Evaluating scale to achieve optimal image classification accuracy in New Hampshire forests

New England forest complexity creates obstacles for land cover classification using satellite imagery. New methodologies such as object-oriented image analysis exhibit potential to improve classification. Although these methods have proven more accurate than traditional methods, it has been unclear what resolution yields the most accurate classification. As high resolution imagery increases classification difficulty and lower resolutions may not provide sufficiently detailed maps, this study explored the use of object-oriented classification to classify several resolutions of satellite imagery (Landsat TM, SPOT, IKONOS) at various spatial scales. Although Landsat TM imagery yielded the highest accuracy, all classification results were unacceptable for practical use. While classification was inaccurate, segmentation successfully delineated forest stands. A comparison of 1-foot resolution aerial photography and 4-meter resolution IKONOS imagery demonstrated little agreement between segmentation of individual tree canopies. This study indicates that finer resolution imagery is needed for segmentation and classification of individual trees