A multilingual large language model agent for wildfire evacuation guidance in limited English proficiency communities
This paper introduces BEACON, an open-source multilingual mobile agent that integrates real-time fire perimeter data, weather-based risk scoring, and polygon-avoidant routing to provide personalized, language-adaptive wildfire evacuation guidance for limited English proficiency communities.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Every year, the United States faces a growing threat from wildfires. The season when these fires burn has stretched to be eighty-four days longer on average than it was in the 1970s. The smoke from these fires is so pervasive that it contributes to an estimated forty thousand deaths annually across the country. When a fire starts, the time between the first spark and the moment a resident must decide to flee is often terrifyingly short. In recent years, fast-moving flames driven by strong winds have overwhelmed warning systems that were designed for slower threats. For the roughly twenty-six million people in the U.S. who have limited English proficiency, the situation is even more dire. While over eighty percent of emergency alerts are still issued only in English, millions of residents speak neither English nor Spanish. If a person cannot read or understand an evacuation order in time, that order effectively does not exist.
To address this gap, a team of researchers has developed a new kind of digital assistant called BEACON. This is not a simple chatbot that answers questions; it is an intelligent agent designed to guide people out of danger in their own language. The system combines real-time data about where fires are burning with weather information to calculate a safe path away from the flames. It then uses advanced language technology to explain that path and provide a personalized checklist of what to take, all while switching languages instantly if the user speaks a different tongue. The goal is to ensure that when a fire approaches, every resident, regardless of their language skills, receives clear, actionable, and life-saving guidance.
The researchers built BEACON as a mobile application that acts as a bridge between complex data and a person in a panic. When a user opens the app, it immediately checks their location against live maps of active fire perimeters. These maps are drawn from data provided by WatchDuty, a service staffed by firefighters and dispatchers who track fires using official sources and radio communications. The system also pulls in hourly weather data, including wind speed, temperature, and soil moisture, to understand how fast a fire might move and in which direction. By combining the distance to the fire with these weather conditions, the app assigns a danger level to the user's location, ranging from low to critical.
One of the most significant challenges the team solved was finding a route that actually avoids the fire. Standard GPS navigation tools are designed to find the fastest path, but they often fail in emergencies because they do not treat fire zones as hard barriers. In tests, commercial GPS services frequently suggested routes that cut directly through active fire areas, even when those areas were explicitly marked. BEACON overcomes this by using a specialized routing system that treats fire perimeters as solid walls that cannot be crossed. It calculates a path that steers clear of the fire and also avoids areas where the weather conditions make the fire likely to spread rapidly. If no safe route can be found, the app does not offer a false sense of security; instead, it immediately instructs the user to call emergency services.
The heart of the system is its ability to communicate in the user's language without any setup. Unlike traditional emergency systems that require a user to select a language from a menu, BEACON detects the language of every single message the user sends. If a family member opens the app in English and then hands the phone to a parent who speaks Spanish, the agent instantly switches to Spanish for the next response. This happens turn by turn, ensuring the conversation remains natural and accessible. The agent also remembers details about the household, such as whether there are children, pets, or livestock, and uses this information to generate a customized checklist. It might remind a user to pack a pet carrier or load a trailer for animals, tailoring the advice to their specific situation.
The system is designed to be resilient, meaning it keeps working even if parts of the technology fail. If the routing service cannot find a path, or if the memory system that stores user preferences is temporarily unreachable, the app does not crash or go silent. Instead, it degrades gracefully, continuing to provide the most useful information it can while explicitly telling the user to seek help if necessary. During live demonstrations, the team showed that the system could reliably identify a user's language, generate a safe route that avoided fire zones, and provide a full situational update from the very first message. The researchers found that by combining real-time fire data, weather modeling, and an intelligent language agent, they could create a tool that offers a level of personalized, multilingual guidance that was previously unavailable to millions of residents.
This work highlights that the technology to protect vulnerable populations exists today. While federal mandates for multilingual emergency alerts are not scheduled to take full effect until 2028, the researchers demonstrated that a small team can build a functional, life-saving application using existing public data and modern artificial intelligence. The system does not replace official emergency services but acts as a critical layer of support, ensuring that language barriers do not become a barrier to survival. By making evacuation guidance accessible, personalized, and reliable, BEACON offers a new standard for how communities can prepare for and respond to the increasing threat of wildfires.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.