Temporal-spectral Quality Promotion Methods Of Spatiotemporal Fusion Products From Multi-source Remotely Sensed Images

High precision time series remote sensing data have important guiding significance for the study of regional ground change(such as vegetation cycle growth change,dynamic monitoring of land resource utilization).However,due to the limitation of the sensor itself,the spatial resolution and temporal resolution of satellite image are mutually restrict.In addition,the negative influence of cloud,rain,and other atmospheric condition,which cause not only less number of clear cloudless satellite images of a field,but also contain a lot of noise in satellite images.In this case,the effective application of the acquired satellite images has lots of restrictions.Therefore,how to combine the existing spatial-temporal fusion model and time-series filtering model to generate high precision remote sensing time series data has become a crucial problem.In view of the problem that predicted images based on spatial-temporal fusion model may be inaccurate,two study areas are selected in this paper,which contain different surface feature.One is located in Dunhuang city of Gansu province.The other is located in Saihanba national forest park of Hebei province.And combine the spatial-temporal fusion model and time-series filtering model.First,four spatial-temporal fusion models STARFM,FSDAF,SPFM and SPTSFM,are used to generate high spatial and temporal resolution Landsat TM images by fuse Landsat TM satellite data and MODIS product.Then,three time-series filtering model,A-G,S-G,and HANTS filtering model,are used to reconstruct the reconstruct the time series of the generated high-sequence Landsat TM data.The aim of it is to remove the data noise,which introduced in the fusion process and build high-precision time series data.At the same time,when evaluate the accuracy of the generated time series data,the dynamic time wrapping of similarity measurement,is introduced to verify the accuracy,and verify the validity of it by compare the DTW and other assessment index.The results show that: From the overall analysis,the DTW has a high consistency with other assessment index.And combine spatial-temporal fusion model and time-series filtering model can effectively eliminate the noise,and can obtain more accurate time series data compared with any single method.A-G filtering method has a high reconstructing accuracy for the time series data with periodic changes,S-G filtering method has a high reconstructing accuracy for the time series data with invariable features,and HANTS filtering method has obtained a high reconstructing accuracy for both features.