INCREMENTAL UPDATE IN AN MRDB

K.-H. Anders¹, M. Sester¹, J. Bobrich²

1 - Leibniz Universitaet Hannover, Institute of Cartography and Geoinformatics, Hannover, Germany

2 - Federal Agency for Cartography and Geodesy, Frankfurt, Germany

karl-heinrich.anders@ikg.uni-hannover.de

In this paper we will give an overview about the state of the art in incremental generalisation in a MRDB and open problems. A Multi-resolution/representation-database (MRDB) can be described as a spatial database, which can be used to store the same real-world-phenomena at different levels of precision, accuracy and resolution. Furthermore these phenomena can be stored for different ways of presentation or symbolisation. Also, data stemming from different applications can be stored. All these corresponding objects relating to the same physical objects or phenomena can be stored and linked.

We will discuss the requirements and goals of an MRDB from the standpoint of an NMA, in this case the Federal Agency of Cartography and Geodesy of Germany. For the realization of an MRDB two approaches can be used. The one way is the complete automatic derivation of all target scales from one base Digital Landscape Model (DLM). The other approach is the linking of existing DLM’s by a matching process. We will discuss the advantages and drawbacks of these approaches.

There are several reasons for introducing an MRDB: On the one hand it allows a multi-scale analysis of the data: Information in one resolution can be analysed with respect to information given in another resolution. On the other hand a major reason for National Mapping Agencies to investigate and implement MRDB is the possibility of propagating updates between the scales, which is also called “incremental generalization”. The updates which are to be propagated through such a system can be geometric or semantic changes. With semantic changes we mean updates of attribute values. The advantage of this incremental generalization approach is that an automatic incremental update process for a spatial database storing landscape models of different scales can be implemented which keeps the different models in a consistent state. Therefore only one landscape model (the most detailed) has to be updated manually in future, the derived data sets can be updated automatically by exploiting the links between the corresponding objects.

We will give a detailed discussion of the incremental update process in an MRDB. This complex process contains three main tasks: Change Detection, Model Generalisation, and Cartographic Generalisation. First of all, the objects that have changed have to be identified; this can be done either by assigning a unique identifier to all objects; if this is not possible or available, geometric matching techniques are needed. Then, the changes have to be propagated to the derived scale using generalization techniques. As the links are available, also the range of influence of a change can be limited. In this way, the relevant and possibly affected objects are identified by the link structure and the model generalization is performed. There are already very powerful methods for this, thus the automatic model generalisation can be considered mainly as solved. A still open problem is the complete automatic incremental cartographic generalisation of a DCM (Digital Cartographic Model) which is linked to a specific DLM. One possibility is to introduce the changes that occurred between the two scales as deformation space and apply these deformations also for the changed object. Another option is to enrich the links with more information about the earlier generalization process. Here, especially also the affected and participating objects have to be included. Based on this information, either a re-generalization can be triggered, or inferences can be drawn about the possible influence of the deformation to the changed object. These issues will be discussed in detail in the paper. An additional difficulty arises, when date from different sources have been linked, and thus the links stem from a matching process instead of a generalization process.