Peer-Reviewed Publication: You, S., P. Zhu, et al. 2025: Predicting Tropical Cyclone Intensity Using a Convolutional Neural Network and 20 Years of IMERG Satellite Rainfall Data, Weather and Forecasting, 40, 2317–2331.
This repository contains my code for my HIECNN, HIPCNN, and HRICNN models. All 3 models are part of the Real Time HAI product (in development currently).
The purpose of this project is to create convolutional neural network (CNN) models that can estimate and predict hurricane intensity from satellite imagery. Each model focuses on a different aspect of hurricane intensity estimation: HIECNN estimates current intensity, HIPCNN predicts intensity changes over time, and HRICNN estimates rapid intensification events.
All the models accept satellite images and environmental data as inputs to produce their respective outputs.
- HIECNN (Hurricane Intensity Estimation CNN): Estimates current hurricane intensity from satellite images.
- HIPCNN (Hurricane Intensity Prediction CNN): Predicts future hurricane intensity changes based on current satellite images and environmental data.
- HRICNN (Hurricane Rapid Intensification CNN): Estimates the likelihood of rapid intensification events using satellite imagery and environmental factors.
The models utilize NASA IMERG satellite images as inputs. Below are examples of the input images used.
Below is a diagram illustrating the architecture of the CNN models used in this project.
- Python (TensorFlow, Keras, NumPy, Pandas)
- Jupyter Notebooks (for experimentation and prototyping)
- R (for data analysis and visualization)
- HTML/CSS/JS (for future web interface development)
- Web Demo for future HAI product: YouTube Link
- HIECNN Model Explanation: YouTube Link
- HIPCNN Model Explanation (previously known as HCNN): YouTube Link
- HRICNN Model Explanation: YouTube Link