Analyzing the Radiative Feedbacks Using Kernel and Neural Network Methods

We report two developments concerning the feedback analysis that is critical for understanding climate sensitivity and its difference in climate models. First, we developed a set of radiative sensitivity kernels for analyzing the radiative feedbacks with regard to the surface and top-of-atmosphere radiation budgets together. This kernel dataset is developed by conducting partial radiative perturbations with multi-years’ EAR-interim reanalysis atmospheric profiles and thus is especially suitable for analyzing the observed climate feedbacks. In addition, we explored a Neural Network (NN) method to estimate climate feedbacks. The NN method can calculate cloud and non-cloud radiative feedbacks directly while accounting for the nonlinearity effects between feedback variables. Comparisons are made between the two methods when applied to analyzing the radiative feedback in inter-annual climate variations.