Stochastic approaches within current PBL parameterization schemes

With the acknowledgement that model deficiency is an important limitation to deterministic mesoscale forecast skill, interest in stochastic approaches to sub-grid scale parameterization schemes is growing. PBL parameterization is one component of mesoscale models that is known to suffer from a lack of variability, as compared to observations. They are thus a reasonable candidate for stochastic perturbations aimed at producing appropriate responses in mesoscale forecasts. But the sensitivity of a highly-diffusive system such as a PBL scheme to stochastic perturbations has not been explored and in fact is not guaranteed. One reasonable, if intermediate, approach is to add stochastic terms to existing deterministic parameterization schemes.

This work focuses on measuring PBL parameterization sensitivity to different classes of stochastic perturbations. A column model with WRF land-surface, surface-layer, and PBL parameterization schemes is used as a testbed. Autoregressive models are formulated with both real and synthetic observational statistics, and applied to key parameters in two different PBL parameterizations. Sensitivity is quantified with large ensembles by computing finite-size Lyapunov exponents. Instantaneous error-growth vectors reveal the structures associated with those instabilities. The results have implications for data assimilation of near-surface observations and ensemble forecasting of sensible weather.