Title page

Contents

Highlights 3

Introduction 9

1. Background 11

1.1. Modeling for natural hazard forecasts 11

1.2. AI and machine learning 22

1.3. Machine learning in natural hazard modeling 22

1.4. Policy environment 23

2. Machine Learning May Significantly Enhance Natural Hazard Modeling 25

2.1. Machine learning may speed up forecasting 25

2.2. Machine learning may improve the use of data 27

2.3. Machine learning may improve predictions in the absence of data 29

2.4. Machine learning can help leverage multiple models to improve predictions 30

3. Challenges to Applying Machine Learning in Natural Hazard Modeling 36

3.1. Data gaps, bias, and incompatibility 36

3.2. Trust in machine learning 39

3.3. Limited coordination and collaboration 41

3.4. Workforce and resource needs create barriers to uptake of machine learning 42

4. Policy Options to Help Enhance Benefits or Address Challenges of Using Machine Learning Technologies for Natural Hazard Modeling 45

4.1. Facilitate improved data collection, sharing, and use 45

4.2. Expand machine learning education and training 46

4.3. Address hiring and retention barriers and certain resource shortfalls 47

4.4. Take steps to mitigate bias and foster trust in data and machine learning models 48

4.5. Maintain status quo efforts 49

5. Agency and Expert Comments 52

Appendix I: Objectives, Scope, and Methodology 54

Appendix II: Expert Participation 58

Appendix III: GAO Contacts and Staff Acknowledgments 60

Table 1. Key federal agencies involved in modeling for natural hazard forecasts 11

Table 2. Selected machine learning algorithms 23

Figure 1. Basic steps of forecasting a natural hazard event 12

Figure 2. Representation of an emulator replacing part of a traditional model 26

Figure 3. A digital twin recreating a wildfire 27