| Snow is an important part of land cover.Using high spatial and temporal resolution remote sensing data to dynamically monitor snow cover is an effective method.However,it is difficult to obtain temporally continuous high-spatialresolution remote sensing images due to the influence of cloud layers and sensor revisit cycles.The classical ESTARFM spatio-temporal data fusion model obtains images with both high temporal and spatial resolution by fusing high spatial resolution images and high temporal resolution images.It breaks through the limitations of a single sensor.The Sentinel-2 A/B MSI satellite images launched by the European Space Agency(ESA)have high spatial and temporal resolution,high spectral resolution and radiometric resolution,which are very suitable for the study of snow cover changes at the basin scale.Therefore,this study takes the Babao River Basin in the east of Qilian Mountain as the research area,and attempts to apply ESTARFM to Sentinel-2 MSI and MODIS data fusion to capture the information of rapid changes in snow cover.However,the ESTARFM model has the following two shortcomings: 1)When searching for similar pixels,using a fixed-size sliding window may cause computational redundancy.In areas with high heterogeneity,the number of similar pixels in the sliding window may be insufficient,which affects the accuracy of data fusion.Therefore,it is necessary to establish a dynamic sliding window to filter similar pixels.2)The increase of the sliding window will lead to the homogenization of the spectral reflectance of similar elements within the window.Therefore,the spatial weights of the neighboring similar pixels should be enhanced while establishing the dynamic sliding window to reduce the homogenization effect between pixels.Firstly,in this paper,the sliding window size for screening similar image pixels and the spatial distance weights of similar image pixels in this model are improved to address the shortcomings of ESTARFM model.The fusion accuracy of the improved model is verified by Sentinel-2 and GF-2 images.Secondly,the improved model is used to fuse the Sentinel-2 MSI and MODIS images in the study area to obtain the Sentinel-2 time-series data from July 25,2019 to July 4,2020 in the Babao River basin and extract the snow cover information.Finally,the temporal and spatial variation characteristics of snow cover in the Babao River basin were analyzed by combining the near-surface air temperature and precipitation data.The main conclusions are as follows:(1)The improved ESTARFM model has a good fusion effect in both qualitative and quantitative evaluations.In the qualitative analysis,the fused images of the improved model have some improvement in the clarity and accuracy of feature boundaries and better visualization.In the qualitative analysis,the fused images of the improved model have some improvement in the clarity and accuracy of feature boundaries and better visualization.In the quantitative analysis of the four sample areas,the band correlation coefficients(R)between the fused images of the improved model and the original Sentinel-2 images were almost all above 0.9.And many quantitative indexes such as root mean square error(RMSE)and mean absolute error(MAE)of the improved model are better than the ESTARFM model.(2)The improved ESTARFM model has better snow information fusion effect than the original model.The snow fusion effect of the improved model is evaluated based on the snow map extracted from GF-2.It is found that compared with the ESTARFM model,the snow texture of the fused image of the improved ESTARFM model is closer to that of the GF-2 image,and the deviation of snow cover is reduced compared with the original model.It indicates that the improved ESTARFM model has good and improved snow fusion effect,in which the deviation of snow cover in flat area E is reduced the most,which is 4.78%.(3)Within a snow accumulation year,the snow area in the Baba River basin shows the characteristics of increasing first and then decreasing.Spatially,it shows the characteristics of accumulating first in the surrounding mountainous areas and melting first in the river valley areas.The changes of snow cover in the whole basin are closely related to meteorological factors.The area of unstable snow in the basin accounts for about the total area of the basin in general,mainly distributed in the river valley area.The area with more snow days accounts for about 1/4 of the watershed area,mainly distributed in the high altitude mountains around the basin. |
