Vegetation structural composition mapping of a complex landscape using forest cover density transformation and random decision forest classifier: a comparison

Forest cover density (FCD) transformation and random decision forest (RDF) classification have been widely used for vegetation mapping. Nevertheless, a comparison of their capabilities in complex tropical landscapes is still rarely carried out. This study compared the two methods using Landsat-8 OLI imagery which includes the blue up to thermal bands for vegetation structural composition mapping in a complex landscape of Central Java, Indonesia. We used the FCD transformation with six indices to generate 11 classes, while the RDF classified the same 11 classes based on training areas and used a random process involving various number of splits and trees. The results showed that the FCD transformation achieved 69.32% accuracy, while the RDF was able to classify the 11 classes with various accuracies depending on the parameter setting, i.e. from 70.76% to 75.19%. Regarding the obtained accuracies, problems associated with the terrain and vegetation characteristics have been discussed for further recommendation.

Automated summary

This study compared the capabilities of forest cover density (FCD) transformation and random decision forest (RDF) classification in complex tropical landscapes. Landsat-8 OLI imagery was used for vegetation structural composition mapping in Central Java, Indonesia. The FCD transformation achieved an accuracy of 69.32%, while the RDF classification achieved accuracies ranging from 70.76% to 75.19% depending on the parameter setting.