Related Experiment Video
Updated: May 24, 2025
00:05
In Silico Modeling Method for Computational Aquatic Toxicology of Endocrine Disruptors: A Software-Based Approach Using QSAR Toolbox
Published on: August 28, 2019
13.8K
Developing a real-time water quality simulation toolbox using machine learning and application programming interface.
Gi-Hun Bang1, Na-Hyeon Gwon2, Min-Jeong Cho2
1Department of Integrated Water Management, Yeungnam University, Daehak-ro 280, Gyeongsan-si, Water Campus, Korea Water Cluster, Gukgasandan-daero 40-gil, Guji-myeon, Dalseong-gun, Gyeongsangbuk-do, Daegu, Republic of Korea.
Journal of Environmental Management
|March 1, 2025
Summary
A new machine learning toolbox provides real-time river water quality modeling. Random Forest models trained on random sampling data showed the best performance for key water quality parameters.
Area of Science:
- Environmental Science
- Water Resource Management
- Computational Hydrology
Background:
- Rivers are crucial for human life but face increasing pollution.
- Real-time river water quality modeling is essential for effective management and protection.
- Existing methods require enhancement for timely and accurate assessments.
Purpose of the Study:
- To develop a real-time river water quality simulation toolbox using machine learning (ML) and an application program interface (API).
- To construct and compare ML models for simulating key water quality parameters: chlorophyll a (Chl-a), dissolved oxygen (DO), total nitrogen (TN), total organic carbon (TOC), and total phosphorus (TP).
- To optimize model performance through hyperparameter tuning and evaluate the impact of data sampling methods and ML algorithms.
Main Methods:
- Developed a toolbox integrating ML algorithms (Artificial Neural Network, Random Forest, Support Vector Machines) with API data.
- Constructed water quality simulation models for the Nakdong River.
- Employed hyperparameter optimization and compared model performance using random sampling versus time-series data.
Main Results:
- Models trained with random sampling data generally outperformed those trained with time-series data.
- The Random Forest (RF) algorithm demonstrated the best performance among the tested ML models using random sampling.
- RF achieved R-squared values of 0.79 for DO, 0.65 for TN, 0.74 for TP, 0.45 for Chl-a, and 0.48 for TOC.
- Dissolved oxygen, total nitrogen, and total phosphorus models showed superior performance compared to chlorophyll a and total organic carbon models.
Conclusions:
- The developed ML-based toolbox effectively simulates real-time river water quality.
- Random Forest models utilizing random sampling data offer a promising approach for water quality prediction.
- Further improvements for Chl-a and TOC models require diversifying input variables; sensitivity and uncertainty analyses enhance model understanding.
