AIRWISE

An Airborne Wireless Sensor Network for Ambient Air Pollution Monitoring

Orestis Evangelatos and Jos

e D. P. Rolim

Department of Computer Science, University of Geneva, Geneva, Switzerland

Keywords:

Airborne Systems, Wireless Sensor Networks, Air Quality, Pollution Monitoring.

Abstract:

Over the last decades with the rapid growth of industrial zones, manufacturing plants and the substantial

urbanization, environmental pollution has become a crucial health, environmental and safety concern. In par-

ticular, due to the increased emissions of various pollutants caused mainly by human sources, the air pollution

problem is elevated in such extent where signiﬁcant measures need to be taken. Towards the identiﬁcation

and the qualiﬁcation of that problem, we present in this paper an airborne wireless sensor network system for

automated monitoring and measuring of the ambient air pollution. Our proposed system is comprised of a

pollution-aware wireless sensor network and unmanned aerial vehicles (UAVs). It is designed for monitoring

the pollutants and gases of the ambient air in three-dimensional spaces without the human intervention. In

regards to the general architecture of our system, we came up with two schemes and algorithms for an au-

tonomous monitoring of a three-dimensional area of interest. To demonstrate our solution, we deployed the

system and we conducted experiments in a real environment measuring air pollutants such as: NH

, CH

, O

along with the temperature, relative humidity and atmospheric pressure. Lastly, we experimentally

evaluated and analyzed the two proposed schemes.

1 INTRODUCTION

The atmospheric composition has been continuously

changing over the past thousands of years but it is

just after the industrial revolution of the 18th cen-

tury when the atmosphere started to be signiﬁcantly

effected. The huge growth of urbanization and the

massive construction of polluting factories and indus-

trial cities, coupled with the lack of legislation and

standards for the atmospheric pollutants, led to a pro-

gressively increase of the concentrations of danger-

ous gases in the air. As the atmosphere is essential

to support life on our planet, air pollution has long

been recognized as a serious threat to human health

and to the whole ecosystem. In that context, over the

last few decades, governments and NGO’s have set

rules in the emissions of harmful substances in the

atmosphere. Since the early 1970s the EU Air Qual-

ity Directive (EUA, ) and the U.S National Ambient

Air Quality Standards (NAAQS) (Chow et al., 2007)

have been working on improving the air quality by

controlling those emissions and deﬁne maximum at-

mospheric concentrations.

Due to the hazardous effects of the air pollution

to the people and to the environment, air quality eval-

uation is playing an important role in the assessment

of the limits in the exposure of the population and the

minimization of health impacts. Human exposure to

air pollutants may have serious health effects depend-

ing on several factors such as: duration, magnitude

and frequency of the exposure. People in their ev-

ery day life come in contact with various pollutants in

the air both indoors and outdoors. As a matter of fact,

air quality monitoring is crucial not only for assessing

the exposure of the population to the air pollution but

it can also be proven extremely useful for scientists in

improving the pollution prediction models. In addi-

tion it can be used to provide emergency information

in the cases of unpredictable disasters. Taking into ac-

count the importance of the air pollution monitoring,

it is very challenging to monitor how the ambient pol-

lutants are dispersed and diluted in the air both hori-

zontally and vertically. In particular is at high interest

a ﬁne grained monitoring in different spatial and tem-

poral distributions.

In relation to the ambient air quality monitoring,

several methods and techniques have been developed.

Traditionally, the monitoring is done with the use of

large monitoring stations placed in static locations

such as on top of towers and buildings. However, due

231

Evangelatos O. and D. P. Rolim J..

AIRWISE - An Airborne Wireless Sensor Network for Ambient Air Pollution Monitoring.

DOI: 10.5220/0005203302310239

In Proceedings of the 4th International Conference on Sensor Networks (SENSORNETS-2015), pages 231-239

ISBN: 978-989-758-086-4

 2015 SCITEPRESS (Science and Technology Publications, Lda.)

to their large size and cost of maintenance, these sta-

tions are deployed in relatively spatial areas and con-

sequently they can not act as mobile stations. One

of the main contributions of our work is the solution

towards this problem; the development of a mobile

monitoring system for the monitoring of the ambient

air pollution.

In this paper we present a WSN system for au-

tomated ambient air quality monitoring. Air qual-

ity sensors integrated with embedded devices enable

the measurement of the air pollution in a very ef-

ﬁcient and low-cost way. Our proposing system is

able to measure with the use of unmanned aerial ve-

hicles (UAVs) and WSNs and without the need of hu-

man intervention, the concentrations of several pol-

lutant, gases and environmental parameters, in three-

dimensional environments. We name our system:

AIRWISE.

The paper is organized as follows: in Section II

the related work and motivation is presented. In Sec-

tion III we propose the theoretical schemes and algo-

rithms as well as the implementation of the AIRWISE.

In Section IV we present the system development to-

gether with our experimental results and their evalu-

ation. Conclusions and future work are presented in

Section V.

2 RELATED WORK

The signiﬁcant advantages in distributed sensor net-

work systems including but not limited to reliability,

scalability, dynamicity and efﬁciency, have brought

the WSN systems into the next generation. WSN sys-

tems play an inevitable role in our everyday life and

they have been widely adopted in sensing and moni-

toring applications. In (Evangelatos et al., 2013) we

have proposed a framework with which we can sense,

monitor and control an environment by using WSNs.

Apart from the use of WSNs in the area of smart en-

vironments, lately they have been used also in the

context of air sensing and monitoring. Such a sys-

tem for example, is described in (Hu et al., 2011),

where sensors have been placed on top of cars form-

ing a vehicular WSN dedicated to measure the pol-

lutants’ concentrations. In addition, the authors in

(Yaacoub et al., 2013) have developed a monitoring

system for ground level air quality analysis in Qatar

using a WSN. A system using WSN devoted to the

monitoring of particular pollutants has been proposed

in (Wang et al., 2010), where carbon monoxide (CO)

sensors were used for the monitoring of the CO lev-

els in the premises of a university campus area. Other

similar systems that have been developed for air qual-

ity monitoring using WSN are proposed in (Chen

et al., 2013) where the authors have designed a WSN

node for remote monitoring of CO and in the (Kavi

K. Khedo, 2010) where it is proposed a simulation

system for air pollution monitoring using WSNs.

Previous work regarding the air quality and the as-

sessment of health impacts near the airports of UK

(Yim et al., 2013) showed that high amounts of pollu-

tants such as CO and NO

are emitted in the air dur-

ing the take off and the approach of a plane in an air-

port. Similar works such as the (Solazzo et al., 2013)

and (Lee et al., 2013) are presenting models and es-

timations on the concentrations and behaviour of the

pollutants in the air. In these regards we believe that

those models and estimations could be veriﬁed and

improved with the help of a WSN which would mea-

sure those pollutants in real environments. The au-

thors in (Barakeh et al., 2014) are proposing a frame-

work with which they can monitor in real time partic-

ulate matter evolution in construction sites in order to

assess the air quality, but although such a system can

provide a lot of important information on air quality,

it is static and bound to the ground.

Due to the recent advancements in robotics, aviation

and material sciences, the gap between airborne sys-

tems and WSNs has started to be shortened. Drones

are being used in a great variety of applications rang-

ing from supporting search and rescue operations

(Waharte and Trigoni, 2010) to aerial robotic con-

structions (Willmann et al., 2012). In addition, with

the technological advancements in 3D printing and

laser-cutting technologies, it is possible to manufac-

ture low-cost drones with individual features (Nickel

et al., 2014). The prior work of (Valente et al.,

2011) has used a quadrocopter-drone for implement-

ing a cropping monitoring system in the research

ﬁeld of precision agriculture using WSNs. In (Jude

et al., 2007) the authors have developed a WSN com-

posed of bird-sized micro aerial vehicles and ground

nodes in which they have analyzed networking perfor-

mances, such as RSSI behaviour and packet loss rates.

Experimental results on the integration of UAVs and

WSNs have been presented in (Teh et al., 2008).

Systems and deployments that have been pro-

posed so far are mainly investigating individually, or

in the most relevant works two out of the three follow-

ing domains: air quality monitoring, WSN and UAV.

To the best of our knowledge there has not been yet

proposed a system that combines WSNs, drones and

air pollution monitoring systems. Our work presents

a low-cost, automated pollution monitoring system

which is comprised of a wireless network with sensors

dedicated for measuring the concentration of air pol-

lutants and UAV for performing the measurements in

SENSORNETS2015-4thInternationalConferenceonSensorNetworks

232

different altitudes, latitudes and longitudes. We came

up with two schemes and algorithms resulting in a

system’s application that can monitor in ﬁne-grained

resolution and in near-real time, the ambient air qual-

ity in real three-dimensional spaces using WSNs. The

information acquired from the system regarding the

the pollutants’ concentrations in the ambient air could

be provided as proﬁtable resource data to air quality

scientists for improving their environmental models,

to governments as prerequisite information for index-

ing the air quality of their districts and last but not

least as inﬂuential dissemination information to the

people in order to uphold their environmental aware-

ness.

3 ARCHITECTURE OF AIRWISE

In our paper we present an Airborne WSN system

with which we can monitor the ambient air pollution

in three-dimensional real space environments. The

measurement of the pollutants in the air is being done

by pollution sensors which are placed on top of un-

manned aerial vehicles (drones). Drones have the

ability to ﬂy and hover in the air both manually and

automatically. The general design, algorithms and ar-

chitecture of the AIRWISE system is divided in the

following two categories: A. the theoretical models

and B. the implementation design. In the following

subsections we present, ﬁrstly the theoretical models

and afterwards the system’s implementation design.

3.1 Theoretical Models, Schemes and

Algorithms

In our work, in order to deal with the measurement

of the three-dimensional air space environment, we

propose the following general approach to facilitate

exposition: we divide the three-dimensional area we

want to investigate (denoted hereinafter as D) into

”small” equally tessellated cubic-subareas (named as

monitor-cubes). The three-dimensional area D with

its monitor-cubes is depicted in Figure 1. By dividing

the whole area of interest D, into these monitor-cubes,

we are able to distributively monitor the concerned

environment and extract individual pollution data for

each of them separately. This allows us to create sep-

arate ”heat” and history pollution maps for each dif-

ferent physical subareas as well as of the whole area

D. The size of each subarea (monitor-cube) can be

deﬁned by the user in accordance with the location

and the circumstances of the monitoring area. At the

same time, this tessellation gives us the possibility of

conducting both ﬁne-grained and macro-scaled mea-

surements. We designate that the measurements in

each monitor-cube regarding the pollutants, are taken

from their center. Our approach, deﬁnitions, schemes

and algorithms described below hold for both types of

measurements; ﬁne-grained and macro-scaled.

3.1.1 General Deﬁnitions

Prior to the description of our approach and models,

we need make the following general deﬁnitions:

Monitor-Cubes (Subareas S

(x,y,z)

). To facilitate the

exposition of our schemes and algorithms, we as-

sume without loss of generality, that the area D is cu-

bic. As described above, the three-dimensional area D

for monitoring the air quality, is tessellated into sev-

eral ”small” cubic subareas S, which we denote as:

(x,y,z)

, where x | x ∈ [0, k] (respectively y | y ∈ [0, l]

and z | z ∈ [0, m]) and k+1 (respectively y+1 and z+1)

is the number of division of the ﬁrst dimension (re-

spectively of the 2nd and the 3rd) of D. The area D

and its tessellation into the subareas S is depicted in

Figure 1.

Concentration of Pollutant ”a” (

−−→

CPa): There are

several pollutants existing in the air such as: Nitro-

gen Oxides (NO

), Carbon Oxides (CO

), Ammonia

(NH

) etc, and their concentrations vary depending

on a number of several parameters such as the: loca-

tion, altitude, temperature etc. We deﬁne the vector:

Concentration of Pollutant ”a” (

−−→

CPa) which repre-

sents the measured concentration of a pollutant ”a” in

the air. The value of this parameter is obtained by the

pollution sensor and its metric is usually in ppm (parts

per million). Subsequently, the

−−→

CPa is normalized

between [0,1] in respect to the minimum and maxi-

mum concentration values the pollution sensor is able

to measure.

Weight (

(x,y,z,i)

). For each subarea S

(x,y,z)

we deﬁne

a weight W

(x,y,z,i)

where i ∈ N. The weight W

(x,y,z,i)

represents the arithmetic mean of the measured con-

centration of the pollutant

−−→

CPa in a speciﬁc subarea

(x,y,z)

of the iteration (monitoring) cycle i. The term

iteration cycle represents one completed monitoring

of the whole area D and its value i represents the i-th

cycle.

Measuring Rate (MR). As MR we deﬁne the con-

stant value which represents the measuring rate with

which the pollution sensor is collecting pollution data

from its nearby environment. The MR can be deﬁned

as MR = Samples / Second.

Duration of Measurement (DM). As the pollutants

in the air some times could be burdensome to mea-

sure, long time measurements might be required to be

collected. Therefore, we deﬁne the value: Duration of

AIRWISE-AnAirborneWirelessSensorNetworkforAmbientAirPollutionMonitoring

233

Figure 1: Area D for monitoring and its tessellation to Sub-

areas S

(x,y,z)

Measurement (DM), to represent the duration of the

measuring process. Depending on the environmen-

tal variables of the speciﬁc time and location, short

time measurements might sufﬁce to collect trustwor-

thy data. However in situations such as toxic or harsh

environments, long time measurements might be re-

quired to obtain more accurate results.

3.1.2 Schemes

In this section we present two different schemes with

which we approach the problem of monitoring the

ambient air pollution in 3-D spaces. For each of them

we present as well their respective algorithms.

Sequential Monitoring Scheme In the Sequential

Monitoring Scheme, the routing of the drone and sub-

sequently the collection of the pollution data by the

sensors it carries on, are done in a sequential man-

ner. This means that the drone is routed in a static

and predeﬁned trajectory whereas the sensors are col-

lecting data systematically from the center of each

subarea S. The sensing process and hence the routing

pattern starts from the subarea S

(0,0,0)

and it covers

progressively all the subareas until it will arrive to the

subearea S

(x,y,z)

. At that point one iteration cycle (i)

will have been completed. Then the ﬂying and sens-

ing process will restart from the subarea S

(0,0,0)

Sequential Monitoring Algorithm (SMA). The

pseudo-code of the Sequential Monitoring Algorithm

(Algorithm 1) representing the sequential monitor-

ing scheme is presented below. For each iteration

cycle and for each subarea, the algorithm measures

the concentration of the pollutants and calculates their

Weight W .

Dynamic Monitoring Scheme. In order to use

more efﬁciently the limited and constrained resources

of the airborne systems and the WSNs, we propose

another monitoring scheme which acts in a dynamic

Algorithm 1: Sequential Monitoring Algorithm

(SMA).

Input: Values of: MR, DM, k,l,m

Output: The Weight W

(x,y,z,i)

MR ← default Measuring Rate

DM ← default Duration of Measurement

k, l, m ← size of each axis of area D

max i ← maximum iteration cycles

x, y, z, i ← 0

begin

while i < max i do

for z ← 0 to z = m do

for y ← 0 to y = l do

for x ← 0 to x = k do

−−→

CPa ← Take MR · DM

samples of pollutant a

(x,y,z,i)

← Arithmetic mean

−−→

CPa

x + + %next subarea of x

axis

y + + %next subarea of y axis

z + + %next subarea of z axis

i + + %next iteration cycle

x, y, z ← 0 %restart from S

(0,0,0)

return W

(x,y,z,i)

end

way. In this scheme the subareas are given a potential

of being monitored or not, depending on their previ-

ous weight values. We consider a subarea as stable

when its most recent weights W do not alter ”much”

during a speciﬁc time frame. In that case, we can

avoid visiting and consecutively avoid monitoring a

stable subarea. As a result we can use more efﬁcient

the limited energy of both the drone and the sensors

whilst increasing the time efﬁciency of the system as

well. Lower energy consumption could be translated

into monitoring of larger areas and for longer periods.

In order to better describe the dynamic monitor-

ing scheme, some further deﬁnitions in extension to

the general ones (mentioned for the sequential moni-

toring scheme), are needed to be made;

Minimum Iteration Cycles (min i). The parameter

min i is used to deﬁne the number of minimum iter-

ation cycles (monitoring cycles) for which the algo-

rithm will keep collecting data from all the subareas,

before it will enter into the dynamic mode.

Threshold (Thr). The Thr, threshold parameter is

an upper bound of the mean accumulated difference

between subsequent weights over a speciﬁc number

of consecutive iteration cycles. Subareas for which

their most recent weights are remaining ”almost” in-

variable, are deliberated as stable subareas. More-

over, the Thr describes the sensitivity of the algo-

rithm. With the term sensitivity we refer to the degree

of the pollution variation each subarea is allowed to

SENSORNETS2015-4thInternationalConferenceonSensorNetworks

234

sustain in order to be considered as stable. It is an

important parameter, as it allows the adjustment of

the tradeoff between the sensitivity of the monitoring

process versus the time and the energy needed to com-

plete an iteration cycle i.

Last iteration Cycles to Compare (LiC): The pa-

rameter LiC delineates the number of the most recent

iteration cycles whose W will be used in the compar-

ison with the threshold Thr.

Idle value (Idle

(x,y,z)

). The Idle

(x,y,z)

is a parameter

which represents the number of iterations for which a

subarea remains in stable mode and thus is not being

monitored.

Maximum Idle state (maxId). The maxId bounds

the maximum iteration cycles for which a subarea is

allowed to stay inj Idle considered as stable subarea.

It is used to ensure the reliability of the algorithm in

terms of avoiding the formation of holes and to guar-

antee the refreshness rate of each subarea. It assures

that there will not exist any ”ghost-subareas” i.e. ar-

eas which might remain unmonitored for a ”long” pe-

riod of time.

Dynamic Monitoring Algorithm (DMA). In this

subsection we present the dynamic monitoring al-

gorithm which represents the dynamic monitoring

scheme. In this algorithm, for each subarea and for

each iteration cycle, the concentration and the weights

of their pollutants are measured. The same concep-

tion holds for the SMA with the main difference that

the DMA takes into consideration the property that a

subarea might be monitored or not depending on its

stability parameter. Initially the algorithm will moni-

tor the area D for a minimum iteration cycles (min i)

before it will start taking into account the stability

parameter of each subarea. The maximum iteration

cycles for which the algorithm will be executed is

set by max i and the maximum idle iteration cycles

(Idle

(x,y,z)

) for which a subarea can remain at stable

is set by maxId. The algorithm 2 is presented below.

3.1.3 Complexity

In this section we present and compare the time com-

plexity of our two proposed algorithms. In the ﬁrst

scheme (SMA), the visiting pattern of the subareas by

the drone and hence their monitoring by the sensors,

is done in a continuous-sequential way, in which all

the subareas are monitored in every monitoring cy-

cle. To measure the time complexity of the two algo-

rithms, we consider the number of measurements per-

formed assuming the following: k=l=m=n-1 (in par-

ticular the x axis is tessellated in n equal parts and the

same holds for the y and z axis); the transportation

Algorithm 2: Dynamic Monitoring Algorithm

(DMA).

Input: The values:MR, DM, k,l,m, min i, Thr, LiC,

maxId

Output: The Weight W

(x,y,z,i)

MR ← default Measuring Rate

DM ← default Duration of Measurement

k, l, m ← size of each axis of area D

min i ← minimum iteration cycles

max i ← maximum iteration cycles

T hr ← Threshold deﬁning an subarea as ”stable”

LiC ← Last iteration Cycles to Compare

maxId ← Maximum Idle-state value

x, y, z,i ← 0

begin

while i < max i do

for z ← 0 to z = m do

for y ← 0 to y = l do

for x ← 0 to x = k do

if i > min i and

∑

i−LiC

(x,y,z,i−1)

−W

(x,y,z,i)

LiC

T hr

and Idle

(x,y,z)

< maxId then

(x,y,z,i)

← W

(x,y,z,i−1)

Idle

(x,y,z)

+ +

else

−−→

CPa ← Take MR · DM

samples of pollutant a

(x,y,z,i)

← Arithmetic mean

−−→

CPa

Idle

(x,y,z)

← 0

x ++ %next subarea of x axis

y ++ %next subarea of y axis

z ++ %next subarea of z axis

i ++ %next iteration cycle

x, y, z ← 0 %restart from S

(0,0,0)

return W

(x,y,z,i)

end

time Tt needed to move from one subarea to a neigh-

boring one in comparison to the monitoring time

needed to monitor a subarea (Tm = MR·DM) is neg-

ligible, i.e. Tt<<Tm. Therefore the time complexity

of the SMA algorithm is: n

·MR·DM . In the second

scheme (DMS), the visiting pattern of the drone and

the monitoring of the area D, is done in a dynamic

way based on the decision of whether a subarea is

stable or not. Considering a subarea as stable allows

the system to bypass it and move to the next subarea.

The efﬁciency of this algorithm lies in the fact that

some subareas might not be monitored which results

in less power consumption of the whole system, or

in extended monitor space. The time complexity of

the DMA algorithm is O(n

·MR·DM), but depending

AIRWISE-AnAirborneWirelessSensorNetworkforAmbientAirPollutionMonitoring

235

on the algorithm’s input values and the environmental

parameters the DMA could perform even better than

that.

3.2 Implementation of AIRWISE

As far as the implementation of the AIRWISE sys-

tem is concern, we had to face the following chal-

lenges : the limited energy resources of the unmanned

aerial vehicles and the sensor nodes; the assembly of

a lightweight UAV which would be able to carry on

the additional payload of the sensor node; the integra-

tion of a ﬂying mechanism that could enable the UAV

to ﬂy also autonomously; and lastly the development

a WSN system that would be able to support the mo-

bility of the UAV in a three-dimensional environment

and transmit its data in near-real time. The imple-

mentation and our proposing solution towards those

challenges are divided in two subsystems which we

present below: the airborne-ﬂying subsystem and the

WSN subsystem.

3.2.1 Airborne Subsystem

Due to the nature of the problem of monitoring the

ambient air quality, one of the key requirements that

we needed to face was the implementation of a sys-

tem that would be able to take measurements in

the air in three-dimensional spaces. The solution

that we propose towards this challenge is the use of

unmanned aerial vehicles (UAVs) and in particular

quadrocopters. Quadrocopters have the ability to take

off and land horizontally, they are also able to spin

around their vertical axis and most importantly hover

in the air. Their ability of hovering in the air allow us

to maintain them in the air at speciﬁc positions for as

long as it is needed. Alternative airborne systems that

are using small planes are not able to hover and thus

are not suitable for our application. The drone (the

term is used interchangeably with the term quadro-

copter) that we use in our system is shown in Figure 2

(Left) and we self assembled it from parts which are

produced by 3DRobotics. It is a lightweight and pow-

erful APM Copter with a load capacity of approxi-

mately 600gr. It beneﬁts from mechanical simplic-

ity and design ﬂexibility and despite its small size it

is capable of lifting small payloads. The four blades

of the drone as well as its communications are con-

trolled via the ardupilot, which is an open source

UAV platform able to autonomously control multi-

copters. We equipped the drone with a GPS antenna

and with a telemetry set operating at 433Mhz. In

our implementation we used the version of ArduPi-

lotMega 2.6 which gives us a lot of advantages such

as: autonomous ﬂight; automatic stabilization; nav-

igation using GPS; reception of telemetry informa-

tion and control of the drone in real-time using the

MAVLink protocol.

3.2.2 WSN Subsystem

To achieve the main goal of our work (i.e. to auto-

matically monitor the ambient air and extract infor-

mation regarding its quality) we use a wireless sen-

sor network. This network is comprised of two nodes

with gas sensing capabilities and one basestation for

receiving the data from those nodes. One node was

dedicated for the airborne measurements and the other

one for ground measurements used for comparisons.

Both of them were transmitting their collected data

to the basestation. The nodes are comprised of the

following components: a) an electronic board for ac-

commodating the gas sensors, b) the gas sensors, c) an

external antenna for communicating with the basesta-

tion, d) a main board with the processor, e) a GPS

module and f) a rechargeable battery. Due to the fact

that the nodes and their components are very sensi-

tive and fragile we designed and 3D-printed a cover

box to protect them. The complete assembled node,

its cover box and the basestation are shown in Fig-

ure 2 (Right). Both the nodes and the basestation we

used are manufactured by Libellium (lib, ). As far

as the nodes are concerned, we used as their main

board the Waspmote v1.2. The Waspmote node runs

with the ATmega 1281 microcontroller at a frequency

of 14.7456 MHz and with a memory of 128kB. On

top of the mainboard, an 2dBi XBee pro 802.15.4

antenna was integrated for communicating with the

basestation. In addition, a sensor board with tem-

perature, humidity, atmospheric pressure and gases

sensors was integrated. In particular, the gases sen-

sors that we installed were: Molecular Oxygen (O

Ammonia (NH

), Methane (CH

) and Carbon Diox-

ide (CO

) manufactured by Figaro(ﬁg, ). Moreover,

we equipped the nodes with a GPS module so that

we could time-stamp and position-stamp the measure-

ments taken by the sensors. The energy supply of

Figure 2: (Left) Drone used in AIRWISE, including GPS

and telemetry antenna. (Right) Basestation, wireless sensor

node and node’s cover box.

SENSORNETS2015-4thInternationalConferenceonSensorNetworks

236

the nodes was provided by a Li-Ion rechargeable bat-

tery with a capacity of 6600mAh. The size of the

box including all the components was 8x8x7 cm and

it weighted in total 300gr. with a battery weighting

200gr.

On the other endpoint of our WSN subsystem,

the basestation was equipped with a 5dBI XBee pro

802.15.4 antenna. It was connected via a USB to a

computer for receiving and propagating the informa-

tion to the AIRWISE program, which we developed

in C#. This program was designed to be responsible

for logging all the information that is receiving, an-

alyze them in order to calculate the concentration of

the pollutants

−−→

CPa and their weights W

(x,y,z,i)

and as

well visualize them.

4 EXPERIMENTS AND

EVALUATION

4.1 Overall Experimental Set Up

The overall experimental set up of the system can be

seen in Figure 2. The weight of the drone itself was:

1.5kg and the additional weight of the sensor node

was 0.3kg resulting in a total weight of 1.8kg. For

our experiments we chose an area of 6.3 hectares in

a heterogeneous environment in-between of a small

forestall area and residential buildings.

The experiments we conducted regarding the AIR-

WISE system were divided in the three following cat-

egories: a) WSN behaviour, b) Airborne system be-

haviour and c)integration of the WSN and airborne

system.

4.1.1 WSN Experiments

Firstly we run experiments to determine the behaviour

of the WSN subsystem. In order to achieve highly ac-

curate calibration of the gas sensors, speciﬁc chemi-

cal gas tubes need to be used. However, as the mea-

surements of the pollutants with high laboratory ac-

curacy is out of the scope of this paper, the calibra-

tion of the gas sensors was done based on trial and

error. Nonetheless, even if we could not achieve high

accuracy we could obtain very accurate variations in

the concentration of the pollutants between different

measurements.

For our experiments we installed gas sensors for

, CH

, NH

and O

, along with sensors for en-

vironmental parameters of temperature, humidity and

atmospheric pressure. The raw data acquired from the

gas sensors, the environmental sensors and the GPS,

Figure 3: AIRWISE program for receiving, analyzing and

visualizing the data collected from the sensor node.

were sent to the basestation using the XBee antenna in

four separate packets. Once the packets were received

by the basestation, the AIRWISE backbone program

running on a laptop analyzed the raw data and visual-

ized them in a user friendly way. A screenshot of the

program while it was receiving data from the wire-

less node is shown in Figure 3. In order to complete

one data gathering cycle (from the sensors described

above) at one speciﬁc location, it was required 1 min.

and 15 sec. This relatively big amount of time intro-

duced some energy and time related problems that we

will discuss below.

4.1.2 Airborne System Experiments

As far as the experiments of the airborne system are

concerned, we were able to operate the quadrocopter

described previously in two different ﬂying modes:

the automatic and the manual one. The automatic

ﬂying mode uses the APM 2.6, a GPS receiver, an

accelerometer and the ”mission planner” software in-

stalled on a laptop. Via this software we were able

to set speciﬁc waypoints in a area and program the

drone ﬂy towards those waypoints. Once we set up

the waypoints on a graphical interface, we uploaded

them to the APM of the drone using the MAVLINK

protocol. The beneﬁt of the automatic ﬂying mode

enables the drone to take off and land without our in-

tervention. Moreover, we could send commands to

the drone,in real time, while it was ﬂying to change

its direction. This was proven especially useful when

the pollution in some areas was higher than expected.

In the second ﬂying mode of the drone i.e. the manual

one, we used a Futaba 7-Channel Radio Transmitter.

The auto-stabilization system of the APM stabilized

the drone even in the presence of strong winds. In or-

der to maintain the safety precautions, the drone was

landing when its battery was at 20%. Its maximum

ﬂying time with a fully charged 5000mAh 11.1V LiPo

AIRWISE-AnAirborneWirelessSensorNetworkforAmbientAirPollutionMonitoring

237

battery without any payload, was approximately 15

minutes.

4.1.3 WSN and Airborne System Integration

Experiments

In the last set of experiments we combined and tested

the integration of the WSN and the drone. For this

category of experiments, we deﬁned a fraction of our

overall experimental area, a small cubic area D. The

edges of this cubic area of interest were 39 meters

long with a volume of 59319m

. This area was tes-

sellated in 3x3x3 subcubes where the centers of each

subcube (subarea S) were 13 meters apart from each

other. Every time the measurements were gathered

from each subarea, the collected data were sent to the

basestation in near-real time, and simultaneously they

were also saved locally. Due to the additional weight

of the sensor node and its battery, the maximum ﬂight

time of the drone was reduced from 15 to 12 minutes.

Initially we set up the sensor node to collect data from

all of its sensors (i.e. pollutants, environmental pa-

rameters and GPS). In these initial experiments, the

time needed to perform measurements from one sub-

area was 1 min. and 15 sec. and compared to the 12

min. of maximum ﬂight time of the drone, we were

able to gather measurements only from 9 subareas.

Those 9 subareas correspond to only 0.33 iteration cy-

cles and for covering the whole area D we needed at

least 27 measurements ( i.e. one iteration cycle). The

traveling time from the endpoint of one layer to the

starting point of an other was in average 4 seconds.

4.2 Evaluation

In order to evaluate better our algorithms in this real

world development, we set the WSN subsystem to

measure only the CO

in the air, including though

GPS and environmental parameters. This shortened

signiﬁcantly the subarea’s data gathering cycle to 15

seconds. The experience we acquire from this fact is

that for the time being the batteries of the drones, de-

spite being off the shelf, are not yet adequate to per-

form complex tasks. For this reason we do need to

develop efﬁcient mechanisms to overcome those en-

ergy constraints.

Figure 4 shows the results we obtained from mea-

suring the CO

using the SMA and DMA during the

12 min. lifespan (ﬂying time) of the drone. The SMA

scheme reported in this lifespan, measurements from

45 subareas. These 45 subareas correspond to 1.67

iteration cycles. On the other hand, using the DMA

scheme (with: Threshold at 0.5%, min

and LiC at 5

and the maxId at 10), for the same 12 min. life-span,

Figure 4: CO

reported concentrations using SMA and

DMA algorithms during the lifespan of the drone.

a total of 74 measurements were reported. These 74

measurements correspond to 2.74 iteration cycles.

Comparing the performance of the two algo-

rithms, we observe that the DMA algorithm performs

better and in particular it can report 29 more measure-

ments than the SMA with a 0.5% tolerance in the CO

concentration. Consequently, the DMA is approxi-

mately 64% more energy efﬁcient. In Figure 4 we

can observe also that the two algorithms report almost

identical measurements. The only drawback using the

DMA scheme is that more messages need to be sent

and received which impacts negatively in the energy

consumption of the sensor node. Speciﬁcally, using

the DMA, the battery of the sensor was reduced by

4% whereas using the SMA it was reduced by 2%.

However, comparing the battery depletion rate of the

sensor node to the one of the drone, the difference is

almost negligible.

Due to the design of the DMA, it is let on the free-

dom of the system operator to decide the tradeoff be-

tween the sensitivity of the measurements and their

quantity. Meaning that a bigger value in the Threshold

would allow for more measurements while a smaller

value would allow for more precise ones. The advan-

tage of the near-real time monitoring of our system,

is that a meteorologist for example in a scenario of a

volcanic eruption, could change on-the-ﬂy the trajec-

tory of the drone towards another area of interest. In

addition, in emergency pollution situations, by using

the architecture of the AIRWISE system, more drones

with more sensors could be dispatched for a more de-

tailed monitoring. The AIRWISE backbone system

which can be run on a laptop, makes the whole system

easily portable and transferable.

5 CONCLUSIONS AND FUTURE

WORK

In this paper we investigated the challenges of the

air quality monitoring and we presented a system-

SENSORNETS2015-4thInternationalConferenceonSensorNetworks

238

solution using WSNs and UAVs. We proposed a sys-

tem’s architecture together with a theoretical frame-

work and two schemes for monitoring the air pollu-

tion in 3D spaces. Furthermore, we showed the imple-

mentation of our approach with which the automatic

monitoring of the ambient air can be facilitated. We

have extended the capabilities of airborne systems by

coupling them with WSNs. In particular, we imple-

mented the AIRWISE system which is able to moni-

tor pollutants in the air such as: NH

, CH

, CO

, the

percentage and environmental parameters such as

temperature, humidity and atmospheric pressure. We

developed the system, we run experiments with it and

lastly we evaluated and compared our schemes and

algorithms in a real deployment scenario.

Our future work plans include scaled up experi-

ments with more drones and sensors acting in a col-

laborative way. In addition, we plan to investigate

the direct interconnectivity between the wireless node

and the autopilot system of the drone.

ACKNOWLEDGEMENTS

This work was partially supported by the EU/FIRE

IoT Lab project - STREP ICT-610477.

REFERENCES

European union air quality sector, http://ec.europa.eu/

environment/air/quality ,retrieved sept. 2014,.

www.ﬁgaro.co.jp/en , retrieved sept. 2014.

www.libelium.com , retrieved sept. 2014.

Barakeh, Z. A., Delbart, V., and Bonnet, F. (2014). Multiple

gas sensors system for environmental and air quality

assessments. a way to perform environmental moni-

toring in smart cities. International Conference on

Sensor Networks, SENSORNETS, pages 360 – 364.

Chen, Z., Shi, Z., and Guo, Q. (2013). Design of wireless

sensor network node for carbon monoxide monitoring.

Telecommunication Systems, 53(1).

Chow, J. C. et al. (2007). Will the circle be unbroken: A

history of the u .s . national ambient air quality stan-

dards.

Evangelatos, O., Samarasinghe, K., and Rolim, J. (2013).

Syndesi: A framework for creating personalized smart

environments using wireless sensor networks. In Dis-

tributed Computing in Sensor Systems (DCOSS), 2013

IEEE International Conference on.

Hu, S.-C., Wang, Y.-C., Huang, C.-Y., and Tseng, Y.-C.

(2011). Measuring air quality in city areas by vehicu-

lar wireless sensor networks. J. Syst. Softw.

Jude, A. et al. (2007). Sensorﬂock: an airborne wireless

sensor network of micro-air vehicles. In Proc. of the

5th international conference on Embedded networked

sensor systems, SenSys ’07.

Kavi K. Khedo, Perseedoss R., M. A. (May, 2010). A

wireless sensor network air pollution monitoring sys-

tem. International Journal of Wireless and Mobile

Networks (IJWMN), 2.

Lee, H., Olsen, S. C., Wuebbles, D. J., and Youn, D. (2013).

Impacts of aircraft emissions on the air quality near

the ground. Atmospheric Chemistry and Physics Dis-

cussions, 13(1):689–727.

Nickel, V., Ottersbach, P., Reichert, R., and Schfer, M.

(2014). Smart fabrication of robotic systems. Inter-

national Conference on Sensor Networks, SENSOR-

NETS, pages 345 – 350.

Solazzo, E. et al. (2013). Evaluating the capability of

regional-scale air quality models to capture the ver-

tical distribution of pollutants. Geoscientiﬁc Model

Development, 6(3):791–818.

Teh, S., Mejias, L., Corke, P., and Hu, W. (2008). Exper-

iments in integrating autonomous uninhabited aerial

vehicles (uavs) and wireless sensor networks. In

Proceedings of the 2008 Australasian Conference on

Robotics and Automation (ACRA 2008).

Valente, J., Sanz, D., Barrientos, A., Cerro, J. d., Ribeiro, n.,

and Rossi, C. (2011). An air-ground wireless sensor

network for crop monitoring. Sensors, 11(6):6088–

6108.

Waharte, S. and Trigoni, N. (2010). Supporting search and

rescue operations with uavs. In Int. Conf. on Emerging

Security Technologies (EST), 2010.

Wang, D. et al. (2010). Monitoring ambient air quality

with carbon monoxide sensor-based wireless network.

Commun. ACM.

Willmann, J. et al. (2012). Aerial robotic construction. In-

ternational Journal of Architectural Computing.

Yaacoub, E., Kadri, A., Mushtaha, M., and Abu-Dayya, A.

(2013). Air quality monitoring and analysis in qatar

using a wireless sensor network deployment. In Wire-

less Communications and Mobile Computing Confer-

ence (IWCMC).

Yim, S. H., Stettler, M. E., and Barrett, S. R. (2013). Air

quality and public health impacts of {UK} airports.

part ii: Impacts and policy assessment. Atmospheric

Environment, 67(0):184 – 192.

AIRWISE-AnAirborneWirelessSensorNetworkforAmbientAirPollutionMonitoring

239