Impact of a Stochastic Energy Backscatter Scheme on Climate and Variability across Timescales and Resolutions

Stochastic physics is one of most widely used methods to represent model uncertainty in ensemble prediction systems of weather and climate models. These schemes aim to represent absent or poorly simulated process whose scales are below the truncation scale, they have been proven to be a skilful tool against the common underdispersiveness (lack of internal variability) of these models, as well as theoretically able to improve the mean climate through a noise-induced drift (better variability leads to a better mean climate). However, the formulation of these schemes often relies in pragmatic assumptions with limited scientific basis, and their physical realism is often challenged. The stochastic energy backscatter method is one of the main formulations of stochastic physics. It is designed to stochastically simulate upscale cascades of energy coming from numerical dissipation, convective subgrid-scale events or subgrid mountain drag. This scheme has been successfully implemented in many of the most important numerical weather prediction models across the world. It improves the ensemble skill scores, and under some configurations the mean climate too. In order to understand the impacts of the stochastic energy backscatter concept in a deterministic framework, we use the Stochastic Kinetic Energy Backscatter (SKEB2) scheme in the Met Office Unified Model (MetUM). We explore the impact of SKEB2 across timescales and resolutions in terms of usual model evaluation metrics such as biases or root mean error square, as well as some process-based techniques to diagnose the simulation of tropical and extra-tropical variability, such as cyclone tracking, Lorenz Energy Cycle or Madden Julian Oscillation diagnostics. Our results show that the extra kinetic energy added by SKEB2 can improve the representation of key processes that drive the atmospheric variability, leading to a slight improvement of climate biases. However it degrades the skill of short-range (less than 5 days) deterministic forecasts.