Research On Medium And High Spatial Resolution NDVI Time Series Reconstruction

NDVI time series data with different spatial resolutions have been widely used in vegetation phenology monitoring and landcover change detection.The spatial resolution of NDVI time series mainly includes medium spatial resolution and high spatial resolution.However,due to cloud,data collecting restriction,limitation of sensors and so on,there are noise and missing data in time series,which may reduce the data quality.For two different spatial resolution NDVI time series,researchers propose many methods to reconstruct time series and improve the quality of time series.At present,the reconstruction methods proposed for two different spatial resolution NDVI time series are divided into two strategies.For the middle spatial resolution NDVI time series,the reconstruction method is mainly to reduce noise and outliers.For the high spatial resolution NDVI time series,the reconstruction method mainly adopts the spatial and temporal fusion to reconstruct time series by filling the missing value and improving the temporal resolution.Existing noise reduction reconstruction methods have problems of overfitting local normal low values and limited ability to deal with continuous missing values.Existing fusion reconstruction methods have weak ability to generate long time series and are sensitive to noise in time series.In this article,in view of the deficiency of two different reconstruction strategies,we develop two new reconstruction methods by using information from neighboring similar pixel and the historical data set.STSG filter is for medium spatial resolution NDVI time series.At the same time GFSTF fusion is for high spatial resolution NDVI time series.STSG employs neighboring pixels to assist in the noise reduction of the target pixel in a particular year,and the relationship between the NDVI of neighboring pixels and that of the target pixel can be obtained from multi-year NDVI time series.We test STSG in some areas of China mainland.Results show that,compared with other filtering methods,such as asymmetric gaussian filtering,double logical model filtering,fourier transform filtering and savitzky-golay filtering,STSG has more advantages and can deal with the continuous missing data.There is no overfitting of true low value in SG.GFSTF use the re-fusion strategy and improve the quality of initial NDVI fusion data by using the relationship in all years.Our tests show that,compared with FSDAF,GFSTF has a lower mean absolute error in different vegetation types and greatly improves the quality of the initial NDVI fusion time series.Meanwhile,although GFSTF employs NDVI data from other years,it is not sensitive to changes of landeover and vegetation greenness between different years.In this paper,we systematically compare and summarize the limitations of medium and high spatial resolution NDVI time series and conclude their different reconstruction strategies.We enrich the reconstruction strategies and give new time series reconstruction technology specifically based on STSG and GFSTF algorithms.