针对经验模态分解(empirical mode decomposition, EMD)降噪过程中采用相关系数(p)准则确定分界本征模态函数(intrinsic mode function, IMF)分量K值存在不准确性的问题，选用复合评价指标(T)，综合考虑曲线平滑度(r)与均方根误差(RMSE)2个指标值，提出一种改进的EMD降噪方法。利用9个模拟数据和2个陆态网基准站的实测GPS高程时间序列数据进行验证，结果表明，复合评价指标比单一的相关系数指标所确定的分界IMF分量K值更准确，能够更可靠地识别噪声与信号的分界点，使降噪效果更佳。

In the process of empirical mode decomposition(EMD) denoising, there is inaccurate problem when using the correlation coefficient criterion to determine the inaccurate problem of the intrinsic mode function(IMF) K value. This paper adopts the composite evaluation index(T), considering the curve smoothness(r) and root mean square error(RMSE) of two indices, presenting an improved EMD denoising method. We verify the method using 9 simulated data and measured GPS elevation time series data from two reference stations of CMONOC. The experimental results show that the composite evaluation index is more accurate than single correlation coefficient index in determining K value, which can more reliably identify the boundary point between noise and signal, and the noise reduction effect is better.