Abstract

A neural network model is developed to measure shortwave radiative feedbacks. This model is trained with the atmospheric data of the fifth-generation European Centre for Medium-Range Weather Forecasts Reanalysis to emulate the atmospheric shortwave radiation fluxes, which thus allows direct and computationally efficient evaluation of the radiative feedbacks following the straightforward perturbation method. This model is tested for computing the radiative feedbacks of surface albedo, water vapor and cloud in the Arctic climate change. In both idealized and real climate change scenarios, the model can accurately quantify the albedo feedback compared to the truth computed by a radiative transfer model, while the linear kernel method severely overestimates the albedo feedback during Arctic warming. The neural network model can capture the nonlinearity in the surface albedo feedback and its dependency on the cloud amount.