Andrew Markham, Niki Trigoni
Oxford University Computing Laboratory, Oxford, U.K.
Stephen Ellwood
Wildlife Conservation Research Unit, Oxford University, Oxford, U.K.
Wireless sensor network, Experimental, Link quality, RSSI.
Existing work has shown that rainfall has an effect on link quality. Some authors report a positive effect in
moist conditions, whereas others demonstrate a significant decrease in link throughput as a result of rainfall or
fog. The precise cause of these variations has not yet been conclusively established. This paper reports on long
term (26 day) link quality results from 12 nodes deployed in a forest. We found that rainfall has the effect of
decreasing the performance of 28% of good links (classified as those having above 90% packet reception), but
simultaneously increasing the performance of 34% of poor links (those having below 50% packet reception).
In addition, it was found that variations in link quality persisted for a few days after rainfall. This suggests that
link variations are not a result of rain induced fading, but rather due to water sitting on node packaging. We
present experimental evidence which demonstrates that changes in link quality (both positive and negative)
are indeed due to the presence of water, capacitively loading the antenna, altering its radiation pattern.
Although there have been a number of experimental
studies of link quality variation over time in a wire-
less sensor network, most have been over a very short
time period (typically a few hours) (Srinivasan and
Levis, 2006; Becker et al., 2009). Furthermore, many
have involved indoor deployments, where the effects
of weather are minimal (Becker et al., 2009; Wang
et al., 2007; Cerpa et al., 2005). Accurate estimates
of link quality are essential for the deployment of a
long-lived and robust sensor network. By using link
estimates, bad links can be avoided. Overall network
performance can be greatly improved by choosing re-
liable links (Wang et al., 2007).
In particular, we are concerned with the impact
that rainfall has on link performance. Some re-
searchers have reported an improvement in link qual-
ity during moist periods (Thelen et al., 2005), whereas
others have reported a detrimental effect (Anastasi
et al., 2004). It is thus not clear whether rainfall is
an environmental factor that should be welcomed or
This study was conducted in order to assess the
best strategy to adopt for forwarding bulk data bun-
dles (i.e. hundreds of kilobytes) with varying delay
requirements. For bulk transfer of information, what
is important is that data is sent over reliable, shortest
path links to the end destination. This means avoid-
ing retransmissions and asymmetry where possible by
building and maintaining a reliable tree. Changes to
the network topology can incur a very large energy
cost if a node has a large volume of data that has to
‘backtrack’ through the tree. If a link which was pre-
viously good fails, the question is whether to delay
transmission or to reroute to another node. In order
to decide these questions, it is necessary to know how
links vary over time, and what level of stability can be
In this study, we attempt to investigate the causes
behind link variation in an outdoor deployment due
to rainfall. This paper presents results from an out-
door deployment in a forest environment over a pe-
riod of 26 days. The quality (both RSSI and LQI)
of all possible links was measured every 5 minutes.
Correlations were performed with data provided by
an advanced meteorological station, part of a system
measuring climate change, located only 500 m away.
Markham A., Trigoni N. and Ellwood S. (2010).
In Proceedings of the International Conference on Wireless Information Networks and Systems, pages 148-153
DOI: 10.5220/0002959201480153
0 20 40 60 80 100 120
X distance [m]
10 11 12
Y distance [m]
Figure 1: Layout of nodes within the deployment area.
Unlike other work, the sole purpose of this work was
to monitor link quality. Thus, there are no interfering
effects from network traffic or MAC layer dependen-
A total of 12 T-Mote SKY nodes were deployed in
Wytham Woods, Oxford, from the period 17/02/2009
to 14/03/2009 (26 days). The nodes were powered by
two Energizer lithium AA batteries and placed within
a waterproof, protective plastic enclosure. Metal
shelving brackets were used to affix the nodes to tree
trunks, at an approximate height of 1.8 m above the
Nodes transmit short beacons (10 bytes, includ-
ing CRC) with an average period of 3 s, uniformly
dithered within this period. The random number gen-
erator was seeded with the node’s unique MAC ID so
that nodes would not be synchronized with one an-
other. The beacons contain the node’s ID and its cur-
rent timestamp. Nodes transmit the beacons at the
highest power level (+0 dBm). Every 5 minutes (uni-
formly dithered), nodes switch to receiving mode and
listen to incoming beacons for a total of 6 s. They
record their own timestamp, the incoming node’s ID
and its timestamp, and also the LQI and RSSI as pro-
vided by the CC2420 radio. This information is stored
in the 1 Mbyte external flash, and is uploaded at the
end of the study. When not transmitting or receiving,
nodes power down their radio modules and enter low
power sleep mode.
The radio modules were configured to use Zigbee
Channel 24. In the deployment area, there were no
interfering WiFi signals. The layout of nodes in the
deployment area is shown in Figure 1.
3.1 Long-term Link Stability
In this section we examine the stability of links, i.e.
if a link that is good on a certain day, will remain
good for the duration of the study. This is impor-
tant for MAC protocols as it indicates the necessary
frequency of route repair. The performance of every
(unidirectional) link is shown in Figure 2. The av-
erage packet reception rate is indicated by the inten-
sity of each cell, with good links represented by white
and bad links represented by black. Links have been
sorted according to their initial reception rates.
Figure 2 demonstrates that the majority of links,
in general, show little long term variation. Some
links do show significant variation, changing from
very high throughput to very low throughput (e.g. link
number 70). However, what is interesting about the
diagram of evolution is that it shows that there are cor-
related variations in link performance, e.g. on Febru-
ary 23 there is significant variation across all links.
A similar pattern is seen around March 4. This sug-
gests that there is some common effect which is af-
fecting all links. Note however that the variation is not
all in the same direction some links have improved
whereas others have become worse. In the following
sections we investigate causes for these variations.
3.2 Correlation with Environmental
In this section, we examine the correlation of the
number of good links in the network (those with over-
all throughput greater than 90%) with a variety of en-
vironmental variables. Data for this study were col-
lected by an automated weather monitoring system
located approximately 500 m from the deployment
area. Table 1 shows the linear (Pearson’s product mo-
ment) correlation coefficients for the various parame-
ters. The results show that the two parameters which
show negative significant correlation (at the 95% con-
fidence level) are rainfall and ambient temperature.
Surface wetness, which indicates the presence of wa-
ter droplets also shows a negative correlation. Corre-
lations for the other measured variables are all low.
3.3 Variation in Link Performance with
The previous section indicated that rainfall was the
physical variable which had the strongest impact on
link performance. In order to quantify the effect of
Figure 2: Evolution of links over time, using a three hour binning. The intensity of each cell represents the packet reception
rate, with white representing 100% and black 0%. Links have been sorted according to their packet reception rate in the first
three hours of operation, with bad links at the top of the figure and good links at the bottom of the figure.
Table 1: Correlation coefficients.
Solar radiation 0.056
Net radiation -0.033
Relative Humidity -0.068
Ambient Temperature -0.247
Wind Speed -0.008
Rainfall -0.291
Albedo (Sky) 0.056
Albedo (Ground) 0.069
Surface wetness -0.154
rainfall on link stability, we characterized links on
March 1 into categories based on their throughput.
Good links are those with a throughput of over 90%,
whereas poor links are those with a throughput lower
than 50%. Fair links are defined as those with a
throughput between 50% and 90%. The purpose of
this investigation is to determine the effect of rain on
different link categories.
The variation of good links over time is shown in
Figure 3. This shows that on the onset of rainfall on
March 3, 27.5% of links which were good on March 1
became poor or fair. Note that links did not instantly
become good again once the rain had stopped. By
March 6, all links are back to being good again. When
it rains again on March 7, 10% of links degrade in per-
formance to become fair.
The variation of poor links over time is shown in
Figure 4. This shows that rainfall has the effect of
actually improving link throughput, as evidenced by
the fact that 37.5% of poor links have become fair in
performance. Similar to the behaviour of good links,
by the 6 March, all but one link has returned to poor
3.4 An Investigation into Rain-induced
The correlation results show that there is a negative
correlation between precipitation and link success.
The results from Section 3.3 demonstrate that some
links improve, whereas others become worse. We
first investigate whether this variation could be due to
fading caused by rainfall itself. An approximate em-
pirical relationship between attenuation A and rainfall
rate R is given by
A = αR
where α and β are parameters that are dependent on
frequency and temperature (Olsen et al., 1978). A
number of models provide values for these parame-
ters depending on weather conditions. The path loss
at 2.5 GHz under extreme rain conditions of R = 250
mm/hr, using the worst case thunderstorm distribution
of Joss et al. (J-T), is calculated to be 0.124 dB/km
(α = 1.63 × 10
;β = 1.2). With a maximum link
range of 100 m, the expected path loss due to rain-
fall is negligible at 0.01 dB. It should be noted that
higher frequencies (e.g. 10 GHz and above) suffer
heavily from rain fading and this attenuation is actu-
ally used in weather radars to measure precipitation
rate and rainfall drop size (Meneghini and Nakamura,
However, there is a clear rain related variation in
link quality, one which has been reported in other
work (Thelen et al., 2005). As this loss is not due
to the channel itself, there has to be another cause. As
we were using the onboard antenna, we conjectured
that this variation in link performance could be due to
WINSYS 2010 - International Conference on Wireless Information Networks and Systems
Mar 1 Mar 3 Mar 5 Mar 7 Mar 9
Number of links
Peak rainfall rate [mm/hr]
Poor Fair
Good Peak Rainfall Rate (mm/hr)
Figure 3: Evolution of links which were classified as good
on March 1. Notice how the number of good links decreases
greatly with the heavy rain on March 3.
water pooling on top of the plastic enclosure, creating
a reflective plane. Water has a relative static permit-
tivity (ε
) of approximately 80. This would have the
effect of altering the antenna radiation pattern which
could account for these changes.
In an attempt to determine if this could indeed ex-
plain the significant variations in link quality, we con-
ducted a simple experiment, the details of which are
described in the following section.
The deployment demonstrated that there was a causal
link between precipitation and link quality. However,
this effect could not be ascribed to path loss, as atten-
uation by rainfall at 2.4 GHz was shown to be negligi-
ble. To investigate the effect of water droplets present
on the plastic enclosure, two nodes were placed 10 m
apart in a laboratory environment. One node (within a
plastic enclosure) was placed on the ground and con-
figured to transmit beacons every 0.5 s. The other
node was connected directly to a PC and used to log
the RSSI of the incoming beacons. Water was sprayed
gradually onto the plastic housing of the transmitting
node over a period of 10 minutes.
In order to indicate moisture presence on the en-
closure, a simple moisture sensor was fabricated from
a small piece of perforated prototyping board. The
resistance of this device indicates the presence (low
resistance) or absence of water (high resistance). The
output from a potential divider with the sensor con-
nected to ground and a 10 k resistor connected to
the 3V supply was applied to the ADC0 input of the
T-Mote SKY. A threshold ADC reading was chosen
which allowed the sensor to report the presence or ab-
sence of moisture.
The results from this experiment are shown in Fig-
Mar 1 Mar 3 Mar 5 Mar 7 Mar 9
Number of links
Peak rainfall rate [mm/hr]
Poor Fair
Good Peak Rainfall Rate (mm/hr)
Figure 4: Evolution of links which were classified as poor
on March 1. Notice how the number of fair links rises with
the heavy rain on March 3.
ure 5. Initially, the RSSI level is -77 dBm. At time t
= 100 s, water was sprayed on the plastic enclosure.
Every further 60 s, more water was added. The RSSI
plot demonstrates that the addition of water first has
the effect of increasing the received RSSI by approx-
imately +4 dB. However, as more water is added, the
RSSI decreases below its original value by approx-
imately -6 dB. Overall, this is a 10 dB variation in
signal strength.
These results show that the presence of water on the
plastic enclosure has the effect of altering the received
signal strength. However, this variation can be posi-
tive or negative, depending on the amount of water
present. Water, with its high dielectric constant, alters
the radiation pattern of the antenna, essentially acting
as capacitive loading. Large amounts of water in close
proximity to the antenna have the effect of detuning
it, reducing its efficiency. These results demonstrate
that signal strength variation is due to the presence of
water on the enclosure, not channel attenuation due
to precipitation. As condensation can occur on nodes
without rain falling (e.g. dew), precipitation is thus
not the only indicator of channel variation.
Due to the causal link between water on the en-
closure and channel variation, it would be beneficial
for nodes to be able to measure this presence. Nodes
could then take informed decisions as to the cause of
a link failure/improvement. We presented a simple
moisture sensor based on measuring the resistance be-
tween two electrodes. This method however has the
drawback that the sensor needs to be mounted on the
exterior of the box. Due to the presence of water, cor-
rosion of the electrodes will occur over time. To some
degree this can be lessened by applying an AC drive
current to the sensor. A better method of measuring
water presence would be to place two parallel plates
to the interior of the enclosure and measure the ca-
0 100 200 300 400 500 600 700
Time [s]
RSSI [dBm]
Figure 5: Experimental results demonstrating variation in
RSSI with water presence on plastic enclosure. At 100 s,
water was sprayed on the plastic housing. Approximately
every 60 s thereafter, more water was added. Note how
the received signal strength initially increases and then de-
creases once a large amount of water is present. Node sep-
aration was approximately 10 m in a laboratory setting.
pacitance. Water present on the exterior of the box
will increase the capacitance. This would be a much
more robust technique, as the sensor element is not
exposed. Some microcontrollers even incorporate a
module for measuring capacitance e.g. the Charge
Time Measurement Unit (CTMU) of the PIC24F se-
ries of microcontrollers (Microchip, 2010), allowing
accurate measurements to be made without requiring
external components.
As identified above, links vary in different ways
according to the presence of water. In a delay-tolerant
network setting, nodes can defer exchange of bulk
data until conditions are favourable for network trans-
mission. This can save considerable energy, by avoid-
ing lossy links. To learn this information, periodic
probe packets can be sent, and the relationship be-
tween link quality and surface wetness modelled at
each node. This could easily be done using a simple
histogram, but a polynomial model relating link qual-
ity to wetness could also be utilized.
One of the major investigations into channel propa-
gation in outdoor wireless sensor networks was con-
ducted in potato fields (Thelen et al., 2005). Mica2dot
nodes were used, with external whip antennas with
a transmission frequency of 433 MHz. The authors
found that transmission was better during conditions
of high humidity, such as overnight and during rain-
fall. They speculated that the cause for this could be
reflection from the vegetation canopy, but did not con-
duct further experiments to investigate whether this
was in fact the cause for variation. Another study
which used both Mica and Mica2dot motes, this time
operating at 868 MHz, found very significant reduc-
tions in transmission range, a reduction from 55 m
to 10 m, during rain and fog (Anastasi et al., 2004).
However, no details were given on the amount of rain-
fall or the duration of the reduction. The authors state
that attenuation of the electromagnetic wave is due to
absorption by water particles, but provide no valida-
tion of this claim.
There is a much larger body of work which re-
ports on link quality in an indoor setting. The work
in this paper has considered link variations over long
periods of time (hours), whereas short term link esti-
mation (milliseconds) is considered by (Becher et al.,
2008) and in (Cerpa et al., 2005). Short term estima-
tion also requires frequent probe packets. An alter-
native approach to link estimation is link characteri-
zation which uses bounds on link metrics instead of
instantaneous measurements (Lin et al., 2008). In a
much more detailed investigation, a similar approach
is specified in order to estimate link quality based
on probe packets sent at the start of the deployment
(Meier et al., 2008). Over 8 million packets were
recorded for each trial. A supervised learning tech-
nique for classifying links is presented in (Wang et al.,
In summary, we have demonstrated that there is a rela-
tionship between rainfall and link performance. This
is a result which has been reported on in other work,
but the causes of this variation have not yet been ad-
equately explained. We showed that the actual fac-
tor which alters link performance is not rainfall itself,
but the presence of water on the exterior of the enclo-
sure. This explains why links take time to recover
from a precipitation event as they slowly dry over
time. Water capacitively loads the antenna, which al-
ters its radiation pattern. In some instances links be-
come worse, but poor links can also improve in per-
formance. It is a simple matter to equip a node with
a sensor that can measure the presence of water. This
information can be used by the MAC and networking
layers to take informed decisions about routing and
link management.
Based on our initial investigation, we plan to in-
corporate capacitance based moisture sensors in a fur-
ther deployment, and to use knowledge of moisture
to influence routing. We also intend to investigate
whether external antennas suffer from water presence
to the same degree.
WINSYS 2010 - International Conference on Wireless Information Networks and Systems
This work was supported by EPSRC WildSens-
ing grant (EP/E013678/1). Meteorological data for
Wytham Woods was kindly supplied by Michele Tay-
lor, Environmental Change Network.
Anastasi, G., Falchi, A., Passarella, A., Conti, M., and Gre-
gori, E. (2004). Performance measurements of motes
sensor networks. In Proceedings of the 7th ACM inter-
national symposium on Modeling, analysis and simu-
lation of wireless and mobile systems, pages 174–181.
ACM New York, NY, USA.
Becher, A., Landsiedel, O., Kunz, G., and Wehrle, K.
(2008). Towards Short-Term Wireless Link Quality
Estimation. In Proceedings of the 5th IEEE Workshop
on Embedded Networked Sensors (IEEE HotEmNets
Becker, M., Beylot, A., Dhaou, R., Gupta, A., Kacimi, R.,
and Marot, M. (2009). Experimental Study: Link
Quality and Deployment Issues in Wireless Sensor
Networks. In Proceedings of the 8th International
IFIP-TC 6 Networking Conference, page 25. Springer.
Cerpa, A., Wong, J. L., Potkonjak, M., and Estrin, D.
(2005). Temporal properties of low power wireless
links: modeling and implications on multi-hop rout-
ing. In Proceedings of the 6th ACM international sym-
posium on Mobile ad hoc networking and computing,
pages 414–425, New York, NY, USA. ACM.
Lin, S., Zhou, G., Wu, Y., Whitehouse, K., Stankovic, J. A.,
and He, T. (2008). Achieving stable network perfor-
mance for wireless sensor networks. In SenSys ’08:
Proceedings of the 6th ACM conference on Embedded
network sensor systems, pages 453–454, New York,
Meier, A., Rein, T., Beutel, J., and Thiele, L. (2008). Cop-
ing with unreliable channels: Efficient link estimation
for low-power wireless sensor networks. In Proc. 5th
Intl Conf. Networked Sensing Systems (INSS 2008).
Kanazawa, Japan: IEEE, pages 19–26.
Meneghini, R. and Nakamura, K. (1990). Range profiling
of the rain rate by airborne weather radar. Remote
Sensing of Environment, 31(3):193–209.
Microchip (2010). PIC24F: Charge Time Measurement
Unit. Available online: www.microchip.com.
Olsen, R., Rogers, D., and Hodge, D. (1978). The aR b rela-
tion in the calculation of rain attenuation. IEEE Trans-
actions on Antennas and Propagation, 26(2):318–
Srinivasan, K. and Levis, P. (2006). RSSI is under appre-
ciated. In Proceedings of the Third Workshop on Em-
bedded Networked Sensors (EmNets 2006). Citeseer.
Thelen, J., Goense, D., and Langendoen, K. (2005). Radio
wave propagation in potato fields. In 1st Workshop on
Wireless Network Measurements. Citeseer.
Wang, Y., Martonosi, M., and Peh, L.-S. (2007). Predicting
link quality using supervised learning in wireless sen-
sor networks. SIGMOBILE Mob. Comput. Commun.
Rev., 11(3):71–83.