ABSTRACT
In recent years, several algorithms have been developed to detect spectral change in pixels over time. However, in order to attribute spectral change into disturbance agents (e.g., logging, fire), reference data are required. This chapter presents a good practice framework for the creation of an example reference dataset that takes advantage of an existing forest inventory plot network. Although several methods have been recommended by the literature for reference data creation, the strength of this method lies in utilizing a sampling framework that is stratified across an entire jurisdiction using a fixed sampling design. It is unbiased and comprehensive over a large area, and statistically robust. The created reference dataset consists of almost 8000 reference pixels over a large area in Victoria, Australia. A number of ancillary datasets and information are used by trained interpreters to attribute disturbance information. This reference dataset is then used in a machine learning environment to produce classified disturbance maps over a 28 year period.
