Researchers have developed the first systematic precipitation retrieval algorithm for China's Fengyun-3G (FY-3G) satellite, converting passive microwave observations into an operational product that describes surface rainfall and vertical hydrometeor profiles. The method, published on August 11, 2026, in the Journal of Remote Sensing, combines Bayesian retrieval with a physically optimized prior database to produce estimates consistent with established Global Precipitation Monitoring (GPM) products while improving cloud liquid water representation.
Accurate precipitation measurements are crucial for weather forecasting, climate research, water management, and disaster preparedness, yet rainfall remains a major source of uncertainty in environmental models. Passive microwave satellites can monitor precipitation over oceans, mountains, and areas poorly covered by ground instruments, providing wider coverage than radar. FY-3G carries the advanced MWRI for the rainfall mission (MWRI-RM) radiometer with 26 channels spanning 10.65–183 GHz, but these observations lacked a dedicated algorithm for comprehensive operational retrieval.
The research team from the University of Science and Technology of China, the National Satellite Meteorological Center, the Meteorological Observation Centre of the China Meteorological Administration, and collaborators built a Bayesian retrieval around a prior database linking hydrometeor profiles and atmospheric states with microwave brightness temperatures. They combined GMI observations, GPROF profiles, ERA5 variables, and radiative-transfer calculations, then applied a one-dimensional variational method to adjust cloud liquid water, rainwater, ice water, and surface emissivity for physical consistency. The database was applied to FY-3G MWRI-RM observations with separate retrievals for ocean, coast, low-elevation land, and high-elevation regions.
The algorithm generated a multiple-hydrometeor profile (MHP) product that agreed well with two established GPM products—the Goddard profiling algorithm (GPROF) and the combined radar-radiometer algorithm (CMB)—in surface precipitation rate, rainwater path, and ice water path patterns. Surface precipitation estimates generally showed relative mean biases within 14% over ocean and land. A key advantage emerged for cloud liquid water: MHP showed a markedly stronger relationship between cloud liquid water path and precipitation intensity than either comparison product, indicating a more physically realistic cloud–rain linkage. The method also retrieves vertical profiles of cloud liquid water, rainwater, and ice water, offering more information on storm microphysics than rainfall-only products.
Calibration tests showed strong agreement between simulated and observed MWRI-RM brightness temperatures. Under clear skies, Pearson correlations exceeded 0.94 across all channels, mean biases stayed within ±0.4 K, and RMSEs were below 2.5 K over land and 2.0 K for most ocean channels. In a typhoon case, all three products captured the eyewall and three spiral rainbands; MHP achieved a cloud-liquid-water-to-rainfall spatial correlation of 0.84, versus 0.72 for GPROF and 0.52 for CMB. In a land rainfall case, correlations were 0.84, 0.76, and 0.47. Validation against 2,486 rain gauges showed that extending accumulation from 10 to 60 minutes increased correlation from 0.455 to 0.542 and reduced Root Mean Square Error from 1.743 to 1.341 mm/h. Remaining biases reflected heavy-rain underestimation and uncertainties related to surface type and ice-particle assumptions. Oceanic rainwater path was underestimated by 27% versus GPROF and 19% versus CMB.
The authors note that "the MHP product exhibits good consistency with GPROF and CMB," adding that its precipitation-related variables can reveal storm microphysics and support improved model parameterization. The team also sees potential for extending the retrieval framework beyond FY-3G to other Fengyun microwave imagers and future satellite missions. The prototype could strengthen operational precipitation monitoring over China and globally, especially where ground networks are sparse, and may support typhoon and flood monitoring, cloud–precipitation research, climate-model evaluation, and improved microphysical parameterization. Further refinement of heavy-rain representation, high-elevation retrievals, and ice-particle models could improve accuracy and broaden future operational use. The study is available at DOI: 10.34133/remotesensing.1057.
