Recent advances in digital airborne sensors and satellite platforms make spatially accurate, high-resolution multispectral imagery readily available. These advances provide the opportunity for a host of new applications to address and solve old problems. High-resolution imagery is particularly well suited to urban applications. Previous data sources (such as Landsat TM) did not show the spatial detail necessary to provide many urban planning solutions. This paper provides an overview of a project in which one-meter digital imagery was used to produce a map of pervious and impervious surfaces to be used by the city of Scottsdale, Arizona for storm-water runoff estimation. The increased spatial information in onemeter or less resolution imagery strains the usefulness of image classification using traditional supervised and unsupervised spectral classification algorithms. This study assesses the accuracy of three different methods for extracting land-cover/land-use information from high-resolution imagery of urban environments: (1) combined supervised/ unsupervised spectral classification, (2) raster-based spatial modeling, and (3) image segmentation classification using classification tree analysis. A discussion of the results and relative merits of each method is included.
[1]
Leo Breiman,et al.
Classification and Regression Trees
,
1984
.
[2]
C. Tomlin.
Geographic information systems and cartographic modeling
,
1990
.
[3]
Russell G. Congalton,et al.
A review of assessing the accuracy of classifications of remotely sensed data
,
1991
.
[4]
T. Bernhardsen.
Geographic Information Systems: An Introduction
,
1999
.
[5]
The HEC Hydrologic Modeling System.
,
1995
.
[6]
K. M. S. Sharma,et al.
A modified contextual classification technique for remote sensing data
,
1998
.
[7]
Russell G. Congalton,et al.
Assessing the accuracy of remotely sensed data : principles and practices
,
1998
.
[8]
Russell G. Congalton,et al.
An Error Matrix Approach to Fuzzy Accuracy Assessment: The NIMA Geocover Project
,
2004
.