Tropical Cyclone Intensity Probabilistic Forecasting System Based on Deep Learning

Figure 1

An illustration of PTCIF. It takes multimodal spatiotemporal sequence data as input and outputs probability distributions and can further obtain prediction intervals. Figure (a) shows the overall structure of PTCIF, the input mainly contains 3D features and statistical features, and the middle of the model contains some feature extraction and fusion components. Figure (b) shows an illustration of the encoder module, and figure (c) shows an illustration of the CBAM module.

(a)

(b)

(c)