An Update to the Stochastically Perturbed Parameterization Scheme of HarmonEPS

Abstract

Abstract High-resolution, limited-area forecasting is strongly affected by errors in the initial atmospheric state, lateral boundary conditions (LBCs), and physical parameterizations used by numerical weather prediction (NWP) models. These errors need to be accounted for through the introduction of uncertainty in an ensemble prediction system (EPS). One approach to account for model error is to use a stochastically perturbed parameterizations (SPPs) scheme. A first version of the SPP scheme of HARMONIE EPS (HarmonEPS) has been tested, with promising improvements in ensemble spread. However, it introduced systematic biases and deteriorated skill scores for some variables. Here, we investigate the performance of an updated version of the HarmonEPS SPP scheme, which includes (i) the use of uniform distributions, (ii) the correlation of stochastic patterns between key SPP parameters, and (iii) the introduction of four additional parameters, in the microphysics and mass-flux schemes. Two five-parameter SPP-based setups are compared against initial and LBC perturbations setups for five forecast periods: (i) 22–28 March 2019, (ii) 6–12 July 2020, (iii) 20–26 February 2021, (iv) 13–26 January 2021, and (v) 20 May–2 June 2021. We find that SPP-based experiments show better probabilistic metrics for near-surface and cloud-related variables than the non-SPP experiments. The SPP-based ensembles show increased spatial spread (as indicated by dFSS), while maintaining similar spatial skill (as indicated by eFSS) with the non-SPP experiment. In addition, the systematic bias in the ensemble members of the previous SPP iteration has been alleviated with the use of uniform distributions. Finally, the use of microphysical and mass-flux perturbations improves the ensemble scores for cloud-related variables, precipitation, and visibility. Significance Statement Ensemble prediction systems are essential for weather forecasting. Their skill depends on the representation of uncertainties arising from errors in the initial atmospheric state and model physics. In this study, we present an update to the stochastically perturbed parameterizations scheme, which represents uncertainty arising from model errors, of the limited-area HARMONIE ensemble prediction system. The update to the scheme includes the introduction of uniform distributions and correlated perturbations for the perturbed parameters, as well as the addition of four new perturbed parameters. Overall, this revision removes systematic biases in ensemble members, which were present in the previous iteration of the scheme, and shows increased ensemble spread (up to 70%) and reduced model errors (up to 5%) for near-surface and cloud-related variables.