This study analyzes land-cover types in the Gee Creek Watershed of southern Washington using the pixel-based and object-based image analysis approaches. Landsat imagery has traditionally been used for pixel-based classification and change detection in land-cover studies. In recent years, the availability of high-resolution satellite and aerial imagery have enabled for land-cover classification to occur at scales not possible using traditional Landsat imagery. High-resolution aerial imagery of 1

more »

... meter or greater has become readily available for free. Yet, commonly found black and white ( or panchromatic) aerial imagery is without the multiple spectrum bands found in Landsat imagery, thereby limiting the accuracy of traditional pixel-based multispectral classification approaches. Instead, object-based image classification can be used as an alternative analysis approach for determining land-cover types on high-resolution imageries. This paper examines and compares two traditional Landsat pixel-based techniques with the high-resolution object-based approach. Both approaches are used to conduct land-cover classification within the highly variable landscape of the Gee Creek Watershed. The high variability found within the Watershed is the result of recent years of development that have changed the landscape from predominantly forest and agriculture to one of the fastest growing suburbia's outside the Portland-Vancouver metropolitan area. Two pixel-based classification analyses are conducted using Landsat imagery; supervised classification of multispectral bands and unsupervised classification of transformed Tasseled Cap bands. These traditional approaches are then compared to object-based classification using 1 meter resolution natural color aerial imagery obtained from the United States Department of Agriculture. The result of this analysis suggests that Landsat pixel-based approaches are only suitable for determining general land-cover types, whereas the use of object-based classification on high-resolution imagery resulted in increased accuracy and ultimately led to a higher number of land-cover classes being distinguished.