Use of the Stochastic-dynamic Approach in a Single Dynamic-Core Storm-Scale Ensemble for Improved Spread and Reliability of QPF and Surface Variables

It is of high socio-economic value to provide reliable forecasts of extreme and hazardous weather events at the storm-scale. Such numerical weather forecast systems can sometimes provide unreliable forecasts, where the actually observed event lies often outside the uncertainty predicted by the ensemble spread. One reason for the lack in spread lies in the need to truncate the underlying equations at a particular scale, leaving the smaller scale features and their variability unrepresented. Several methods to represent such model error have been developed, namely multi-model and multi-parameter approaches as well as stochastic parameterization schemes. NCAR scientists have been working to combine the multi-parameter with the stochastic parameterization approach to achieve a reliable ensemble forecasting system within a single-physics suite/single dynamical core framework.

NCAR scientists, in collaboration with Colorado State University/CIRA, University of Colorado/CIRES, NOAA/Earth System Research Laboratory, and the NOAA/Environmental Modeling Center, developed a Stochastically Perturbed Parameterization (SPP) Scheme by stochastically perturbing key parameters or variables in multiple parameterization schemes, such as the boundary layer, surface layer, gravity wave drag, radiation, microphysics, and horizontal diffusion schemes. Essential is that the stochastic pattern generator have temporal and spatial correlations which represent underestimated meso-scale organization:

Perturbation pattern for three different spatial scales: a) convection-permitting scale, b) meso-scale, c) synoptic scale
Figure: Perturbation pattern for three different spatial scales: a) convection-permitting scale, b) meso-scale, c) synoptic scale.

The research team has shown that a storm-scale ensembles with SPP have comparable skill and reduced systematic biases in comparison with statistically inconsistent multi-model ensembles. This demonstrates that a single dycore/single physics suite together with stochastic parameterizations is a viable alternative to statistically inconsistent multi-model approaches.

While this work was sponsored by NOAA and conducted with the operational HRRR ensemble utilizing the WRF ARW dynamical core and WRF-HRRRE physics, it is anticipated that the scientific findings will carry over to other next-generation high-resolution regional ensemble systems.

These findings were published in Monthly Weather Review, a Journal of the American Meteorological Society.