Sensitivity analysis of a radiation parameterization

Abstract

An assessment of the forecast sensitivity to model and radiation parameters is critical to validating climate change studies. This paper quantifies forecast first‐order sensitivity to the input to the radiation calculations in a numerical weather prediction model. The forecast aspects of interest are the temperature at the surface and near the tropopause. The model used here is a single column model, the AER Local Forecast and Assimilation (ALFA) model, which emulates a grid box of a general circulation model. It includes the NASA Ames rapid model as a broadband radiation parameterization. The sensitivities are calculated for a single time step (tendency) by the adjoint method. They reflect interactions of parameterized physical processes, since they are derived from a single iteration of the forward model and its adjoint. The sensitivities are distinguished by their sign as positive and negative responses to initial perturbation. When the atmosphere is cloudy, the surface temperature tendency shows the highest sensitivity to the single‐scattering albedo in the band centered at 0.69 μm (14,492 cm−1) in the visible window region. When the atmosphere is clear, the highest sensitivity is to the water vapor path in the continuum band, but 1 order of magnitude smaller. Also, in the clear case, the temperature tendency just below the tropopause is most sensitive to the water vapor path in the rotational absorption band. Again, this sensitivity is smaller than the sensitivity to a single‐scattering albedo in a cloud at the same altitude.