On Forest Inventory and Remote Sensing

Karlsruhe

1991

S. 72-82

11 Lit. Ang.

Bandaufführung

13282

76606

Most resource studies which utilize remote sensing rely at least partly on aerial photointerpretation to determine the categories for labeling objects within the study area. The categories produced by photointerpretation are considered ground truth" and are also refereed to as the reference data. Computer classifications, or maps, are judged to be accurate if they predict the correct category for the reference data. This method is widely used throughout the world, yet it has an underlying assumption that is rarely tested. It is assumed that the photointerpreted categories are absolutely correct. In this paper I examine this assumption in the context of forest inventory in the mixed conifer forests of California. If the accuracy of photointerpretation is not sufficiently high then the traditional practice of comparing satellite classification to photointerpretation is not justified. If this hypothesis is true, it is speculated that spectral analysis of advanced digital satellite data SPOT and TM) can be used in conjunction with ancillary ground data to produce forest classifications of the same or better accuracy than by traditional photointerpretation techniques. This latter point will be addressed in subsequent papers.