Spatio-Temporal Graph Neural Networks for Multi-Horizon River-Stage Forecasting in the Amazon: A Case Study on the Manaus–Santarém Corridor
Resumo
Inland waterway transport in the Brazilian Amazon depends critically on river navigability, which is strongly governed by hydroclimatic variability and increasingly severe hydrological extremes. This paper proposes a multi-model framework for daily multi-horizon river-stage forecasting in the Manaus–Santarém corridor, combining official hydrometric observations from ANA with a spatio-temporal graph neural network (STGNN) that encodes upstream–downstream connectivity and travel-time-informed edge weights. The STGNN is evaluated against temporal benchmarks, recurrent baselines, a statistical model, and operational baselines across six forecast horizons (1–90 days) under a unified protocol with a chronological hold-out test set covering 2024. The STGNN achieves the best MAE, RMSE, and NSE at every horizon, maintaining NSE > 0.99 through 30 days, a threshold no competing model reaches, with RMSE gains of 29–35% over the strongest baseline at 14–30 days, the window governed by upstream signal propagation according to the physical travel times encoded in the graph. Results demonstrate that explicitly representing the river network as a directed graph provides consistent and physically interpretable predictive advantages over purely temporal architectures for multi-horizon Amazonian hydrological forecasting.
Referências
Agência Nacional de Águas e Saneamento Básico (ANA) (2025b). Portal de dados abertos da ana. [link]. Acesso em: 10 mar. 2026.
Chagas, V. B., Chaffe, P. L., and Blöschl, G. (2022). Process controls on flood seasonality in brazil. Geophysical Research Letters, 49.
de Lima, L. S., e. Silva, F. E. O., Anastácio, P. R. D., de Paula Kolanski, M. M., Pereira, A. C. P., Menezes, M. S. R., Cunha, E. L. T. P., and Macedo, M. N. (2024). Severe droughts reduce river navigability and isolate communities in the brazilian amazon. Communications Earth and Environment, 5.
Domenighini, C. (2024). Autonomous inland navigation: a literature review and extracontractual liability issues. Journal of Shipping and Trade, 9.
Elneel, L., Zitouni, M. S., Mukhtar, H., and Al-Ahmad, H. (2024). Examining sea levels forecasting using autoregressive and prophet models. Scientific Reports, 14(1):14337.
Espinoza, J. C., Jimenez, J. C., Marengo, J. A., Schongart, J., Ronchail, J., Lavado-Casimiro, W., and Ribeiro, J. V. M. (2024). The new record of drought and warmth in the amazon in 2023 related to regional and global climatic features. Scientific Reports, 14.
Espinoza, J. C., Marengo, J. A., Schongart, J., and Jimenez, J. C. (2022). The new historical flood of 2021 in the amazon river compared to major floods of the 21st century: Atmospheric features in the context of the intensification of floods. Weather and Climate Extremes, 35.
Fassoni-Andrade, A. C., Fleischmann, A. S., Papa, F., de Paiva, R. C. D., Wongchuig, S., Melack, J. M., Moreira, A. A., Paris, A., Ruhoff, A., Barbosa, C., Maciel, D. A., Novo, E., Durand, F., Frappart, F., Aires, F., Abrahão, G. M., Ferreira-Ferreira, J., Espinoza, J. C., Laipelt, L., Costa, M. H., Espinoza-Villar, R., Calmant, S., and Pellet, V. (2021). Amazon hydrology from space: Scientific advances and future challenges.
Li, L. and Jun, K. S. (2024). Review of machine learning methods for river flood routing.
Luo, J., Zhu, D., and Li, D. (2025). Classification-enhanced lstm model for predicting river water levels. Journal of Hydrology, 650:132535.
Marengo, J. A., Espinoza, J. C., Fu, R., Muñoz, J. C. J., Alves, L. M., ROCHA, H. R. D., and Schöngart, J. (2024). Long-term variability, extremes and changes in temperature and hydrometeorology in the amazon region: A review. Acta Amazonica, 54.
Zhao, X., Wang, H., Bai, M., Xu, Y., Dong, S., Rao, H., and Ming, W. (2024). A comprehensive review of methods for hydrological forecasting based on deep learning.
