Content uploaded by Borja Espejo García
Author content
All content in this area was uploaded by Borja Espejo García on Nov 06, 2019
Content may be subject to copyright.
Full Terms & Conditions of access and use can be found at
http://www.tandfonline.com/action/journalInformation?journalCode=tgis20
Download by: [University of Zaragoza] Date: 02 May 2017, At: 03:44
International Journal of Geographical Information
Science
ISSN: 1365-8816 (Print) 1362-3087 (Online) Journal homepage: http://www.tandfonline.com/loi/tgis20
Aggregation-based information retrieval system
for geospatial data catalogs
Javier Lacasta, F. Javier Lopez-Pellicer, Borja Espejo-García, Javier Nogueras-
Iso & F. Javier Zarazaga-Soria
To cite this article: Javier Lacasta, F. Javier Lopez-Pellicer, Borja Espejo-García, Javier
Nogueras-Iso & F. Javier Zarazaga-Soria (2017): Aggregation-based information retrieval system
for geospatial data catalogs, International Journal of Geographical Information Science, DOI:
10.1080/13658816.2017.1319949
To link to this article: http://dx.doi.org/10.1080/13658816.2017.1319949
Published online: 02 May 2017.
Submit your article to this journal
View related articles
View Crossmark data
RESEARCH ARTICLE
Aggregation-based information retrieval system for
geospatial data catalogs
Javier Lacasta , F. Javier Lopez-Pellicer , Borja Espejo-García, Javier Nogueras-
Iso and F. Javier Zarazaga-Soria
Aragon Institute of Engineering Research (I3A), Universidad de Zaragoza, Zaragoza, Spain
ABSTRACT
Geospatial data catalogs enable users to discover and access
geographical information. Prevailing solutions are document
oriented and fragment the spatial continuum of the geospatial
data into independent and disconnected resources described
through metadata. Due to this, the complete answer for a query
may be scattered across multiple resources, making its discovery
and access more difficult. This paper proposes an improved infor-
mation retrieval process for geospatial data catalogs that aggre-
gates the search results by identifying the implicit spatial/thematic
relations between the metadata records of the resources. These
aggregations are constructed in such a way that they match better
the user query than each resource individually.
ARTICLE HISTORY
Received 24 August 2016
Accepted 12 April 2017
KEYWORDS
Geospatial data catalog;
information retrieval;
catalog service for the web;
spatial data infrastructure
1. Introduction
Geographical information is commonly used by organizations, institutions and common
citizens for daily work and leisure activities. In the last years, the number, variety and
goals of geographical data creators and users have increased, thanks to the progressive
cost reduction of the technologies needed for acquiring, processing, analyzing, acces-
sing, presenting and transferring geographical information (Anderson and Gaston 2013,
Paneque-Gálvez et al.2014). Part of this cost reduction is the result of more than two
decades of work by public and private initiatives to promote the generation and use of
geographical information through spatial data infrastructures (SDI).
The geospatial data catalogs are responsible for facilitating the location and access to
spatial resources in SDIs (OGC 2007a). The creation of international standards has
facilitated its adoption. Some examples are the International Organization for
Standardization metadata standards for geographical data (ISO/TC 211 2014,2016),
the Open Geospatial Consortium (OGC) standards for discovering geographical informa-
tion on the web (OGC 2007b) and the implementing rules of the European INSPIRE
Directive that ensure the interoperability of European SDIs (Nogueras-Iso et al.2009).
Technologically, geospatial data catalogs are similar to digital libraries that provide
access to textual documents, images or any other kind of resource described with a
metadata record (Smith 1996). The most basic geospatial data catalogs provide a text/
CONTACT Javier Lacasta jlacasta@unizar.es
INTERNATIONAL JOURNAL OF GEOGRAPHICAL INFORMATION SCIENCE, 2017
https://doi.org/10.1080/13658816.2017.1319949
© 2017 Informa UK Limited, trading as Taylor & Francis Group
keyword-based search system and a location-based search component to filter and sort
the resources by their spatial features (Göbel and Klein 2002). This kind of query is
usually named ‘concept at location’query in the literature (Hübner et al.2004). In the
geographical context, the concepts in a ‘concept at location’query are the themes of the
resources. The text-based search usually provides free-text queries on the metadata
records, and the selection of terms from controlled vocabularies. The location-based
search component usually allows constraining the user query to an area defined by
coordinates or by geographic identifiers. The answer always consists of a metadata
record list with the resources that partially fulfill the query restrictions, sorted by some
similarity criteria.
These approaches are valid in many situations but they have limitations. From a
spatial perspective, geographical information forms a continuum that covers the Earth
surface. However, the data creators divide it into independent resources that cover
different spatial and thematic extents. This division is usually done to fulfill the producer
goals and they ignore the nature of the stored information. For example, governments
develop geographical resources that cover their country surface, but there are several
geographical features such as river basins that are shared between countries. In the case
of rivers, Wolf et al.(1999) estimated that 45.3% of the land surface corresponds with
river basins covering more than one country. In this context, the organization of the
information into datasets spatially delimited by the boundaries of countries and other
kinds of administrative regions enters into conflict with users that need continuous data
(e.g. a pan European provider of road maps, or a flood risk manager covering different
autonomous regions in a country). Moreover, data belonging to related topics in the
same area may be scattered across different datasets. That is, this producer-oriented
approach may enter into conflict with users who need continuous data.
The pan-European INSPIRE geoportal
1
is an example of a system with the previous
limitations. This geospatial data catalog was created for providing support for the
implementation of the INSPIRE Directive. It is a state-focused geospatial data catalog
where each member state describes its geospatial resources. This focus may lead to
incomplete answers for some kind of queries. For example, Figure 1 shows the answer to
a query about ‘road networks’in an area between Italy and France. This kind of query
will always produce incomplete results because there are not resources in the catalog
covering both countries. In this example, the first result describes exclusively the road
networks in France and the second one describes only those in Italy. This data partition
makes ranking an unhelpful feature. Each result only provides a part of the required
information, and the user is forced to review all of them to compose a set of suitable
Figure 1. Example of query answer from the INSPIRE geospatial data catalog.
2J. LACASTA ET AL.
road resources that fulfill his needs. This review task is challenging because the lack of
feedback about how the search parameters define the search results makes difficult the
comparison and interpretation of results (Göbel and Klein 2002). Moreover, in the same
way as there is spatial fragmentation, there is also thematic fragmentation. As each
resource may contain only a small set of the themes of the information available about
an area, in multi-theme queries, the results will also be thematically fragmented.
Nowadays, the geospatial data catalogs of public institutions (such as the INSPIRE
geoportal) are the most technologically advanced, but all of them have similar problems
in terms of results interpretation.
The main contribution of this paper is an improved information retrieval (IR) process
for geospatial data catalogs that aggregates the search results by identifying the implicit
spatial/thematic relations between the metadata records of the resources. These aggre-
gations are constructed in such a way that they match better the user query than each
resource individually. The returned aggregations are composed of metadata records that
describe resources that complement each other and fill the spatial gaps that each
individual resource has for each queried theme. This paper is focused on analyzing
the suitability of the aggregation of the metadata records provided as query results in
the geospatial data catalog context. To analyze the result composition issues, other IR
issues such as terminological heterogeneity or the use of imprecise spatial references in
user queries are left aside. The system performance is evaluated comparing the behavior
of the proposed IR system with another one that is similar to those used in prevalent
geospatial data catalogs.
The rest of the paper is organized as follows. Section 2 reviews other works related to
IR systems for geospatial data catalogs. Section 3 explains the IR issues that we analyze
in this paper. Section 4 describes the proposed IR system, which is evaluated through a
series of experiments described in Section 5.Section 6 discusses the obtained results.
Finally, the paper ends with some conclusions and outlook on future work.
2. State of the art
There are many works that have proposed IR improvements through better similarity
measures for result ranking, and through the increase of the metadata and query
description quality. This section reviews a selection of these works.
Related to the definition of a spatial similarity measure between resources for ranking
purposes, on the web context, Watters and Amoudi (2003) propose as a ranking factor
for queries the distance between the place where the user is located and the place
where the web server with the relevant data is located. They translate URLs into the
spatial coordinates of the place where the web domain is situated and they use the
linear distance of these coordinates to the user spatial location as a ranking factor of the
results. Asadi et al.(2005) analyze the different types of textual queries that involve
spatial information and describe how to adjust their ranking formulas. They review direct
queries about facts, local queries that restrict the relevant results to those describing an
area and location-based queries where the objective is to locate specific entities in an
area (e.g. train stations).
Focused on geospatial data catalogs, Larson and Frontiera (2004) describe a statisti-
cally based ranking formula for geometry-based spatial queries. They make a review of
INTERNATIONAL JOURNAL OF GEOGRAPHICAL INFORMATION SCIENCE 3
previous spatial-based ranking formulas and propose a statistical measure that includes
a corrective factor to deal with the problems caused by the imprecise definition of
bounding boxes in border areas such as coasts. Lanfear (2006) suggests another ranking
method for spatial features in geospatial data catalogs that takes into account the
overlap between the query area and the resource, and the dimensions of the area
outside the overlap. More recently, Renteria-Agualimpia et al.(2016) detect incoheren-
cies in metadata collections by comparing the explicit geographical extension defined
by coordinates and an implicit one defined by geographic identifiers found in metadata
records. Their use of the Hausdorffdistance to detect how similar are two geometries
can be directly used in the IR context to determine a spatial similarity measure of a
resource with the user query.
In addition to the previous works, there are ranking proposals that focus on the
integration of different relevance measures. Göbel and Klein (2002) propose a linear
ranking formula that in addition to the similarity with the spatial feature of the query
(both coordinate and gazetteer based), it includes the degree of thematic coincidence
and temporal overlap. Martins et al.(2005) compare different approaches to generate a
combined ranking value from individual spatial and thematic distances. This comparison
includes the use of a linear combination, the product, the maximum similarity and a
step-linear function. Finally, Megler and Maier (2011) present a ranking method for
integrating spatial and temporal query features based on the mean between the spatial
and temporal distances to the center of the selected period or selected area.
Another approach frequently used to obtain better IR systems for geospatial data
catalogs has been to improve the resource and the query descriptions. Lieberman (2006)
describes an SDI architecture where online resources are able to self-describe them-
selves. This solution requires a semantic facade on top of OGC standard services that
describes their content through the use of ontologies. Lutz et al.(2009) propose instead
a semantic catalog where geospatial resources are described using roles and concepts
from a domain ontology. Somehow related, Janowicz et al.(2010) propose a transparent
semantic layer for SDI. They annotate the resource descriptions using ontologies and
they relate these ontologies through a reasoning service implemented as a profile of an
OGC catalog and a processing service, respectively. Finally, Florczyk et al.(2010) add
semantics into an SDI catalog with a linked data administrative geography ontology that
is used for data integration and referencing geographic themes.
Related to the query description improvement, other works have focused on the
identification of textual patterns describing locations or location-based references (e.g.
north of X). Works such as Sallaberry (2013), Ferrés and Rodríguez (2015) and Kim et al.
(2017) show that the identification of textual patterns describing locations can greatly
improve the quality of the results when spatial description in metadata records is
textual. However, in geospatial data catalogs, these solutions are less relevant because,
by design, the metadata records of spatial resources specify their spatial limits as
coordinates.
Our process focuses on automatically providing improved aggregated search results
for geospatial data catalogs by using raw metadata. There have been other proposals
that perform aggregations of resources at the data level but they require either human
intervention or an extra layer of complexity such as adding domain knowledge in the
form of ontologies. For example, Hübner et al.(2004) describe an ontology-based
4J. LACASTA ET AL.
reasoning system that integrates heterogeneous geographical information in ‘concept at
location in time’queries. The user employs provided ontologies to define a query and
the system returns a list of resources sorted by relevance. Then, the system facilitates its
visualization and integration. Similarly, Lutz and Klien (2006) propose a retrieval system
in which features published at Web Feature Service (WFS)s are described in terms of a
shared domain ontology. This system offers a user interface that allows formulating
queries using such ontology. A different approach can be found in Latre et al.(2009).
This work describes a retrieval system that identifies non-explicit relations between
hydrologic feature types published at WFSs and uses this knowledge to expand results.
Finally, Zhu et al.(2015) describe a user-focused spatial data analysis service that unifies
the access to heterogeneous data by creating linked layers after parsing user requests.
3. Spatial and thematic issues in geospatial data catalogs
This section reviews the IR systems used in a representative set of geospatial data
catalogs and describes their features and issues. We have analyzed the pan-European
INSPIRE catalog, and the national catalogs in USA
2
(GeoPlatform), Spain
3
(IDEE), United
Kingdom
4
(Data.Gov) and Canada
5
(GeoDiscovery). Below, we present an analysis of their
user query interfaces and how they answer when the queries include spatial and
thematic constraints. Then, we describe the issues that the process described in this
paper tries to correct.
3.1. Prevalent approaches
Table 1 shows the type of search and ranking provided by each analyzed system. All the
analyzed systems provide free-text search and some kind of controlled topic list or
faceted solution. Additionally, their advanced search components focus on specific
metadata fields, such as the resource type or data format. Among them, only the
Spanish and Canadian systems offer temporal search (periods of time). Regarding the
spatial search, the queries in the UK catalog cannot include textual and spatial features
simultaneously. The remaining systems provide a bounding box-based spatial search
component that can be combined with other query elements.
In order to determine how the search process is performed, we have analyzed the
result of queries using controlled fields, queries using free-text fields, queries using a
spatial bounding box and queries with the three restrictions. The query terms have been
selected so they return multiple resources (e.g. ‘road network’in INSPIRE geoportal) and
we have counted the occurrences of the query terms in each of the obtained metadata
records and the percentage of spatial area in query that they cover.
Table 1. Search and ranking features.
Catalog Country Type of search Type of ranking
INSPIRE EU Spatial, free text, term cloud, topics Relevance
GeoPlatform USA Spatial, free text, facets Relevance, popularity
IDEE Spain Free text, topics, date Rating, relevance, popularity
Data.gov.uk United Kingdom Spatial or free text, facets Relevance, popularity
GeoDiscovery Canada Spatial, free text, date, topics Relevance
INTERNATIONAL JOURNAL OF GEOGRAPHICAL INFORMATION SCIENCE 5
Through this analysis, we have found that the systems do ‘AND’style queries when
the queries involve two or more types of constraints (spatial, controlled or free text).
That is, the responses only contain records that match all the constraints. The results can
be sorted according to a relevance rank (some of them also include popularity and user
rating) or alphabetically by different fields (e.g. title). Regarding the relevance rank, the
number of occurrences of the query term in any part of a metadata record is used as a
ranking factor (the metadata records with more query term occurrences are first).
However, when the query involves controlled fields, only the existence is taken into
account as ranking factor. We also tried to identify if any of the systems uses ontologies,
or any other kind of formal model, for query expansion or refinement. However, since all
the results in the tested systems contain the used query terms, it seems that queries
have not been expanded with additional terms such as synonyms, hypernyms or
hyponyms. In the systems supporting spatial restrictions, the more the query area and
the geographical extent of a resource overlap, the better its rank is.
We have been unable to identify the exact ranking formula used for combining the
spatial and textual rankings in the systems that support ‘concept at location’style
queries (INSPIRE, GeoPlatform and GeoDiscovery). However, we have detected that, in
these systems, a spatially closer resource is ranked first even if it has far fewer occur-
rences of the textual query terms. This indicates that the ranking weight given to the
spatial similarity is higher than the used one for the textual similarity.
The functionalities offered by these systems seem suitable in many situations but
they are problematic when performing queries about multiple themes in an area cross-
ing multiple countries or regions. In these cases, the results obtained are similar to those
described in Figure 1. That is, the results are partially relevant and none can be
considered a complete answer.
3.2. Data fragmentation issues
The search features identified are very common and they can be found in digital libraries
outside the spatial field (e.g. Europeana
6
). It is important to note that the reviewed
systems manage the resources as independent entities. However, in the geospatial con-
text, the resources are related by the themes they cover and by spatial proximity (all this is
indicated in their metadata records). If these relations are not taken into account, when a
user query does not fittheartificial divisions (spatial and thematic) of the data continuum
performed by the data creators, the catalogs will return incomplete results. Depending on
the user query and the spatial and thematic data fragmentation, the answer may suffer
from under-coverage, over-coverage and partial coverage of the results with respect to
the query. Below, we are going to characterize each of these issues.
The first issue is the under-coverage of the results with respect to the query. At
spatial level, this happens when the results include resources that only slightly inter-
sect with the query bounding box. At thematic level, it happens when the result
includes resources about a small subset of the query themes. This is a problem
because resources that only slightly fulfill the query may be considered very relevant
results. As an example of the spatial under-coverage, Figure 2(A1) shows the bounding
box of a query focused on the ‘Castilla la Mancha’region in Spain (continuous line) and
a resource focused on ‘Valencia’region that only slightly intersects with the query
6J. LACASTA ET AL.
(discontinuous line). Figure 2(A2) shows a thematic under-coverage example. It con-
tains a query about many subjects (‘roads’,‘rivers’and others) and a result (R) detailing
only ‘roads’. In both cases, the amount of information provided with respect to the
requested query is small. Thus, although they fulfill the query, they have little
relevance.
The second issue is the over-coverage of the results with respect to the query. At spatial
level, it happens when the results include resources that cover an area much bigger than the
requested one. At thematic level, it occurs when a result contains information about many
more themes than the requested ones. Over-coverage is a problem in the sense that the
amount of irrelevant information in the results makes difficult to identify the desired
content. Any other result more adjusted to the query is probably a better option for the
user. The spatial over-coverage example is shown in Figure 2(B1). It depicts the same
‘Castilla la Mancha’query, but in this case, the result covers the desired region and the
rest of the Iberian Peninsula. Figure 2(B2) shows a thematic over-coverage example with a
query about ‘roads’and a result (R) containing information about ‘roads’,‘rivers’and many
other themes. In both cases, the result contains not only relevant information but also a
disproportionate amount of irrelevant data.
The last issue happens when there are many partial results to a query. All of them are
partially relevant, but none of them completely fulfill the query specification. Existent
systems sort these results according to a spatial/theme similarity criterion (usually some
spatial/theme overlap variant). However, when results are presented in this way, it is
difficult to distinguish which areas/themes of the query are described by each resource
and how they complement each other. An example of the spatial aspect of the partial
coverage is shown in Figure 2(C1). It shows the same ‘Castilla la Mancha’query with
multiple partial relevant results. Figure 2(C2) continues with the previous ‘roads’and
‘rivers’query but providing as answer a resource about ‘roads’and a different one about
Figure 2. Spatial and thematic issues in the IR system of geospatial data catalogs. (a) under
coverage, (b) over coverage, (c) resource composition.
INTERNATIONAL JOURNAL OF GEOGRAPHICAL INFORMATION SCIENCE 7
‘rivers’to show the thematic partial coverage. In both cases, none of the results are a
complete answer to the query.
These spatial and thematic issues can happen in any combination in a ‘concept at
location’style query, especially when resources only cover part of the spatial area and
part of the requested themes. In this case, as previously indicated, the results generated
by the reviewed systems are not able to completely fulfill the query restrictions. Next
section proposes an IR system able to deal with these issues by aggregating the
metadata records in the result list into collections of compatible records that, as a set,
are a better answer to the query. The construction of these aggregations helps to solve
the composition issue and mitigates under and over-coverage problems.
4. Generating thematically and spatially aggregated results
In a classic IR system, when performing the intersection between the metadata of a
resource and the query to determine if it is relevant, only a subset of the themes may
intersect, and only a part of the query area may be covered. This means that each
retrieved resource only provides a partial result. However, combining it with others, a
more complete result could be obtained. For example, in a query about ‘highways’in
‘Spain,’we can find a resource about highways in the south of Spain. This result is
incomplete but if combined with other one covering the north, we can compose a good
result. The same happens with respect to the themes. For example, in a query about
‘highways’and ‘motorways’in Spain, a resource about Spanish motorways may be the
perfect complement to another one depicting the Spanish highways.
Figure 3 shows the main steps of the IR process created to generate aggregations of
metadata records as results of ‘concept at location’queries. The query analysis step is a
simple decomposition process where the query is processed to separate spatial (bound-
ing box) and thematic (keywords) requirements. The next step is to obtain the metadata
records describing resources that are partially relevant to the query. This is done using a
spatial and an inverted textual index. Only those that intersect spatially with the query
area and contain at least one of the query keywords are returned. Then, the obtained
metadata records are sorted according to their relevance degree. Finally, the IR process
aggregates the records in suitable groups. This section focuses on describing how the
ranking and aggregation process is performed.
Figure 3. Process for aggregation of query results.
8J. LACASTA ET AL.
Equation (1) shows the similarity formula used for result ranking. It represents the
similarity with a value between 0 and 1, 0 being ‘irrelevant’and 1 ‘perfect match’. In the
formula, we use the following symbols: dH GQ;GR
ðÞrepresents the Hausdorffdistance
between the query geometry (GQ) and a metadata record geometry (GR); Max DH is the
biggest Hausdorffdistance of all the partially relevant resources with respect to the
query; size TQ\TR
indicates the number of themes in common between the metadata
record (TR) and the query (TQ); and size TQ
ðÞindicates the number of themes in the query.
SimilarityðQ;RÞ¼Max DH dH GQ;GR
ðÞ
Max DH size TQ\TR
ðÞ
size TQ
ðÞ (1)
The Hausdorffdistance is the greatest of all the distances between any point in a
geometry and the closest point in another geometry. Since the Hausdorffdistance
between geometry A and B may be different from the Hausdorffdistance between
geometry B and A, the maximum is used. The Hausdorffdistance of overlapping
geometries of similar size is smaller than the equivalent one between overlapping
geometries with very different dimensions. Therefore, it is very appropriate for ordering
resources of different administrative levels (e.g. region vs. country size). Additionally, the
Hausdorffdistance can be used with complex geometries. Thus, replacing the metadata
bounding boxes with approximate geometries would directly increase the quality of
results without having to modify the IR system. This is also valid for resources with
multiple disjoint geometries (e.g. Iberian peninsula and Canary islands) that can be
represented as a single multi-polygon geometry for distance measure purpose. A
problem of the Hausdorffdistance is that it can give small distances to nonoverlapping
resources (if they are similar in size and they are spatially close). However, this issue is
not relevant for our system because the ranking is only applied on resources that
overlap.
Once the ranking of the results has been performed, the last step aggregates the
results that are spatially and thematically compatible. We consider that two metadata
records (and therefore the resources they describe) are spatially compatible if the
combined area for all their themes is significantly closer to the query area than each
record individually. Regarding the thematic compatibility, we only consider compatible
those that are thematically disjoint (they do not have any common query themes), and
those that share one query theme and also half of the rest of the keywords. This avoids
aggregating records of resources that describe a theme in very different ways. For
example, if we are making a query about ‘river basins’, we may find a resource that
focuses on the ‘water flow’and another one describing the ‘geology’of the basin. In this
case, they are too different to be in the same aggregation. When the records do not
share a query theme, they can always be aggregated because they are fulfilling disjoint
restrictions in the query.
Algorithm 1 describes the method used to aggregate the list of ranked results
obtained with the previous steps. For each metadata record obtained as result, the
process searches other results that complement it. The aggregation process is per-
formed only if there is a relevant part of the query area that is not covered in all the
themes. The coverageFactor indicates the percentage of area in the query (summing up
all themes) that can be left uncovered. The smaller it is, the more complete the results
INTERNATIONAL JOURNAL OF GEOGRAPHICAL INFORMATION SCIENCE 9
are. However, it is better to not completely cover the query bounding box with results to
deal with imprecisions in the definition of the query or the resources. For example, it is
counterproductive to complement a resource with a 99% of query coverage just to
cover a small gap in a border. The value selected in the experiment section is a
compromise between a complete coverage of the query and the management of
deficiencies in the query formulation and the resources.
Algorithm 1 Spatio-thematic aggregation of results.
function AGGREGATIONSTEP(results;query)
aggregationList ;
for result 2results do
aggregation result
reducedQuery query aggregation
resultExtended true
while areaðreducedQueryÞ>coverageFactor areaðqueryÞ&resultExtended do
resultExtended false
possibleAggregated getBestResultðresults;reducedQuery;aggregationÞ
if possibleAggreatedÞ;then
reducedQuery reducedQuery possibleAggregated
aggregation aggregation [possibleAggregated
resultExtended true
end if
end while
aggregationList aggregationList [aggregation
end for
return duplicateRemovalðaggregationListÞ
end function
The search of results that complement a given one is done by the function
getBestResult depicted in Algorithm 2. This process is repeated until no more suitable
resources for the aggregation are found. The identification of suitable complementary
results is done by removing those that are spatially and thematically incompatible with
the current aggregation (thematicFilter and spatialFilter functions). Then, the rest are
sorted according to the dimension of the uncovered part of the query, and the closest
one is selected.
Algorithm 2 Function to obtain a new element for an aggregation.
function GETBESTRESULT(results;reducedQuery;aggregation)
filteredResults spatialFilterðresults;reducedQuery;infoFactorÞ;
filteredResults thematicFilterðfilteredResults;reducedQuery;aggregationÞ;
sortedResult rankResultsðfilteredResults;reducedQueryÞ
if sortedResult Þ;then
10 J. LACASTA ET AL.
return sortedResult½0
else
return ;
end if
function SPATIALFILTER(results;reducedQuery;infoFactor)
filteredResult ;;
for result 2results do
if areaðintersectionðresult;reducedQueryÞÞ >areaðreducedQueryÞinfoFactor then
filteredResult filteredResult [result;
end if
end for
return filteredResult
end function
function THEMATICFILTER(results;reducedQuery;aggregation)
filteredResult ;;
for result 2results do
themesInQuery ¼themesðresultÞ\themesðreducedQueryÞ;
themesInAggr ¼themesInQuery \themesðaggregationÞ;
commonThemes ¼themesðresultÞ\themesðaggregationÞ
if result‚aggregation ^ððthemesInQueryÞ;^themesInAggr ¼¼ ;Þ_
ðthemesInAggrÞ;^sizeðcommonThemesÞsizeðthemesðaggregationÞÞ2Þthen
filteredResult filteredResult [result;
end if
end for
return filteredResult
end function
In the process to identify the best possible record to add to an aggregation depicted
in Algorithm 2, the spatial and thematic filters avoid adding resources that only improve
the results in a negligible amount and the creation of thematically heterogeneous
aggregations. The spatial filter removes the metadata records of resources that do not
cover a relevant part of the query area uncovered by the aggregation. This behavior is
adjusted by the infoFactor parameter that represents the amount of new information a
resource has to provide to be included in the aggregation. A low value generates more
complete aggregations, but some of their components may provide very little new
information. A high value creates aggregations with more relevant elements, but it
may leave important parts of the query uncovered. The thematic filter removes those
results already in the aggregation and those that share a theme with the query and the
aggregation but do not have in common at least half of the keywords. This is done
because results with fewer keywords in common are likely to be too different between
them to be integrated, even if they share a queried theme. After applying both filters,
the remaining results are ranked according to the similarity formula shown in Equation
(2), and the most similar one is selected as a new element in the aggregation.
INTERNATIONAL JOURNAL OF GEOGRAPHICAL INFORMATION SCIENCE 11
Similarity Uq;RðÞ¼
PT2Uq Max DHTdH GTUq ;GTR
=Max DHT
Size TUq
(2)
Equation (2) is a generalization of Equation (1). Since we try to find the metadata record of
the resource that is the most similar to the area of the themes that is not covered by the
current members of an aggregation, the geometry of each theme in the query is different.
For example, in a query about ‘highways’and ‘motorways’in Spain, we may have con-
structed an aggregation with a resource with the highways in the south of Spain, and
another one covering the motorways in the east. In this context, the extension that is
needed to cover with additional resources is different for the theme ‘motorways’and the
theme ‘highways’. In the equation, we calculate the similarity of a metadata record (R) that
is candidate for the aggregation with respect to the area of the query themes not covered
by the aggregation (Uq). It is calculated as the sum of the spatial similarity of each theme of
the query with some spatial extension uncovered with respect to the metadata record
extension for these themes, divided by the number of query themes that have a spatial
part uncovered (Size TUq
). The spatial similarity for each theme is obtained in a way
analogous to Equation (1). In Equation (2), the following symbols are used: MaxDHT
represents the maximum Hausdorffdistance between the theme extension of all the
candidates and the uncovered extension of the query for this theme, and dH GTUq;GTR
is the Hausdorffdistance of the theme of the metadata record that is being analyzed (GTR)
with respect to the uncovered part of the query for the theme (GTUq ).
This process may generate redundant aggregations with the same elements in different
order (e.g. it can aggregate the 1st result with the 10th one and then aggregate the 10th
result with the 1st one) and aggregations that are a superset of another one (in this case,
some elements in the superset are not relevant). The last step removes these redundancies.
5. Experiments
This section compares the performance of the proposed IR process (Aggregated IR
System) with a basic IR system (Basic IR System) similar to the ones used in the geospatial
data catalogs described in Section 3.1.
Our Aggregated IR System has been tuned to try to create aggregations with at least a 90%
of query coverage (coverageFactor ¼0:1) and to add elements to the aggregation even if
they only provide a small improvement of the result (infoFactor ¼0:1). The Basic IR System
used for comparison applies a similarity measure that behaves similarly to the geospatial data
catalogs described in Section 2 (see Equation (3)). This measure performs a combination
between the spatial intersection of the query and the metadata record of the resource, and
the theme intersection. The spatial similarity is obtained as the area of the intersection
between the query (AQ)andtherecord(AR), divided by the maximum area of intersection
between all the resources and the query. For the thematic similarity, we have directly used the
Jaccard coefficient, which is calculated as the number of themes in common between the
query (TQ)andtherecord(TR) divided by the total number of themes. Finally, this similarity
values are weighted with αand βfactors to be able to adjust the weight of the spatial aspect
(α) of the query with respect to the thematic ones (β). To avoid giving any advantage to our
proposal, the experiments with the Basic IR System have been performed multiple times with
12 J. LACASTA ET AL.
different αand βvalues, and the best obtained results have been the ones used in the
comparison.
SimilarityðQ;RÞ¼α\AQ;AR
ðÞ
maxAreaIntersection
þβJaccard TQ;TR
ðÞðÞ(3)
5.1. Evaluation methodology
For this experiment, we have used the metadata records provided through the
Geoportal of the Spanish National Spatial Data Infrastructure
7
(IDEE) in 2015. This
collection contains 97,867 records describing geographical resources created by differ-
ent Spanish governmental institutions. This includes themes so different such as topol-
ogy, environment, mineral resources, industry and infrastructures, among other themes.
The performance of an IR system to solve the issues described in Section 3 cannot be
simply described in terms of classical precision/recall measures. These measures are
often based on a binary classification results as relevant and nonrelevant as a whole
(Baeza-Yates and Ribeiro-Neto 2011). In our system, only the metadata records of
resources that contain part of the selected area and some of the query themes are
returned. Therefore, all the results contain at least a bit of relevant information. The
problem here is to measure the degree of relevance for result ordering.
The proposed system is focused on improving the results of ‘concept at location’
queries. The objective of this type of queries is to return first the results that have an
exact match with the query restriction, second those that cover the selected area but
include much additional information (over-coverage), third those that have only a partial
coverage, and finally results that only slightly fulfill the query restrictions (under-
coverage).
To evaluate the ranking of the two systems, we have used the discounted cumulative
gain (DCG) measure shown in Equation (4) (Baeza-Yates and Ribeiro-Neto 2011). This
measure calculates the gain of adding each document to the result set based on its
position in the result list. To obtain this measure, it is needed to describe the gain that
each result adds to the result list (Gi). For this task, we have used the gain criteria
described in Table 2. In these criteria, higher values indicate that the result is more
adjusted to the spatial or thematic restrictions in the query. The lower ones indicate that
there is less similarity. The spatial and thematic content of each result of the analyzed
queries has been classified according to these criteria. The final gain of each result is
calculated as the mean of the spatial and the thematic gains.
Table 2. Criteria values used to determine the quality of a result.
Gain value Meaning Description
3 Exact match The spatial or thematic features of the result are approximately
equal to the query
Over-coverage The spatial or thematic features of the result approximately cover
the query but they are much more extensive
Partial coverage The spatial or thematic features of the result just cover a part of the
query
Under-coverage The spatial or thematic features of the result just slightly cover a
part of the query
INTERNATIONAL JOURNAL OF GEOGRAPHICAL INFORMATION SCIENCE 13
DCG½i¼ i¼1;G1
iÞ1;Gi=log2ðiÞDCGi1
(4)
5.2. Description of the experiments and results obtained
The main advantage of our system with respect to a basic one is that it is able to identify
subsets of low gain results and transform them into higher gain aggregations. For
example, it can combine several results with partial coverage of the query to obtain
an exact match. To evaluate how the system performs, we have selected four themes
commonly used in fields such as hydrology, ecology, infrastructure planning, industry or
agriculture [some recent examples on the interest in these areas can be found in Graser
et al.(2015) and Pereira et al.(2015)] that have a high presence in the collection and four
spatial areas that contain information about these themes. The themes are ‘elevation’
(model), ‘road network’,‘soil use’and ‘hydrography’. Using these themes and areas, we
have generated all the possible queries that include one or two of the themes and one
spatial area. This makes a total of 40 different queries (combinations without repetitions
of 2 themes selected from the 4 original ones plus the ‘empty’theme, and the 4 different
areas, i.e. 5
2
4). For these queries, we have obtained the DCG for the 10 first
positions of their result lists, and we have calculated the mean DCG at each position.
This mean takes into account that some queries return less than 10 results. Figure 4
compares the mean DGC of the two systems at each position of their result list (number
of query result). It can be observed how the aggregated system has always a higher
mean gain. This means that the obtained resources and the way they are positioned in
the result list are better in the aggregated system than in the basic one.
To explicitly show how the system behaves with respect to the undesired effects
described in Section 3, we analyze in detail the results of a small set of the selected
queries. These queries show how our system behaves in two main scenarios: when there
are results that perfectly match the spatial and/or thematic query restriction, and when
there are not close matches. Table 3 summarizes the selected queries (Q1–4). The table
shows the query bounding box (min and max longitude, latitude), a toponymical
reference (Location) of the Spanish region containing the bounding box (for illustrative
Figure 4. Mean DGC comparison.
14 J. LACASTA ET AL.
purposes) and the themes in the query (Themes). In the case of the first query (Q1), there
are results that perfectly match the query restrictions. The detailed analysis of the result
ordering is illustrative of the difference in behavior between our system and the basic
solution. The rest of the queries display result sets that do not contain any result that
perfectly match the query. Specifically, Q2 focuses on the spatial features. It requests
information about a theme (‘elevation’) in a region (a part of ‘Andalucía’) where there are
not resources that fit well with the selected area for the selected theme. However, there
are resources that have over-coverage and others with partial coverage. Q3 focuses on
the thematic features. It includes multiple themes (‘road networks’and ‘soil use’) and it
selects a region (Galicia) that contains resources that match the spatial aspects of the
query but only partially match the thematic aspects. Q4 describes the more general case
where none of the resources match well the query area (a part of Galicia and Castilla y
León) and the selected themes (‘road networks’and ‘soil use’again).
Table 4 shows a summary of the performance of each system, measured as the spatial
and thematic similarity of the results with respect to the query specification. It includes
the number of results obtained from each query (Num Res), the mean spatial coverage of
the results with respect to the spatial restriction in the query (Mean SpCov), the mean
thematic overlap between the results and the thematic restrictions in the query (Mean
ThCov) and the size of the biggest aggregation obtained (Max AggSize). The spatial and
thematic mean coverage visualize the degree of fulfillment of the user needs, while the
size of the biggest aggregation indicates the number of individual results that are
needed to fulfill the query in the worst case. Table 5 details the first three results of
each query. It includes the title of the result in each position (result order), the percen-
tage of spatial (SpCov) and thematic (ThCov) coverage of each result with respect to the
query and the gain value (Gain) obtained according to the criteria indicated in Table 2.
The results that aggregate several resources to compose a better result are marked with
ðAÞ. In the figure, it can be observed the difference between the results of the basic
system, where most of them have over-coverage and partial coverage, and the ones
obtained in the aggregated system, which generates aggregations closer to the query
constraints.
Table 4. Comparison of system results.
Basic IR Aggregated IR
Num Res
Mean
SpCov (%)
Mean
ThCov (%) Num Res
Mean
SpCov (%)
Mean
ThCov (%)
Max
AggSize
Q1 12 51.37 100 6 99.98 100 1
Q2 9 55.55 100 6 100 100 4
Q3 25 100 50 24 100 100 2
Q4 31 81.15 50 29 97.5 100 4
Table 3. Queries selected for the evaluation.
Number Bounding box (min; max) Location Themes
Q1 −6.02, 37.37; −5.93, 37.41 Andalucía Elevation
Q2 −6, 37.35; −5.9, 37.41 Andalucía Elevation
Q3 −8.38, 42.25, −7.50, 43.08 Galicia Road network, soil use
Q4 −7.8, 42.21; −5.9, 43.29 Galicia, Castilla y León Road network, soil use
INTERNATIONAL JOURNAL OF GEOGRAPHICAL INFORMATION SCIENCE 15
Table 5. Detailed comparison of system results.
Basic IR Aggregated IR
Order Result title
SpCov
(%)
ThCov
(%) Gain Result title
SpCov
(%)
ThCov
(%) Gain
Q1 1 Orography of Andalucía 100100 2.5 Andalucía EDM 98433 99 100 3.0
2 Contour lines 100100 2.5 Orography of Andalucía 100100 2.5
3 Digital Terrain Model 100100 2.5 Contour lines 100100 2.5
Q2 1 Orography of Andalucía 100100 2.5 (A) Andalucía EDM 98433/34/43/44 100 100 3.0
2 Contour lines 100100 2.5 Orography of Andalucía 100100 2.5
3 Digital Terrain Model 100100 2.5 Contour lines 100100 2.5
Q3 1 Topographic Base of Galicia 100 50 2.0 (A) Topographic Base of Galicia/Map of Coverages and
Soil Uses
100 100 3.0
2 Map of Coverages and Soil Uses 100 50 2.0 (A) Map of Coverages and Soil Uses/Basic Cartography of
Galicia
100 100 3.0
3 Basic Cartography of Galicia 100 50 2.0 (A) Topographic Base of Galicia/Soil Uses, Polygons 100 100 3.0
Q4 1 CORINE Land Cover 1990 10050 1.5 (A) Topographic Base of Galicia/Transport Network CyL/
Map of Coverages and Soil Uses/Land Cover CyL
97 100 3.0
2 CORINE Land Cover 2000 10050 1.5 (A) Basic Cartography of Galicia/Transport Network CyL/
Map of Coverages and Soil Uses/Land Cover CyL
97 100 3.0
3 CORINE Land Cover changes 1990–2000 10050 1.5 (A) Soil Uses. Polygons/Topographic Base of Galicia/
Transport Network CyL/Land Cover CyL
96 100 3.0
Spatial coverage percentages marked with a star (*) indicate a big spatial over-coverage (they are resources at country level for queries about a small region).
16 J. LACASTA ET AL.
Q1 is representative of the situation when a query has a perfect match with the
collection. It has been selected because there is a resource in the collection that matches
at 99% the query bounding box (tile 98433 of the ‘Andalucía Elevation Digital Model’).
Additionally, there are five resources relevant for the query theme but that cover all
Andalucía/Spain (they have spatial over-coverage). Finally, there are other six themati-
cally relevant resources with spatial under-coverage. In the basic system, the most
relevant resource is provided as the sixth result and the previous places are occupied
by the resources with spatial over-coverage that completely cover the query area. The
results with under-coverage are placed last. In the proposed system, no aggregation is
generated for this query, but the ordering is improved since the best result is placed first
and those with over-coverage are sorted according to the spatial similarity with the
query. Additionally, the resources with spatial under-coverage are not returned because
the aggregation process identifies that they are only reliable if complemented with
another one that is reliable by itself.
Q2 analyzes the behavior of the systems when there are resources completely cover-
ing the thematic restrictions but not the spatial ones. For Q2, the collection contains
nine relevant resources, five with spatial over-coverage (the same in Q1) and four with
partial spatial coverage. They are four tiles of the ‘Andalucía Elevation Digital Model’
(‘98433’,‘98434’,‘98443’and ‘98444’). In the basic system, the five first results are those
with spatial over-coverage, and the last four ones are those that have partial coverage. In
the proposed system, the four resources with partial spatial coverage are aggregated
into a single result that perfectly matches the query. This aggregation is provided first in
the result list. The rest of the results are sorted according to their spatial distance with
respect to the query.
Q3 analyzes a scenario with resources that cover the spatial aspects of the query but
only partially the thematic ones. For this query, there are 25 resources focused on Galicia
and Spain about ‘road networks’and ‘soil use’, but none containing both. In the basic IR
system, the obtained results are not distinguishable since they all completely cover the
query area and contain one query theme. In the proposed system, 24 compatible
aggregations that fulfill the user needs are found. These aggregations add to each result
(i.e. focused on one theme) the spatially closest result of the other theme. For example,
the first result is the aggregation consisting of the two first results of the basic system,
i.e. ‘Topographic Base of Galicia’(providing road networks) and ‘Map of Coverages and
Soil Uses’(providing soil use).
Finally, Q4 focuses on the most general case: a query that has no clear candidate
for the thematic and spatial query restrictions. It uses the same themes as the third
query, but it covers an area that includes part of Galicia and their neighbor region of
Castilla y León. In the collection, there are 31 relevant resources, and all of them have
partial coverage of the spatial or the thematic query restrictions. In the basic system,
as in the third query, the results only cover a single theme and they are sorted
according to the degree of intersection with the query bounding box. This places the
results with spatial over-coverage upper in the result list. For example, the first five
results are different versions of the CORINE land cover project about ‘Soil Uses’in
Spain. The result 17 is the first one about ‘Soil Uses’focused on a region close to the
query bounding box (‘Land cover of Castilla y León’), and the result 19 is the first
about ‘road networks’(‘Topographic Base of Galicia’). The proposed system
INTERNATIONAL JOURNAL OF GEOGRAPHICAL INFORMATION SCIENCE 17
aggregates resources focused on Galicia and Castilla y León to form results that
almost perfectly fit the user needs. Figure 5 compares graphically the first result
obtained in both systems (queries are showningray,resultsinblack).Thebasic
system provides an unfocused result about a single query theme (Soil Use) and the
query area is only a small fragment of the area covered by the resource. Regarding
the proposed system, it returns an aggregation containing four results that provide
an almost complete answer to the user query.
As a final comparison between the two systems, Figure 6 shows the DCG of the four
selected queries (discounted cumulative gain) at each position of their result lists
(number of query results) for the basic and the aggregated IR systems. It can be
observed how the proposed system behaves better for all the query types, being
especially advantageous in the most general case (Q4). In this case, none of the
collection resources perfectly match the query but there are several partial matches.
Therefore, the aggregation process can show its maximum potential.
Figure 5. Graphical comparison of the first result of Q4 in both systems.
18 J. LACASTA ET AL.
6. Discussion
The management of the spatial information as a continuous set is needed for many
tasks such as analyzing the morphology of a river or identifying routes in a road
network. However, when dealing with geospatial data catalogs, usually continuous
information can be found divided into individual resources that do not provide the
implicit spatial/thematic connection between them. This problem is not architectural, it
is related to how data producers manage information. The technologies used for IR are,
in most cases, general purpose solutions not adapted to the nature of spatial data.
The IR system proposed in this paper identifies the spatial and thematic relations in a
collection of metadata records of geospatial resources to produce results closer to the
query restrictions. To identify the spatial closeness of the records with respect to a query,
it uses criteria similar to the one indicated in Lanfear (2006), but using the Hausdorff
distance as the spatial similarity measure. The thematic similarity is the ratio between
the common themes in the record and the query. Finally, our system integrates the
spatial and thematic similarity with a ranking formula like the one detailed in Martins
et al.(2005). What makes our system very different from them is the addition of a
processing layer that identifies the spatial and thematic relations between the results to
generate collections of metadata records as query results. The system uses these rela-
tions to combine the results into coherent aggregations that are closer to the user query
constraints than the individual resources.
In this aspect, the paper has similarities with Lieberman (2006) or Lutz et al.(2009)in
the sense that we use a layer on top of a basic catalog to provide improved results. The
difference is that they use ontologies to generate individual results and we focus on the
use of raw metadata to produce aggregations of results. They deal with resources as
individuals while our system goes a step beyond that by considering that a result can be
a composition of several metadata records.
Figure 6. Detailed DCG comparison of the analyzed queries.
INTERNATIONAL JOURNAL OF GEOGRAPHICAL INFORMATION SCIENCE 19
The aggregation of the catalog metadata records can be seen as a data integration
task that helps to improve the quality of a catalog IR process. From this perspective, our
work is related to proposals such as Hübner et al.(2004) and Lutz and Klien (2006) but
working at metadata level instead of at data level.
Our aggregation proposal shows that taking into account the collection context
in geospatial data catalogs is a way to provide more complete results. However, we
have observed a limitation caused by the difficulty of generating consistent aggre-
gations. In general, the themes related to geographical information are a quite
homogeneous terminology set where the aggregations are really meaningful (land-
forms, infrastructures, cadaster). However, even if two resources share the same
thematic, they may not be completely compatible if they provide too different
information. For example, a resource containing the geometry of parcels and their
type of crop is not very compatible with an aerial thermal image used for crop
analysis. The problem has been mitigated using all the keywords of the resources
as integration context. However, its effectiveness depends on the content of a
single metadata field. A more sophisticated solution would require taking into
account additional metadata elements to avoid noise in the aggregations.
Additional elements to take into account as factors for data integration would be
the data information models, formats, scales, or resolutions.
With respect to the processing time, the step for computing the aggregations does
not significantly delay the search process because the spatial operations required to
perform the aggregation are restricted to the resources in the result list. Including the
time to access the spatial and textual indices, all the queries have returned their results
in less than 1 s.
7. Conclusions
This paper has identified and analyzed three issues (under coverage, over coverage,
partial coverage) from ‘concept at location’style queries in geospatial data catalogs.
These issues are caused by the lack of adaptation of prevalent IR engines used in these
geospatial data catalogs to the specific nature of geoinformation. As a solution, we have
proposed an IR method that yields aggregations of search results that match better
‘concept at location’query restrictions.
The proposed IR method takes all the metadata records of resources that partially
fulfill a query (intersect the bounding box or the themes) and finds the spatial and
thematic relations between them. Next, it uses these relations to generate sets of
metadata records that are a better answer to the query than each one individually.
To evaluate its performance in archetypical ‘concept at location’queries, we have
compared the performance of our proposalwithrespecttoanIRsystemsimilarto
those used in prevalent geospatial data catalogs. The results have shown that this
approach may complement the traditional plain list of results of geospatial data
catalogs.
Additionally, our proposed aggregation-based functionality could be easily offered in
any geospatial data catalog by extending the catalog service for the web (CSW) interface
provided by OGC consortium (OGC 2007a). In the CSW, the GetRecords operation is
responsible for locating resources according to the user query specified restrictions. CSW
20 J. LACASTA ET AL.
standard establishes three levels of detail in query results (Brief, Summary and Full). To
provide interoperability between systems, Brief and Summary results structure is
restricted to the Dublin Core (DCMI 2007) based schema defined by OGC. In the case
of Full results, the standard allows the definition of profiles that extend the service
functionality. Through these profiles, the result structure could be redefined to provide
aggregations as a new type of resource that can be returned by the CSW.
Since the structure of the Brief and Summary levels of detail is restricted, it is not
possible to completely describe the aggregations: just the type of returned resource
would indicate that is a collection (e.g. using dct:Collection as dc:type value), and the
identifiers of the elements conforming it would be referenced (using dc:relation field).
However, the Full basic description could be extended as needed to indicate the themes
and area of the query added by each resource in the aggregation. This approach would
make the aggregation-based system compatible with any existent CSW client. In all the
three views, a client could obtain the description of resources with standard metadata
fields, and only in the Full view, it would need to be specifically adapted to process the
details of the aggregation composition.
Future work will explore the use of other metadata elements to solve problems
related to scale and information content. Clustering the resources that only differ in
representation fields, such as the scale, before the indexing process would reduce the
heterogeneity of the results, showing the alternative scales and the type of content
available in each cluster. Including temporal information can be also used to extend the
proposed method for dealing with ‘concept at location’at time queries. Finally, once we
have aggregated metadata results, it could be possible to produce virtual resources that
give access to the associated resources in an integrated way, even if this information
was originally scattered across multiple resources.
Notes
1. http://inspire-geoportal.ec.europa.eu/
2. https://www.geoplatform.gov/
3. http://idee.es/
4. https://data.gov.uk
5. https://geodiscover.alberta.ca/geoportal/
6. http://www.europeana.eu/portal/
7. http://www.idee.es/csw-inspire-idee/srv/spa/catalog.search#/home
Acknowledgments
We are grateful to the collaboration of the Spanish National Geographic Institute to provide us
with data for the experiments.
Disclosure statement
No potential conflict of interest was reported by the authors.
INTERNATIONAL JOURNAL OF GEOGRAPHICAL INFORMATION SCIENCE 21
Funding
The work of Borja Espejo-Garcia has been partially supported by Aragon Government through the
grant number [C38/2015].
ORCID
Javier Lacasta http://orcid.org/0000-0003-3071-5819
F. Javier Lopez-Pellicer http://orcid.org/0000-0001-6491-7430
Javier Nogueras-Iso http://orcid.org/0000-0002-1279-0367
F. Javier Zarazaga-Soria http://orcid.org/0000-0002-6557-2494
References
Anderson, K. and Gaston, K.J., 2013. Lightweight unmanned aerial vehicles will revolutionize
spatial ecology. Frontiers in Ecology and the Environment, 11, 138–146. doi:10.1890/120150
Asadi, S., et al., 2005. Searching the World Wide Web for local services and facilities: a review on
the patterns of location-based queries. In: W. Fan, Z. Wu and J. Yang, eds. Advances in web-age
information management, lecture notes in computer science. Cham, Switzerland: Springer, Vol.
3739, 91–101.
Baeza-Yates, R. and Ribeiro-Neto, B., 2011.Modern information retrieval. The concepts and technol-
ogies behind search. Reading, MA: Addison-Wesley.
DCMI, 2007.DCMI abstract model. Dublin Core Metadata Initiative, Technical report. Seoul, Korea:
National Library of Korea.
Ferrés, D. and Rodríguez, H., 2015, Evaluating geographical knowledge re-ranking, Linguistic
processing and query expansion techniques for geographical information retrieval. In:C.
Iliopoulos, S. Puglisi and E. Yilmaz, eds. Proceedings of the 22nd International Symposium, SPIRE
2015, 9309 of Lecture Notes in Computer Science, September, London, UK. New York, USA:
Springer-Verlag, 311–323.
Florczyk, A.J., et al., 2010. Applying semantic linkage in the geospatial web. Lecture Notes in
Geoinformation and Cartography (LNG&C). Geospatial Thinking, 201–220.
Göbel, S. and Klein, P., 2002. Ranking mechanisms in metadata information systems for geospatial
data. In:Proceedings of the Earth Observation & Geo-Spatial Web and Internet Workshop, Ispra,
Itally. Munich, Germany: Fraunhofer, 13–13.
Graser, A., Asamer, J., and Ponweiser, W., 2015. The elevation factor: Digital elevation model quality
and sampling impacts on electric vehicle energy estimation errors. In:Proceedings of the
International Conference on Models and Technologies for Intelligent Transportation Systems (MT-
ITS) [online], 3–5 June 2015. Budapest, Hungary: Domokos Esztergár-Kiss, 81–86.
Hübner, S., et al., 2004. Ontology-based search for interactive digital maps. IEEE Intelligent Systems,
19, 80–86. doi:10.1109/MIS.2004.15
ISO/TC 211, 2014.ISO 19115-1:2014. Geographic information –Metadata –Part 1: fundamentals.
Geneva, Switzerland: International Organization for Standardization.
ISO/TC 211, 2016.ISO 19135-3:2016. Metadata –Part 3: XML implementation of fundamentals.
Geneva, Switzerland: International Organization for Standardization.
Janowicz, K., et al., 2010. Semantic enablement for spatial data infrastructures. Transactions in GIS,
14, 111–129. doi:10.1111/j.1467-9671.2010.01186.x
Kim, J., Vasardani, M., and Winter, S., 2017. Similarity matching for integrating spatial information
extracted from place descriptions. International Journal of Geographical Information Science, 31,
56–80. doi:10.1080/13658816.2016.1188930
Lanfear, K.J., 2006.A spatial overlay ranking method for a geospatial search of text objects. Reston,
VA: USGS, Technical report.
22 J. LACASTA ET AL.
Larson, R. and Frontiera, P., 2004. Ranking and representation for geographic information retrieval.
In: R. Purves and C. Jones, eds. Proceedings of the SIGIR 2004 Workshop on Geographic
Information Retrieval [online], 25–29 July. Sheffield. Available from: http://www.geo.uzh.ch/~
rsp/gir/abstracts/
Latre, M., et al., 2009. An approach to facilitate the integration of hydrological data by means of
ontologies and multilingual thesauri. In: M. Sester, L. Bernard and V. Paelke, eds. Advances in
GIScience. Lecture notes in geoinformation and cartography (LNG&C),02–05 June, Hannover,
Germany. Berlin, Germany: Springer-Verlag, 155–171.
Lieberman, J., 2006.Geospatial semantic web interoperability experiment report. Open Geospatial
Consortium, Technical report. Abingdon, UK: Taylor & Francis.
Lutz, M., et al., 2009. Overcoming semantic heterogeneity in spatial data infrastructures. Computers
& Geosciences, 35, 739–752. doi:10.1016/j.cageo.2007.09.017
Lutz, M. and Klien, E., 2006. Ontology-based retrieval of geographic information. International
Journal of Geographical Information Science, 20, 233–260. doi:10.1080/13658810500287107
Martins, B., Silva, M.J., and Andrade, L., 2005. Indexing and ranking in Geo-IR systems. In: C. Jones
and R. Purve, eds. Proceedings of the Workshop on Geographic information retrieval,04
November, Bremen, Germany. New York, USA: ACM, 31–34.
Megler, V.M. and Maier, D., 2011. Finding haystacks with needles: ranked search for data using
geospatial and temporal characteristics. In: J.B. Cushing, J. French and S. Bowers, eds. Scientific
and statistical database management, no. 6809 of lecture notes in computer science. Berlin,
Germany: Springer Verlag, 55–72.
Nogueras-Iso, J., et al., 2009. Development and deployment of a services catalog in compliance
with the INSPIRE metadata implementing rules. In: B. Van Loenen, J.W.J. Besemer and J.A.
Zevenbergen, eds. SDI convergence: research, emerging trends, and critical assessment.
Amersfoort, Netherlands: The Netherlands Geodetic Commission (NGC), 21–34.
OGC, 2007a.OpenGIS catalogue service implementation specification.Wayland, MA: Open
Geospatial Consoritum, Technical report Version 2.02.
OGC, 2007b.OpenGIS catalogue services specification 2.0.2 - ISO metadata application profile.
Wayland, MA: Open Geospatial Consoritum, Technical report Version 2.02.
Paneque-Gálvez, J., et al., 2014. Small drones for community-based forest monitoring: An assess-
ment of their feasibility and potential in tropical areas. Forests, 5, 1481–1507. doi:10.3390/
f5061481
Pereira, P., et al., 2015. The impact of road and railway embankments on runoffand soil erosion in
eastern Spain. Hydrology and Earth System Sciences Discussions, 12, 12947–12985. doi:10.5194/
hessd-12-12947-2015
Renteria-Agualimpia, W., et al., 2016. Improving the geospatial consistency of digital libraries
metadata. Journal of Information Science, 42, 507–523. doi:10.1177/0165551515597364
Sallaberry, C., 2013.Geographical information retrieval in textual corpora. Hoboken, NJ: Wiley.
Smith, T.R., 1996. A digital library for geographically referenced materials. Computer, 29, 54–60.
doi:10.1109/2.493457
Watters, C. and Amoudi, G., 2003. GeoSearcher: Location-based ranking of searchengine results.
Journal of the American Society for information Science and Technology, 54, 140–151.
doi:10.1002/(ISSN)1532-2890
Wolf, A.T., et al., 1999. International river basins of the world. International Journal of Water
Resources Development, 15, 387–427. doi:10.1080/07900629948682
Zhu, X., et al., 2015. Integrating spatial data linkage and analysis services in a geoportal for China
urban research. Transactions in GIS, 19, 107–128. doi:10.1111/tgis.2015.19.issue-1
INTERNATIONAL JOURNAL OF GEOGRAPHICAL INFORMATION SCIENCE 23
