Development of hybrid machine learning models for improved streamflow modelling in South-Western Ghats, India
The performance of the Soil and Water Assessment Tool (SWAT), standalone machine learning (ML) models, and hybrid SWAT-ML approaches was evaluated for five river basins in Kerala, India. The ML models considered include Support Vector Machine (SVM), Random Forest (RF), Extreme Gradient Boosting (XGB), and a Deep Multiple Learning (DMKL) framework. Hybrid models were developed by integrating SWAT-simulated streamflow as an additional input to the ML algorithms. Results indicate that hybrid SWAT-ML models consistently outperform standalone SWAT and individual ML models across all basins during both training and testing periods. RF-SWAT, SVM-SWAT, and XGB-SWAT exhibit higher NSE values, stronger correlations with observed streamflow, and reduced bias, especially in capturing peak flows and extreme events. Scatter plots and hydrograph comparisons reveal improved results for hybrid models, highlighting their enhanced ability to represent nonlinear hydrological responses. Overall, hybrid-model provides a robust framework for improving streamflow simulation.
Authors
- Rajat Kumar Sharma (ORCID: https://orcid.org/0000-0003-2599-8551)
- Arka Roy (ORCID: https://orcid.org/0000-0002-4603-7760)
- K. Sreelash (ORCID: https://orcid.org/0000-0002-1384-9287)
- Guguloth Praveen Kumar
- N. Renjima
Institutions
- National Centre for Earth Science Studies (IN)
Publication Details
- Journal
- Hydrological Sciences Journal
- Published
- 2026-11-17
- DOI
- https://doi.org/10.1080/02626667.2026.2762191
- Primary Topic
- Hydrological Forecasting Using AI
- Type
- article
- Field-Weighted Citation Impact
- 0.00
Funders
- National Centre for Earth Science Studies