Neagoe et al. (2025) Hybrid LSTM-ARIMA Model for Improving Multi-Step Inflow Forecasting in a Reservoir

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

Identification

• Journal: Water
• Year: 2025
• Date: 2025-10-24
• Authors: Angela Neagoe, Eliza-Isabela Tică, Liana Ioana Vuţă, Otilia Nedelcu, Gabriela Elena Dumitran, Bogdan Popa
• DOI: 10.3390/w17213051

Research Groups

Not explicitly stated in the provided text, but the study focuses on the Izvorul Muntelui–Bicaz reservoir in Romania, suggesting involvement of local or national hydrological or energy research entities.

Short Summary

This study proposes a novel hybrid LSTM-ARIMA model for short-term reservoir inflow prediction, demonstrating significant improvements in accuracy (R² from 0.93 to 0.96, RMSE from 9.74 m³/s to 6.94 m³/s for one-day-ahead forecasts) over standalone LSTM, particularly for multi-step predictions.

Objective

• To develop and evaluate a novel hybrid LSTM-ARIMA model for accurate short-term reservoir inflow prediction, especially in data-scarce environments, by first capturing nonlinear hydrological patterns with LSTM and then modeling residual linear trends with ARIMA.

Study Configuration

• Spatial Scale: Izvorul Muntelui–Bicaz reservoir in Romania.
• Temporal Scale: Daily inflow data from 2012 to 2020 for calibration; 365 days corresponding to 2021 for one-day-ahead prediction testing; multiple seven-day forecasts randomly selected to cover all 12 months of 2021 for multi-step evaluation.

Methodology and Data

• Models used: Hybrid LSTM-ARIMA (LSTM followed by ARIMA for residuals), standalone LSTM (for comparison).
• Data sources: Daily observational inflow data from the Izvorul Muntelui–Bicaz reservoir.

Main Results

• For one-day-ahead forecasting, the proposed hybrid model significantly outperformed standalone LSTM, increasing the R² from 0.93 to 0.96 and reducing the Root Mean Square Error (RMSE) from 9.74 m³/s to 6.94 m³/s.
• For multi-step forecasting (84 randomly selected values, 7 per month), the hybrid model improved R² from 0.75 to 0.89 and lowered RMSE from 18.56 m³/s to 12.74 m³/s.
• The hybrid model effectively captures both seasonal patterns and nonlinear variations in hydrological data, offering notable improvements, especially in multi-step forecasting.

Contributions

• Proposes a novel hybrid LSTM-ARIMA approach that reverses the conventional sequence, using LSTM first to capture nonlinear patterns and then ARIMA to model residual linear trends.
• Demonstrates significant performance improvements over standalone LSTM for both one-day-ahead and multi-step reservoir inflow forecasting.
• Offers a replicable data-driven solution for short-term inflow prediction in reservoirs, particularly valuable in regions with limited physical data on topography, vegetation, and basin characteristics.

Funding

Not mentioned in the provided text.

Citation

@article{Neagoe2025Hybrid,
  author = {Neagoe, Angela and Tică, Eliza-Isabela and Vuţă, Liana Ioana and Nedelcu, Otilia and Dumitran, Gabriela Elena and Popa, Bogdan},
  title = {Hybrid LSTM-ARIMA Model for Improving Multi-Step Inflow Forecasting in a Reservoir},
  journal = {Water},
  year = {2025},
  doi = {10.3390/w17213051},
  url = {https://doi.org/10.3390/w17213051}
}