Reconstruction Methods Of Medium Spatial Resolution Time-series NDVI Data Sets

The detection of time-series vegetation dynamic change is an important area and hot direction of environmental remote sensing research.High-quality time-series data sets have great significant on research and detection of land use change,phenology and climate change,ecosystem variation and biodiversity distribution.However,due to the limitation of existing remote sensing data platforms,it is difficult to acquire high-quality medium and high spatial resolution time-series data sets.Consequently,carrying out the research of time-series data reconstruction method as well as enhancing existing time-series quality would be significant.In this paper,MODIS and Landsat remote sensing data in Changting County,Fujian province were utilized to carry out the research of medium resolution NDVI reconstruction methods.The main research contents and results are summarized as follows:(1)Filtering analysis of MODIS NDVI time-series products.Quality factor(QA)was introduced in the study,and the MODIS NDVI data sets from 2000 to 2013 of Changting County were filtered based on TIMESAT platform.The comparison and analysis of different filtering methods show that Asymmetric Gaussian Function Fitting(AG)and Double Logistic Function(DL)have better high-quality NDVI fidelity,especially AG is superior.However,Savitzky-Golay Fitting(S-G)has the overall low value,and it is easy to bring in new noise in growing season.AG was chosen as the ultimate filter method to obtain the after denoising MODIS NDVI data sets,providing the high-quality data foundation for time-series reconstruction.(2)Research on time-series NDVI reconstruction methods.Three reconstruction methods of NDVI data sets were researched and compared in varies conditions of data inputs:Spatial and Temporal Adaptive Reflectance Fusion Model(STARFM),Enhanced Spatial and Temporal Adaptive Reflectance Fusion Model(ESTARFM)and regression analysis method.The results indicated that there was no one method can be optimal in all situations.ESTARFM showed the best performance in temporal and spatial variation when seasonal differences were apparent.STARFM was obvious affected by seasonal aspect,however,it has most robust property when the predict time was close to the obtained time.Regression analysis method can acquire the best reconstruction results when the predict time has smaller interval from obtained time,nevertheless,it was liable to appear higher or lower than true value.(3)Reconstruction of time-series NDVI data sets of Changting county.Reconstruction scheme of time-series data sets were formulated based on the obtained data,and four seasons per year NDVI data sets with 30 meters spatial resolution of Changting county from 2000 to 2013 were established,which demonstrated high similarity with MODIS NDVI on time variation trends and well consistency with real Landsate value.We evaluated the quality of the data sets we reconstructed from the following aspacts:the similarity of variation tendency,the spatiotemporal trend of typical objects,the comparison between reconstruction results and true value,the rationality of data sets and the evaluation of reconstruction results in complex land type.The result showed that the reconstructed time-series data sets had high similarity of temporal variation trend with original MODIS NDVI data sets.Also,it can reflect the seasonal and annual variation of different land cover types reasonably,which showed fine reconstruction results.