Modelos de Deep Learning em auxílio à previsão de variáveis meteorológicas em aeroportos na Antártica

• Alana de Lima Pontes Gadelha CHM
• Lúcia M. A. Drummond UFF
• Leandro Santiago de Araújo UFF

DOI: https://doi.org/10.5753/eradse.2025.17119

Resumo

This study investigates the use of Deep Learning models for forecasting hourly atmospheric pressure and the temperature at the SCRM Aerodrome in Antarctica, based on nearly 15 years of METAR data. The evaluated models were MLP, CNN, LSTM, and BiLSTM, with the BiLSTM achieving the lowest errors and showing strong ability to capture long-term dependencies. Forecasts were also compared with the state-of-the-art ICON-LAM numerical model: while ICON-LAM demonstrated high accuracy, the BiLSTM achieved competitive performance with much lower computational cost. These results highlight the potential of memory-based Deep Learning models as complementary tools to numerical weather prediction, enhancing forecasting and operational safety in polar regions.

Referências

Géron, A. (2019). Hands-On Machine Learning with Scikit-Learn, Keras, and TensorFlow: Concepts, Tools, and Techniques to Build Intelligent Systems. O’Reilly Media, Sebastopol, CA, 2nd edition.

Iriza-Burcă, D., Andrei, D., Dragoş, C., Mihăilă, A., Rugină, C., Dima, M., Pripon, C., and Mitrea, M. (2024). Comparison of cosmo and icon-lam high-resolution numerical forecast for romanian territory: Case studies and evaluation. Romanian Journal of Physics.

Xu, M. (2024). Comparative analysis of machine learning models for weather forecasting: A heathrow case study. Dean & Francis. ISSN 2959-6157.

Zhang, H., Liu, Y., Zhang, C., and Li, N. (2025). Machine learning methods for weather forecasting: A survey. Atmosphere, 16(1):82. Open Access under CC BY 4.0 License.