Potential of stochastic methods for improving convection-permitting ensemble forecasts of extreme events over the Western Mediterranean

Alejandro Hermoso,Victor Homar,Robert S Plant

doi:10.1016/j.atmosres.2021.105571

Alejandro Hermoso, Victor Homar + Show 1 more

Open Access

https://doi.org/10.1016/j.atmosres.2021.105571

Copy DOI

Abstract

The design of convection-permitting ensemble prediction systems capable of producing accurate forecasts of disruptive events is an extraordinarily challenging effort. The difficulties associated with the detection of extreme events found at these scales motivates the research of methodologies that efficiently sample relevant uncertainties. This study investigates the potential of multiple techniques to account for model uncertainty. The performance of various stochastic schemes is assessed for an exceptional heavy precipitation episode which occurred in eastern Spain. In particular, the stochastic strategies are compared to a multiphysics approach in terms of both spread and skill. The analyzed techniques include stochastic parameterization perturbation tendency and perturbations to influential parameters within the microphysics scheme. The introduction of stochastic perturbations to the microphysics processes results in a larger ensemble spread throughout the entire simulation. Conversely, modifications to microphysics parameters generate small-scale perturbations that rapidly grow over areas with high convective instability, in contrast to the other methods, which produce more widespread perturbations. A conclusion of specific interest for the western Mediterranean, where deep moist convection and local orography play an important role, is that stochastic methods are shown to outperform a multiphysics-based ensemble for this case, indicating the potential positive impact of stochastic parameterizations for the forecast of extreme events in the region.

Highlights

Since the pioneering works of Lorenz in the 1960s, it is extensively recognized that numerical weather forecasts are uncertain, and that small differences amplify with time, eventually rendering a forecast useless (Lorenz, 1963, 1969)
Over the course of phase 1 (Fig. 4a–d), ensembles con taining perturbations to microphysics yield the largest spread and higher rainfall amounts than MPS and Stochastic Perturbed Physics Tendency (SPPT), with 6-h precipitation in excess of 50 mm in the 95th percentile
Stimulated by the typical deficiencies found in many convectivescale probabilistic forecasts concerning underdispersion, an examina tion of multiple methodologies to cope with model uncertainties for a heavy precipitation episode in eastern Spain is conducted

Summary

Introduction

Since the pioneering works of Lorenz in the 1960s, it is extensively recognized that numerical weather forecasts are uncertain, and that small differences amplify with time, eventually rendering a forecast useless (Lorenz, 1963, 1969). The theoretical frameworks of Liouville and Fokker-Planck equations (Hasselmann, 1976; Ehrendorfer, 1994) provide a basis to describe the time evolution of uncertain sys tems, their practical application in operational weather forecasting is unachievable and currently restricted to low complexity systems (Hermoso et al, 2020a). In this context, ensemble prediction systems (EPS) emerge as a feasible alternative tool to account for the inherent uncertainties in numerical weather prediction, initial con dition and model errors. Some operational ensembles introduce perturbations to sea surface temperature and/or soil variables to sample uncertainties asso ciated with these fields (e.g., Tennant and Beare, 2014; Bouttier et al, 2016)

Objectives

Methods

Results

Conclusion