Çırağ et al. (2026) Multi-scale drought analysis and machine learning-based completion of missing streamflow data in the Aras Basin

Identification

• Journal: Physics and Chemistry of the Earth Parts A/B/C
• Year: 2026
• Date: 2026-03-25
• Authors: Burak Çırağ, Cansu Bozkurt
• DOI: 10.1016/j.pce.2026.104410

Research Groups

• Department of Civil Engineering, Faculty of Engineering, Atatürk University, Erzurum, Türkiye
• Department of Construction, Technical Sciences Vocational School, Ardahan University, Ardahan, Türkiye

Short Summary

This study conducted multi-scale meteorological and hydrological drought analyses in the Aras Basin, Türkiye, and developed a machine learning approach (XGBoost) to complete missing streamflow data, demonstrating that data imputation significantly enhances the reliability of early hydrological drought detection.

Objective

• To perform multi-scale meteorological and hydrological drought analyses using precipitation, temperature, evaporation, and streamflow data in the Aras Basin.
• To develop and apply machine learning methods for predicting and completing missing streamflow data to improve the effectiveness and reliability of drought models and early detection.

Study Configuration

• Spatial Scale: Aras Basin, specifically the provinces of Ardahan, Kars, and Erzurum, Türkiye.
• Temporal Scale: Data from 1980 to 2023. Streamflow data gaps occurred after 2011.

Methodology and Data

• Models used:
  • Machine Learning for data imputation: Random Forest (RF), K-Nearest Neighbors (KNN), Support Vector Regression (SVR), eXtreme Gradient Boosting (XGBoost). XGBoost was selected for imputation due to highest performance.
  • Meteorological Drought Indices: Standardized Precipitation Index (SPI), Standardized Precipitation Evapotranspiration Index (SPEI), Reconnaissance Drought Index (RDI).
  • Hydrological Drought Index: Streamflow Drought Index (SDI).
  • Trend Analysis: Mann-Kendall and Sen's Slope techniques.
• Data sources: Observational precipitation, temperature, evaporation, and streamflow data.

Main Results

• The XGBoost model demonstrated the highest performance metrics among tested machine learning methods for predicting and imputing missing streamflow data.
• Accurately completing missing streamflow data significantly increases the reliability of drought analysis.
• The methodology contributes significantly to early hydrological drought detection, particularly in continental climates.
• The study provides a scientific basis for future water resource management and climate change adaptation strategies in the region.

Contributions

• Integration of multi-scale meteorological and hydrological drought analysis with advanced machine learning techniques for streamflow data completion.
• Demonstration of the critical role of accurate data imputation in enhancing the reliability and effectiveness of drought monitoring and early detection systems.
• Specific application and validation of the methodology in the Aras Basin, a continental climate region, addressing a common challenge of incomplete hydrological records.
• Providing actionable insights for water resource management and climate change adaptation strategies based on improved drought analysis.

Funding

• Not mentioned in the provided text.

Citation

@article{Çırağ2026Multiscale,
  author = {Çırağ, Burak and Bozkurt, Cansu},
  title = {Multi-scale drought analysis and machine learning-based completion of missing streamflow data in the Aras Basin},
  journal = {Physics and Chemistry of the Earth Parts A/B/C},
  year = {2026},
  doi = {10.1016/j.pce.2026.104410},
  url = {https://doi.org/10.1016/j.pce.2026.104410}
}