Estimation And Analysis Of Fractional Vegetation Cover And Phenological Parameters Based On Spatio-Temporal Data Fusion

Vegetation is an important component of the global ecosystem.Changes in vegetation can reflect changes in climate factors and are the most direct response of terrestrial ecosystems to environmental changes.With the continuous development of space satellite science,remote sensing technology has become the main means of continuously monitoring regional vegetation.However,due to the ubiquitous surface spatial heterogeneity,there is a certain scale dependence in vegetation information remote sensing inversion,and there must be differences in vegetation parameters inverted using different spatial resolution image data.Therefore,based on the normalized differential vegetation index(NDVI),this study inverted two vegetation parameters,land surface phenology(LSP)and fractional vegetation cover(FVC),to explore the impact of image spatial-temporal resolution on vegetation phenological changes and spatial distribution.High spatial-temporal resolution remote sensing data was generated through a spatiotemporal fusion model to provide data support for vegetation phenology inversion and growth change monitoring.The research on the impact of spatial-temporal resolution on vegetation information extraction will provide a reference for the selection of appropriate resolution image data sources for vegetation remote sensing monitoring.This study used Sentinel-2,Landsat-8,MOD13Q1,and MOD13A1 images as data sources.Firstly,the high spatial resolution NDVI time series data were obtained by fusing the images using the spatiotemporal fusion model ESTARFM(Enhanced Spatial and Temporal Adaptive Reflectance Fusion Model),and the NDVI time series curve was smoothed.Then,the dynamic threshold method was used to invert the growing season start time(SOS),end time(EOS),and length(LOS)of ESTARFM NDVI,MOD13Q1 NDVI,and MOD13A1 NDVI data phenological parameters.The spatial distribution characteristics of LSP extracted from different spatial resolution images were analyzed.After that,the spatial variation function was used to analyze the spatial variation characteristics of vegetation coverage in different seasons,and the seasonal differences and annual temporal changes of FVC extracted from different spatial resolution remote sensing images were compared and analyzed.The main conclusions of this study are as follows:(1)The high spatiotemporal resolution NDVI image generated based on the ESTARFM model combines the advantages of both data sources.The R~2of pixel values between real images and fused images reaches 0.87,indicating that the fusion result can fully reflect the spectral information and pixel values of the image.This provides data support for the study of dynamic monitoring of land surface vegetation.(2)The semivariogram can effectively describe the spatial variation characteristics of FVC.The seasonal differences in spatial variability of FVC indicates as the order of winter>spring>autumn>summer.The differences in the proportion of FVC levels at each resolution in summer and winter are small,while the differences in the proportion of FVC level area at each resolution in spring and autumn are large.The extraction accuracy of FVC with different resolutions indicates as the order of summer>autumn>spring>winter,and the extraction accuracy of FVC in low-resolution images is significantly improved in the summer when vegetation growth is flourishing.Different surface FVC levels have significant differences in their demand for monitoring using high-resolution images.(3)The temporal distribution of average FVC at different resolutions is very similar,and FVC time series curves at different spatial resolutions have consistent seasonal dynamics and amplitudes.However,when the FVC of the 10m spatial resolution image is between0.6-0.7,the FVC extracted from the 250 m/500 m resolution image has the highest accuracy,and the impact of low spatial resolution mixed pixels on FVC extraction is minimal.(4)Differences in tree species result in inconsistent changes in FVC curves during the growth and withering periods.The amplitude of FVC increase and decrease in Xuyi and Ganyu is larger than that in Liyang.Selecting a fusion threshold with a relative error below 12%,determining remote sensing data sources at different time periods,and using a combined resolution approach to extract FVC has improved calculation accuracy and efficiency while reducing computational costs.(5)Most of the forest pixels in the study area have SOS concentrated between the 32nd and 100th day,EOS concentrated between the 224th and 304th day,and LOS concentrated between the 150th and 262nd day.As the spatial resolution of the images decreases,SOS and EOS become more concentrated,resulting in some pixels underestimating SOS and overestimating EOS for forest pixels,thereby lengthening the LOS of forest pixels.Compared to the lower resolution MODIS images,the fused 30-meter high-resolution images provide a more detailed characterization of the spatial distribution of phenology in the study area,preserving more detailed information and having significant advantages on the small-scale research level.