Automatic extraction of terrain features from digital terrain data: A multi-faceted study

Ridge and valley lines are important terrain features to many scientific endeavors and practical applications. They are extracted manually from topographic maps traditionally. Besides being tedious, the manual process involves much arbitrariness by human interpreters and the results are not repeatable nor consistent. The lack of repeatability and consistence undermines the usefulness of the extracted results. The increasing availability of digital terrain data provides an alternative that may remedy the shortcomings of manual extraction.;The automatic delineation of terrain features is a multi-faceted task that is relevant to cognitive issues, terrain modeling, and computer implementation. This research has identified four groups of methods for the extraction of ridge and valley lines. They are: symbolic approach, tracing approach, profiling approach, and hydrological approach. The embedded meaning of ridge and valley lines in each of these methods is investigated. A series of tests are conducted inside computers to evaluate the performance of these methods. A primary investigation on the symbolic method concludes that the difficulty pertaining to the generation of TINs undermines the feasibility of the symbolic approach, thus it is not pursued further. The other three groups of methods have been tested and compared on the basis of the numbers, continuity, agreement, and positional accuracy of the extracted features. It is concluded that the hydrological approach performs the best generally.;The hydrological approach, instead of emulating manual interpretation in computers, takes an innovative approach that makes good use of the computational power of modern computers. By defining ridge and valley lines with accumulation values, this method is less sensitive to local terrain variations and extracts a rather continuous and complete result. In contrast, the tracing and profiling methods attempt to emulate manual process in extraction and the outcomes turn out to be not satisfactory. The various performances of the three methods present a notion that direct replication of human knowledge into computers is not necessarily feasible in the development of automatic methods. Several topics for future research are identified and subject to further study.