Evaluating Past Horizons for U-Net-Based Precipitation Nowcasting with Radar Data in Southeastern Pará, Brazil

  • Rafael Rocha ITV / UFPR
  • Eduardo Carvalho ITV
  • Ewerton Oliveira ITV
  • Sergio Viademonte ITV
  • Douglas Ferreira ITV
  • Ronnie Alves ITV

Resumo


Severe weather events significantly impact daily life, especially during emergencies, affecting human lives and the economy. Decision-making in such events, like heavy rainfall, strong winds, and flash floods, is challenging due to the rapid changes and strong interconnections between the variables involved. Nowcasting models use real-time data, commonly from weather radars, to forecast short-term rain up to 6 hours ahead, supporting decision-making in severe weather situations. These models provide early warnings and precise information about the location, intensity, and duration of these events. Recently, machine learning models for precipitation forecasting have gained prominence due to their ability to learn from data and offer reliable and fast predictions. This study explores precipitation nowcasting using weather radar data in the southeast of Pará, Brazil, focusing on a one-hour forecast horizon utilizing the U-Net architecture. Four models based on U-Net algorithm, investigating past horizons of 30, 60, 90, and 120 minutes, are evaluated using categorical and continuous metrics, and a visual comparison of the 60-minute forecast horizon. The results demonstrate that the model with a past horizon of 120 minutes outperforms the other models in all evaluated metrics, achieving 37.96 and 0.5476 scores in continuous and categorical metrics, respectively, improving the forecast of severe events and decision making up to a 60-minute forecast horizon.
Palavras-chave: machine learning, past horizons, precipitation nowcasting, weather radar

Referências

Ayzel, G., Scheffer, T., and Heistermann, M. Rainnet v1. 0: a convolutional neural network for radar-based precipitation nowcasting. Geoscientific Model Development 13 (6): 2631–2644, 2020.

Bonnet, S. M., Evsukoff, A., and Morales Rodriguez, C. A. Precipitation nowcasting with weather radar images and deep learning in são paulo, brasil. Atmosphere 11 (11): 1157, 2020.

Caseri, A. N., Santos, L. B. L., and Stephany, S. A convolutional recurrent neural network for strong convective rainfall nowcasting using weather radar data in southeastern brazil. Artificial Intelligence in Geosciences vol. 3, pp. 8–13, 2022.

Jorge, A. A. S., Quiles, M. G., Costa, I. C., and Santos, L. B. L. A convolutional LSTM neural network for precipitation nowcasting based on weather radar data. In XXIII Brazilian Symposium on Geoinformatics - GEOINFO, L. B. L. Santos and M. de Arruda Pereira (Eds.). MCTIC/INPE, São José dos Campos, SP, Brazil, pp. 384–388, 2022.

Marshall, J. S. and Palmer, W. M. K. The distribution of raindrops with size. Journal of Atmospheric Sciences 5 (4): 165–166, 1948.

Ronneberger, O., Fischer, P., and Brox, T. U-net: Convolutional networks for biomedical image segmentation. In Medical image computing and computer-assisted intervention–MICCAI 2015: 18th international conference, October 5-9, 2015, proceedings, part III 18. Springer, Munich, Germany, pp. 234–241, 2015.

Shi, X., Chen, Z., Wang, H., Yeung, D.-Y., Wong, W.-K., and Woo, W.-c. Convolutional lstm network: A machine learning approach for precipitation nowcasting. Advances in neural information processing systems vol. 28, pp. 802–810, 2015.

Wang, Y., Gao, Z., Long, M., Wang, J., and Philip, S. Y. Predrnn++ towards a resolution of the deep-in-time dilemma in spatiotemporal predictive learning. In International conference on machine learning. PMLR, Stockholmsmässan, Stockholm, Sweden, pp. 5123–5132, 2018.
Publicado
17/11/2024
ROCHA, Rafael; CARVALHO, Eduardo; OLIVEIRA, Ewerton; VIADEMONTE, Sergio; FERREIRA, Douglas; ALVES, Ronnie. Evaluating Past Horizons for U-Net-Based Precipitation Nowcasting with Radar Data in Southeastern Pará, Brazil. In: SYMPOSIUM ON KNOWLEDGE DISCOVERY, MINING AND LEARNING (KDMILE), 12. , 2024, Belém/PA. Anais [...]. Porto Alegre: Sociedade Brasileira de Computação, 2024 . p. 49-56. ISSN 2763-8944. DOI: https://doi.org/10.5753/kdmile.2024.244469.