Soil heterotrophic respiration (SHR) is important for carbon-climate feedbacks because of its sensitivity to soil carbon, climatic conditions and nutrient availability. However, available global SHR estimates have either a coarse spatial resolution or rely on simple upscaling formulations. To better quantify the global distribution of SHR and its response to climate variability, we produced a new global SHR data set using Random Forest, up-scaling 455 point data from the Global Soil Respiration Database (SRDB 4.0) with gridded fields of climatic, edaphic and productivity. We estimated a global total SHR of urn:x-wiley:08866236:media:gbc21177:gbc21177-math-0001 Pg C yr−1 over 1985–2013 with a significant increasing trend of 0.03 Pg C yr−2. Among the inputs to generate SHR products, the choice of soil moisture datasets contributes more to the difference among SHR ensemble. Water availability dominates SHR inter-annual variability (IAV) at the global scale; more precisely, temperature strongly controls the SHR IAV in tropical forests, while water availability dominates in extra-tropical forest and semi-arid regions. Our machine-learning SHR ensemble of data-driven gridded estimates and outputs from process-based models (TRENDYv6) shows agreement for a strong association between water variability and SHR IAV at the global scale, but ensemble members exhibit different ecosystem-level SHR IAV controllers. The important role of water availability in driving SHR suggests both a direct effect limiting decomposition and an indirect effect on litter available from productivity. Considering potential uncertainties remaining in our data-driven SHR datasets, we call for more scientifically designed SHR observation network and deep-learning methods making maximum use of observation data.