Evaluation of the MPAS Model for the Reproduction of an Extreme Precipitation Event over the North Coast of the State of São Paulo - Brazil

The southeastern region of Brazil has experienced an increase in extreme precipitation events, with more than a thousand events occurring between 2015 and 2020. Several studies have shown an increase in the frequency and intensity of these events, triggering natural hazards with catastrophic consequences. This is related to the location of the South Atlantic Convergence Zone and deep convection associated with low-level anticyclonic circulations. In addition, blocking systems, cold frontal passages, orographic lift and local convective instability are the main rain producing systems near the coast with heavy and persistent rain. The last event occurred on February 18-19, 2023, on the coast of São Paulo State and became the highest rainfall event in the history of Brazil, when almost 700 mm of rain were recorded in less than 24 hours on the North Coast of São Paulo, causing several damages and dozens of deaths (65) in the region. Given the impact and recurrence of these events, the need for accurate numerical weather prediction (NWP) models became critical to assist decision makers and governments in planning effective actions to avoid or mitigate catastrophic consequences. Several studies have highlighted the challenges of accurately predicting rainfall intensity and location, the dependence on appropriate initial conditions, the importance of microphysical schemes, and the appropriate choice of parameterizations. Our main objective was to apply the Model for Prediction Across Scales Atmosphere (MPAS) model with a variable-resolution mesh under convection-permitting resolution to perform the simulation for the extreme rainfall event over the north coast of São Paulo. A 60-3 km variable resolution mesh with refinement centered on the city of São Sebastião was configured. Model simulations were driven by GDAS/FNL analysis and ERA5 reanalysis to investigate sensitivity to initial conditions. Gridded data from MERGE for daily accumulated precipitation were used to evaluate the simulated daily accumulated precipitation. Data from CEMADEN automatic rain gauges distributed over the coast of São Paulo are also used for precipitation evaluation, while INMET automatic weather stations provide weather variables over Brazil for general evaluation. The ERA5-driven simulation showed more precipitation activity on the north coast according to the simulated radar reflectivity compared to the GDAS/FNL-driven simulations. The bias computed with MERGE data as reference shows a general underestimation for all model runs over the main region. The comparison of simulated hourly precipitation with CEMADEN rain gauge data showed in general a better representation of the observed precipitation by the ERA5-driven simulation, but all simulations were not able to reproduce the precipitation intensity and timing exactly. Three microphysical schemes were also tested: Thompson (convection-permitting suit), WSM6, and Kessler. The results show similar performance in reproducing weather variables, especially for Thompson and WSM6, with slightly better results for the WSM6 run. All simulations predicted rainfall in excess of 250 mm/day in the most affected area. These preliminary results provide valuable information on the performance of MPAS, and the evaluation needs to be extended to other physical parameterizations and model configurations.