Data Quality Issues in Environmental Sensing with Smartphones

Tiago C. de Araújo

1,2

, 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.

1 INTRODUCTION

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 AraÃžjo 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

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

issues.

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

issues.

2 ENVIRONMENTAL ENABLED

SMARTPHONES

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

Sensors

Temp. Hum.

Press

Samsung

Galaxy S4

(i9500/9505)

y y y

Galaxy Note 3 y y y

Galaxy J (N075T) y y y

Galaxy Round

(G910S)

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

Seen

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

3 ACCURACY TESTS

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

Scenario.

Figure 3: Humidity readings from first the run of Idle

Scenario.

Figure 4: Temperature readings from the second run of Idle

Scenario.

Figure 5: Humidity readings from the second run of Idle

Scenario.

1:00 2:00 3:00 4:00 5:00 6:00

Temperature (°C)

Elapsed Time (HH:mm)

Reference Samsung S4

Motorola Moto X

100

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)

Reference

Samsung S4

Motorola Moto X

100

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

run

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

run

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

run

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

issues.

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

handled.

100

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)

Ref.Temp.

S4.Temp.

MotoX.Temp

Ref.Humd

S4.Humd.

One han

Two hands

Pocket

100

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)

Ref.Temp.

S4.Temp.

MotoX.Temp

Ref.Humd

S4.Humd.

One hand

Two hands

Pocket

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

0.654.

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)

100

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

Average

Measured

Official

Temperature (°C)

Backpack 24.4

Leg Pocket 27.2

MotoX

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.

4 DISCUSSION AND

CONCLUSIONS

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

warming.

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

sensing.

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.

ACKNOWLEDGMENTS

This work has been supported by COMPETE: POCI-

01-0145-FEDER-007043 and FCT – Fundação para a

Ciência e Tecnologia within the Project Scope:

UID/CEC/00319/2013.

REFERENCES

Argos System, 2016. Satellite Tracking for Environmental

Applications. Available at: http://www.argos-

system.org/files/Publication/public/r1679_f34_3606-

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:

http://linkinghub.elsevier.com/retrieve/pii/S15741192

12001137 [Accessed April 3, 2014].

eMarketer Inc., 2015. Mobile Commerce Roundup.

Available at:

https://www.emarketer.com/public_media/docs/eMark

eter_Mobile_Commerce_Roundup.pdf [Accessed June

16, 2016].

GSM Arena, 2016. Phone Finder. Available at:

http://www.gsmarena.com/results.php3?chkTemperatu

re=selected [Accessed October 10, 2016].

GSMA Intelligence, 2015. Global Mobile Economy Report

(2015), Available at:

http://www.gsmamobileeconomy.com/GSMA_Global

_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:

https://pt.ifixit.com/Teardown/Samsung+Galaxy+S4+

Teardown/13947 [Accessed June 20, 2016].

Instituto Nacional de Pesquisas Espaciais, Sistema

Integrado de Dados Ambientais. Available at:

http://sinda.crn2.inpe.br/PCD/SITE/novo/site/index.ph

p [Accessed June 27, 2016].

IPMA, 2016. Instituto Português do Mar e da Atmosfera.

Available at: https://www.ipma.pt/pt/index.html

[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:

http://ieeexplore.ieee.org/lpdocs/epic03/wrapper.htm?

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:

http://ieeexplore.ieee.org/lpdocs/epic03/wrapper.htm?

arnumber=6296848 [Accessed April 3, 2014].

Open Signal, 2015. Android Fragmentation Visualized

(August, 2015). Available at:

http://opensignal.com/assets/pdf/reports/2015_08_frag

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: http://doi.wiley.com/10.1002/grl.50786

[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:

http://dl.acm.org/citation.cfm?id=2070960.

Rodrigues, H. et al., 2012. Engaging participants for

collaborative sensing of human mobility. UbiComp’12,

pp.5–8.

The Weather Channel, 2016. Natal, Brazil Weather

Forecast and Conditions - The Weather Channel.

Available at:

https://weather.com/en/clima/hoje/l/BRXX0158:1:BR

[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:

http://ieeexplore.ieee.org/lpdocs/epic03/wrapper.htm?

arnumber=6815208.

SENSORNETS 2017 - 6th International Conference on Sensor Networks