Tropical Cyclone Intensity Prediction Using Deep Convolutional Neural Network

In this study, deep convolutional neural network (CNN) models of stimulated tropical cyclone intensity (TCI), minimum central pressure (MCP), and maximum 2 min mean wind speed at near center (MWS) were constructed based on ocean atmospheric reanalysis, as well Best Track hurricane data over 2014–2018. order to explore the interpretability model structure, sensitivity experiments designed with various combinations predictors. The test results show that simplified VGG-16 (VGG-16 s) outperforms other two general (LeNet-5 AlexNet). display good consistency hypothesis perceptions, which verifies validity reliability model. Furthermore, also suggest importance predictors varies in different targets. top three factors are highly related TCI sea surface temperature (SST), 500 hPa (TEM_500), differences between 850 (vertical shear speed, VWSS). VWSS, relative humidity (RH), SST more significant than MCP. For MWS SST, TEM_500, (TEM_850) outweigh variables. This conclusion implies learning could be an alternative way conduct intensive quantitative research.