中文摘要：

在大气多路径和噪声的条件下，高斯白噪声会造成滑动频谱方法获得的弯曲角与真值之间的偏离，无法获得较好的反演结果。为此，文章提出了一种改进的滑动频谱方法，即利用信号的振幅和谱能量信息对滑动频谱方法进行修正，削弱了信号中噪声的影响，与真值较为接近。分别用改进前后两种方法对中华卫星三号计划（ConstellationObservingSystemforMeteorology，IonosphereandClimate，COSMIC）的掩星进行反演，并将其折射率计算结果和通过全谱反演方法获得的折射率一起，与欧洲中期天气预报中心（EuropeanCentreforMediumRangeWeatherForecasts，ECMWF）的分析场资料进行了统计比较。结果表明：改进的滑动频谱方法删除了信号中的部分噪声，减少了系统偏差；与全谱反演方法进行比较，发现两者具有相当的反演精度。

英文摘要：

The limitations of the sliding spectral （SS） method are the restriction of the resolu- tion of bending angle profiles due to the noise in the moist lower troposphere. It has been validated by simulations of atmospheric propagation using multiple phase screens technique under multipath conditions. A modified sliding spectral （MSS） method was developed to reduce these limitations. It demonstrates that MSS method can reasonably identify and decrease noise which may introduce re- trieval errors in classical SS method, as well as further improve the resolution in the moist tropos- pheric regions. About 4500 COSMIC （Constellation Observing System for Meteorology, Ionosphere and Climate） were analyzed using the SS method and MSS method respectively. Statistical compari- sons of the retrieval refractivity profiles with that from ECMWF （European Centre for Medium- range Weather Forecasts） analysis show that MSS method contains smaller systematic biases than SS method. Biases of the MSS method are close to that of full spectrum inversion （FSI） method at all latitudes.