USING OBJECT-ORIENTED CLASSIFICATION TO MAP FOREST COMMUNITY TYPES

The ability to spatially quantify changes in the la ndscape is one of the most powerful uses of remote sensing. The recent release of all of the Landsat imagery has op ened up doors to many entities not previously able to afford this type of data. A strength of Landsat is its tempora l resolution, so a useful application of the imager y is to create land cover maps through time to quantify land cover chan ge. Recent advances in object-based image analysis (OBIA) have also improved classification techniques for de veloping land cover maps. However, when creating land cover maps with specific land cover types, such as forest types, the collection of reference data becomes ex tremely important. When using an OBIA technique, collecting ground data to classify polygons for use as refer ence data may not be straight forward, since polygons general ly contain a variable number and types of pixels. In previous studies, one sample location within a reference uni t was often used to assess cover type for a referen ce unit, however this study shows that one sample may not be enough to accurately classify a polygon reference unit. This study evaluates how many prism sample locations are needed within a forested reference unit to accurately cl assify that reference unit using dominant tree species. In gen eral, anywhere from 7 to 13 points were necessary, depending on the characteristics of the polygon being identified .