Toward consistent applications of remotely sensed data

The purpose of the present study was to review, evaluate and explore methodologies in classifying remotely sensed data, for the purpose of realizing consistent applications of remotely sensed data, obtaining accurate land cover maps and providing reliable landscape characterizations. Remote sensing is an essential integrated component of earth sciences; remotely sensed data have become a main information source for various applications in forestry and natural resources. However the biggest bottleneck was found to be the inadequate accuracy of the information derived from the remotely sensed data. The poor match between the information obtained from remote sensing and the application goals was attributed to unrepeatable methodologies for image data classification and limited understanding of error propagation and accumulation through the remote sensing information processing flow. In this present study, the optimal methodology for achieving consistently high image data classification accuracy was tested using different combinations of data, algorithms and training method, and training scheme with aid of the dataset spatial variance derived by geo-statistical analysis. The relationships of thematic map\u27s accuracy and the reliability of further landscape characterization were evaluated, an explicit index, called Relative Errors of Area (REA), was proposed for assessing and correcting the error of the measurement of individual land cover types. These results are discussed in terms of the approaches of consistent applications of remotely sensed data in the field of forestry and landscape ecology. Relatively high classification accuracy can be steadily achieved with the combination of data with rich information, classifiers with both spectral and spatial considerations, and training samples determined with the geostatistics of multispectral bands. REA helps translate classification errors into a practically meaningful measure for assessing and correcting the errors of land areas of individual land cover types