Time-series reconstruction and mapping of forest aboveground biomass in the Great Xing’an Mountains of China using GEDI, MODIS, and machine learning

Accurate long-term estimation of forest aboveground biomass (AGB) is essential for understanding carbon dynamics and assessing the impacts of climate change and human disturbance. However, generating high-resolution, continuous AGB time series remains challenging due to data limitations and methodological constraints. In this study, we present a 21-year (2000–2020) reconstruction of forest AGB in China’s Great Xing’an Mountains by integrating multi-temporal MODIS imagery with spaceborne LiDAR data from the GEDI L4B product using the AutoGluon stacking ensemble learning algorithm. All models achieved root mean square errors (RMSE) below 25 Mg/ha, with weighted ensemble model yielding superior performance (R 2 = 0.83, RMSE = 13.99 Mg/ha, rRMSE = 14.38 %). Trend analysis based on Sen’s slope and the Mann-Kendall test revealed a significant regional increase in AGB, with 83.36 % of forest area exhibiting upward trends, while 16.64 % showed declines. Fire disturbance emerged as a primary driver of localized AGB loss, particularly in the northern and eastern subregions. From 2000 to 2020, average forest AGB increased by 14.67 Mg/ha, and total biomass rose by 0.53 Pg. These results demonstrate the potential of combining GEDI and MODIS data with machine learning for large-scale, long-term forest biomass monitoring, offering valuable support for carbon accounting, ecological assessment, and forest management in cold-temperate ecosystems.