Geoscience Reference
In-Depth Information
11
CHAPTER
Geostatistical Mapping of Thematic
Classification Uncertainty
Phaedon C. Kyriakidis, Xiaohang Liu, and Michael F. Goodchild
CONTENTS
11.1 Introduction...........................................................................................................................145
11.2 Methods ................................................................................................................................147
11.2.1 Classification Based on Remotely Sensed Data ......................................................147
11.2.2 Geostatistical Modeling of Context .........................................................................148
11.2.3 Combining Spectral and Contextual Information....................................................150
11.2.4 Mapping Thematic Classification Accuracy ............................................................152
11.2.5 Generation of Simulated TM Reflectance Values....................................................152
11.3 Results...................................................................................................................................153
11.3.1 Spectral and Spatial Classifications .........................................................................155
11.3.2 Merging Spectral and Contextual Information ........................................................155
11.3.3 Mapping Classification Accuracy.............................................................................158
11.4 Discussion.............................................................................................................................160
11.5 Conclusions...........................................................................................................................160
11.6 Summary...............................................................................................................................161
References ......................................................................................................................................161
11.1 INTRODUCTION
Thematic data derived from remotely sensed imagery lie at the heart of a plethora of environ-
mental models at local, regional, and global scales. Accurate thematic classifications are therefore
becoming increasingly essential for realistic model predictions in many disciplines. Remotely
sensed information and resulting classifications, however, are not error free, but carry the imprint
of a suite of data acquisition, storage, transformation, and representation errors and uncertainties
(Zhang and Goodchild, 2002). The increased interest in characterizing the accuracy of thematic
classification has promoted the practice of computing and reporting a set of different, yet comple-
mentary, accuracy statistics all derived from the confusion matrix (Congalton, 1991; Stehman,
1997; Congalton and Green, 1999; Foody, 2002). Based on these accuracy statistics, users of
145
