Abstract: Height assignment of semi-transparent clouds needs two variables:Radiation from the semi-transparent clouds themselves and radiation from the background underneath semi-transparent clouds. The background radiation is not directly observed by the satellite, but can be estimated with infrared/water vapor scatter diagrams of satellite measurements around tracer image boxes. According to the cloud amount and type observed, tracer image boxes can be divided into 4 types:With different levels of clouds inside, near the cloud edge, with only high or middle-high clouds inside and with only middle-low clouds or ground in tracer box. By studying characteristics of infrared/water vapor scatter diagrams of satellite measurements in 8 typical cases of 4 types which are variety, different estimation methods are discussed and a new estimation method for only high or middle-high clouds in tracer box is proposed. When there are cloud-free pixels in tracer boxes, infrared maximum brightness temperature may represent the background radiation for infrared channel; on the other hands, observation pixels with the highest infrared but relative lower water vapor brightness temperatures are used to estimate the background radiation for water vapor channel. Observation pixels with high infrared, but relative lower water vapor brightness temperatures are located in the clear sky region near the cloud edge, where atmospheric stratification conditions are relatively wetter than the region far away from clouds. The clear sky region near the cloud edge relatively better represent the condition underneath clouds better. Due to absorption and retransition processes, the water vapor brightness temperatures with wetter atmospheric stratification conditions in the clear sky region near the cloud edge are relative lower. When there are no cloud-free pixels in the tracer image box, the searching area should be expanded in the east-west direction to find cloud free pixels, until the background radiation can be estimated.Based on the new estimation method, FY-2 wind vector height assignment algorithm is improved, and effectiveness is analyzed by comparing with ECMWF reanalysis data, showing that FY-2 wind vector quality including biases and standard deviations is improved obviously.