BALANCING GROUND TRUTH AND CARTOGRAPHICAL EXPRESSIVENESS: AN OPEN SOURCE SOLUTION FOR HEIGHT CONTOUR FILTERING
For more than a century, height contours on Danish maps were based on material from 19th-century military surveys. While not having high absolute precision, these contours were known for conveying correct impressions of the landscape, as decades of revision by skilled cartographers perfected the graphical communication aspect.
But with modern requirements for high precision heights (hydrology, orthophotos), the need for a new DEM became increasingly evident. Hence, a new national coverage, high resolution DEM was released in 2009.
With a new DEM, we also need new contours. But raw contours from direct interpolation in the grids are way too detailed for cartographical communication. And simplification by Douglas-Peucker (or similar algorithms) typically results in hard-to-interpret, topologically problematic contours. This is far from the highly regarded cartographical expressiveness of the old contours.
To regain some cartographical expressiveness, we developed a contour generalization based on filtering of the raw grid, followed by a small dose of Douglas-Peucker generalization. The real challenge was to get the cartographical expression right since the design criterion of the filter is not given by a skill metric, but rather buried in the skulls of skilled cartographers. In order to leverage these skills, we used an iterative filter design, gradually modifying the filtering process, while doing production runs and evaluations with a group of very experienced cartographers.
The final filtering design included iterative application of both isotropic gaussian filter kernels, and the highly anisotropic Perona/Malik filter. But several devils lurk in the details: additional complications come from the wish to avoid growth of data-void areas, and to avoid modifications to local extrema. But with these complications accounted for, we finally have a new set of contours that approximate the cartographical expressiveness of the original contour set, while having much better absolute heights and consistency with the gridded DEM.
Thomas Knudsen - National Survey & Cadastre,