Zhang et al. (2026) An Enhanced Informer Deep Learning Model for Nationwide Groundwater Level Predictions: A Comparative Study Across 34 Monitoring Stations in China

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

Identification

• Journal: Hydrology
• Year: 2026
• Date: 2026-06-08
• Authors: Y Zhang, Gan Luo, Yanxia Liu
• DOI: 10.3390/hydrology13060149

Research Groups

[Not specified in the provided text]

Short Summary

The study develops a dual-path Informer-p model integrating residual theory to improve the accuracy, generalization, and interpretability of long-sequence groundwater level predictions across diverse ecosystems in China.

Objective

• To overcome the limitations of traditional deep learning models in long-term groundwater forecasting, specifically regarding weak generalization, reduced long-term accuracy, and limited interpretability.

Study Configuration

• Spatial Scale: National scale (China), encompassing 34 monitoring stations across five major ecosystems.
• Temporal Scale: Long-term groundwater observation sequences.

Methodology and Data

• Models used: Informer-p (a dual-path model consisting of a nonlinear main path for temporal dependencies and a linear residual path for stability), and SHAP (SHapley Additive exPlanations) for feature interpretability.
• Data sources: Long-term groundwater level observations from 34 monitoring stations.

Main Results

• Predictive Performance: At representative stations (e.g., Ailao Mountain), the Informer-p model achieved an RMSE of 0.05 m, MAPE of 1.2%, $R^2$ of 0.95, and KGE of 0.95.
• Comparative Improvement: Compared to the original Informer model, Informer-p reduced RMSE by 37.5% and MAPE by 52% at representative sites and outperformed the original model at 22 of the 34 stations, with the highest improvements seen in forest ecosystems.
• Feature Importance: SHAP analysis identified the window maximum, original groundwater level, and window minimum as the most dominant predictive features.

Contributions

• Proposes a novel dual-path architecture that combines nonlinear temporal dependency capturing with a stable linear baseline, enhancing the model's robustness to extreme events and its ability to represent local fluctuations in groundwater levels.

Funding

[Not specified in the provided text]

Citation

@article{Zhang2026Enhanced,
  author = {Zhang, Y and Luo, Gan and Liu, Yanxia},
  title = {An Enhanced Informer Deep Learning Model for Nationwide Groundwater Level Predictions: A Comparative Study Across 34 Monitoring Stations in China},
  journal = {Hydrology},
  year = {2026},
  doi = {10.3390/hydrology13060149},
  url = {https://doi.org/10.3390/hydrology13060149}
}