Budamala et al. (2026) Hierarchical prediction of streamflow and organic carbon at event scale using a physics-guided AI hybrid framework

Identification

• Journal: Journal of Hydrology
• Year: 2026
• Date: 2026-08-13
• Authors: Venkatesh Budamala, Euiyoung Choi, Minchang Kim, Sangchul Lee, Hyunglok Kim
• DOI: 10.1016/j.jhydrol.2026.136248

Research Groups

Department of Environment and Energy Engineering, Gwangju Institute of Science and Technology, the Republic of Korea
Department of Environmental Science and Ecological Engineering, Korea University, Seoul 02841, the Republic of Korea

Short Summary

This study presents a physics-guided spatial-temporal hybrid framework (PhyST-HC) for hierarchical prediction of streamflow and organic carbon at event scale in agricultural watersheds. The framework integrates uncalibrated SWAT-C simulations with Graph Neural Networks (GNNs) and Transformer-based attention mechanisms.

Objective

• Investigate the potential of physics-guided hybrid learning for accurate and transferable event-scale hydro-carbon prediction in agricultural watersheds.

Study Configuration

• Spatial Scale: Watershed scale, specifically the Tuckahoe Creek Watershed (TCW), USA.
• Temporal Scale: Event scale, with a focus on storm-driven export of streamflow and organic carbon.

Methodology and Data

• Models used: Physics-guided spatial-temporal hybrid framework (PhyST-HC) combining SWAT-C simulations with Graph Neural Networks (GNNs) and Transformer-based attention mechanisms.
• Data sources: Distributed SWAT-C-derived hydrological and carbon-state variables, along with observational data from the Tuckahoe Creek Watershed.

Main Results

• The PhyST-HC framework achieved comparable performance to a SUFI-2-calibrated SWAT-C model for streamflow prediction (KGE = 0.82) without watershed-specific calibration.
• The framework substantially outperformed both SUFI-2 and SWAT-LSTM models for organic carbon prediction, with KGE values of 0.68 for POC and 0.74 for DOC.

Contributions

• This study demonstrates the potential of physics-guided hybrid learning for accurate and transferable event-scale hydro-carbon prediction in agricultural watersheds.
• The framework improves representation of event-scale POC flushing, delayed DOC transport, and watershed connectivity under contrasting storm conditions.

Funding

• This research was funded by the National Research Foundation of Korea (NRF) through the Korea University Grant.

Citation

@article{Budamala2026Hierarchical,
  author = {Budamala, Venkatesh and Choi, Euiyoung and Kim, Minchang and Lee, Sangchul and Kim, Hyunglok},
  title = {Hierarchical prediction of streamflow and organic carbon at event scale using a physics-guided AI hybrid framework},
  journal = {Journal of Hydrology},
  year = {2026},
  doi = {10.1016/j.jhydrol.2026.136248},
  url = {https://doi.org/10.1016/j.jhydrol.2026.136248}
}