News

High-resolution GNSS troposphere tomography powered by explainable deep learning - new IGiG paper in “Satellite Navigation”

05-09-2025

Researchers from IGiG UPWr lead the first high-resolution (HR) GNSS troposphere tomography, leveraging WRF outputs and a Super-Resolution GAN (SRGAN). The approach was evaluated over Poland and California, representing diverse geographic and meteorological regimes.

Key findings:

• SRGAN substantially improves tomographic fields: RMSE reductions up to 62% (Poland) and 52% (California) versus original tomography, outperforming Lanczos3, the starting resolution (input) was 50 km, output resolution was 4 km (10 times)
  
• Explainable AI highlights hot-spot regions, including western Poland and the lee side of the Transverse Ranges in California. 
  
• Validation used radiosondes and rainy-epoch detection from GPM IMERG.

Publication:
S. Haji-Aghajany, S. Izanlou, M. Tasan, W. Rohm, M. Kryza, High-resolution GNSS troposphere tomography through explainable deep learning-based downscaling framework, Satellite Navigation 6, 22 (2025), open access, DOI: 10.1186/s43020-025-00177-6. (Published: 14 Aug 2025).

Paper was also featured on NVIDIA Author blog, demonstrating use of AI in weather related research. NVIDIA