Beyond Bias: Systematic Analysis of 3D Cloud Radiative Effects on Surface Irradiance and Their Dependence on Cloud Properties, Albedo, and Solar Angle

Neglecting three-dimensional (3D) cloud–radiation interactions in weather and climate models can significantly distort surface irradiance estimates. Understanding how these effects depend on factors such as cloud properties, surface albedo and solar angle is the central goal of the DFG research unit C3SAR. Accurate representation of surface irradiance is becoming increasingly important as numerical weather prediction (NWP) and cloud-resolving models advance toward ever higher spatial resolutions.

Traditionally, analyses have focused on mean bias, which is computationally efficient - especially when using Monte Carlo methods - since pixel-to-pixel noise cancels out. This allows large-scale studies of realistic, high-resolution cloud scenes. However, relying solely on mean bias can obscure important 3D effects, such as sharper and displaced cloud shadows (Gristey, 2019).

We present a method to go beyond bias by computing pixel-wise root-mean-square differences (RMSD) between 1D and 3D simulations, as well as metrics of surface flux variability. Using novel statistical techniques applicable to general Monte Carlo simulations, this approach retains the computational efficiency of traditional bias calculations. This enables large-scale analysis not only of mean bias but also of pixel-wise differences and variability across statistically representative ensembles of high-resolution surface irradiance fields derived from physically consistent radiative transfer simulations.

We apply this method to systematically investigate how 3D bias, local deviations, and variability depend on solar angle and cloud properties, using MYSTIC (Mayer, 2009) radiative transfer simulations on large-eddy simulation (LES) datasets at kilometer-scale domains with 10 m horizontal resolution. Furthermore, by coarsening the LES fields, we assess how cloud resolution influences 3D effects and determine the resolutions required to capture them accurately. Comparing fully 3D radiative transfer with independent-column approximations (ICA), we quantify the limitations of ICA in representing 3D cloud–radiation interactions, providing valuable guidance for next-generation cloud-resolving models.