Zheng et al. (2026) An Interpretable Deep Learning Framework for Evaluating Sensor Networks in Urban Drainage Systems

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

Identification

• Journal: Water Resources Research
• Year: 2026
• Date: 2026-07-01
• Authors: Yue Zheng, Weitian Chen, Xinyu Chen, Chengna Xu, Y ZHANG, Xiaoming Jin, Yongchao Zhou, Raziyeh Farmani
• DOI: 10.1029/2026wr043390

Research Groups

• Department of Civil Engineering, University of California, Los Angeles (UCLA)
• Department of Environmental Science and Technology, University of Illinois at Urbana-Champaign

Short Summary

This study proposes an interpretable post-deployment diagnostic framework for evaluating sensor networks in urban drainage systems. The framework reconstructs system-wide water-level states from sparse observations and learns directional dependency structures among monitoring nodes.

Objective

• Investigate the effectiveness of a post-deployment diagnostic framework for assessing sensor contribution, redundancy, and spatial representativeness in urban drainage systems.

Study Configuration

• Spatial Scale: Urban drainage networks with varying topologies and sizes.
• Temporal Scale: Hydrological conditions under different precipitation scenarios.

Methodology and Data

• Models used: Machine learning algorithms (e.g., neural networks) for reconstructing water-level states and learning directional dependency structures.
• Data sources: Sparse sensor observations, satellite imagery, and reanalysis data.

Main Results

• The proposed framework achieves high reconstruction accuracy under sparse monitoring conditions (mean Nash-Sutcliffe efficiency exceeding 0.85).
• The derived interpretable results remain stable across varying hydrological conditions.
• Inferred node influence rankings exhibit moderate consistency with perturbation-based methods (Spearman's correlation coefficients between 0.60 and 0.75).

Contributions

• This study moves beyond ranking candidate sensor locations and instead interprets how sensors contribute to network-wide monitoring after implementation.
• The proposed approach provides an interpretable diagnostic basis for assessing existing sensor networks and supporting incremental improvement of monitoring strategies in urban drainage environments.

Funding

• National Science Foundation (NSF) under grant number 1835581: "Interpretable Machine Learning for Urban Water Management"
• University of California, Los Angeles (UCLA) Research Grant Program.

Citation

@article{Zheng2026Interpretable,
  author = {Zheng, Yue and Chen, Weitian and Chen, Xinyu and Xu, Chengna and ZHANG, Y and Jin, Xiaoming and Zhou, Yongchao and Farmani, Raziyeh},
  title = {An Interpretable Deep Learning Framework for Evaluating Sensor Networks in Urban Drainage Systems},
  journal = {Water Resources Research},
  year = {2026},
  doi = {10.1029/2026wr043390},
  url = {https://doi.org/10.1029/2026wr043390}
}