Neighborhood contextual merging of segments for landform detection and classification

Contextual merging of segments using spatial relations of the neighborhood and the geomorphometric variable of the neighbor segments is introduced as a method to improve the over-segmentation of object-based analysis. The scope of the approach is to provide better candidates for the detection and delineation of landforms with simple geomorphometric signatures like sinkholes and aeolian closed depressions. The approach is simple and effective and can be implemented in many GIS software. We exemplify it for the aeolian closed depression from a part of the Romanian Plain by using elevation standard deviation to merge the segments of such closed depressions. While generally, these can be considered as concave features; actually, these have multiple shapes: a flat bottom surrounded by concave, straight, and convex parts. The results are promising, but the approach needs to be better framed in a GUI GIS environment to be implemented for specific landforms detection and delineation tasks.