A Comparison of the Impact of Using Downscaled and Convection-Permitting Analyses on Model Behavior

Forecast error in numerical weather prediction models can be attributed to two main sources: initial condition error and model error. Convection-permitting forecast systems are typically initialized with analyses downscaled from a coarser resolution. These forecasts, when initialized from downscaled analyses, require longer model spin-up time due to the lack of fine-scale structures present at initial time and/or the different model used to generate the analyses. Here, we examine the impact of using analyses at the same resolution as the convection-permitting forecast model. During each integration, forecast models in the system often tend to drift towards their own model climate, often attributed to misrepresentation in the model physics. We investigate how the convection-permitting model spins up to its model climate given a particular representation of the initial state specified using either downscaled or convection-permitting analyses. Using an ‘initial tendency’ method [Klinker and Sardeshmukh (1992); Rodwell and Palmer (2007)], we further decompose the model spin-up process into its dynamical and physical components. We use the advanced Weather Research and Forecasting (WRF-ARW) model supported within the Data Assimilation Research Testbed (DART) for both the analysis and forecast components. We will present results from a series of retrospective springtime convective forecasts over the central United States.