<p>All-sky infrared radiance data assimilation has been challenging, due to nonlinear features and complex error statistics. Covariance localization, which is essential for an ensemble Kalman filter applied to high-dimensional geophysical systems, needs to consider the characteristics of all-sky radiances. To effectively extract information from all-sky radiances that are sensitive to clouds, two adaptive localizations are proposed here. Compared to the previously proposed global group filter (GGF-FS) that adaptively estimates localization lengthscales based on sample correlations, one simultaneously takes the vertical location and localization lengthscales into account (GGF-VS), and the other further considers the impacts of cloud top pressure and brightness temperature on the adaptive localization parameters (GGF-VLS). Both GGF-VS and GGF-VLS have broader vertical localization lengthscales and better capture the TC structure than GGF-FS, and GGF-VLS has larger variations of localization parameters than GGF-VS. Data assimilation experiments for assimilating all-sky infrared radiances confirm that GGF-VS has smaller errors of state variables than GGF-FS, and GGF-VLS further reduces the errors compared to GGF-FS and GGF-VS. Moreover, the more detailed adaptive localization parameters are beneficial for the TC intensity forecast. GGF-VS and GGF-VLS produce smaller errors and ensemble spread for the minimum sea level pressure and maximum wind speed than GGF-FS. GGF-VLS further improves the TC intensity forecast than GGF-VS, and both produce more coherent TC structures that are favored for intensified TC than GGF-FS, including the stronger warm core, more moisture at low levels, and enhanced primary and secondary circulations.</p>