Zhao et al. (2025) Linking deterministic and probabilistic paradigms: a peak-sensitive prediction framework for heterogeneous runoff processes

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

Identification

• Journal: Geomatics Natural Hazards and Risk
• Year: 2025
• Date: 2025-11-28
• Authors: Xuehua Zhao, Jiatong An, Bowen Zhu, Qiucen Guo, Huifang Wang, Xiaoqi Guo
• DOI: 10.1080/19475705.2025.2588704

Research Groups

Not specified in the provided text.

Short Summary

This study introduces a peak-sensitive hybrid framework combining time-varying filtering-based empirical mode decomposition (TVF-EMD) with deep learning to improve seasonal runoff forecasting under hydroclimatic nonstationarity and human regulation. The framework delivers superior point predictions and well-calibrated, peak-aware prediction intervals, supporting risk-informed water-resources management.

Objective

• To develop a peak-sensitive hybrid framework for seasonal runoff forecasting that can handle hydroclimatic nonstationarity and human regulation, providing both superior point predictions and robust, well-calibrated prediction intervals.

Study Configuration

• Spatial Scale: Not specified in the provided text.
• Temporal Scale: Monthly.

Methodology and Data

• Models used: Time-varying filtering–based empirical mode decomposition (TVF-EMD), Bayesian optimization (BO), Convolutional Neural Network (CNN), Bidirectional Gated Recurrent Unit (BiGRU), Self-attention (SA), Observation-linked error correction (OLEC), Quantile Regression (QR), Error-sensitive focal loss (ESFL).
• Data sources: Monthly runoff observations (specific source not detailed).

Main Results

• The deterministic backbone (BO-tuned CNN-BiGRU-SA with OLEC) achieved superior point predictions across four metrics: Willmott’s index (WI) ranged from 0.9990 to 0.9997, Nash-Sutcliffe efficiency (NSE) from 0.9960 to 0.9988, with low mean absolute error (MAE), and percent bias (PBIAS) between -1.1682% and 0.7844%.
• For the complex component, quantile regression with an error-sensitive focal loss produced calibrated and sharper prediction intervals.
• At 90% nominal coverage, the prediction interval coverage probability (PICP) spanned 98.6111% to 100%, and the prediction interval normalized average width (PINAW) spanned 6.8663% to 23.9493%.
• The framework consistently yielded lower Winkler scores compared to the BO-CNN-BiGRU-SA-QR baseline.
• Overall, the framework provides narrower, well-calibrated, and peak-aware prediction intervals.

Contributions

• Introduction of a novel peak-sensitive hybrid framework that effectively couples time-varying filtering–based empirical mode decomposition with deep learning (CNN-BiGRU-SA) for seasonal runoff forecasting.
• Development of a robust methodology that addresses the challenges of hydroclimatic nonstationarity and human regulation in runoff prediction.
• Integration of an observation-linked error correction mechanism for deterministic forecasts and quantile regression with an error-sensitive focal loss for producing calibrated and sharper prediction intervals.
• Provides a comprehensive solution that offers both highly accurate point predictions and reliable, peak-aware uncertainty quantification, enhancing risk-informed water-resources management.

Funding

Not specified in the provided text.

Citation

@article{Zhao2025Linking,
  author = {Zhao, Xuehua and An, Jiatong and Zhu, Bowen and Guo, Qiucen and Wang, Huifang and Guo, Xiaoqi},
  title = {Linking deterministic and probabilistic paradigms: a peak-sensitive prediction framework for heterogeneous runoff processes},
  journal = {Geomatics Natural Hazards and Risk},
  year = {2025},
  doi = {10.1080/19475705.2025.2588704},
  url = {https://doi.org/10.1080/19475705.2025.2588704}
}