Ensemble hindcasting of winds and waves for the coastal and oceanic region of Southern Brazil

When hindcasting wave fields of storm events with wave models, the quality of the results strongly depends on several factors such as the computational grid resolution and the accuracy of the atmospheric forcing. In an effort to minimize the uncertainties involved in this process, three ocean wave and surface wind ensemble hindcast systems were established using the Simulating WAves Nearshore (SWAN) model and Weather Research and Forecasting Model (WRF) with atmospheric data from ERA5 Ensemble of Data Assimilation (EDA) as well as deterministic high-resolution systems. We established three ensemble systems to tackle this: SWN-ERA5EDA using ERA5-EDA global reanalyses winds, SWN-WRFERA5 employing WRF downscaling of ERA5-EDA, and SWN-WRFPPar incorporating WRF multi-physics runs for dynamical downscaling. This study focuses on extreme events in southern Brazil during an austral winter, highlighting the importance of increasing the resolution of ocean wave and surface wind data to provide more accurate and reliable forecasts for coastal and marine activities. Our analyses revealed that atmospheric downscaling performed with WRF not only increased the ensemble spread by significant amounts but also enhanced the sharpness of the wave ensemble hindcast compared with those based solely on the ERA5 EDA. Specifically, for the Rio Grande buoy location, the significant wave height ($H_s$) from the SWN-WRFERA5 system showed an increase of 0.5 over SWN-ERA5, and $H_s$ from the SWN-WRFPPar system increased by 0.6. Additionally, the wave peak period ($T_p$) for both SWN-WRFERA5 and SWN-WRFPPar systems experienced an increase of 1.2 compared to SWN-ERA5. Additionally, the ensemble produced with the WRF multi-physics approach captured peaks in the significant wave height registered by the buoy that were not reproduced by other ensemble systems, demonstrating an improvement in predictive accuracy, despite presenting a smaller correlation between spread and strong localized wave variations. Besides quantifying the hindcast error, the methodology presented in this work also offers a way to generate alternative and improved representations of past extreme events. This approach significantly contributes to our ability to sample recent climatic conditions and expand the dataset for statistical analyses, which is especially valuable for ocean and coastal engineering projects. This study underscores the critical role of enhancing computational and methodological approaches in wave modeling for better understanding and mitigating the impacts of extreme weather events on coastal and oceanic regions.