Spectral variability related to forest inventory polygons stored within a GIS 1

The union of geographical information systems (GIS) with remotely sensed data provides opportunities to update GIS attributes. The ability to update GIS polygons with regional scale remotely sensed data, such as Landsat TM, is limited by the distribution of variance of the spectral content both within and between the GIS polygons. In this study, we address the distribution of spectral values in relation to forest inventory polygon information to benchmark the strength of the relationships between forest inventory information and image spectral response. Following the investigation of the distribution of variance we undertake to predict polygon labels from co-registered Landsat TM data. Management units selected to represent softwood and hardwoods are identified correctly in approximately 64% and 72% of the possible cases respectively; whereas, the management unit selected to represent mixed woods was identified correctly in 15% of possible instances. When considering the results within the context of a polygon, filtering is possible to indicate the most likely class. Both the investigation of the polygon variance structure and the polygon classification indicate that given a sub-optimal variance structure a cohort classification approach is preferred over a per pixel classification.