New adaptive localization methods advance all-sky infrared radiance assimilation and improve tropical cyclone forecasts
image: (A–C) temperature anomaly relative to a reference profile at the radius of 500 km, (D–F) specific humidity, (G–I) radial wind, and (J–L) tangential wind, for experiments (A, D, G, J) GGF-FS, (B, E, H, K) GGF-VS, and (C, F, I, L) GGF-VLS. The “×”, “+”, and black lines denote the locations of warm core and RMW, and ranges of gale-force wind, respectively.
Credit: ©Science China Press
Effective assimilation of satellite infrared (IR) radiances is essential for improving numerical weather prediction, especially for high-impact weather systems such as tropical cyclones (TCs). However, under cloudy conditions, IR radiances become highly nonlinear and exhibit non-Gaussian error distributions, making all-sky assimilation a longstanding challenge. To address this issue, a research team led by Prof. Lili Lei from the State Key Laboratory of Severe Weather Meteorological Science and Technology at Nanjing University has developed two new adaptive localization methods for all-sky IR radiance assimilation.
The researchers note that the previously proposed Global Group Filter (GGF-FS) can estimate adaptive localization scales, but does not fully account for vertical variations in cloud processes and their impacts on radiances. To address this limitation, they propose two refined methods. The first, GGF-VS, allows localization scales to vary with peak height, producing more consistent vertical correlation structures between radiances and model state variables. The second, GGF-VLS, further extends this approach by directly incorporating radiance brightness temperature and cloud-top height, enabling localization parameters to respond more flexibly to changing cloud conditions.
Cycling assimilation experiments show that both new methods reduce prior errors in model state variables compared with GGF-FS, with GGF-VLS delivering the best overall performance. The forecast improvements are especially notable for TCs. Compared to GGF-FS, GGF-VS obtains reduced forecast errors and smaller ensemble spread of minimum sea-level pressure and maximum wind speed; GGF-VLS further improves TC intensity forecasts. These two new methods also produce more physically consistent TC structures, characterized by a stronger warm core, increased low-level moisture, and enhanced primary and secondary circulations. These structures are favorable for TC intensification.
This study demonstrates that advanced adaptive localization techniques can better extract cloud-sensitive information from all-sky IR radiances, which has been a major bottleneck in all-sky radiance assimilation. These findings provide a new technical pathway for improving forecasts of TCs and other hazardous weather systems, while supporting the development of next-generation high-resolution numerical weather prediction systems.
See the article:
You F, Lei L, Zhou L, Tan Z M, Zhang Y. 2026. Adaptive localizations for all-sky infrared radiances in an ensemble Kalman filter. Science China Earth Sciences, 69(3): 992–1003, https://doi.org/10.1007/s11430-025-1806-3