Study On The Spatial-temporal Data Fusion Method Using Tasseled-Cap-Transform Indices From Landsat TM And MODIS

High spatial resolution and multi-spectral characteristics of Landsat TM data makes it widely used in many fields. However a longer revisit cycle and frequent cloud have limited the application in time series analysis. Otherwise, high temporal resolution of MODIS data is more suitable for time series analysis, but MODIS data has less details with250m to1000m spatial resolution, that makes it more suitable for study in large-scale.Temporal-resolution data fusion method based on Landsat TM and MODIS data obtains new data with higher spatial resolution of TM and higher temporal resolution of MODIS for time series analysis to monitor changes in higher spatial resolution.STARFM (Spatial and Temporal Adaptive Reflectance Fusion Model) is one of the models based on reflectance that are used more often currently and has higher accuracy. This paper obtains series of fused images using STARFM with adjusting the size of the parameters in original algorithm, and the results are compared with the reference image to determine the optimal combination of parameters. On this basis, the GSTARFM(GVI STARFM) is presented. Time weighting factor consisted of the acquired time of two pairs of input images based on the cyclical and short gradient features of vegetation greenness index is used in GSTARFM to improve the quality of fused images. The main contributions are as follows:(1) This study sets a range of different values for the moving window, number of class, uncertainty values, distance weighting to get fused images based on tasseled cap transformation component of Landsat TM and MODIS data,comparing with the tasseled cap transformation component of the actual acquired Landsat TM,and the results show that:The increase of moving window and adjusting the number of classes will help to improve the quality of integration, and the change of the constant for distance weighting does not affect the quality of fusion results. When the uncertainty of MODIS and TM images is a non-zero value, the result was less affected. In summary, the adjustments of the parameter in STARFM will affect the fused images, but the improvement is limited.(2) This paper discusses the effect on fused images based on Landsat TM and MODIS acquired in2007. The results are follows:1) The larger the time difference between input images and fused images, the lower the precision of the fused images.2) The accuracy is better when the acquired time of input images is at the peak of vegetation growth.. And over time, while vegetation growth condition changes significantly, the accuracy will be reduced.(3) GSTARFM that assumes GVI shows a uniform change in the short term is presented for GVI in this paper based on STARFM. Compared with STARFM, the improvements are that GSTARFM selected similar neighboring pixels from six Tasseled-Cap-Transform indices of TM images at two moments and that time weighting factor introduced to calculate weights.(4) Fused images time series can show the basic features of vegetation growth. The greenup-. maturity and senescense of the growing reason are clearly showed from the curve and the sort of peaks and the time are consistent with the field survey results. All of that show the effectiveness of GSTARFM.