文献类型： 外文期刊

作者： Yin, Xiong ¹ ; Bai, Mengqi ¹ ; Chen, Bangqian ² ; Lai, Hongyan ² ; Wang, Guizhen ² ; Kou, Weili ³ ; Chen, Yue ⁴ ; Li, Mingshi ¹ ; Zhang, Xiaowei ⁵ ;

作者机构： 1.Nanjing Forestry Univ, Coinnovat Ctr Sustainable Forestry Southern China, Nanjing 210037, Peoples R China

2.Chinese Acad Trop Agr Sci CATAS, Rubber Res Inst RRI, State Key Lab Incubat Base Cultivat & Physiol Trop, Hainan Danzhou Agroecosystem Natl Observat & Res S, Haikou 571101, Peoples R China

3.Southwest Forestry Univ, Coll Big Data & Intelligence Engn, Kunming 650224, Peoples R China

4.Southwest Forestry Univ, Coll Forestry, Kunming 650224, Peoples R China

5.Zhejiang Forest Resources Monitoring Ctr, Hangzhou 310020, Peoples R China

6.Zhejiang Forestry Survey Planning & Design Co Ltd, Hangzhou 310020, Peoples R China

关键词： Landsat time-series; Disturbance agent; LandTrendr; Random forest

期刊名称：ECOLOGICAL INDICATORS （ 影响因子：7.4； 五年影响因子：7.2 ）

ISSN： 1470-160X

年卷期： 2025 年 178 卷

页码：

收录情况： SCI

摘要： Forests in northeast China are crucial for maintaining ecological functions and providing diverse ecosystem services. However, forests have experienced various disturbances at distinct spatio-temporal scales during succession. In-depth investigation of the types of disturbances in this region and their driving mechanisms is essential for sustainable forest management and biodiversity conservation. To meet these demands, we adopted a two-stage approach to attribute the forest disturbance events occurred in this particular region. Firstly, we detected annual disturbances in northeast China from 1986 to 2023 by integrating the LandTrendr algorithm with six spectral indices derived from Landsat time-series imagery. To avoid confusing forest disturbances with spectral changes in agricultural land, we introduced three-year median composite data of Tasseled Cap Brightness, Greenness, and Wetness centered on 1987 and 2022. These data were combined with variables in disturbance detection and the random forest (RF) model to refine the results of forest disturbance detection. Secondly, based on the spectral, topographic, and shape characteristics of the forest disturbance patches, we further classified forest disturbances into three categories: fire, logging, and other (e.g. insects, drought and road construction), again using the RF algorithm. Independent validation results demonstrated that the refined forest disturbance map had an overall accuracy at 95.09 %, with all F1-scores above 0.9. The overall accuracy for disturbance agents was 85.10 %, with all F1-scores above 0.85. Between 1986 and 2023, the estimated area of forest disturbance was about 2.47 x 105 km2, with substantial regional differences and interannual fluctuations. Further, fire, logging, and other disturbance types were responsible for 35.45 %, 18.66 %, and 45.88 % of the forest disturbance area, respectively, and were mainly concentrated before 2000, with a substantial decrease after 2000. In addition, precipitation and soil moisture and vapor pressure difference were the main drivers of different forest disturbance agents. This analysis provides detailed, spatio-temporal information on forest disturbances and their agents in northeast China, enabling forest change comprehension and future management and biodiversity conservation initiatives.