Geoscience Reference
In-Depth Information
propagation (e.g., Kyriakidis and Dungan [2001]), thus allowing one to go beyond simple map
accuracy statistics and address map use (and map value) issues.
11.6 SUMMARY
Thematic classification accuracy constitutes a critical factor in the successful application of
remotely sensed products in various disciplines, such as ecology and environmental sciences. Apart
from traditional accuracy statistics based on the confusion matrix, maps of posterior probabilities
of class occurrence are extremely useful for depicting the spatial variation of classification uncer-
tainty. Conventional classification procedures such as Gaussian maximum likelihood, however, do
not account for the plethora of ancillary data that could enhance such a metadata map product.
In this chapter, we propose a geostatistical approach for introducing contextual information
into the mapping of classification uncertainty using information provided only by the training pixels.
Probabilities of class occurrence that account for context information are first estimated via indicator
kriging and are then integrated in a Bayesian framework with probabilities for class occurrence
based on conventional classifiers, thus yielding improved maps of thematic classification uncer-
tainty. A case study based on realistically simulated TM imagery illustrates the applicability of the
proposed method: (1) regional accuracy scores indicate relative improvements over traditional
classification algorithms in the order of 10% for overall accuracy and 34% for the Kappa coefficient
and (2) maps of pixel-specific accuracy values tend to pinpoint class boundaries as the most
uncertain regions, thus appearing as a promising means for guiding additional sampling campaigns.
REFERENCES
Atkinson, P.M. and P. Lewis, Geostatistical classification for remote sensing: an introduction,
Comput. Geosci.
,
26, 361-371, 2000.
Benediktsson, J.A. and P.H. Swain, Consensus theoretic classification methods,
IEEE Trans. Syst. Man
Cybernet.
, 22, 688-704, 1992.
Benediktsson, J.A., P.H. Swain, and O.K. Ersoy, Neural network approaches versus statistical methods in
classification of multisource remote sensing data,
IEEE Trans. Geosci. Remote Sens.
, 28, 540-552,
1990.
Bonham-Carter, G.F.,
Geographic Information Systems for Geoscientists
, Pergamon, Ontario, 1994.
Congalton, R.G., A review of assessing the accuracy of classifications of remotely sensed data,
Remote Sens.
Environ.
, 37, 35-46, 1991.
Congalton, R.G., Using spatial autocorrelation analysis to explore the errors in maps generated from remotely
sensed data,
Photogram. Eng. Remote Sens.
, 54, 587-592, 1988.
Congalton, R.G. and K. Green,
Assessing the Accuracy of Remote Sensed Data: Principles and Practices
,
Lewis, Boca Raton, FL, 1999.
Cressie, N.A.C.,
Statistics for Spatial Data
, John Wiley & Sons, New York, 1993.
De Bruin, S., Predicting the areal extent of land-cover types using classified imagery and geostatistics,
Remote
Sens. Environ.
, 74, 387-396, 2000.
Deutsch, C.V., Cleaning categorical variable (lithofacies) realizations with maximum a-posteriori selection,
Comput. Geosci.
, 24, 551-562, 1998.
Deutsch, C.V. and A.G. Journel,
GSLIB: Geostatistical Software Library and User's Guide
, 2nd ed., Oxford
University Press, New York, 1998.
Foody, G.M., Status of land-cover classification accuracy assessment,
Remote Sens. Environ.
, 80, 185-201,
2002.
Foody, G.M., N.A. Campbell, N.M. Trood, and T.F. Wood, Derivation and applications of probabilistic
measures of class membership from the maximum-likelihood classifier,
Photogram. Eng. Remote
Sens.
, 58, 1335-1341, 1992.
