Study On The Reconstruction Of Time-series MODIS-NDVI Dataset

NDVI data set is a vital data source to extract forest disturbance information, vegetation coverage, vegetation phenological pattern, winter fallow fields etc. It’s also of great importance that it provides information on correlation between NDVI and climate change. Due to the influence of satellite sensor, clouds and other atmospheric conditions, there is rudimental noise in NDVI time series data sets. Therefore, the reconstruction method of NDVI time series data has important theoretical significance and application value for the reconstruction of long time series remote sensing data.greatIn this paper, reconstruction method for NDVI time series data set has accomplished based on MODIS-NDVI products from 2001 to 2003 in Jiangxi Province. The main conclusions are as follows:(1) Comparison between two sets of quality identification data of NDVI dataset which are VI usefulness(QA) and pixel reliability(VIP) has been made. Statistics of NDVI mean value acquired by QA, VIP and the combined QA and VIP quality identifier has been analyzed. A new quality identifier(QAVIP quality identifier) has been developed according to the description information of QA and VIP, and it shows better corresponding relationship with quality of NDVI pixel values.(2) A new quality weight coefficient, which is converted from the classification of that QAVIP quality identifier is applied for new quality weight Savitzky-Golay(S-G) filter method. Comparison of bias, standard deviation and correlation coefficient among original data set, quality weight S-G filter results and QAVIP quality weight S-G filter results reveals that these two reconstruction methods both perform well for high quality NDVI values, and that QAVIP quality weight S-G filter results behave better than quality weight S-G filter results for low quality NDVI values and noise.(3) Based on that QAVIP quality identifier and the description information of VIP, replacement of pixels of low quality and noise has been conducted. A novel algorithm has been designed by means of the following procedures. Firstly, the mean values of high quality pixels whose vegetation type are same as noise pixels should be used for replacing values of noise pixels. It can be achieved with the aid of sliding window techniques. Then Savitzky-Golay(S-G) filter is applied to reconstruct the above results. The original NDVI values of high quality have been also remained.The proposed method preserves and stabilizes more high quality NDVI data, and it does act well in reconstructing MODIS-NDVI time series products.(4) Results of various reconstruction methods for NDVI time series data of Jiangxi Province from 2001 to 2003 exhibit the following achievements. All the methods have certain effects on reconstruction of discrete noise. Reconstruction NDVI values of adaptive S-G filter, quality weight S-G filter and mean value iterative(MVI) methods are still low, while values based on BISE, QAVIP quality weight S-G filter, quality weight S-G filter, pixel quality analyzed S-G filter and improved MVI reconstruction methods show consistent with the reference image. As for continuous noise, adaptive S-G filter and BISE has no obvious effect. There is a certain effection of noise by MVI, and an improved impact on noise by improved MVI. Reconstruction values of pixel quality analyzed S-G filter and QAVIP quality weight S-G filter are much higher than values of quality weight filter. Meanwhile, pixel quality analyzed S-G filter and QAVIP quality weight S-G filter show better performance than other reconstruction methods.(5) Dynamic threshold of the original data extraction fallow fields far below the phenological information threshold to extract the winter fallow fields and 2008 winter fallow field area. Owing to the reconstruction of NDVI time series data set by various methods, the extraction of winter fallow fields based on NDVI dynamic threshold exhibits much more accurate than dynamic threshold based on raw data. While the area of winter fallow fields based on the extraction of reconstruction image is between area of phenological information extraction and the fallow area in 2008.