Geoscience Reference
In-Depth Information
15
CHAPTER
The Effects of Classification Accuracy
on Landscape Indices
Guofan Shao and Wenchun Wu
CONTENTS
15.1 Introduction...........................................................................................................................209
15.2 Methods ................................................................................................................................210
15.2.1 Relative Errors of Area (REA).................................................................................211
15.3 Results...................................................................................................................................213
15.4 Discussion.............................................................................................................................214
15.5 Conclusions...........................................................................................................................217
15.6 Summary...............................................................................................................................219
Acknowledgments ..........................................................................................................................219
References ......................................................................................................................................219
15.1 INTRODUCTION
Remote sensing technology has advanced markedly during the past decades. Accordingly,
remote sensor data formats have evolved from image (pre-1970s) to digital formats subsequent to
the launch of Landsat (1972), resulting in a proliferation of derivative map products. The accuracy
of these products has become an integral analysis step essential to evaluate appropriate applications
(Congalton and Green, 1999). During the past three decades, accuracy assessment has become
widely applied and accepted. Although methodologies have improved, little attention has been
given to the effects of classification accuracy on the development of landscape metrics or indices.
Thematic maps derived from image classification are not always the final product from the
user's perspective (Stehman and Czaplewski, 1998). Because all image processing or classification
inevitably introduces errors into the resultant thematic maps, any subsequent quantitative analyses
will reflect these errors (Lunetta et al., 1991). Landscape metrics are commonly derived from remote
sensing-derived LC maps (O'Neill et al., 1988; McGarigal and Marks, 1994; Frohn, 1998). Metrics
are commonly used to compare landscape configurations through time or across space, or as
independent variables in modeling linking spatial pattern and process (Gustafson, 1998). Therefore,
conclusions drawn directly or indirectly from analyzing landscape metrics contain uncertainties.
The relationships between the accuracy of LC maps and specific derived landscape metrics are
209
