Using nudging to investigate biases in a global 3 km resolution simulation with GFDL's X-SHiELD model

Abstract:

Global high resolution models are expensive to run, even on large supercomputers. This typically limits the length of simulations. For example, simulations for the DYAMOND project were carried out for just 40 days. Estimating model biases in such short free-running simulations is complicated by the fact that their meteorology diverges from reality just days after initialization, yet the simulations are not long enough to estimate robust climate statistics. Nudging a model to observed meteorology offers a compromise in that it ensures aspects of the model's state follow observations, allowing us to directly compare results of the simulation to reality; however, it still requires that the model physics diagnose many quantities such as radiative fluxes and precipitation. In addition, diagnosing the tendencies imposed on the nudged variables can inform us about how strongly the model must be corrected to match observations. Here we present results from a 40-day global 3 kilometer resolution simulation using GFDL's X-SHiELD model, where the horizontal winds, temperature, and surface pressure are nudged with a one-day timescale to those in the NCEP analysis. We use the results of this simulation to evaluate the model's biases in top of atmosphere radiative fluxes and precipitation, and results from additional nudged and free-running simulations at coarser grid scales to assess the sensitivity of the biases and nudging tendencies to resolution and presence of nudging itself. We have found significant biases in top of atmosphere upward shortwave and outgoing longwave radiation, which have informed model development and tuning efforts toward improving the representation low and high clouds by the boundary layer and microphysics parameterizations used in X-SHiELD, and also found that wind nudging tendencies in areas of steep topography are reduced when running at the fine 3 kilometer scale.