Search

Seven Landsat images near Prince George, British Columbia, Canada, representing a range of within-year and between-year dates, were acquired to assess spectral variability and the concomitant impact upon hyperclustering classification results. Top-of-atmosphere (TOA) radiometric corrections, dark target subtractions and geometric corrections were applied to the imagery. Following application of an unsupervised hyperclustering procedure which employed the K-means classifier, post-classification comparisons examined the differences in spectral response patterns for several target classes, and area summaries were generated to compare the variability in the total area of classes as identified in each image. Finally, the kappa coefficient of agreement was used to quantify the degree of correspondence between the classified images. The results indicated that the spectral variability of the within-year image set exceeded the variability in the between-year image set and differences in class area were highly variable over the range of image acquisition dates. These findings suggest that off-year imagery (acquired on or near anniversary dates) may be preferred to off-season imagery when building large-area Landsat mosaics for land cover classification using the hyperclustering procedure. [ABSTRACT FROM AUTHOR]

The identification and classification of mountain pine beetle, Dentroctonus ponderosa (Hopkins), red-attack damage patterns in a mature lodgepole pine (Pinus contorta) forest located in the Fort St. James Forest District, British Columbia, was accomplished using 1999 Landsat TM satellite imagery, 1999 mountain pine beetle field and aerial survey point data, and G15 forest inventory data. Unrelated variance in the observed spectral response at mountain pine beetle field and aerial survey points was reduced following image stratification with the GIS forest inventory data and removal of other factors uncharacteristic of red-attack damage. Locations of known mountain pine beetle infestation were used to train a maximum-likelihood algorithm; overall classification accuracy was 73 percent, based on an assessment of 360 independent validation points. If local stand variability is reduced prior to signature generation, accuracies and map products can be useful for those involved in active forest management decisionmaking regarding mountain pine beetle infestations. [ABSTRACT FROM AUTHOR]