derived impervious surface estimates, high-resolution planimetric GIS data, vector-to-raster con-
version methods, and raster GIS overlay methods to derive a level of agreement between the subpixel
classified estimates and the planimetric truth in the Dead Run watershed, a small (14-km
watershed in the Mid-Atlantic physiographic region. From the planimetric data we produced a per-
pixel reference data estimate of impervious surface percentage as a means for assessing the accuracy
of preliminary subpixel estimates of impervious surface cover derived from TM imagery. The spatial
technique allows for multiple accuracy assessment approaches. Results indicated that even though
per-pixel-based estimates of the accuracy of the subpixel data were poor (28.4%, Kappa = 0.19),
the accuracy of the impervious surface percentage estimated using whole-area and rank correlation
approaches was much improved (70.9%, Spearman correlation = 0.608). Our findings suggest that
per-pixel-based approaches to the accuracy assessment of subpixel classified data need to be
approached with some caution. Per-pixel-based approaches may underestimate the actual whole-
area accuracy of the MOI map, as derived from subpixel methods, when applied over large
geographic areas. The raster overlay technique easily extracted the data necessary to derive these
assessments. Although the ArcView Avenue script used here was primarily suited for the assessment
of data at the subpixel level, it can be utilized to derive the accuracy of any classified data set in
which higher-resolution digital truth data are available.
We gratefully acknowledge the United States Geological Survey and the Mid-Atlantic RESAC
for providing the preliminary subpixel derived impervious surfaces estimates and Baltimore County,
Maryland for providing the planimetric GIS data. The U.S. Environmental Protection Agency (EPA),
through its Office of Research and Development (ORD), funded and performed the research
described. This manuscript has been subjected to the EPA's peer and programmatic review and has
been approved for publication. Mention of any trade names or commercial products does not
constitute endorsement or recommendation for use.
Civco, D.L. and J.D. Hurd, Impervious Surface Mapping for the State of Connecticut, in Proceedings of the
American Society of Photogrammetry and Remote Sensing (ASPRS) 1997, American Society of
Photogrammetry and Remote Sensing, Bethesda, MD, 1997.
Congalton, R.G. and K. Green,
Assessing the Accuracy of Remotely Sensed Data: Principles and Practices
Lewis, Boca Raton, FL, 1999.
Huguenin, R.L., Subpixel analysis process improves accuracy of multispectral classifications,
, 3, 37-40, 1994.
Huguenin, R.L., M.A. Karaska, D.V. Blaricom, and J.R. Jensen, Classification of bald cypress and tupelo gum
trees in Thematic Mapper imagery,
63, 717-727, 1997.
Ichoku, C. and A. Karnieli, A review of mixture modeling techniques for sub-pixel land cover estimation,
Photogam. Eng. Remote Sens.,
Remote Sens. Rev.
, 13, 161-186, 1996.
Introductory Digital Image Processing: A Remote Sensing Perspective
, Prentice-Hall, Englewood
Cliffs, NJ, 1986.
Ji, M. and J.R. Jensen, Effectiveness of subpixel analysis in detecting and quantifying urban imperviousness
from Landsat Thematic Mapper,
14, 31-39, 1999.
Mid-Atlantic RESAC, Impervious Surface Mapping Using Multi-Resolution Imagery, available at