Data Quality Issues in Environmental Sensing with Smartphones
Tiago C. de Araújo
, Lígia T. Silva
and Adriano J. C. Moreira
Centro de Investigação Algoritmi, Universidade do Minho, Azurém, Guimarães, Portugal
Centro de Investigação CTAC, Universidade do Minho, Azurém, Guimarães, Portugal
Keywords: Smartphones, Mobile Sensing, Participatory Sensing.
Abstract: This paper presents the results of a study about the performance and, consequently, challenges of using
smartphones as data gatherers in mobile sensing campaigns to environmental monitoring.
It is shown that there are currently a very large number of devices technologically enabled for tech-sensing
with minimal interference of the users. On other hand, the newest devices seem to broke the sensor diversity
trend, therefore making the approach of environmental sensing in the ubiquitous computing scope using
smartphones sensors a more difficult task.
This paper also reports on an experiment, emulating different common scenarios, to evaluate if the
performance of environmental sensor-rich smartphones readings obtained in daily situations are reliable
enough to enable useful collaborative sensing. The results obtained are promising for temperature
measurements only when the smartphone is not being handled because the typical use of the device pollutes
the measurements due to heat transfer and other hardware aspects. Also, we have found indicators of data
quality issues on humidity sensors embedded in smartphones. The reported study can be useful as initial
information about the behaviour of smartphones inner sensors for future crowdsensing application developers.
According to an industry track report from a
specialized website (eMarketer Inc. 2015), it is
expected that almost 70% of the world population
will be using a smartphone in 2017, corresponding to
an absolute number of 5.13 billion of people, among
which 2.97 billion will be using internet regularly on
their smartphones. Also, the potential of the new
mobile devices due to its embedded sensors and
processing capabilities goes beyond of what most
users probably can perceive. The recently announced
smartphones from the two market-share leaders
(Samsung Galaxy S7 and Apple iPhone 7) have
hardware capabilities and performance comparable to
high-end computers of few years ago.
Facing these facts, researchers of ubiquitous and
pervasive computing have conducted several works
to investigate the capabilities of smartphones for
collaborative sensing, participatory sensing and
correlated topics. For example, D’Hondt et al. (2013)
implemented a model for measuring noise levels in
urban spaces through a proprietary application using
smartphone’s microphone, when idle, for estimation
of outdoors noise levels. Through the GPS metadata,
the authors could compare their results with official
levels measured by proper devices, and they found a
high correlation between the results leading to the
conclusion that participatory sensing can, under
appropriate conditions, be an alternative to the
conventional monitoring systems.
Investigating the potential of subjective analysis
of the environment using the participatory sensing
approach, Kotovirta et al. (2012) shared their
experiences with the observation of algae presence in
lakes through the feedback of non-specialist users.
The users, willingly, when near a lake send their
evaluation about the presence of algae in it, based
only on visual perception. The data were compared to
those collected by specific biologic monitoring
instruments and, despite the absolute error, they
found a strong qualitative correlation between
observations provided by users and the results
measured by the instruments.
Towards a systemic view of the urban
environment, Kanhere (2011) provided an analysis of
key challenges and possibilities of crowdsourcing
using smartphones. Air quality monitoring, noise
pollution and traffic conditions were cited as potential
areas of research and development. Yet in efforts
de AraÞjo T., Silva L. and Moreira A.
Data Quality Issues in Environmental Sensing with Smartphones.
DOI: 10.5220/0006201600590068
In Proceedings of the 6th International Conference on Sensor Networks (SENSORNETS 2017), pages 59-68
ISBN: 421065/17
2017 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
focused in urban centres, Overeem et al. (2013)
proposed an Android application (Weather Signal)
that uses an algorithm to estimate the air temperature
from the battery temperature of smartphones through
a heat-transfer model considering some additional
parameters. To evaluate the performance of this new
proposal, they compared official air temperature data
from official entities with the data from their
experiment in defined time intervals, and found very
positive indicators, but still requiring some
adjustments in the heat-transfer model.
In a more recent work with smartphones and
urban sensing, the HazeWatch project (Hu et al. 2016)
used the smartphone as an intermediary between a
proprietary data-collection platform and the end-
users. The project relies in the mobility of the
platform, often carried by taxis, bicycles and
voluntaries to identify phenomena that can be unseen
by stationary and official platforms. The
communication between smartphones and the
platforms is through Bluetooth, the smartphone
process the data and then upload them to the cloud
using mobile Internet access networks. They reached
very solid results, but identified the cost of the
platform and its weight as a limitation that hinds the
spread of this initiative, due to motivations-related
There are also efforts towards the user’s
motivation and engagement in collaborative sensing.
When using smartphones, the main issue relates to
battery consumption. People avoid to use applications
that drain too much energy from batteries and has too
few to offer in exchange. Rodrigues et al. (2012)
investigated the “engagement of users” in
participatory mobile campaigns. As they identified
the energy drain as one predominant negative aspect
to attract more – and keep the existing – smartphone
users, the authors introduced a desktop application to
be used in laptops, that are also pervasive, in a study
of human mobility. As results, they appointed that the
initial attraction of users to get involved, in low and
medium quantities, is not difficult, but the main
challenge found is how to keep these users active for
long-periods, as well as to reach a massive numbers
of users even when rewards are considered.
Yet on motivation and engagement studies, the
authors in Zaman et al. (2014) demonstrated that
collaborative campaigns often emerges from
common concerns of a group of people or
community, and proposed a conceptual framework
for management and orchestration of community
campaigns driven by citizens. The most relevant a
subject is for a group of people or community, the
higher are the chances of more users getting involved.
So, keeping these users active through time is also
dependent on how the main subject of a campaign is
important for each individual, on the role each person
can play in it (citizen participation), and also in the
quality of data generated by the campaign and made
available to its users (closing the loop).
Relying on these efforts, and on the fact that there
are a reasonable number of people carrying
smartphones with environmental sensors everywhere,
the idea of using these embedded sensors for a
ubiquitous, collaborative and smart sensor grid
emerged inside the smart cities and environmental
monitoring contexts. Thus, the motivation used as
ground to this investigation is the importance to
develop an analysis about the potential role of
smartphones for the environmental monitoring, either
in urban centres or indoors, using its own hardware in
the data-collection stage through some technical
considerations about the data quality and other related
In recently years we observed an empowerment of
smartphones capabilities through the aggregation of
several features such as GPS, accelerometers,
gyroscopes and lux meters. The presence of these
sensors transformed the mobile phones into versatile
devices. Thus, the embedding of temperature,
pressure and humidity sensors in popular phones,
such as the Samsung Galaxy S4
(iFixit, 2013),
highlighted the possibility of a totally new way of
environmental sensing using smartphones.
Table 1 shows the current models of smartphones
with environmental sensors embedded, obtained from
screening in smartphone-specialized websites
databases. The first conclusion is that environmental
sensors were hugely deployed in 2013 by Samsung,
but they not maintained the trend to the current days
(some of these sensors were not included in newer
models). The only environmental sensor that stills
currently being embedded in a considerable
percentage of devices is the barometric sensor,
probably due to its function in altitude positioning.
This is corroborated by the data extracted from the
Open Signal crowd sensing campaign for Android
devices, where it is possible to see that
environmental-enabled smartphones in activity
decreased in number from 2014 to 2015 (Open
Signal, 2015).
Bearing this information, Table 2 depicts the
quantitative of devices listed in Table 1 that was seen
SENSORNETS 2017 - 6th International Conference on Sensor Networks
by Open Signal in its Android fragmentation study.
From more than 550 thousand devices seen in 2014,
and more than 540 thousand devices seen in 2015,
10% had environmental sensors in 2014, and 7% in
2015. However, as this data was mostly acquired in
European countries, it may not be representative of
worldwide distributions mainly because of
heterogeneity of the worldwide market profiles.
Furthermore, according to GSMA Intelligence
(2015), the penetration of smartphones, per unique
subscriber, in Europe has reached 78.9% in 2014. In
absolute numbers, it is about 585 million people
carrying smartphones. Assuming that the
crowdsensing campaign led by Open Signal has an
acceptable margin of error in Europe resulting in a
good sampling of device diversity and fragmentation
in that region, and crossing this information (Table 2)
with the smartphones penetration given by the GSMA
Intelligence report, it is possible to estimate that there
were about 60 million smartphones with
environmental sensors in Europe in 2014, and about
42 million in 2015. If a linear trend is maintained, it
is expected about 25 million actives smartphones with
environmental sensors in Europe by 2016.
Despite the decay estimative of active
smartphones with environmental-ready sensors, it
still is a considerable absolute number of devices. It
justifies and reinforces our motivation and
investigation objectives. For comparison effect, the
Argos consortium for environmental monitoring and
oceanography has today about 22 thousand actives
transmitters around the world (Argos System, 2016),
and the Brazilian Environmental Data Collection
System has about 1 thousand active platforms through
Brazilian terrestrial and maritime boundaries
(Instituto Nacional de Pesquisas Espaciais, 2016).
Table 1: Smartphones with environmental sensors. (GSM
Arena, 2016).
Brand Model
Temp. Hum.
Galaxy S4
y y y
Galaxy Note 3 y y y
Galaxy J (N075T) y y y
Galaxy Round
y y y
Motorola Moto X (2
Gen.) y n y
Huawei Ascend P6 y n n
Xiaomi Mi3 y n y
Table 2: Smartphones with environmental sensors seen in
the Open Signal crowdsensing.
Number of Devices
Year: 2014 2015
Total number: 558770 542648
Samsung Galaxy S4 36903 24456
Samsung Galaxy Note 3 16603 12409
Motorola Moto X (2
Gen.) 3244 1448
Huawei Ascend P6 897 587
Xiaomi Mi 3 771 652
Devices with environmental sensors: 58418 39552
Towards the utilization of sensor-rich smartphones as
a centric data-collector element in environmental
sensing, we elaborated a sequence of sensibility tests
to investigate the behaviour of smartphone’s
environmental sensors under different situations that
are inherent to participatory sensing scenarios. The
analysis is done through comparison of datasets
generated by a reference and the subjects involved in
the experiences. Due to budget constraints,
availability reasons and due to the higher popularity
among the environmental sensing-enabled
smartphones (see Table 1), the chosen subjects was
the Samsung Galaxy S4 that carries the SHTC1 sensor
- from Sensirion - for temperature and relative
humidity, and one Motorola Moto X, which uses an
internal sensor to monitor the battery, in a similar
approach as the one reported in Overeem et al. (2013).
In this way, we will verify both types that the
temperature sensors can appear in smartphones.
As reference data, a pair of brand new AM2302
sensors were used. This sensor model has, nominally,
accuracy of ±1°C for temperature and ±2% for
relative humidity. The average value between
readings of both sensors was used as guide to
minimize discrepancies and enhance the accuracy.
The logging from smartphones was made through the
Android application “Telemetry”. The data
acquisition from the AM2302 was made using the
hardware platform Arduino, and for data-logging it
was used a computer running a Python script to read
– through USB – and store the data into a CSV file
with the proper timestamps in HH:mm:ss format.
The experiment was divided into 4 scenarios: idle
(for stability check), handling, dynamic and outdoors.
All tests took place in Natal, Brazil, and, all tests were
synchronized to the official local time and
Data Quality Issues in Environmental Sensing with Smartphones
parameters. Figure 1 depicts the devices involved in
this work. The details of each stage, and its respective
discussion, are given below.
Figure 1: Devices involved in this work.
3.1 Idle Scenario
The objective of “idle scenario” was to check the
stability of smartphone’s readings and its accuracy
when they are not being used during a long period.
For redundancy, the experiment was performed twice
in different days and different hours, obtaining the
highest possible number of samples and trying to
cover some environmental variation. Based on
information obtained from direct contact with the
Portuguese Institute of Ocean and Atmosphere
(IPMA, 2016), the sampling frequency was set to one
sample per minute.
The test was executed near a window in a room
with good air circulation, to get the most of
temperature and relative humidity from outside air.
Also, to ensure that the smartphones would not be
artificially warmed, there was no handling of the
devices during the experiment. In order to avoid
residual heat due to the battery charging process, the
measurement procedure started 30 minutes after the
complete charge of the smartphones.
Each measurement run lasted about 6 hours and
370 samples were collected. The readings from the
first run are shown in Figure 2 and Figure 3 for
temperature and humidity respectively. The readings
from the second run are shown in Figure 4 for
temperature and Figure 5 for humidity. The analysis
of the variation and deviations between the reference
sensor and smartphone responses was made using the
following statistics parameters: Maximum Absolute
Error (MxAE), Mean Absolute Error (MnAE), Root
Mean Square Error (RMSE) and Pearson product-
moment correlation (R). Table 3 and 4 illustrates this
analysis for temperature readings from Samsung S4
and Motorola Moto X, respectively, and Table 5 for
Samsung’s relative humidity readings (the Motorola
smartphone does not have a humidity sensor).
Figure 2: Temperature readings from the first run of Idle
Figure 3: Humidity readings from first the run of Idle
Figure 4: Temperature readings from the second run of Idle
Figure 5: Humidity readings from the second run of Idle
1:00 2:00 3:00 4:00 5:00 6:00
Temperature (°C)
Elapsed Time (HH:mm)
Reference Samsung S4
Motorola Moto X
1:00 2:00 3:00 4:00 5:00 6:00
Relative Humidity (%)
Elapsed Time (HH:mm)
Reference Samsung S4
01:00 02:00 03:00 04:00 05:00 06:00
Temperature (°C)
Elapsed Time (HH:mm)
Samsung S4
Motorola Moto X
01:00 02:00 03:00 04:00 05:00 06:00
Relative Humidity (%)
Elapsed Time (HH:mm)
Reference Samsung S4
SENSORNETS 2017 - 6th International Conference on Sensor Networks
Table 3: Statistical parameters obtained from Samsung's
temperature sensor compared to reference.
Parameter 1
run 2
MxAE 2.0°C 1,4°C
MnAE 1.39°C 0.84°C
RMSE 1.40°C 0.87°C
R 0.926 0.980
Table 4: Statistical parameters obtained from Motorola's
inner temperature sensor compared to reference.
Parameter 1
run 2
MxAE -5.3°C -4.8°C
MnAE -2.09°C -3.21°C
RMSE 2.14°C 3.23°C
R 0.780 0.967
Table 5: Statistical parameters from Samsung's humidity
sensor compared to reference, in percentage points.
Parameter 1
run 2
MxAE 10.3 % 16.5 %
MnAE 8.4 % 13,0 %
RMSE 8.5 % 13,3 %
R 0.587 0,895
From both visual and numerical analysis, we
observe promising results for temperature readings
from the Samsung Galaxy S4, and reasonable
readings from the Motorola Moto X. The S4 kept an
average error of 1.39°C in the first run, and less than
1°C in the second, always underestimating the true
temperature, but with a very high Pearson correlation
(0.926 and 0.980 for first and second run,
respectively). On the other hand, despite the constant
shift of Moto X, with the mean absolute error of about
2 and 3 degrees always overestimating the true
temperature, it kept its contour similar to the
reference, resulting in a strong positive Pearson
correlation of 0.780 and 0.967.
In humidity readings, we found indicators
suggesting poor data quality in both runs. The average
error found was high: 8.4 and 13 percentage points in
the first run; in the second run, even with higher
maximum absolute and average absolute errors (16.5
and 13 percentage points, respectively) than the first
run, it was found a good Pearson correlation
indicating that there may exist a systematic error with
this type of sensor, as, for example, a non-linear
response along the operating range or response time
3.2 Handling Scenario
The handling scenario objective was to verify the
response time from smartphone’s temperature and
humidity sensors. Was assumed that the heat from
hands and legs affect the sensor readings, and this test
aims to verify how much it occurs.
This test was performed by submitting the
smartphone to common situations whilst it sensor
stores the measurements each 10 seconds. In addition
to the idle situation, three handling cases were
considered: simple handling (one-handed; emulating
reading or texting), dual handling (two hands on;
emulating the usage of hardware capabilities, e.g.
running a game), inside leg pocket (simulating the
common storage and transportation during daily tasks
or walking). The reference sensors were put together
in the same place the measurements were performed,
as closest as possible.
Each situation lasted approximately 3 minutes,
with an idle situation gap of 30 seconds between each
case due to handling and positioning process, and also
to verify the cooldown time. To double check the
behaviour of the devices under these conditions, this
scenario was performed twice. The first run time
series is depicted in Figure 5 and the second run in
Figure 6. The results met the expectations.
In the first run, the ambient temperature was
stable around 28°C, and relative humidity around
90%. It was observed that handling the Samsung
smartphone can raise its temperature readings up to
4°C from its initial value, whether for utilization with
one hand or two (0:01 to 0:04; and 0:05 to 0:08,
respectively), and about 1.5°C when kept in the
pocket (from about 0:08:30). The Motorola Moto X
did not suffer the same amount of interference mainly
because the nature of its sensor, but it also raised 4°C
from its initial value, but always above the reference,
reaching 9°C of difference in dual-hand utilization.
The relative humidity measured from the Samsung
raised from 69% to 82% after the two hands
utilization, suggesting there is also interference on
humidity readings.
The second run was deployed with ambient
temperature of 27°C and 90% of relative humidity. It
was observed a maximum increase of temperature of
5°C for Samsung (at 0:05) and Motorola (at 0:07),
suggesting that, regardless of the nature of the
embedded sensor (either external or internal), the use
of the smartphone causes the same amount of
pollution on temperature measurements.
Data Quality Issues in Environmental Sensing with Smartphones
Figure 5: Readings from first run of handling scenario.
Figure 6: Readings from second run of handling scenario.
3.3 Dynamic Scenario
The dynamic scenario objective is to verify, as a
complement to the previous experiment, the dynamic
of smartphones sensors, and also to obtain a
correlation, quantifying the accuracy of a mobile
phone as an environmental data collector, and
illustrating even more our evaluation protocol.
This test was performed by submitting each
smartphone and the reference sensors to artificial
variations of temperature. This scenario is divided
into three stages: idle at room ambiance (2 minutes);
heating by a heat source (hair dryer) at a safe distance
(2 minutes); cooldown inside a fridge (4 minutes).
The process is repeated three times changing the
power of the dryer and the intensity of the fridge
trying to distribute the readings equally over the
range. As we do not have access to appropriate
equipment to change the humidity without heating the
sensors, the humidity dynamics will be presented as a
time series to be compared to the reference values.
The sampling frequency set for this experiment was
10 samples per minute.
The temperature result for the Samsung Galaxy
S4 is shown in Figure 7, and for the Motorola Moto
X in Figure 8. For a better visual analysis, both scatter
graphs contain an upward vertical line in grey
indicating where the ideal coefficient of
determination (R² = 1.00) should be, and a dotted line
in black indicating the trend line of the coefficient of
determination (R
) achieved.
The temperature readings for the Samsung
showed acceptable values, considering the
smartphone was idle (as explored in 3.1). The
maximum amplitude (temperature variation)
observed was 40°C by the sensor, and 35°C by the
smartphone. From the numerical analysis of this
dataset, it was observed very strong positive
indicators: Spearman’s Correlation (ρ) of 0.975 and
coefficient of determination (R²) of 0.927;
corroborating the information observed in section 3.1.
On can thus conclude that the external sensor of this
smartphone is reliable when the device is not being
0:00 0:01 0:02 0:03 0:04 0:05 0:06 0:07 0:08 0:09 0:10
Relative Humidity (%)
Temperature (°C)
Elapsed Time (H:mm)
One han
Two hands
0:00 0:01 0:02 0:03 0:04 0:05 0:06 0:07 0:08 0:09 0:10
Relative Humidity (%)
Temperature (°C)
Elapsed Time (H:mm)
One hand
Two hands
SENSORNETS 2017 - 6th International Conference on Sensor Networks
Figure 7: Samsung’s S4 scatter plot for readings in dynamic
experiment for temperature.
Figure 8: Motorola's Moto X scatter plot for readings in
dynamic experiment for temperature.
On the other hand, the results obtained for the
Moto X do not show the same accuracy as the
Samsung’s did. It is important to highlight that this
version of Motorola Moto X is using an internal
sensor in the battery, and that different behaviours
were expected between the two smartphones. The
maximum amplitude of temperature observed by the
Moto X was 14°C. The numerical analysis of the
dataset generated by Motorola’s sensor provided a
moderate positive correlation (R
) of 0.411 and a
“moderate-strong” Spearman Coefficient (ρ) of
As the humidity experiment was not performed
using the proper method, the results for this parameter
are presented only for reference, and are a
consequence to the temperature sweep. A time series
to visual analysis of the behaviour of the Samsung’s
Humidity module under the dynamic variations is
shown in Figure 9.
Figure 9: Humidity timeseries from dynamic scenario.
By observing the time series for humidity, the
poor data quality indicative observed in section 3.1
becomes more evident, reinforcing that the embedded
humidity sensor has some peculiarities. There is both
a shifted and inaccurate reading from the
smartphone’s sensor. While the reference sensor
changed from 97% (ambient humidity at the time the
experiment was performed) to 18% during the test,
the smartphone sensor changed from 80% to 30%.
This suggests that this sensor has a higher inertia
when compared to the temperature module, and
consequently requires much more time to reach the
reference value.
3.4 Outdoor Test
The outdoor test was designed to simulate a mobile
sensing node inserted into a participatory sensing
campaign where people are carrying their devices to
different places while it collects data samples. Also,
it is expected to quantify how much the GPS function
can pollute the measurements due to the increase on
battery and hardware usage.
This test consists in keeping the smartphones
continuous logging the temperature (and humidity,
when possible) while also collecting GPS coordinates
and time stamps during an outdoor walk. These
parameters would be the data used in a sensing
campaign with space and time granularity focused on
urban sensing or environmental monitoring using
smartphones. Thus, trying to cover the different ways
people can carry their devices, we made two opposite
R² = 0,9266
10 20 30 40 50 60
Samsung's S4 Temperature (°C)
Reference Temperature (°C)
R² = 0,4112
10 20 30 40 50 60
Motorola Moto X Temperature (°C)
Reference Temperature (°C)
0:00 0:03 0:06 0:09 0:12 0:15 0:18 0:21
Relative Humidity (%)
Elapsed Time
Reference Humidity S4's Humidity
Data Quality Issues in Environmental Sensing with Smartphones
situations: one measurement run during a 20-minute
walk carrying the smartphone in a backpack, exposed
to the air, with no user intervention (idle); and a 20-
minute walk with the smartphone in the leg pocket
(getting some heat from body). This last situation was
thought to verify if the heat transfer from the body
observed in section 3.2 can be amplified in the walk
process and GPS usage. The walk took place along a
residential area in Natal, Brazil, during the night of 25
of September. Due to the expected lower variations of
the humidity and temperature of this scenario, the
sampling frequency was set to 2 samples per minute.
As reference, official data provided by a
meteorological entity in the moment the test was
performed was used, as illustrated in Figure 10.
Figure 10: Temperature and Humidity observed for Natal,
Brazil, at the time this scenario was performed (The
Weather Channel, 2016).
The obtained results are shown in Table 6, where
it is possible to see the average temperature and
humidity measured in each situation compared with
the official air parameters.
Table 6: Obtained results from outdoor scenario.
Smartphone Situation
Temperature (°C)
Backpack 24.4
Leg Pocket 27.2
Backpack 30.2
Leg Pocket 32.2
Relative Humidity (%)
Backpack 82.5
Leg pocket 77.9
Note that there were acceptable results for
Samsung S4 readings when in the backpack, both for
temperature (+2.4°C error) and humidity (-0.5%
error) considering that the official temperature was
extracted in a different point of the city, or through
the average of multiple observation points, or even
with a different sampling rate. The Moto X
temperature readings were, again, always above the
reference value due to its higher battery consumption,
and consequent heating when its GPS function is
enabled, and also due to its sensor placement inside
the smartphone.
When inside the pocket, as expected, the heat
from legs was transferred to the smartphones and
polluted the measurements. In numbers, the average
temperature increased by 2.8°C for the Samsung S4,
and 2°C degrees for the Moto X, when compared to
the backpack situation. When compared to the
reference given, the temperature was increased by
5.2°C for Samsung S4, and 10.2°C for Motorola
Moto X.
In this work we evaluated the potential of using
smartphones in environmental monitoring through
participatory sensing given that the Samsung and
other manufacturers started to embed these sensors in
their products a few years ago. From our analysis, two
essential aspects that affect the implementation of
useful environmental sensing campaigns using
smartphones could be highlighted: quantity of
devices to cover urban spaces entirely (high space and
time granularity); and data quality to properly and
accurately monitor the environment. Each one
deserves the proper efforts to elucidate, enumerate
and overcome the challenges and difficulties.
Regarding the quantitative of devices, we
analysed the available data from market specialized
websites and the crowdsourcing project led by Open
Signal, and we found that there is a soft downward
trend on the utilization of smartphones with built-in
environmental sensors, and a lack of new models
carrying these sensors. Thus, there is an industry
dependence that hinders the geographic spread of
these devices, and consequently, makes the
engagement of users a more challenging task due to
the reduction of the target people, and to the reduction
of active devices with environmental sensors in urban
spaces. In addition, we have not found any
information that justifies the reason why
manufacturers have stopped putting such sensors in
their newest models.
Regarding the quality of collected data, we have
found that smartphones can collect acceptable
readings for temperature when idle, but the utilization
of the device pollutes the measurements by virtue of
heat transfer from hands and by the hardware natural
warm up from battery, CPU and GPU activity. A
context-aware application to identify if the
smartphone is being handled could potentially
SENSORNETS 2017 - 6th International Conference on Sensor Networks
overcome some of these limitations. For example, by
monitoring the CPU usage, lux meter, accelerometer
and gyroscope data, it would be possible to detect if
the smartphone is idle or not and then trigger the
sensor logging. On the other hand, smartphones
equipped with temperature sensors inside batteries
requires a much more sophisticated context-aware
detection and temperature estimation process because
there is a constant power transfer causing a natural
Yet on quality aspects, we also concluded that
humidity sensors provides inaccurate measurements
even when the device was idle. The position of the
sensor in smartphone’s hardware probably creates a
shield effect, making it difficult to detect the outside
air and quick changes in it, and in the environmental
monitoring a fast time response is essential.
Therefore, to date we concluded that the existing
smartphones are not ready yet to act as discrete,
autonomous, complete and user-centric data-
collectors in participatory sensing campaigns using
embedded environmental sensors, or battery sensors,
without any “data-treatment”, mainly due to the
observed issues in data quality when the devices are
being handled or used. There is a need to a
complementary application to estimate the context
the smartphone is inserted into, and that is not
guaranteed to work equally for all models.
Nevertheless, there are other roles in environmental
sensing that smartphones can play reliably, acting as
supporting devices. There are satisfactory results
from studies using smartphones as data mules or as
data transmitters from peripheral sensors, for example
the work reported by Tong & Ngai (2012) and by Park
& Heidemann (2011).
Considering that the proven ubiquity of
smartphones makes them a great instrument for
crowdsensing; that micro sensors are neither
expensive nor drain too much battery; that the
evolution of MEMS technologies behind these
sensors are enabling the development of more
accurate devices; and considering the results we
obtained from our experiments, it would be positive
to see manufacturers integrating these sensors again
in their future smartphones. It would enable a totally
pervasive and friendly perspective for environmental
monitoring through participatory and collaborative
With this work we hope to have contributed with
“first step” information for future developers who
plan to develop participatory sensing applications –
and campaigns – focused on environmental
monitoring or on observation of urban climate
phenomena using smartphones, despite the budget
constraints that limited the subject smartphones
utilized in the experiments. As a work in progress,
and as a consequence of this report, we are evaluating
the performance and data quality issues of low-cost
sensors often used in urban environment monitoring
systems by “DIY” initiatives.
This work has been supported by COMPETE: POCI-
01-0145-FEDER-007043 and FCT – Fundação para a
Ciência e Tecnologia within the Project Scope:
Argos System, 2016. Satellite Tracking for Environmental
Applications. Available at: http://www.argos-
CLS_ARGOS-HD.pdf [Accessed June 27, 2016].
D’Hondt, E., Stevens, M. & Jacobs, A., 2013. Participatory
noise mapping works! An evaluation of participatory
sensing as an alternative to standard techniques for
environmental monitoring. Pervasive and Mobile
Computing, 9(5), pp.681–694. Available at:
12001137 [Accessed April 3, 2014].
eMarketer Inc., 2015. Mobile Commerce Roundup.
Available at:
eter_Mobile_Commerce_Roundup.pdf [Accessed June
16, 2016].
GSM Arena, 2016. Phone Finder. Available at:
re=selected [Accessed October 10, 2016].
GSMA Intelligence, 2015. Global Mobile Economy Report
(2015), Available at:
_Mobile_Economy_Report_2015.pdf [Accessed June
27, 2016].
Hu, K. et al., 2016. Design and Evaluation of a Metropolitan
Air Pollution Sensing System. IEEE Sensors Journal,
16(5), pp.1448–1459.
iFixit, 2013. Samsung Galaxy S4 Teardown. Available at:
Teardown/13947 [Accessed June 20, 2016].
Instituto Nacional de Pesquisas Espaciais, Sistema
Integrado de Dados Ambientais. Available at:
p [Accessed June 27, 2016].
IPMA, 2016. Instituto Português do Mar e da Atmosfera.
Available at:
[Accessed October 1, 2016].
Kanhere, S.S., 2011. Participatory Sensing: Crowdsourcing
Data from Mobile Smartphones in Urban Spaces. 2011
Data Quality Issues in Environmental Sensing with Smartphones
IEEE 12th International Conference on Mobile Data
Management, pp.3–6. Available at:
arnumber=6068482 [Accessed March 30, 2014].
Kotovirta, V. et al., 2012. Participatory Sensing in
Environmental Monitoring -- Experiences. 2012 Sixth
International Conference on Innovative Mobile and
Internet Services in Ubiquitous Computing, pp.155–
162. Available at:
arnumber=6296848 [Accessed April 3, 2014].
Open Signal, 2015. Android Fragmentation Visualized
(August, 2015). Available at:
mentation_report.pdf [Accessed June 24, 2016].
Overeem, A. et al., 2013. Crowdsourcing urban air
temperatures from smartphone battery temperatures.
Geophysical Research Letters, 40(15), pp.4081–4085.
Available at:
[Accessed March 28, 2014].
Park, U. & Heidemann, J., 2011. Data muling with mobile
phones for sensornets. Proceedings of the 9th ACM
Sensys Conference, pp.162–175. Available at:
Rodrigues, H. et al., 2012. Engaging participants for
collaborative sensing of human mobility. UbiComp’12,
The Weather Channel, 2016. Natal, Brazil Weather
Forecast and Conditions - The Weather Channel.
Available at:
[Accessed September 26, 2016].
Tong, X. & Ngai, E.C.H., 2012. A ubiquitous
publish/subscribe platform for wireless sensor networks
with mobile mules. Proceedings - IEEE International
Conference on Distributed Computing in Sensor
Systems, DCOSS 2012, pp.99–108.
Zaman, J. et al., 2014. Citizen-friendly participatory
campaign support. 2014 IEEE International
Conference on Pervasive Computing and
Communication Workshops (PERCOM
WORKSHOPS), pp.232–235. Available at:
SENSORNETS 2017 - 6th International Conference on Sensor Networks