Automatic Generalization up to 80% Correct

Contrary to common belief in the past, critical parts of the map production process, including cartographic generalization, can be automated to produce quality results. This saves valuable time and resources. In a recently published article in 'Geomatik Schweiz' (May 2010) swisstopo, the Swiss Federal Office of Topography, talks with facts and figures about how they are achieving excellent results using the automatic generalization in axpand. The following is an excerpt: 


The national maps of Switzerland are high-quality topographic maps at 1:25000 to 1:1 000 000 scales. In order to satisfy the ever-increasing demand for vector GIS data, the Federal Office of Topography swisstopo is working on changes to their production process ranging from the compilation of topographic base data to the production of maps. Examples are used here to illustrate why automation is necessary for successful restructuring of the production of cartographic GIS data and in which areas automation can be successfully implemented.

From copper plates to glass engravings to CAD cartography

When the Federal office decided in 1935 to create a complete sets of maps in 1:25 000 to 1:1 000 000 scales, no one imagined that the creation of the 1:50 000 scale map series would take 26 years (1937-63) and the 1:25 000 series would take 29 years. During this time, not only were the maps produced, but the production methods and processes were improved in order to simplify and accelerate the work.

At the beginning of the 1950s, copper engraving was replaced by glass engraving for original production and offset printing was introduced. This allowed for the printing of more than 100 sheets per hour for the first time. By the end of the 1980s, numerous other critical technical innovations, eg. typesetting, followed. This allowed for the continuous improvement of the efficiency of the updating process. The majority of the work, especially cartographic editing, continued to demand complex manual work. Cartography was seen as an area of specialization that required highly demanding handicraft skills that could not be automated -- a sort of small, gallic village in the midst of increasing automation. The first trials for the digital updating of topographic maps was attempted in 1989, but glass engraving weren't replaced by digital CAD cartography until the beginning of 2001. The CAD sytem provided new tools, such as zooming or the editing of Bezier curves, which allowed for rationalization of the work and simplified the rectification of errors significantly. Even though there were very few automated process steps at this point in time, the efficiency of cartographic editing was significantly increased.

From CAD to GIS Cartography

Already at the end of the 90s the rapid development of the internet made it clear that the demand for location-based information in electronic form would continue to grow. A clear advantage of this vector GIS data is that it can be enhanced with additional information and linked to other data. For this reason, it is often referred to as 'intelligent data'. At the beginning of the new century, before the advent of Google Maps, which dispelled all doubts about these developments, swisstopo launched two large projects that were designed to move future production completely to vector GIS data production: TLM and OPTINA-LK. The goal of the TLM project is to build a uniform, high-resolution 'topographic landscape model' (TLM, see illustration 1). The project has since been completed and the production of TLM began in 2008. The goal of OPTINA-LK is to make the production of a digital cartographic model (see illustration 2) and the derivation of topographic maps from the TLM possible. Object-type landscape elements with vector geometry and diverse attributive characteristics are modelled in the TLM and DCM models. The objects are structured according to object types with similar characteristics which are in turn structured in themed blocks such as streets, railways, buildings, waterways, borders, etc. The TLM model is available completely in 3D with three coordinates per interpolation point (x,y,z), while the DCM model is stored as a 2D model (x,y). Raster data is also a part of these models, the TLM includes a high-resolution landscape model 'DTM' and the DCM includes layers such as cliffs, boulders or scree and relief. The new data opens up considerably expanded and more flexible use - the most important motivation for the change to GIS data. This data opens up many opportunities - including the automation of a large number of previously manual process steps - at a level that was impossible in the past. Automation is, however, not an end in itself. Compared to the 29 years it took to build up the 1:25 000 scale topographic map series using glass engraving, building up the DCM25 model should take a mere 6 years. Because the creation of the data is much more time intensive than updating is, a high level of automation is critical. The following example, based on the production process for the DCM25, shows which steps can be automated in the future.


Automatic generalization

Once the data is available in a cartographic source model, generalization - the automation step most important from a cartography point of view - takes place in the system "SysDab". SysDab is being developed within the framework of the project OPTINA-LK by Axes Systems AG and is based on "axpand", a data-based cartographic GIS system. SysDab is responsible for the fully automatic generalization of the source model (illustration 3). Generalization, alongside text placement, is the core of cartography. Generalization is concerned with making maps readable in spite of limited space in which to display the data. Where necessary, important elements are emphasized and less important elements are simplified or eliminated. Generalization is an extremely complex process in which numerous content, geometric and topological rules come into play. For a long time generalization was considered to be a step that could not be automated. Cartographers require a four-year professional education and significant professional experience in order to learn how to manually generalize the various topographic map scales. The challenge for automated generalization is to define contextual and geometric rules in a format that the computer can understand and then implement them in an effective process. The operators that use these rules need to work precisely and take neighboring map elements into consideration in almost all operations. The result of the automated generalization in SysDab (system goal) must be so good that fewer than 25% of the objects need to be edited manually. SysDab already fulfills this goal. Recent tests using mid-land map sheets showed that and average of 80% of the objects in the DCM25 were correctly automatically generalized according to generalization rules. This is a very good result. An important element is an excellent building generalization during which complex forms are simplified and objects are aligned to each other, as well as a number of other well-functioning generalization operators. There is potential to further improve the automatic generalization in SysDab for the DCM25.



The examples above have pioneer character. Thanks to technical developments and innovative solutions, it is possible today to automate much of the work of cartographers. This way cartographers can concentrate on more complex challenges that still require know-how, intuition and a trained eye. Cartographers will not be replaced by automation since data-base supported GIS cartography is extremely know-how intensive. The job description of cartographers will change, however, and new knowledge will be required. Fortunately, the professional education for cartographers has been wisely combined with that of geo-informatics and surveying to one educational direction which is geomatics. This will ensure that the knowledge necessary for GIS cartography can be sustainably developed and maintained over time.

The full article in the original German can be accessed as a pdf using this link.