Changes affecting generalization of land cover features in a smaller scale.
Papsiene, Lina ; Papsys, Kestutis
1. Introduction
National mapping agencies (NMAs) often maintain reference and
thematic spatial data sets to represent various spatial data identifying
natural and anthropogenic phenomena of the world (Kazemi 2003). Usually
they are stored in several scales, e.g. reference spatial data in
Lithuania are collected at 1:10 000 (basic scale), 1:50 000 and 1:250
000 (Papsiene, Papsys 2011), whereas, for example, in Belgium it makes
1:10 000 and 1:50 000 (Bayers 2010). In reference data sets, polygon
features serve to store the features representing land cover, such as
forests, arable land, built-up territories, hydrographic features, etc.
The main task to quickly and effectively update spatial data of a
smaller scale is to use spatial data sets of a larger scale more often
updated as those of a smaller scale. For example, reference spatial data
in Lithuania at a scale of 1:10 000 are updated constantly while those
at the scales of 1:50 000 and 1:250 000 - on a 5 year basis or even
more.
The automatic generalization of spatial data is one of the most
appropriate ways employed in the creation and update of a spatial data
set. The use of automatic procedures in the update of spatial data is
affected by three principal factors: reduction in work and time
resources, the qualification and subjectivity of specialists
(Kilpelainen 2000) and data accuracy achieved by manual update (McHaffie
2002). As a rule, along with generalization methods applied in the
update of spatial data sets, all features of a larger scale are
generalized regardless of whether those features have changed compared
to the earlier spatial data set of a larger scale. Therefore, the
features that have not really changed are also updated. In this way,
each update produces absolutely new features (new data set) having no
relation with the earlier feature version. Such update process requires
high technological and human resources, as it takes time to generalize
all features anew, revise the result later and evaluate whether it meets
the set requirements. Accordingly, such generalization is more
appropriate for creating rather than for updating a spatial data set
based on larger scale data. While updating spatial data sets it is best
to generalize only changes reducing costs. The changed features can be
identified in the two following ways:
-- comparing earlier and later versions of a feature through
various queries;
-- supporting unique IDs of the features that have to be
implemented across all reference spatial data sets at all scales and
tracking changes in the life cycle of the feature (Stankevicius 2008;
Beconyte et al. 2009; Stankevicius et al. 2010). Feature IDs must be
unique throughout the data set and remain unchanged all through the life
cycle of the feature (INSPIRE ... 2009).
2. Generalization of Polygon Features
The generalization process is defined as the process of selecting
and simplifying representation details appropriate to the scale and/or
purpose of the map (ICA ... 1973). Digital cartography distinguishes
three types of generalization by defining a process from reality to
cartographic products (Grunreich 1985; Weibel, Dutton 1999; Cecconi
2003):
-- Feature generalization is employed to create the initial
abstract image of a phenomenon of the real world (e.g. from satellite
images, GPS measurements). Feature generalization produces a primary
data model.
-- The process of model generalization performs controlled
reduction in data. Model generalization is used for creating and
updating a data set of a smaller scale from spatial data of a larger
scale. Model generalization produces a secondary data model.
-- Cartographic generalization is used for developing a
cartographic product. This process comprises visualization operations
and is employed for the generalization of the spatial features of a
primary or secondary model in order to get a cartographic product.
Many years of scientific research had not seen a uniform and
systematic classification of generalization operations. McMaster and
Shea (1992) identified 12 different generalization functions that were
classified according to the transformation type used for generalization:
1. spatial (graphic) transformation: simplification, amalgamation,
refinement, displacement, smoothing, merging, exaggeration, aggregation,
collapse and enhancement;
2. attribute transformation: classification and symbolization.
The AGENT project in 1999 specified the former classifications.
Thus, generalization functions of spatial transformation were identified
depending on the transformation type (attribute transformation and
spatial transformation). In addition, generalization operations under
spatial transformation have been furthermore divided depending on the
features they can be applied for (AGENT ... 1999):
-- individual features: simplification (weeding, unrestricted
simplification) collapse, enhancement (enhancement with regards to
geometric constraints, enhancement with regards to semantic
constraints);
-- individual features or a set of features: selection/
elimination, displacement;
-- set of features: aggregation (amalgamation, combine,
typification).
Furthermore, in 2006, Li presented a systematic classification of
generalization operations depending on what generalization could be
applied for various geometric elements representing features. This
classification identifies the groups of operations used for individual
and a group of point, polyline or polygon features. Individual polygon
features may be applied for the operations of collapse (including
area-to-point, area-to-line and partial), displacement, exaggeration
(including directional thickening, enlargement and widening),
elimination, (shape) simplification, split, whereas a group of polygon
features is applied for aggregation, agglomeration, amalgamation,
dissolving, merging, relocation, (structural) simplification and
typification.
Based on Li classification (2006), Table 1 specifies generalization
operations that may be employed upon the model generalization of land
cover features of a reference data set:
-- elimination and simplification of individual polygon features;
-- aggregation and dissolving a group of polygon features.
[TABLE 1 OMITTED]
In order to properly perform the generalization of spatial features
it is, first of all, necessary (Papsiene, Papsys 2011) to:
-- determine requirements for features, i.e. the density of
features, geometry resolutions, min. area;
-- select algorithms and parameters of generalization operations;
-- determine the priority of selected algorithms;
-- model the generalization process.
The generalization of features must be done in separate object
groups represented by the same phenomena of the world. The
generalization of land cover features requires, primarily, the selection
of proper features according to quality parameters (e.g. selection of
deciduous forests). It should be mentioned that first we cannot
eliminate at once features according to both their attributes and
geometric features (e.g. select only forests with the area over 10 ha).
The reason is that in the next step, the aggregation of the selected
features according to a minimum distance preliminary defined between the
neighbouring features, the features small in the area may, after
aggregation, form the conglomerates of a significant size. The size of
all features must be evaluated and the features that fail to meet
requirements for a minimum feature area must be eliminated only after
aggregation. The last step is the simplification of features. The
conception process of the generalization of land cover is presented in
Fig. 1.
3. Relation Between the Type of Polygon Changes and Generalization
Process
The identification of changes in spatial data includes the analysis
of feature versions at different periods (Singh 1989). The primary task
of identifying changes in features is to decide which features have
changed compared to the earlier version of a spatial data set and what
is the type of changes in features that can be evaluated by answering
several questions presented in Table 2.
[FIGURE 1 OMITTED]
The choice of generalization operations depends on the type of
changes in features. Table 2 shows that some types of changes in a
larger scale may affect different changes in a spatial data set of a
smaller scale. For this reason, it is impossible to make an unambiguous
decision as to what generalization is to be applied as long as all
changes in features and likely influence on neighbouring features are
not analyzed and evaluated.
A new feature in a smaller scale must be created in two cases (Fig.
2):
-- a new feature is identified in a spatial data set of a larger
scale with a quality or quantity attribute represented in a smaller
scale;
-- a feature in a larger scale acquires a new quality or quantity
attribute represented in a smaller scale.
[FIGURE 2 OMITTED]
In these two cases, a feature is selected from a spatial data set
of a larger scale applying a simplification operation; the achieved
result is integrated into a spatial data set of a smaller scale.
If a feature is to be eliminated, no generalization operation is
performed.
A feature in a smaller scale will have to be deleted in cases
opposite to those of creating a new feature, i.e. when (Fig. 3):
-- the deleted feature is identified in a spatial data set of a
larger scale having a quality or quantity attribute usually represented
in a smaller scale;
-- the feature in a larger scale acquires a new quality or quantity
attribute not represented in a smaller scale.
[FIGURE 3 OMITTED]
A feature in a smaller scale is updated when its quality or
quantity attributes or shape in a spatial data set of a larger scale are
changed (Fig. 4). In the first case, feature attributes and in the
second, the feature shape is updated.
[FIGURE 4 OMITTED]
If only the attributes of a feature have changed, no generalization
is needed (only the attribute is updated) while in case of changes in
the shape, feature simplification is to be carried out.
Additionally, the evaluation of the above cases shows it is
necessary to evaluate the distance to the neighbouring features with the
same attribute, i.e. whether it is above or below the minimum distance
allowed:
1. a new feature or the feature that "moved towards" the
neighbouring feature will be aggregated with it, i.e. the feature in a
smaller scale will enlarge (e.g. when a new residential area emerged
close to a former built-up territory) (Fig. 5);
2. upon elimination or "receding" the feature, that was
earlier aggregated with the neighbouring one, will have to be eliminated
from the aggregated polygon feature in a smaller scale, i.e. the feature
will be reduced (e.g. gardening was started in one of the adjacent
fields of the arable land) (Fig. 6);
3. a feature of the changed shape in a larger scale will have
effect on the shape of the feature produced by aggregating neighbouring
features (e.g. a part of one of adjacent forests was cut down) (Fig. 7).
[FIGURE 5 OMITTED]
[FIGURE 6 OMITTED]
[FIGURE 7 OMITTED]
The first case demands a feature simplification operation as well
as aggregation with a neighbouring feature from a spatial data set of a
smaller scale. In the second case, the "unsuitable" feature
needs to be eliminated from the earlier aggregated polygon feature in a
smaller scale. However, instead of the elimination function, it is
enough to newly simplify and integrate the changed features
(additionally, the aggregation function is used for merging the rest of
the features). In the third case, similarly to the first one, the
simplification of changed features--aggregation with adjacent
features--is to be performed.
4. Possibilities of Evaluating the Significance of Changes in the
Polygon Feature
When identifying changes in spatial features, it is essential to
determine significant and to reject insignificant changes (Richard et
al. 2005). A change will be significant when:
-- the acquired new attribute is represented in a spatial data set
of a smaller scale;
-- a change in the feature shape will be seen in a spatial data set
of a smaller scale.
Upon evaluating the significance of changes in features, a single
changed feature needs to be analyzed establishing how its change will
influence surrounding features according to the types of changes
specified in the above section.
The evaluation of changes in features is to be conducted in groups
according to the type of changes in the following procedure.
Group: update of the feature attribute. When evaluating the
significance of changes in the feature attribute of a larger scale, it
suffices to know what attributes are significant for (i.e. attributable
to) the feature in a smaller scale and comparing different versions of
the feature (feature version before and after the update in a larger
scale) to single out the features that have acquired new
"significant" attributes, e.g. the name of the lake has been
specified.
Evaluation in this group of changes requires:
-- to link feature attributes before and after the update through
unique feature IDs (if there are any) or spatial join that creates a
table join in which the field attributes of features before and after
the update are presented based on the relative locations of the
features;
-- to find changed appropriate attributes through queries
("AttributeValueBeforeUpdate" <>
"AttributeValue_AfterUpdate").
Group: creating a feature. The appearance of a new feature will be
significant in cases similar to those of updating the feature attribute,
i.e. if a new feature has proper quality and quantity attributes.
Evaluation requires:
-- to eliminate, through queries, features (before and after the
update) lacking set quality and quantity attributes, i.e. to select the
proper ones;
-- to find new features using information on the life cycle of the
feature (if there is any) or selecting features "after the
update" not intersecting the feature "before the update".
Group: eliminating a feature. Deleting a feature will be
significant in a larger scale if the feature was earlier represented in
a smaller scale.
Evaluation requires finding the eliminated features by using
information on the life cycle of the feature (if there is any) or
selecting features "before the update" not intersecting the
feature "after the update"
Group: a feature under aggregation. The changed feature will be
significant in respect of aggregation operation in case it "has
approached" closer than the minimum distance allowed between
neighbouring features considering the attributes of the same quality and
quantity. Searching for such changes may be easily implemented through
spatial analysis query looking for intersections between the buffer
around the feature of a changed shape and/or new features and a
neighbouring feature. The width of the buffer under creation must be
equal to the defined minimum distance between neighbouring features.
Group: changing the shape of features. The evaluation of the
significance of changes in the feature shape is more complex than that
found in the cases mentioned above.
The significance of changes in the feature shape is determined by
comparing the size of changes referring to the fixed minimum change
allowed, which, first of all, should depend on the scale (this scale
affects the resolution of the map) and specificity of a spatial data
set. Additionally, the expected changes in the size of the object have
to be evaluated after generalization.
Before the analysis of changes in the feature shape is started, the
features with changed shape are immediately rejected, if they are not
represented in smaller scale maps according to parameters of quality and
quantity.
A model has been developed to evaluate the significance of changes
in the feature shape using spatial analysis queries (by ESRI ArcGIS
software). The purpose of the model is to compare the features of a
smaller scale before and after the update, to find changes in the
feature shape, to evaluate them and to select the significant ones that
have to be simplified. The model has been developed on the presumption
that changes in the feature shape will be significant if the
vertex/vertexes of changes (expressed in polygon) will be moved from the
feature boundary (before the update) at a distance higher than the set
(s) is placed, which must, as mentioned above, depend on the scale
(usually it makes 0.05 cm of the map scale).
1. The actions in the query of spatial analysis follow the order
below (Fig. 8).
2. Union of spatial features before (sdata.vi) and after the update
(sdata.v2). Selecting the changed part of the feature (ChangesMinusall,
ChangesPlusall).
3. Determining the significance of buffer width. The size of the
buffer depends on specified resolution that relates to the resolution of
spatial data.
4. Creating buffer zones (according to the specified significance
of buffer width) around the source feature before the update (Buffer).
5. Selecting changes in feature geometry outside the buffer zone
(ChangesMinus, ChangesPlus).
6. Simplifying changes in the geometry of the selected feature
(ChangesMinus simplify, ChangesPlussimplify).
7. Creating buffer zones (according to the specified significance
of buffer width) around updated objects (Buffer).
8. Selecting changes in the geometry of the objects outside the
buffer (ChangesMinusSignificant, ChangesPlusSignificant). The resulted
changes will be significant.
[FIGURE 8 OMITTED]
5. Evaluation Test on Changes in Polygon Features
The period from 2009 to 2010 faced the development of Lithuanian
digital raster orthophotographic map ORT10LT at a scale of 1:10 000,
which served as a base for updating the features of Lithuanian reference
spatial data set at a scale 1:10 000 in 2011. Furthermore, the period
from 2011 to 2012 has been witnessing the update of Lithuanian reference
spatial data set at a scale of 1:50 000 using the already updated
spatial data at a scale of 1:10 000.
Research task: check the correctness of the above described
evaluation methods for feature changes. Research object: features
representing built-up territories in Lithuanian reference spatial data
sets at a scale of 1:10 000. The target territory: the municipality of
Moletai Region.
Accordingly, the following features have been analyzed and compared
with:
Lithuanian reference spatial data set at a scale of 1:10 000 before
the update (2010 version);
Lithuanian reference spatial data set at a scale of 1:10 000 after
the update (2011 version).
The research results are presented in Table 3.
6. Conclusions
The update of polygon features identifying land cover territories
has to be carried out using larger scale generalization, which would
include only significant changes rather than all features from a
reference spatial data set. Changes in specific features must be
evaluated according to their influence on the update of the data of a
smaller scale. The accepted significant changes are those
"seen" in a spatial data set of a smaller scale, i.e. having
"appropriate" quality and quantity attributes for a smaller
scale and meeting minimum requirements for a geometric attribute.
Depending on the type of feature changes, feature update may vary in a
spatial data set of a smaller scale. Besides, the type of a change
affects different analysis of its significance. Therefore, the
evaluation of changes in features must be performed consequently
according to the type of changes in the feature and follow the
procedure: updated attributes, new features, deleted features, features
that have to be aggregated, features having an updated shape.
doi: 10.3846/20296991.2012.728045
References
AGENT Consortium 1999. Selection of Basic Algorithms, version 2.3.
Technical report. Department of Geography, University of Zurich, Zurich.
Available from Internet: http:// agent.ign.fr/deliverable/DD2.pdf
Bayers, E. 2010. Updating topographic reference data (processes,
methods, needs, research): progress report in the Belgian NGI-IGN, in
ISPRS Archives, vol. 38, part 4-5-2/W9: 185-188. Available from
Internet: http://www.isprs.org/proceedings/
XXXVIII/4_8_2-W9/papers/final_190_NGI_Belgium_Paper_Eric_BAYERS_EN.pdf
Beconyte, G.; Kryzanauskas, A.; Papsiene, L.; Papsys, K.;
Stankevicius, Z. 2009. Lietuvos geografines informacijos infrastruktura
--kelias i bendra geografijos metodologija, Geografija [Geography]
45(1): 1-10. ISSN 1392-1096.
Cecconi, A. 2003. Integration of Cartographic Generalization and
Multi-Scale Database for Enhanced Web Mapping: Ph.D. Dissertation.
Zurich University, Switzerland. 155 p. Available from Internet:
http://www.geo.uzh.ch/fileadmin/files/
content/abteilungen/gis/research/phd_theses/thesis_AllessandroCecconi_2003.pdf
Grunreich, D. 1985. Computer assisted generalisation, in Papers
CERCO-Cartography Course. Frankfurt am Main, Institut fur Angewandte
Geodasie.
ICA-International Cartographic Association. 1973. Multilingual
Dictionary of Technical Terms in Cartography. Wiesbaden, F. Steiner. 573
p.
INSPIRE-Infrastructure for Spatial Information in Europe. 2009.
INSPIRE Generic Conceptual Model, version 3.2. Available from Internet:
http://inspire.jrc.ec.europa.eu/
documents/Data_Specifications/D2.5_v3.2.pdf
Kazemi, F. 2003. A generalization framework to derive multi-scale
GEODATA, in Proc. of the Spatial Science Conference. September, 2003.
Canberra, Australia, 1-12.
Kilpelainen, T. 2000. Maintenance of multiple representation
databases for topographic data, The Cartographic Journal 37(2): 101-107.
Li, Z. L. 2006. Algorithmic Foundation of Multi-scale Spatial
Representation. CRC Press (Taylor & Francis Group), Bacon Raton. 310
p. ISBN 9781420008432.
McHaffie, P. H. 2002. Towards the automated map factory: Early
automation at the U.S. Geological Survey, Cartography and Geographic
Information Science 29(3): 193-206. Available from Internet:
http://www.geography.wisc.edu/histcart/ v6initiative/07mchaffie.pdf
McMaster, R.; Shea, S. 1992. Generalization in Digital Cartography.
Association of American Geographers, Washington, USA. 134 p.
Papsiene, L.; Papsys, K. 2011. Possibilities of updating
small-scale basic spatial data in Lithuania using generalization
methods, Geodesy and Cartography 37(4): 143-148. ISSN 2029-6991.
Richard, J. R.; Srinivas, A.; Omar, A. K.; Badrinath, R. 2005.
Image change detection algorithms: a systematic survey, IEEE
Transactions on Image Processing 14(3): 294-307.
http://dx.doi.org/10.1109/TIP.2004.838698
Singh, A. 1989. Digital change detection techniques using
remotely-sensed data, International Journal of Remote Sensing 10(6):
989-1003. http://dx.doi.org/10.1080/01431168908903939
Stankevicius, Z. 2008. Feasibility study of united national
cartographical model, in 7th International Conference
"Environmental Engineering': Selected papers, vol. 3. Vilnius:
Tech nika, 1483-1487.
Stankevicius, Z.; Beconyte, G.; Kalantaite, A. 2010. Automation of
update of digital national geo-reference databases, Technological and
Economic Development of Economy 16(2): 254-265.
Weibel, R.; Dutton, G. 1999. Generalising Spatial Data and Dealing
with Multiple Representations, in Geographic Information Systems:
Principles and Technical Issues, vol. 1. Ed. by Longley, P. A.;
Goodchild, M. F.; Maguire, D. J.; Rhind, D. W. New York: Wiley, 125-155.
Lina PAPSIENE. PhD student at the Department of Geodesy and
Cadastre, Vilnius Gediminas Technical University, Sauletekio al. 11,
LT-10223 Vilnius, Lithuania; Ph +370 5 274 4703, Fax +370 5 274 4705,
e-mail: gkk@vgtu.lt. A graduate from Vilnius Gediminas Technical
University (MSc in measurement engineering / geodesy and cartography,
2000), actively participates in the processes of developing spatial
information infrastructure and producing basic (reference) spatial data
at the scales of 1:10 000, 1:50 000, 1:250 000 in Lithuania. Research
interests: generalization and harmonization of spatial data, SDI.
Kestutis PAPSYS. PhD student at the Centre of Cartography, Vilnius
University, M. K. Ciurlionio g. 21, LT-03101 Vilnius, Lithuania; Ph +370
5 272 4741, Fax +370 5 373 7723, e-mail: kestutis.papsys@gf.stud.vu.lt.
A graduate from Vilnius University (MSc in geography, 1999). Actively
participates in the process of developing spatial information
infrastructure in Lithuania. Research interests: GIS methods for
predicting natural, social and ecological hazards, modelling with GIS,
SDI.
Lina Papsiene (1), Kcstutis Papsys (2)
(1) Vilnius Gediminas Technical University, Sauletekio al. 11,
LT-10223 Vilnius, Lithuania
(2) Vilnius University, M. K. Ciurlionio g. 21, LT-03101 Vilnius,
Lithuania
E-mails: (1) gkk@vgtu.lt (corresponding author); (2)
kestutis.papsys@gf.stud.vu.lt
Received 10 June 2012; accepted 21 September 2012
Table 2. Possible types of changes in features
Question Type of change
in a larger scale in a smaller scale
Is the feature new? New feature New feature
Has an attribute of Updated feature New feature or
the feature changed? attribute Updated feature
attribute
Has the shape of the Updated feature Updated feature
feature changed? shape shape
Has the minimum distance Updated feature Updated feature
between neighbouring shape shape (aggregated
features changed? feature)
Has the feature been Deleted object Deleted object or
deleted? Updated feature
shape (no
aggregated feature)
Table 3. Test results
Group of changes Build-up territories
Changes in Changes significant
1:10 000 for 1:50 000
Update of the 0 0
feature attribute
Creation of the 1184 1038 *
feature
Elimination of the 198 199
feature
Feature under Not applicable 2034 **
aggregation 199 ***
Changes in the shape 27333 11932
of the feature
* new features bigger than 0.01 ha.
** features have to be aggregated in a smaller scale
(distance less than 5 m)
*** features have to be eliminated from aggregated
features in a smaller scale (distance more than 5 m)