Resumo:
On the afternoon of February 27, 2023, in approximately 40 minutes, more than 50 mm of rain occurred in Itajubá, southern Minas Gerais, resulting in significant damages and losses associated with flooding, landslides, fallen trees, and collapsed walls. In this context, the objective of this study is to evaluate the performance of the numerical forecast conducted by the Center for Weather and Climate Studies and Forecasting of Minas Gerais (CEPreMG) at the Federal University of Itajubá (UNIFEI), which operationally runs the Weather Research and Forecasting (WRF) model. Additionally, the study aims to conduct eight sensitivity tests with the WRF model to identify a potentially better configuration for forecasts in southern Minas Gerais. Thus, in addition to the control simulation, which uses the same configuration as the operational CEPreMG setup, the sensitivity tests consider different cloud microphysics parameterization schemes: (1) WSM3, (2) WSM6, and (3) WDM6; initial and boundary conditions: (4) GFS, (5) GDAS, and (6) ERA5; (7) soil moisture; (8) sea surface temperature (SST); and (9) a combination of the best configurations obtained from the eight experiments. To evaluate the model's performance, various statistical measures are applied to compare the simulated precipitation with observed data from meteorological stations, satellite estimates, and radar observations. Statistical analyses indicated that for microphysics, the WSM3 scheme showed the best accuracy, while initialization data from the ERA5 reanalysis led to a better simulation of the studied case. Regarding soil moisture, the climatological data provided with the WRF produced a simulation more similar to the studied event, whereas for SST, weekly data resulted in the best simulation performance. Statistical analyses of the experiment using the best-obtained configurations indicated that this simulation outperformed the previous experiments. Therefore, it is suggested that this configuration be used for case studies in southern Minas Gerais, although it cannot be implemented operationally due to the latency of the ERA5 reanalysis.
