Skip to main content
Article
100%
HD

Does your GPS sometimes seem a bit lost? The culprit might be the very air we breathe. Yes, the atmosphere, with its humidity and pressure, can 'stretch' satellite signals, causing positioning errors. Until now, correcting this effect relied on global weather models, a bit slow like a forecast from an older computer.

But now comes Pangu-Weather, an AI trained on weather data that promises to be a champion. A new study tested this model on hundreds of monitoring stations across China, comparing it to the traditional GFS model. The result? Pangu-Weather proves surprisingly effective, especially in the complex regions of southern China, where the climate is a real puzzle.

What makes it special is its ability to make much more frequent forecasts, every hour instead of every day, and to be more stable over long periods. In practice, it's like having an always updated and accurate weather report, even when traveling for days. This discovery is good news for all tech and geolocation enthusiasts, as it paves the way for increasingly reliable GPS, even in difficult weather conditions.

In summary, AI isn't just replacing our tools; it's making them smarter and more responsive to nature's challenges.

Bonus
100%
HD

The study is based on an innovative approach called 'ensemble forecasting', which combines multiple predictions from the Pangu-Weather model starting from slightly different initial conditions. This method, borrowed from traditional meteorology, allows for estimating not only the most likely value of the tropospheric delay (ZTD, Zenith Total Delay) but also its uncertainty. The validation data come from 250 GNSS stations of the CMONOC (Crustal Movement Observation Network of China) network in 2020.

Key results show that Pangu-Weather outperforms the reference GFS (Global Forecast System) model in many regions, particularly in southern China, characterized by high humidity and mountainous terrain. However, in arid, high-altitude areas (between 2 and 4 km), the GFS model maintains a slight advantage for short-term forecasts (up to 9 hours). An interesting finding is that Pangu-Weather's error grows more slowly over time: after 10 days (240 hours), its RMSE (Root Mean Square Error) is lower than that of the empirical GPT3 model, demonstrating greater stability for long-range forecasts. This feature is crucial for real-time positioning applications that require stability over extended periods.

Ask AI

Explore key aspects of this story with 3 quick insights

Source

Authors:
Shuangping LiBin ZhangHaohang BiLiangke HuangBo ShiQingsong Ai
Date Published:
Original Article www.mdpi.com
Fact-check verified on 29/08/2026
Read our Editorial Guidelines & AI Ethics Policy
Creative Commons License CC BY-NC-SA 4.0

Keep exploring

Discover the full archive of science bites or head back to home

Riconnessione al server... Aggiornamento della sessione in corso
Connessione interrotta Impossibile stabilire il collegamento
Sessione aggiornata Ricarica per visualizzare i contenuti