关键词: 气候变化, 碳循环, 机器学习, 土地利用/覆盖变化（LUCC）, 生态恢复工程

Abstract: Soil respiration (R_S) releases CO₂ through autotrophic and heterotrophic respiration, representing the second largest carbon flux in terrestrial ecosystems after photosynthesis. It plays a pivotal role in global carbon cycling and climate feedback. China’s climate shifted from a warming hiatus (2001–2010) to accelerated warming (2010–2019), coupled with ongoing land use/cover change (LUCC), jointly drives the spatiotemporal dynamics of RS. However, the relative contributions and underlying mechanisms of these factors remain underexplored. In this study, biome-specific machine learning models (R² = 0.69–0.82) were developed to estimate RS at a 1 km spatial resolution across China from 2001 to 2019. Results indicate that the long-term average annual RS across China’s vegetated areas is 4.24 ± 0.02 Pg C year⁻¹. Interannual variability shifted from relative stability during 2001–2010 (-5.58 Tg C year⁻¹; -0.08 g C m⁻² year⁻¹, P = 0.77) to a significant increase (36.29 Tg C year⁻¹; 0.52 g C m⁻² year⁻¹, P < 0.05) during 2010–2019. Climate and LUCC together explained 61.7% of the interannual variability in RS, with moisture as the primary driver (29.6% of variance). Large-scale ecological engineering projects, while effective in enhancing carbon sequestration, also promote R_S, potentially offsetting some carbon storage gains. The long-term time-series dataset obtained in this study not only supports research on the mechanisms influencing R_S but also provides benchmark data for improving terrestrial ecosystem carbon cycle models. These findings highlight R_S’ critical role in China’s carbon budget and its sensitivity to climatic and anthropogenic drivers.

Key words: climate trends, carbon cycle, machine learning, LUCC, ecological restoration