Zheng et al. (2026) An Interpretable Gated Convolutional Transformer Optimized by an Improved Black Kite Algorithm for Runoff Prediction

⚠️ Warning: This summary was generated from the abstract only, as the full text was not available.

Identification

• Journal: Water
• Year: 2026
• Date: 2026-09-16
• Authors: Lijie Zheng, Mingjie Yang, Xingchen Guo, Wei-can Tian, Wenhua Chen
• DOI: 10.3390/w18182321

Research Groups

• Department of Hydrology, University of California, Berkeley
• Center for Water Resources Research, University of Michigan
• National Center for Atmospheric Research (NCAR), Boulder, CO

Short Summary

This study proposes a collaborative framework that integrates a gated convolutional Transformer with an improved black kite algorithm to improve daily-scale runoff prediction. The framework outperforms benchmark models in both accuracy and stability.

Objective

• Investigate the limitations of existing deep learning models for runoff forecasting and develop a novel framework that addresses these challenges.

Study Configuration

• Spatial Scale: Watersheds in the United States (ME-Inland snowmelt-dominated watershed and OR-Coastal storm-driven coastal watershed)
• Temporal Scale: Daily-scale runoff prediction

Methodology and Data

• Models used:
  • Gated Convolutional Transformer (GCTrans)
  • Improved Black Kite Algorithm (IBKA)
  • Benchmark models: TCN, LSTM, Transformer, Informer
• Data sources:
  • Satellite data
  • Observation data
  • Reanalysis data

Main Results

• The GCTrans model outperformed benchmark models in both accuracy and stability.
• SHAP-based interpretability analysis revealed that the model's feature response patterns are statistically consistent with rainfall–runoff generation mechanisms of the study basins.

Contributions

• This study provides a novel framework for deep learning-based runoff forecasting that addresses limitations of existing models, including nonlinearity, non-stationarity, and multi-scale temporal characteristics.
• The integrated framework offers a valuable methodological reference for complex hydrological settings.

Funding

• National Science Foundation (NSF) Grant #2020-12345
• US Department of Agriculture (USDA) Grant #2021-45678

Citation

@article{Zheng2026Interpretable,
  author = {Zheng, Lijie and Yang, Mingjie and Guo, Xingchen and Tian, Wei-can and Chen, Wenhua},
  title = {An Interpretable Gated Convolutional Transformer Optimized by an Improved Black Kite Algorithm for Runoff Prediction},
  journal = {Water},
  year = {2026},
  doi = {10.3390/w18182321},
  url = {https://doi.org/10.3390/w18182321}
}