Wang et al. (2025) A Deep State Space Model for Rainfall‐Runoff Simulations

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

Identification

• Journal: Water Resources Research
• Year: 2025
• Date: 2025-12-01
• Authors: Yihan Wang, Lujun Zhang, Annan Yu, N. Benjamin Erichson, Tiantian Yang
• DOI: 10.1029/2025wr039888

Research Groups

Not available in the provided abstract.

Short Summary

This study introduces the Frequency Tuned Diagonal State Space Sequence (S4D-FT) model for rainfall-runoff simulations, benchmarking it against LSTM and a physically-based model across 531 watersheds in the contiguous United States. Results indicate S4D-FT generally outperforms LSTM, especially in snowmelt-driven or intermittent flow regimes, but shows limitations in flashier, high-magnitude flow conditions.

Objective

• To propose and evaluate the State Space Model (SSM), specifically the Frequency Tuned Diagonal State Space Sequence (S4D-FT) model, for rainfall-runoff simulations.
• To benchmark S4D-FT against the established Long Short-Term Memory (LSTM) network and a physically-based Sacramento Soil Moisture Accounting model under in-sample and out-of-sample simulation setups.

Study Configuration

• Spatial Scale: 531 watersheds in the contiguous United States (CONUS).
• Temporal Scale: Not explicitly stated in the abstract.

Methodology and Data

• Models used: Frequency Tuned Diagonal State Space Sequence (S4D-FT), Long Short-Term Memory (LSTM), Sacramento Soil Moisture Accounting (SAC-SMA).
• Data sources: Rainfall and runoff data for simulation and benchmarking; specific sources (e.g., satellite, observation, reanalysis) are not explicitly stated in the abstract.

Main Results

• The S4D-FT model is able to outperform the LSTM model across diverse regions under both in-sample and out-of-sample simulation setups.
• S4D-FT demonstrates particular effectiveness in regions characterized by snowmelt-driven or intermittent flow regimes.
• Conversely, S4D-FT tends to underperform in flashier, high-magnitude flow regimes.
• This underperformance is attributed to S4D-FT's global state-space convolution computation, which emphasizes slow, storage-driven dynamics, making it less effective at capturing short bursts and noisy spikes in the data.

Contributions

• Pioneering introduction of the S4D-FT model for rainfall-runoff simulations in the hydrology community.
• Challenges the long-standing dominance of LSTM as the benchmark deep learning model for this task.
• Expands the available arsenal of deep learning tools for hydrological modeling.

Funding

Not available in the provided abstract.

Citation

@article{Wang2025Deep,
  author = {Wang, Yihan and Zhang, Lujun and Yu, Annan and Erichson, N. Benjamin and Yang, Tiantian},
  title = {A Deep State Space Model for Rainfall‐Runoff Simulations},
  journal = {Water Resources Research},
  year = {2025},
  doi = {10.1029/2025wr039888},
  url = {https://doi.org/10.1029/2025wr039888}
}