Artificial intelligence can enhance our ability to manage natural disasters. However, understanding and addressing its limitations is required to realize its benefits. Here, we argue that interdisciplinary, multistakeholder, and international collaboration is needed for developing standards that facilitate its implementation.Acute events of natural origin (e.g., atmospheric, hydrologic, geophysical, oceanographic, or biologic) can result in disruption and devastation to society, nature, and beyond 1,2 . Such events, which disproportionately impact certain regions (e.g., least developed countries 3 ) and populations (e.g., women and children 4 ), are often referred to as natural disasters by experts in the geoscience and disaster risk reduction communities, as reflected in the scientific literature and in Sustainable Development Goals 11.5 and 13.1.Recently, interest has grown in leveraging innovative technologies such as artificial intelligence (AI) to bolster natural disaster management 5 . In many fields, such as medicine and finance, AI has gained traction due to advances in algorithms, a growth in computational power, and the availability of large data sets. Within natural disaster management, it is hoped that such technologies can also be a boon: capitalizing on a wealth of geospatial data to strengthen our understanding of natural disasters, the timeliness of detections, the accuracy and lead times of forecasts, and the effectiveness of emergency communications.This Comment looks at successes and limitations of data collection methods and AI development for natural disaster management. It then examines the challenges and solutions surrounding AI implementation. It is shown that, although AI has the promise to enhance our ability to manage natural disasters, its effective adoption depends on collaborative efforts to address these challenges.
Successes and limitations to dataThe foundation of any AI-based approach is high-quality data. A recent success is the emergence of new (and novel use of traditional) data collection methods. For example, sensor networks now help us to gather data from topographically complex regions, which are otherwise difficult to monitor, at high spatiotemporal resolutions. Such networks have proven successful for flash