Patil et al. (2026) Cyclone Intensity Prediction Using Piecewise CNN and Multispectral Satellite Imagery: A Deep Learning Approach

Identification

• Journal: Lecture notes in networks and systems
• Year: 2026
• Date: 2026-01-01
• Authors: Seema J. Patil, Bhagwat Biradi, Dr. Prof. Pallavi Baviskar, Shweta Joshi, Mallanagouda Patil
• DOI: 10.1007/978-3-032-18141-1_10

Research Groups

• Department of Computer Engineering and Technology, Dr. Vishwanath Karad MIT-World Peace University, Pune, India
• Department of Peace Studies, Dr. Vishwanath Karad MIT-World Peace University, Pune, India
• Sandip Institute of Engineering and Management, Sandip Foundation, Nashik, India
• KJ’s Trinity Academy of Engineering, Pune, India

Short Summary

This paper proposes a deep learning approach using a Piecewise Convolutional Neural Network (CNN) and multispectral satellite imagery to predict cyclone intensity. The method divides satellite images into intensity levels and trains distinct CNN regression models for each level, achieving improved prediction quality with a Mean Absolute Error of 3.95 m/s and a Root Mean Squared Error of 5.19 m/s.

Objective

• To develop and evaluate a deep learning technique for determining cyclone intensity utilizing historical records from the National Hurricane Center’s HURDAT2 archive and an imagery-centric dataset, aiming to improve prediction quality across a range of cyclone strengths.

Study Configuration

• Spatial Scale: Atlantic and Pacific basins.
• Temporal Scale: Historical records, with six-hourly wind speed readings.

Methodology and Data

• Models used: Piecewise Convolutional Neural Network (CNN), which divides satellite images into intensity levels and sends them to distinct CNN regression models, each trained for a particular intensity level.
• Data sources: National Hurricane Center’s HURDAT2 archive (historical records), imagery-centric dataset, multispectral satellite imagery.

Main Results

• The proposed Piecewise CNN model achieved a Mean Absolute Error (MAE) of 7.67 knots (approximately 3.95 m/s) in cyclone intensity prediction.
• The model achieved a Root Mean Squared Error (RMSE) of 10.09 knots (approximately 5.19 m/s).
• The two-stage system, which processes images based on intensity levels, improved prediction quality across a range of cyclone strengths.

Contributions

• Introduction of a novel Piecewise CNN framework that segments satellite images by intensity level and applies specialized CNN regression models, enhancing prediction accuracy across the full spectrum of cyclone intensities.
• Demonstrates a practical deep learning solution for cyclone intensity monitoring with potential for real-time forecasting and future integration of multimodal meteorological inputs.

Funding

• Not specified in the provided text.

Citation

@article{Patil2026Cyclone,
  author = {Patil, Seema J. and Biradi, Bhagwat and Baviskar, Dr. Prof. Pallavi and Joshi, Shweta and Patil, Mallanagouda},
  title = {Cyclone Intensity Prediction Using Piecewise CNN and Multispectral Satellite Imagery: A Deep Learning Approach},
  journal = {Lecture notes in networks and systems},
  year = {2026},
  doi = {10.1007/978-3-032-18141-1_10},
  url = {https://doi.org/10.1007/978-3-032-18141-1_10}
}