Advances in remote sensing and geographical information science (GIS) resulted in proliferation of medium- to-high resolution digital elevation models (DEMs) of global or continental-scale spatial extents. A DEM requires further algorithmic processing in order to provide information for geomorphologic and/or geologic analysis. The standard DEM-derived product is a map of shaded relief that provides a three-dimensional depiction of the terrain. It enables visual terrain analysis, but it is not meant for an automated, algorithm-based analysis. This research project will develop a methodology for auto-generating a thematic map of landforms and tools for its automated analysis. Unlike shaded relief maps, which depict variations in elevation field, the new "geomorphometric" map will depict semantic interpretation of terrain. It will offer an alternative tool for visual geomorphologic analysis, and it will be susceptible to automated terrain analysis. Machine vision-based methodology will be used to generate a robust geomorphometric map in a manner that is computationally efficient. The investigators will deliver a geomorphometeric map of the entire conterminous Unites States based on 30 meters per pixel DEMs. Automated analysis of terrain will be performed by measuring a quantitative similarity between any two "landscapes" for small sections of the map. Similarity-based analysis will be used to determine landscape archetypes, which are characteristic patterns of landforms found across the conterminous Unites States. The new methods also will used to calculate a degree of correspondence between real and simulated landscapes, thus assessing relative success of different landscape evolution models.
This project will use machine-vision and similarity-measure techniques to provide better tools for terrain analysis. The project will yield a method for robust and efficient generation of geomorphometric maps, which will provide significant enhancements to shaded relief, thereby facilitating more complete analysis of terrestrial and planetary terrains. The project will provide objective means of landscape classification, a new direction in the field of geomorphometry. It will enable objective, easy-to-interpret comparison of synthetic landscapes to ground truth and produce much-needed quality metric to landscape evolution models. The project further integrates domains of geography and spatial science with computer science domains of pattern recognition and content-based image retrieval.
