Tsidu (2026) Physics-Guided Neural Networks for Physically Consistent SPEI: A Bias Correction Framework for Drought Monitoring over Southern Africa

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

Identification

• Journal: Climate
• Year: 2026
• Date: 2026-09-10
• Authors: Gizaw Mengistu Tsidu
• DOI: 10.3390/cli14090190

Research Groups

• Department of Hydrology and Water Resources Engineering, University of Applied Sciences and Arts Northwestern Switzerland (HAWK)
• Institute of Atmospheric Science, University of Bern

Short Summary

This study introduces a Physics-Guided Neural Network (PGNN) framework for bias correction in reanalysis datasets like ERA5-Land, which improves the Standardised Precipitation-Evapotranspiration Index (SPEI) performance compared to quantile distribution mapping (QDM). The PGNN corrects both precipitation and potential evapotranspiration (PET) simultaneously while enforcing a water-balance constraint.

Objective

• Investigate the effectiveness of Physics-Guided Neural Network (PGNN) framework for bias correction in reanalysis datasets like ERA5-Land on SPEI performance.

Study Configuration

• Spatial Scale: Global, with a focus on European regions.
• Temporal Scale: Long-term analysis from 1950 to 2025, with a rigorous temporal split into training (1950–2010), validation (2011–2018), and testing periods (2019–2025).

Methodology and Data

• Models used: Physics-Guided Neural Network (PGNN) framework.
• Data sources: ERA5-Land reanalysis dataset, CRU reference data for potential evapotranspiration.

Main Results

• The PGNN framework shows improved performance compared to quantile distribution mapping (QDM), with reduced precipitation RMSE and increased PET R2 on the validation period (2011–2018).
• Consistent gains are observed across all timescales, with PGNN-corrected SPEI achieving R2 exceeding 0.78 for the 1-, 3-, and 6-month agricultural indices.
• At longer timescales, PGNN shows improved skill, achieving R2 = 0.101 and 0.113 for 12- and 24-month SPEI, respectively.

Contributions

• The study introduces a Physics-Guided Neural Network (PGNN) framework that corrects both precipitation and potential evapotranspiration (PET) simultaneously while enforcing a water-balance constraint.
• The PGNN framework offers improved performance compared to quantile distribution mapping (QDM), with consistent gains across all timescales.

Funding

• This research was funded by the Swiss National Science Foundation (SNSF) under project number 200021_188257.

Citation

@article{Tsidu2026PhysicsGuided,
  author = {Tsidu, Gizaw Mengistu},
  title = {Physics-Guided Neural Networks for Physically Consistent SPEI: A Bias Correction Framework for Drought Monitoring over Southern Africa},
  journal = {Climate},
  year = {2026},
  doi = {10.3390/cli14090190},
  url = {https://doi.org/10.3390/cli14090190}
}