A Methodological Knowledge Acquisition of Cartographic Generalization Practices at Mapping Agencies through an International Survey Overcoming the Knowledge Acquisition in Cartographic Generalization: A Heuristic Natural Knowledge Transfer from Cartographers to Artificial Intelligence Systems

Artificial Intelligence has played an important role in spatial data management and map production processes across Geographical Information Systems (GIS) applications. Studies by the authors show that knowledge acquisition within existing generalisation systems e.g. generalization of line and area features has not been fully implemented. Also many of the implemented generalization algorithms are generic and unable to offer a total solution for the operational environment. The motivation of this research is to develop a workflow for derivation of multiple scale maps from a master database. For the automation of the map generalization process, it is necessary to integrate cartographers’ experience and intuition with the generalization operations within GIS. This paper aims to present the process of acquiring knowledge from a Cartographic Generalisation Survey. A survey of cartographic generalisation practices was conducted from November 2005 to May 2006 at several National Mapping Agencies (NMAs), state mapping agencies and a number of software vendors, in order to capture the cartographers’ knowledge about the principles of cartographic generalisation and their experience with existing generalisation software. The survey was designed to collect experts’ recommendations in relation to new technologies and future generalisation research that could be undertaken by universities and the spatial information industry. The survey results are being utilised to build a knowledge-based expert system which is known as “Generalization Expert System”, built in JavaPython delivering automated generalisation of lines and polylines. Cartographers’ feedback was provided in the form of broad qualitative statements and was analysed to obtain the most pertinent comments. Statistical responses were assessed in quantitative terms. This paper provides key findings from the analysis of the survey results which input criteria into the development of a conceptual framework to generalise spatial databases and a practical realisation of the framework in order to deliver coherent capabilities and automate the generalisation of features as much as possible for “derivative mapping” applications. 1.0 INTRODUCTION Knowledge discovery has lead to the amassing of very large repositories of customer, operations, scientific, and other types of data using a number of techniques such as predictive modelling (Provost and Kolluri 1999). Survey research in general aims to collect information from sample representatives of the total population of survey targets. Then the information gathered from the survey sample is used to make a generalisation about the view of the total population with the limitation of random errors. Two major criticisms are regularly made in the literature when discussing surveys of this nature, one is the disregard for sampling errors due to the sample size, and the other is the disregard for responses vs. non-responses bias (Wunsch, 1986). It is essential that these two issues should be addressed when designing a quantitative