Biomass carbon plays a crucial role in the global carbon cycle and carbon sinks in terrestrial ecosystems. The Tibetan Plateau is an amplifier of global climate change, and the spatial distribution of biomass in the region responds with high sensitivity to climate change. In addition to climate change, human activities interventions also greatly affect the carbon cycle and carbon balance in the alpine ecosystem of the Tibetan Plateau. In this study, biomass carbon density and spatial distribution on the Tibetan Plateau from 2001 to 2020 was conducted using three machine-learning (ML) algorithms, including random forest (RF), support vector machines (SVM) and extreme gradient boosting (XGBoost). Multi-source environmental variables were comprehensively considered, including topographic factors, climatic factors, biological factors, soil properties and socioeconomic factors. The results showed that: (1) The average aboveground biomass carbon density was 3.864 kg C/m2, and the average belowground biomass carbon density was 0.918 kg C/m2. (2) Among the three ML algorithms, the validation accuracy of RF was the highest, and its accuracy for AGB were higher (R2 = 0.519, RMSE = 1.283 kg C/m2) than BGB (R2 = 0.378, RMSE = 0.705 kg C/m2). (3) Soil pH, precipitation, and livestock volume were important indicators for the prediction of AGB, with importance values of 18.93%, 13.50%, and 7.18%, respectively; latitude, precipitation, and soil organic carbon were important indicators for the prediction of BGB, with importance values of 7.23%, 7.09%, and 6.11%, respectively. (4) Spatially, the distribution pattern of AGB and BGB of Tibetan Plateau was consistent, and the high value was located in the east and low in the west, which was consistent with the distribution pattern of the sampling sites data. In a word, accurate accounting and spatiotemporal dynamics of biomass carbon density can provide important theoretical basis for carbon sink assessment.