Está en la página 1de 13

Environmental Modelling & Software 79 (2016) 354e366

Contents lists available at ScienceDirect

Environmental Modelling & Software


journal homepage: www.elsevier.com/locate/envsoft

Design and prototype of an interoperable online air quality


information system
Stefan Wiemann*, Johannes Brauner, Pierre Karrasch, Daniel Henzen, Lars Bernard
€t Dresden, Germany
Chair of Geoinformatics, Technische Universita

a r t i c l e i n f o a b s t r a c t

Article history: This paper focuses on the design and development of a Spatial Data Infrastructure (SDI)-compliant online
Received 31 July 2014 system for air quality information retrieval, including support for real-time monitoring. This system
Received in revised form assesses exposure to ambient air to mitigate potential health risks, which is crucial for susceptible in-
27 October 2015
dividuals, health practitioners and decision makers. Particular attention is paid to the development of an
Accepted 29 October 2015
interoperable, applicable and transferrable approach to the application of robust and flexible air quality
Available online 12 November 2015
modeling as required for early warning systems on the Web. Moreover, the design provides different
access levels to system components for both non-expert and scientific users and supports extension with
Keywords:
Spatial data infrastructure
external standard compliant services. The developed Web-client Time2Maps enables the user to view,
Air quality analyze and download requested air quality information and serves as a portal to the designed online
Sensor observations system.
© 2015 Elsevier Ltd. All rights reserved.

1. Introduction further data sources and related decision support tools. This paper
describes and discusses the design and implementation of a flexible
The adverse impact of poor air quality on human health, espe- and interoperable online system to provide air quality information
cially the respiratory and cardiovascular systems, is well known as a basis for subsequent exposure and health risk assessment in a
and documented (e.g., Janssen and Mehta, 2006; EEA, 2012). During timely fashion. Accordingly, the presented system design focuses
previous decades, various legislative initiatives and technical de- on three key points:
velopments have helped to improve air quality in many countries.
However, this topic is still considered a significant issue (WHO, 1. Interoperability: The standards and specifications that are used
2009). Consequently, timely information about current air quality for the development and establishment of Spatial Data In-
conditions is considered crucial, not only for individuals, who must frastructures (SDI) serve as the reference framework to achieve
be alerted about potentially adverse air pollution levels or meteo- interoperability. Related standards from the International
rological extremes, but also for a vast group of stakeholders, such as Standardization Organization (ISO) and the Open Geospatial
health practitioners and decision makers, who monitor potential Consortium (OGC) are applied and tested for their suitability to
health risks and initiate mitigation strategies. support the provision of spatio-temporal air quality informa-
Today, a huge number of different online air quality information tion. By using and extending state-of-the-art technology and
sources are already available (Johansson et al., 2015). Related in- standards, this system is implemented as a set of interoperable
formation services are not only provided by weather services but Web Services, which can easily be integrated into existing SDI
also offered by various environmental administrations and and Geographical Information Systems (GIS).
different private or commercial providers. However, these sources 2. Applicability: Interoperability is one key aspect to ease applica-
still lack interoperable air quality information services that can bility. Additionally, the system builds on available, official, in situ
easily integrate provided air quality maps and indicators into air quality observation networks, on remote sensing and on
different information systems to efficiently support fusion with additional topographic, traffic and socio-economic data sources
that are typically available. The application of easily understood
and robust statistical methods to calculate air quality maps
should support an easy and ad-hoc deployment of the devel-
* Corresponding author.
E-mail address: stefan.wiemann@tu-dresden.de (S. Wiemann). oped components. The resulting system serves continuous

http://dx.doi.org/10.1016/j.envsoft.2015.10.028
1364-8152/© 2015 Elsevier Ltd. All rights reserved.
S. Wiemann et al. / Environmental Modelling & Software 79 (2016) 354e366 355

spatio-temporal air quality information to both non-expert and Continuous monitoring builds the foundation to observe the
scientific users. The information can be directly visualized via status of and changes in air quality. Existing in situ sensors typically
interactive Web maps or integrated into other information and provide time series of different air quality parameters as point-
analysis systems via the offered Web services. based measurements that are valid within the direct vicinity of
3. Transferability: In addition to interoperability and applicability, the sensor station. Following the principle of subsidiarity,
the provision and usage of easily available software components numerous national or regional environment agencies provide up-
(ideally well-maintained and open source) and commonly and to-date air quality observations within their administrative areas.
globally available data sets (e.g., digital elevation models, Because of legal reporting obligations, those observations are
remote sensing derived land cover, open street map data, etc.) usually aggregated on higher levels (e.g., the European air quality
should ease transfer towards other areas. database, Airbase1). Although the online availability of air quality
data continuously increases, the harmonization of air quality
The methods and tools that are presented in this paper mainly monitoring standards and processes is recognized as a major issue
focus on applications to describe the spatio-temporal distribution (Engel-Cox et al., 2013). Many technical and organizational aspects
of particulate matter with diameters less than 10 mm (PM10), which are addressed by the INSPIRE directive (Directive 2007/2/EC
is one of the most prominent pollutants in research and the media establishing an Infrastructure for Spatial Information in Europe, EU
because of its well-known health implications. However, this sys- 2007), which lays down a general framework for publishing envi-
tem is expected to be easily extendable to other air pollutants or ronmental data in the European Union, comprising the application
meteorological parameters. of open standards for data encoding and exchange and regulations
Chapter 2 introduces air quality monitoring and the current regarding the general availability and accessibility of data. The
state towards interoperable, Web-based environmental informa- INSPIRE data specifications on Atmospheric Conditions and on
tion services. The general design of the system components Environmental Monitoring Facilities are of particular importance in
(chapter 3) includes a description of the applied statistical terms of air quality. As a logical consequence, the CAFE directive
modeling approach and a discussion regarding the applicability of explicitly demands compatibility to the INSPIRE directive with
current remote sensing products for air quality monitoring. The regards to its implementation.
actual system implementation is described in chapter 4. Chapter 5 Even if in situ air quality monitoring is legally required,
concludes with related findings and identifies future research and continuous measurements of current sensor networks provide
development issues. rather coarse measurements to stay cost efficient. Accordingly, the
derivation of spatially covering air quality maps typically requires a
2. Air quality monitoring and web-based information spatial densification of the air quality observations (Engel-Cox et al.,
systems 2013). Such a densification is typically achieved by using geo-
statistical models or a combination of geostatistical and deter-
2.1. Air quality monitoring and modeling ministic models. Deterministic models are based on emission rates
and simulate actual meteorological and chemical processes in the
Legislation and regulations provide the general framework for atmosphere and their effects on air pollutant distribution. However,
today's official air quality monitoring. As a prominent example, the exact emission rates and the distinction between anthropogenic
European CAFE Directive (Directive 2008/50/EC on ambient quality and natural sources are often unknown or subject to uncertainty
and cleaner air for Europe; EU, 2008) mandates European Union (Klingner and Sa €hn, 2008). In contrast, statistical models mainly
member states to assess national ambient air quality by operating rely on sensor observations and their statistical relationships and
appropriate measurement systems (observations and models). The use interpolation methods or land-use regression. For reviews on
directive requires an implementation of clean air plans to keep corresponding approaches, refer to Ryan and LeMasters (2007),
harmful air pollution below defined thresholds. In addition to Hoek et al. (2008) and Gulliver et al. (2011). Models that involve
thresholds for a number of air pollutants, the directive defines Web-based implementation can be divided into two categories: (1)
general rules on the setup of observation networks, including the simple to use and automatable models that can directly be accessed
spatial and temporal resolution and requirements for the sampling and executed online and (2) complex, ideally automatable models
locations. In the case of particulate matter, the distribution of sta- that can be used for the calibration and validation of previous
tions is determined by population density and PM assessment models.
thresholds. For example, a midsized city with approximately half a By using remote sensing, information can be obtained from a
million inhabitants is typically monitored with two or three sta- distance (ex situ) by measuring and analyzing the interaction of
tions. Measurements shall be available on a daily or, where electromagnetic radiation with a feature of interest, which results
possible, on an hourly basis with a data availability of 90% and a in certain spectral and radiometric properties. Remote sensing is
maximum uncertainty of 25% at fixed measurement stations (EU, used to measure the transmission characteristics of radiation
2008). throughout the atmosphere in terms of air quality. The transition is
For PM10, particles with a diameter less than 10 mm, an annual influenced by aerosols and trace gases and allows conclusions
average of 40 mg/m3 is defined as the threshold for the protection of regarding the concentration and distribution of air pollutants.
human health. Moreover, a concentration of 50 mg/m3 shall not be Compared to in situ measurement networks, remote sensing (1)
exceeded for more than 35 days a year. The annual average promises not only point based observations but observations that
threshold for PM2.5, particles with a diameter less than 2.5 mm, is cover an area (i.e., Pixel) and (2) could offer global coverage of
set to 25 mg/m3. However, because PM2.5 is considered harmful at observation products of an agreed and harmonized quality level,
any concentration, the directive stipulates an explicit reduction thus promising good applicability and transferability as required for
target for all member states. In addition to the regulation of PM10 the intended online system. However, the major challenge is to
and PM2.5 concentrations, the measurement and assessment of the derive air quality at ground level, where it actually poses a risk to
ultra-fine fraction of particulate matter (PM0.1) has become more human health. An overview of satellite systems and their
important. A comprehensive review of corresponding studies
together with recommended environmental actions are summa-
1
rized by the WHO REVIHAAP report (WHO, 2013). http://www.eea.europa.eu/themes/air/air-quality/map/airbase.
356 S. Wiemann et al. / Environmental Modelling & Software 79 (2016) 354e366

application for the determination of aerosols and different trace controlled, it can be used as an indicator to alert for potentially
gases can be found in Martin (2008). harmful air quality conditions close to the sensor site and, where
appropriate, to model initial air quality estimates in the area.
2.2. Interoperability for air quality information systems However, these estimates are considered to be preliminary results
and must be interpreted with caution. Laboratory and time-
Bernard et al. (2014) suggested three interoperability levels to consuming gravimetric mass determination are demanded by
provide a rough classification of the different approaches towards the EU (2008) for the quality control of PM10, which makes pre-
sharing and integrating spatio-temporal environmental models, liminary data inevitable for early warnings. After quality control,
observations and simulation results in SDI: all the measurements are published according to corresponding
legal regulations, e.g., the CAFE directive and its national imple-
1. Collective model development of a community model, which mentations. At this stage, these data are considered to be suitable
typically occurs within a certain domain and addresses domain for regular air quality modeling. Consequently, modeling is divided
specialists and scientists, such as the model developers. into (1) air quality estimation, which uses an ad hoc model to
2. Sharing and coupling models and model components is achieved instantly provide information about the spatial variability of air
via standardized interfaces (e.g., the Open Modeling Interface) quality, e.g., as required for early warning; and (2) the application
to build integrated models, either within a domain or by of more complex and reliable models, which replace the first es-
bridging two or more disciplines, and again mainly addresses timates by refined air quality information as required for official
specialists and scientists, such as the model integrators. warnings, health risk assessment and reports. The latter can also
3. Provision of observations and simulation results via (standardized) be used to validate, train and further improve the ad hoc modeling
Web services and, more specifically, by applying OGC Web ser- approaches.
vice standards. The provided simulation results stem either The workflow to generate air quality information from different
from pre-calculated model runs or from an ad-hoc execution of real-time sensor measurements with an integrated air quality
a suitable robust (pre-parameterized) model. The services model and with a focus on near real-time air quality estimation is
typically address a wider group of non-expert users. shown in Fig. 2. Conceptually, three major tasks can be identified:

The means for sharing models, interoperable model compo- 1. Observations: sensor measurements on specific air quality pa-
nents and model integration (interoperability levels 1 and 2) are rameters are gathered by in situ and/or remote sensing systems
the subject of ongoing research (e.g., Nativi et al., 2013; in a real-time or near real-time fashion. The data are archived
Hamilton et al., 2015). Thus, current operational systems usu- and used as major input for air quality model building and
ally fall within level 3 and support sharing observations and modeling.
modeling results, primarily as online air quality maps and re- 2. Modeling: the model building process builds on existing sensor
ports. Consequently, past, current and forecasted air quality measurements and additional environmental data and is based
conditions are modeled offline and put on the Web for static or on either deterministic or statistical approaches. The aim is to
interactive visualization. derive an integrated air quality model that can process real-time
A prominent example for an online air quality information sensor measurements in an ad hoc manner. A refinement is
system is the Copernicus Atmosphere Monitoring Service,2 which applied to regularly validate and refit the final model once new
combines an ensemble of models to derive data products on the input data are available. Corresponding notifications can be
global and regional level. This service primarily uses in situ data to, either pull-based (regular check for changed inputs) or push-
inter alia, serve modeled air quality data on the Web. In addition, it based (changes in inputs triggers refinement).
offers capabilities to compare models against each other and 3. Dissemination: model results are published as (near) real-time
against ground-truth in situ observations. However, it focuses on a air quality maps on the Internet. This step comprises air qual-
continental level and provides no means for SDI-based data ity information and visualization for non-expert users (e.g.,
download or processing. As another example, an interpolation environment policy makers) and raw data and intermediate
service for in situ air quality observations is described by the modeling results for expert users (e.g., scientists).
INTAMAP project (Pebesma et al., 2011). This service provides
functionality via the OGC Web Processing Service (WPS) standard Several requirements must be met to implement air quality
and can therefore be integrated into SDI service chains. However, modeling and assessment workflow. In general, air quality models
this project is no longer maintained. must be easy to use, customize and maintain to support the us-
ability and interpretability of results. Consequently, the selection of
3. Design of an interoperable online air quality information suitable models and modeling strategies constitutes a major chal-
system lenge. A generic online system could integrate a variety of models
for different purposes. To facilitate application in a plug-and-play
3.1. Workflow for online air quality modeling manner, each model must be complemented with sufficient hu-
man- and machine-readable lineage information, which comprises
A set of idealized workflows sketch the potential usage sce- input and output restrictions, the applied modeling strategy and
narios for the online air quality information system, help to quality characteristics, e.g., compliant to ISO 19115 (Geographic
document the overall requirements and provide the basis for the Information e Metadata). Approaches to describe the provenance
following system design. Timely access to information is required issues of models are discussed by Henzen et al. (2013). Links to
to mitigate short-term human health risks that are related to air existing publications on a model are regarded as helpful, especially
quality. As depicted in Fig. 1, the starting point is an observation in the scientific domain.
that is made by a specific in situ sensor, which provides raw Although a variety of tools and strategies exist for air quality
measurements. Although this data point is not yet quality modeling, identifying optimal solutions for a particular use remains
challenging. The following prerequisites must be fulfilled to facili-
tate flexible and interoperable modeling in an ad hoc manner and
2
https://www.gmes-atmosphere.eu. with good performance in terms of quality and response time:
S. Wiemann et al. / Environmental Modelling & Software 79 (2016) 354e366 357

Fig. 1. Conceptual workflow for online air quality monitoring and modeling.

Fig. 2. Schematized workflow for building an integrated air quality model within an online system.

 Processed input data must be available as frequently as outputs  Ideally, by using standardized Web Service interfaces, the
are expected, preferably openly accessible via standardized Web components (most importantly, the model) must be exchange-
Service interfaces on the Internet. able and replaceable by alternatives by using the same interface
 For exposure and health risk assessment and for regulatory without tampering with the workflow and infrastructure as a
monitoring, the air quality must be modeled at ground level, whole. This feature allows better comparison, transferability
where people are actually exposed to it. and reuse of models in other applications, eases model updating
 Tools for modeling air quality must be well tested and validated and supports dynamic service failover strategies in a real-time
and preferably openly accessible via standardized interfaces for environment.
Web-based geoprocessing.
 All tools must be well documented by respective metadata to 3.2. System architecture
allow for further development and maintenance.
 The overall system performance must allow near real-time ac- Today's Web-based air quality information systems mainly focus
cess to air quality observations and modeling results. on the efficient provision of air quality observations and modeling
results. Less emphasis is given to the interoperability, flexibility and
Additionally, the following requirements must be met in terms extensibility of the system components. To address this issue, SDIs
of the overall design of an online air quality information system: provide concept, infrastructure and required architecture essentials
to build Web-accessible information systems. To implement the
 Web Services must allow automated, computer-invoked runs of workflows that were described above, an infrastructure that is
the whole workflow. Such automation is feasible with stateless mainly based on Web Services as standardized by the Open Geo-
Web Services, which do not require human interaction during spatial Consortium (OGC) is designed and shown in Fig. 3.
runtime. This feature does not necessarily affect the model Users gain access to air quality maps either by accessing a Web-
building process, but rather the online execution of the inte- based client application via their Web browser or by using a
grated air quality model. standalone desktop application; both communicate with Web
 Air quality maps must be digitally and interactively available in Mapping Services (WMS), which provide the air quality maps. To
Web browsers; no further expert (desktop-based) software promote respective existence, dataset, service, and client metadata
should be required. are registered in a Catalog Service for the Web (CSW), which is
 Experts must be able to download unprocessed source data and provided in conjunction with a geoportal.
intermediate results on demand in a variety of formats and Client requests to a WMS contain parameters such as the loca-
customized according to their needs (e.g., in terms of the spatial tion, spatial resolution, time, and visualization details (e.g., styling
and temporal resolution). hints). According to these parameters, the WMS itself requests a
358 S. Wiemann et al. / Environmental Modelling & Software 79 (2016) 354e366

Fig. 3. SDI infrastructure for air quality modeling based on standardized Web Services. The applied standardized interfaces are the Web Feature Service (WFS), the Web Coverage
Service (WCS), the Sensor Observation Service (SOS), the Web Processing Service (WPS), the Web Mapping Service (WMS) and the Catalog Service for the Web (CSW).

Web Processing Service (WPS) for an on-demand calculated air the sensor location and sensor technology are mapped to O&M,
quality map. Once the map is received from the WPS, the WMS all further data collection and customized data provision (e.g.,
sends it back to the client. data for a particular location at a certain point in time) are
A WPS offers a standardized interface to run almost any geo- automatically handled by the SOS. Second, required remote
processing functionality (Foerster et al., 2011) and is thus used to sensing data are offered by a WCS. The WCS provides an interface
provide the developed air quality model. A similar approach was to customize georeferenced image data for download or further
used for hydrological modeling by Castronova et al. (2013) and for usage in other Web Services. WCS downloads can be parameter-
geoprocessing workflow composition by Vitolo et al. (2015) and ized to include data in various spatio-temporal and thematic
Yue et al. (2015). The WPS can wrap almost any legacy (geo- dimensions.
processing) functionality that is invoked without user interaction Automated Web service communication is possible because
(e.g., a Python or R script), including the ability to parameterize the each service provides standardized interfaces and thus knows how
process properly. Because almost any legacy software that is used in to communicate with other services. Additionally, each service can
the geospatial domain can be invoked by scripts or Application be replaced by a similar service that offers the same standardized
Programming Interfaces (API), model implementers can use soft- interface. This concept supports interoperability, applicability and
ware with which they are already familiar to prepare their models transferability as key aspects of the proposed online system and is a
and provide scripts that are executable by WPS. During the model's major requirement and facilitator for an SDI-based architecture's
runtime, a WPS requests required data, e.g., the latest observations, success.
which are provided by additional Web Services (e.g., the Sensor
Observation Service (SOS) or Web Coverage Service (WCS)). Sub- 3.3. A robust model for the estimation of air quality
sequently, the WPS computes corresponding air quality maps based
on these latest observations. In terms of the preconditions that were set in Chapter 2, only a
An important aspect when dealing with processing services for few air quality models fulfill the requirements for real-time
air quality modeling is the awareness and handling of the temporal modeling as a basis for online mapping in SDI. A huge number of
reference of the data. Elements that must be formalized and models are either not available at reasonable costs, operate only on
encoded in the metadata of the processing results comprise the specialized input data, or are very complex to parameterize and
timestamp of the creation, the timestamp or timeframe of the operate. Consequently, a new model has been designed and
contained air quality information and the timeframe of the validity implemented by using available approaches as blueprints. This
of the provided information. This information can be used by a model targets a robust and flexible online air quality modeling
client to search for suitable data, assess the usability for a certain framework and performing adequately in terms of both quality and
purpose and conduct additional processing, such as time series computation time. In this sense, “robust” means that the model is
analysis. However, the access to spatio-temporal OGC Services is relatively insensitive to outliers or missing values.
currently hampered by a lack of client applications that address The developed model builds on Tobler's 1st law of geography
time-enabled data and maps. (“Everything is related to everything else, but near things are more
Two standardized Web Service types are used to provide ac- related than distant things”, Tobler, 1970). However, in this case,
cess to historical and real-time sensor observations. First, an SOS distance not only refers to geographic distance but represents an
provides access to in situ sensor network observations for the air artificial multivariate distance that is based on a number of
quality model on the one hand, and scientists on the other. A descriptive attributes. This model builds on the concept of land-use
common underlying data model that encapsulates the raw sensor regression (LUR; see Ryan and LeMasters, 2007 or Hoek et al., 2008
observations and provides them with accompanying metadata is for a review of LUR) and adapts the concept of affinity areas, which
required to enable such multiple usages. The standardized Ob- was introduced by McGregor (1996) and more recently applied by
servations & Measurements (O&M) data model is used for this Vienneau and Briggs (2013).
purpose. Once raw sensor data and various metadata regarding The general idea is to determine to which extent a certain area
S. Wiemann et al. / Environmental Modelling & Software 79 (2016) 354e366 359

can be represented by a single in situ station. Therefore, a multi- space should have a verifiable and non-interrelated influence on
dimensional feature space is created from attributes that are the air quality. To allow for a comparison of attribute distances,
related to air quality (e.g., topographic data, land use, etc.), with these attributes must be transformed to the same, e.g., normalized,
each attribute representing a single dimension. A distance between scale. The type of transformation depends on the attribute char-
two points in this space is referred to as attribute distance. acter and value distribution. The proposed model is applicable for
To map a certain geospatial location on the feature space, all point-based in situ observations and can serve as a basis for ad
required attribute values must be calculated for that particular hoc air quality estimation.
location. Because a location with respect to air quality can be seen
as representative of its surroundings, all the attributes that span the 3.4. Application of remote sensing
feature space are calculated based on the location's surroundings.
In a study by Janssen et al. (2008), circles with a radius of 2 km were Remote sensing can be used to (1) complement the in situ model
found to be most appropriate to describe the characteristics of an in on an ad hoc basis and (2) regularly validate the in situ model
situ station. However, because the optimal area might vary among within the online system. For particulate matter, the Aerosol Op-
locations and considered parameters, we generalize this assump- tical Depth (AOD) is found to have a linear relationship with
tion and propose a calculation of attribute values based on the ground-level concentrations of PM10 (Koelemeijer et al., 2006; Hoff
weighted sum of different location surroundings, as shown in Eq. and Christopher, 2009; Emili et al., 2010). The majority of studies
(1). rely on AOD data that are derived from the Moderate-resolution
Imaging Spectroradiometer (MODIS), which is described in Remer
XNR
ui ai ðri Þ et al. (2005). Modeling is based on studies that were conducted
a¼ PNR NR ¼ number of radiuses (1) by Koelemeijer et al. (2006) and Emili et al. (2010). Basically, a linear
i¼0 j¼0 j
u
relationship between the satellite AOD values is assumed (Eq. (4)).
A single attribute value a for the characterization of an in situ
PM10 ¼ a*AOD þ b (4)
station is based on the weighted sum of values from a number of
different surrounding areas, which are defined by their circle radius However, meteorological conditions during the satellite over-
ri. The particular weight ui depends on the representativeness of a pass, such as the relative humidity (RH) in the atmospheric column
surrounding area and is subject to prior investigation, which yields and the boundary layer height (BLH), must be considered. For the
a maximum correlation between the attribute values a and the latter, the distribution of PM10 is assumed to be homogenous, even
observation measurements at each station. though a high variability in the BLH was found by Emeis and
Following the assignment of attribute values, every in situ sta- Scha€fer (2006). The relative humidity is considered by using the
tion can be located in the feature space. Because this concept also H€anel parametrization (Eq. (5), H€ anel, 1976).

 g
1  RH g ¼ Ha€nel extinction coefficient
f ðRHÞ ¼ (5)
1  RH0 RH0 ¼ reference value of relative humidity ðground levelÞ

applies to any other location, the corresponding air quality estimate This physical background facilitates the formulation of second-
can be calculated by applying the Inverse Distance Weighting (IDW, ary models that consider the meteorological parameters BLH and
cf. Shepard, 1968) method as given in Eq. (2). RH (Eq. (6), cf. Emili et al., 2010).

X
NO AOD
ui vi PM10 ¼ a* þb (6)
vest ¼ PNO NO ¼ number of in situ observations (2) BLH*f ðRHÞ
i¼0 u
j¼0 j
Furthermore, value-pairs of PM10 and the AOD, or the
The estimated value vest is based on the sum of specifically normalized AOD with RH and BLH taken into account, frequently do
weighted in situ sensor measurements vi. The specific weight ui for not have a Gaussian distribution; thus, symmetrizing the distri-
each in situ station is calculated by Eq. (3). bution with a logarithmic transformation is necessary (Eq. (7), cf.
Emili et al., 2010).
X
NA gi 1pi
di NA ¼ number of attributes log PM10 ¼ a*log AOD þ b*log BLH þ h*log f ðRHÞ þ d (7)
u¼ PNA (3)
g di ¼ single attribute distance ða  ai Þ
i¼0 j¼0 j
A sufficient number of remote sensing and in situ measure-
The specific weight u is the inverse attribute distance between ments are required to determine the regression parameters. Spatial
representations of two locations in the feature space. Similar to (e.g., nearest neighbor, bilinear) and temporal (e.g., linear) inter-
standard IDW, the distance weight power pi can be set to reflect a polation must be applied to the remote sensing data to overlay and
potential decrease in the influence of an attribute with increasing compare different PM10 and AOD measurements at one point in
distance di. The additional parameter gi determines the weight of a time and space.
single attribute on the total attribute distance and depends on the
impact of an attribute on the considered air pollutants. The optimal 4. A prototype system for online air quality information
weights gi can be determined by analyzing the interrelationship
between the attribute distance and the correlation of observation A number of components are implemented to demonstrate the
measurements between stations. basic functioning of the system design within an SDI. All the
To achieve reliable results, the attributes that span the feature components communicate via open standards, in particular, OGC
360 S. Wiemann et al. / Environmental Modelling & Software 79 (2016) 354e366

standards for spatial data provision, processing and visualization.

4.1. In situ model application and validation

The proposed model is applied for the area of the German Free
State of Saxony, Germany, by using the official sensor network that
is operated by the responsible authority LfULG (Saxon State Office
for Environment, Agriculture and Geology). For demonstration
purposes, PM10 measurements at 26 stations from a study period
from 2003 to 2007 are selected for a proof-of-concept imple-
mentation of the model. All the input attributes are based on the
European Environment Agency (EEA) 1 km INSPIRE reference grid,3
which serves as a harmonized reference system for the exchange of
environmental information in Europe. The availability and coverage
of attributes are key facilitators for the transferability of the
implementation to other regions. The focus on free and preferably
globally available data led to the following list of selected attributes
as proxies to PM10 modeling:

 Geographic distance (GD) between two locations on the Earth's


surface.
 Land cover information (LC) from the 2006 Corine Land Cover
(CLC) dataset. To transform CLC classes to metric scale, a land
use indicator is calculated by following the methodology in Fig. 4. Scatterplot for attribute distances and the correlation of measurements be-
Janssen et al. (2008). tween each in situ station for PM10 with the weights from Table 1.

 Mean elevation data (EL) by averaging elevation data from the


NASA Shuttle Radar Topographic Mission (SRTM; resolution: 3
Table 1
arc-seconds) over the model reference grid. Optimal IDW weights for modeling PM10, in percent.
 Population density data (PD) from the EEA, which are corrected
LC EL TC EM GD R2
by using official population statistics at the municipality level
(inhabitants per square kilometer). PM10 18 30 22 10 20 0.47
 Road density (RD) information from OpenStreetMap (OSM),
which is calculated from main street feature classes (road kilo-
meter per square kilometer). whole layer stack is always 1. Based on Eq. (2), the air quality es-
 Official traffic census (TC) data for major roads in Saxony (ve- timate is calculated by multiplying each layer with the corre-
hicles per day per square kilometer). sponding in situ measurement value and subsequently adding up
 Traffic-related PM10 emission data (EM) from the LfULG at the layers.
street level and aggregated to square kilometers (tons per year). To obtain information regarding the model accuracy and un-
certainty, a number of validation procedures are applied to the
An apparent problem is that a correlation between attributes prototype, in particular:
can potentially distort the modeling results. Therefore, the corre-
lation between the attributes was determined for the study area. As  Leave-p-out cross validation: A certain number p of stations are
outcome, population density and traffic density are removed from randomly removed from the model building and taken as a
the model building because of their very high correlations with validation subset. This step is performed for each day within the
other attributes. Principal component transformation is not applied study period from 2003 to 2007 and results in an R2 of 0.75 for
to assure the interpretability and transferability of the intermediate p ¼ 1 and 0.69 for p ¼ 5 (Fig. 6). As previously concluded by Ryan
and final results. Based on the remaining attributes, the optimal and LeMasters (2007), the number and type of removed stations
weights are calculated by maximizing the correlation between (e.g., rural, urban background, traffic, etc.) influences the model
attribute distances and the correlation of measurements between result. Nevertheless, the applicability and robustness of the
each in situ station. Therefore, a variety of combinations with var- model for estimating PM10 is shown.
iable attribute weights (gi in Eq. (3)) and variable weights for the  External validation: The annual mean values that are derived
location's surroundings (ui in Eq. (1)) are computed. The combi- from the model are compared to the official annual means that
nation with the highest coefficient of determination (R2 ¼ 0.47) for are provided by the LfULG as the responsible authority, which
PM10 is depicted in Fig. 4. The corresponding optimal weights that relies on both statistical and dispersion modeling. The results of
are used for further processing are shown in Table 1. the comparison are shown in Fig. 7. In general, these values
Affinity areas are determined for each in situ station based on indicate similar trends but also show differences, especially an
the attribute weights (Eq. (3)), which indicate whether and to overestimation in rural areas. Because the timeframes only
which extent an in situ station can represent an area. Corre- partially overlap (developed model: 2003e2007; LfULG refer-
sponding examples for three stations are shown in Fig. 5. The ence: 2001e2005), the results must be considered with caution.
modeling result is a stack of layers, with each layer containing Nevertheless, these results are reasonably comparable and
weights of a certain in situ station measurement based on the suggest plausible results from the model.
reference grid cells. The sum of all weights for each cell over the
Typically, statistical methods for air quality modeling are not
regarded to be suitable to assess short-term exposures because
3
http://www.eea.europa.eu/data-and-maps/data/eea-reference-grids-2. they do not consider the prevailing meteorological conditions
S. Wiemann et al. / Environmental Modelling & Software 79 (2016) 354e366 361

Fig. 5. Attribute distance for selected in situ stations for the parameter PM10; green ¼ low attribute distance, red ¼ large attribute distance; Stations: Carlsfeld (left, rural
background with high elevation), Dresden-Nord (middle, traffic), Hoyerswerda (right, urban background). (For interpretation of the references to color in this figure legend, the
reader is referred to the web version of this article.)

Fig. 6. Scatterplot for the internal cross validation of PM10 with p ¼ 1 (left) and p ¼ 5 (right); all measurements in mg/m3.

Fig. 7. Annual means for the presented model compared to official data from the LfULG; all concentrations in mg/m3.

(Gulliver et al., 2011). However, as indicated by the validation, the to be applicable for early warning purposes, as presented in Fig. 1.
model provides robust air quality estimates from a relatively small Moreover, the model, by design, allows for the incorporation of
number of observations on a regional scale and is thus considered additional (e.g., meteorological) attributes to improve the modeling
362 S. Wiemann et al. / Environmental Modelling & Software 79 (2016) 354e366

results. The model building process can be repeated, either regu-  As expected, the exclusion of outliers results in a lower standard
larly or upon a change in the underlying environmental data, to deviation but does not significantly affect the correlation.
update and refit the model accordingly.
In summary, an ideal set of parameters that cover all or a
4.2. Determination of PM10 with remote sensing comparable set of in situ stations unfortunately cannot be found.
Moreover, only approximately 10% of the data for the year 2006 is
An experimental study to derive PM10 from satellite images was available at the highest two quality levels (q2 and q3). Thus, the
conducted for the year 2006 to investigate the applicability of PM10 data that are derived from remote sensing are not sufficiently
remote sensing with respect to the proposed online system. reliable and transferable for ad hoc complementing the modeling of
Therefore, the relationship of AOD and ground level PM10, which air quality in the online system. However, depending on the un-
was obtained from the Saxonian in situ sensor network, was certainty requirements of a particular case, these data can be a
analyzed. The AOD values were obtained from NASA for both starting point for the continuous validation of air quality modeling
MODIS instruments (mounted on Aqua & Terra satellite) at a res- results.
olution of 10 km. All the values had a quality flag attached, which
provides information on its confidence (Hubanks, 2007). BLH data 4.3. Service-based implementation
were obtained from the ECMWF (European Centre for Medium-
Range Weather Forecasts) at a resolution of 0.75 . The reference Central components are implemented as Web Services to
value RH0 (cf. Eq. (5)) was taken from corresponding in situ mea- demonstrate the capability and suitability of the designed system.
surement stations, whereas RH was not available and was These components provide the modeled air quality information by
constantly set to 50%, as recommended by Van Donkelaar et al. using standardized Web interfaces and protocols. The overall sys-
(2010). The H€ anel coefficient g was taken from a comparable tem architecture is organized around OGC Web Services (WMS,
study by Randriamiarisoa et al. (2006) and was set to a value of 1.04. WFS, SOS and WPS; Fig. 8).
The analysis was performed for the different modeling ap- The implementation aims for an ad hoc provision of air quality
proaches that were described in chapter 3.3 by using different estimates for visualization and further analysis. The main compo-
interpolation techniques and quality levels. Furthermore, the sen- nents are as follows:
sibility of the models was tested by identifying and excluding
outliers.  A data service that provides sensor observations is based on the
As an example, representative regression results are shown in 52 North SOS framework4 and offers air quality measurements
Table 2 for one rural in situ station in Niesky, Saxony, which uses Eq. that are encoded in a lightweight O&M profile (Jirka et al., 2012)
(7) with outliers excluded. via the OGC SOS interface. In addition to historical data, up-to-
The high correlation coefficients in Table 2 are not representa- date measurements that are provided by the LfULG are auto-
tive of the results of all 25 available in situ stations. The average matically downloaded, transformed and written into the
correlation coefficient over all stations is comparatively low (0.42) observation database with a delay of one day. This intermediate
and varies between 0.33 and 0.87. The same is true for the stan- step is required because official observations are only provided
dard deviation, which has an average value of 13.69 and a range as 24-h means for the previous day and are not yet readily
between 4.65 and 84.36. The results allow for the following con- available via standardized interfaces.
clusions to be drawn for this study:  A data download service that offers reference features for spatial
aggregation, in particular, postal code areas and administrative
 Good correlations (>0.8) and standard deviations (<10 mm/m3) boundaries, via the Web Feature Service (WFS) interface is set
are found for regression modeling only at particular in situ up by using GeoServer.5
stations.  Three processing services that are implemented with the
 The optimal parameter sets for the regression model signifi- 52 North WPS framework6 offer processing capabilities via the
cantly vary between different in situ stations with no significant OGC WPS interface to the online system. The provided func-
pattern. tionalities include the following:
 The interpolation methods for aligning AOD and PM10 values in o Air quality estimation based on the sensor observations by
space and time only play a minor role in the regression. using the modeling approach in chapter 4.1. The model is
 The best results are achieved for the quality levels q2 and q3, implemented by using the statistical software R7 with the
while the results for the levels q0 and q1 are significantly worse RServe package to access functionality via TCP/IP (Trans-
and, therefore, not usable. mission Control Protocol/Internet Protocol) over the Web.
o Health risk assessment based on the relative risk function for
Table 2 all-cause mortality and short-term exposure to PM10, as
Correlation between AOD and PM10 and the corresponding standard deviation (mg/ determined by Ostro (2004). This service also uses R to
m3) for the rural in situ station in Niesky, Saxony (with outliers excluded).
compute the relative risk index.
Data interpolation Quality level Correlation (Pearson) Standard deviation o Spatial aggregation of either air quality or health risk modeling
Nearest neighbor >¼q0 0.46 12.2 outputs by using polygon reference features. This feature is
>¼q1 0.55 10.4 implemented by using Python with GDAL (Geospatial Data
>¼q2 0.82 9.6 Abstraction Library) bindings in the service backend.
¼q3 0.83 9.8
 A mapping service proxy that listens through the OGC WMS
Bilinear >¼q0 0.49 12.2
>¼q1 0.57 10.4 interface and delegates requests to one of the processing
>¼q2 0.81 9.8
¼q3 0.82 10.2
Bilinear þ time >¼q0 0.49 12.1 4
http://52north.org/communities/sensorweb.
>¼q1 0.53 11.7 5
http://www.geoserver.org.
>¼q2 0.82 9.7 6
http://52north.org/communities/geoprocessing.
¼q3 0.83 10.0 7
http://www.r-project.org.
S. Wiemann et al. / Environmental Modelling & Software 79 (2016) 354e366 363

Fig. 8. Interaction between the implemented Web Services for online air quality modeling and health risk assessment.

services for information retrieval. Accordingly, this service can the offered interfaces for raw data retrieval (via WFS and SOS) and
provide raw or aggregated air quality and health risk maps. Any intermediate processing results of the online system (via WPS)
client that supports the WMS interface can be used to access the requires more sophisticated clients, e.g., those offered by legacy GIS
implemented mapping service and act as a simple client. or statistical software. The corresponding service endpoints are
 An advanced Web-client, named Time2Maps,8 is implemented linked from Time2Maps. This option is intended for domain spe-
for interactive visualization and analysis. This client is mainly cialists to utilize data for their own studies and to experiment with
based on JavaScript and supports the visualization and explo- the provided models. Both access alternatives are enabled by
ration of spatio-temporal data. standardized and, thus, interoperable Web Services as required by
the SDI.
The workflow for retrieving a health risk map is exemplarily
illustrated in Fig. 9 and starts with a WMS request that contains a
5. Conclusions & future challenges
certain timestamp by using the time parameter as defined by the
WMS specification. The WMS delegates the request to the related
A better understanding of environmental phenomena and re-
WPS for health risk modeling. As input, this WPS requests aggre-
lationships and their impact on human life is part of the next-
gated air quality coverage from the aggregation WPS, which itself
generation Digital Earth as promoted in Craglia et al. (2012).
requests PM10 concentrations from the WPS that provides the air
Moreover, the demand for tools and methods to model the envi-
quality modeling. The latter requests observation measurements
ronment is central to the ‘Model Web’, which was described by
via the SOS interface (encoded in O&M) by using the provided
Nativi et al. (2013). Building upon those visions, this paper
timestamp and calculates the corresponding air quality coverage.
described the design and prototypical implementation of an online
This coverage is encoded as GeoTIFF and is subsequently aggregated
air quality information system specifically to assess and mitigate
by the aggregation WPS by using pre-defined reference feature
health risks. The implementations proved the general applicability
boundaries, which are retrieved via the WFS. The aggregated fea-
of current OGC standards to enable interoperable access to near
tures are encoded in GML and sent back to the health risk WPS,
real-time spatio-temporal simulations and to allow different access
which calculates the health risk index. The result is then returned
paths and granularities, including (1) spatio-temporal visualiza-
as GML to the WMS Proxy, which renders the data and finally
tions of the simulation results, (2) simulation results such as spatio-
returns the map image to the original requestor. Thus, communi-
temporal coverages or results that are aggregated to administrative
cation between the components is solely based on existing Web
units as geographic feature collections, and (3) the execution and
standards for data transfer and OGC standards for spatial data and
control of the simulation and spatial-temporal aggregation as
services. In a local test environment, which has low network la-
geoprocessing services. The implemented services were success-
tency, the whole workflow takes less than 10 s and decreases to
fully tested in conjunction with the SDI Geoportal for Saxony.
approximately 1 s for repeated requests by caching already
Moreover, ongoing discussions with local authorities revolve
computed information. All the components, including the source
around the possibility to enhance authoritative services by imple-
code and documentation, can be accessed via the Time2Maps
menting parts of the described online system.
client, which has been made available to the community as an open
This paper also focused on robust and efficient air quality
source tool.
modeling techniques that can be applied to mesoscale air quality
As depicted in Fig. 8, two ways exist to access the functionality
simulations in an online system. The modeling is based on estab-
that is offered by the online system. First, a simple WMS client can
lished approaches (spatial statistics, land use regression, affinity
use the offered mapping service for presentation purposes. Because
areas) to pragmatically and quickly derive air quality maps from an
the complexity of the workflow is hidden behind the widely sup-
in situ sensor network. The use of free and preferably globally
ported WMS interface, the user does not need to address under-
available input data for modeling facilitates transferability to other
lying spatial formats and interaction mechanisms. The Time2Maps
regions. A number of validation methods proved the suitability of
can be used as an access portal to the online system for time-
the approach, but further examination in terms of geostatistical
enabled visualization and statistical information. A screenshot of
methods, such as Kriging, is open to additional research. Future
the application is depicted in Fig. 10. An alternative way to access
work must focus on the automation of applied validation pro-
cedures to (1) implement automated validation services for online
information regarding potential model failures and (2) better
8
http://time2maps.dyndns.org/saxony. support the transferability of the model into other application
364 S. Wiemann et al. / Environmental Modelling & Software 79 (2016) 354e366

Fig. 9. Sequence diagram for online air quality and health risk assessment as implemented by the online system.

Fig. 10. Screenshot of the Time2Maps client for time-enabled air quality information visualization with the map viewer (left), selection menu (right) and time slider (bottom). The
selected map shows the city of Dresden and its surrounding area.

domains as a central requirement for integrated environmental system with provenance metadata, allowing the Web-based dis-
modeling (Laniak et al., 2013). covery and transferability of applied simulation models and input
Communicating model accuracy is another issue: the accuracy parameters, as described in Henzen et al. (2013) and Bernard et al.
of modeling results is usually estimated by the model or statistically (2014).
determined. Modeling results can also be compared to each The integrated usage of in situ and remote sensing data proved
another to determine commonalities and differences. In the pre- to be difficult and might not be possible in a number of cases.
sented study, the accuracy of the results is statistically determined Modeling ground level air quality information from remote sensing
and is included as part of the documentation. However, uncertainty data requires complex modeling of the atmosphere and involves an
often remains difficult to interpret. Although identified as a major inherent time lag for data provision. Moreover, cloud coverage is
requirement for intelligent GIS by Burrough (1992), ways to one of the most limiting factors for using remote sensing data,
formalize, exchange and visualize uncertainty are still a major especially in mid-latitude regions. Thus, remote sensing data
issue, especially on the Web. This concept has been addressed by a currently seems suitable for validation purposes and historical air
number of authors, such as MacEachren et al. (2005), who quality modeling at certain timestamps. The EU Copernicus Pro-
described possibilities for uncertainty visualization; Williams et al. gram may pave the way for the adoption of data access and avail-
(2009), who proposed the UnCertML language to formalize uncer- ability standards and reduce delay in future remote sensing data
tainty; and Bastin et al. (2013), who described strategies for un- provisions (Schulte-Braucks, 2013). However, to what extent this
certainty management in distributed environments. Regarding air approach improves the modeling of ground-level air quality re-
quality, the DELTA tools that were described by Thunis et al. (2012) mains to be seen.
analyze the model performance with respect to the European CAFE The concept of using ‘citizens as sensors’ (Goodchild, 2007),
Directive. Nevertheless, available solutions still must gain accep- which implies a combination of crowdsourcing approaches and
tance and dissemination. A related issue concerns data and simu- low-cost sensors, is considered to be promising but still faces major
lation provenance, which intends to enhance the described online quality issues (cf. Boulos et al., 2011). Further research is required to
S. Wiemann et al. / Environmental Modelling & Software 79 (2016) 354e366 365

investigate whether and how low-cost sensors can be integrated to Hershey, PA, pp. 245e286.
Goodchild, M.F., 2007. Citizens as sensors: the world of volunteered geography.
densify spatio-temporal air quality monitoring systems and how
GeoJournal 69 (4), 211e221.
they can contribute to the Digital Earth's nervous system, as Gulliver, J., de Hoogh, K., Fecht, D., Vienneau, D., Briggs, D., 2011. Comparative
described by De Longueville et al. (2010). assessment of GIS-based methods and metrics for estimating long-term ex-
Additional research challenges are imposed by the transition of posures to air pollution. Atmos. Environ. 45, 7072e7080.
Hamilton, S.H., ElSawah, S., Guillaume, J.H.A., Jakeman, A.J., Pierce, S.A., 2015. In-
the implemented prototype into an operational online system. A tegrated assessment and modelling: overview and synthesis of salient di-
first issue is related to data access: even if legislation mandates free mensions. Environ. Model. Softw. 64, 215e229.
Ha€nel, G., 1976. The properties of atmospheric aerosol particles as functions of the
and open access to environmental data, obtaining access to near
relative humidity at thermodynamic equilibrium with the surrounding moist
real-time environmental observations is still difficult because of a air. Adv. Geophys. 19, 73e188.
low take-up of standards regarding data access and transfer, the Henzen, C., M€ as, S., Bernard, L., 2013. Provenance information in geodata in-
limited spatio-temporal resolution of data, large temporal delays in frastructures. In: Vandenbroucke, D., Bucher, B., Crompvoets, J. (Eds.),
Geographic Information Science at the Heart of Europe, 2013. Lecture Notes in
data provision or the considerable preparation effort that is Geoinformation and Cartography. Springer, pp. 133e151.
required to utilize data. This issue applies to both in situ and remote Hoek, G., Beelen, R., de Hoogh, K., Vienneau, D., Gulliver, J., Fischer, P., Briggs, D.,
sensing observations. A mediation layer for sensor integration, 2008. A review of land-use regression models to assess spatial variation of
outdoor air pollution. Atmos. Environ. 42, 7561e7578.
which was described by Bro €ring et al. (2011), could improve
Hoff, R.H., Christopher, S.A., 2009. Remote sensing of particulate pollution from
accessibility at this point but still needs further elaboration and space: have we reached the promised land? J. Air Waste Manag. Assoc. 59 (6),
persuasion efforts, especially within public administrations. Alter- 645e675.
Hubanks, P.A., 2007. MODIS Atmosphere QA Plan for Collection 005 e Deep Blue
natively, the model itself can be transported to the data for
Aerosol Update. Version 3.5. NASA. Technical Report.
execution by using the moving code paradigm (Müller et al., 2013). Janssen, N., Mehta, S., 2006. Human Exposure to Air Pollution. World Health Or-
This approach is the preferred choice if large datasets are involved ganization, pp. 61e85 (Chapter 3).
and the data transportation time exceeds the time that is required Janssen, S., Dumont, G., Fierens, F., Mensink, C., 2008. Spatial interpolation of air
pollution measurements using corine land cover data. Atmos. Environ. 42,
to run actual calculations, or if the data must be protected and stay 4884e4903.
intra muros. € ring, A., Kjeld, P., Maidens, J., Wytzisk, A., 2012. A lightweight approach
Jirka, S., Bro
for the sensor observation service to share environmental data across Europe.
Trans. GIS 16, 293e312.
Acknowledgments Johansson, L., Epitropou, V., Karatzas, K., Karppinen, A., Wanner, L., Vrochidis, S.,
Bassoukos, A., Kukkonen, J., Kompatsiaris, I., 2015. Fusion of meteorological and
air quality data extracted from the web for personalized environmental infor-
The research leading to these results has received funding from
mation services. Environ. Model. Softw. 64, 143e155.
the European Community's Seventh Framework Programme (FP7/ Klingner, M., S€ ahn, E., 2008. Prediction of PM10 concentration on the basis of high
2007-2013) under grant agreement no. 244100, EO2HEAVEN. resolution weather forecasting. Meteorol. Z. 17, 263e272.
Koelemeijer, R., Homan, C., Matthijsen, J., 2006. Comparison of spatial and temporal
variations of aerosol optical thickness and particulate matter over Europe.
References Atmos. Environ. 40, 5304e5315.
Laniak, G.F., Olchin, G., Goodall, J., Voinov, A., Hill, M., Glynn, P., Whelan, G.,
Bastin, L., Cornford, D., Jones, R., Heuvelink, G.B., Pebesma, E., Stasch, C., Nativi, S., Geller, G., Quinn, N., Blind, M., Peckham, S., Reaney, S., Gaber, N., Kennedy, R.,
Mazzetti, P., Williams, M., 2013. Managing uncertainty in integrated environ- Hughes, A., 2013. Integrated environmental modeling: a vision and roadmap for
mental modelling: the uncertweb framework. Environ. Model. Softw. 39, the future. Environ. Model. Softw. 39, 3e23.
116e134. MacEachren, A.M., Robinson, A., Hopper, S., Gardner, S., Murray, R., Gahegan, M.,
Bernard, L., Ma €s, S., Müller, M., Henzen, C., Brauner, J., 2014. Scientific geodata in- Hetzler, E., 2005. Visualizing geospatial information uncertainty: what we
frastructures: challenges, approaches and directions. Int. J. Digital Earth 7 (7), know and what we need to know. Cartogr. Geogr. Inf. Sci. 32 (3), 139e160.
613e633. Martin, R., 2008. Satellite remote sensing of surface air quality. Atmos. Environ. 2,
Boulos, M.N.K., Resch, B., Crowley, D.N., Breslin, J.G., Sohn, G., Burtner, R., Pike, W.A., 7823e7843.
Jezierski, E., Chuang, K.S., 2011. Crowdsourcing, citizen sensing and sensor web McGregor, G., 1996. Identification of air quality affinity areas in Birmingham, UK.
technologies for public and environmental health surveillance and crisis Appl. Geogr. 16, 109e122.
management: trends, OGC standards and application examples. Int. J. Health Müller, M., Bernard, L., Kadner, D., 2013. Moving code - sharing geoprocessing logic
Geogr. 10, 67. on the web. ISPRS J. Photogrammetr. Remote Sens. 83, 193e203.
€ring, A., Maue
Bro , P., Janowicz, K., Nüst, D., Malewski, C., 2011. Semantically-enabled Nativi, S., Mazzetti, P., Geller, G.N., 2013. Environmental model access and inter-
sensor plug & play for the sensor web. Sensors 11, 7568e7605. operability: the GEO ModelWeb initiative. Environ. Model. Softw. 39, 214e228.
Burrough, P.A., 1992. Development of intelligent geographical information systems. Ostro, B., 2004. Outdoor Air Pollution. World Health Organization. Technical Report.
Int. J. Geogr. Inf. Sci. 6 (1), 1e11. Pebesma, E., Cornford, D., Dubois, G., Heuvelink, G., Hristopulos, D., Pilz, J.,
Castronova, A.M., Goodall, J.L., Elag, M.M., 2013. Models as web services using the Sto€ hlker, U., Morin, G., Skøien, J., 2011. Intamap: the design and implementation
open geospatial consortium (OGC) web processing service (WPS) standard. of an interoperable automated interpolation web service. Comput. Geosci. 37,
Environ. Model. Softw. 41, 72e83. 343e352.
Craglia, M., de Bie, K., Jackson, D., Pesaresi, M., Remetey-Fülo € pp, G., Wang, C., Randriamiarisoa, H., Chazette, P., Couvert, P., Sanak, J., Me gie, G., 2006. Relative
Annoni, A., Bian, L., Campbell, F., Ehlers, M., van Genderen, J., Goodchild, M., humidity impact on aerosol parameters in a Paris suburban area. Atmos. Chem.
Guo, H., Lewis, A., Simpson, R., Skidmore, A., Woodgate, P., 2012. Digital Earth Phys. 6, 1389e1407.
2020: towards the vision for the next decade. Int. J. Digital Earth 5 (1), 4e21. Remer, L.A., Kaufman, Y.J., Tanre , D., Mattoo, S., Chu, D.A., Martins, J.V., Li, R.-R.,
De Longueville, B., Annoni, A., Schade, S., Ostlaender, N., Whitmore, C., 2010. Digital Ichoku, C., Levy, R.C., Kleidman, R.G., Eck, T.F., Vermote, E., Holben, B.N., 2005.
Earth's nervous system for crisis events: real-time sensor web enablement of The MODIS aerosol algorithm, products, and validation. J. Atmos. Sci. 62 (4),
volunteered geographic information. Int. J. Digital Earth 3 (3), 242e259. 947e973.
EEA, 2012. Air Quality in Europe e 2012 Report. European Environment Agency. Ryan, P.H., LeMasters, G.K., 2007. A review of land-use regression models for
Technical Report. characterizing intraurban air pollution exposure. Inhal. Toxicol. 19 (S1),
Emili, E., Popp, C., Petitta, M., Riffler, M., Wunderle, S., Zebisch, M., 2010. PM10 127e133.
remote sensing from geostationary SEVIRI and polar-orbiting MODIS sensors Schulte-Braucks, R., April 2013. Observing the land beyond. Int. Innov. 44e47.
over the complex terrain of the European alpine region. Remote Sens. Environ. Shepard, D., 1968. A two-dimensional interpolation function for irregularly spaced
114, 2485e2499. data. In: Proceedings of the 1968 23rd ACM National Conference, ACM, New
Emeis, E., Sch€afer, K., 2006. Remote sensing methods to investigate boundary-layer York, NY, USA, pp. 517e524.
structures relevant to air pollution in cities. Bound. Layer Meteorol. 121 (2), Thunis, P., Georgieva, E., Pederzoli, A., 2012. A tool to evaluate air quality model
377e385. http://dx.doi.org/10.1007/s10546-006-9068-2. performances in regulatory applications. Environ. Model. Softw. 38, 220e230.
Engel-Cox, J., Oanh, N.T.K., Van Donkelaar, A., Martin, R.V., Zell, E., 2013. Toward the Tobler, W., 1970. A computer movie simulating urban growth in the Detroit region.
next generation of air quality monitoring: Particulate Matter. Atmos. Environ. Econ. Geogr. 46, 234e240.
80, 584e590. Van Donkelaar, A., Martin, R.V., Brauer, M., Kahn, R., Levy, R., Verduzco, C.,
EU, 2008. Directive 2008/50/EC of the European Parliament and of the Council of 21 Villeneuve, P.J., 2010. Global estimates of ambient fine particulate matter con-
May 2008 on Ambient Air Quality and Cleaner Air for Europe. centrations from satellite-based aerosol optical depth: development and
Foerster, T., Schaeffer, B., Baranski, B., Brauner, J., 2011. Geospatial web services for application. Environ. Health Perspect. 118, 847e855.
distributed processing: applications and scenarios. In: Zhao, P., Di, L. (Eds.), Vienneau, D., Briggs, D.J., 2013. Delimiting affinity zones as a basis for air pollution
Geospatial Web Services: Advances in Information Interoperability. IGI Global, mapping in Europe. Environ. Int. 51, 106e115.
366 S. Wiemann et al. / Environmental Modelling & Software 79 (2016) 354e366

Vitolo, C., Elkhatib, Y., Reusser, D., Macleod, C.J.A., Buytaert, W., 2015. Web tech- WHO, 2013. Review of Evidence on Health Aspects of Air Pollution e REVIHAAP
nologies for environmental big data. Environ. Model. Softw. 63, 185e198. Project. World Health Organization, Regional Office for Europe. Technical
Williams, M., Cornford, D., Bastin, L., Pebesma, E., 2009. Uncertainty Markup Lan- Report.
guage (UnCertML). OpenGIS Discussion Paper. Version 0.6. Yue, P., Zhang, M., Tan, Z., 2015. A geoprocessing workflow system for environ-
WHO, 2009. Global Health Risks: Mortality and Burden of Disease Attributable to mental monitoring and integrated modelling. Environ. Model. Softw. 69,
Selected Major Risks. World Health Organization. Technical Report. 128e140.