Filling the gaps in global rainfall maps
Global rainfall maps are built from observations collected by multiple satellites, but they have an important limitation. The satellites that provide these measurements cannot see the whole Earth at once. As a result, some regions of these maps can have no valid observations.
Global rainfall maps currently have limitations due to gaps in satellite observations. Researchers at Chiba University, Japan, developed a method to fill these gaps using a machine learning approach. Leveraging information from neighboring time periods and surrounding regions, the team created a 3D U-Net model combined with a 3D condition encoder.
This combination allowed the model to learn precipitation patterns across space and time, learning how rainfall would be distributed in the missing areas. The approach was tested on the Global Satellite Mapping of Precipitation (GSMaP) data, which was trained using ERA5 reanalysis data. The model achieved high prediction accuracy and image quality when applied to 2024 data, outperforming other deep learning methods.
Sensitivity analysis revealed that precipitation data and infrared satellite images were the most crucial inputs for accurate reconstruction. When applied to real GSMaP maps, the model successfully reconstructed missing regions, demonstrating its potential for practical use in weather monitoring, water resource management, and disaster risk reduction.
Written by urgent.news from Phys.org's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
