AI Data Classification of Atmospheric Ducts for Advanced Radar Modeling

Authors

  • Charlie Paterson University of Portsmouth, School of Mathematics and Physics, Portsmouth, PO1 3FX, United Knigdom
  • Rebecca Caves QinetiQ, Portsmouth, PO6 3RU, United Kingdom
  • Chloe Peet QinetiQ, Portsmouth, PO6 3RU, United Kingdom
  • William Dawber QinetiQ, Portsmouth, PO6 3RU, United Kingdom

DOI:

https://doi.org/10.59973/emjsr.243

Keywords:

Atmospheric Refractivity, Neural Network, Ducting, Evaporation Duct, Hybrid Duct, Machine Learning, Regression, Propagation Modelling, Refractivity Prediction, Environmental Data, AI

Abstract

This study explores the development of neural network models for predicting atmospheric refractivity profiles from the input of sea clutter returns, with a focus on two distinct duct types: evaporation ducts and hybrid ducts. Accurate prediction of these profiles is critical for various defense and communication applications, where atmospheric refractivity significantly influences radar performance and signal propagation. To capture the differing characteristics of each duct type, three models were evaluated: two specialized models trained independently on evaporation and hybrid duct data, and a joint model trained on both. Prior to training, the dataset was balanced by down-sampling the more frequent hybrid duct cases to prevent bias.
Model performance was assessed using standard regression metrics including Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and the coefficient of determination (R²). The specialized evaporation model achieved an R² score of 0.9894 and an RMSE of 0.10322, while the hybrid duct model demonstrated superior precision with an R² of 0.9996 and RMSE of just 0.02095. The joint model maintained strong overall performance (R² = 0.9965, RMSE = 0.0591), though slight performance degradation was noted in evaporation duct cases, suggesting mild overfitting to the structurally complex hybrid ducts. Model predictions closely matched true refractivity profiles across both duct types, including cases with steep gradient transitions. These results underscore the capacity of neural networks to generalize well across atmospheric
conditions. While joint models offer practical efficiency, specialized models may provide enhanced accuracy for operational scenarios where precision in refractivity prediction is critical.

References

Kirandeep Kaur and Bharat Gurnani, Refraction of Light (2023), https://www.ncbi.nlm.nih.gov/books/NBK594230/

Samudrapom Dam, Understanding Snell’s Law in Light Refraction (2024), https://www.azooptics.comArticle.aspx?ArticleID=2708

RadarTutorial, Anomalous Propagation of Electromagnetic Waves (n.d.), https://www.radartutorial.eu/07.waves/wa17.en.html

Kulessa, A.S., Barrios, A., Claverie, J., Garrett, S., Haack, T., Hacker, J.M., Hansen, H., Horgan, K., Hurtaud, Y., Lemon, C., Marshall, R., McGregor, J., McMillan, M., Periard, C., Pourret, V., Price, J., Rogers, L., Short, C., Veasey, M. and Wiss, V., 2016. The Tropical Air-Sea Propagation Study (TAPS). Bulletin of the American Meteorological Society. DOI: 10.1175/BAMS-D-14-00284.1. DOI: https://doi.org/10.1175/BAMS-D-14-00284.1

Kieran Berryman, What is Temperature Inversion? (2024), https://www2.purpleair.com/blogs/blog-home/what-is-temperature-inversion?

Barrio-Zhang, H., Ruiz-Guti´errez, ´E., Orejon, D., Wells, G. and Ledesma-Aguilar, R., 2024. Droplet motion driven by humidity

gradients during evaporation and condensation. The European Physical Journal E, 47. DOI: https://doi.org/10.1140/epje/s10189-024-00426-7. DOI: https://doi.org/10.1140/epje/s10189-024-00426-7

Chai, X., Li, J., Zhao, J., Hu, Y. and Zhao, X., 2022. NWPP-EDH: Numerical weather prediction products evaporation duct regional prediction model. Frontiers in Earth Science, 10, p.1031186. Available at: https://www.frontiersin.org/articles/10.3389/feart.2022.1031186/full]. https://doi.org/10.3389/feart.2022.1031186. DOI: https://doi.org/10.3389/feart.2022.1031186

Zhu, X., Yan, K., Jiang, L., Tian, L. and Tian, B., 2024. A method for elevated ducts refinement based on convolutional neural network. Radio Science, 59, p.1. https://doi.org/10.1029/2023RS007789. [Accessed 5 Jul. 2025]. DOI: https://doi.org/10.1029/2023RS007789

Ronaghan, S., 2018. Deep Learning: Overview of Neurons and Activation Functions. [online] Available at: https://srnghn.medium.com/deep-learning-overview-of-neurons-and-activation-functions-1d98286cf1e4.

Downloads

Published

2025-07-19 — Updated on 2025-07-19

Versions

How to Cite

Paterson, C., Caves, R., Peet, C., & Dawber, W. (2025). AI Data Classification of Atmospheric Ducts for Advanced Radar Modeling. Emerging Minds Journal for Student Research, 3, P36-P49. https://doi.org/10.59973/emjsr.243

Issue

Section

Physics

Most read articles by the same author(s)