Deriving overlapped cloud motion vectors based on geostationary satellite on Typhoon Mulan

We present a novel Overlapped Cloud Motion Vectors (OCMVs) deriving algorithm using the Himawari-8 satellite. A multi-layer Cloud Top Heights (CTHs) retrieving model based on multi-spectral observed radiances is constructed using neural networks to reduce the substantial uncertainty of CMVs over multi-layer clouds. The retrieved CTHs are assigned to upper ice and lower water cloud layers, then they are used as respective tracers for deriving OCMVs based on an optical flow algorithm. -The files of trainmodel1.py and trainmodel1.py are the codes for training the multi-layer Cloud Top Heights (CTHs) retrieving models; -The six csv files concludes all the matched datasets we used in the neural network (NN ) models; -The file of datapre.py is codes for retrieving CTHs for Typhoon Mulan -The file of optical_fb_cwv12_mulan_save.py is the code for estimating CMVs from the multi-layer Cloud Top Heights (CTHs) -The CTH_20220808-20220810 zip file concludes the derived CTH from the NN models during Mulan period; -The CWV_20220808-20220810 zip file concludes the derived CWV from the NN models during Mulan period; --The FuXi-forecast zip file concludes the 12-hour forecasted wind vectors by FuXi in 20220809.00:00 and 20220809.03:00;