Guidelines and Challenges Towards the Implementation of Intelligent
Sensing Techniques in a Water Quality Prediction Application
Marcos X.
´
Alvarez, Mois
´
es S
´
anchez, Olga Zlydareva, Gregory M.P. O’Hare and Michael J. O’Grady
CLARITY: Centre for Sensor Web Technologies, School of Computer Science and Informatics,
University College Dublin, Belfield, Dublin 4, Ireland
Keywords:
Environmental Wireless Sensor Networks (EWSNs), Water Quality Forecast, AFME, Crowdsourcing.
Abstract:
SmartCoasts is an INTERREG 4A project aimed at providing novel solutions for real-time monitoring and
forecasting of coastal water quality. The intended predictive system relies on freely available online weather
forecasts and a suite of real-time meteorological data measured across a river catchment. In a preliminary
stage, a prototype has been developed taking the real-time data from GPRS loggers deployed at strategically
located stations according to a centralised architecture. Even though such system has proven its suitability
providing accurate predictions, certain pitfalls that hamper usability have been detected. Adding intelligent
capabilities to the sensing nodes might help to overcome such situation. This paper presents a general overview
of the current situation and discusses some of the major challenges and difficulties that need to be faced in
order to set up a really smart Environmental Wireless Sensor Network.
1 INTRODUCTION
For more than a decade, Wireless Sensor Networks
(WSNs) have been widely adopted both in indoor
and outdoor monitoring applications. While indoor
locations are usually less exposed to external inter-
ferences and weather hazards, which eases durabil-
ity, maintenance and repeatability of the underlying
hardware and software, outdoor applications are more
prone to be affected by the surrounding conditions.
Nevetherless, applications where broad coverage area
is a must, such as climate monitoring (Arampatzis
et al., 2005), water quality monitoring (Le Dinh et al.,
2007) and precision agriculture (Hwang et al., 2010),
to cite some, are becoming nowadays increasingly
popular.
The initial achievements in deploying Environ-
mental Wireless Sensor Networks (EWSNs) as a reli-
able solution for data collection and observation were
mostly based on centralised architectures where sen-
sor nodes simply act as data collectors and transmit-
ters towards one or more base stations and lack any
processing capabilities, as this happens in a single
server (Martinez et al., 2004). Such approach is sim-
ple and robust, but lacks flexibility on scaling and
software upgrading. Moreover, maintenance is com-
plicated due to the remote location of the nodes.
On the other hand, distributed EWSNs are use-
ful in many scenarios, such as water quality moni-
toring and forecasting, where wide area continuous
monitoring and minimum human intervention are re-
quired. To achieve such goal, the nodes in the net-
work should collaborate in a self-managed manner,
taking autonomous decisions about gathering, pro-
cessing and transmitting useful data at minimum en-
ergy cost.
This paper is a case study on the deployment of an
EWSN aimed at feeding a marine water quality fore-
casting system for short and mid-term predictions. It
is our goal to outline the main lessons learnt from
the work conducted until now, providing some gen-
eral ideas that could contribute to enable the reliable
implementation of generic environmental monitoring
applications. In particular, we consider that upgrading
from a centralised architecture, based on stand-alone
data loggers, to a distributed one might contribute to
improve the general performance of the forecasting
system.
The rest of the paper is organised as follows. In
Section 2 the currently deployed system and its main
components are described. Section 3 identifies the
drawbacks which we faced during the exploitation of
SmartCoasts, while Section 4 introduces the proposed
solutions, finishing with the conclusions in Section 5.
428
X. Álvarez M., Sánchez M., Zlydareva O., M. P. O’Hare G. and O’Grady M..
Guidelines and Challenges Towards the Implementation of Intelligent Sensing Techniques in a Water Quality Prediction Application.
DOI: 10.5220/0004902104280431
In Proceedings of the 3rd International Conference on Sensor Networks (MOEOD-2014), pages 428-431
ISBN: 978-989-758-001-7
Copyright
c
2014 SCITEPRESS (Science and Technology Publications, Lda.)
Figure 1: SmartCoasts deployment area map. Measurement
locations: - River level; - Weather station; - Rain
gauge; - Microbial sampling; - ADCP buoy; - Water
quality validation point.
2 GENERAL OVERVIEW OF THE
SMARTCOASTS PROJECT
Smart Coasts (SCs) (SmartCoasts, 2013) is an Irish
and Welsh INTERREG 4A project broadly motivated
by the EU Bathing Water Directive, which encour-
ages the development of information and communi-
cations technology (ICT) tools and real-time public
information systems in order to improve the main-
tenance of public health and increase the number of
beaches passing the new EU standards, thus quali-
fying for awards such as the Blue Flag. SCs is cur-
rently being validated in the Bray Bay (Ireland), and
across the catchment of the local River Dargle. Fig-
ure 1 shows a map of the area under study.
The development of a water quality monitoring
and prediction framework as the one described here
leans on contributions coming from a wide spectrum
of research fields, such as Microbiology, Civil Engi-
neering, Computer Science and Telecommunications.
This is an implicit consequence of the holistic ap-
proach followed in the SCs project. Thus, a rich va-
riety of sources act as system inputs, as Figure 1 sug-
gests. The core of the marine water quality forecast-
ing system comprises, on the one hand, an integrated
catchment-coastal model to simulate flow and con-
taminant transport from the Dargle to the coast, and
on the other, a database that provides a simple way
to access and manage the large amounts of input and
output data required and generated by the system –a
detailed description of both components and the sys-
tem inputs can be found in (Bedri et al., 2013)–. Fig-
ure 2 traces out the architecture of the described sys-
tem as well as all the data flows involved.
2.1 Main Features of the Deployed
Network Nodes
The primary meteorological phenomenon driving the
integrated model is forecasted rainfall, which is
amended by real-time observations when a significant
deviation is detected. In-field data across the catch-
ment are collected by commercial data loggers devel-
oped by Isodaq, known as Frogs (IsodaqFrog, 2013)
(see Figure 4). These devices run on an 8-bit micro-
controller and are equipped with a GSM/GPRS mo-
dem, which can be connected to an internal or exter-
nal antenna. Several digital, analogue (16-bit A/D)
and SDI-12 sensor inputs are provided.
A battery life of up to 7 years is possible but, under
the current configuration –the Frogs are transmitting
the collected data on a daily basis–, the estimation is
reduced to two years. This figure will be dramatically
reduced should the sampling rate be increased, as de-
sired for example when big rainfall events occur. In
such situations, the integrated model should get up-
dates every 15 minutes, which would drop battery life
expectancy to only 3 months if the Frogs would keep
working continuously at this rate.
3 DRAWBACKS REVEALED BY
THE CURRENT NETWORK
IMPLEMENTATION
The network currently in operation cannot be actually
considered an EWSN, since the Frogs do not interact
with each other and follow a point-to-point commu-
nication schema through a GPRS base station, thus
completely relying on the public data network infras-
tructure. Such approach has demonstrated its validity
within the scope of the research work described here,
but also unveils a number of weaknesses.
First of all, some of the course sections of the
River Dargle and its tributaries are located in dark or
marginal areas where the GSM/GPRS signal coverage
is precarious, as Figures 3 illustrate. This has strongly
conditioned the final location of the Frog data loggers,
as not the most suitable places in terms of data rele-
vance have been chosen, but those as close as pos-
sible where the base station was reachable. More-
over, signal fading events, unpredictable in the radio
propagation study made during the design phase of
the project, have been sporadically experienced as a
consequence of the changing surrounding conditions,
strongly affecting the radio link reliability.
Secondly, since the density of the deployed log-
gers is very low and their functionality is restricted
GuidelinesandChallengesTowardstheImplementationofIntelligentSensingTechniquesinaWaterQualityPrediction
Application
429
Figure 2: SmartCoasts architecture and data flow schematic.
by the features that the commercial infrastructure of-
fers, the reactivity to the changes in the environment
conditions is very limited. This is because the Frogs
can only be scheduled to send the collected data on a
fixed-period basis. On the contrary, it is desirable that
the network initiates a sequence of events, should a
certain event exceed a predefined threshold. For ex-
ample, for the sake of our application, it would be
really interesting to increase the sampling rate down-
stream, when the rainfall at the river spring outnum-
bers a certain amount, in order to better capture the
flow rise and the evolution of the faecal contamina-
tion washed off.
Finally, the determinism that the communication
scheme shows, allows a quite accurate estimation of
the power consumption in the long term, which is def-
initely a benefit, but on the other hand, provides a sub-
optimal solution. The lack of processing capabilities
in the sensing nodes impedes taking autonomous de-
cisions about who, when and what to transmit, which
would improve the energy management and thus, the
Figure 3: GSM (left) and GPRS (right) signal coverage es-
timation for the River Dargle catchment area. The colour
bars indicate an increasing strength, meaning the left-most
colour a poor signal and the right-most one a good signal.
No signal areas are represented in white.
life expectancy of the network nodes. Given the rela-
tively high costs, both in economic and energy terms,
of transmitting to the base station at this stage, it is
desirable to enable nodes with short-range communi-
cation features when possible, since not always it is
necessary to transmit all the collected information to
the base station, should that transmission could be de-
livered to a neighbouring node.
4 PROPOSED SOLUTIONS
Until now, our research work has mainly focused on
developing a functional prototype that demonstrates
the feasibility of the intended water quality forecast-
ing system. This is why most of the attention has been
given to the integration aspect, with special focus on
data acquisition and management. The next step
currently in progress– is aimed at finding an optimal
solution to the trade-off between energy consumption
and transmission rate. Some improvements both in
the hardware and software levels are proposed.
In the short term, the sensor nodes are being up-
graded from an 8- to a 32-bit microcontroller archi-
tecture. At this moment we are developing initial
lab tests with the ARM-Cortex M3 family, harness-
ing a SAM3S-EK2 evaluation board (see Figure 4).
The newly designed devices offer enough processing
power to deal with the computational requirements
in the future scenario, where intelligence will be dis-
tributed over the entire network, as the nodes are sup-
posed to react immediately when and where an unex-
pected event happens. That functionality is achieved
as a result of implementing the Java ME-based Agent
Factory Micro Edition (AFME) agent platform on all
SENSORNETS2014-InternationalConferenceonSensorNetworks
430
Figure 4: Isodaq’s Frog RX data logger (left) and SAM3S-
EK2 evaluation board (right).
the network nodes. The final goal behind this deci-
sion is maximizing the network life expectancy, as
a consequence of diminishing the number and data
length of the costly transmissions to the base station,
which would be replaced when possible by cheaper
short-range communications between the nodes. Be-
sides, enabling the Java technology in the new nodes
opens the door to wider interoperable infrastructures
that will be able to run the same code on different
nature platforms, thus easing environmental data in-
terchange.
Finally, in the long-term it is expected to allow
the aggregation of crowdsourced data to the sensed
values. With the ubiquitous availability of mobile de-
vices, citizens have turned into a valuable source of
environmental data, as suggested by other state-of-
the-art research works (CobWeb, 2013). In our opin-
ion, such information would contribute to fill the tem-
poral and spatial gaps that might arise in wide area
deployments like this. For instance, an angler by the
river could detect a hazardous discharge, take a geo-
referenced photo and upload it to the system, and this
would trigger the network to increase the sampling
rate in the area and advice the local authorities to take
safety measures. The database-centered approach that
has been already followed will facilitate a smooth in-
tegration of the crowdsourced inputs, as it works as
an abstraction layer that masks the complexity of han-
dling both the data uploaded by the human users and
the outputs from the different sensors involved. This
is a remarkable feature, given the fact that one of the
major bottlenecks in the deployment of environmen-
tal monitoring systems is the great number of differ-
ent proprietary data formats that the sensors can pro-
duce. Furthermore, such database will be the core
of an environmental information system that will pro-
vide short- and mid-term forecasts about the bathing
water quality conditions, as well as historical reports
delivered in open source formats.
5 CONCLUSIONS
In this paper the case study of a bathing water qual-
ity forecasting system based on stand-alone data log-
gers was presented. We consider the deployed sys-
tem as a starting point towards the implementation of
a fully-featured Environmental Wireless Sensor Net-
work with intelligent data managing and routing.
The main pitfalls encountered until now and the
solutions envisaged to overcome those difficulties
have been described. In our opinion, the ideas out-
lined here can benefit several other general-purpose
outdoor environment monitoring applications.
ACKNOWLEDGEMENTS
Smart Coasts is supported by the European Re-
gional Development Fund (ERDF) through the Ire-
land Wales Program (INTERREG 4A) and by Sci-
ence Foundation Ireland (SFI) under the grant
07/CE/I1147.
REFERENCES
Arampatzis, T., Lygeros, J., and Manesis, S. (2005). A sur-
vey of applications of wireless sensors and wireless
sensor networks. In Proceedings of the 20th IEEE
International Symposium on Intelligent Control, ISIC
’05, and the 13th Mediterranean Conference on Con-
trol and Automation, MED ’05, pages 719–724.
Bedri, Z., Corkery, A., O’Sullivan, J.,
´
Alvarez, M. X.,
Erichsen, A., Deering, L., Demeter, K., O’Hare, G.
M. P., Meijer, W. G., and Masterson, B. (2013). An
integrated catchment-coastal model for real-time wa-
ter quality predictions. Environmental Modelling and
Software. Submitted, under review.
CobWeb (2013). http://cobwebproject.eu. Accessed
November 30, 2013.
Hwang, J., Shin, C., and Yoe, H. (2010). Study on an
agricultural environment monitoring server system us-
ing wireless sensor networks. Sensors, 10(12):11189–
11211.
IsodaqFrog (2013). http://www.isodaq.co.uk/product/frog-
rx-data-logger. Accessed November 30, 2013.
Le Dinh, T., Hu, W., Sikka, P., Corke, P. I., Overs, L., and
Brosnan, S. (2007). Design and deployment of a re-
mote robust sensor network: Experiences from an out-
door water quality monitoring network. In Proceed-
ings of the 32nd IEEE Conference on Local Computer
Networks, pages 799–806. IEEE Computer Society.
Martinez, K., Hart, J. K., and Ong, R. (2004). Environmen-
tal sensor networks. Computer, 37(8):50–56.
SmartCoasts (2013). http://www.smartcoasts.eu. Accessed
November 30, 2013.
GuidelinesandChallengesTowardstheImplementationofIntelligentSensingTechniquesinaWaterQualityPrediction
Application
431