A hybrid regional climate downscaling for the southern Brazil coastal region

A hybrid method is applied to generate a high‐resolution regional downscaling of atmospheric conditions to the southern coast of Brazil. The method consists of applying a principal component analysis to daily fields of the sea level pressure (SLP) data from the NCEP‐CFSR reanalysis. A cluster analysis (K‐means) is then applied to the 87 principal components that explained 95% of the variance of the time series. Daily atmospheric conditions were clustered into 36 weather types which represent the most predominant conditions observed in the study area. The estimated weather types were able to represent the major atmospheric systems affecting local climate, including the cyclones and anticyclones that are usually present in this region. Then, we applied the numerical Ocean‐Land‐Atmosphere Model (OLAM) to dynamically downscale the atmospheric condition that is closest to the centroid of each cluster. The model was set with a global grid and a refining approach with 6 km grid spacing over the coastal region of south Brazil. This approach allowed us to represent simultaneously the planetary waves and the local mesoscale systems, and their mutual interactions. The results provided new high‐resolution atmospheric fields for the coastal region and showed that the model was capable of resolving the major local mesoscale features. The main advantage of applying such a method is in reducing the number of numerical simulations (lower computational cost) at the same time it represents the totality of the atmospheric conditions observed in the study area. The final results consist in detailed information of the local climate that can be related to injuries to the coastal area and thus is useful to support decision‐makers. Global climate models provide valuable information about the climate but their grid spacing is often too low to give precise information about the local meteorological features. Meanwhile, climatological simulations with regional models are very demanding on computer resources. In this work, we use a method for selecting representative weather types for high‐resolution simulations. Those weather types are capable of representing the major climate variability for the study area and can also provide information valuable for decision‐makers.