NWS AI-Generated Weather Predictions Are Making Up New Towns

How’s the weather looking in Whata Bod, Idaho? Only AI can tell you. The National Weather Service recently posted a weather map on social media that was AI-generated and included several entirely made-up town names.

In a post that has since been deleted and replaced by a seemingly still AI-generated but less error-riddled version, the official National Weather Service X account for Missoula, Montana, shared a map that showed a 10% chance of high winds in “Orangeotild” and lesser gusts in “Whata Bod,” two towns that do not exist. The map also included geographical errors, according to the Washington Post, the result of NWS deciding to leave the generation of this information up to AI.

It’s not the first time this has happened. In November, the NWS office in Rapid City, South Dakota, posted a wind map to X that included misspelled town names. That image, which is still live on the account, included a Google Gemini watermark in the lower right-hand corner, indicating the image was generated with Google’s AI model.

A spokesperson for NWS told Gizmodo that the use of AI for public-facing information like these weather maps is uncommon, but it isn’t prohibited. “Recently, a local office used AI to create a base map to display forecast information, however the map inadvertently displayed illegible city names,” the spokesperson said in an email. “The map was quickly corrected and updated social media posts were distributed.”

Last August, the General Services Administration announced an agreement with Google to allow federal agencies to use Gemini for Government, Google’s enterprise AI suite designed to comply with government requirements. Contained within that collection of tools is Gemini’s image generation capabilities. Presumably, NWS offices have been using these tools to experiment with information generation. The office’s parent agency, the National Oceanic and Atmospheric Administration, recently announced an embrace of AI for weather prediction models, which included a partnership with Google DeepMind.

There is some promise to AI-enhanced weather prediction, including a recent paper published in Nature that showed some AI models are capable of providing accurate 10-day forecasts at smaller scales than traditional models. But even these models call for a human forecaster to confirm the information. That is getting harder to do over at NOAA and NWS, where the Trump administration has announced plans to cut 17% of staff. Getting the little things wrong is a great way to erode public trust, especially when the information is coming from official government accounts. After all, who cares how accurate your forecast is if it’s a forecast for a made-up town?