Sezen et al. (2025) A multi-hybrid model approach optimizing discharge forecasts in karst catchment under climate change

Identification

• Journal: Journal of Hydrology
• Year: 2025
• Date: 2025-11-20
• Authors: Cenk Sezen, Mojca Šraj
• DOI: 10.1016/j.jhydrol.2025.134620

Research Groups

• Ondokuz Mayis University, Faculty of Engineering, Samsun, Turkey
• University of Ljubljana, Faculty of Civil and Geodetic Engineering, Ljubljana, Slovenia

Short Summary

This study developed a novel multiple hybrid hydrological model (TUW-CemaNeige GR5J-SR-GP) to enhance daily rainfall-runoff simulations in the complex karst Ljubljanica River catchment, demonstrating significant improvements in simulating extreme flows under both observed and future climate change scenarios.

Objective

• To develop and evaluate a novel multiple hybrid hydrological model (TUW-CemaNeige GR5J-SR-GP) to enhance daily rainfall-runoff simulations, particularly for extreme flows, in the karst Ljubljanica River catchment under observed and future climate change scenarios (Representative Concentration Pathway 4.5 and 8.5).
• To reveal the mathematical relationships between model inputs and discharge outputs through Symbolic Regression-Genetic Programming (SR-GP), enabling interpretation of variable importance and hydrological significance.
• To enlarge the hybrid modelling framework by implementing Grey Wolf Optimization (GWO) for parameter calibration, Ensemble Empirical Mode Decomposition (EEMD) for input decomposition, and Recursive Feature Elimination (RFE) for feature selection.

Study Configuration

• Spatial Scale: Ljubljanica River catchment, Slovenia, with an area of approximately 1890 square kilometers and elevations ranging from 300 to 1800 meters above sea level.
• Temporal Scale:
  • Observed Data: Daily data from January 1, 2010, to December 31, 2023.
  • Projected Data: Daily data for the period 2010–2100, divided into four sub-periods: 2010–2023, 2024–2050, 2049–2075, and 2074–2100.
  • Data Split: 70% for calibration/training and 30% for validation/testing, with a 1-year warm-up period for the CemaNeige GR5J model.

Methodology and Data

• Models used:
  • Conceptual Hydrological Models: Technische Universität Wien (TUW), Génie Rural à 5 paramètres Journalier (GR5J) integrated with the CemaNeige snow module (CemaNeige GR5J).
  • Data-Driven Model: Symbolic Regression-Genetic Programming (SR-GP).
  • Hybrid Model: TUW-CemaNeige GR5J-SR-GP.
  • Optimization Algorithm: Grey Wolf Optimisation (GWO) for model parameter calibration.
  • Data Decomposition: Ensemble Empirical Mode Decomposition (EEMD) for input data.
  • Feature Selection: Recursive Feature Elimination (RFE).
  • Sensitivity Analysis: Monte Carlo sensitivity analysis based on Latin hypercube sampling (Sobol' indices).
• Data sources:
  • Areal Precipitation: Daily data from Topol pri Medvodah, Logatec, Postojna, Ljubljana, and Nova vas na Blokah meteorological stations, processed using the Thiessen polygon method.
  • Mean Air Temperature: Daily data from Ljubljana and Postojna meteorological stations.
  • Discharge: Daily data from the Moste discharge gauging station on the Ljubljanica River.
  • Potential Evapotranspiration (PET): Calculated daily using the Oudin equation.
  • Climate Projections: Averaged daily projected precipitation, air temperature, and potential evapotranspiration from four Global Climate Models (GCMs) and Regional Climate Models (RCMs) (CM1, CM2, CM3, CM4) under Representative Concentration Pathway (RCP) 4.5 and 8.5 scenarios.

Main Results

• The hybrid TUW-CemaNeige GR5J-SR-GP model achieved superior daily discharge simulation performance for the observed dataset, with Nash-Sutcliffe Efficiency (NSE) and Kling-Gupta Efficiency with knowable-moments (KGEkm) values greater than or equal to 0.90.
• For the observed dataset, the hybrid model significantly improved low-flow (baseflow) simulation performance by 41% for the TUW model and 36% for the CemaNeige GR5J model (based on NSE).
• The hybrid model also enhanced monthly peak discharge simulation performance for the observed dataset by 5% for the TUW model and 13% for the CemaNeige GR5J model (based on NSE).
• Under RCP4.5 and RCP8.5 climate change scenarios, the hybrid model improved discharge simulations by up to 14% in the validation period compared to stand-alone conceptual models.
• Under climate change scenarios, baseflow simulation performance improved by up to 28% (TUW) and 12% (CemaNeige GR5J), while high-flow performance improved by up to 32% (TUW) and 31% (CemaNeige GR5J) using the hybrid model.
• Symbolic Regression-Genetic Programming (SR-GP) equations revealed that for the observed dataset, the TUW model's surface runoff component (Q0) and the CemaNeige GR5J model's routing store outflow component (QR) were the most influential variables on discharge.
• Under climate change scenarios, the CemaNeige GR5J model's slow-flow component (QR) and the TUW model's subsurface flow (Q1) or baseflow (Q2) components were identified as dominant predictors for discharge simulation.
• Sensitivity analysis using Sobol' indices indicated that the influence of the CemaNeige GR5J's QR component on discharge simulation significantly increases from 2010–2023 to 2074–2100 under future climate change scenarios.

Contributions

• Introduction of a novel multiple hybrid modelling approach (TUW-CemaNeige GR5J-SR-GP) for daily rainfall-runoff forecasting in a complex karst catchment.
• Comprehensive assessment of the hybrid model's performance under both observed and future climate change scenarios (RCP4.5 and RCP8.5), demonstrating its applicability for discharge projections.
• Integration of advanced techniques including Grey Wolf Optimization (GWO) for parameter calibration, Ensemble Empirical Mode Decomposition (EEMD) for input decomposition, and Recursive Feature Elimination (RFE) for feature selection within the hybrid framework.
• Explicit consideration of snow accumulation and melt processes through the TUW and CemaNeige GR5J models, which are critical in the study catchment.
• Derivation of algebraic equations from the SR-GP model, providing interpretable mathematical relationships between model inputs and discharge outputs, and enabling hydrological significance analysis.
• Detailed evaluation of the hybrid and stand-alone models' performance in simulating extreme runoff events (low and high flows).

Funding

• Slovenian Research and Innovation Agency (ARIS) (Grant P2-0180)
• Slovenian national committee of the IHP UNESCO research programme (UNESCO IHP C3330-20-456010)
• UNESCO Chair on Water-related Disaster Risk Reduction

Citation

@article{Sezen2025multihybrid,
  author = {Sezen, Cenk and Šraj, Mojca},
  title = {A multi-hybrid model approach optimizing discharge forecasts in karst catchment under climate change},
  journal = {Journal of Hydrology},
  year = {2025},
  doi = {10.1016/j.jhydrol.2025.134620},
  url = {https://doi.org/10.1016/j.jhydrol.2025.134620}
}