Optimizing an atmospheric general circulation model for weather prediction (10km) and climate (100km) resolutions.

At the GMAO we have developed a multi-scale modeling and assimilation system (GEOS-5) that is used at very different resolutions for numerical weather prediction (NWP), reanalysis, seasonal forecasting, and longer-term climate applications. We have tried to develop tunings of the AGCM parameterizations that allow us to use the model in this range of resolutions (Molod et al, 2015). In this presentation, we focus on several parameters that (in our system, at least) that appear to be most effectively tuned for achieving the desired seamlessness: (1) the critical relative humidity used in the precipitation PDF scheme, (2) the maximum allowed entrainment in the deep convective parameterization, (3) the timescales for shallow and deep convection, and (4) the physics/dynamics coupling timestep itself. Although some of these parameters require only choosing optimum values at each resolution, we find that those involved with the balance between resolved and parameterized deep convection must be varied in a flow-dependent way. In the presentation we describe a scheme to do this that uses low-level moisture in the free atmosphere to identify regions with active resolved convection. We will also argue, based on the sensitivity we see to physics/dynamics coupling timestep, that it may be necessary to bring water vapor transport and simple condensation and heating into the dynamics, since their time scales are comparable to those of resolved vertical motions.

We will also discuss the tuning procedure, and in particular, the metrics used to select parameters. Using high-resolution observations of global brightness temperature, precipitation, and water vapor we explore the characteristics of weather produced within a well-tuned climate simulation. Re-forecasting experiments can be used to explore the ideal tuning for producing the best anomaly correlation and root mean squared (RMS) error statistics for the basic state variables in a five day forecast validated against high resolution analyses with GEOS-5. The optimal NWP tuning configuration can then be used to produce a typical AMIP climate simulation to examine the differences in the climate produced from the earlier "well-tuned" climate simulation.