PhD Defence Ms Xi Zhao

Department of Earth Observation Science

Title of defence

Random sets to model uncertainty in remotely sensed objects

Summary

Remotely sensed images as a major data source to observe the earth have been extensively used in spatial-temporal analysis in environmental research. Information on the spatial distribution and spatial-temporal dynamics of natural landscapes recorded by series of images, however, usually bears various kinds of uncertainties. This thesis proposes a random set method to deepen our insight into the uncertainties that are inherent in these observations of natural phenomena from images. The general objective of this research is to develop different techniques based on random sets to represent image objects with indeterminate boundaries, quantify their extensional uncertainties, and address uncertainty modeling in a spatial temporal change analysis. The methods are applied to classifying wetland vegetation and monitoring wetland inundation in the Poyang Lake area in China.

Main concepts and statistical characteristics of random sets are introduced in the context of geoinformation science in Chapter 2. They characterize the spatial distribution of a random set, including covering function, p-level sets, support set, core set, median set, mean set and variance. Chapter 3 represents extensional uncertainties of extracted image objects by random sets and quantifies their degree of extensional uncertainty by derived indices. The number of iterations to achieve a stable covering function and the sum of the variances are good indicators of boundary sharpness. The coefficient of variation has a positive relation with the degree of uncertainty. An asymmetry ratio reflects the uneven gradual changes along different directions where broad boundaries exist. Results show that several characteristics of extensional uncertainty of segmented objects can be quantified numerically and spatially by random sets. Chapter 4 addresses the accuracy assessment of segmented uncertain objects modeled by random sets. Results show that significant correlations exist between a covering function of random set model and actual vegetation coverage, and that their probability distributions are similar. The accuracy of random set varies from good to moderate according to kappa coefficients derived from the confusion matrixes of support, mean, median, and core sets.

Three objects on the NDVI image derived from Landsat TM (left); Contours of the core sets, mean sets and support sets of the three random regions

Chapter 5 develops a mixed Gaussian model with three components to serve as a general method to parameterize random sets in a multi-temporal analysis. Transition zones between wetland vegetation and open waters are identified and random set related indices are adopted to show spatial uncertainties of inundated areas. Results reveal that random sets provide detailed spatial configurations of the wetland water regimes as presented by water covering days (WCD). Random sets contribute to WCD mapping by means of smoothing the WCD variation captured by limited snapshots and by providing more details than when using a crisp method.

Four maps of random sets for illustrating the spatial pattern of wetland extents in different seasons

In Chapter 6, a random spread process model (RSP) is constructed for monitoring the spatial-temporal pattern of wetland inundation, in which random sets are used to model the spatial extent of each stage. Periodicity, trend and random components are captured by monthly and yearly random sets. The Cov-Dist matrix and related operators are used to summarize the spatial pattern and quantify the similarity of different stages in the process. Results improve our understandings of the substantial seasonal dynamic pattern of the lake and reveal a subtle interannual change trend in the monitoring period. Therefore, the random spread process serves as a valuable addition to include uncertainty in wetland inundation modeling.

In conclusion, this research shows that random sets provide a general framework for describing uncertainties of natural landscape extracted from remote sensing images. It enriches spatial and spatio-temporal modeling of phenomena which are uncertain in space and dynamic in time.

Biography

Xi Zhao was born on the 29th of April in Hubei province, China. From 2001 to 2005, she studied in the School of Resource and Environmental Sciences, Wuhan University, China, and obtained Bachelor degree on Geographical Information System there. From 2005 to 2007, she attended a joint master's program between Wuhan University and the International Institute for Geo-Information Science and Earth Observation (ITC) in the Netherlands. She received her M.Sc. degree with distinction in geo-information science and earth observation, with a specialization in environmental systems analysis and management. From 2007 onwards, she has undertaken the doctoral study with a joint PhD scholarship offered by ITC, the Netherlands, and the State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, China. Her research interest is in uncertainty modeling, image mining and spatial-temporal change detection.

Zhao, X., Stein, A. (Promotor) and Chen, X. (assistant promotor) Random sets to model uncertainty in remotely sensed objects. Enschede, University of Twente Faculty of Geo-Information and Earth Observation ITC, 2012. (ITC Dissertation 203), ISBN: 978-90-6164-328-9.

   Full text

previous page
more events