Wang et al. (2026) Interpretable hierarchical Bayesian modeling of monthly streamflow for heterogeneous basins: A comparative study of two basins

Identification

• Journal: Journal of Hydrology
• Year: 2026
• Date: 2026-03-27
• Authors: Hui Wang, Manoj Shrestha, Tirusew Asefa, Dingbao Graduation Wang
• DOI: 10.1016/j.jhydrol.2026.135320

Research Groups

• Tampa Bay Water, Clearwater, FL, United States
• Department of Civil, Environmental and Construction Engineering, University of Central Florida, Orlando, FL, United States
• Patel College of Global Sustainability, University of South Florida, Tampa, FL, United States

Short Summary

This study develops and compares two interpretable Bayesian hierarchical models (BHMs) for simulating monthly streamflow in two heterogeneous basins, demonstrating their ability to leverage shared information while capturing basin-specific storage dynamics and seasonal variations. The models provide an interpretable framework for understanding storage-driven responses and offer a robust pathway for operational streamflow forecasting in multi-basin settings.

Objective

• To investigate and compare two interpretable Bayesian hierarchical models (BHMs), M0 (baseline) and M1 (with seasonal adjustments), for simulating monthly streamflow in heterogeneous basins, specifically assessing their ability to capture both cross-basin commonalities and basin-specific storage contrasts and seasonal regimes.

Study Configuration

• Spatial Scale: Two heterogeneous basins in the Tampa Bay region, Florida, USA: the Alafia River (ALA) and the Hillsborough River at Morris Bridge (MBR).
• Temporal Scale: Monthly streamflow data. Models were fitted and evaluated using data from 1981 to 2018, with skill comparison for a validation period from 2019 to 2024.

Methodology and Data

• Models used:
  • Bayesian Hierarchical Models (BHMs)
  • M0: A multibasin baseline BHM without explicit seasonality.
  • M1: An augmented version of M0 that includes seasonal intercept adjustments, dividing months into summer and non-summer periods.
• Data sources: Observed monthly streamflow and rainfall data for the specified basins and periods (1981–2024).

Main Results

• The developed BHMs effectively capture both cross-basin commonalities and distinct storage-driven responses across heterogeneous basins.
• Model M1, which incorporates seasonal intercept adjustments, significantly reduces conditional bias in seasonal regimes with minimal loss of parsimony compared to M0.
• The model parameters (persistence coefficients for lagged streamflow, parameters for lagged rainfall, and seasonal intercepts) possess direct hydrologic interpretations, reflecting basin retention, delayed runoff contributions, and seasonal level shifts, respectively.
• Partial pooling across the Alafia River and Hillsborough River basins stabilizes inference while preserving distinct lag weights that reveal contrasting storage behavior.
• Model performance was evaluated using posterior predictive checks, including mean square error, correlation between observed and simulated values, and Pareto-Smoothed Importance Sampling Leave-One-Out (PSIS-LOO) cross-validation.

Contributions

• Development of an interpretable Bayesian hierarchical simulation framework that explains differences in storage-driven responses across nearby, heterogeneous basins.
• Provision of a general and reproducible approach for robust operational streamflow forecasting, particularly valuable in multi-basin settings where shared climate forcing interacts with basin-specific storage and routing characteristics.

Funding

No explicit funding information was provided in the paper text.

Citation

@article{Wang2026Interpretable,
  author = {Wang, Hui and Shrestha, Manoj and Asefa, Tirusew and Wang, Dingbao Graduation},
  title = {Interpretable hierarchical Bayesian modeling of monthly streamflow for heterogeneous basins: A comparative study of two basins},
  journal = {Journal of Hydrology},
  year = {2026},
  doi = {10.1016/j.jhydrol.2026.135320},
  url = {https://doi.org/10.1016/j.jhydrol.2026.135320}
}