.jpg)
Accurate prediction of intense precipitation remains one of the greatest challenges in operational meteorology. Traditionally, numerical weather prediction models assimilate GNSS satellite observations to capture atmospheric water vapor through the zenith wet delay (ZWD), a magnitude directly related to columnar humidity. However, the latest generation of machine learning models have barely exploited this data source. A recent study has integrated ZWD obtained from GNSS into Aurora, a foundational weather prediction model, for the first time, achieving significant improvements in forecasts of accumulated rainfall over six hours, especially in extreme events. Results show an 8.8% increase in the Equitable Threat Score for the 99th percentile and a more realistic precipitation power spectrum at synoptic and planetary scales. This advance demonstrates that direct GNSS observations contain valuable information that artificial intelligence models can leverage to anticipate high-impact phenomena.
For a company like Q2BSTUDIO, specialized in custom applications and artificial intelligence, this case illustrates how integrating heterogeneous data with machine learning techniques can transform critical sectors. Developing models that combine satellite signals with deep learning architectures requires robust and scalable software, areas in which we offer AWS and Azure cloud services and cybersecurity solutions to protect data flow. Likewise, the visualization and analysis of these results can be enriched with business intelligence tools such as Power BI, and process automation through AI agents streamlines continuous model validation. Ultimately, collaboration between data science and custom software development is key to bringing innovations like this to operational environments, improving response capacity to extreme weather events.
