Google expands AI tools to anticipate natural disasters

Google is using artificial intelligence to predict floods, track wildfires and improve alerts for earthquakes and extreme weather. The company says its goal is to give communities useful information before danger arrives, not afterward.

The announcement brings together a decade of work on forecasting and emergency response tools. The idea is simple: know where a disaster could occur, how much warning there may be and which areas need to prepare first.

Floods: up to seven days’ warning

The Flood Hub system already provides forecasts in flood-prone areas covering 2 billion people in more than 150 countries. Forecasts for river flooding can arrive up to seven days in advance.

Google has also developed a model for flash floods in urban areas, which can develop very quickly. In these cases, the system can provide up to 24 hours of warning. To train it, the company created a dataset with Groundsource based on 20 years of public reports and published both the data and its hydrological framework.

In practice, these forecasts can help emergency services evacuate people earlier, protect infrastructure and deliver financial assistance to families before the water rises. In Nigeria and Bangladesh, organizations such as GiveDirectly and the International Rescue Committee have already used them to activate this kind of early support.

Greater accuracy for hurricanes and wildfires

For extreme events such as cyclones, the WeatherNext 2 model generates hourly forecasts for the entire planet in minutes. It calculates variables including wind speed and direction, precipitation and atmospheric pressure.

Google says that during the 2025 hurricane season, the model predicted cyclone paths and intensity several days in advance. During Hurricane Melissa, the United States National Hurricane Center used it to anticipate the impact on Jamaica five days ahead, allowing the country’s weather services to warn the public.

For wildfires, Google uses satellite imagery to track their boundaries in Search and Maps. This coverage now reaches 34 countries, including seven added this year.

The company is also involved in FireSat, a satellite network developed with Earth Fire Alliance and Muon Space. If the planned constellation is completed, with more than 50 satellites, it could detect fires as small as 5 by 5 meters anywhere in the world and update the information every 20 minutes. This is a planned capability, not coverage that is already available.

From forecasts to action

Predicting a disaster is not enough if the information does not arrive in time or does not explain what to do. That is why Google combines its climate and geospatial models in the Google Earth AI collection, which can analyze imagery, population data and environmental conditions to answer more complex questions.

For example, an organization could estimate where a hurricane will make landfall, which communities are most vulnerable and what measures they should take. The company also uses aerial and satellite imagery to study the reflectivity of buildings, a factor that can help cities decide where to install cool roofs to reduce surface heat.

For the general public, these tools appear in products people already use:

• Search and Maps SOS alerts bring together information from authorities and trusted media outlets.
• Public Alerts distributes official warnings in collaboration with organizations in more than 90 countries.
• Extreme heat alerts provide warnings and recommendations in more than 100 countries.
• Android’s earthquake alert system can warn users before vibrations reach them.
• Google Maps displays updated air quality data in more than 30 countries.

Last year, Google helped connect people with information about crises more than 10 million times a day, on average. The figure shows the potential reach of these tools, although their usefulness depends on accurate data, action from authorities and communities being able to receive the help they need.

The next stage is not just about generating better predictions. It is about turning them into concrete decisions: evacuating, protecting a home, sending money or closing an area before the danger is obvious. AI can extend the time available to react, but local preparation and response still determine how much damage is avoided.