IEEE COMPUTER, MANUSCRIPT ID 1

Environmental Sensor Networks

Kirk Martinez, Member, IEEE, Jane K. Hart, Royan Ong

Abstract— Sensor networks for the natural environment require an understanding of earth science, combined with sensor,
communications and computer technology. We discuss the evolution from data logging to sensor networks, describe our
research from a glacial environment and highlight future challenges in this field.

Index Terms— ( wireless sensor networks, radio communications, environmental monitoring, low power design).

—————————— ‹ ——————————

R ECENT developments in wireless network technology

and miniaturisation mean that for the first time, the
not necessarily use a simple one-way data stream over a
communications network. Elements of the system will
natural environment can be realistically monitored. These make decisions about what data to pass on, such as local-
systems can potentially provide new data for environ- area summaries and filtering in order to minimise power
mental science (e.g. climate models) as well as vital hazard use while maximising information content.
warnings (e.g. flood alerts). This is particularly important in
remote or dangerous environments where many fundamen-
WWW
tal processes have rarely been studied due to their inacces-
sibly.
Chong and Kumar [1] suggest that the development of
sensor networks requires technologies from three different Sensor
research areas: sensing, communication and computing. Network

increasing power, memory, size

Within the field of environmental sensor networks, an es- Server
sential fourth component is the application of domain
increasing mobility

knowledge. Before any system is designed and installed, a

detailed understanding of its physical environment and
deployment is required. The systems must be designed to
withstand specific conditions, such as temperature, pres-
sure or vibration. The importance of the data collected, as Base Base
well as their interpretation also needs to be understood. Station Station
This will dramatically affect the design of the communica-
tions and data protection. There are many general surveys
on sensor networks [2], so in this article we discuss the spe-
cific issues of an environmental sensor network. We show
the evolution from environmental logging to a pervasive
sensor network, illustrate the effects of domain knowledge
Sensor
on the three elements of sensor networks from a glacial en-
Nodes
vironment and discuss the challenges associated with envi-
ronmental sensor networks.
Fig. 1. Generic sensor network architecture.

WHAT IS AN ENVIRONMENTAL SENSOR NETWORK?

Figure 1 shows a simple sensor network. The sensor
Environmental monitoring has a long history, including nodes gather data autonomously and a network is used to
analogue loggers such as early paper plotters measuring pass the data to one or more base stations, which forward
barometric pressure and the recording of specific environ- the data to a Sensor Network Server (SNS). We envisage a
mental parameters. Loggers record data at specific intervals wide range of data from different networks being made
and require manual downloading by a maintenance team. available on such servers. A satellite image or map may be
A sensor network is designed to transmit the data from required together with some raw data to aid in interpreta-
an array of sensors to a data repository on a server. They do tion or visualisation. Web publishing through interfaces
———————————————— such as Web Services, give researchers seamless access to
• K.Martinez and R.Ong are members of the School of Electronics and Com- information.
puter Science, University of Southampton, SO17 1BJ UK. Phone +44 Moving from sensor nodes through to the data server the
(0)2380594491, email: [email protected]. systems generally increase in compute power, data storage
• J.K.Hart is a member of the School of Geography, email: [email protected]
and power availability. Sensor nodes may only have power
Manuscript received (insert date of submission if desired).
xxxx-xxxx/0x/$xx.00 © 2004 IEEE
2 IEEE COMPUTER, MANUSCRIPT ID

to survive a few months or years but even base stations will deep). Each probe is equipped with pressure, temperature
have a limited life due to their power supply and harsh and orientation (tilt in three dimensions) sensors. The
conditions. The mobility of sensor nodes or base stations probes are not recoverable.
may be high and require location systems. If a large number
of sensor nodes is possible for one base station then they Base Station
Reference
would typically be organised as an ad-hoc set of clusters Station
with representative nodes communicating a group’s data to
the base. Ice
The evolution from logging to a pervasive sensor net-
SNS
work can be illustrated by the Argus video system for near- Sediment
shore monitoring [3] which has evolved from passive video
recording in 1992 to the active intelligent processing of im- Fig. 2. GLACSWEB System overview
ages. This is a good example of where data reduction is
necessary within the sensor network because otherwise
video data would overload the communications infrastruc- The Base Station (Figure 3) doubles as a communication
ture. relay between the Probes and the Reference Station, and as
Another example is the automated monitoring of soil wa- the controller for autonomous operation that orchestrates
ter sensors in 1992 [4], which could be seen as a forerunner the entire system. It has a differential GPS (dGPS) to allow
of the current small-scale habitat modelling sensors. These ice movement to be followed. It also has temperature and
include systems made by the Center for Embedded Net- tilt sensors, a snow meter and camera. It is controlled by a
work Sensing, which use an array of sensors to measure the PIC and uses PICs to interface to some modules. A real-
population of birds and other species [5], NASA’s Hunting- time-clock is used to wake the system up.
ton Gardens project [6] and the Berkeley’s habitat model-
Long range
ling at Great Duck Island [7]. communication
On a much larger scale, the development of Environ-
Short range Transceiver Radio modem
mental Observations and Forecasting System (EOFS) which communication (433 MHz) (466 MHz)
combine real time in-situ monitoring with distribution net-

RS232
works that carry data to centralised processing sites. One Solar
Panel
example of this is the CORIE project which studies the Co-
lumbia river estuary [8]. Similarly the FLOODNET project GPS dGPS
PIC module
[9] plans to provide a flood warning in the UK and SE- PIC Micro-
Power controller
(Sealed Lead-acid)

COAS [10] will monitor coastal erosion around small is- supply
12V battery

lands intended as wind-farms. GSM

The next step must be to monitor in more remote or hos- PIC

tile environments. These include projects NASA’s project in I2C

Antarctica [11], and further plans to use sensor networks on Snow

Mars [12] and Europa [13]. GLACSWEB is a preliminary 10
11 12 1
2
PIC
9 3 FLASHROM
step in the monitoring of a hostile environment, and many 8
7 6 5
4
(32kB)
of the problems encountered in developing a system to Realtime Clock

work in a glacial environment, will be transferable to these Fig. 3. System diagram of the Base Station
remote environments both on earth and elsewhere.
The Reference Station is a mains-powered Linux-based
gateway for transferring data. It is the position reference
GLACSWEB
point and records a dGPS file daily. This PC relays the data
Monitoring the behaviour of ice caps and glaciers is an from the probes, base station and dGPS to the data server in
important part of our understanding of the Earth’s climate. Southampton on a daily basis (via ISDN).
Understanding the subglacial environment is an ongoing In order to survive for one year most of the system is
research area which is addressed in this project [14, 15]. To powered-off between readings and controlled by a real-time
accurately study this environment the system must autono- clock (RTC). The power-budget allows the probes to wake-
mously record glaciers over a reasonable geographic area up every 4 hours to take readings. However the communi-
and over a relatively long time. It also needs to be as non- cation channel is only opened once a day during a system-
invasive as possible to allow the sensor nodes (probes) to wide window. The daily sequence of events is shown in
mimic the movement of stones and sediment (till) under the Table I. At the end of each period, the probe and base sta-
ice. tion configure their RTCs to the next “wake-up” time before
The system described here consists of: probes inserted in shutting down.
the glacier, a Base Station on the glacier surface and a Ref-
erence Station that relays data to the SNS in Southampton,
as shown in figure 2. Nine probes were deployed in 2003; a
TABLE I. SYSTEM TIMELINE
majority at the ice-till boundary (between 50m to 80m
MARTINEZ ET AL.: ENVIRONMENTAL SENSOR NETWORKS 3

Time Probe Base Station Ref. Station

Transceiver
(433 MHz)
0000 Data log Short range
communication
0300 GPS log GPS log
RS232
0400 Data log
Orientation
0800 Data log Power supply
Micro-
sensor
(switch-mode)
1200 Data log controller

(Lithium Thionly Chloride)

1600 Comms Comms Comms (16F876)
Pressure

7.2V battery
sensor
Transfer to
1900 SNS I2C

2000 Data log

11 12 1
10
9
2
3
FLASHROM Temperature
8 4 (16kB) sensor
Sensor Nodes 7 6 5

Realtime Clock
Sensor nodes have the following requirements: Fig. 5. Simplified system diagram of a Probe
• Low-cost – so many units can be produced.
• Low power – for long-term operation. Communications
• Automated – maintenance free The nature of the environment meant the communications
must meet the following requirements:
• Robust – withstand errors and failures.
• High-power omnidirectional for probes
• Non-intrusive – low environmental disturbance.
• long-range for base to reference
• Low pollution • Low data-rate
The electronics and sensors are enclosed in a sealed plas- • Error-detection and correction
tic cylindrical capsule (Figure 4). Each has one 100psi pres- • Backup channels are needed
sure sensor, two dual-axis micro-electromechanical system Because any part of the communication chain can be
(MEMS) tilt sensors and a temperature sensor. The sensor faulty we use a store-and-forward mechanism for data
readings are read and stored (in FlashROM) by the PIC mi- transfer. The ring-buffer technique in the probes is also
crocontroller. Two PICs were used to enable failsafe control used in the base station so that data flows when communi-
and remote reprogramming. Communication with the base cation channels are available. A long-range (2.5km) hop
station is via a transceiver with an omnidirectional antenna between the base and reference stations is linked with a
(Figure 5). 500mW 466MHz radio-modem with built-in error-handling
(9.6kbit/s suffices). If this fails a backup GSM phone is used
by the base to send data directly to the server in Southamp-
ton. This actually occurred when one radio modem failed in
the reference station.
One consequence of using PIC microcontrollers is that the
use of TCP/ip was ruled out. A custom protocol allowed a
lower overhead and a greater degree of experimentation. A
packet-based protocol with error detection was devised
which also allowed a multi-master bus-like network topol-
ogy could be employed. An extensive use of store-and-
Fig. 4. an open probe (10 cm long) showing its PCB forward, time-outs, checksums and retries allows the sys-
As communication problems can be expected (especially tem to tolerate communication errors. Broadcast packets
during the wet summer and deployment period) the data is allow system time synchronisation for example.
stored in a 16k ring-buffer so it can tolerate a 6 month isola-
Computing
tion. The PIC can receive and interpret commands which
include reprogramming and time schedule changes. A range of different computer systems and software are
The power source had to survive in cold conditions (es- required to build a sensor network:
• Microcontrollers – for sensor nodes
timated around 0˚C) so four Lithium Thionyl Chloride AA
• Small OS – for nodes
batteries were used (3.6V, 2 AH). The efficient switch-mode
• Low-power systems – for base stations
regulator is controlled by the real-time clock and tolerates a • Routing and message-passing
voltage drop from the batteries. • Server – for the sensor network server
• Publishing software – visualisation and services
The use of simple microcontrollers means that non-
standard operating systems have to be used. To fit into the
small memory available on a PIC we did not use an OS at
all. Many researchers use something like TinyOS and this
would lead to fewer bugs and easier maintenance. How-
ever the small memory on most systems still means it is
4 IEEE COMPUTER, MANUSCRIPT ID

difficult to implement complex algorithms. tributed to the loss of communications when the Probe
Base stations need to communicate with many systems slipped into the water-filled zone at the bottom of the hole.
using both custom and standard devices. It is beneficial to Under such circumstances, communication is impossible
use a low power board with a low-power suspend mode until the water freezes in the winter, when we expect each
with a real OS for easy development. Our initial use of PIC probe to transmit their back-log of data.
rulled out the use of standard WiFi networking for exam- Results from the Base Station show battery voltage (Fig-
ple, but meant that some hardware could replace software ure 8), tilt and temperature (Figure 9).
(e.g. RS232 routing). We are currently investigating the use
of a low power StrongARM based board running Linux.
14.50
A Xeon based Linux server is used as the final destina-
tion for the data and all web publication. A key element of 14.00

our development is to publish web services and data de- 13.50

scriptions so that future search engines can gather the data.

Volts
13.00

PRELIMINARY RESULTS 12.50

The prototype system was installed at Briksdalsbreen, 12.00

4 6 8 10 12 14 16 18 20 22 24 26 28 30 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 1 3 5
Norway, in August 2003. Holes were drilled with a modi- day

fied high pressure hot jet-wash after a ground-penetrating

radar survey to map out rivers in and under the ice. The Fig. 8. Base station battery Voltage (from August to October 2003).
base station (Fig. 6) used a tripod sitting on the ice to hold
The base station battery level fluctuated between 12.5V
the antennas, in case snow built up during the winter.
and roughly 13.5V over the course of 65 days (Figure 8).
The overall trend shows that battery charge actually in-
creases during the summer due to the 10W solar panel
mounted on top of the box.
The Base Station’s tilt sensors indicate reasonably stable
on a 15˚ slope and that it moves during warm periods. The
temperature on the glacier (Figure 9) is slowly decreasing
due to the onset on winter and is within the operating
range of the components. The equipment itself remained
stable for two sunny/rainy months and is anchored using
rocks and a backup anchor to a submerged pole. The dGPS
readings indicate that the ice surface melted down around
2m over the summer, which confirmed that drilled poles
can not be used as anchors.
15

Fig. 6. Base station and antenna tripod at Briksdalsbreen. 10

Temperature and pressure (Figure 7) readings were re- 5

Temperature and Tilt

ceived from one Probe (No. 8) for nine days after deploy- 0

ment within the glacier. The temperature was a constant -5

0.8˚C, the tilt readings were constant throughout this period

-10
and are not shown.
-15

20 -20

18 -25
11 18 25 1 5 8 12 15 19 22 26 29 3 6 10 13 17 20 24 27 31 3 7 10 14 17 21 24 28 31
Pressure (psi)

16 day (from Sept to Dec 2003)

14

12
Fig. 9. Base Station temperature/˚C (black) and tilt/degrees (white)
showing winter cooling and occasional box movements.
10

8
These preliminary results show that the system architec-
31 1 2 3 4 5 6 7 ture is sound. The probe communication links were ham-
Day pered by water in the holes, which tended to sink surface
streams. So no data was received for the first few months. It
Fig. 7. Probe 8 Pressure Readings (from July 31, 2003)
is expected that the holes will close with ice deformation
Probe 8 was tightly wedged approximately 20m into the and freezing during the winter.
th
glacier. The cessation of data after the 7 of August is at- Future work includes designing a position measuring
MARTINEZ ET AL.: ENVIRONMENTAL SENSOR NETWORKS 5

system to locate the probes and further miniaturising of communications window could reduce the scalability be-
their electronics. Inter-probe communications and extra cause if many probes had to send a back-log of data the
sensors are also planned for the second system to be in- time could be insufficient. However in our case they would
stalled in 2004. simply send more data the next day or commands can be
sent to probes to keep them awake for longer. Arrays of
base stations or gateways will typically be needed in order
CHALLENGES FOR ENVIRONMENTAL SENSOR
to increase scalability.
NETWORKS
Extracting data gathered by the sensor nodes in remote Remote Management
locations involves some unique challenges. GLACSWEB Systems in isolated locations can not be visited regularly
has tackled many of these issues and led to a greater under- so remote access is essential. Bugs need fixes, subsystems
standing of the solutions. might need shutting down and schedules changed. In our
case we found that a camera on the base station would be
Miniaturisation needed in order to monitor the physical status of the site
Miniaturisation is essential as many systems are de- and systems. Power control to completely isolate subsys-
ployed in confined spaces and may have to be unobtrusive. tems was found to be essential not only for power man-
For low frequency radio the antenna size can be a limita- agement but for workarounds (the Duck Island project
tion. Dielectric antennas measuring only 5x7x0.5mm were found that their sensors could short circuit their power
used to save space as well as for their properties. Surface when wet). Custom communications make remote access
mount components were used together with double-sided more complex because normal logins and routing are not
boards but some some further integration could be available. More software development and failure scenario
achieved through the use of programmable logic for exam- testing is required in order to achieve good remote man-
agement.
ple. The miniaturisation of subsystems has to be balanced:
the limiting factors are still battery size and radio power Usability
requirements. If sensor networks are to be deployed by teams who buy
them off the shelf they need to become easier to install,
Power Management
maintain and understand. In the GLACSWEB prototype an
Power management is essential for long-term operation.
earth-scientist could not install the system because of the
In common with other projects we used a time schedule in range of computer and electronics technologies exposed
order to manage power and employed high-efficiency regu- without simple interfaces. This can be compared to install-
lated switch-mode power supplies. We felt it was risky to ing a scanner, printer and network at home. Plug-and-play
use an extremely adaptive scheme from the start, due to the style developments will help in this area [16].
unknown communication losses and reliability issues. As more data becomes available a major issue is how re-
However a rate of change driven data capture system and searchers actually access effectively. The BeanWatcher [17]
inter-probe ad-hoc communications would theoretically addresses semi-automatic monitoring and management.
reduce power use further. Systems requiring a long boot or
resume time have to be avoided as this can become the Standardisation
dominant factor. Compatibility between off-the-shelf modules such as
dGPS units or weather stations is very low and in practice
Radio Communication separate code needs to be written for every module that is
Radio communications success is unpredictable in wet integrated. In some cases drivers are available, for a web-
and windy locations. We found that theoretical calculations cam for example, but without the correct operating system
of radio losses in glacier ice were a poor guide to reality this is unusable. A future challenge will be to standardise
and this will be common in other environments, e.g. leaf the interface and even some radio networking to allow in-
cover changes in forest habitats. The ability to alter trans- teroperability between different sensor network vendors
ceiver power and the use of lower frequency or acoustic [18].
Standardising the publication of the final data is essen-
fall-back systems will be commonly used in the future.
tial and can be done using semantic web technologies. This
Communications between the sensor nodes is useful where
would essentially join sensor networks into the semantic web
some may be out of the base station’s range. However the
[19], but will be complex unless the community can agree
energy used to set up ad-hoc networks and clusters has to
on some common ontologies to describe the domains. The
be minimised. Generic Exchange Models [20] and IRISnet [21] are a move
Scalability in this direction.
Not only do groups of sensors need to be added regu- Security
larly to environmental systems, but potentially large num- Security issues are important at all levels of a sensor
bers need to be managed. network from physical to data interference. Systems need to
Our initial network topology allowed up to 256 unique blend into the environment and where appropriate carry
devices to be connected to one base station. The use of a warnings, information and possibly alarms. This may be
6 IEEE COMPUTER, MANUSCRIPT ID

less problematic in remote areas. Some systems can cope [14] R.B. Alley, “Continuity comes first: recent progress in under-
with the loss of one or more nodes due to failure or dam- standing subglacial deformation,” In: Maltman, A. J., Hubbard,
age. Data may need to be protected against deliberate and B., and Hambrey, M. J. (Eds.). Deformation of Glacial Materials,
accidental alteration. However security should not be used Geological Society Special Publication, Vol. 176, 2000, pp.171-180.
to hamper public access to information. A balance between [15] J.K. Hart and J. Rose, “Approaches to the study of glacier bed
deformation,” Quaternary International, vol. 86, 2001, pp. 45-58.
security and information needs to be reached so that all
[16] Y. Gsottberger, et al., “Embedding Low-Cost Wireless Sensors
parties can trust the systems, this will dramatically affect
into Universal Plug and Play Environments,” EWSN, 2004, pp.
their development and implementation.
291-306.
[17] A. Lins, et al., “BeanWatcher: A tool to generate multimedia
CONCLUSIONS monitoring applications for wireless sensor networks,” Manage-
ment of Multimedia Networks and Services, Proceedings , Lecture Notes
Environmental sensor networks provide exciting techni- in Computer Science, vol. 2839, 2003, pp. 128-141.
cal challenge. For the first time, data sets of different types [18] F.C. Delicato, et al., “A flexible middleware system for wireless
and scales can be merged together to enhance our under- sensor networks,” Middleware 2003, Proceedings Lecture Notes in
standing of the Earth as a whole. Designing sustainable Computer Science, vol. 2672, 2003, pp.474-492.
sensor networks for the natural environment is a demand- [19] http://www.semanticweb.org
ing task. Communications engineering, power manage- [20] D. Seber, Doganet al., “GEON: Cyberinfrastructure for the Geo-
ment, deployment, weather-proofing, stability and remote sciences” Abstract submitted to American Geophysical Union,
diagnostics have all provided interesting problems. The Fall Meeting, 2003, San Francisco, CA.
convergence of sensor networks with the semantic web will [21] P.B. Gibbons et al., “IrisNet: An architecture for a worldwide
complete the chain from raw data to useful information sensor web,” IEEE Pervasive Computing, vol. 2, no 4, 2003, pp. 22-
available globally. 33.

ACKNOWLEDGMENTS Kirk Martinez is a Senior Lecturer in the School of Electronics and

Computer Science at the University of Southampton, UK. His research
The authors thank Intellisys, BTExact and Topcon for their interests include sensor networks, image processing and the semantic
support and Sue Way, Daniel Miles, David Robertson and web. He has a PhD in Electronics Systems Engineering from the Uni-
versity of Essex. He is a member of the IEEE and the IEEE Computer
Al Ridoch for field assistance and Inge and Gro Melkevoll Society. Contact him on [email protected].
for the use of their café to site the Reference Station. http://www.ecs.soton.ac.uk/~km

Jane Hart is a Reader in the School of Geography at the University of

REFERENCES Southampton, UK. Her research interests include glaciology, glacial
[1] C.Y. Chong and S.P. Kumar, “Sensor networks: Evolution, oppor- sedimentology and climate change. She has a PhD in Environmental
tunities, and challenges,” Proceedings of the IEEE, vol. 91, n.8, 2003, Science from the University of East Anglia.
pp. 1247-1256.
Royan Ong is a research assistant in the School of Electronics and
[2] Akyildiz IF, Su WL, Sankarasubramaniam Y, et al. A survey on Computer Science at the University of Southampton, UK. His research
sensor networks IEEE COMMUN MAG 40 (8): 102-114 AUG 2002 interests include sensor networks and control systems.He has a PhD
[3] R. Holman, J. Stanley and T. Ozkan-Haller T, “Applying video in Electronics Systems Engineering from the University of Leicester.
sensor networks to nearshore environment monitoring,” IEEE
Pervasive Computing, vol 2, no. 4, 2003, pp. 14-21.
[4] M. E. Grismer, “Field sensor networks and automated monitoring
of soil-water sensors,” Soil Science, vol. 154, no. 6, 1992, pp. 482-
489.
[5] B. Charny, “Wireless research senses the future ZDNat News
[Online]”, http//zdnet.com.com/2100-1105-976377.html, 2002,
Dec.
[6] http://sensorwebs.jpl.nasa.gov/resources/huntington_sw31.shtml
[7] R. Szewczyk, et al., “Lessons from a Sensor Network Expedition,”
In the 1st European Workshop on Wireless Sensor Networks
(EWSN '04), January 2004, Berlin, Germany, pp 307-322.
[8] D.C. Steere, et al., “Research Challenges in Environmental Obser-
vations and Forecasting Systems,” Proc. ACM/IEEE Int. Conf. Mo-
bile Computing and Networking (MOBICOMM), 2000, pp. 292-299.
[9] http://envisense.org/floodnet.htm
[10] http://envisense.org/secoas.htm
[11] K.A. Delin, et al., “Sensor Web in Antarctica: Developing an Intel-
ligent, Autonomous Platform for Locating Biological Flourishes in
Cryogenic Environments,” 34th Lunar and Planetary Science Confer-
ence, 2003.
[12] http://sensorwebs.jpl.nasa.gov/
[13] http://fuego.jpl.nasa.gov/uvvm/baseline.htm