Master´s thesis defense by Kasper Hintz – Niels Bohr Institute - University of Copenhagen

Niels Bohr Institute > Calendar > NBI Calendar 2015 > Master´s thesis defens...

Master´s thesis defense by Kasper Hintz

The Sensitivity of Extreme Precipitation in Relation to Changing Time Step in the HIRLAM NWP Model

Extreme precipitation related to convection is often poorly predicted, even for small forecast lead times. In this thesis the sensitivity of extreme precipitation forecasts to changes of the time step in the DMI-High Resolution Limited Area Model (HIRLAM) -based Numerical Weather Prediction (NWP) nowcasting model is investigated, with the aim to improve the prediction of severe precipitation events. By comparing model output to radar estimated precipitation rates, it is found that the skill of the forecasted precipitation rates increases significantly as the time step is decreased. The forecasted precipitation pattern is investigated by using the methods of Fractional Skill Score (FSS), from which the results show an improvement of the forecasted precipitation pat- tern as the time step decreases. The NWP nowcasting model is tested with both the Soft Transition Condensation (STRACO) scheme and the Kain-Fritsch (KF) convective parameterization scheme. Furthermore, simulations with no convective parameteriza- tion have been performed, which shows that the model dynamics are very sensitive to a changing time step. It is found that the model dynamics can not properly resolve Deep Moist Convective (DMC) processes that occur on small scales in space and time with the current operational time step (100s, 3km grid size, 40 vertical levels).

It is preferable not to decrease the time step permanently, as this would increase the computation time without getting any benefit the vast majority of the time. Instead, for an operational implementation, it is suggested that a simple prior adaptive time step scheme is implemented, such that the time step is decreased when convection is active or likely to occur in the next model run.