Assessing parametrization uncertainty associated with horizontal resolution in numerical weather prediction models

Area of Science:

• Atmospheric Science
• Meteorology
• Computational Fluid Dynamics

Background:

• Ensemble weather prediction systems require accurate representation of model uncertainty.
• Stochastic algorithms are used to model sub-grid-scale variability, but their accuracy is often unproven.
• Existing methods may misrepresent the true sources of model uncertainty.

Purpose of the Study:

• To quantify uncertainty in physical parameterization tendencies within the ECMWF Integrated Forecasting System.
• To investigate the relationship between model error and horizontal resolution deficiency.
• To develop an improved stochastic parameterization scheme.

Main Methods:

• Compared high-resolution "truth" forecasts with low-resolution "target" forecasts.
• Coarse-grained forecasts to common spatial and temporal resolutions.
• Defined model error and examined its probability distribution function.

Main Results:

• Temperature tendency errors from convection and phase changes follow a Poisson process (variance proportional to mean).
• Radiation temperature tendency errors exhibit a different mean-variance relationship.
• The ECMWF's current scheme, assuming standard deviation proportional to mean, may be suboptimal.

Conclusions:

• The findings support the applicability of the Craig and Cohen model to parametrized convection.
• A prototype scheme based on variance proportional to mean improved forecast skill.
• The developed scheme shows potential for enhancing ensemble weather prediction accuracy.