Towards a process-oriented understanding of the impact of stochastic perturbations on the model climate

Abstract

Stochastic parametrisation techniques have been used by operational weather centres for decades to produce ensemble forecasts and to represent uncertainties in the forecast model. Their use has been demonstrated to be highly beneficial, as it increases the reliability of the forecasting system and reduces systematic biases. Despite the random nature of the perturbation techniques, the response of the model can be nonlinear, and the mean state of the model can change. In this study, we attempt to provide a process-based understanding of how stochastic model perturbations affect the model climate. Previous work has revealed sensitivities of the occurrence of diabatically driven, rapidly ascending airstreams to the stochastically perturbed parametrisation tendencies (SPPT) scheme. Such strongly ascending airstreams are linked to different weather phenomena, such as precipitation and upper-tropospheric ridge building in the midlatitudes, which raises the question of whether these processes are also influenced by stochastic perturbations. First, we analyse if rapidly ascending airstreams also show sensitivities to a different perturbation technique – the stochastically perturbed parametrisations (SPP) scheme, which directly represents parameter uncertainty in parametrisations and has recently been developed at the European Centre for Medium-Range Weather Forecasts (ECMWF). By running a set of sensitivity experiments with the Integrated Forecasting System (IFS) and by employing a Lagrangian detection of ascending airstreams, we show that SPP results in a systematic increase in the occurrence of ascending air parcel trajectories compared to simulations with unperturbed model physics. This behaviour is very similar to that of SPPT, although some regional differences are apparent. The one-sided response to the stochastic forcing is also observed when only specific parametrisations are perturbed (only convection parametrisation and all parametrisations but convection), and we hypothesise that the effect cannot be attributed to a single process. Thereafter, we link the frequency changes in ascending airstreams to closely related weather phenomena. While the signal of increased ascending motion is directly transmitted to global precipitation sums for all analysed schemes, changes to the amplitude of the upper-level Rossby wave pattern are more subtle. In agreement with the trajectory analysis, both SPPT and SPP increase the waviness of the upper-level flow and thereby reduce a systematic bias in the model, even though the magnitude is small. Our study presents a coherent process chain that enables us to understand how stochastic perturbations systematically affect the model climate. We argue that weather systems which are characterised by threshold behaviour on the one hand and that serve as a dynamical hinge between spatial scales on the other hand can convert zero-mean perturbations into an asymmetric response and project it onto larger scales.