The Three River Head-water Region (TRHR) is located in the hinterland of the Tibetan Plateau, is the source of the Yangtze River, Yellow River and Lancang River. As an important water source and ecological function conservation area of China, accurate monitoring of the spatio-temporal variation in the grassland aboveground biomass (AGB) is important in the TRHR. In this study, based on field observation, remote sensing, meteorological and topographical data, we estimated the grassland AGB in the Three River Source National Park (TRSNP) and analyzed its spatiotemporal change and response to climatic factors. Four machine learning (ML) models (random forest (RF), cubist, artificial neural network (ANN) and support vector regression (SVR) models) were constructed and compared for AGB simulation. The AGB results estimated with the four ML models were then applied in the integrated analysis via Bayesian model averaging (BMA) to obtain more accurate and stable estimates (r=0.88; RMSE=71.60g/m²). The results showed that the spatial distribution of grassland AGB in the TRSNP had obviously spatial heterogeneity, showing a decreasing trend from southeast to northwest. The grassland AGB in the Yangtze River Source National Park, Yellow River Source National Park and Lancang River National Park were 82.96g/m², 117.54g/m², and 168.39g/m², respectively. In the Yellow River and Yangtze River Source Parks, from 2000 to 2018 grassland AGB showed a non-significant increasing trend due to the influence of temperature increase; in the Lancang River region, the interannual dynamics statistics showed a non-significant downward trend due to the low annual precipitation in 2015 and 2016.