asthma emergency room visits and hospitalizations due to exposure to smoke fine particulate matter (PM2.5) in the Western US in the 2050s. Isolating the amount of PM2.5 from wildfire smoke is both difficult to estimate and, thus, utilized by relatively few studies. In this study, we use a sophisticated modeling approach to estimate future increase in wildfire smoke exposure over the reference period (2003-2010) and subsequent health care burden due to asthma exacerbation. Average increases in smoke PM2.5 during future fire season ranged from 0.05-9.5 µg/m3 with the highest increases seen in Idaho, Montana, and Oregon. Using the Integrated Climate and Land-Use Scenarios (ICLUS) A2 scenario, we estimated the smoke-related asthma events could increase at a rate of 15.1 visits per 10,000 persons in the Western US, with the highest rates of increased asthma (25.7-41.9 per 10,000) in Idaho, Montana, Oregon, and Washington. Finally, we estimated healthcare costs of smoke-induced asthma exacerbation to be over $1.5 billion during a single future fire season. Here we show the potential future health impact of climate-induced wildfire activity, which may serve as a key tool in future climate change mitigation and adaptation planning.
Over the past several decades, there has been a constant increase in the use of unmanned aircraft systems (UASs). Hence, there has also been a growth in the number of control algorithms to service many applications embodied by these vehicles. Initially, UASs were made popular by the military for reconnaissance, intelligence, surveillance, and target acquisition (RISTA) applications. Nowadays, UASs are used for everything from crop surveys to tourism. Nowhere is this more evident than with a multi-rotor unmanned aerial vehicle (UAV). This paper presents a survey of control methods for multi-rotor systems, namely quadrotors. In doing so, we hope to visualize a clear path to what additional capabilities might be needed in the future. In our examination, we review many of the notable research organizations and their efforts to expand the utility of multi-rotor aircraft. We also summarize the basic literature definitions and control strategies for autonomous quadrotors.