Simulating the impact of midlatitude ice cloud horizontal inhomogeneity on spaceborne passive submillimeter‐wave simulations

Abstract The advantage of spaceborne passive submillimeter wavelengths lies in their sensitivities to the size, ice water content, and shape of ice cloud hydrometeors, leading to improvement of the representation of ice cloud characteristics in numerical weather prediction models. However, plane‐parallel clouds and independent distribution assumptions are commonly adopted for simplicity in traditional forward simulations and retrieval methods, ignoring the fine‐scale structures within the footprint. This study aims to investigate the impacts of midlatitude ice clouds' horizontal inhomogeneity on spaceborne passive submillimeter radiometers and proposes an approximation method for three‐dimensional (3D) radiative transfer based on one‐dimensional (1D) radiative transfer simulations. Specifically, an idealized midlatitude 3D cloud scenario is constructed to examine the horizontal inhomogeneity effects by comparing the brightness temperature differences (BTDs; which could be defined as the brightness temperature under the clear sky minus that in the cloudy scene) at the top of atmosphere between 3D and 1D radiative transfer simulations. The idealized single‐pixel cloud model reveals that BTDs show dual‐phase sensitivity: below saturation thresholds they amplify with cloud centralization, whereas under 664 GHz saturation they increase with expanding coverage but diminish with spatial heterogeneity due to reduced effective saturation zones. Furthermore, though increased ice water path heterogeneity enhances BTDs at subsaturation frequencies (183–448 GHz) due to 3D scattering effects, it reverses at 664 GHz, where saturation causes BTD reduction with greater spatial variability, governed by cloud geometry. Then, an approximation method is proposed to approximate the 3D BTDs based on the 1D radiative transfer results. The analysis reveals that corrected result from the average‐value approximation using regression line achieves the highest approximation accuracy, with the root‐mean‐square error of 1.27–2.55 K.