A Spatial-Temporal Data Fusion Method Based On Landsat 8 OLI And MODIS Data

The global change science urgently needs to monitor the seasonal landscape changes at a fine resolution.The high spatial resolution and multi-spectral characteristics of Landsat image contributed to its application in vast fields,but it's application in landscape change was greatly hindered by the insufficiency of data,which was caused by its long revisit period and the impact of cloud.The MODIS image can cover the entirely earth every 1 or 2 days,it is more applicable in the timing analysis and reflection of the seasonal changes of different landscapes,while the spatial resolution(250?100 m)makes it impossible to be applied in small-scale researches.The study of changes of different surface coverage types in heterogeneous environments using remote sensing method requires that data have high temporal and spatial resolution,whereas data from single satellite often cannot meet this requirement.This study was applied in the Midwest of Shandong province.Firstly,impacts on quality of images after fusion were analyzed trough the analysis of data characteristic and the change of parameters,hence the best-input parameters were determined.second,comparative analysis was executed on the STIFM(spatial and temporal data fusion method),STARFM(spatial and temporal adapti-ve reflectance fusion model)and FSDAF(flexible spatiotemporal data fusion method),and the result shows that the image generated by the FSDAF can get better results than other methods when it was applied in areas where the landscape changes quickly.However,FSDAF is sensitive to the quality of input image,the complexity of its algorithm also makes it time consuming,and it cannot take the full advantage of the spatial superiority of fine resolution image and the temporal superiority of the coarse image.At last,we proposed an image fusion method called the IFSDAF(Improved FSDAF)method.In this method,we used the Disturbance Index to describe the change of reflectance,and combined the decomposition of mixed pixels and the FSDAF.The research content and main results were listed as below:(1)The integration of the MODIS daily surface reflectance.Through the analysis of relationship between data characteristics of MODIS and surface reflectance of MOD09GA,we obtained the phenology information through the MODIS NDVI time series data(every 8 days),which was generated by the cooperative double-star method.Then the surface reflectance data of MOD09GA was reconstructed in a step of eight-day to ensure that the surface reflectance of a specific period can be fully utilized.Data processing was fulfilled through program written in the Python scripting language,and MODIS reflectance data of higher quality was achieved.(2)The comparative analysis of commonly used spatial-temporal data fusion algorithm.In this study,three commonly used spatial temporal data fusion method were used to generate image of high spatial and temporal resolution in a specific period,quantitative assessments were executed on the result images,and the paper discussed the influences of classification number on these algorithms,respectively.Meanwhile,the fusion results in different land cover types(dry land,woodland,grassland,urban construction land and rural residential land)were quantitatively assessed from the aspect of spectral and image structure.The applicability of these fusion methods were analyzed,and it showed that the FSDAF algorithm could get more accurate fusion results.(3)Improved flexible spatial-temporal data fusion method.It is found that the fusion results of the FSDAF algorithm was better than other algorithms.But it is also sensitive to data quality,has high computational complexity and did not take full advantage of the spatial detail information of the high spatial resolution image and the timing change of the low spatial resolution images.Based on the analysis of different algorithms,this paper improves the FSDAF algorithm by introducing the surface disturbance index and linear spectral mixed decomposition theory,proposes IFSDAF algorithm and improves the quality of image fusion in the study area.At the same time,we also discuss the influence of parameter on the image quality generated by IFSDAF algorithm.(4)Analysis of Vegetation Growth Timing Characteristics.We used the fused data generated by IFSDAF to extract NDVI time series data and then the extracted data were compared with the original MODIS NDVI timing data(every 8 days).The result shows that the increase,peak and decline of vegetation were consistent with the MODIS NDVI data,and it proved the validity of the IFSDAF algorithm.