=Paper=
{{Paper
|id=Vol-1630/paper1
|storemode=property
|title=Semantic Data Layers In Air Quality Monitoring For Smarter Cities
|pdfUrl=https://ceur-ws.org/Vol-1630/paper1.pdf
|volume=Vol-1630
|authors=Jean-Paul Calbimonte,Julien Eberle,Karl Aberer
|dblpUrl=https://dblp.org/rec/conf/semweb/CalbimonteEA15
}}
==Semantic Data Layers In Air Quality Monitoring For Smarter Cities==
https://ceur-ws.org/Vol-1630/paper1.pdf
Semantic Data Layers in Air Quality Monitoring
for Smarter Cities
Jean-Paul Calbimonte, Julien Eberle, and Karl Aberer
EPFL, Switzerland
{name.surname}@epfl.ch
Abstract. Air pollution is one of the key indicators for quality of life
in urban environments, and is also the subject of global health concern,
given the number of mortal diseases associated to exposure to pollutants.
Assessing and monitoring air quality is an important step in order to bet-
ter understand the impact of pollution on the health of the population.
Nevertheless, in order to scale to the city level, traditional high-quality
stationary sensing stations are not enough. Limitations include lack of
coverage, the cost of deployment and maintenance, as well as the res-
olution of the observed phenomena. The OpenSense2 project aims at
providing a city-level sensing deployment that combines different lev-
els of air quality sensing: reference stations, mobile sensing on public
transportation, and participatory crowdsensing. In this paper we high-
light some of the key challenges of managing the data captured by such
infrastructure, taking the city of Lausanne as a driving use-case. Fur-
thermore, we present a semantics-based approach for characterizing and
exposing the air quality data, so that it can be made available to citizens
and application developers in a way that it can be usable and understood
effectively.
Keywords: Sensors, Internet of Things, Semantic sensors, Semantic an-
notation
1 Introduction
Cities can be seen as live systems that act and react to internal and external
stimuli, as complex entities that have distinct behaviors and characteristics that
make them unique. Citizens are one of the driving internal components of a city,
although not the only ones. They share a physical, geographical and environ-
mental space with natural elements such as flora & fauna, with environmental
conditions, with artificial entities such as traffic, industry, agriculture, etc. In
these complex systems, it is important to assess or monitor the overall health
condition of a city. Being such unique entities, cities are sometimes fragile and
vulnerable to the multiple factors and characteristics that surround them. One
of the most important conditions that measure the health of a city is related to
the pollution levels, which may include soil contamination, acoustic and water
pollution, littering, air pollution, radioactive contamination, etc. In particular,
air pollution is one of the most important and most studied of these, given its
direct impact, not only on the health and lifestyle of citizens, but also on the
rest of the components of an urban settlement [17]. In fact, it is estimated that
around 7 million premature deaths are attributable to air pollution [12] world-
wide. Cardiopulmonary, cardiovascular and ischemic heart diseases have been
linked to sustained exposure to certain air pollutants in urban environments,
making it a global health concern.
Understanding how the air pollutants are produced and how they disperse
in a certain urban area is key to be able to explain the resulting air quality con-
ditions. Different tools have been studied to cope with this challenge, including
dispersion models, pollution map interpolation models, regression models, etc.
These require a set of input data in order to be generated, including accurate ref-
erence air quality measurements, pollutant sources, city street city models, wind
direction, temperature and other weather data, among others. However, it has
been evidenced that air pollution conditions present highly localized patterns,
which can vary greatly from one street to the next one. Moreover, coverage of
city-wide air quality observations can be impractical and unfeasible using stan-
dard stations, as it would be prohibitive in terms of costs, and also because of
mobility restrictions.
The OpenSense1 project aims at integrating air quality measurements cap-
tured by heterogeneous mobile and crowdsensed data sources, in order to under-
stand the health impacts of urban air pollution exposure, and providing high-
resolution air quality maps. In this paper we focus on the data management chal-
lenges of such an infrastructure, and how semantic representations and models
can help defining different layers of air quality data provision. We argue that
this approach can help bridging the gap between the sensed raw data and the
understanding of air quality conditions over space and time in a city. We present
the progress achieved so far, focusing on the deployment in the city of Lausanne,
where the platform is already running for about two years.
The remainder of the paper is structured as follows. First we describe the
overall approach of OpenSense in Section 1. In section 2 we provide more concrete
details about the Lausanne deployment. Section 3 provides a discussion about
the use of semantics in OpensSense, while Section 4 describes more concretely
the data management support for health studies. Section 5 presents some related
work before concluding.
2 OpenSense Air Quality Monitoring
Ambient air quality in the context of a city can be assessed in function of different
pollutants present in the air. Different air pollutants are emitted and produced
by different sources and have distinct characteristics, and of consequently differ-
ent ways of being detected. The most commonly targeted pollutants, according
to competent bodies such as the European Environment Agency include: Carbon
1
OpenSense project: http://opensense.epfl.ch
monoxide (CO), Nitrogen dioxide (NO2), Nitrogen monoxide (NO), Sulfur diox-
ide (SO2), Ground level Ozone (O3), Particulate matter (PM) and Lead (Pb).
CO is produced by incomplete combustion processes, including transportation,
industrial and household combustion. NO2 and NO (or NOx) are mainly emit-
ted by the transportation sector. This is also one of the main sources of PM, as
well as heating and industrial processes. PM exists in different ranges of sizes,
e.g. 10, 2.5 nm and ultrafine particules. O3 is formed by reactions of other pol-
lutants such as NOx and volatile organic compounds. Considering the relatively
low levels of Lead and SO2 in the targeted cities of the OpenSense project, we
do not consider them for the air quality monitoring deployment, and focus on
CO, NO2, O3 and PM, which are also relevant for the health studies that are
carried out simultaneously.
Air quality sensing in OpenSense envisions a hybrid and heterogeneous sens-
ing environment (see Figure 1) where reference stationary stations, mobile sens-
ing devices on public transportation, and crowdsensing, contribute to an overall
city view of the status of air quality conditions. This combination has well iden-
tified advantages, from different points of view:
Accuracy Depending on the characteristics of the sensing device one may ob-
tain more or less accurate results. A reference station typically complies
with well defined data quality requirements and produces high fidelity ob-
servations. More affordable but less accurate equipment can be mounted on
public transportation, still providing acceptable data accuracy. Crowdsens-
ing devices need to be considerably cheaper and consequently less accurate.
However, the number of participants can compensate this factor, and still
provide indicative observation measurements.
Coverage Clearly, the public transportation vehicles provide the means to reach
the main streets and avenues of the city, covering most of the urban geogra-
phy. However, for more fine-grained measurements and less accessible places,
we need to rely on the crowdsensing platform.
Maintenance & Costs Reference stations need expensive set-up and super-
vision. The mobile sensors on public transportation have the advantage of
having their own power supply and virtually zero transportation costs. Main-
tenance is still necessary, although components are less expensive and easier
to replace if necessary.
Reliability Problems in one node or in one layer of the sensing platform can fall
back on the other layer, providing a resilient infrastructure that is flexible
in case of unexpected events.
The data collection from the sensors produces time series of the air pollu-
tant measurements, which are only the first step in the OpenSense pipeline. The
resulting dataset consists of raw observations that cannot be directly used by
citizens and external applications. To understand the semantics of this data,
additional processing is required, exploiting first the spatial characteristics of
the data points. Given the highly localized nature of air pollutant concentra-
tions (due to physical behavior characterized in models such as street canyons),
Fig. 1: The OpenSense2 approach for Air Quality monitoring in Smart Cities: different planes pro-
vide data at different granularities: raw observations, spatio-temporal aggregations, and air quality
products and applications, including air pollution maps [9], models [18], mobile recommendations,
etc.
we need to project the observations to street segments and perform spatial ag-
gregation and interpolation. Similarly, in terms of time, we may be interested
in different time granularity (e.g. hourly, weekly, monthly) and consider that
the sampling can be very irregular in case of crowdsensing. Once the spatio-
temporal distributions have been computed and made available, pollution maps
can be generated, complemented or validated with this data, leading to more
advanced air quality models: e.g. log-linear regression models, lagrangian dis-
persion models, etc. Finally, using these models, end-user applications can be
built, leveraging on the available processed datasets. Applications include per-
sonalized recommendations for reducing the exposure to pollutants, or support
for health studies related to diseases associated to air pollution.
In this context, multiple challenges arise to make this smart city deploy-
ment possible. In particular, we set the scope of the analysis around the data
management issues:
Heterogeneity Different types of stationary and mobile sensors collaborate in
the platform, as described above. Hence the need for high-level representa-
tions of the meta-information that is collected: e.g. time. location, device
characteristics, units of measurement, etc., as well as semantic information
of the observation themselves: e.g. raw CO2 measurement, aggregated CO2
average over a certain location, etc.
Provenance Given that different layers of data abstraction can be constructed
(i.e. from raw measurements to high-level concepts such as spatio-temporal
air quality conditions), it is important to keep trace of the processing ac-
tions or steps that were taken to get a certain data product. This can help
reproducing results or health studies, as well as understanding the datasets
that led to those results.
Mobile Data quality For the crowdsensing scenarios and the health studies,
people are expected to use mobile phones to collect sensed data, and contex-
tual information such as accelerometer or GPS coordinates to detect their
activity and compute their exposure. However the data collected may con-
tain outliers and faulty measurements given the low guarantees on data qual-
ity. Nevertheless, peer correlation and self-correlation can help understand
the value of certain sections of the captured time series. Explicit semantic
representations of data quality and the measurement context, such as in-
door/outdoor, can help applications taking decisions about whether to use
or not certain pieces of data contributed by mobile phones.
Privacy Participatory sensing helps building collective knowledge, but this comes
sometimes at the cost of private data disclosure. A clear example in OpenSense
is related to location privacy, as participants continuously disclose where they
are. It is possible to use obfuscation and hiding techniques given the semantic
and geographic sensitivity of a certain place, not compromising substantially
the quality of the service.
Energy consumption The crowdsensing scenarios may require smart switch-
ing between public transportation sensors, and personal mobile sensors.
Given that the latter are more vulnerable to battery depletion, they can
be used only if the public network is not available in the surrounding area,
or if other peers are not available either.
Urban air quality requires a complex deployment of sensors, as the one we
propose in OpenSense. This setting allows not only monitoring the concentration
of pollutants, but also letting this information be visible and accessible publicly,
so that citizens can learn and adapt their behavior to the measured conditions.
Moreover, the health studies allow the city to self-diagnose and provide recom-
mendations to the crowdsensing participants. Smart cities should be able, not
only to monitor and observe the events and conditions that surround it, i.e.
in this case the quality of the air, but also should be able to evaluate these
conditions and act upon the results.
3 OpenSense Lausanne Deployment
In this section we describe the deployment of the OpenSense platform in the
city of Lausanne. Ten buses have been equipped with electrochemical sensors
for CO and NO2, a resistive O3 sensor, and a Naneos Partector for particulate
matter. The buses cover different areas of the city, ranging from the lake-side to
the city center and the northern suburbs. A bus may change its line and path
depending on days of the week, peak hours and vacation periods. The sensor
boxes collect the measurements and send them via GPRS communication to a
base station every five minutes. The base station runs an instance of GSN [1]
(Global Sensor Networks), which is devoted to manage the data life-cycle of the
sensor observations.
The sensors deployed on the buses constitute the backbone of the smart city
air quality monitoring platform. The reliability of the sensors on these buses,
Fig. 2: OpenSense Lausanne deployment: 10 sensor boxes installed on top of buses of the local
transportation network, are linked to GSN via GPRS communication. Sensor boxes report CO,
NOx, O3 , CO2 and UFP measurements.
as well as their number and sampling frequency offer minimal conditions for
providing relatively good data quality, at least in the main streets of the city.
However, for the locations where buses are not accessible, OpenSense needs to
rely on other sources of data. One is a computed GRAMM/GRAL model2 that
allows simulating the air flow in complex urban terrain as well as the dispersion of
air pollutants at building scales. Using these models it is possible to approximate
concentration levels of NOx and PM10 hourly at 5 m resolution. Nevertheless, the
pollution maps generated with these models do not take into account unforeseen
events or anomalies that are not captured in the model. If this is the case, then
the OpenSense Lausanne deployment needs to use the data from participatory
sensing mobile devices. As part of this platform, we have developed tinyGSN [5],
an android application that allows mobile phones to collect sensor data from
built-in or external sensors, and -if necessary- report to a central GSN instance.
With tinyGSN we are able to provide a common data acquisition, storage and
processing environment for mobile and participatory sensing.
3.1 GSN Data Processing and Access
We use the GSN (Global Sensor Networks) middleware as the core back-end
server for stream processing and data publishing. The inherently decentralized
system architecture of GSN allows different instances to co-exist in a distributed
deployment, and each instance can expose a number of different virtual sensors.
In each GSN instance, a pool of deployed virtual sensors is administered by the
virtual sensor manager. This includes handling the lifecycle of a virtual sensor
and managing the incoming streams provided through a wrapper. Wrappers are
the interfaces between physical sensing devices and a virtual sensor [4].
GSN provides already a set of ready-to-use wrappers (e.g. UDP, serial port,
HTTP, etc.) while it is easily extensible if new data sources are needed. The
streams produced by each virtual sensor have an output structure composed
of one or more fields, which can be defined in terms of a continuous query
running over one or more sources. Once the data has been acquired, it can be
2
Lausanne GRAMM/GRAL: http://www.empa.ch/plugin/template/empa/*/159292
Fig. 3: GSN distributed architecture: different GSN instances may communicate remotely through
inter-GSN communication. Each GSN instance may use different wrappers for data acquisition, and
processing classes for additional data operations.
processed, as specified in the corresponding GSN processing class. Then it can
be stored, in persistent or temporary storage resources, depending on the virtual
sensor configuration parameters. Next, at the query manager layer, the system
can evaluate running queries, acting directly on the streams produced at the
lower layers. The query capabilities are exposed through the service interfaces,
currently implemented as an HTTP RESTful interface that can be accessed by
external users and applications.
Each GSN instance can be accessed (and interact remotely with other in-
stances) through a native interface (inter-GSN communication) implemented on
top of a message queue, ØMQ (ZeroMQ)3 . This native interface helps providing
a wide range of possible deployment set-ups, such that GSN instances can be
located in different physical machines or data centers. Finally, an access layer on
top of the services allows defining permissions over the virtual sensors and the ob-
servations they produce. The system has been implemented in Java, while some
out-of-the box wrappers are implemented in other languages. The entire project
is open-source, and is available in Github, as a standalone project, with an ex-
isting and growing community of users and developers. The existing community
of developers and users inherited from GSN positions this software project as
one of the most comprehensive and extensible tools for IoT data management,
as it has been shown in several real life deployments and environmental scientific
research.
In the case of the Lausanne deployment, it presented a set of challenges that
we had to surpass to successfully deliver the measurements as a coherent and
queryable dataset. These challenges include: the particularities of the commu-
nication protocol with the bus sensor boxes, the data acquisition flow, and the
need for health monitoring of the system. The modularity and flexibility of GSN
contributed to address these issues satisfactorily. We have chosen a layered data
flow consisting of three main layers: the Bridge, Splitter and Join virtual sen-
sors, as shown on Figure 4. Communication among layers is operated through
3
ZeroMQ: http://zeromq.org/
the ZeroMQ wrapper, providing asynchronous communication, and data mes-
saging with other remote instances of GSN. Experiments on this architecture
have shown [4] that GSN can scale to sustain the load of low to mid-level rates
of stream elements per virtual sensor when using the ZeroMQ-based internal
communication system. We describe below the complete process in detail.
Fig. 4: OpenSense data flow for the Lausanne deployment: The Bridge partially parses the blocks
and sends them to the Splitter. At this point the data payload is parsed and later the Join adds the
geo-location information to each packet.
At the first level of the dataflow, the Bridge Virtual Sensor receives the
raw data packets from the bus sensor boxes through the Connector Wrapper.
Every 5 minutes a node initiates a connection to the GSN server through the
GSM network and sends the contents of its buffer. The buffer is emptied if the
transmission is successful, otherwise it is resent the next time with the additional
data collected during that time. The data is organized into packets, each of
which contains a partially ordered timestamp, the measurements, and a simple
checksum to ensure its integrity. At the beginning of each transmission, the
datetime special packets are sent, and then used to reconstruct the timestamps
in the wrapper. The packets are not completely parsed at this first step and
the payload that contains the actual measurements is stored as a byte array.
The virtual sensor responsible for this wrapper forwards the data, without any
processing, but has been modified to allow propagating back commands to the
wrapper.
The second layer is composed of processing classes that parse the different
packets contents, one for each packet type. Until this point, every group of sen-
sors (grouped in one packet type) has its own stream of measurement, such as
GPS, accelerometer or CO/NO2. The last layer takes care of geo-referencing the
measurements, by joining the stream from the GPS with each other stream. But
as sensors are not necessarily synchronized (some are push-based and other pull-
based) it is not trivial to join the streams. Therefore we need to interpolate the
GPS stream, more specifically the latitude, longitude, altitude and HDOP. For
this purpose we developed a Join virtual sensor that receives two streams and
maps one onto the other. Incoming data is kept in buffers until the timestamps
of both streams overlaps and in the case of large gaps in the GPS data, the mea-
surements are also discarded to avoid tagging with inaccurate locations. Apart
from the SMS alerts, the only way of getting information about the health status
of the sensing nodes is via the data transmissions. Except for the fine particle
sensor and the GSM link quality, the nodes are not reporting status information
to the server; therefore it needs to be inferred from the connection metadata.
Namely, at every transmission, GSN logs the number of packets received, the
checksum errors, the time needed to transmit all the data, and if it got inter-
rupted. Moreover, GSN has been instrumented in such a way that it can report
various metrics about the virtual sensors data rates and processing errors. An
external tool, namely Graphite, takes care of aggregating the metrics and pro-
ducing alerts whenever some issue is detected on a node or the GSN server.
4 OpenSense Data Semantics
The data monitoring capabilities of OpenSense would be fruitless if they do
not include the means to generate meaningful and semantically understandable
data. One of the key features of the OpenSense project is that it aims at offering
transparent access to the data that is produced, so that it can be reused as
feedback to citizens, hoping that they can take the necessary actions to avoid
health related issues associated with air pollutants. Without proper meaningful
information this goal cannot be effectively achieved. As an example, consider the
raw CSV data that GSN originally provides from the sensors in the buses:
# vsname:geo_osanm
# geographical:Lausanne, Switzerland
# description:Map GPS onto ANM
# time,station,co,no2,co2,latitude,longitude,altitude,hdop
2015-04-15T08:26:40.050+0200,392,245,930,486,46.535785403723,6.5665168648746475,395.82602,1.26
2015-04-15T08:26:41.050+0200,392,226,927,486,46.53578453126755,6.566519090658208,395.946,1.26
2015-04-15T08:26:41.050+0200,192,219,2659,480,46.53578453126755,6.566519090658208,395.946,1.26
2015-04-15T08:26:42.050+0200,392,241,939,486,46.53578366816291,6.566520951726367,396.066,1.26
Each reading is represented by a line in the CSV file, and as it can be seen,
the same table-like structure is used for all buses, differentiated by the station
column. Although some minimal metadata is provided in the CSV headers we
are missing important metadata (e.g. sensor details, measurements capabilities,
units, etc.), and more importantly, we lack any useful information about the
streets or avenues where data was collected, and in fact it is also mixed among
concurrently reporting buses (i.e. data from different buses may be interleaved).
In this form, the data provided is of little use, unless data pre-processing and
massaging is performed. It is precisely due to this fact that in OpenSense we
propose a layered semantic data management model, where at each layer we
provide additional value to the data, e.g. through semantic annotations that
describe data cleaning and pre-processing, temporal and spatial aggregations,
and finally event annotations, as depicted in Figure 5
Fig. 5: OpenSense layered processing: from raw measurements to spatial and temporal aggregation
and derivation of events. Raw data observation points are projected into bus line segments, then
aggregated over time. Spatial aggregation allow finding patterns over time in these segment regions.
In this example, the bottom layer corresponds to raw, data as it is output
from the GSN virtual sensors (notice that GSN itself has different data lay-
ers as well, this raw data is already more processed than the unparsed binary
payload that is received from the sensor boxes). At this level, each data point
still needs to be analyzed and aligned with the geographical representation of
the bus trajectory. the GPS location traces have a non-negligible error margin
that needs to be fixed (e.g. by projecting the point into the segment of the bus
line). This spatial processing may require even pruning erroneous values, and
interpolating if some values are missing in the trace. From this level, we can
derive temporal aggregations of the bus traces, considering that several buses
can traverse the same segment in one or more days, several times. This allows to
capture information such as when was the PM concentration at its peak during
this week? or What was the CO concentration at a particular time of the day.
Furthermore, spatio-temporal aggregations over these segment time series can
allow discovering patterns over time (e.g. daily, weekly peaks, summer trends,
weekend patterns, etc.), which can help better understanding the dynamic of air
pollution exposure in terms of time and space. These findings can be summarized
or transformed into particular events with a given duration and location. These
high level semantic representations can be linked successively to the underlying
data that originated it, using provenance annotations.
As an example of embedded metadata that can be provided for the OpenSense
data sets, consider the JSON snippet below. It represents a description of a GSN
CSV output of sensor data, that maps to RDF following the specifications of the
CSV on the Web Working Group4 .
{
"@context": ["http://www.w3.org/ns/csvw", {"@language": "en"}],
"url": "opensense.csv",
"tableSchema": {
"columns": [{
"name": "time","titles": "Time",
"aboutUrl": "#obs-{_row}",
"propertyUrl": "ssn:observationResultTime"
"datatype": {"base": "datetime","format": "yyyy-MM-ddTHH:mm" },
}, {
"name": "station","titles": "Bus sensor",
"aboutUrl": "#obs-{_row}",
"propertyUrl": "ssn:observedBy"
}, {
"name": "co","titles": "CO concentration",
"aboutUrl": "#obs-{_row}",
"propertyUrl": "ssn:observationResult"
}, {
"name": "type_event","virtual": true,
"aboutUrl": "#obs-{_row}",
"propertyUrl": "rdf:type",
"valueUrl": "opensense:CO_Observation"
}, {
"name": "unit","virtual": true,
"aboutUrl": "#obs-{_row}",
"propertyUrl": "qu:unit",
"valueUrl": "unit:mgm3"
}]
}
}
The example has been simplified for space and didactic reasons. It provides
an explicit description of the columns of an OpenSense CSV table, and describes
how RDF triples should be generated from this source. For example, the first
column time will generate a triple (an observation) that will be linked to the
date-time through the ssn:observationResultTime. Similarly the station will link to
the sensor that produced the observation, and the definition in the co column
will generate the relationship with the actual value. In addition, the CSV on
the Web specification allows defining virtual columns, which are useful to create
more than one subject per CSV row (e.g. for the units, and for the observation
type).
This solution results interesting for data consumers that are not necessarily
familiar with Semantic Web standards, as it comes in the form of a seemingly or-
dinary CSV. Nevertheless, thanks to the CSV-to-RDF standard transformations
encoded in the metadata, it is trivial to interpret the dataset as RDF triples.
An important issue in this respect is related to the capability of current RDF
triple stores to handle streams of very dynamic RDF data. While it is certainly
technically possible to store and archive data in such databases [4], it remains a
challenge to perform continuous processing and stream data querying over RDF
Streams. In this context, the ongoing work at the W3C RDF Stream Community
4
CSV on the Web http://www.w3.org/TR/csv2rdf/
Group5 , and the current prototypes that follow this new processing paradigm
can use the stream of data produced by the OpenSense deployment as an input
for more advanced continuous query processing.
For the other layers the datasets are not centered on point-in-time and point-
in-space observations, but rather on time series that span a trajectory in the
earth (e.g. bus traces), or that are aggregated over a certain time duration. In
both cases, the current capabilities of the SSN ontology remain limited. Semantic
descriptions on trajectories allow to perform advanced queries about air quality
in a certain region, taking into account that two streets can have entirely different
air circulation patterns, even if they are geographically close to each other.
5 Air Quality and Health Studies
Impact of air pollutants exposure on citizen’s health is one of the medium-long
term goals of OpenSense. However, it is often not trivial to carry out experi-
ments in this area, given the many technical difficulties of human sensing. We
have started to tackle some of the technical issues in data management and pro-
cessing for air pollution-related health studies, mainly on two fronts: the mobile
sensing platform, and the activity recognition mechanisms. The mobile sensing
platform, based on tinyGSN, provides the necessary building blocks to offer data
acquisition through customizable wrappers, communication and integration with
GSN nodes, and basic processing and cleaning.
Fig. 6: OpenSense location and accelerometer traces captured by tinyGSN. In the left Figure (a),
the location of a participant is plotted. Her trace allows to discern where she was and also how she
moved (means of transportation). In the right hand side Figure (b), different accelerometer traces
allow discovering basic activity patterns: stationary, walking, cycling, running, climbing stairs, etc.
In preliminary tests, we have used Android phones to capture location (Fig-
ure 6a) and accelerometer (Figure 6a) measurements on volunteers during a lim-
ited amount of time, and a controlled environment. TinyGSN has been used to
collect and process these traces, which are afterwards recuperated in a full GSN
instance. TinyGSN has been designed to efficiently use the battery resources it
5
W3C RSP Community Group: http://www.w3.org/community/rsp
has, while still providing the necessary rates of data. As it can be seen in the
example, users could be traced as they wander around the city of Lausanne,
and we are even able to characterize their activities based on the accelerometer
data and their location. Activity recognition is a key element to provide per-
sonalized recommendations based on air quality levels. For example exposure
while performing outdoor exercise may have totally different consequences than
stationary exposure. Logically, location and activity recognition can be consid-
ered as sensitive by a good number of users. To avoid unwanted disclosure, it is
possible to use obfuscation techniques that partially hide the user location while
preserving quality of service [2].
6 Related Work
Existing air quality sensing systems have focused on the device and sensor layers,
while disregarding the data management aspect [15]. In most of these systems
the semantics of the data is hidden or implicit in ad-hoc data schemas and data
source descriptions.
About semantic sensor data management, a good number of works has fo-
cused on Linked Data publishing, considering streams as the fundamental build-
ing block. More specifically, there have been projects focusing on different as-
pects related to sensing, including coastal flood monitoring [8], earth observation
remote sensing [11], or natural hazards [16].
Air Quality monitoring has been widely studied, and different approaches
have been used, ranging from sensors on bicycles [6, 13], vehicles [3, 10] and
hand-held devices [7]. Interactions between different types of sensors in this case,
remains an important question, and it is still needed to understand how different
types of sensor networks can be combined effectively. This becomes even more
complex in the context of crowdsensing, where incentives will need to play a
role [14], and malicious behavior needs to be countered.
7 Conclusions & Future Work
We have described the challenges that we have faced, and some that we still have
to deal with, to assemble an air quality monitoring system for Smart Cities, such
that it can provide a multi-layer approach that uses semantic information about
the sensors and its metadata. We have described a data life-cycle approach that
handles data in different layers, so that it can provide the right abstractions
for the heterogeneous users and applications that will use this dataset. This
approach allows users to choose the data layer that best suits their intended
purpose, i.e. raw data, temporal and spatial aggregations, pollution map repre-
sentations, etc.
In this way, it is possible to generate aggregated views (e.g. over space or
time), and annotate them using RDF. Also, these air quality time series can
serve as a form of diagnosis for a Smart City, which can in turn act quickly to
reduce the risks of morbidity related to air pollutants exposure. The semantic
annotations can range from anomaly labeling, daily/hourly patterns, outlier de-
tection, etc. Nevertheless, as it is important to trace the origin of the data that
is processed by the OpenSense platform, we propose maintaining provenance
hyperlinks, so that at any time a citizen or scientist can access the raw data for
further processing or for quality control. We have also presented the building
blocks of a more complex air pollution sensing scenario, which includes not only
public transportation sensors but also reference stations and crowdsensing.
In the next months we plan to finish and launch into production the Lausanne
OpenSense deployment platform, including the semantic aspects discussed in
this work, beyond the current capabilities offered by GSN natively. We also plan
to further investigate mechanisms for privacy in crowdsensing scenarios, which
is one of the main problems to make this option viable. We plan performing
the necessary evaluation tests to verify that GSN and tinyGSN can effectively
support the loads and variability of the described scenarios, including those
where coverage is an issue, or when energy consumption limits the usefulness
of mobile sensing. Furthermore, we plan to evaluate behavior recommendations
based on the activity recognition results.
Finally we can foresee that this type of deployment can be the basis for an
ecosystem of applications and modules that form part of a city-wide infrastruc-
ture that makes the city capable of self-monitoring, diagnosing, and potentially
self-healing, using semantic data management tools and technologies.
Acknowledgments Partially supported by the Swiss National Science Foun-
dation Nano-Tera.ch OpenSense2 project.
References
1. Aberer, K., Hauswirth, M., Salehi, A.: A middleware for fast and flexible sensor
network deployment. In: Proc. 32nd International Conference on Very Large Data
Bases VLDB, pp. 1199–1202. VLDB Endowment (2006)
2. Agir, B., Calbimonte, J.P., Aberer, K.: Semantic and sensitivity aware location
privacy protection for the internet of things. In: Privacy Online: Workshop on
Society, Privacy and the Semantic Web Privon 2014 (2014)
3. Buchli, B., Yuecel, M., Lim, R., Gsell, T., Beutel, J.: Demo abstract: Feature-
rich platform for wsn design space exploration. In: Information Processing in Sen-
sor Networks (IPSN), 2011 10th International Conference on. pp. 115–116. IEEE
(2011)
4. Calbimonte, J.P., Sarni, S., Eberle, J., Aberer, K.: Xgsn: An open-source seman-
tic sensing middleware for the web of things. In: Proc. of the 7th International
Workshop on Semantic Sensor Networks (2014)
5. Eberle, J., Calbimonte, J.P., Aberer, K.: Efficiently gather-
ing contextual information for health studies. http://www.nano-
tera.ch/pdf/posters2015/OpenSense252.pdf (2015)
6. Elen, B., Peters, J., Poppel, M.V., Bleux, N., Theunis, J., Reggente, M., Standaert,
A.: The aeroflex: a bicycle for mobile air quality measurements. Sensors 13(1), 221–
240 (2012)
7. Elen, B., Theunis, J., Ingarra, S., Molino, A., Van den Bossche, J., Reggente,
M., Loreto, V.: The everyaware sensorbox: a tool for community-based air quality
monitoring. Sensing a Changing World (2012)
8. Gray, A.J., Sadler, J., Kit, O., Kyzirakos, K., Karpathiotakis, M., Calbimonte, J.P.,
Page, K., García-Castro, R., Frazer, A., Galpin, I., et al.: A semantic sensor web
for environmental decision support applications. Sensors 11(9), 8855–8887 (2011)
9. Hasenfratz, D., Saukh, O., Walser, C., Hueglin, C., Fierz, M., Thiele, L.: Pushing
the spatio-temporal resolution limit of urban air pollution maps. In: Pervasive
Computing and Communications (PerCom), 2014 IEEE International Conference
on. pp. 69–77. IEEE (2014)
10. Hedgecock, W., Völgyesi, P., Ledeczi, A., Koutsoukos, X., Aldroubi, A., Szalay, A.,
Terzis, A.: Mobile air pollution monitoring network. In: Proceedings of the 2010
ACM Symposium on Applied Computing. pp. 795–796. ACM (2010)
11. Koubarakis, M., Sioutis, M., Garbis, G., Karpathiotakis, M., Kyzirakos, K., Niko-
laou, C., Bereta, K., Vassos, S., Dumitru, C.O., Espinoza-Molina, D., et al.: Build-
ing virtual earth observatories using ontologies, linked geospatial data and knowl-
edge discovery algorithms. In: On the Move to Meaningful Internet Systems: OTM
2012, pp. 932–949. Springer (2012)
12. Organization, W.H.: News release. http://www.who.int/mediacentre/news/
releases/2014/air-pollution/en/. (March 2014)
13. Peters, J., Van den Bossche, J., Reggente, M., Van Poppel, M., De Baets, B.,
Theunis, J.: Cyclist exposure to ufp and bc on urban routes in antwerp, belgium.
Atmospheric Environment 92, 31–43 (2014)
14. Radanovic, G., Faltings, B.: Incentives for truthful information elicitation of contin-
uous signals. In: Twenty-Eighth AAAI Conference on Artificial Intelligence (2014)
15. Sanchez, L., Muñoz, L., Galache, J.A., Sotres, P., Santana, J.R., Gutierrez, V.,
Ramdhany, R., Gluhak, A., Krco, S., Theodoridis, E., et al.: Smartsantander: Iot
experimentation over a smart city testbed. Computer Networks 61, 217–238 (2014)
16. Sheth, A., Henson, C., Sahoo, S.S.: Semantic sensor web. Internet Computing,
IEEE 12(4), 78–83 (2008)
17. Tsai, D.H., Guessous, I., Riediker, M., Paccaud, F., Gaspoz, J.M., Theler, J.M.,
Waeber, G., Vollenweider, P., Bochud, M.: Short-term effects of particulate matters
on pulse pressure in two general population studies. Journal of hypertension 33(6),
1144–1152 (2015)
18. Un, E.C., Eberle, J., Kim, Y., Aberer, K.: A model-based back-end for air quality
data management. In: Proceedings of the 2013 ACM conference on Pervasive and
ubiquitous computing adjunct publication. pp. 1143–1150. ACM (2013)