Content-based landscape retrieval using geomorphons

We introduce a concept of content-based landscape retrieval (CBLR). Our CBLR system retrieves morphometrically similar landscapes from a large DEM thus acting as a search engine for landscapes. The system works on the principle of query-byexample, a reference landscape is given and the system outputs a map showing degree of similarity between a reference and all the other local landscapes across the spatial extent of the DEM. Landscape is defined as pattern of landform elements. A DEM is converted into a map of landform elements using the geomorphons method. The core of the CBLR is the similarity function between two landscapes that encapsulates a degree to which their patterns of landform elements correspond to each other. The search relies on exhaustive evaluation of similarity using an overlapping sliding window approach. In the featured case study we use our method to delineate a spatial extent of the characteristic landscape formed by the end moraine associated with the latest glaciations across the country of Poland. Good agreement is found between the region delineated by our method and a range of end moraine manually delineated from geomorphic, geological, and paleogeographical information. The CBLR can be implemented as a GeoWeb application and serve as a rapid and convenient tool for exploration of very large DEM datasets.