Grass yield is the basis of livestock carrying capacity, which is of great significance to livestock production. In this paper, based on the hyperspectral remote sensing satellite (GF-5B) images, combined with the grass yield data collected in the field, a multiple linear stepwise regression (MLR) method was used to establish an inverse grass yield model in the Dangqu basin at the source of Yangtze River, and to achieve the near-real-time large-area assessment of grass yield. The results showed that the grass yield (dry) of alpine meadow in the Dangqu basin varied from 0 to 238.41g/m2, with an average value of 61.26g/m2, and the grass cover rate was about 57%. The high value area of grass yield was mainly distributed in the southeastern Chatan wetland, and this situation may be closely related to the distribution of precipitation and temperature in the Dangqu basin. The research results can improve the rational use of alpine grassland in the source of Yangtze River, provide auxiliary suggestions for the intensive, large-scale, ecological and digital development of animal husbandry, guarantee the sustainable and healthy development of animal husbandry in the source of Yangtze River, and protect the ecological security of the source of Yangtze River.