Sovremennye problemy distantsionnogo zondirovaniya Zemli iz kosmosa, 2025, V. 22, No. 4, pp. 76-86
Problems in forecasting oceanographic series, using the example of Black Sea surface temperature and Ekman index for the southern part of Canary Islands upwelling
1 Institute of Natural Technical Systems, Sevastopol, Russia
Accepted: 23.06.2025
DOI: 10.21046/2070-7401-2025-22-4-76-86
The paper presents the experience of forecasting oceanic phenomena using recurrent neural networks and long-term–short-term memory structures. In particular, the problems arising in forecasting Black Sea surface temperature and Ekman index for the southern Canary Islands upwelling (13–21° N) are analyzed using a one-dimensional model for a 60-month forecast as an example. It is shown that one of the main problems in forecasting oceanographic series, such as wind speed and direction, sea surface temperature, etc., is the unpredictability of variations in these parameters on small time scales (up to a year). These variations of natural origin do not allow the model to use the training experience for reliable forecasting. Improving the quality of the model (complication) at the training stage leads to overfitting, which is reflected in model testing; simplification of the model leads to smoothing of the forecast. In addition, it is confirmed that the accuracy of forecasting oceanographic parameters largely depends on the quality and length of the original data. Due to the development of satellite technologies, access to a vast archive of remote sensing data is provided. However, some data, as shown in the article, related, for example, to parameters of coastal upwellings, are still quite noisy.
Keywords: LSTM model, recurrent model, gradient descent, Ekman upwelling index, coastal upwelling, sea surface temperature
Full textReferences:
- Chollet F., Deep learning with Python, Shelter Island, NY: Manning, 2017, 384 p.
- Chollet F., Watson M., Deep learning with Python, 3rd ed., Shelter Island, NY: Manning, 2025, 600 p.
- Géron A., Hands-on machine learning with Scikit-Learn, Keras, and TensorFlow: Concepts, tools, and techniques to build intelligent systems, 2nd ed. O’Reilly Media, 2019, 848 p.
- IPCC, 2021: Summery for policymakers, In: Climate Change 2021: The Physical Science Basis. Contribution of Working Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change, Cambridge University Press, 2021, pp. 1–32.
- Jebri F., Srokosz M., Jacobs Z. L. et al., Earth observation and machine learning reveal the dynamics of productive upwelling regimes on the Agulhas Bank, Frontiers in Marine Science, 2022, V. 9, Article 872515, DOI: 10.3389/fmars.2022.872515.
- Polonsky A. B., Serebrennikov A. N., What is the reason for the multiyear trends of variability in the Benguela upwelling?, Izvestiya, Atmospheric and Oceanic Physics, 2022, V. 58, No. 12, pp. 1450–1457, DOI: 10.1134/S0001433822120192.
- Wang S., Fu G., Song Y. et al., Ocean-Mixer: A deep learning approach for multi-step prediction of ocean remote sensing data, 2024, J. Marine Science and Engineering, V. 12, No. 3, Article 446, DOI: 10.3390/jmse12030446.