Proceedings of the Conference on Bridging Engineering Disciplines with AI and Machine Learning (BEDAIML 2026)

<Previous Article In Volume
Next Article In Volume>

PatchTSTSpike: A Patch-Based Transformer Framework for Cloudburst-Like Extreme Rainfall Detection Using ERA5 Reanalysis

Authors
Ankit Kumar1, Prashant Kumar1, *, Yukiharu Hiasaki2, Rajni Rajni3
1Department of Mathematics, National Institute of Technology, Delhi, 110040, India
2Department of Physics and Earth Sciences Faculty of Science, University of the Ryukyus, Nishihara, Japan
3Jindal Global Business School, O.P Jindal Global University, Sonipat, Haryana, 131001, India
*Corresponding author. Email: prashantkumar@nitdelhi.ac.in
Corresponding Author
Prashant Kumar
Available Online 4 June 2026.
DOI
10.2991/978-94-6239-697-5_30How to use a DOI?
Keywords
Himalayan Meteorology; Rare-event Detection; Transformer Models; ERA5 Reanalysis; Extreme Rainfall
Abstract

The occurrence of cloudburst-like heavy rainfall events in the Himalayan region is very rare but highly destructive on the other hand predicting them is even more difficult, These event initiates from very complex atmospheric crossover like change in pressure, temperature, Wind etc. In this paper, we have developed PatchTST, that is, a custom spike aware temporal patch based transformer model [5][11][15]. It is designed to analyze extreme rainfalls using Fifth Generation ECMWF Reanalysis data (ERA5 reanalysis data)[Link] [4][14]. We evaluate the model using variable window length ranging from 30 to 90 days. Among all of these windows, the 30 -day look-back window provides the most promising detection performance, getting a ROC - AUC of 0.6007 . While using longer temporal windows, particularly 60 days, slight improvement in precision on the cost of reduced recall, implying a give and take between event detection sensitivity and prediction certainty. Overall, the used framework offers a subtle transformer-based approach for studying spike event like precipitation in mountains or hilly areas and provides a practical base for in-future data-driven research on Himalayan weather and it’s extremes.

Copyright
© 2026 The Author(s)
Open Access
Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

<Previous Article In Volume
Next Article In Volume>
Volume Title
Proceedings of the Conference on Bridging Engineering Disciplines with AI and Machine Learning (BEDAIML 2026)
Series
Advances in Intelligent Systems Research
Publication Date
4 June 2026
ISBN
978-94-6239-697-5
ISSN
1951-6851
DOI
10.2991/978-94-6239-697-5_30How to use a DOI?
Copyright
© 2026 The Author(s)
Open Access
Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Cite this article

risenwbib
TY  - CONF
AU  - Ankit Kumar
AU  - Prashant Kumar
AU  - Yukiharu Hiasaki
AU  - Rajni Rajni
PY  - 2026
DA  - 2026/06/04
TI  - PatchTSTSpike: A Patch-Based Transformer Framework for Cloudburst-Like Extreme Rainfall Detection Using ERA5 Reanalysis
BT  - Proceedings of the Conference on Bridging Engineering Disciplines with AI and Machine Learning (BEDAIML 2026)
PB  - Atlantis Press
SP  - 357
EP  - 369
SN  - 1951-6851
UR  - https://doi.org/10.2991/978-94-6239-697-5_30
DO  - 10.2991/978-94-6239-697-5_30
ID  - Kumar2026
ER  -