Experimentos Fatoriais Fracionados Assimétricos para Avaliação de Modelos para Previsão de Chuva no Nordeste do Brasil

Abstract

In the state of Rio Grande do Norte, episodes of heavy rains have caused disorders such as land slides, flood and chaotic traffic. Thus, rainfall is a major concern for researchers and other agencies linked to weather research. The accurate prediction of extreme events, such as heavy rainfall, is an essential factor in public polices to mitigate the effects caused by meteorological phenomena. Since the mid-twentieth century, the use of quantitative methods to conduct weather forecast by numerical simulation has stood out in meteorology. With the technological advancement of computing, more and more atmospheric phenomena can be simulated in high resolution. In these models, identifying combinations and/or atmospheric physics parameterization settings is still a challenge for researchers. In this sense, the main objective of this work is to implement experimental design techniques to select parameterization settings and evaluate numerical weather forecasting models. This paper presents the use of asymmetric fractional factorial designs (EFFA) to evaluate the performance of Weather Research and Forecasting (WRF), using various physical parameterizations, to simulate a weather event in 2008 in Rio Grande do Norte State. We selected four cumulus parameterization schemes (Grell-Deveyi, NSAS, Tiedtke and Kain-Fritsch), two schemes of microphysics (WSM6 and Thompson), two planetary boundary layer (YSU and MYJ) and two model top settings (10mb and 50mb), for three different domains (36Km, 12Km and 4km) and conducted 32 simulations. We use Lenth plots to analyse the experiment and guide the identificationof appropriat parameterization schemes. The final results of this study are promising and confirm that statistically designed experiments is of utmost importance to assess WRF, bringing significant progress in this field. It is whorth mentioning that the same procedure can be used for other models