Study On Ontology-Driven Typical Landform Extraction In The Loess Plateau Using DEM

Precise division of landforms is very important for a profound understanding of occurrence, development and evolution of regional environmental systems, thus it has a realistic and broad application requirement. Loess Plateau, with an area of322,300square kilometers, is located in the inland areas of western China. Sedimentary strata of the Loess Plateau are homogeneous in high level. Typical geomorphic types can be represented as combinations of slope units. Within the Loess Plateau region, geological disasters such as soil erosion, landslides, mudslides and so on occur frequently. Therefore, it’s necessary to timely, accurately monitor the occurrence of geological disasters and mitigate the effects of disasters. Most of the currently completed Loess Plateau landform maps are a rough outline of landforms with a small scale, therefore it is difficult to accurately reflect the location and distribution of typical landforms of the Loess Plateau. It is important and necessary to study the automatic extraction of loess plateau typical landform types in a large-scale. In the last10years, with the development of satellite and remote sensing technology, the resources of global high-resolution digital elevation model are becoming richer, and the object-oriented image analysis techniques have emerged and rapidly developed, all of this creates favorable conditions for our work.Digital elevation models contain not only rich terrain information but also rich geomorphic information. It is still a problem to extract geomorphic information automatically from digital elevation models. When landform mapping experts interpret landform types from satellite imagery or DEM data through traditional pixel-based techniques, expert knowledge is often not clear enough, thus there are dissimilarities between the different users of knowledge, and therefore landform mapping experts will face the challenge of variability of landform forms, thus it is difficult to automatically extract landform types.Over the past10years, with the development of object-based image analysis techniques, a method made??new progress, which is to analyze and extract landform information through geographic information ontology modeling approaches. In this paper, based on geographic information ontology modeling approach and the thought that typical landform features are represented as a variety of combinations of slope forms, we established the conceptualized, formal ontology model of the Loess Plateau landform types, and then designed the classification rule sets for object-based image classification. Then we segmented30m spatial resolution ASTER DEM, and extracted three typical landform types:loess tableland, loess ridge, loess hillock plateau automatically. The results are well consistent with the visual interpretation, thus can achieve a higher level of accuracy.The main contribution of this paper is:(1) We designed the quaternary structure of landform ontology of Loess Plateau, and the three steps of ontology modeling of loess tableland, loess ridge and loess hillock plateau (knowledge description, ontology characteristics determination, ontology model implementation). In the ontology modeling process, by introducing the thought of slope form combinations we linked the concepts of landform types between the real world and the image world and build a bridge between them, thus the link of landform domain knowledge and remote sensing image information is possible during the object-oriented analysis process, and so we made up the weakness of traditional pixel-oriented landform extraction method which breaking the integrity of landform units.(2) Combined the thought of ontology and slope forms, we determined the layers involved in division during the object-based classification process, and using landform ontology modeling method we obtained the landform classification rules set based on ASTER GDEM digital elevation model, and achieved the automatic extraction of landform types. In the accuracy evaluation, we compared the extracted results of Shaanxi Yijun region with results from other major extraction methods. The results of accuracy evaluation show that the method herein can reach a satisfactory level of accuracy.(3) Compared with the published landform maps, the extraction results herein can be more accurate when represent the location and distribution of loess tableland, loess ridge and loess hillock plateau. And the mapping method herein is convenient, fast and accurate, thus can meet the needs of medium and large scale landform mapping to a certain extent, and provide a way for the medium and large scale landform mapping of Loess Plateau from the perspective of thinking and method, and also provide a more objective and effective way for the study of landforms.