Park et al. (2025) Assessing the Applicability of the LTSF Algorithm for Streamflow Time Series Prediction: Case Studies of Dam Basins in South Korea

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Water
• Year: 2025
• Date: 2025-11-10
• Authors: Jiyeon Park, Ju‐Young Shin, Sunghun Kim, Jihye Kwon
• DOI: 10.3390/w17223214

Research Groups

[Information not available in the provided text.]

Short Summary

This study systematically assessed the applicability of two Long-Term Time Series Forecasting (LTSF) linear models, NLinear and DLinear, for hydrological inflow prediction at eight major dams in South Korea, comparing their performance against conventional AI models like LSTM and XGBoost. While LSTM generally achieved the highest R2 and lowest NRMSE, DLinear minimized NMSE, and NLinear showed superior hydrological consistency, demonstrating the potential of LTSF models for this domain but highlighting site-specific performance variations.

Objective

• To systematically assess the applicability of Long-Term Time Series Forecasting (LTSF) linear models (NLinear and DLinear) for hydrological inflow prediction.

Study Configuration

• Spatial Scale: Eight major dams in South Korea.
• Temporal Scale: 24 h input window for training, evaluated for 24 h lead times.

Methodology and Data

• Models used: NLinear, DLinear (LTSF linear models); Long Short-Term Memory (LSTM), eXtreme Gradient Boosting (XGBoost) (conventional AI models).
• Data sources: Hydrological inflow data.

Main Results

• LSTM consistently achieved the highest coefficient of determination (R2) and the lowest normalized root mean square error (NRMSE).
• DLinear minimized the normalized mean square error (NMSE).
• NLinear delivered superior hydrological consistency as measured by Kling–Gupta efficiency (KGE).
• XGBoost showed comparatively larger variability across sites.
• Spatial heterogeneity was evident, with sites grouped into high-performing, transition, and vulnerable categories.
• Peak-flow analysis revealed amplitude attenuation and phase lag at longer forecasting horizons.

Contributions

• First systematic assessment of LTSF linear models (NLinear and DLinear) for hydrological inflow prediction.
• Comparative analysis of LTSF models against conventional AI models (LSTM, XGBoost) in the context of dam inflow forecasting.
• Provided insights into the performance characteristics, including spatial heterogeneity and peak-flow behavior, of different models for long-horizon hydrological forecasting.

Funding

[Information not available in the provided text.]

Citation

@article{Park2025Assessing,
  author = {Park, Jiyeon and Shin, Ju‐Young and Kim, Sunghun and Kwon, Jihye},
  title = {Assessing the Applicability of the LTSF Algorithm for Streamflow Time Series Prediction: Case Studies of Dam Basins in South Korea},
  journal = {Water},
  year = {2025},
  doi = {10.3390/w17223214},
  url = {https://doi.org/10.3390/w17223214}
}