Attribution of forest disturbance types based on the Dynamic World class probability data: A case study of Myanmar

• A method is developed based on probability data from Dynamic World. • Forest disturbance types can be rapidly attributed at various spatial extents. • Overall accuracy was 93.3% and most disturbance classes achieved good accuracy. • A Google Earth Engine app of the developed method has been published. Attribution of forest disturbance types using satellite remote sensing is practicable and several methods have been developed to automate the procedure. However, limited by commonly used data and the methodology, achieving accurate and rapid attribution of forest disturbance types over broad spatial extents remains challenging. In this study, we developed a method for attributing forest disturbance types using Dynamic World class probability data (i.e., probabilities for Dynamic World land use land cover types). Specifically, we first obtained a high-quality probability time series by pre-processing the class probability data. Then, we segmented the entire time series into several subseries and classified them according to the hypothetical trajectories. Finally, we completed the attribution of forest disturbance types using the variables derived from the probability time series and the results of the subseries classification. We used the developed method to investigate the forest disturbance types in Myanmar from 2017 to 2023 and validated its effectiveness by conducting unbiased accuracy assessment. The overall accuracy of the type for the acquired map was approximately 93.3%, and the overall accuracy of the year was approximately 96.7%, proving that the method is feasible. This method is based on the Google Earth Engine, which allows users to attribute forest disturbance types in different areas rapidly by simple parameter adjustments. Even if available classes do not satisfy users’ needs, the method can facilitate more detailed attribution of disturbance types.