Panchal et al. (2026) Analysis of ensemble and control forecasts from GEFS and NEPS for reservoir inflow prediction

Identification

• Journal: Stochastic Environmental Research and Risk Assessment
• Year: 2026
• Date: 2026-04-01
• Authors: Ayushi Panchal, S. M. Yadav
• DOI: 10.1007/s00477-026-03200-7

Research Groups

• Department of Civil Engineering, Sardar Vallabhbhai National Institute of Technology (SVNIT), Surat, India

Short Summary

This study comprehensively analyzes and compares the performance of ensemble and control forecasts from GEFS and NEPS for reservoir inflow prediction at the Ukai Reservoir, India. It demonstrates that bias-corrected ensemble forecasts significantly outperform deterministic control forecasts, particularly at longer lead times (1-5 days), thereby enhancing flood risk management and operational decision-making.

Objective

• To conduct an in-depth comparative analysis of deterministic control and probabilistic ensemble forecasts from GEFS and NEPS for reservoir inflow prediction.
• To integrate hydrometeorological precipitation forecasts with a hydrodynamic model to predict reservoir inflows, enabling improved water resources management and flood risk assessment.
• To evaluate and recommend the most effective bias-correction method for simulated reservoir inflows.

Study Configuration

• Spatial Scale: Ukai Reservoir, located on the Tapi River in the Middle Tapi Basin, India, covering an area of 6588 square kilometers (from Sarangkheda to Ukai gauging site). GEFS and NEPS precipitation forecasts were used at a 0.5° × 0.5° resolution.
• Temporal Scale:
  • Hydrodynamic model calibration: 10 years (2007–2016) of daily data.
  • Hydrodynamic model validation: Three major flood events (1998, 2006, 2019).
  • Forecast evaluation: Two recent flood events (September 16 – October 15, 2021, and August 8 – August 25, 2022).
  • Forecast lead times: 1 to 5 days.

Methodology and Data

• Models used:
  • Hydrodynamic Model: MIKE HYDRO River (incorporating the NAM rainfall-runoff model, a lumped conceptual hydrologic model).
  • Ensemble Prediction Systems (EPS):
    • Global Ensemble Forecasting System (GEFS) from the Indian Institute of Tropical Meteorology (IITM), Pune (20 perturbed ensemble members + 1 control forecast).
    • NCMRWF Ensemble Precipitation System (NEPS) from the National Centre for Medium Range Weather Forecasting (NCMRWF), Noida, India (11 perturbed ensemble members + 1 control forecast).
• Data sources:
  • TIGGE (The International Grand Global Ensemble) platform for GEFS and NEPS precipitation forecasts (GRIB format).
  • Observed daily rainfall data from the India Meteorological Department (IMD).
  • Observed daily discharge data from the Central Water Commission (CWC), Surat, and MTBO Gandhinagar.
  • Observed daily evaporation data from Jal Sinchai Bhavan, Surat, and Tapi Jal Bhavan, Ukai.
  • Bias correction methods: Delta Change Method (DCM), Quantile Mapping (QM), and Quantile Delta Mapping (QDM).

Main Results

• The developed hydrodynamic model showed good performance, with a correlation coefficient (R²) of 0.75 during calibration (2007-2016) and R² values of 0.853, 0.925, and 0.866 for the 1998, 2006, and 2019 flood validation events, respectively.
• Bias correction significantly improved forecast accuracy, with the Delta Change Method (DCM) consistently outperforming Quantile Mapping (QM) and Quantile Delta Mapping (QDM). DCM-corrected forecasts achieved R² values exceeding 0.95 and Nash-Sutcliffe Efficiency (NSE) values above 0.90, even for 4- and 5-day lead times, with a 30-45% reduction in Root Mean Square Error (RMSE).
• Ensemble-based inflow forecasting, particularly after DCM bias correction, consistently outperformed control (deterministic) forecasts, especially for longer lead times (2-5 days).
• NEPS demonstrated superior performance for 1-day lead time forecasts, while GEFS performed better for 2- to 5-day lead times.
• Ensemble forecasts provided a valuable range of expected inflows (minimum and maximum estimates), enhancing uncertainty quantification and enabling reservoir operators to plan releases with a 1- to 5-day lead time for flood mitigation.
• Statistical scores (Probability of Detection, POD; False Alarm Rate, FAR; Critical Success Index, CSI) indicated high detection capabilities (high POD) for ensemble forecasts, but FAR generally increased and CSI decreased with increasing lead time, reflecting growing forecast uncertainty.

Contributions

• This is the first known research to apply and comparatively evaluate both GEFS and NEPS data for reservoir inflow forecasting at a global scale, specifically for the Ukai Reservoir in India.
• Provides a comprehensive framework for integrating operational ensemble and control precipitation forecasts with a hydrodynamic model for improved reservoir inflow prediction and flood risk management.
• Identifies the Delta Change Method (DCM) as the most effective bias-correction technique for enhancing the accuracy and reliability of ensemble-based hydrological forecasts in the study region.
• Demonstrates the practical utility of ensemble forecasts in providing uncertainty-aware predictions, which is crucial for real-time reservoir operations and timely flood mitigation measures.
• Offers valuable insights for atmospheric scientists and forecast developers to further refine ensemble precipitation prediction systems, particularly in monsoon-driven hydrological contexts.

Funding

No funds have been received to carry out the present study.

Citation

@article{Panchal2026Analysis,
  author = {Panchal, Ayushi and Yadav, S. M.},
  title = {Analysis of ensemble and control forecasts from GEFS and NEPS for reservoir inflow prediction},
  journal = {Stochastic Environmental Research and Risk Assessment},
  year = {2026},
  doi = {10.1007/s00477-026-03200-7},
  url = {https://doi.org/10.1007/s00477-026-03200-7}
}