Copyright (c) by the paper's authors. Copying permitted for private and academic purposes. In: A. Comber, B. Bucher, S. Ivanovic (eds.): Proceedings of the 3rd AGILE PhD School, Champs sur Marne, France, 15-17-September-2015, published at http://ceur-ws.org
Mapping applications, i.e. websites where a multi-scale navigation in topographic maps is available, may be used for wayfinding purpose for instance. Such mapping applications display multi-scale maps, i.e. set of maps at different scale. However, the displayed representations can present large differences (Fig. 1). To recognize the different representations of the same real world entity across scales, users sometimes have to make a quite important cognitive effort. We make the assumption that adding intermediate representations may be one way to reduce the gaps between the existing levels and thus to ease the user navigation.
Scale Fig. 1. These different levels of a multi-scale map may present large differences (IGN-France). Developing methods to automatically derive these intermediate representations requires dealing with the following research issues: – At which scales some intermediate representations are needed? – How could the intermediate representations be designed? – How can we automatically derive these intermediate representations? 2
To produce a map, a cartographic vector dataset and its symbolization must be defined according to the map scale and purpose. To derive the less detailed levels in the multi-scale map, the producers may use less detailed sources, or cartographic generalisation. This process aims at simplifying the map content and symbolization to fit a given scale, ensuring the map legibility. The generalisation operators and their parameters have to be chosen depending on the target scale and the map purpose. An iterative adjustment of the generalisation process is often needed to refine the result.
To ensure the map legibility, the generalisation process may reduce the
quantity of information and simplifies their geometry and symbolization.
Modelling this process for a multi-scale map forces us to define a smooth and
consistent evolution of the map content. These issues are addressed by the
literature related to level of detail and degree of generalisation [
Researchers in the generalisation field are currently working on its automation,
but it is still an issue to model this process [8]. Some structures, such as the
ScaleMaster system [1], provide a multi-scale generalisation framework. This
system allows the modelling of a generalisation process on a whole scale range
(Fig. 2). Its implementation on a Java platform [
As inconsistencies between scales have been identified as a disturbing factor
[2], the added intermediate levels should be consistent with the existing levels. We
will specially care about the spatial relationships consistency, such as maintaining
topological relationships or spatial patterns, which can guide the user navigation.
To ensure this consistency, the generalisation process should take the existing less
detailed representations into account, as proposed in [4]. It supposes that the links
between multiple representations are available. It could be managed with a
MultiRepresentation Database (MRDB) for instance [
Studying existing mapping applications helped us to identify potential approaches for navigation improvement. We analysed a group of sixteen general and national multi-scale maps (e.g. Google Maps or the French geoportal). The identification of heterogeneous representations was our first concern, as they are difficult to interpret by users and thus represent the problems to solve. If we are able to detect these disturbing patterns, we know when an intermediate level is needed and which type of transition it has to improve.
We will have to define at which scale intermediate levels should be derived. The study of mapping applications allowed us to compare the different used strategies for the scales distribution. For instance, a common scale in most applications is probably a useful scale and maybe a required one. Then, we will try to define a difference threshold between two levels. User tests could help us to define this maximal distance beyond which the feeling is disturbing.
We will then need a difference measurement. Two levels can be different in
terms of content, symbolization or both. Regarding the content difference, the
generalisation process may decrease the number of objects (selection, aggregation
or typification), but also modify their shape or move them to ensure their
legibility. Measuring the number of deleted, aggregated and modified objects may
give us an idea of the content difference between two levels. On the other hand,
the symbolization difference may be owed to a simplification of the semantic
abstraction, with fewer classes for the roads for instance. To measure this specific
difference, we could compare the number of classes in each level or maybe more
meaningfully, the number of objects which have changed class. More generally,
visual density and clutter measures [
Then, we will model a multi-scale generalisation process to provide the automated levels derivation. For each intermediate level to produce, we will have to choose the source datasets, the generalisation operators and their parameters. Knowledges to guide this process will come from the study of generalisation strategies in existing maps. Then, we will use the ScaleMaster system [1,11 to orchestrate this complex process. The generalisation modelling will be guided by iterative tests, where the results will be validated by common generalisation evaluation methods [5,9], but also with the difference measurement previously mentioned.
We will have to improve the ScaleMaster system [
Finally, user tests could help us to validate our hypothesis than adding intermediate levels can ease the user navigation in a multi-scale map. Acknowledgments This work is supported by the French National Research Agency, as part of the MapMuxing project [ANR-14-CE24-0011-01]. 4