A Time Series Analysis-Based Forest Disturbance Time Backtracking Method in Northeastern China, the Changbai Mountain

Forest disturbance is a crucial factor in evaluating carbon sink changes in ecological studies. Current research on forest disturbances mainly focuses on forest disturbance detection, forest disturbance classification, and other analyses and statistics of forest changes methods. These studies are mostly based on time-series images or forest remote sensing features to achieve dynamic monitoring of forest changes. However, it is often impossible to construct disturbance information with a lack of images from early remote sensing. This paper addresses this limitation by proposes a method of forest disturbance time backtracking which can get historical information on forest disturbance over a wider temporal distribution. Using this method, we ultimately achieved forest disturbance detection on early remote sensing images from the 1960s. This method is based on machine learning and deep learning algorithms to construct two forest disturbance datasets to achieve this goal. The first dataset, for categorizing forest disturbances, is the first of its kind in Northeast China. This data can be widely used in similar studies. The second dataset is the forest disturbance recovery year dataset. We estimate the timing of early forest disturbance by classifying this dataset. The experimental results showed that the first dataset has an accuracy of over 90%, and the backtracking historical disturbance results are consistent with the actual historical information. This method is significant in its ability to fill in gaps in the forest disturbance classification present in Northeast China and enhance the confidence and time span of the forest disturbance detection method.