Guniganti et al. (2025) A hybrid hydrologic modeling framework-role of spatial resolution, calibration approaches and error modeling

Identification

• Journal: Journal of Hydrology Regional Studies
• Year: 2025
• Date: 2025-12-17
• Authors: Surya Kiran Guniganti, Satish Kumar Regonda, Balaji Rajagopalan
• DOI: 10.1016/j.ejrh.2025.103053

Research Groups

• Department of Civil Engineering, IIT Hyderabad, India
• Department of Climate Change, IIT Hyderabad, India
• Department of Civil, Environmental and Architectural Engineering, University of Colorado, Boulder, USA
• Cooperative Institute for Research in Environmental Sciences, University of Colorado, Boulder, USA

Short Summary

This study develops a hybrid hydrologic modeling framework to assess the roles of spatial resolution, multi-site calibration strategies, and machine learning-based error modeling in semi-distributed streamflow prediction for the Narmada River Basin, India. It finds that a medium spatial resolution combined with a novel Sequential Ungauged Basin calibration approach and Random Forest error correction provides a computationally efficient and skillful framework, particularly improving high flow predictions and demonstrating strong potential for ungauged basins.

Objective

• To assess the optimal spatial resolution for a semi-distributed hydrological modeling approach in the Narmada River Basin.
• To evaluate two multi-site calibration techniques (Basin Specific (BS) and Sequential Ungauged Basin (SQ)) for streamflow estimation, considering the role of routed flows from upstream sub-basins.
• To determine if a Random Forest (RF) error-correction model can improve the accuracy of streamflow predictions from the hydrological model.

Study Configuration

• Spatial Scale: Narmada River Basin, India (area: 98,796 km²). Four spatial resolutions were tested, resulting in 7, 21, 29 (medium resolution, MR), and 41 sub-basins, derived from stream definitions ranging from 4600 km² to 1400 km². Digital Elevation Model (DEM) resolution was 30 m. Six gauge-discharge stations were used for evaluation.
• Temporal Scale: Daily time step. Streamflow data from 1980 to 2012. Calibration period: 1991–2012 (with 1991 as warmup). Validation/Testing period: 1980–1990. Analysis focused on the monsoon season (June-September).

Methodology and Data

• Models used:
  • Hydrological model: Sacramento Soil Moisture Accounting (SAC-SMA)
  • Hydraulic routing scheme: Kinematic Wave Routing (KWR)
  • Calibration algorithm: Dynamically Dimensioned Search (DDS)
  • Error modeling: Random Forest (RF) machine learning algorithm
• Data sources:
  • Daily streamflows: Water Resources Information System for India (WRIS).
  • Daily gridded precipitation: India Meteorological Department (IMD) 0.25° × 0.25° product.
  • Daily gridded temperature: IMD 1° × 1° product.
  • Potential Evapotranspiration (PET): Estimated using Hamon’s formula.
  • Digital Elevation Model (DEM): Shuttle Radar Topography Mission (SRTM) 30 m spatial resolution (NASA).
  • River channel parameters: Derived from SRTM DEM and Google Earth (average width), Manning’s roughness coefficient (n) assumed 0.035.

Main Results

• Spatial Resolution: Model performance significantly improved from the coarsest (7 sub-basins) to coarser (21 sub-basins) resolutions, then plateaued for medium (29 sub-basins) and fine (41 sub-basins) resolutions. The medium resolution (MR) was selected as optimal due to similar performance to finer resolutions with reduced computational cost.
• Multi-site Calibration: The novel Sequential Ungauged Basin (SQ) approach consistently outperformed the Basin-Specific (BS) approach. SQ demonstrated Nash-Sutcliffe Efficiency (NSE) improvements ranging from 0.05 to 0.37 for all flows and 0.08 to 0.69 for high flows compared to BS. The SQ approach showed better performance under wet basin conditions (Standardized Precipitation Index (SPI) > 1), indicating its potential for flood forecasting.
• Error Modeling: Random Forest (RF) error modeling substantially improved overall model performance, particularly for the BS approach, with NSE improvements exceeding 20% for three upstream stations during the training period. Percentage Bias (PBIAS) values decreased by approximately 90% in the training period. Improvements were more evident for lower 80th percentile flows, while high flows (above 80th percentile) showed mixed or decreased performance in the testing period. Rainfall, impervious runoff, and interflow were identified as key variables influencing error estimation.
• Framework Validation: The integrated framework (MR configuration with BS calibration and RF error correction) was found to be computationally efficient and skillful. Validation on two ungauged locations demonstrated its strong potential for reliable streamflow estimation in data-scarce or ungauged interior regions.

Contributions

• Developed and assessed a novel hybrid hydrological modeling framework integrating the semi-distributed SAC-SMA model, Kinematic Wave Routing, and a Random Forest-based error correction module.
• Provided a systematic assessment of spatial resolution impact on semi-distributed model performance, identifying an optimal "medium" resolution that balances accuracy and computational efficiency.
• Introduced and validated a novel Sequential Ungauged Basin (SQ) multi-site calibration approach that explicitly accounts for upstream routed flows and pseudo streamflows for ungauged sub-basins, demonstrating its superior performance, especially for high flows and wet conditions relevant to flood forecasting.
• Quantified the significant improvements in streamflow prediction accuracy achieved through Random Forest-based error modeling, particularly for lower-performing calibration strategies, and identified key hydrological variables influencing these errors.
• Demonstrated the transferability of calibrated parameters to ungauged and data-scarce basins, highlighting the framework's potential for regionalization and application in similar regions globally.

Funding

• Ministry of Earth Sciences, Government of India [IITM/MM-II/Univ.Colorado/2018/INT-7]
• Fulbright-Kalam Fellowship (for the third author)

Citation

@article{Guniganti2025hybrid,
  author = {Guniganti, Surya Kiran and Regonda, Satish Kumar and Rajagopalan, Balaji},
  title = {A hybrid hydrologic modeling framework-role of spatial resolution, calibration approaches and error modeling},
  journal = {Journal of Hydrology Regional Studies},
  year = {2025},
  doi = {10.1016/j.ejrh.2025.103053},
  url = {https://doi.org/10.1016/j.ejrh.2025.103053}
}