A data-driven estimate of litterfall and forest carbon turnover and the drivers of their inter-annual variabilities in forest ecosystems across China

Strong influences of climate and land-cover changes on terrestrial ecosystems urgently need to re-estimate forest carbon turnover time (tau(forest)), i.e., the residence time of carbon (C) in the living forest carbon reservoir in China, to reduce uncertainties in ecosystem carbon sinks under ongoing climate change. However, in absence of accurate carbon loss (e.g., forest litterfall), tau(forest) estimate based on the non-steady-state assumption (NSSA) in forest ecosystems across China is still unclear. In this study, thus, we first compiled a litterfall dataset with 1025 field observations, and applied a Random Forest (RF) algorithm with the linkage of gridded environmental variables to predict litterfall from 2000 to 2019 with a fine spatial resolution of 1 km and a temporal resolution of one year. Finally, tau(forest) forest was also estimated with the data-driven litterfall product. Results showed that RF algorithm could well predict the spatial and temporal patterns of forest litterfall with a model efficiency of 0.58 and root mean square error of 78.7 g C m(-2) year(-1). Mean litterfall was 205.4 +/- 1.1 Tg C year(-1) (mean +/- standard error) with a significant increasing trend of 0.65 +/- 0.14 Tg C year(-2) from 2000 to 2019 (p < 0.01), indicating an increasing carbon loss from litterfall. Mean tau(forest) was 26.2 +/- 0.1 years with a significant decreasing trend of -0.11 +/- 0.02 years (p < 0.01) from 2000 to 2019. Climate change dominated the inter-annual variability of tau(forest) in high latitude areas, and land-cover change dominated the regions with intensive human activities. These findings suggested that carbon loss from vegetation to the atmosphere becomes more quickly in recent decades, with significant implication for vegetation carbon cycling-climate feedbacks. Meanwhile, the developed litterfall and tau(forest) datasets can serve as a benchmark for biogeochemical models to accurately estimate global carbon cycling.