Scholarlink

Contaminant levels in water quality downstream of four wastewater treatment works in the Crocodile West Catchment, as well as their spatial- and seasonal variations were examined. A total of 24 samples were collected during dry and wet seasons from four sampling site locations: Lethabile, Losperfontein, Rietfontein, and Brits. Physicochemical parameters such as pH, EC, temperature, suspended solids, chemical oxygen demand, dissolved oxygen, sulphates, phosphates, nitrogen, and ammonia were determined. Spatial and seasonal variations in water quality parameter concentrations were evalauted using the Kruskal-Wallis test and Mann-Whitney U test respectively, at a significance level of p < 0.05). The results indicated that pollution levels followed the order; Lathabile > Losperfontein > Rietfontein > Brits. Values of R2 greater than 0.6 obtained from random forest models of predicting pollutant concentrations from environmental variables indicates strong predictive performance and low prediction error. Urban and industrial wastewater treatment plants were identify as the main sources of contamination, while pollution indices, including total organic pollution and eutrophication, reached their maximum level during the rainy season. All sampling sites were classify as heavily polluted and eutrophic. These results suggest that machine learning models, such as random forest, are effective tools for predicting eutrophication and detecting water pollution in complex aquatic systems.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Forecasting the Pollution and Eutrophic Status of the Upper Crocodile West Catchment, South Africa by a Random Forest Regression-Based Machine Learning Approach

  • 章节

  • 发表日期：

    2026年

查看全文
  • Ernestine Atangana,
  • Paul J. Oberholster,
  • Heba Bedair,
  • Ashkan Khalife,
  • Soumya Ghosh

摘要

Contaminant levels in water quality downstream of four wastewater treatment works in the Crocodile West Catchment, as well as their spatial- and seasonal variations were examined. A total of 24 samples were collected during dry and wet seasons from four sampling site locations: Lethabile, Losperfontein, Rietfontein, and Brits. Physicochemical parameters such as pH, EC, temperature, suspended solids, chemical oxygen demand, dissolved oxygen, sulphates, phosphates, nitrogen, and ammonia were determined. Spatial and seasonal variations in water quality parameter concentrations were evalauted using the Kruskal-Wallis test and Mann-Whitney U test respectively, at a significance level of p < 0.05). The results indicated that pollution levels followed the order; Lathabile > Losperfontein > Rietfontein > Brits. Values of R2 greater than 0.6 obtained from random forest models of predicting pollutant concentrations from environmental variables indicates strong predictive performance and low prediction error. Urban and industrial wastewater treatment plants were identify as the main sources of contamination, while pollution indices, including total organic pollution and eutrophication, reached their maximum level during the rainy season. All sampling sites were classify as heavily polluted and eutrophic. These results suggest that machine learning models, such as random forest, are effective tools for predicting eutrophication and detecting water pollution in complex aquatic systems.