An October study transfers a cloud neural network between Fengyun satellites by translating its inputs. Nominally similar channels respond differently to the light reaching them. Polynomial conversions, fitted to physical simulations, turn FY-4B measurements into FY-4A equivalents.[1]
The reference comes from a different kind of instrument. CALIPSO sent laser pulses into the atmosphere and measured their reflections, probing the heights of clouds and airborne particles. Passive imagers instead collect reflected sunlight or emitted radiation. These are complementary views: the laser supplies information about vertical structure that a passive observation does not measure in the same way. NASA and France’s CNES ended CALIPSO’s science mission on 1 August 2023, leaving its measurements as an archive for later comparisons.[2]
Against 517,958 matched CALIPSO profiles from June 2022–March 2023, conversion reduced cloud-amount-weighted mean height root mean square error by 0.08 kilometres relative to unconverted FY-4B inputs; correlation rose from 0.841 to 0.851. The uncertainty analysis of the correlation increase ignores possible dependence between neighbouring profiles; the error reduction is descriptive.[1]
Sources & further reading
- Physics-constrained transfer learning Atmospheric Chemistry and Physics · 9 Oct 2026 · Accessed 29 Oct 2026
- First Long-Duration Lidar Satellite Mission CALIPSO Ends NASA · 26 Sep 2023 · Accessed 29 Oct 2026