Limits of Surface-Based Machine Learning Post-Processing for Severe Wind Gust Forecasting: A Case Study in Southern Brazil

Dataset Description This repository contains the processed datasets used in the study "Limits of Surface-Based Machine Learning Post-Processing for Severe Wind Gust Forecasting: A Case Study in Southern Brazil". The data supports the evaluation of Machine Learning architectures (CNN, LSTM, Transformer, Linear Regression) for correcting ECMWF-IFS wind gust forecasts against observational data from the SIMEPAR network. Content The repository consists of: NetCDF Files (.nc): These files contain the aligned time series for surface meteorological variables. The data is derived from both the ECMWF Integrated Forecasting System (IFS) ensemble (50 members) and surface observations from the System of Technology and Environmental Monitoring of Paraná (SIMEPAR). The dataset covers the period from January 2021 to January 2024. The input features include wind gust at 10 m, zonal (u) and meridional (v) wind components at 10 m, 2 m temperature, and 2 m dew point temperature. Scaling Parameters (JSON): A supplementary JSON file containing the normalization parameters (min/max values) used for scaling the input tensors. As the input tensors are normalized, this file is required to reconstruct the physical values.