The Government Accountability Office (GAO) identified the application of machine learning, a subset of artificial intelligence (AI) utilizing algorithms to discern patterns in data, in forecasting models pertaining to natural hazards. These hazards encompass severe storms, hurricanes, floods, and wildfires, often precursors to natural disasters. Notably, certain machine learning models have transitioned to operational use in routine forecasting, showcasing potential enhancements in warning times for severe storms. While a subset of machine learning applications is deemed near operational readiness, others necessitate extensive development and testing over several years.
The Government Accountability Office (GAO) recognized the prospective advantages associated with incorporating machine learning into this domain, which include:
- Efficiency Enhancement: Streamlining forecast timelines by substituting sluggish components within models, thereby mitigating operational costs.
- Enhanced Accuracy: Augmenting model precision through the comprehensive utilization of available data, incorporating non-traditional data sources inaccessible to traditional models, and generating synthetic data to address informational gaps.
- Uncertainty Mitigation: Diminishing output uncertainty by refining ensemble modeling—enhancing the amalgamation of predictions from diverse models—and optimally leveraging historical data.
The Government Accountability Office (GAO) also pinpointed challenges associated with the adoption of machine learning. These challenges encompass:
- Data Constraints:Limitations in data impede the training of machine learning models, potentially diminishing accuracy, particularly in regions with sparse weather observations, such as rural areas.
- Trust and Understanding: A lack of trust and comprehension regarding algorithms, coupled with concerns about potential biases, can instill hesitation among forecasters and other users in embracing machine learning models.
- Coordination and Collaboration Deficiencies: Limited coordination and collaboration pose impediments to the comprehensive development of certain machine learning models. Notably, forecasters have expressed a lack of opportunities to engage with researchers and articulate their specific requirements.
- Workforce and Resource Challenges: Workforce and resource gaps present hurdles in this context. High upfront costs for the development and implementation of machine learning models are noted, and some companies engaged in this domain may not possess a complete understanding of the data and phenomena they are modeling, as highlighted by academic researchers.
The Government Accountability Office (GAO) has outlined five policy options aimed at mitigating the challenges identified. These options are designed to provide guidance to policymakers, including Congress, federal and state agencies, academic and research institutions, and industry stakeholders. The status quo option represents a scenario wherein government policymakers refrain from taking additional actions beyond ongoing efforts.
Given that natural disasters, on average, result in significant casualties and substantial economic losses in the U.S. annually, effective forecasting through computer modeling is paramount for preparedness and response efforts. Such forecasting is instrumental in saving lives and safeguarding property. The integration of Artificial Intelligence (AI) in this context serves as a powerful tool, automating processes, expeditiously analyzing vast datasets, facilitating novel insights for modelers, and enhancing overall efficiency.
This report, focusing on the application of machine learning in natural hazard modeling, addresses (1) the evolving and current utilisation of machine learning for modelling severe storms, hurricanes, floods, and wildfires, along with the associated potential benefits; (2) the challenges associated with machine learning implementation; and (3) policy options to either alleviate challenges or amplify the benefits derived from the application of machine learning.
To compile this report, GAO scrutinized the use of machine learning in modeling severe storms, hurricanes, floods, and wildfires across both developmental and operational phases. Additionally, GAO conducted interviews with a diverse array of stakeholder groups, encompassing government entities, industry representatives, academic and research institutions, and professional organizations. A meeting of experts was convened in collaboration with the National Academies, and pertinent reports and scientific literature were thoroughly reviewed. The culmination of these efforts is the identification of policy options presented in this comprehensive report.
Read the full GAO report here.