This study evaluates the ability of the Model for Prediction Across Scales - Atmosphere (MPAS version 8.1.0) to reproduce an unprecedented extreme precipitation event that occurred along the coast of the state of São Paulo (SP), in southeastern Brazil, on February 18-19, 2023. A set of sensitivity experiments was conducted using global variable-resolution (VR) meshes (60–3, 60–10 and 46–12 km) and a regional 3-km mesh. The 60-3 km experiments examine the influence of lead time, microphysical parameterizations, physics suites, and modeling framework (regional vs. global VR). Model simulations were evaluated against precipitation data from the CEMADEN rain gauge network and radar estimates from FCTH. Results show that simulation skill is highly sensitive to model configuration. While the 60–3 km global VR mesh with the CP physics suite simulates precipitation associated with orographic lifting along the Serra do Mar Mountain, the best overall performance is obtained with coarser 60-10 km global VR mesh using MR physics suite. This configuration provides improved representation of atmospheric moisture transport and PBL processes. The regional 3 km experiment simulates precipitation intensity close to the observed maximum (675 mm/day vs 683 mm/day), although the rainfall core is displaced. These findings highlight the challenges of predicting highly localized extreme rainfall over coastal complex terrain and provide guidance for the application of MPAS in forecasting high-impact precipitation events in southeastern Brazil. • This is the first study to use the MPAS model to simulate an unprecedented extreme precipitation event over the São Paulo coast in southeastern Brazil. • Experiments with low resolution (10 km) global variable-resolution meshes over the study area outperformed those with high resolution (3 km). • Physical parameterizations associated with the representation of atmospheric moist and PBL processes play a critical role in the reproduction of precipitation. • Regional MPAS with 3 km resolution and no convective parameterization reproduced the precipitation intensity observed during the event.
Júnior et al. (Sun,) studied this question.