Combining location and classification error sources for estimating multi-temporal database accuracy

Citation
Y. Carmel et al., Combining location and classification error sources for estimating multi-temporal database accuracy, PHOTOGR E R, 67(7), 2001, pp. 865-872
Citations number
40
Language
INGLESE
art.tipo
Article
Categorie Soggetti
Optics & Acoustics
Journal title
PHOTOGRAMMETRIC ENGINEERING AND REMOTE SENSING
ISSN journal
0099-1112 → ACNP
Volume
67
Issue
7
Year of publication
2001
Pages
865 - 872
Database
ISI
SICI code
Abstract
Defection and quantification of temporal change in spatial objects is the s ubject of a growing number of studies. Much of the change shown in such stu dies may be an artifact of location error and classification error. The bas ic units of these two measures are different (distance units for location e rror and pixel counts for classification error). The lack of a single index summarizing both error sources poses a constraint on assessing and interpr eting the apparent change. We present an error model that addresses locatio n and classification error jointly. Our approach quantifies location accura cy in terms of thematic accuracy, using a simulation of the location error process. We further develop an error model that combines the location and c lassification accuracy matrices into a single matrix, representing the over all thematic accuracy in a single layer. The resulting time-specific matric es serve to derive indices for estimating the overall uncertainty in a mult i-temporal dataset. In order to validate the model, we performed simulation s in which known amounts of location and classification error were introduc ed into raster maps. Our error model estimates were highly accurate under a wide range of parameters tested. We applied the error model to a study of vegetation dynamics in California woodlands in order to explore its value f or realistic assessment of change, and its potential to provide a means for quantifying the relative contributions of these two error sources.