Kulbekova et al. (2026) Machine Learning Classification of Elevated Discharge in the Koksu River Basin, Kazakhstan: Benchmarking Against Persistence and Illustrative DEM-Based Inundation Scenarios

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

Identification

• Journal: Water
• Year: 2026
• Date: 2026-09-21
• Authors: Sholpan Kulbekova, Abzal Kalygulov, Ranida Arystanova, Asset Arystanov, О.С. Курманбаев, A.N. Munaitpasova, Talgat Usmanov, R. V. Yussupov, Jay Sagin, Sangchul Lee
• DOI: 10.3390/w18182352

Research Groups

• Department of Hydrology, University of Kazakhstan
• Institute of Geography, National Academy of Sciences of Kazakhstan

Short Summary

This study compares the performance of machine learning models (Random Forest, XGBoost, and LSTM) for daily elevated-discharge classification in a Central Asian watershed, finding that simple persistence outperforms these models. The results highlight the importance of routine benchmarking against persistence when adopting machine learning approaches for early warning systems.

Objective

• Investigate the effectiveness of machine learning classifiers (RF, XGBoost, and LSTM) compared to persistence in predicting daily elevated-discharge events in a data-sparse Central Asian watershed.

Study Configuration

• Spatial Scale: 1614 km2 Koksu River basin, Zhetysu Region, Kazakhstan.
• Temporal Scale: Daily discharge classification for the period 2005–2023.

Methodology and Data

• Models used: Random Forest (RF), XGBoost, Long Short-Term Memory (LSTM)
• Data sources: Verified discharge, precipitation, and temperature records from four monitoring stations.

Main Results

• Simple persistence outperformed machine learning classifiers in predicting daily elevated-discharge events.
• The Critical Success Index (CSI) for the persistence baseline was 0.821 (95% block-bootstrap CI [0.725, 0.891]), narrowly ahead of XGBoost (CSI = 0.813), Random Forest (CSI = 0.809), and LSTM (CSI = 0.763).
• Feature-importance analysis revealed that lagged discharge was the primary driver of predictive skill.

Contributions

• This study highlights the importance of routine benchmarking against persistence when adopting machine learning approaches for early warning systems in Central Asian watersheds.
• The findings underscore the necessity of carefully evaluating the performance of machine learning models before implementation.

Funding

• This research was funded by the Kazakhstan Ministry of Education and Science (project code: 0117RK00834) and the National Academy of Sciences of Kazakhstan.

Citation

@article{Kulbekova2026Machine,
  author = {Kulbekova, Sholpan and Kalygulov, Abzal and Arystanova, Ranida and Arystanov, Asset and Курманбаев, О.С. and Munaitpasova, A.N. and Usmanov, Talgat and Yussupov, R. V. and Sagin, Jay and Lee, Sangchul},
  title = {Machine Learning Classification of Elevated Discharge in the Koksu River Basin, Kazakhstan: Benchmarking Against Persistence and Illustrative DEM-Based Inundation Scenarios},
  journal = {Water},
  year = {2026},
  doi = {10.3390/w18182352},
  url = {https://doi.org/10.3390/w18182352}
}