Uncertainty in numerical weather forecasts arising from an imperfect knowledge of the initial condition of the atmospheric system and the discrete modeling of physical processes is addressed with ensemble prediction systems. The breeding method allows the creation of initial condition perturbations in a simple and computa-tionally inexpensive way. This technique uses the full nonlinear dynamics of the system to identify fast-growing modes in the analysis fields, obtained from the difference between control and perturbed runs rescaled at regular time intervals. This procedure is more suitable for the high-resolution ensemble forecasts required to reproduce small-scale high-impact weather events, as the complete nonlinear model is employed to generate the perturba-tions. The underdispersion commonly observed in ensemble forecasts emphasizes the need to develop methods that increase ensemble spread and diversity at no cost to forecast skill. In this sense, we investigate the benefits of different breeding techniques in terms of ensemble diversity and forecast skill for a mesoscale ensemble over the Western Mediterranean region. In addition, we propose a new method, Bred Vectors Tailored Ensemble Perturbations, designed to control the scale of the perturbations and indirectly the ensemble spread. The combina-tion of this method with orthogonal bred vectors shows significant improvements in terms of ensemble diversity and forecast skill with respect to the current arithmetic methods.
All Science Journal Classification (ASJC) codes
- Atmospheric Science