Spectral Decomposition of Cloud Radiative Feedbacks: A Kernel Approach

The concepts of broadband cloud radiative feedback have been extensively used in the climate model evaluation as well as in the study of climate change. Due to compensating biases among different spectral bands, the simulated broadband cloud radiative feedback can be seemingly correct but for the wrong reasons. From this perspective, understanding spectral decomposition of the cloud radiative feedback can offer us more insights about the cloud radiative feedbacks.

Recently a kernel method has been introduced to infer observed broadband cloud radiative feedbacks using satellite observations without the use of any off-line radiative transfer calculation. We extended the method to model simulations and it can obtain a broadband cloud radiative kernel similar to the one derived using an off-line radiative transfer model as in Zelinka et al (2012). The advantage here is that this method can provide cloud radiative feedback at each spectral band used in the radiation scheme of the climate model. Using the CESM as an example, we used this kernel method to derive cloud radiative feedbacks for the +2K SST run, for the 2xCO₂ run, and for the 4xCO₂ run and compare them with the counterparts derived using the adjust method as in Soden et al. (2008). The broadband cloud radiative feedbacks derived from two methods are similar to each other, for both the magnitude and spatial distribution. However, the results from the adjust method have different spectral decompositions compared to the results from the kernel method. The spectral decomposition of cloud feedbacks from the kernel method changes little among three runs examined here, while the spectral decompositions of the results from the adjust method have obvious change from the +2K SST run to 2xCO2 and 4xCO2 runs. The implication for such spectral decomposition is then discussed and the derived short-term cloud radiative feedback from the CESM simulation and from observations are compared with each other.