Abstract: We developed a method of comprehensive forest disturbance identification system based on the distur-bance characteristics of forest ecosystem and the integrated multi-source remote sensing data to evaluate the overall forest disturbance intensity of Heilongjiang Province from 2001 to 2023. We further conducted disturbance extraction and types identification. The results showed that forest disturbance intensity peaked in 2003, primarily due to large-scale forest fires. The spatial consistency between disturbance detection using the LandTrendr method and the Global Forest Change dataset exceeded 90%. Forest disturbances could be categorized into three types, including fire disturbance, pest disturbance, and logging disturbance. The overall classification accuracy for disturbance types was 87.8% (Kappa coefficient=0.81). Different spectral indices had different responses to disturbance types. Specifically, the normalized burn ratio was the most sensitive to fire disturbance. The normalized difference vegetation index was more responsive to overall vegetation change. The normalized difference moisture index made a more significant contribution to the identification of pest disease, while the modified greenness difference index could assist in detecting logging activities. In conclusion, the integrative analysis of multi-spectral indices and the fusion of temporal features could effectively improve the accuracy of identifying forest disturbance types, which would provide a scientific basis for forest ecosystem management in cold temperate zone of Northeast Asia.

Key words: forest disturbance, remote sensing monitoring, LandTrendr, time series analysis, machine learning