A new study celebrates artificial intelligence for translating National Weather Service forecasts into multiple languages, claiming accuracy above 95 percent and dramatically faster turnaround times. Yet the history of weather communication shows that translation failures often stem from culture, context, and human interpretation—areas where AI still faces significant challenges.
CHAMPAIGN, Ill. — Nearly 69 million people in the United States speak a language other than English at home, yet weather warnings have long been issued almost exclusively in English. A new study documents how the National Weather Service is using artificial intelligence to change that, developing a comprehensive translation program that delivers life-saving forecasts and alerts in Spanish, Chinese, Vietnamese, Samoan, French and other languages.
The new multidisciplinary study was led by University of Illinois Urbana-Champaign climate, meteorology and atmospheric sciences professor Joseph Trujillo-Falcón and NWS scientist Monica Bozeman and is published in the journal Artificial Intelligence for the Earth Systems.
“Translating weather forecasts has always been a critical, time-consuming task, often added to the plates of bilingual forecasters managing full operational responsibilities, but it provides information for the 68.8 million people who do not speak English at home,” said Trujillo-Falcón, who also leads the ALERTAS lab in partnership with the department of communication at Illinois.
The urgency for accurate and culturally sensitive translation of weather forecasts in the U.S. intensified in the 1970s and ’80s. In 1987, an F4 tornado struck the town of Saragosa, Texas, and 151 of the town’s 183 residents suffered injuries or casualties. The town, which had many Spanish-only speakers, had only one Spanish-language radio station, which did provide the NWS warning; however, the literal translation of the English message failed because the word “warning” has no Spanish equivalent.
The new program partners the NWS with the AI translation platform LILT, whose patented training process enables large language models to adapt neural machine translation tools for weather terminology and messaging.
“Before this program, bilingual NWS forecasters spent up to an hour manually translating a single product. The AI system now completes the same translation in 5 to 7 minutes for some NWS products, with accuracy scores above 95%,” Trujillo-Falcón said. “More than 30 NWS offices from across the country are using it and, starting with the 2025 hurricane season, the National Hurricane Center officially began issuing AI-translated Spanish advisories for all hurricanes in the Pacific and Atlantic coasts, providing lifesaving information to the U.S. and even Latin America.”
Beyond the operational aspect of the new program, the team is also very interested in how it will benefit the American people and economy.
“Tourists from across the world come back time and time again, benefiting local businesses, and we can provide them with life-saving information,” Trujillo-Falcón said. “And with the World Cup approaching, the NWS has made an agency-wide effort to provide decision-support services in the event of dangerous storms approaching the stadiums.”
Trujillo-Falcón emphasized the importance of the multidisciplinary focus the U. of I. brings to the program. For example, Illinois geography and geographic information science graduate student and study co-author Liam Llewellyn has overlaid geographic information systems, weather and census data to identify which language translations are needed for all 122 NWS offices across the country.
“We plan to expand these collaborations in the future here at Illinois,” Trujillo-Falcón said. “For example, I have been working with Spanish and Portuguese professor Salvatore Callesano, who is helping with the social science evaluation of how the public responds to AI tools.”
Researchers from the University of North Dakota, Pace University, Colorado State University and the University of Oklahoma also contributed to this study.
