插补GNSS坐标时间序列的改进DINEOF算法

针对GNSS坐标时间序列插补中数据插值经验正交函数（data interpolating empirical orthogonal functions,DINEOF）算法受低相关度站点影响、连续长空缺插值效果欠佳的问题，提出相关数据插值经验正交函数（coefficient data interpolating empirical orthogonal functions,CDINEOF）算法。首先计算目标站点与其周围站点数据的相关度，然后筛选出相关度较高的数据构建观测矩阵，最后采用DINEOF算法对观测矩阵中的缺失数据进行插补。通过模拟数据和实测数据验证新方法的可行性，并与DINEOF算法和多项式插值法的结果进行对比分析。模拟数据实验结果表明，当观测数据存在连续长空缺时，CDINEOF算法的插值效果优于DINEOF算法和多项式插值法。实测数据实验结果表明，CDINEOF算法在保留方差最大化方面效果最好，与DINEOF算法和多项式插值法相比分别提升了11.8%和6.7%。

Improved DINEOF for Interpolating GNSS Coordinate Time Series

To solve the problem that the data interpolating empirical orthogonal functions(DINEOF) in GNSS coordinate time series interpolation are affected by low correlation sites and poor interpolation effect of continuous long vacancies, we propose the coefficent data interpolating empirical orthogonal functions(CDINEOF) method. Firstly, we calculate the correlation between the target site data and its surrounding site data, then filter out the site data with higher correlation to construct the observation matrix, and finally use DINEOF to interpolate the missing data in the observation matrix. Experiments are conducted to verify the feasibility of the new method by simulated and actual data, and the results are compared and analyzed with those of DINEOF method and polynomial interpolation method. The experimental results of simulated data show that the interpolation effect of CDINEOF method is better than that of DINEOF method and polynomial interpolation method when there are continuous long vacancies in the observed data; the experimental results of measured data show that CDINEOF method is the best in maximizing the retained variance, and the maximum variance is improved by 11.8% and 6.7% based on DINEOF method and polynomial interpolation method, respectively.