Landsat temporal discrimination of forest cover types of the Carmanah Valley, British Columbia

• Main Author: Yatich, Sammy Kibet
• Language: English
• Published: 2009
• Online Access: http://hdl.handle.net/2429/5842

The classification of forest cover types in the Pacific Coastal Rainforest is a difficult task. The complexity and variability of species compositions and sites makes various cover types arduous to define and identify. This study utilised multitemporal TM satellite data and a combined classification approach to determine if it is possible to discriminate forest cover types of the Carmanah Valley. Landsat TM data from June and September were used to investigate whether seasonal variability can aid in the discrimination. Due to the huge dataset involved, principal component analysis (PCA) was applied to reduce data dimensionality to a level that could be easily handled or processed with available image analysis software and enhance the meaning. Results from PCA were used in visual analysis and supervised classification. Classification accuracies for three temporal data sets, June-early summer, September-late summer and a multitemporal set, were compared. Finally, TM data were assessed for their potential to provide information regarding forest age and stocking classes. Results indicated that the degree of correlation between any two bands was related to the amount of spectral contrast. The higher the correlation, the less the spectral contrast and the lower the correlation, the more the spectral contrast. A trend was observed between stand age and digital values. With an increase in stand age, digital values decreased in all TM bands considered. However, the main funding was that the specific forest cover types of the Carmanah Valley can be identified and that time of the year can significantly affect cove-type classification accuracy. Comparison of the classifications of forest cover type showed that the multitempoal approach was significantly better than the single-date classifications.