Creating Grid-Based Machine Learning Severe Weather Guidance for Watch-to-Warning Lead Times in the Warn-on-Forecast System - Thesis

Aug 2, 2026

Upload by Violet

This thesis develops grid-based machine learning guidance to improve watch-to-warning lead times within the Warn-on-Forecast (WoFS) system. It builds a workflow spanning dataset construction, feature engineering from WoFS ensemble forecasts, and several baseline and advanced modeling approaches. Traditional machine learning and deep learning methods are trained and evaluated with verification metrics, including objective skill, case studies, stratified verification by initialization time, and feature ablation to quantify contributions across severe hazards.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages102

Reading Time4 h 17 min

AccessFree to download

Acknowledgments

List Of Tables

List Of Figures

Abstract

Introduction

Literature Review

Watch-to-Warning Guidance

Machine Learning for Severe Weather

Data & Methods

Warn-on-Forecast System

Dataset

Warn-on-Forecast Datasets

Feature Engineering

Target Data

Baseline and Machine Learning

Baseline Models

Logistic Regression

Tree-based Machine Learning

Convolutional Neural Networks and U-nets

Deep Learning Methods

Verification Metrics

Results

Performance of Traditional Machine Learning Techniques

Stratified Verification

Feature Ablation

Performance of Models on Final Dataset

Performance of Deep Learning Techniques

Conclusions and Summary

Discussion

Summary and Future Work

Reference List

UNIVERSITY OF OKLAHOMA GRADUATE COLLEGE CREATING GRID-BASED MACHINE LEARNING SEVERE WEATHER GUIDANCE FOR WATCH-TO-WARNING LEAD TIMES IN THE WARN-ON-FORECAST SYSTEM A THESIS SUBMITTED TO THE GRADUATE FACULTY in partial fulfillment of the requirements for the Degree of MASTER OF SCIENCE By SAMUEL VARGA Norman, Oklahoma 2024 CREATING GRID-BASED MACHINE LEARNING SEVERE WEATHER GUIDANCE FOR WATCH-TO-WARNING LEAD TIMES IN THE WARN-ON-FORECAST SYSTEM A THESIS APPROVED FOR THE SCHOOL OF METEOROLOGY BY THE COMMITTEE CONSISTING OF Dr. Corey Potvin, Chair Dr. Montgomery Flora Dr. Aaron Hill Dr. Cameron Homeyer © Copyright by SAMUEL VARGA 2024 All Rights Reserved. Acknowledgments To begin, I thank my research advisors: Dr. Corey Potvin and Dr. Montgomery Flora. Throughout my two years at the University of Oklahoma, they have continued to challenge me to grow professionally and academically. Their continuous support and guidance have been instrumental to my success and growth over the last two years. I also thank the members of the WoFS team for welcoming me and for all of the opportunities they have provided. The ability to consult with them has been paramount to both my growth and the success of this project. Quite literally, this project would not be possible without their hard work. Finally, I thank my family and friends for their support. They may not have understood what I was saying, yet they listened all the same. This material is based upon work supported by the Joint Technology Transfer Initiative Program within the NOAA/OAR Weather Program Office under Award No. NA22OAR4590171 . Table of Contents Acknowledgments iv List Of Tables vii List Of Figures viii Abstract xiii 1 Introduction 1 2 Literature Review 3 2.1 Watch-to-Warning Guidance ........................ 3 2.2 Machine Learning for Severe Weather ................... 5 3 Data & Methods 10 3.1 Warn-on-Forecast System ......................... 10 3.2 Dataset ................................... 11 3.2.1 Warn-on-Forecast Datasets ..................... 11 3.2.2 Feature Engineering ........................ 13 3.2.3 Target Data ............................. 15 3.3 Baseline and Machine Learning ...................... 18 3.3.1 Baseline Models ........................... 18 3.3.2 Logistic Regression ......................... 19 3.3.3 Tree-based Machine Learning ................... 20 3.3.4 Convolutional Neural Networks and U-nets ............ 22 3.4 Deep Learning Methods .......................... 28 3.5 Verification Metrics ............................. 29 4 Results 34 4.1 Performance of Traditional Machine Learning Techniques ........ 34 4.1.1 Objective Skill ........................... 34 4.1.2 Case Study ............................. 37 4.2 Stratified Verification ............................ 46 4.2.1 Skill by Initialization Time ..................... 46 4.2.2 Impact of Removing Cases ..................... 48 4.3 Feature Ablation .............................. 56 4.4 Performance of Models on Final Dataset ................. 62 4.4.1 Objective Skill ........................... 62 4.4.2 Case Study ............................. 63 4.5 Performance of Deep Learning Techniques ................ 71 4.5.1 Objective Skill ........................... 71 4.5.2 Case Study ............................. 72 5 Conclusions and Summary 77 5.1 Discussion .................................. 77 5.2 Summary and Future Work ........................ 80 Reference List 85 List Of Tables 3. 1 Fields extracted from WoFS ensemble forecasts and used to create features. Fields 1-8 are intrastorm while the remaining fields are classed as environmental. The ensemble statistics for each field type are also listed. Each statistic is calculated three times per field due to the three different smoothing radii, resulting in 174 predictor fields 14 3.2 The standard 2x2 contingency table, also k

FAQs

What problem does the thesis address in severe weather forecasting?

It targets improved watch-to-warning lead times by generating guidance using grid-based machine learning inside the Warn-on-Forecast (WoFS) system.

How is the dataset constructed for model training and evaluation?

The work uses WoFS ensemble forecasts to create features, defines target data for the warning-related guidance, and compiles datasets spanning initial and final case sets.

What verification approaches are used to assess model performance?

Performance is evaluated using objective skill measures, case studies, stratified verification by initialization time, feature ablation studies, and metrics suitable for probabilistic verification.

Aug 2, 2026

Upload by Violet