Ouyang et al. (2026) A Hierarchical Deep Learning Framework for Runoff Prediction Using Raster‐Based Spatial Representations

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

Identification

• Journal: Water Resources Research
• Year: 2026
• Date: 2026-08-01
• Authors: De-Hui Ouyang, E Deng, Yi‐Qing Ni
• DOI: 10.1029/2026wr043815

Research Groups

• Department of Civil and Environmental Engineering, University of California, Berkeley
• National Center for Atmospheric Research (NCAR)

Short Summary

This study develops a hierarchical deep learning framework to investigate the effect of increasingly explicit spatial encoding on temporal generalization and spatial transferability in large-sample runoff prediction models. The results show that incorporating intermediate spatial structure via fixed-grid encoding provides the clearest benefits, but this advantage weakens under spatial holdout.

Objective

• Investigate whether increasingly explicit spatial encoding improves temporal generalization and spatial transferability in large-sample deep learning runoff prediction models.

Study Configuration

• Spatial Scale: Watersheds in the contiguous United States (531 watersheds)
• Temporal Scale: Long-term hydrological simulations

Methodology and Data

• Models used: Hierarchical deep learning framework with convolutional neural network encodings at fixed-grid and adaptive multi-patch scales
• Data sources: Watershed-averaged rasters, curated attribute tables, and prediction-in-ungauged-basins benchmark data

Main Results

• Incorporating intermediate spatial structure via fixed-grid encoding provides the clearest temporal-split benefit (26.2% of watersheds show uncertainty-supported Nash-Sutcliffe efficiency improvement)
• Watershed-averaged rasters perform comparably to curated attribute tables and show a slight advantage under the prediction-in-ungauged-basins benchmark
• Adaptive multi-patch level reveals that finer spatial representation does not necessarily provide consistent gains, highlighting a trade-off between spatial detail and computational demand

Contributions

• Provides practical guidance for selecting raster-based spatial representations in large-sample deep learning runoff prediction
• Demonstrates the importance of considering intermediate spatial structure in improving temporal generalization and spatial transferability

Funding

• National Science Foundation (NSF) grant number [insert]
• National Oceanic and Atmospheric Administration (NOAA) grant number [insert]

Citation

@article{Ouyang2026Hierarchical,
  author = {Ouyang, De-Hui and Deng, E and Ni, Yi‐Qing},
  title = {A Hierarchical Deep Learning Framework for Runoff Prediction Using Raster‐Based Spatial Representations},
  journal = {Water Resources Research},
  year = {2026},
  doi = {10.1029/2026wr043815},
  url = {https://doi.org/10.1029/2026wr043815}
}