����������ý�Ļ���Ʒ

Professor Amir Behzadan joined an “invitation-only” group of experts from academia, government and the private sector to discuss the potential for artificial intelligence (AI) to transform weather prediction at the AI for Numerical Weather Prediction (AI4NWP) workshop in Washington, D.C. The event, held on May 6, was jointly hosted by the White House Office of Science and Technology Policy and the National Oceanic and Atmospheric Administration (NOAA). 

“We are witnessing an upward trend in the number and frequency of major climate events across the country and around the world,” said Behzadan, of ����������ý�Ļ���Ʒ’s Department of Civil, Environmental and Architectural Engineering and a faculty research fellow of the at the . “Storms are becoming more severe and moving far more inland. Tornadoes, intense rainfalls and flash floods are happening in places that were once deemed safe.” 

Behzaden said that early AI models demonstrate improved skill in forecasting events like hurricanes, winter storms and heat waves. These models are potentially valuable tools for alerting residents earlier, ultimately saving lives and property, he said.

During the event, Behzadan served on a scientific panel on building and maintaining trust in AI systems, where he shared his views on how human trust in AI-generated weather predictions can be formed, calibrated, measured and used to inform decision-making. 

Event participants, including Biden-Harris administration leaders, covered a wide range of topics pertaining to the benefits and challenges of AI for weather prediction, building and maintaining trust in AI systems and exploring scientific unknowns that pose the biggest challenges. 

Through the workshop, NOAA strengthened its commitment to working with partners from academia and the private sector to build the necessary infrastructure for trustworthy weather AI models that support decision-making. 

����������ý�Ļ���Ʒ sat down with Professor Behzadan to ask a few questions about the use of AI in weather prediction. 

How can AI be used to improve weather prediction?

AI can enhance weather prediction in several ways. For example, AI algorithms can analyze large volumes of data from various sources, including weather satellites, ground stations, in-situ sensors and crowdsourced information. Machine learning (ML) techniques can be then applied to identify patterns and correlations in this data, leading to more accurate forecasts. 

Neural networks, a type of ML model inspired by the structure and function of the human brain, can analyze complex atmospheric dynamics and predict weather phenomena that are indicative of storms, hurricanes or droughts. 

AI can also improve our understanding of cloud formations or atmospheric pressure gradients, allowing meteorologists to generate more reliable forecasts, particularly in regions with limited data. AI models can analyze and learn from the available data, discern patterns and relationships within the data and adapt and improve over time as they receive more data. These capabilities allow the model to adapt to evolving weather patterns, increase the accuracy of long-term predictions and enhance our ability to mitigate risks due to climate disasters.

How can scientists increase public trust in AI for weather prediction?

To build and strengthen trust in AI for weather prediction, transparency and communication are key. For example, providing clear explanations of how AI algorithms work and their limitations can help build understanding and confidence among end users. 

Also, for an AI model to be fully adopted, it must demonstrate the underlying uncertainty and accuracy by comparing its forecasts with traditional methods. Collaborating with meteorologists and incorporating their expertise can reassure users that human oversight is integral to system design and deployment. Continuous validation against observed weather data and transparent communication of AI predictions’ successes and failures will further build trust in its capabilities. 

Should we trust AI for weather forecasting?

As for whether humans should trust AI for weather prediction, this is a question of choice and weighing the evidence; after all, "trust" is a subjective willingness of individuals or communities to rely on each other or on a technology. 

While emerging AI systems have the potential to significantly improve forecast accuracy and lead to better preparedness for extreme weather events, it’s essential to recognize their limitations and the possibility of errors. 

To sustain user trust, potential users should have ample opportunities to safely interact with AI systems, with human oversight and interpretation remaining central to AI integration, especially in complex or unprecedented situations, such as rapidly intensifying storms, flash floods and wind storms. This is mainly because in those situations, the variability and uncertainty of data increases, potentially leading to unstable AI outcomes that need to be more closely scrutinized by a human expert. Ultimately, as with any high-stake application, a balanced approach that leverages both AI technology and human expertise is likely the most effective strategy for reliable weather prediction.