Global variability of belowground autotrophic respiration in terrestrial ecosystems

Abstract. Belowground autotrophic respiration (RA) is one of the largest, but highly uncertain carbon flux components in terrestrial ecosystems. It has not been explored globally before and still acted as a “black box” in global carbon cycling. Such progress and uncertainty motivate a development of global RA dataset and understand its spatial and temporal pattern, causes and responses to future climate change. This study used Random Forest to study RA's spatial and temporal pattern at the global scale by linking the updated field observations from Global Soil Respiration Database (v4) with global grid temperature, precipitation and other environmental variables. Globally, mean RA was 43.8 ± 0.4 Pg C a −1 with a temporally increasing trend of 0.025 ± 0.006 Pg C a −1 over 1980–2012. Such increment trend was widely spread with 58 % global land areas. For each 1 °C increase in annual mean temperature, global RA increased by 0.85 ± 0.13 Pg C a −1 , and it was 0.17 ± 0.03 Pg C a −1 for 10 mm increase in annual mean precipitation, indicating a positive feedback of RA to future climate change. At a global scale, precipitation was the main dominant climatic drivers of the spatial pattern of RA, accounting for 56 % of global land areas with widely spread globally, particularly in dry or semi-arid areas, followed by shortwave radiation (25 %) and temperature (19 %). Different temporal patterns for varying climate zones and biomes indicated uneven response of RA to future climate change, challenging the perspective that the parameters of global carbon stimulation independent on climate zones and biomes. The developed RA database, the missing carbon flux component that is not constrained and validated in terrestrial ecosystem models and earth system models, will provide insights into understanding mechanisms underlying the spatial and temporal variability of belowground carbon dynamics. RA database also has great potentials to serve as a benchmark for future data-model comparisons. The RA product is freely available at https://doi.org/10.6084/m9.figshare.7636193 .