Skip to main navigation Skip to search Skip to main content

Decoding dissolved oxygen dynamics using ensemble machine learning models: insights across multi-regional rivers

  • Wang Jing
  • , Ali Omran Al-Sulttani
  • , Raad Z. Homod
  • , Ricky Anak Kemarau
  • , Sajjad Firas Abdulameer
  • , Iliass Naouadir
  • , Omer A. Alawi*
  • , Billel Salhi
  • , Mehvish Bilal
  • , Zaher Mundher Yaseen
  • *Corresponding author for this work

Research output: Contribution to journal › Article › peer-review

Abstract

Dissolved oxygen (DO) is a key indicator of river water quality (RWQ) as it reflects aquatic ecosystem health, microbial and biological activity, and the balance of physicochemical processes within river systems. The study evaluates ensemble machine learning (ML) models to predict DO in the Brisbane River (Australia), the Piabanha River (Brazil), and two river systems in Hong Kong. Random Forest (RF), Extreme Gradient Boosting (XGBoost), Light Gradient-Boosting Machine (LightGBM), Extra Trees (ET), and Gradient Boosting Decision Trees (GBDT) models were developed using hierarchical feature combinations to assess model robustness under varying input dimensionality. Results indicated that the RF model achieved the best generalization performance for the Australian dataset, with a coefficient of determination (R2 = 0.87) during the testing phase, whereas the ET model achieved the best performance in Brazil (R2 = 0.95). In the case of Hong Kong, the ET model achieved the highest performance (R2 = 0.94) at Pai Min Kok, whereas the LightGBM model showed consistent performance at Kai Tak River (R2 = 0.85–0.86). Further, research results indicated that complete multivariate predictor sets generally improved ML model prediction accuracy. Overall, the ensemble models effectively captured the nonlinear variability of DO across diverse river systems, demonstrating their potential for data-driven river water quality assessment and prediction.

Original languageEnglish
Article number136343
JournalJournal of Hydrology
Volume680
DOIs
StatePublished - Nov 2026

Bibliographical note

Publisher Copyright:
© 2026 Elsevier B.V. All rights are reserved, including those for text and data mining, AI training, and similar technologies.

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

  1. SDG 3 - Good Health and Well-being
    SDG 3 Good Health and Well-being
  2. SDG 6 - Clean Water and Sanitation
    SDG 6 Clean Water and Sanitation

Keywords

  • Dissolved oxygen
  • Ensemble machine learning
  • Multi-regional modeling
  • River water quality

ASJC Scopus subject areas

  • Water Science and Technology

Fingerprint

Dive into the research topics of 'Decoding dissolved oxygen dynamics using ensemble machine learning models: insights across multi-regional rivers'. Together they form a unique fingerprint.

Cite this