EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC-ND 4.0

Comparative Analysis of Machine Learning Models for Water Level Forecasting towards Flood Early Warning in the Mekong River

Tanapattara Wongkhamchan, Ariya Namvong, Jakub Michel, Chayada Surawanitkun, Wullapa Wongsinlatam, Tawun Remsungnen, Thalerngsak Wiangwiset, Wichian Tong-Gate, Apirat Siritaratiwat

ECTI Transactions on Electrical Engineering, Electronics, and Communications · 06.11.2026 · 25 (1)

DOI: 10.37936/ecti-eec.2027251.262931 ↗

Kurz gefasst

Fachartikel in der Zeitschrift ECTI Transactions on Electrical Engineering, Electronics, and Communications, online erschienen am 06.11.2026. Verfasst von Tanapattara Wongkhamchan und 8 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Flooding is a recurring hazard in Thailand, particularly in Nong Khai, causing significant socioeconomic impacts. Traditional hydrological models struggle to capture the complex nonlinear behavior of multi-station water-level systems. This study compares deep learning architectures with classical machine learning models for water-level forecasting. The research evaluates four distinct models:two recurrent neural network (RNN) architectures—Long Short-Term Memory (LSTM) and Gated Recurrent Units (GRU)—and two non-RNN baselines—Random Forest (RF) and Support Vector Machine (SVM). All models are trained and validated on a 19-year hydrological time-series dataset (2006–2025) comprising daily water level measurements from five key monitoring stations along the Mekong River. Performance was assessed using Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) metrics. Results indicate that while all four models successfully capture primary seasonal trends, the RNN based models (LSTM and GRU) outperform the classical ensemble (RF) and kernel-based regression (SVM) methods. The LSTM model demonstrated the highest predictive accuracy, achieving the lowest error rates with an RMSE of 0.3459 and an MAE of 0.1196. These results suggest that LSTM’s sequence-based memory gates provide superior capability in modeling complex hydrological fluctuations. Consequently, the LSTM is the foundational architecture for a future hydrological Digital Twin of the Mekong River Basin, designed to integrate tributary data for holistic flood simulation and operational decision support.

Zitieren

APA

Wongkhamchan, T., Namvong, A., Michel, J., Surawanitkun, C., Wongsinlatam, W., Remsungnen, T., Wiangwiset, T., Tong-Gate, W., & Siritaratiwat, A. (2026). Comparative Analysis of Machine Learning Models for Water Level Forecasting towards Flood Early Warning in the Mekong River. ECTI Transactions on Electrical Engineering, Electronics, and Communications, 25(1). https://doi.org/10.37936/ecti-eec.2027251.262931

Vancouver

Wongkhamchan T, Namvong A, Michel J, Surawanitkun C, Wongsinlatam W, Remsungnen T, et al. Comparative Analysis of Machine Learning Models for Water Level Forecasting towards Flood Early Warning in the Mekong River. ECTI Transactions on Electrical Engineering, Electronics, and Communications. 2026;25(1). doi:10.37936/ecti-eec.2027251.262931

BibTeX
@article{Wongkhamchan2026,
  author = {Wongkhamchan, Tanapattara and Namvong, Ariya and Michel, Jakub and Surawanitkun, Chayada and Wongsinlatam, Wullapa and Remsungnen, Tawun and Wiangwiset, Thalerngsak and Tong-Gate, Wichian and Siritaratiwat, Apirat},
  title = {Comparative Analysis of Machine Learning Models for Water Level Forecasting towards Flood Early Warning in the Mekong River},
  journal = {ECTI Transactions on Electrical Engineering, Electronics, and Communications},
  year = {2026},
  volume = {25},
  number = {1},
  doi = {10.37936/ecti-eec.2027251.262931}
}
Kostenlos starten →