Bhardwaj et al. (2026) Evaluating Data-Driven and Physically-Based Models for Streamflow Forecasting in a Himalayan Catchment

Identification

• Journal: Lecture notes in civil engineering
• Year: 2026
• Date: 2026-01-01
• Authors: Shyam Sundar Bhardwaj, Madan Kumar Jha, Ataur Rahman
• DOI: 10.1007/978-3-032-18708-6_11

Research Groups

• AgFE Department, IIT Kharagpur, Kharagpur, West Bengal, India
• School of Engineering, Design and Built Environment, Western Sydney University, Sydney, Australia

Short Summary

This study compares the performance of a physically-based model (SWAT) with data-driven machine learning models (XGBoost, Random Forest, LSTM) for daily streamflow forecasting in the Bhagirathi River Basin, a Himalayan catchment. The findings indicate that data-driven models, particularly Random Forest, outperform the SWAT model, highlighting their potential for operational hydrological forecasting in data-scarce, flood-prone mountainous regions.

Objective

• To compare the performance of the physically-based Soil and Water Assessment Tool (SWAT) model with data-driven machine learning models (XGBoost, Random Forest, and Long Short-Term Memory networks) for streamflow simulation in the Bhagirathi River Basin.

Study Configuration

• Spatial Scale: Bhagirathi River Basin, a Himalayan catchment.
• Temporal Scale: Daily.

Methodology and Data

• Models used: Soil and Water Assessment Tool (SWAT), XGBoost, Random Forest, Long Short-Term Memory (LSTM) networks.
• Data sources: Hydrological data (streamflow observations and meteorological inputs, implied for model training and evaluation).

Main Results

• The SWAT model showed good performance during calibration (Nash-Sutcliffe Efficiency (NSE) = 0.79, Root Mean Square Error (RMSE) = 61.95 m³/s, Percent Bias (PBIAS) = 0.96%) and validation (NSE = 0.74, RMSE = 79.75 m³/s, PBIAS = −5.58%).
• Among the machine learning models, Random Forest demonstrated superior performance in validation, achieving an NSE of 0.872 and the lowest RMSE of 55.8 m³/s.
• Overall, data-driven machine learning models, especially Random Forest, outperformed the physically-based SWAT model for streamflow forecasting in the study area.

Contributions

• Provides a comparative evaluation demonstrating the superior performance of data-driven machine learning models (Random Forest, XGBoost, LSTM) over a traditional physically-based model (SWAT) for streamflow forecasting in a complex, data-scarce Himalayan catchment.
• Underscores the potential for integrating data-driven machine learning models into operational hydrological forecasting systems and adaptive water management strategies, particularly in flood-prone mountainous basins.

Funding

• Not explicitly mentioned in the provided text.

Citation

@article{Bhardwaj2026Evaluating,
  author = {Bhardwaj, Shyam Sundar and Jha, Madan Kumar and Rahman, Ataur},
  title = {Evaluating Data-Driven and Physically-Based Models for Streamflow Forecasting in a Himalayan Catchment},
  journal = {Lecture notes in civil engineering},
  year = {2026},
  doi = {10.1007/978-3-032-18708-6_11},
  url = {https://doi.org/10.1007/978-3-032-18708-6_11}
}