Reducing biases in coarse-resolution models by coarse-graining fine-resolution model output

Abstract:

Global fine-resolution weather and climate models can faithfully resolve atmospheric processes across a range of scales, but are expensive to run, even on large supercomputers, while coarser-resolution models can have substantial biases in their simulated states. Thus, methods which can translate atmospheric states between fine- and coarse-resolution models are useful to leverage the relative strengths of each. We develop a method to coarse-grain fine resolution model states in order to both instantiate a coarse-resolution model and to compute the apparent sources of heat and moisture that are not resolved in the coarse model. We apply this approach using a non-hydrostatic atmospheric model with hybrid terrain-following coordinates. We demonstrate that the model’s sloping, terrain-following coordinates require coarse-graining along pressure levels rather than model levels for unbiased coarse-grained states over topography. We also evaluate the structure of the apparent sources that are not resolved by a coarse-resolution model integrated over the same grid as a fine-resolution model that has been coarse-grained, showing the importance of these sources for resolving deep convection and the hydrologic cycle. Finally, we show that because the coarse-graining apparent sources can reasonably represent the differences in tendencies between the different model resolutions, they allow for nudging of the coarse-resolution model to reduce its states’ biases relative to the high-resolution model.