Sommaire de la publication

Sommaire de la publication

Levesque, M.-A., Y. BĂ©dard, M. Gervais, R. Devillers, 2007, Towards managing the risks of data misuse for spatial datacubes, Proceedings of the 5th International Symposium on Spatial Data Quality, June 13-15, Enschede, Netherlands


At the end of the 1990s, a new category of decision-support applications specifically designed for exploring and analyzing spatio-temporal data entered the geomatics field. These applications, called Spatial OLAP (SOLAP), add a cartographic component to the traditional OLAP databases designed for decision-support. They allow users to interactively explore and cross-tabulate their spatial data at different levels of granularity for different themes. There is no longer a need to be a GIS specialist to explore spatial data, no need to learn a query language such as SQL, and no need to wait more than 10 seconds to get the answer regardless of each theme level of granularity and of the complexity of the analysis. With a few mouse clicks, SOLAP users reach a level of power never achieved before. They can interactively visualize the data in different ways (maps, tables and diagrams), drill-down for more details, drill-up to obtain the global picture, drill-across for comparisons, create comparison maps, synchronize views, etc.

Such facility and speed improve the usersÂ’ capability to analyze spatial data and support decision processes. However, they also open the door to a larger group of users who can directly interact with the data but may not completely understand the underlying limitations of the data, especially when several themes are cross-tabulated and temporal data are involved. Furthermore, expert users may easily forget certain details and nuances about present and past data when facing such a level of interactivity. Thus, more than ever before, such factors may lead to faulty interpretations.

These issues that have been also encountered in the field of Geographical Information Systems (GIS) have lead to research on «error-sensitive GIS», «error-aware GIS» and «error-quality GIS». In addition, recent studies have also demonstrated the legal obligations of data providers to warn or advise spatial data users, especially for non-experts, to reduce uncertainty and prevent the misuse of data. Consequently, the arrival of SOLAP technologies and spatial datacubes requires to revisit these issues in order to face these new challenges.

This paper aims at proposing a generic approach for designing and integrating a warning system into SOLAP applications in order to draw usersÂ’ attention to elements that could cause problems in their analysis. Warnings have been used for many years by product manufacturers (e.g. warnings in a user manual or on the box of the product), organizations (e.g. warnings in public swimming pools), or even by software companies (e.g. Microsoft Windows exceptions messages) to prevent people against potential hazards related to the use of their products. Similar to other fields, the communication of warnings to users could be one of the potential solutions to prevent the misuse of datacubes. Despite a higher need of prevention with SOLAP applications, these applications nevertheless constitute a more controlled and defined environment of analysis than GIS applications. This increases the feasibility of implementing a solution such as a warning system.

Our research provides guidelines for categorizing, designing and integrating warnings into SOLAP applications and a way to communicate them to the end-user. Different cases were identified when warnings should be provided to the end-user, especially when the decision was taken to not implement some integrity constraints in the database, when changes in the semantic or the structure of the data took place over time, when semantic or geometric aggregation of the data was not straightforward, etc. A large majority of those cases can be detected during the design of the datacubes since they are the results of a design decision. We propose in this sense to improve system development methods by formally identifying and documenting the cases when warnings are found to be necessary, and then to make those warnings context-sensitive in the system. These warnings would eventually be classified according to the level of risk that they represent and to end-user pr