A Machine-Learning Approach to Generalisation of GIS Data

A. Forghani, S. Kazemi, D. Bruce

Research output: Contribution to journalArticlepeer-review

5 Citations (Scopus)
29 Downloads (Pure)


The automation of map generalisation in this study involves an expert system approach that consists of four main components including knowledge acquisition, an inference engine, knowledge representation and a user interface. The acquired knowledge was then utilised to build a knowledge-based solution: a ‘Generalisation Expert System’ (GES) developed in Java, Python and C programming environments for the delivery of generalised geographical features. Its capabilities are demonstrated in a case study through generalising several line and polyline databases over the study area in Canberra, Australia. The cartographic and GIS software communities will benefit from this study through access to a set of tools, guidelines and protocols that incorporate a standardised cartographic generalisation methodology. The results of the trials utilising GES were analysed: a series of generalisation routines were performed to assess the quality of simplification results for different spatial layers. Cartometric measures such as the total length and number of line or polyline segments were used as indices of generalisation to quantify generalisation performance for the target small scale. For example, there are 101,228 segments in 1:250,000 scale and 9,491 segments in 1:500,000 scale contours over the study area. This requires a reduction in the complexity and the density of elevation data. Changes in the representation of contour features at 1:250,000 and 1:500,000 scales as a result of generalisation were quantified. Outputs from map derivation have been analysed applying the Radical Law, this determines the retained number of objects for a given scale change and the number of objects of the original source map. Testing demonstrated that the implemented algorithms in GES are able to extract characteristic vertices on the original entity lines and polylines (e.g. for roads) while excluding non-characteristic lines and polylines to reduce irrelevant computation. This study has demonstrated reasonable improvements in Vertex Reduction, Classification and Merge, Enhanced Douglas-Peucker and Douglas-Peucker-Peschier algorithms. The test results show that GES generalises line features accurately while still maintaining their geometric relations. Existing generalisation software requires advanced technical skills from users; GES however, has a basic and user friendly Graphical User Interface (GUI) which is an advantage to users with basic technical skills and understanding of spatial data management. Changes to geographic parameters should be updated in multi-scale maps and spatial databases in near real-time. GES can be developed as a potential tool for generalising large-scale maps into smaller scales, and creating maps of different themes across a variety of scales. Test results have also demonstrated that the methodology developed improves the efficiency of line and polyline generalisation. This study aims to contribute to generalisation system design and the production of a clear framework for users. Experiments presented in this book can be applied to real world problems such as the generalisation of road networks and area features using GES. Future research should be directed towards developing web mapping platforms with generalisation functionality at varying scales.

Original languageEnglish
Pages (from-to)41-59
Number of pages19
JournalInternational Journal of Geoinformatics
Issue number2
Publication statusPublished - Mar 2021
Externally publishedYes


  • Machine-Learning
  • GIS
  • Data


Dive into the research topics of 'A Machine-Learning Approach to Generalisation of GIS Data'. Together they form a unique fingerprint.

Cite this