Study On The Spatiotemporal Dynamics Of Snow Cover In The Northern Hemisphere Based On Multi-source Remote Sensing Big Data And Cloud Computing Platfor

Snow cover regulates the water cycle and energy exchange in most parts of the Northern Hemisphere.The snow cover days and snow cover area showed a significant downward trend.This research is based on the powerful remote sensing image processing ability of Google Earth Engine cloud computing platform and LSTM neural network model.The daily binary snow cover data set in the Northern Hemisphere is generated by using MODIS,Landsat and other multi-source remote sensing data.This study uses Landsat satellite data to verify the snow recognition accuracy of MOD10A1 satellite,and analyzes the relationship between MOD10A1 snow recognition accuracy and snow month and altitude.The temporal and spatial changes and distribution characteristics of snow cover frequency,snow cover area and snow cover phenology in the Northern Hemisphere from 2000 to 2019 were studied;A snow cover area prediction model based on STL-LSTM neural network is proposed,and the prediction results of different time steps and different network models are analyzed;The correlation between snow cover phenology and the regional contribution to the change of snow cover phenology in the Northern Hemisphere were discussed.The specific research contents are as follows:(1)Using the 30 m Landsat data to verify the 500 m MOD10A1 data in the Northern hemisphere,the F1 score based on the binary snow map MOD10A1 was 85.4%,and the overall accuracy was 91.4%.The false positive rate of MOD10A1 snow recognition is significantly lower than the false negative rate.(2)In the Northern Hemisphere,there are notable interannual and regional changes in snow cover frequency based on Google Earth Engine Cloud computing platform.The snow cover frequency from 2000 to 2019 was in the regions near Eurasia(55°-65°N,30°-120°E),The Mongolian Plateau,Stanov Plateau and Greenland declined at a rate of0.4-0.6 d/a,and showed an increasing trend of 0.3-0.4 d/a in central Kazakhstan,Eastern Siberia and Northeast China.(3)From 2000 to 2019,the snow cover area in the Northern Hemisphere decreased significantly in spring and winter,and the average annual winter decline trend reached1.9×10~5 km~2;The STL-LSTM snow area prediction model is constructed by combining STL method and LSTM neural network.When the time step is 6,the prediction effect of STL-LSTM is the best,with an average R~2 of 0.96 and an average MAPE of 4.18.Compared with other models,the effectiveness of STL time series decomposition method combined with LSTM neural network method is proved.(4)From 2000 to 2019,the snow cover days in the vicinity of Kazakhstan and the central plains of North America increased significantly,with an average increase of 1.1 d per year.The change of snow cover phenology in the Northern Hemisphere is primarily due to the change in snow cover end date,and the correlation between the two reaches0.89;the analysis of regional contribution to the change of snow phenology in the Northern Hemisphere shows that the change of snow phenology in the Northern Hemisphere is mainly caused by the snow phenology of Eurasia,with a contribution rate of roughly 60%.Studying the temporal and spatial dynamics of snow cover frequency,snow cover area and snow cover phenology in the Northern Hemisphere is of great significance to global climate change,it would provide evidence for further revealing the interaction mechanism between snow cover and climate change.