Publication Date

1997

Description

Satellite image mapping of grasslands is problematic when species diversity occurs at a sub-pixel scale. We propose a method, called melody classification, to map ground cover units that group several spectral classes (colours). Melodies are defined as the normalized expected frequencies of each class within the ground cover unit. Starting from an unsupervised classification, an image is created showing the probability of finding each spectral class in the vicinity of each pixel. Each pixel is classified by comparing the melody in its neighbourhood with that of each ground cover unit. Accuracies are greatly enhanced over those of supervised classification. Melody classification can be applied to detect and monitor occurrence of particular species or groups of species in rangeland management.

Share

COinS
 

Integrating Satellite Images and Species-Based Vegetation Maps to Manage Native Grasslands

Satellite image mapping of grasslands is problematic when species diversity occurs at a sub-pixel scale. We propose a method, called melody classification, to map ground cover units that group several spectral classes (colours). Melodies are defined as the normalized expected frequencies of each class within the ground cover unit. Starting from an unsupervised classification, an image is created showing the probability of finding each spectral class in the vicinity of each pixel. Each pixel is classified by comparing the melody in its neighbourhood with that of each ground cover unit. Accuracies are greatly enhanced over those of supervised classification. Melody classification can be applied to detect and monitor occurrence of particular species or groups of species in rangeland management.