Late Seral Forest Classification Using ETM+ Remote Sensing Imagery for the Central Pacific Coastal Forests

To conserve the Earth's most extraordinary expressions of temperate biodiversity in the Pacific Northwest (PNW), USA, the mapping of late seral (old and mature) conifer forests plays a critical role. For this paper, we define old conifer forests as >150 years and mature conifer forest between 50 and 150 years. We offer a new Optimal Iterative Unsupervised Classification (OIUC) procedure for mapping late seral conifer forests over an eight-ecoregion area. The key steps of the OIUC classification were: (1) fully using the Landsat 7 Enhanced Thematic Mapper Plus (ETM+) 15 m panchromatic channel merged with other 30 m bands 3 and 5 to make a pan-sharpened false color composite for high resolution image interpretation; (2) splitting the ETM+ scene by ecologically distinct areas, or ecoregions, to create relatively homologous images for classification; (3) using a procedure similar to cluster busting where multiple iterative manipulation of the ISODATA clusters was employed; and (4) edge matching of sub-scenes to form ecoregions, then later merged together to form a map for the entire PNW. Supporting data and information included ancillary spatial GIS data layers, aerial photos, Digital Ortho Quad images (DOQs), field investigations, and previously reported forest age results. Classification accuracy was assessed using 2081 stratified random locations on 105 individual DOQs covering the entire region. Approximately 4.7 million ha (not, vert, similar19%) of the PNW was classified as old conifer forest (>150 years). Another 4.8 million ha (not, vert, similar19%) was classified as mature conifer forest (50 to150 years). Over 9.48 million ha (not, vert, similar38%) of the PNW was late seral conifer forest. The extent of late seral forests (old and mature conifer cover classes) varied greatly between the eight ecoregions. The Central and Southern Cascades and Klamath and Siskiyou ecoregions contained the highest amount of late seral forest in the region. The results showed high accuracy of the late seral forest classification for the PNW with an overall accuracy of 90.72% and KAPPA test K value 0.8534. Producer's (Omission) accuracy for old and mature forests were 91.36% and 80.40%, User's (Commission) accuracy were 89.42% and 80.59%, respectively. Accuracy levels differed for the different ecoregions examined. In general, mature conifer forests exhibited higher levels of confusion than did old conifer forests, due to the spectral influences of high-density young conifer stands and terrain shadow effects. The results fill an important data gap needed for ongoing conservation planning purposes throughout the region. We found that for relatively large geographic areas the OIUC method is an efficient and cost-effective alternative that yields high quality results.