Avaliação da confiabilidade da previsão de tempo por ensemble para a região de Angola /África

Abstract

The reliability of ensemble weather forecast was evaluated for the region of Angola, through two experiments where the physics of WRF model was disturbed. The focus of this experiment was directed to evaluate the uncertainty of the physical model process, the sensitivity of physical parameterization, as well as the performance of deterministic forecasts. To achieve this goal, an array with 16 settings of WRF model was created using different physical layout combinations, where four different schemes microphysics, two different schemes of planetary boundary layer and two different schemes of cumulus parametrization. The model was integrated for a period of 102 hours over two operational weather forecast experiments for the periods; of 03/19/2013 to 23/03/2013 and 12/04/2014 to 16/04/2014. Data of maximum temperature, minimum air temperature and accumulated precipitation weather stations referenced by WMO, Luanda, Uige, Saurimo, Luena, Huambo and Lubango, were compared with the results of the WRF model simulations in the respective grid points. The reliability of the ensemble mean was evaluated based on two metrics, the modified box plots and the discrepancy between the RMSE and SPRD, which produced consistent results. The sensitivity analysis ace physical changes was determined based on the relative influence of different parameterizations in bias. But the performance of deterministic forecasts was based on the average RMSE ranking analysis. The results show that the predictions presented uncertainty that may be related or the parameterization schemes adopted, initial conditions, or by smoothing the topography of the template. In general, the model showed maximum and minimum temperatures consistently lower than observed. The minimum temperature was best represented by the ensemble than the maximum temperature. The ensemble forecast, unable to satisfactorily reproduce the amount of the precipitation occurred, as well as detect isolated rains coming from convective systems. None of the combinations adopted was efficient to simulate all variables in all locations.