University of Twente Student Theses


Uncertainty in lake extent trends related to time and frequency of observation

Ijumulana, Julian (2010) Uncertainty in lake extent trends related to time and frequency of observation.

[img] PDF
Abstract:Increasing availability of large mixed archives of remotely sensed data have motivated their use in time series analysis in geographic phenomena. However, this increase requires appropriate image mining method as well as algorithms in which uncertainty inherent in these datasets is explicitly stated. In this study object oriented image analysis with subsequent fuzzy-rule based classification have been used to extract meaningful information about uncertain lake extent from a series of images acquired at irregular time interval by ASTER and ETM+ sensors between 1999 and 2009. The method involves transforming a discrete image into homogeneous regions that correspond to (part of) real world phenomenon through the process of segmentation. The multiresolution segmentation algorithm was used to generate a network of segments technically known as image objects. The algorithm applies optimization procedure that locally minimizes average heterogeneity of resulting image objects for a given resolution. Fuzzy object-based classification was used in this study in which membership function values were determined based on various features of image objects. Each classified object was attributed with a membership function value expressing its uncertainty. Since the lake extent was determined by the total number individual objects, the weighted average membership was determined to express uncertainty in lake extent at a moment in time. Accuracy assessment of the resulting classes was evaluated by determining the class stability in which the least and best classified image object of that class was retrieved. Lake extent time series was performed using linear regression modelling. Least squares method was used to estimate the best fit line representing the trend in lake extent. A combination of observations from the two sensors was perfomed after evaluating how strongly they are linearly related. The correlation coefficient of 0.71 was obtained revealing that the two sensors can be combined to obtain required number of observations for time series analysis in uncertain lake extent. While analysing time series in lake extent, 97% of variation in the observations was well modelled by the prediction line during rainy season, 90% during dry season and 91% in all seasons. However, trend analysis in lake extent is complex because of wetlands with dominant vegetation whose size also varies with seasons. It is also difficulty due to natural and unpredictable water level fluctuations. With these preliminary results of time series analysis using remotely sensed data from mixed archives at irregular time interval, it is possible to study various geographic phenomena whose speed of change requires large number of observations for their detection. Key words: Lake extent, uncertainty, object-oriented image analysis, multiresolution segmentation, fuzzy object-based classification, time series analysis
Item Type:Essay (Master)
Faculty:ITC: Faculty of Geo-information Science and Earth Observation
Programme:Geoinformation Science and Earth Observation MSc (75014)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page