Seminar Abstract:

Investigations in uncertainty modeling are becoming key topics in spatial statistical studies. Initially, point based approaches resulted into geostatistical developments. Currently, there is an increased interest in modeling objects and their uncertainty. This presentation shows how random sets are being applied to classify and represent the extensional uncertainty of spatial objects. Random sets model uncertainties in vegetation patches and in the lake extent in the Poyang Lake area in China. Vegetation patches are of objects of interest as they have both sharp and vague boundaries, whereas the lake extent is uncertain because of wetland inundations changes within a year and between years. Extensional uncertainty of extracted objects is quantified by the statistical characteristics of random sets. The accuracy of the classified uncertain objects is addressed by relating the covering function with the measured variables, by testing the similarity of their distributions and by estimating the thematic accuracy of the classifications. Significant correlations exist between the covering function and vegetation coverage. Random sets reveal subtle changes of wetland hydrology that were not visible from a crisp approach. We conclude that several characteristics of extensional uncertainty of segmented objects can be quantified using random sets. The random set model thus enriches spatial modeling of uncertain spatial phenomena.
