A smarter eye in orbit can now tell steady rain from snow and hail far more reliably

From orbit, telling one kind of falling water from another is surprisingly hard — and it matters, because those labels feed weather models, flood planning and the long-term climate record. The Global Precipitation Measurement satellite's radar has relied on fixed rules that can mislabel snow, heavy convective rain and hail in tricky cases. A team from Southern Illinois University Edwardsville, the University of Oklahoma and the National Severe Storms Laboratory trained a deep-learning model on the same radar signals, checked against detailed ground-radar labels. Its overall classification score rose from 0.66 to 0.87 on their benchmark, it stopped confusing snow with convective storms, and it placed hail cores more precisely inside storm cells. It is not in operational use yet, but it points to cleaner, more consistent rainfall records from space.

SDG 9 · Industry, innovation and infrastructure — Mechanism: tech innovation