Abstract
We introduce TropDS, a downscaling framework that combines physical residual modeling with a U-Net AI refinement to produce high-resolution tropospheric delays from coarse global grids. Global ZHD & ZWD accuracies reach 90.97% & 85.76%, with RMSE < 2 mm and inference taking ~0.04 s per epoch. It also remarkably boosts early-stage forecast precision when coupled with AI weather models like Pangu-Weather, GraphCast and FengWu. A robust solution for precise atmospheric corrections in geodesy.
Related Projects
Related Links
BibTeX
Total Visits: