Professional Documents
Culture Documents
Ecological Modelling: Angela Lausch, Andreas Schmidt, Lutz Tischendorf
Ecological Modelling: Angela Lausch, Andreas Schmidt, Lutz Tischendorf
Ecological Modelling
journal homepage: www.elsevier.com/locate/ecolmodel
Data mining and linked open data – New perspectives for data analysis
in environmental research
Angela Lausch *, Andreas Schmidt, Lutz Tischendorf
Department of Computational Landscape Ecology, Helmholtz Centre for Environmental Research – UFZ, Permoserstr, 15/D-04318, Leipzig, Germany
A R T I C L E I N F O A B S T R A C T
Article history: The rapid development in information and computer technology has facilitated an extreme increase in
Available online xxx the collection and storage of digital data. However, the associated rapid increase in digital data volumes
does not automatically correlate with new insights and advances in our understanding of those data. The
relatively new technique of data mining offers a promising way to extract knowledge and patterns from
Keywords: large, multidimensional and complex data sets. This paper therefore aims to provide a comprehensive
Data Mining
overview of existing data mining techniques and related tools and to illustrate the potential of data
Knowledge Discovery Process
mining for different research areas by means of example applications. Despite a number of conventional
Data Mining Tools
Linked Open Data
data mining techniques and methods, these classical approaches are restricted to isolated or “silo” data
Semantic sets and therefore remain primarily stand alone and specialized in nature. Highly complex and mostly
RapidMiner interdisciplinary questions in environmental research cannot be answered sufficiently using isolated or
area-based data mining approaches. To this end, the linked open data (LOD) approach will be presented
as a new possibility in support of complex and inter-disciplinary data mining analysis. The merit of LOD
will be explained using examples from medicine and environmental research. The advantages of LOD
data mining will be weighed against classical data mining techniques. LOD offers unique and new
possibilities for interdisciplinary data analysis, modeling and projection for multidimensional, complex
landscapes and may facilitate new insights and answers to complex environmental questions. Our paper
aims to encourage those research scientists which do not have extensive programming and data mining
knowledge to take advantage of existing data mining tools, to embrace classical data mining and LOD
approaches in support of gaining more insight and recognizing patterns in highly complex data sets.
ã 2014 Elsevier B.V. All rights reserved.
1. Introduction Schmid (2013) points out that with the increasing networking and
ever-growing computing capacity we find ourselves in a “profound
The rapid technological advances in information technology of transformation process” of the 21st century where more and more
the 21st century are intrinsically linked to a gigantic increase in “decision-making processes have to be outsourced to technical
data and information. This is the result of increased networking systems” in order to be at all able to function. Internet services like
and globalization, continuous improvement in computer engi- Facebook, Google and Wolfram Alpha aim “to make all systematic
neering, media storage, highly sophisticated data bases, the knowledge immediately computable and accessible to everyone”
Internet as a platform for communication and of the enormous (Alpha, 2013).
expansion of automated data collection via sensors, monitoring Estimates suggest that the entire amount of data in the world
systems and mobile and smartphone applications. The ever doubles every 20 months (Runkler, 2010). According to Meyer and
smaller-scale automated measuring, cataloging and monitoring Lüling (2003), researchers at the University of Berkeley have
of so many areas of people’s lives as well as their social calculated that around 1 exabyte (= 1 million terabytes) of data is
environment and networks leads to a “datafication” of the world generated every year. And so by the end of 2015, the annual
(Kreissl, 2013). Rekow (2013) refers to the phenomenon of growing Internet data traffic will amount globally to 1 zettabyte
data volume in all areas of life also as “information overload”. (1 zettabyte = 1000 exabytes). For the sake of comparison, “it
would take over five years to watch the amount of videos which
will be transferred per second through the Internet by the year
* Corresponding author. Tel.: +49 3412351961. 2015” (Computer, 2013). Whereas in the 1990s there was still a
E-mail address: [email protected] (A. Lausch). deficit in the availability of digital data, the relation between data
http://dx.doi.org/10.1016/j.ecolmodel.2014.09.018
0304-3800/ ã 2014 Elsevier B.V. All rights reserved.
Please cite this article in press as: Lausch, A., et al., Data mining and linked open data – New perspectives for data analysis in environmental
research. Ecol. Model. (2014), http://dx.doi.org/10.1016/j.ecolmodel.2014.09.018
G Model
ECOMOD 7323 No. of Pages 13
availability and existing evaluation methods and algorithms has 2.1. Data mining vs. statistical approaches
changed completely.
Nowadays, generating data does not mean insight. Only a Data mining and conventional statistical analyses have different
fraction of the data is used to gain insight. As always, the statement purposes. Whereas classical statistical approaches focus primarily
applies that “data is not insight”. Those who have a lot of data “have on verifying stated hypotheses, data mining methods search
possibilities to gain insight, but those who have the right analytical through many possible, mostly unknown hypotheses (Witten and
tools and instruments, also hold the key to acquiring insight”. Eibe, 2001). Coupling statistical and data mining methods will be
While large amounts of available data have, for the most part, been the only way to gain insight and knowledge from the ever
archived and stored, only small parts have really been analyzed, increasing amount of digital data.
used and understood and processed in human-understandable As Witten et al. (2011) pointed out, analyzing diverse and
ways (Begum, 2013). complex data in the future will not only require a coupling of data
In order to understand data and to gain insight, the data has to mining and statistical methods but the merging of disciplines and
be sorted, transformed, harmonized and processed both statisti- methods such as pattern recognition, data bases, artificial
cally and analytically. The potential growth increase in data bases intelligence and machine learning algorithms (Fig. 1).
in all areas makes the analysis of huge amounts of more complex
information more difficult and less clear. Using the procedures of 2.2. Data mining process
classical statistics that have been common up to now, it is
becoming apparent that very limited information and knowledge Data mining involves the entire process from the provision of
can be gleamed and insufficient progress made from the huge and data right up to the projection and application of model findings to
complex amounts of data. Only if future research succeeds in new, unknown data structures. This process includes (a)
extracting information from this complex amount of data can this techniques to preprocess data, (b) the actual data mining system
be the basis for good operative and strategic decisions and (DM system) and c) interpreting and evaluating data.
projections. Necessary preprocessing steps for data mining include
In recent years, an enormous development in the area of data selection, preprocessing and transforming data into suitable
complex data analysis has been made with data mining. Data data formats. The DM system itself is at the core of the actual data
mining, often termed knowledge discovery in databases KDD, is mining process, which is made up of three Phases: Training, Test
defined as the “non-trivial process of identifying valid, new and and Validation (Fig. 2). Within this process the objective of data
potentially useful and understandable patterns in data bases” mining is to repeatedly attempt to determine an estimated value
(Saake and Heuer, 1999). Data mining is a rather new field in (e.g., the buying behavior of a customer) based on the researched
computer science that pursues the objective of extracting data (e.g., market data), which is compared with a predetermined
knowledge and analyzing complex data to find existing associa- reference value (target variable) (Witten and Eibe, 2001). This
tions, to extract structures, patterns and regularities in large and process is repeated iteratively until the comparison of estimated
complex data bases (Witten et al., 2011; Joseph et al., 2013). and reference values results in an acceptable value. This obtained
Currently, there are a number of common data mining techniques model now forms the basis for Phase 3 of the interpretation and
to analyze texts, web contents, images, pictures, videos and spatial evaluation as well as for deriving knowledge based on other, as yet
data. Compared to classical statistical techniques, all of these data unknown data.
mining techniques exhibit very good results when it comes to
discovering and analyzing patterns and coherencies of the 2.3. Types of data mining
researched data.
Environmental research is an interdisciplinary research field Data mining systems can be categorized depending on their
that is complex, multifaceted and very dynamic. In this field, data objectives. Fig. 2 provides an overview.
mining holds a particular promise to gain true insight by means of Data Mining: Data mining includes the analysis of numeric and
unveiling associations and relationships. But at the moment, data categorical data in large and complex data sets. Often this term is
mining methods are only used in some areas. And here, too, an used to generally describe more specialized techniques, such as
interdisciplinary analysis approach is not given sufficient support. text, web or spatial data mining.
LOD is a new development in support of internet based data
mining where complex, highly dynamic and, for the first time,
interdisciplinary data mining analyses are becoming possible. The
goal of this paper is: (1) to provide an overview of existing data
mining techniques, tools and methods, (2) to show the potential
data mining holds for research using examples, (3) to give an
overview of the limitations and necessary requirements of data
mining for interdisciplinary environmental research, (4) to present
the LOD approach as a new possibility in complex and interdisci-
plinary data mining research, (5) to encourage non-mathemati-
cians and non-computer scientists from all scientific disciplines
who do not have programming knowledge to use these novel tools
in their research.
Please cite this article in press as: Lausch, A., et al., Data mining and linked open data – New perspectives for data analysis in environmental
research. Ecol. Model. (2014), http://dx.doi.org/10.1016/j.ecolmodel.2014.09.018
G Model
ECOMOD 7323 No. of Pages 13
Fig. 2. Data mining process with the data mining system (DM-S) in the Phases (1) Training Phase, (2) Test Phase and (3) Validation Phase. The data mining process works in a
comparable way in all types of data mining types like text mining or web mining (changed according to Fayyad et al., 1996 and Tanner, 2013).
Text Mining: Text mining includes algorithms to analyze lexical example is “Google Correlate” (http://www.google.com/trends/
and grammatical aspects of texts (Dörre et al., 2001). In text mining correlate).
algorithms the actual text is broken down into specific structures Spatial Data Mining (SDM): Spatial data mining pursues the goal
whereby patterns and core information of the text are recorded of discovering patterns in large, multi-dimensional spatial data
and grouped and classified using data mining methods. The first sets as created by remote sensing techniques in earth observation.
text mining tools could record contents and structures of simple Shekhar et al. (2010) point out that extracting interesting patterns
text documents like Microsoft Word and Acrobat PDF documents. and associations from complex and multi-dimensional spatial data
New developments in text mining now scan and analyze bearing spatial dependencies and spatial-temporal autocorrela-
unstructured text in s, memos, surveys, chats, notes, white papers, tions is more difficult than from traditional numeric and
forums and presentations (Begum, 2013). categorical data.
Web-Mining: According to Bensberg and Weiß (1999) web If, in addition to spatial data, additional time series are
mining is understood to be the application of data mining methods incorporated into the pattern analysis, the term spatial-temporal
on information compiled on the Internet. In web mining, a data mining is used to describe this.
distinction is made between (1) web content mining, which is the
analysis of website contents; (2) web structure or relationship 2.4. Methods of data mining
mining, i.e. the analysis of in- and outbound hyperlinks of websites
and (3) web usage mining, which records and analyses user There are various methods in support of data mining, which can
interaction with websites by scanning log files. be grouped according to the objective in question. Here we provide
Image-Mining: The goal of image mining techniques is to an overview over the most commonly used data mining methods.
analyze and extract spatial patterns in image data which are not We encourage the reader to also refer to Witten et al. (2011) and
explicitly stored in the images (Mohanty et al., 2013). Pattern Chu (2014) for a comprehensive overview of the various methods
extraction is done in a variety of ways based on recognizing the available for data mining.
existence and distribution of colors, texture, shape, distances and Association and sequence analysis: With the help of association
intensities in the image data. analysis, relationships between objects can be made and quanti-
Mining of Picture, Video data and Music: Pattern mining fied. Using specific indicators, which, in most cases, include the
techniques are used more and more to recognize features in support, confidence and lift values, the strength of the identified
pictures, videos and music data. This involves, on the one hand, associations can be evaluated.
detecting disturbances in pictures, video and music files, but also Grouping and clustering: These are procedures which pool
using similarities in data. Furthermore, data mining is the only way similar objects into groups with the goal of having very similar
to handle the huge amounts of complex data which is generated, objects within one group and having the groups differ as much as
for instance, on Google Image or YouTube. Of prime importance is possible from each another. Numerous metrics are used for
the quick activation of content-based images/videos retrieval, characterizing similarities in groups.
indexing, classification and tracking. Regression: With the assistance of regression analysis, func-
Mining of Time Series data: In the analysis of time series, tional dependencies are determined among the variables within a
temporal relations in time data are researched by means of a data set. Regression models are used to estimate or predict
special distance function like dynamic time warping. The goal is to variables. To represent dependencies there are also linear and non-
recognize similarities over the course of the time series, even when linear (e.g., square, logistic or Poisson) regression approaches in
these similar features are time-shifted over the course. A good addition to linear ones (Witten et al., 2011).
Please cite this article in press as: Lausch, A., et al., Data mining and linked open data – New perspectives for data analysis in environmental
research. Ecol. Model. (2014), http://dx.doi.org/10.1016/j.ecolmodel.2014.09.018
G Model
ECOMOD 7323 No. of Pages 13
Classification: The goal of classification is to find functions and The four most commonly used open source data mining tools are
models with whose help data objects can be assigned to previously RapidMiner, R, WEKA and KNIME. They are described briefly
identified classes. Deriving a model is based usually on a set of below:
objects for which the respective class allocation is known. With the RapidMiner: RapidMiner (http://www.rapidminer.com) is an
help of a model that has to be determined, objects are to be open source framework which supports various types of data
classified that have no known classification. The models can be mining such as text, web, image or linked open data mining.
determined with the help of neuronal networks, discriminant Through its sophisticated graphical user interface, data mining
analyses or decision trees (Breiman et al., 1984) and random forest processes can be implemented and executed quickly, intuitively
(Chu, 2014). Naïve Bayes methods as well as support vector and without computer programming knowledge (Fig. 9). Further-
machines (SVM) can also be used to classify data (Witten et al., more, RapidMiner offers many necessary pre- and post-process
2011; Chu, 2014). stages in support of data integration, extract, transform and
There are a number of other procedures such as time series loading (ETL), analysis and reporting in one single and coherent
analysis or visualization und evolutionary algorithms. Witten et al. graphical user interface (Hofmann and Klinkenberg, 2013).
(2011) as well as Goele and Chanana (2012) provide a compre- RapidMiner furthermore offers a comprehensive selection of data
hensive overview of these. mining operations as well as an integration of the WEKA and R
libraries. RapidMiner is implemented in Java and can integrate
2.5. Tools for data mining processes developed by users as plug-ins. Moreover, it also offers
an application programming interface (API), through which it can
In recent years, data mining tools have become much more be used in other programs as a Java library.
available, interactive and explorative. This development has been R/Rattle: R (http://www.r-project.org/) and Rattle (R Analytical
pioneered by systems, which enable scientists from different Tool To Learn Easily) http://rattle.togaware.com/) have been used
disciplines without specialized expertise in mathematics and widely as an open source programming language in research for
computer programming to conduct data mining in their research statistical data analysis. R supports many statistical and data
areas. mining techniques, like non-linear modeling, classical statistical
The report of the 14th annual KDnuggets Software Poll tests, time-series analysis, classification clustering, Partial Least
(KDnuggets Annual Software Poll, 2013) reveals the use of open Squares Regression (PLSR), random forest and many more. Using R
source and commercial data mining tools during 2012 and 2013 for data mining requires learning the R programming language, an
(Fig. 3). According to this poll, the open source data mining tools effort that should not be underestimated. However, the R package
RapidMiner and R have been the most frequently used tools. The data mining plugin - Rattle GUI - is available (Williams, 2011) and
report also depicts that 29% of voters only used commercial provides the user with a user-friendly introduction to the methods
software, 30% used open source software and 41% used both of data mining. Rattle uses a graphical user interface for an easy use
commercial and free software. of data mining algorithms based on the R-framework. Rattle GUI
activates data mining functions in R that have been implemented.
3. Open source tools for data mining WEKA: WEKA (Waikato Environment for Knowledge Analysis)
(http://www.cs.waikato.ac.nz/ml/weka/index.html) is a Java
There are some capable open source data mining tools available library with a large selection of sophisticated machine learning
with good application potential in environmental and interdisci- algorithms. WEKA can be used in custom Java applications or via
plinary research even for users without extensive IT and computer the integrated graphical user interface. WEKA contains tools for
programming skills. Hirudkar and Sherekar (2013) provide a data preprocessing, classification, regression, clustering, associa-
comparative analysis of these data mining tools and techniques. tion rules and visualization. WEKA is probably the oldest and most
Fig. 3. Report of the KDnuggets Annual Software Poll: Using data mining tools in 2013xps13#(1880 voters, KDnuggets Annual Software Poll, 2013).
Please cite this article in press as: Lausch, A., et al., Data mining and linked open data – New perspectives for data analysis in environmental
research. Ecol. Model. (2014), http://dx.doi.org/10.1016/j.ecolmodel.2014.09.018
G Model
ECOMOD 7323 No. of Pages 13
successful open source data mining library and software which very clear structure due to its modularity and is therefore easy to
was later integrated as libraries in RapidMiner and R. Since its GUI use without any prior knowledge of computer programming. Its
is rather rudimentary and offers low user comfort, the usage of ease of use is based on “streamed data processing” that is similar to
WEKA tools has fallen sharply in recent years in favor of the knowledge flow mode of WEKA and RapidMiner. All functions
RapidMiner and R. such as input, visualization, data manipulation or mining
KNIME: (Konstanz Information Miner, http://www.knime.org). algorithms are packed into so-called “knots”. These knots can be
Similar to the above, KNIME was developed at the University of moved at will using drag and drop and connected by so-called
Konstanz in 2004 as a machine learning tool. Written in Java, pipes. In the basic version KNIME only offers basic data mining
KNIME aimed to provide high operational modularity, while at the functions such as data import and export, database access, data
same time being simple to upgrade. KNIME provides open source manipulation and visualization as well as decision trees, associa-
data integration under the Gnu Public License (GPL). KNIME has a tion rules, regression, meta-learning or Bayes clustering. These
Table 1
Comparison of open-source data mining tools.
Please cite this article in press as: Lausch, A., et al., Data mining and linked open data – New perspectives for data analysis in environmental
research. Ecol. Model. (2014), http://dx.doi.org/10.1016/j.ecolmodel.2014.09.018
G Model
ECOMOD 7323 No. of Pages 13
basic functions can be extended in KNIME using pre-assembled components and moisture patterns. In this way, important soil
plugins from WEKA or R. KNIME is also a plug-in of the popular process-vegetation patterns and interactions are modeled based
integrated development environment (IDE) “Eclipse”. A simple on data mining methods (Lausch et al., 2013).
comparison of open-source data-mining tools can be seen in The advantages of such spatial data mining techniques arise
Table 1. For more comprehensive comparisons of open-source as from their ability to capture and determine unknown, interesting
well as commercial data mining tools see Hirudkar and Sherekar, and significant patterns and knowledge in spatial as well as
2013. temporal high resolution EO data. The highly multi-dimensional
300–400 spectral bands as well as multiple EO datasets over the
4. Data mining and linked open data for environmental year necessitate partial and fully automated data-mining techni-
research ques for research. Such knowledge about patterns and processes is
an absolute necessity for subsequent modelling and projections
A Web of Science report provides an overview of the use of data about environmental issues Table 3.
mining techniques over the last 20 years (Fig. 4). In 2012, roughly The greatest disadvantage of data mining compared to LOD, (see
10% of all scientific papers used data mining methods. An analysis Section 4) is that with data-mining methods, patterns and
of data mining usage in research areas reveals that data mining knowledge can only be extracted from closed or “silo” data sets,
techniques are primarily used in computer science and engineer- where all of the data required for the analysis are associated
ing. To date, environmental sciences account for only 1.8% of data through known relationships as established in relational
mining usage. databases. Consequently, it is the job of the analyst to couple all
Data mining methods are used in many areas such as of the necessary data and information that could potentially
environmental research and natural hazards. Different applica- influence patterns and processes. Should the analyst make an
tions require specific data mining like spatial data mining for incorrect decision in the process of selecting suitable variables for
natural hazards, text and web mining in social research and image the model, this cannot be taken into account during the modelling
mining in medical research (Table 2). process. Hence, it is possible that new or unknown patterns and
Another important area for data mining is earth observation knowledge are not recorded due to incorrect decisions that have
based on remote sensing. Remote sensing techniques produce an been made previously.
enormous amount of spatial, multi-spectral and multi-dimension-
al earth observation (EO) data. Such data volumes and their 5. Demands of linking future research and data mining
inherent complexity necessitate the automatic recognition of
spatial patterns, knowledge and insight. All of the previously discussed applications of data mining in
Hyperspectral remote sensing data with a spatial resolution of different research areas show that the usage of data mining
0.5 to 3 m and a spectral resolution of 300–400 spectral bands in a techniques is mostly tailored to certain areas. For example, social
spectral range of 400–2500 nm are the most complex EO data sciences primarily use text and web mining for analyzing surveys,
available today. Feilhauer et al. (2014), as well as Lausch et al. markets or socio ecological patterns whereas medical research
(2013, 2013a,b,c) have been able to use such hyperspectral data primarily uses methods of image mining to identify and predict
recorded from aircraft to establish numerous process-pattern cancer or classical data mining to identify relationships between
interactions between soil characteristics and vegetation in the demographics of patients with critical illnesses. Environmental
environmental research (see Table 2). Fig. 5 shows an example research extracts patterns based on multidimensional and
for spatial data mining based on airborne hyperspectral complex spatio-temporal remote sensing data for the automated
data–AISA-EAGLE/HAWK (Airborne Imaging Spectrometer for discovery of spatial knowledge. This mainly area-based data
Applications) at the TERENO–Schäfertal test site. The goal of these mining approach will, however, no longer be expedient.
hyperspectral measurements is to identify important soil charac- In order to gain an understanding of ecological patterns and
teristics like organic matter, texture parameters, clay content or their importance and effects on the driving process in environ-
moisture structures and important biochemical–biophysical vege- mental research, economic, social, environmental and ecological
tation characteristics (chlorophyll, water- cellulose content). The indicators have to be included in modeling and projections in
vegetation parameters are the proxy or functions of soil addition to the ecological patterns.
Fig. 4. Citation report in the Web of Science with the topics - Data mining–(a) Using Data Mining in Research area, (b) Data Mining in Research–Web of Science from
1993xps14#2012.
Please cite this article in press as: Lausch, A., et al., Data mining and linked open data – New perspectives for data analysis in environmental
research. Ecol. Model. (2014), http://dx.doi.org/10.1016/j.ecolmodel.2014.09.018
G Model
ECOMOD 7323 No. of Pages 13
Table 2
Overview of conventional data mining applications in different research areas.
Fig. 5. (a–b) Airborne hyperspectral AISA-Eagle/HAWK remote sensor mounted on Piper, (c) CIR-image from hyperspectral sensors of the AISA-EAGLE/HAWK (AISA-DUAL)
400–2,500 nm with data cube, 367 spectral bands with 2 m recorded ground resolution, date of recording Mai 2012 with a Piper, Region Schäfertal - Bode Catchment, (d)
Spectral curve of ground truth sampling points for soil and vegetation in the test site. (e) Spectral intensity curves of imaging hyperspectral data, (f) Data Mining Model, (g)
Application of the best data mining model on airborne hyperspectral image data for quantification and recognition organic content patterns and (h) Pattern of Clay content.
Please cite this article in press as: Lausch, A., et al., Data mining and linked open data – New perspectives for data analysis in environmental
research. Ecol. Model. (2014), http://dx.doi.org/10.1016/j.ecolmodel.2014.09.018
G Model
ECOMOD 7323 No. of Pages 13
Table 3
Requirements of data mining tools, techniques and platforms for future interdisciplinary research and possible solutions.
Therefore environmental research will face several challenges associated under certain conditions. In Fig. 6 a simple semantic
now and in the future. These are: network is shown using the example of the sentence “Uta makes
coffee for Lutz”. The subject and object stand for nodes, whereas
Dealing with the enormous increase in data volume and the predicates and verbs are represented by edges.
complexity from all disciplines. To create LOD and to carry out LOD queries, clearly defined rules
A very interdisciplinary nature and interactions of various need to be established for the associative relationships between
disciplines like sociology, economics, law, ecology, toxicology the objects. The rules for the statements - Oskar “is a” man
and others. and - Antje “is a” woman would for example be as follows (adapted
As a result of the impact of economic, social, environmental and after Dengler, 2012):
planning decisions, complex data that is currently available has
to be brought into perspective with its historical development if B “is the child of” A and A “is a” man, then A “is the father of” B
and historical change. if B “is the child of” A and A “is a” woman, then A “is the mother
of” B
6. Linked open data–A new data mining approach in research A, B = variables for the objects to be investigated
From the above it is possible to make the following two
LOD is a recent development whereby all data sets that are statements: (1) All men who have children are fathers, and (2) all
freely available on the Internet are interrelated by way of semantic women who have children are mothers. Fig. 7 shows a more
nets and techniques (Dengel, 2012) in an effort to publish these in a complex semantic network to describe the relationships between
machine readable way and to facilitate analyzing them in human- the family members in Oscar’s family. Semantic networks can be
understandable form (Heath and Bizer, 2011). static or change over time. At the same time, semantic networks
can expand very rapidly, in which case the derivation of knowledge
6.1. Semantic methods and networks is required using data mining. Further information on semantic
networks can be found in Dengel (2012) and Reichenberger (2012).
Semantic networks are a visual form of knowledge representa-
tion (Dengel, 2012). With the help of semantic networks, state- 6.2. Concept of LOD
ments about objects using nodes and the relationships between
them can be graphically displayed. This helps the meaning of an The concept of LOD is based on the idea of linking publicly
object (its definition) to become ascertainable through its available data “silos” on the internet by means of semantic
associative relationship with other objects (definitions, data). methods and networks. By linking data, all of the data objects
Associative relationships can express the relationship between (e.g., man, woman) become related to each another. By determin-
two completely independent objects (definitions) that do not have ing a number of rules about these relationships, such inter-linked
to have any connection with each other, but can be loosely data can be “understood” by machines and algorithms, which
enables global data mining approaches and the discovery of truly
new associations, patterns and knowledge.
LOD is based on the Resource Description Framework (RDF)
data model, which formulates syntax and rules about data and
resources as well as their location on the internet. The RDF-model
resembles classic methods for concepts such as the Entity-
Relationship-Model (ERM) or the Unified Modelling Language
(UML) class diagram. RDF models have a formal semantic that is
based on digraphs. The RDF syntax extends the Extensible Hyper
Text Markup Language (XHTML) of the web by semantic
annotation in support of linking up data. Along with RDF there
Fig. 6. Semantic network for the sentence “Ute makes coffee for Lutz”. are also other standards such as the Web Ontology Language
Please cite this article in press as: Lausch, A., et al., Data mining and linked open data – New perspectives for data analysis in environmental
research. Ecol. Model. (2014), http://dx.doi.org/10.1016/j.ecolmodel.2014.09.018
G Model
ECOMOD 7323 No. of Pages 13
Fig. 7. Semantic network to describe the relationships of Oscar’s family members with one another.
(OWL) that enable more complex semantic annotations. More 7. LOD in medical research
detailed literature on the subject can be found in Dengel (2012) as
well as under http://www.w3.org. The semantic search in medical images was first successfully
The LOD approach is not about expanding linkage of documents implemented in the research project THESEUS MEDICO (Lavrac and
via hyperlinks–something which has been common on the Novak, 2013). The objective of LOD was, on the one hand, to
Internet - but, rather, about the “linking of all publically accessible simplify access to online information and data and, on the other
data” together (Heath and Bizer, 2011). Publically accessible data hand, to network data and generate new knowledge by linking
includes many areas of interest, such as weather, events, politics, complex and heterogeneous data.
business, news, geographic spatial data, videos, pictures, images Imaging techniques like Computer Tomography (CT) and
and much more. In this context, Berners-Lee (2006, 2009) refers to Magnetic Resonance Tomography (MRT) are key procedures in
the LOD approach as the next generation of the web. medical diagnostics and treatment and aim to recognize and
diagnose sicknesses early on and treat them correctly. The
Implementation of LOD approaches requires adherence to the symptoms identified in CT and MRT are often difficult to diagnose
four basic components as formulated by Berners-Lee (2006) and as a result of the tremendous complexity stemming from various
Dengel (2012): causes. A targeted treatment is, for this reason, not always possible
because the causes and interactions of symptoms cannot be clearly
1. Use Uniform Resource Identifiers (URI) to uniquely identify data. ascertained. By linking complex and not just stand alone specialist
2. Use the Hypertext Transfer Protocol (HTTP) so that people, web information and data, the linked open data approach offers
agents and data mining tools can access and refer to data. tremendous new potential for future research.
3. A URI has to refer to usable information that can be provided
Fig. 8 shows a conceptual LOD model of the semantic linking of
with the RDF and queried with the Simple Protocol and RDF open data in medical research using the example of disease
Query Language (SPARQL). diagnosis through computer tomography. In the LOD approach,
4. Links to other RDF resources should be established in support of
heterogeneous and complex data like texts, images or lab data are
growing a word wide network of publically available data and semantically linked with each other.
allowing for truly inter-disciplinary data mining.
The advantage of the LOD concept here is that while there is a
“subject core” - in this case the CT/MRT symptom - a number of
According to Berners-Lee and his call in a TED lecture (Berners-
links to other disciplines like sociology, toxicology, biochemistry
Lee, 2009) for “raw data now!” more and more people, institutions,
can be made as a result of linked open data, a fact which makes it
companies and research centers are putting their data on the web
possible to identify patterns in interactions between symptoms
in a machine readable form as per the LOD approach thereby
and complex causes.
making it available to the general public for complex and
interdisciplinary data analysis. In this way, the linked data on
the web results in a worldwide LOD network which according to 8. LOD in environmental research
Bizer et al. (2009) is also referred to as the “linked open data cloud”.
Google, for instance, uses LOD to improve search functions. Coupling of Database–EUROSTAT: Paulheim et al. (2013) devel-
DBpedia is an LOD version of the open and free online encyclopedia oped the first linked open data extension for the open source tool
Wikipedia. In August 2011, the LOD cloud numbered roughly RapidMiner. This example shows how to read and analyze data
31.6 billion RDF Tripel, linked over 389 million RDF links (Dengel, from the linked open data source, Eurostat (Paulheim and
2012). Fürnkranz, 2012). The corresponding RapidMiner workflow
Please cite this article in press as: Lausch, A., et al., Data mining and linked open data – New perspectives for data analysis in environmental
research. Ecol. Model. (2014), http://dx.doi.org/10.1016/j.ecolmodel.2014.09.018
G Model
ECOMOD 7323 No. of Pages 13
Fig. 8. Conceptual LOD Modell of semantic link of open data in medical research using the example of disease diagnosis through magnetic resonance tomography (MRT). The
clinical MRT picture (yellow box) shows the classical symptoms of chronic tension headache (example). This illness is effect by different causes. The importance of causes can
be analysed with data mining of linked open data approach.
(Fig. 9) implements data mining techniques of Eurostat data for and associated (ground truth and meta) data. A world-wide
analyzing patterns based on the semantic linked open data. Based network of remotely sensed data will facilitate rapid data access
on a classification tree as data mining method the semantically and enable a fully automated data processing framework. New
linked LOD data from Eurostat is researched with regard to which tools and cost-effective data mining approaches will make this
factors explain high alcohol consumption in European countries. potential available for researchers of all fields, independent of their
Remote Sensing Research: The growing volume of Earth expertise in computer programming and mathematics (ESA, 2014).
Observation (EO) data (through global monitoring, long-term More examples of applications of semantic networks and LOD are
archives–(European Remote Sensing Satellite - ERS, Environmental shown in Table 4.
Satellite - Envisat) and the associated growth in diversity and
complexity of additional data, methods and models requires the 8.1. Advantages and tools of the LOD approach
creation of data networks and Thematic Exploration Platforms
(TEP, ESA, 2014). The principal idea of TEP is “to move processing to The advantage of the LOD approach relies on the possibilities of
the data, rather than the data to the users” (ESA, 2014). First retrieving and subsequently analyzing data that have been related
successful applications of TEP involve the Geo-spatial Processing through associative links. The data range, the depth of branching
on Demand (URL, 2014), environment. The “Earth Observation and thus the range of topics that are linked with one another
Exploitation Platform (EOXP, Bitto et al., 2014) prototype illustrates depend on the opening, provisioning and linking of data all of
another possible implementation of TEP. TEP in combination with which are constantly growing on the Internet through RDF
LOD (Balazinska et al., 2007) will be the foundation for a world- semantic. Through LOD it will therefore become increasingly
wide inter-connected data archive, which will serve as a base for possible to discover completely new relationships, patterns,
truly global and interdisciplinary data mining of remotely sensed insights and knowledge.
Please cite this article in press as: Lausch, A., et al., Data mining and linked open data – New perspectives for data analysis in environmental
research. Ecol. Model. (2014), http://dx.doi.org/10.1016/j.ecolmodel.2014.09.018
G Model
ECOMOD 7323 No. of Pages 13
Fig. 9. Application for linking open data of EUROSTAT based on the open source data mining tool RapidMiner. (a, b) Process procedure of data mining in RapidMiner, (c) Singel
results of LOD data mining analysis from Eurostat based on LOD, Electricity Production Capacity [MWh]/Country, (d) Data mining application–Classification tree of EUROSTAT
data to explain the high alcohol consumption in different countries (modified after Paulheim and Fürnkranz, 2012).
Table 4
Overview of LOD applications in different research areas.
Please cite this article in press as: Lausch, A., et al., Data mining and linked open data – New perspectives for data analysis in environmental
research. Ecol. Model. (2014), http://dx.doi.org/10.1016/j.ecolmodel.2014.09.018
G Model
ECOMOD 7323 No. of Pages 13
Table 5
Commercial and open source tools for storing, publishing and managing LOD.
Another advantage of LOD relies on the fact that the data are The LOD approach is a new concept for data integration with a
very interdisciplinary and there is a diversity of possible data huge potential for analyzing interdisciplinary, complex and
formats that can be integrated into a LOD analyses. Because the distributed data sets. According to Berners-Lee (2006, 2009)
coupling of LOD data is standardized by means of RDF rules and LOD will be the next generation of the Internet by associating
syntax, no further steps are required for data transformation or publicly accessible data in a standardized way. One huge advantage
data integration. of LOD compared to traditional data mining is that the analyst does
The development of tools that support LOD queries is not have to link associated data prior to analysis, but that machine
constantly growing. Table 5 contains commercially available and learning algorithms will be able to analyze an open set of
open source tools for storing, publishing and managing open interdisciplinary databases.
linked data. The LOD approach will facilitate to find novel insights from and
novel associations between interdisciplinary and complex data
9. Conclusion sets and based on such knowledge we will be able to solve complex
problems and make better predictions.
The rapid development of information technology in the 21st
Century is associated with an enormous increase in data volumes
and data complexity. The insights gained from such data References
complexity and data volumes are not holding up to our expect-
ations as modern data analyzing techniques are not yet widely Alonso-Peral, M.M., Sun, C., Millar, A.A., 2012. MicroRNA159 can act as a switch or
tuning microRNA independently of its abundance in Arabidopsis. PLoS ONE 7,
known and used. e34751.
Data mining offers one promising solution to this situation with Alpha, W., 2013. http://www.wolframalpha.com/about.html (Date: 1.12.2013).
numerous approaches for knowledge discovery in multi-dimen- J. Aryal A. Morshed R. Dutta 2014. Land cover class extraction in GEOBIA using
environmental spatial temporal ontology Proceeding of the South-Eastern
sional and complex data. Unfortunately, data-mining techniques European Journal of Earth Observation and Geomatics Special Issue, 21–24 May
are currently exclusively used in analytical research areas such as 2014, Thessaloniki, Greece, ISSN 2241 2014 429–434.
computer science, physics or mathematics and often require Balazinska, M., Deshpande, A., Franklin, M., Gibbson, P.B., Nath, J.G., Hansen, M.,
Liebhold, M., Szalay, A., Tao, V., 2007. Data Management in the Worldwilde
extensive programming expertise. Sensor Web. IEEE Comput. Soc. 1536–1568.
The objective of this paper was therefore to make researchers Battistini, A., Segani, S., Manzo, G., Catani, F., Casagli, N., 2013. Web data mining for
with little or no experience in computer programming aware of the automatic inventory of geohazards at national scale. Appl. Geogr. 43, 147–158.
R. Baker T. Barnes J.E. Beck. Educational Data Mining. 1st. International Conference
significance and tremendous potential of data mining in a on Educational Data Mining, Proceedings. Montreal, Quebec, Canada.
demonstrative manner. To achieve this, data mining processes, Begum, S.H., 2013. Data Mining Tools and Trends–An Overview. International. J.
types and methods were portrayed in a comprehensible manner Emerg. Res. Manage. Technol. 2278–9359, 6–12.
Bensberg, F., Weiß, T., 1999. Web Log Mining als Marktforschungsinstrument für das
and important references were provided to existing open source
World Wide Web. Wirtschaftsinformatik 41, 426–432.
data mining tools, which are characterized by their simple and Berners-Lee T., 2006. Linked Data. W3C Design Issue. URL http://www.w3 org/
intuitive use in support of complex data analysis. DesignIssues/LindedData.html, (Date: 20.07.2014).
Good, simple and easy-to-learn open source data mining tools Berners-Lee T., 2009. The Next Web. URL http://www.ted.com./talks/tim_berner-
s_lee_on_the_next_web.html, (Date: 20.07.2014).
are RapidMiner and KNIME, which enable very complex data- Bhise, R.B., Thorat, S.S., Supekar, A.K., 2013. Importance of Data Mining in Higher
mining processes to be generated without any previous computer Education System 6, 18–21.
programming knowledge with the help of intuitive graphical user Bitto, F., Castracane, P., Graziano, A., Lapaola, M., Farres, J., Zelli, C., 2014. The Earth
Observation Exploitation Platform (EOXP) Prototype, Technical Note, March
interfaces. Many data mining functions can be added to those tools 2014. https://earth.esa.int/documents/10174/1157689/EOXP-ACS-ESA-IIM
by means of plugins and connectors to WEKA and R/Rattle are also (Date: 20.07.2014).
available. Bizer, C., Heath, T., Berners-Lee, T., 2009. Linded Data–The Story So Far. Int. J. Seman.
Web and Info. Syst. 5, 1–22.
LOD is a novel approach that can help us to discover new Boic, B., Peters-Anders, J., Schimak, G., 2014. Ontology Mapping in Semantic Time
insights and knowledge from truly multi-disciplinary and inter- Series Processing and Climate Change Prediction. Proceedings of International
linked data sets publically available on the internet. This paper Environmental Modelling and Software Society (iEMSs), San Diego, http://
www.iemss.org/society/index.php/iemss-2014-proceedings.
comprehensively explains the concept of LOD. Reference is made to
Brady, S.M., Provart, N.J., 2009. Web-queryable large-scale data sets for hypothesis
the basic elements of semantic networks, LOD, RDF, SPARQL and generation in plant biology. Plant Cell 21, 1034–1051.
OWL and supported by examples whenever possible. In compari- Breiman, L., Friedman, J.H., Olshen, R.A., Stone, C.J., 1984. Classification and
Regression Trees. Monterey, CA, Wadsworth.
son with traditional data mining techniques, reference is also made
R. Burget V. Uher R. Cervenec Rapid Image Feature Extraktion and Mining Fischer, S
to the advantages of LOD as well as existing commercial and open- Mierswa, I. (Eds.) Proceedings of the 2nd Rapid Miner Community Meeting and
source LOD tools. Conference, RCOMM Aachen 2011 2011 123–132.
Please cite this article in press as: Lausch, A., et al., Data mining and linked open data – New perspectives for data analysis in environmental
research. Ecol. Model. (2014), http://dx.doi.org/10.1016/j.ecolmodel.2014.09.018
G Model
ECOMOD 7323 No. of Pages 13
Chu, W.W., 2014. Data Mining and Knowledge Discovery for Big Data. Methodolo- sensing-based indicators in laboratory, field and landscape. Environmental
gies, challenge and Opportunities. Springer Verlag Berlin Heidelberg. Monitoring Assess. 185, 1215–1235.
Computer, 2013. Statista–Das Statistik-Portal. Das Internet im Jahr 2015. Computer Lausch, A., Pause, M., Schmidt, A., Salbach, C., Gwillym Margianto, S., Mehrbach, I.,
Bild 22, 116-117. 2013b. Temporal imaging hyperspectral monitoring of chlorophyll: LAI and
Curtarolo, S., Hart, G.L.W., Nardelli, M.B., Mingo, N., Sanvito, S., Levy, O., 2013. The water content of barley during a growing season. Can. J. Remote Sensing 39,
high-throughput highway to computational materials design. Nat. Mater. 12, 191–207.
191–201. Lausch, A., Zacharias, S., Dierke, C., Pause, M., Kühn, I., Doktor, D., Dietrich, P.,
Del Rio N., Villanueva-Rosales, N., Pennington, D., Benedict, K., Stewart, A., Grady, C. Werban, U., 2013c. Analysis of vegetation and soil pattern using hyperspectral
J., 2014. ELSEWeb Meets SADI: Supporting Data-to-Model Integration for remote sensing, EMI and Gamma ray measurements. Vadose Zone J doi:http://
Biodiversity Forecasting. Discovery Informatics: AI Takes a Science-Centered dx.doi.org/10.2136/vzj2012.0217.
View on Big Data AAAI Technical Report FS-13–01, http://www.aaai.org/ocs/ Lavrac, N., Novak, P.K., 2013. Relational and Semantic Data Mining for Biomedical
index.php/FSS/FSS13/paper/viewFile/7631/7488. Research. Informatica 37, 35–39.
Dengel, A., 2012., Semantische Technologien. Spektrum Akademischer Verlag. Ledolter, J., 2013. Data Mining and Business Analytics with R., Wiley & Sons. New
J. Dörre P. Gerstl R. Seiffert Text Mining. In: Hippner H., Küsters U., Meyer, M., Wilde, Yersey, Hoboken.
K.D. (eds.): Handbuch Data Mining im Marketing. Wiesbaden 2001 2001 Maity, B., Chatterjee, B., 2012. Forecasting GDP Growth Rates of India.: An Empirical
465–488. Study. Int. J. Econom. Manage. Sci. 1 (9), 52–58.
ESA, 2014. THEMATIC EXPLOITATION PLATFORMS. Official Tender of the ESA, 2014. Merz, B., Kreibich, H., Lall, U., 2013. Multi-variate flood damage assessment: a tree-
AO7870, http://emits.sso.esa.int/emits/owa/emits.main, (Date: 20.07.2014). based data-mining approach. Natural Hazards Earth Syst. Sci. 13, 53–64.
Fayyad, U., Shapiro, G., Smyth, P., 1996. The KDD process for extracting useful Meyer, M. Lüling, M., 2003. Data Mining in Forschung und Lehre in Deutschland 4,
knowledge from volumes of data. Commun. ACM 39, 27–34. Schriften zur Empirischen Forschung und Qualitativen Unternehmensplanung
Feilhauer, H., Doktor, D., Lausch, A., Schmidtlein, S., Schulz, G., Stenzel, S., 2014. 15, 2003. Ludwig-Maximilians-Universität München. http://www.imm.bwl.
Mapping Natura 2000 habitats and their local variability with remote sensing. uni-muenchen.de/forschung/schriftenefo/ap_efoplan_15 pdf. (Date: 1.12.2013).
Appl. Veg. Sci doi:http://dx.doi.org/10.1111/avsc.12115. Mohanty, A.K., Senepati, M.R., Lenka, S.K., 2013. A novel image mining technique for
Goele,S., Chanana, N., 2012. Data Mining Trend In Past, Current And Future. classification of mammograms using hybrid feature selection. Neural Comput.
International Journal of Computing & Business Research, In Proc. I-Society 2012. Appli doi:http://dx.doi.org/10.1007/s00521–012-0881-x.
http://www.researchmanuscripts.com/isociety2012/15. pdf, (Date: Murthy, D., Gross, A., Takata, A., Bond, S., 2013. Evaluation and Development of Data
20.07.2014). Mining Tools for Scial Netwirk Analysis. Özyer, T. (eds.) Mining Social Networks
Goodyear, P., Banks, S., Hodgson, V., & McConnell, D. 2004. Research on networked and Security Informatics Lecture Notes in Social Networks. Springer Science
learning: An overview. In P. Dillenbourg, M., Baker, C., Bereiter, Y. Engeström, G., +Business Media Dordrecht, DOI 10.1007/978–94-007–6359-3_10. 183–202.
Fischer, H. U. Hoppe, D. McConnell (Eds.), Advances in Research on Networked Nieland, S., Moran, N., Kleinschmit, B., Förster, M., 2014. Using semantic-based
Learning,Springer, 1–9. spatial reclassification for interoperable data management in Natura 2000 mon-
J.K. Hart K. Martinez Environmental Sensor Netzworks: A revolution in the earth itoring. Proceedings of International Environmental Modelling and Software
science? Earth-Science Reviews 78 2006 177–191. Society (iEMSs), San Diegeo, http://www.iemss.org/society/index.php/iemss-
Hautier, G., Jain, A., Ong, S.P., 2012. From the computer to the laboratory: materials 2014-proceedings.
discovery and design using first-principles calculations. J. Mater. Sci. 47, Paulheim, H., Mitichkin, E., Ristoski, P., Bizer, C., 2013. RapidMiner Linked Open Data
7317–7340. Extension. Manual Version 1.2,11/29/13. University of Mannheim, Data and Web
Heath, T., Bizer, C., 2011. Lined Data: Evolving the Web into a Global Data Space. Science Group, http://dws.informatik.uni-mannheim.de/fileadmin/lehr-
Morgan & Claypool, 1st. Edition. http://linkeddatabook.com/, (Date: 1.12.2013). stuehle/ki/research/RapidMinerLODExtension/RapidMinerLODExtensionMa-
C.A. Henson J.K. Pschorr A.P. Sheth K. Thirunarayan SemSOS: Semantic Sensor nual.pdf (Date: 15.12.2013).
Observation Service. Proceedings of the 2009 International Symposium on Paulheim, H., Fürnkranz, J., 2012. Unsupervised Generation of Data Mining Features
Collaborative Technologies and Systems (CTS 2009), Baltimore, MD May 2009 from Linked Open Data. In: International Conference on Web Intelligence,
18–22. Mining, and Semantics (WIM’s).
Hirudkar, A.M., Sherekar, S.S., 2013. Comparative Analysis of Data Mining Tools and H. Rekow Einführung in das Data Mining. GRIN Verla–Publisher 2013.
Techniques for Evaluating of Database Systems. Int. J. Comput. Sci. Appl. 6, Reichenberger, K., 2012. Kompendium semantischer Netz: Konzepte, Technologie,
232–237. Modellierung. X.media.press. Springer.
Hofmann, M., Klinkenberg, R., 2013. RapidMiner–Data Mining Use Cases and T.A. Runkler Data Mining. Vieweg-Teubner Verlag, Wiesbaden. Saake, G., Heuer,
Business Analytics Applications. CRC Press, Taylor & Francies Group. A.,1999. Datenbanken–Implementierungstechniken, ISBN 2010 3–8266–0513–
Ikonomovska, E., Gama, J., Dzeroski, S., 2011. Learning model trees from evolving 6.
data streams. Data Mining Knowledge Discov. 23, 128–168. Saravanan, D., Srinivasan, S., 2013. Matrix based indexing technique for video data. J.
Jain, R., Singh, D., 2013. Data Mining and Analysis of Economic Data. International of Comput. Sci. 9, 534–542.
Advandced Research in Computer Science and Software Engineering–Interna- Schmid, T., 2013. Wie man zwischen den Zahlen liest. Data-Mining und
tional Journal of Advanced Research in Computer Science and Software computergestützte Vorhersagen am Beispiel Bioinformatik. In: Arbeitstitel–
Engineering 3, 683–688. Forum für Leipziger Promovierende 5, 13–29.
Joseph, M.V., Sadath, L., Rajan, V., 2013. Data Mining: A Comparative Study on S. Shekhar P. Zhang Y. Huang Spatial Data Mining Maimon, O., Rokach, L. (eds.) Data
Various Techniques and Methods. Int. J. Adv. Res. Comput. Sci. Software Eng. 2, Mining and Knowledge Discovery Handbook, Springer Science+Business M LLC
106–113. 2010 837–854.
Kalampokis, E., Tambouris, E., Tarabanis, K., 2013. Linked Open Government Data Tanner, R., 2013. Data Mining–das etwas andere Eldorado. Technologie IT-Methoden
Analytics, M.A., Wimmer, M. Janssen, and H.J. Scholl (Eds.): EGOV 2013. LNCS 8, 37–42.
8074, 99–110. IFIP http://dx.doi.org/10.1007/978–3-642–40358-3_9. Thushari, S., Vilas, W., Hitesh, N.S., 2012. Disaster mitigation and preparedness using
KDnuggets Annual Software Poll, 2013 KDnuggets Annual Software Poll: Using Data linked open data. J. Ambient Intelligence Humanized Comput doi:http://dx.doi.
Science software in 2013. (http://www.kdnuggets.com/2013/06/kdnuggets- org/10.1007/s12652–012-0128–9.
annual-software-poll-rapidminer-r-vie-for-first-place.html, (Date: 01.12.2013). W.G. Touw, J.R. Bayjanov, L. Overmars, L. Backus, J. Boekhorst, M. Wels, S.A.F.T.
M. Koubarakis M. Karpathiotakis K. Kyzirakos C. Nikolaou S. Vassos G. Garbis M. Hijum. Data mining in the Life Sciences with Random Forest: a walk in the park
Sioutis K. Bereta S. Manegold M.L. Kersten M. Ivanova H. Pirk Y. Zhang C. Kontoes or lost in the jungle? Brief Bioinform 14, 315–326 2013 10.1093/bib/bbs034.
I. Papoutsis T. Herekakis D. Mihail M. Datcu G. Schwarz O. Dumitru D. Espinoza- URL, 2014; http://www.adb.org/themes/urban-development.
Molina K. Molch U.D. Giammatteo M. Sagona S. Perelli E. Klien T. Reitz R. Gregor Villa, F., Athanasiadis, I.N., Rizzoli, A.E., Emilio, A., 2009. Modelling with knowledge:
Building virtual earth observatories using ontologies and linked geospatial data A review of emerging semantic approaches to environmental modelling.
Krötzsch, M. Straccia, U., (Eds.) Web Reasoning and Rule Systems, RR 2012, LNCS Environ. Model. Software 24, 577–587.
7497 2012 Springer–Verlag Berlin Heidelberg, 229–233 Vintrou, E., 2013. Data Mining, A Promising Tool for Large-Area Cropland Mapping.
Kumar, R.A., Saravanan, D., 2013. Content based image reterival using color IEEE J. Sel. Top. Appl. 99. 1–7.
histogram. Int. J. Comput. Sci. Inform. Technol. 4, 242–245. Williams, G., 2011. Data Mining with Rattle and R: The Art of Excavating Data for
Kreissl, R. 2013. Datenspuren: Komplette Umkehr der Beweislast. New Scientist, Knowledge Discovery (Use R!), Springer Science Business Media, LLC,2011.
http://irissproject.eu/?p=325, (Date: 22. Februar 2013). I.H. Witten ,F. Eibe. 2001. Data Mining. Praktische Werkzeuge und Techniken für das
Landeghem, S.V., De Bodt, S., Drebert, Z., Inzé, D., Van de Peer, Y., 2013 The Potential maschinelle Lernen. Carl Hanser Verlag München Wien.
of Text Mining in Data Integration and Network Biology for Plant Research: A I.H. Witten F. Eibe M.A. Hall Data Mining Practical Maschine Learning Tools and
Case Study on Arabidopsis Plant Cell 25, 794–807. Techniques 2011 Morgan Kaufmann Publishers.
Lausch, A., Heurich, M., Gordalla, D., Dobner, H.-J., Gwillym-Margianto, S., Salbach, Wu, F., Zhan, J., Yan H., Shi C., Huang, J., 2013. Land cover mapping based on
C., 2013. Forecasting potential bark beetle outbreaks based on spruce forest multisource spatial data mining approach for climate simulation: a case study
vitality using hyperspectral remote-sensing techniques at different scales. in the farming-pastoral ecotone of North China, Advances in Meteorology,
Forestest Ecol. Manage. 308, 76–89. http://dx.doi.org/10.1155/2013/520803, (Date: 01.012.2013).
Lausch, A., Pause, M., Merbach, I., Zacharias, S., Doktor, D., Volk, M., Seppelt, R., Yuan, J., Yang, M.,Wu, Y., 2011. Mining discriminative co-occurrence patterns for
2013a. A new multi-scale approach for monitoring vegetation using remote visual recognition. CVPR. 2777–2784.
Please cite this article in press as: Lausch, A., et al., Data mining and linked open data – New perspectives for data analysis in environmental
research. Ecol. Model. (2014), http://dx.doi.org/10.1016/j.ecolmodel.2014.09.018