Convection-permitting mechanism to enhance the thunderstorm forecasting over Sao Paulo state in Brazil

In this work we tunned three NWP models - ICON, WRF and MPAS - for simulation of severe thunderstorms considering convection permitting mechanism. This feature implies explicitly solving atmospheric convective processes, rather than using approximations, especially in vertical air movements that drive storms. Nowadays, it is well known that NWP models solve weather events such as precipitation and wind gusts with relatively high accuracy by means of equations describing the physical processes of the atmosphere. These processes comprise heat transfers, moisture exchanges, turbulent movements, interactions with the surface, and radiation. Convection is an important process in the atmosphere, as it contributes to the formation of circulation from large scales (hundreds of kilometers) to local intense precipitation due to storms. In weather forecasting models, these physical processes are solved by mathematical equations at grid points with horizontal spacing ranging from a few kilometers to hundreds of kilometers. However, not always models can adequately capture the underlined features of them. Regions with complex geography, such as mountains and coastlines, are especially challenging for NWP models to accurately describe storm systems, primarily due to abrupt variations in air movements and interactions between land and sea. In order to mitigate these imperfections, mathematical solutions known as parametrizations are used to estimate the effects of convective atmospheric moisture on cloud systems represented by the model. Generally, this convection mechanism in models operates on grids larger than 5 km. However, for short-term forecasting, typically using smaller grid sizes (typically between 1 to 3 km), it is useful to explicitly describe the thermodynamic cycle of convective processes, which is handled in a hybrid manner where deep convection in large clouds is explicitly solved by the model, while shallow convection with small sub-grid clouds that do not produce precipitation is parametrized. The ICON, WRF and MPAS models have the capability to explicitly solve deep atmospheric convection, which is a crucial feature for the applications of short-term forecasting of severe convective events. Therefore, we configure a two-way nested simulation, hybrid convection scheme and by considering a large domain grid of about 7 km mesh and an inner grid of about 3 km mesh, for ICON and WRF and a 3 km target grid mesh for the MPAS, which covers Central, Southeast and South parts of Brazilian's regions. We then applied this configuration to 3 strong thunderstorms events, which were propagate from South Brazil to Sao Paulo state, happened on October 2024, September 2025 and November 2025. AWS observations and images from GOES-19 satellite were used to evaluate the simulations. The results indicate that precipitation forecasts are more organized with explicit convection compared to when parametrized with shallow convection. Additionally, the improvement of simulation variables within the inner grid was made possible by the convection-permitting mechanism, which explicitly solves large-scale convective clouds and only parametrizes shallow sub-grade processes that do not produce precipitation or are very weak. The experiments were crucial as they involve significant improvements for forecasting storms, enhancing the model NWP's nowcasting and monitoring severe events.