Grassland biomass is a basic index of ecosystem function and production efficiency of alpine meadow ecosystem. However, during non-growing season of the half of the year in winter takes a gap of grass observations period on Qinghai-Tibet Plateau, the work of grass observation which is completely suspended. This study based on monthly synchronous field observation on grass parameters and hyperspectral in alpine grassland of Haibei, Qinghai, from August 2020 to April 2021, observations of biomass, apparent state and spectral characteristics of grasses in different months and at different states of attenuation, and analysis of their dynamic processes. The results showed that wilted grass biomass in winter is generally in two stages:rapid decline and relative stability, a rapid decline period from August to October with the biomass was decreased sharply from 9225 kg/hm² in August to 3536 kg/hm² in October, there is a distinct decrement of nearly 160% compare to the biomass in August, and then, a gently and less variation period performed on the time of November to April of the next year. A Revised Dead Grass Vegetation Index (R-DGVI) was proposed based on the relationship between ground observed grass biomass and reflectance spectrum, which showed a good ability for wilted grass identification especially in area of lower or higher vegetation cover, compare to Normalized Difference Vegetation Index, it represented the stronger responsive and wider threshold range on wilted grass biomass monitoring. Furthermore, a remote sensing estimation model for winter wilted grass biomass using MODIS satellite data was established with R² reached at 0.5627 (P<0.01), then, a five-level classified system related to R-DGVI value at different grades grassland was provided, which was suitable for remote sensing monitoring on wilted grass biomass in winter. The results will provide a further information to grass change pattern in winter and some basic methods and technical for biomass monitoring.