Scholarlink

<p>Accurate water quality (WQ) parameter estimation is essential for effective environmental monitoring and management. Traditional methods often involve extensive field sampling, which can be time-consuming and costly. This study proposes a novel machine learning approach using random forest (RF) and support vector regression (SVR), leveraging hyperspectral reflectance data to estimate key WQ parameters. Hybrid RF-SVR, RF-Harris hawk optimization (RF-HHO), and SVR-HHO algorithms we employed and compared to assess the WQ parameters. The models utilize preprocessed hyperspectral remote sensing reflectance (Rrs) as inputs and various WQ parameters as outputs to design the input-output structures of the proposed models. The proposed models were trained and validated on datasets comprising in situ hyperspectral surface reflectance measurements along with various WQ parameters, including chlorophyll-a concentration (Chla), total suspended solids (TSS), and absorption by colored dissolved organic matter at 440&#xa0;nm (aCDOM440). The hybrid models generally outperformed the standalone RF and SVR models in predicting certain parameters and the data preprocessing methods had an obvious impact on model accuracy. Among them, the RF-SVR model emerged as the most effective, achieving correlation coefficients (R) and root-mean-square errors (RMSE) of 0.96 and 15.69 (mg m<sup>− 3</sup>) for Chla, 0.78 and 18.09 (g m<sup>− 3</sup>) for TSS, and 0.83 and 0.45 (m<sup>− 1</sup>) for aCDOM440, respectively, during the testing stage. Although the proposed models demonstrated high accuracy in national and local regions across Europe, data availability remains critical in achieving reliable WQ parameter estimates. The proposed models were also evaluated in other regions, showing promising potential for wider application in WQ parameter modeling. These findings highlight the potential of machine learning and hyperspectral remote sensing techniques as powerful tools for large-scale, cost-effective WQ monitoring and improved prediction accuracy under varying environmental conditions. The proposed framework offers a scalable solution for real-time WQ assessment and explores integration techniques for further environmental monitoring applications.</p>

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

Hyperspectral reflectance-driven estimation of water quality parameters using hybrid random forest and support vector regression techniques

  • 期刊论文

  • 发表日期：

    2025年6月23日

查看全文
  • Tamer ElGharbawi,
  • Mosbeh R. Kaloop,
  • Jong Wan Hu,
  • Fawzi Zarzoura

摘要

Accurate water quality (WQ) parameter estimation is essential for effective environmental monitoring and management. Traditional methods often involve extensive field sampling, which can be time-consuming and costly. This study proposes a novel machine learning approach using random forest (RF) and support vector regression (SVR), leveraging hyperspectral reflectance data to estimate key WQ parameters. Hybrid RF-SVR, RF-Harris hawk optimization (RF-HHO), and SVR-HHO algorithms we employed and compared to assess the WQ parameters. The models utilize preprocessed hyperspectral remote sensing reflectance (Rrs) as inputs and various WQ parameters as outputs to design the input-output structures of the proposed models. The proposed models were trained and validated on datasets comprising in situ hyperspectral surface reflectance measurements along with various WQ parameters, including chlorophyll-a concentration (Chla), total suspended solids (TSS), and absorption by colored dissolved organic matter at 440 nm (aCDOM440). The hybrid models generally outperformed the standalone RF and SVR models in predicting certain parameters and the data preprocessing methods had an obvious impact on model accuracy. Among them, the RF-SVR model emerged as the most effective, achieving correlation coefficients (R) and root-mean-square errors (RMSE) of 0.96 and 15.69 (mg m− 3) for Chla, 0.78 and 18.09 (g m− 3) for TSS, and 0.83 and 0.45 (m− 1) for aCDOM440, respectively, during the testing stage. Although the proposed models demonstrated high accuracy in national and local regions across Europe, data availability remains critical in achieving reliable WQ parameter estimates. The proposed models were also evaluated in other regions, showing promising potential for wider application in WQ parameter modeling. These findings highlight the potential of machine learning and hyperspectral remote sensing techniques as powerful tools for large-scale, cost-effective WQ monitoring and improved prediction accuracy under varying environmental conditions. The proposed framework offers a scalable solution for real-time WQ assessment and explores integration techniques for further environmental monitoring applications.