<p>Belowground or &#8216;soil&#8217; autotrophic respiration (RAsoil) depends on carbohydrates from photosynthesis flowing to roots and rhizospheres, and is one of the most important but uncertain components in forest carbon cycling. Carbon allocation plays an important role in forest carbon cycling and reflects forest adaptation to changing environmental conditions. However, carbon allocation to RAsoil is rarely measured directly and has not been fully examined at the global scale. To fill this knowledge gap, the spatio-temporal patterns of RAsoil with a spatial resolution of half degree from 1981 to 2017 were predicted by Random Forest (RF) algorithm using the most updated Global Soil Respiration Database (v5) with global environmental variables; carbon allocation from photosynthesis to RAsoil (CAsoil), was calculated as the ratio of RAsoil to gross primary production (GPP); and its temporal and spatial patterns were assessed in global forest ecosystems. We found strong temporal and spatial variabilities of RAsoil with an increasing trend from boreal forests to tropical forests. Globally, mean RAsoil from forests was 8.9 &#177; 0.08 Pg C yr<sup>-1</sup> (mean &#177; standard deviation) from 1981 to 2017 increasing at a rate of 0.0059 Pg C yr<sup>-2</sup>, paralleling broader soil respiration changes and indicating an increasing carbon loss respired by roots. Mean CAsoil was 0.243 &#177; 0.016 and showed a decreasing trend over time, although there were interannual variabilities, indicating that CAsoil was sensitive to environmental changes. The temporal trend of CAsoil varied greatly in space, reflecting uneven responses of CAsoil to environmental changes. The spatio-temporal variability of carbon allocation should be considered in global biogeochemical models to accurately predict belowground carbon cycling in an era of ongoing climate change.&#160;</p>
