Skip to main content
SLU publication database (SLUpub)

Abstract

Questions related to environmental health call for a new generation of statistical tools. We describe some statistical methods that are applicable under nonstandard conditions that frequently arise in environmental problems and in remote sensing applications. Error rates of traditional remote sensing classification methods are usually quite high but can be improved by (1) denoising the images using wavelet transforms, (2) using the Gibbs sampler for reclassification, and (3) applying new classification algorithms such as probabilistic classifiers. These classifiers can be used for quality assessments at the pixel level. Resampling of blocks is used for accuracy assessments. An example is determining the total volume of dead wood over some region. Estimates are obtained by summing the predicted volumes over all pixels. A nonparametric method that estimates the variance and the distribution for estimators of this kind has been developed; the approach allows for spatial dependence and for nonstationarity. Consistent variance estimators for totals and functions of totals, e.g. ratio-estimators, together with convergence rates are provided

Keywords

Classification; nearest neighbor methods; Gibbs sampler; probabilistic classifiers; quality assessment ; resampling ; wavelet transforms; remote sensing

Published in

Publisher: JSM

Conference

Joint Statistical Meetings

SLU Authors

UKÄ Subject classification

Probability Theory and Statistics

Permanent link to this page (URI)

https://res.slu.se/id/publ/4975