Mapping Annual Forest Disturbance from 1986 to 2021 at 30-M Resolution in China Using the Modified Cold Algorithm

Abstract

Accurate monitoring of forest disturbances, along with the generation of datasets that capture their timing and spatial distributions, are crucial for assessing carbon emissions and formulating effective forest management strategies in pursuing China's carbon neutrality goals. However, previous efforts to monitor forest disturbances in China using existing algorithms have faced several challenges, such as limited focus on specific types of disturbance, incomplete temporal coverage, and reduced accuracy due to the intrinsic constraints of disturbance monitoring algorithms. This study aims to modify the COLD algorithm (mCOLD) to generate a 30-m resolution dataset of China's Annual Forest Disturbance (CAFD) from 1986 to 2021. mCOLD introduces three improvements by incorporating spatial information, confirming disturbances with minimal observations and durations, and bidirectional monitoring techniques, offering a more comprehensive detection of forest disturbance patches than COLD. The CAFD dataset, validated against 6,164 reference samples, achieved an overall accuracy (OA) of 92.28%, with a producer’s accuracy (PA) of 88.09% and a user’s accuracy (UA) of 89.93%, representing an 8.16% improvement over COLD. This dataset indicates a cumulative forest disturbance area of 88.6 million hectares in China from 1986 to 2021, accounting for 38.4% of its total forest area in 2021. Compared to Hansen's forest loss and gain dataset, which exhibited a pronounced imbalance with a high UA of 94.76% but a notably lower PA of 54.29%, CAFD had a longer temporal coverage and achieved a 4.98% improvement in OA and a 34.66% improvement in PA, with a more harmonious balance between UA and PA. This was attributed to CAFD's ability to record multiple disturbance events, a feature lacking in Hansen's dataset. Notably, 23.1% of CAFD's disturbed areas experienced two disturbances, and 13.8% more than two. The mCOLD algorithm holds promise for global forest disturbance monitoring, and the CAFD dataset, accessible at https://doi.org/10.6084/m9.figshare.24464278 (Shang et al., 2025), will enhance our understanding of the multifaceted roles forests play in carbon cycling, climate regulation, air quality, and overall ecosystem health.

Keywords: forest disturbance, spatial integration, bidirectional monitoring, Change detection, Landsat time series