An integrated framework for real-time river sediment monitoring: fusing low-cost hydrometry with a comparison of LSTM and N-BEATS models and uncertainty analysis

Document Type : Original Article

Authors

1 Department of Water Resources Engineering, Faculty of Civil, Water and Environmental Engineering , Shahid Beheshti University, Tehran, Iran.

2 Department of Water Resources Engineering, Faculty of Civil , Water and Environmental Engineering, Shahid Beheshti University, Tehran, Iran.

Abstract

Objective: The objective of this study was to develop an integrated framework for real-time monitoring and prediction of the median bed-sediment diameter (D50) in rivers and to compare the performance of two deep learning architectures, Long Short-Term Memory (LSTM) and Neural Basis Expansion Analysis for Time Series (N-BEATS), in predicting sediment size. An additional objective was to quantify the uncertainty of LSTM predictions using Monte Carlo (MC) dropout.
 
Method: High-frequency water-level time-series data were collected from the Roudak River in Tehran using a custom-designed, low-cost ultrasonic stage sensor. The water-level data were integrated with grain-size analysis results obtained from bed-sediment samples collected during multiple field campaigns. Two deep learning models, LSTM and N-BEATS, were developed and evaluated for predicting D50. In addition, Monte Carlo dropout was applied to the LSTM model to assess the uncertainty and confidence of its predictions.
 
Results: Both deep learning models demonstrated a strong capability to learn the complex relationships between river hydraulic conditions and bed-sediment characteristics. However, the N-BEATS model achieved superior predictive performance, with an R² of 0.871 and an RMSE of 0.054, compared with an R² of 0.861 and an RMSE of 0.056 for the LSTM model. These results indicate that N-BEATS provided more accurate and robust predictions of D50.
 
Conclusions: The study demonstrated that integrating low-cost real-time water-level monitoring with advanced deep learning techniques provides a practical and cost-effective approach for river sediment monitoring. The superior performance of N-BEATS compared with LSTM highlights the potential of modern time-series architectures for improving computational hydrology and supporting more effective sediment and water-resources management.

Keywords

Main Subjects

References

Adibrad, R., Khosravinezhad, B., & Adibrad, M.H. (2025). Application of outlier robust extreme learning machine (ORELM) in river flow prediction. Advanced Technologies in Water Efficiency, 5(3), 44-55. https://doi.org/10.22126/atwe.2025.11918.1159 [In Persian]
Akbari, M., Pourreza-Bilondi, M., Khashei-Siuki, A., Kalanki, E., & Mianabadi, A. (2024). Uncertainty analysis of DSSAT model parameters in estimating cotton yield using GLUE method. Advanced Technologies in Water Efficiency, 4(4), 19-34. https://doi.org/10.22126/atwe.2024.10960.1135[In Persian]
Boiten, W. (2005). Hydrometry. Taylor & Francis Publications, London, United Kingdom. https://doi.org/10.1201/9780203971093
Bowler, A. L., Bakalis, S., & Watson, N. J. (2020). Monitoring mixing processes using ultrasonic sensors and machine learning. Sensors, 20(7), 1813. https://doi.org/10.3390/s20071813
Bowler, A. L., Pound, M. P., & Watson, N. J. (2022). A review of ultrasonic sensing and machine learning methods to monitor industrial processes. Ultrasonics, 124, 106776. https://doi.org/10.1016/j.ultras.2022.106776
De Camargo, E. T., Spanhol, F. A., Slongo, J. S., da Silva, M. V. R., Pazinato, J., de Lima Lobo, A. V., ... & Martins, L. D. (2023). Low-cost water quality sensors for IoT: A systematic review. Sensors, 23(9), 4424. https://doi.org/10.3390/s23094424
Ji, H. W., Yoo, S. S., Lee, B. J., Koo, D. D., & Kang, J. H. (2020). Measurement of wastewater discharge in sewer pipes using image analysis. Water, 12(6), 1771.‏ https://doi.org/10.3390/w12061771
Huang, C. C., Chang, M. J., Lin, G. F., Wu, M. C., & Wang, P. H. (2021). Real-time forecasting of suspended sediment concentrations in reservoirs by the optimal integration of multiple machine learning techniques. Journal of Hydrology: Regional Studies, 34, 100804.  https://doi.org/10.1016/j.ejrh.2021.100804
Kratzert, F., Klotz, D., Brenner, C., Schulz, K., & Herrnegger, M. (2018). Rainfall–runoff modelling using long short-term memory (LSTM) networks. Hydrology and Earth System Sciences, 22(11), 6005-6022. https://doi.org/10.5194/hess-22-6005-2018
Masoudimoghaddam, M., Yazdi, J., & Shahsavandi, M. (2025). A low-cost ultrasonic sensor for online monitoring of water levels in rivers and channels. Flow Measurement and Instrumentation, 102, 102777. https://doi.org/10.1016/j.flowmeasinst.2024.102777
Pirouzmehr, M., Izadbakhsh, M. A., Shabanlou, S., Rajabi, A., & Yousefvand, F. (2025). Simulation of groundwater level fluctuations using support vector regression (SVR) and extreme learning machine (ELM) models. Advanced Technologies in Water Efficiency, 5(3), 1-19. https://doi.org/10.22126/atwe.2025.11658.1150 [In Persian]
Roushangar, K., Shahnazi, S., & Azamathulla, H. M. (2022). Sediment transport modeling through machine learning methods: review of current challenges and strategies. River dynamics and flood hazards: studies on risk and mitigation, 223-240. https://doi.org/10.1007/978-981-19-7100-6_13
Saberian, M., Samadi, V., & Popescu, I. (2024). Probabilistic hierarchical interpolation and interpretable configuration for flood prediction. Hydrology and Earth System Sciences Discussions, 2024, 1-41. https://doi.org/10.5194/hess-2024-261
Sahoo, A., & Udgata, S. (2019). A novel ANN based adaptive ultrasonic measurement system for accurate water level monitoring. IEEE Transactions on Instrumentation and Measurement, 69(6), 3359–3369. https://doi.org/10.1109/TIM.2019.2939932
Wang, D., Li, Q., Liu, S., Pan, L., & Li, J. (2025). Long-term river flow forecasting: An integrated deep learning model with multi-scale feature extraction. Expert Systems with Applications, 281, 127387.‏ https://doi.org/10.1016/j.eswa.2025.127387
Ye, F., Jin, Z., Zhang, P., Xu, D. Z., Lan, L., & Sun, J. (2025). Research on the digital twin polder area system driven by integrating the Xin'anjiang model and the N-BEATS model. International Journal of Intelligent Systems, 2025, 8899669. https://doi.org/10.1155/int/8899669
Yu, Y., Zhu, Y., Li, S., & Wan, D. (2014). Time series outlier detection based on sliding window prediction. The Scientific World Journal, 2014, 879736. https://doi.org/10.1155/2014/879736
Advanced Technologies in Water Efficiency
Volume 6, Issue 3
September 2026
Pages 117-132

Files

History

  • Receive Date: 12 July 2025
  • Revise Date: 27 November 2025
  • Accept Date: 02 February 2026

Share

How to cite

Statistics