Supporting data for: Random Forest classification of water masses and reconstruction of thermohaline profiles in the São Sebastião Channel

This dataset contains the code and input data used in the paper "Random Forest classification of water masses and reconstruction of thermohaline profiles in the São Sebastião Channel (South Brazil Bight)". The repository includes: - Python scripts: Python files implementing Random Forest classifiers and regressors using Scikit-learn- Trained models: Pre-trained Random Forest models for water mass classification and profile reconstruction- Input data: Processed temperature and salinity profiles from 13 monitoring stations in the São Sebastião Channel (1991-2018), along with ERA5 wind reanalysis and satellite SST data The models presented in this work achieve: - Water mass classification: F1-scores > 0.92 for all four classes (Coastal Water, Tropical Water, South Atlantic Central Water, and Mixed)- Profile reconstruction: R² = 0.992, mean absolute errors of 0.14°C (temperature) and 0.06 (salinity) The framework requires only satellite sea surface temperature and ERA5 wind reanalysis as inputs, both globally available.