考虑全时间序列信息的NDVI变分重建方法

A Variational Method for Reconstructing NDVI Time Series Considering Full Time Information

NDVI数据受云覆盖等因素影响,普遍存在信息缺失和噪声污染问题。传统基于滤波或曲线拟合的方法对NDVI时间序列的先验特征考虑不足,难以有效解决时间连续缺失的难题。该文在变分框架下,通过正则化方式对NDVI时间序列的局部平滑性和非局部年际相似性进行刻画,提出一种考虑全时间序列信息的NDVI重建方法;选取Savitzky-Golay(SG)滤波、时间序列谐波分析(HANTS)和Whittaker滤波3种方法作为对比方法,在长江经济带两个区域的MODIS NDVI产品上进行模拟实验和真实实验。结果表明:该文方法在随机缺失和时间连续缺失情况下的MAE、RMSE和CC均更优,且能有效滤除不平滑的原始NDVI时间序列曲线中的噪声,获得平滑曲线,在处理时间连续的云污染方面表现更好。

Cloud-induced disturbances and other contamination are widespread in current NDVI products,which become obstacles for subsequent high-level applications.Although a number of temporal approaches have been developed for filtering or smoothing NDVI time series,they ignore the prior features of NDVI time series and tend to encounter the challenge of reconstructing temporally continuous missing values.This paper proposes a novel method to reconstruct NDVI time-series data via a variational regularization framework.The temporally local smoothness information and inter-annual similarity information of NDVI time-series data are combined to formulate a regularization model.In addition,pixel reliability index data is integrated to the model by imposing a weight on the fidelity term.The proposed method is tested in two regions of the Yangtze River Economic Belt in China based on MODIS NDVI products with a 16-day temporal resolution by comparing it with three widely used temporal methods(SG,HANTS,Whittaker filter).Quantitative experimental results show that the proposed method has obvious advantages in missing data reconstruction.It can obtain better quantitative indicators(lower mean absolute error and root mean square error,and higher correlation coefficient)under different data missing ratios,and can also more accurately reconstruct the missing information of continuous time series.The real experimental results show that the proposed method can effectively filter out the noise in the NDVI time series,obtain smooth results,and have better performance in dealing with temporally continuous missing values.Therefore,the proposed method is very promising for the reconstruction of NDVI time-series data.