Uncertainty2: Stochastic Perturbations in a Convective-Scale Ensemble

Due to sensitivity to initial-condition error, it is of paramount importance that numerical weather prediction models capture uncertainty in the state of the atmosphere. However, there is uncertainty regarding the uncertainty: for instance, what is the spatial variation of cloud condensation nuclei? This presentation outlines some recent developments in stochastic perturbations, the newer breed of “variance generation” in an ensemble system, applied in a 3-km limited-area ensemble during severe weather outbreaks. Amongst five stochastic schemes implemented in the ensemble, we present two new methods: a fractal field applied to soil moisture, and a land-surface “morphing” algorithm inspired by music computer software. Four ensemble configurations are evaluated for their spread and skill characteristics: two with a fixed suite of parameterisation, with and without stochasticism, and two with mixed parameterisations, again with and without stochasticism. We find, as expected from previous literature, that the ensemble with the most “variance generation” (i.e., mixed parameterisations and stochastic perturbations) performs best, suggesting stochasticism has an important role to play in optimising ensemble performance. These improvements are maximised at large spatial scales and lower precipitation thresholds, which suggests future work is required to optimise stochasticism for the convection-allowing models.