Using Machine Learning to Improve the NSSL’s Warn-On-Forecast System’s Prediction of Thunderstorm

Aug 2, 2026

Upload by Anda

This thesis develops and evaluates machine learning approaches to improve the NSSL Warn-on-Forecast (WoFS) system’s prediction skill for thunderstorms. The work centers on deep learning methods applied to WoFS specifications, including U-Net based architectures and a patching strategy for training data. Model performance is assessed using verification metrics, with comparisons against the WoFS baseline and case studies spanning best, average, and worst scenarios. Verification results are complemented by explainability techniques to clarify learned patterns, and findings are used to guide iterative goals for future model improvements.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages67

Reading Time2 h 49 min

AccessFree to download

Acknowledgements

List Of Tables

List Of Figures

Abstract

Introduction

Literature Review

Nowcasting Thunderstorms

Warn-on-Forecast System Previous Works

Deep Learning and the Atmosphere

Methods and Data

Warn-on-Forecast System Specifications

U-Nets

Dataset

Patching Scheme

Deep Learning Model Methods

Results

Verification Metrics

WoFS Baseline

Initial Deep Learning Performance

Current Deep Learning Performance

Explainability

Conclusion and Future Work

Reference List

UNIVERSITY OF OKLAHOMA GRADUATE COLLEGE USING MACHINE LEARNING TO IMPROVE THE NSSL’S WARN-ON-FORECAST SYSTEM’S PREDICTION OF THUNDERSTORM LOCATION A THESIS SUBMITTED TO THE GRADUATE FACULTY in partial fulfillment of the requirements for the Degree of MASTER OF SCIENCE By CHAD WILEY Norman, Oklahoma USING MACHINE LEARNING TO IMPROVE THE NSSL’S WARN-ON-FORECAST SYSTEM’S PREDICTION OF THUNDERSTORM LOCATION A THESIS APPROVED FOR THE SCHOOL OF METEOROLOGY BY THE COMMITTEE CONSISTING OF Dr. Corey Potvin, Chair Dr. Amy McGovern, Co-Chair Dr. Montgomery Flora Dr. Cameron Homeyer © Copyright by CHAD WILEY 2023 All Rights Reserved. Acknowledgements This thesis would not have been possible without the help and support of my advisors, friends and family, and colleagues. Without the support and sacrifices made by each person, I would not have been able to find the success I have found in my time at the University of Oklahoma and NSSL. I would like to first and foremost thank my advisors, Dr. Corey Potvin, Dr. Amy McGovern, and Dr. Montgomery Flora. Their guidance, mentorship, and expertise led to me reaching to goals I had set before myself. The bi-weekly meeting with Dr. Potvin and Dr. Flora challenged me to become a better scientist through discussion of findings, thought processes, and communication of my work, and I feel I have learned so much from that time spent meeting. Dr. Flora was also instrumental in the guidance of verification and explainability techniques in machine learning and proved crucial to the project. Dr. Amy McGoven provided guidance, advice, and new ideas to try with my project. Her support and guidance in our weekly meetings often re-centered me on the task, and I would often leave the meeting feeling revitalized and motivated to tackle the challenges the master’s degree presented. I would also like to extend thank yous to Dr. Randy Chase and Tobias Schmidt, who both provided code and were patient teachers. Dr. Chase’s extensive knowledge of deep learning models and supercomputing saved me time and enabled me to learn so much. Mr. Schmidt’s patching code, debugging expertise, and friendship also saved me immense time and was always someone I could rely on for help when I was stuck. I also need to extend my deepest gratitude to my family and partner, Jess. Their love, support, and visits allowed me to disconnect from the challenges of the program and recharge. I am so blessed to have such an amazing support system. Lastly, I would be remiss if I failed to mention thanks to my two favorite sports teams, the Philadelphia Phillies and Eagles. Their long post-season runs provided me with a much-needed distraction and nightly break from my work. Their subsequent losses in the championships then also motivated me to focus more on my work afterward so that I could forget the heartbreaking losses both teams inflicted on me. Overall, the time each and everyone spent on me during my two years at OU lead me to grow into a better scientist, and I will be forever grateful. This material is based upon work supported by the National Science Foundation under Grant No. ICER-2019758 . Funding was provided by NOAA/Office of Oceanic and Atmospheric Research under NOAA–University of Oklahoma Cooperative Agreement NA21OAR4320204, U.S. Department of Commerce. The computing for this project was performed at the OU Supercomputing Center for Education & Research (OSCER) at the University of Oklahoma (OU) . The statements, findings, conclusions, and recommendations are those of the author(s) and do not necessarily reflect the views of NOAA or the U.S. Department of Commerce. Table of Contents Acknowledgements iv List Of Tables viii List Of Figures ix Abstract xii 1 Introduction 1 2 Literature Review 6 2.1 Nowcasting Thunderstorms ........................ 6 2.2 Warn-on-Forecast System Previous Works ............... 9 2.3 Deep Learning and the Atmosphere ................... 11 3 Method

FAQs

What is the main goal of this thesis?

To improve the NSSL Warn-on-Forecast (WoFS) system’s prediction of thunderstorms by developing and evaluating machine learning methods.

Which deep learning approach and data preparation methods are used?

The thesis uses U-Net based models and a patching scheme, including an initial and a current patching strategy, combined with WoFS specification inputs.

How are the models evaluated and interpreted?

Performance is measured with verification metrics and compared to the WoFS baseline through best/average/worst case studies. Explainability techniques are also applied to provide insight into model behavior.

Aug 2, 2026

Upload by Anda