Scholarlink

<p>Dissolved oxygen (DO) is a crucial indicator of river water quality, directly influencing aquatic ecosystem health and biogeochemical processes. Accurate estimation of DO in tropical river systems remains challenging due to limited in situ observations, high environmental variability, and persistent cloud cover that constrains satellite-based monitoring. This study addresses this gap by proposing a hybrid modelling approach that combines stepwise regression (SR) with two ANN architectures, Backpropagation Neural Network (BPNN) and Radial Basis Function Neural Network (RBFNN), to enhance DO prediction in the Pahang River, Malaysia, using Landsat 8 imagery. Stepwise regression was first applied to identify optimal predictor variables, reducing model complexity and mitigating overfitting before ANN training. The selected features were then used to develop and evaluate standalone ANN and hybrid SR–ANN models. Performance assessment using the coefficient of determination (R-squared) and root mean square error indicated that the hybrid SR–RBFNN achieved the best results, achieving an R-squared value of 0.991 and a root mean square error of 0.045&#xa0;mg/L on the testing set. Analysis of hidden layer configurations further revealed that excessive neuron numbers may lead to overestimation, underscoring the importance of balanced network design. Spatial distribution maps of DO were generated for upstream, midstream, and downstream river segments, revealing distinct spatial variability influenced by salinity gradients and cloud contamination. Overall, the hybrid SR–ANN framework offers a cost-effective and robust solution for satellite-based DO estimation in data-scarce tropical regions. The approach demonstrates the value of combining feature selection with neural networks and supports digital decision-making for sustainable river basin management.</p> Graphical abstract <p></p> <p>Based on the graphical snapshot, this study aimed to enhance the estimation of dissolved oxygen (DO) in the Pahang River, Malaysia, using Landsat 8 imagery and a hybrid machine learning approach. Secondary data were obtained from monitoring stations operated by the Department of Environment (DOE) Malaysia, providing comprehensive records of water quality parameters. A total of 25 sampling points along the river, surveyed in 2018, 2020, and 2021, were used to extract water quality information through remote sensing. Parameters were selected based on their availability and retrievability from Landsat 8 imagery. The proposed method integrates stepwise regression (SR) for optimal feature selection with artificial neural networks (ANNs), specifically Backpropagation Neural Network (BPNN) and Radial Basis Function Neural Network (RBFNN), to improve DO predictions. SR reduces model complexity and mitigates overfitting, while the hybrid SR–ANN models leverage the selected predictors to achieve enhanced accuracy. Performance evaluation demonstrated that the SR–RBFNN hybrid outperformed other models, achieving an R² of 0.991 and RMSE of 0.045&#xa0;mg/L. Analysis of hidden layer configurations revealed that excessive neuron numbers may lead to overestimation, highlighting the importance of balanced network design. Spatial DO maps were generated for upstream, midstream, and downstream river segments, showing distinct variability influenced by salinity gradients and cloud cover. Overall, this hybrid SR–ANN framework offers a cost-effective, reliable, and robust tool for satellite-based DO estimation in data-scarce tropical rivers. The approach supports informed and sustainable river management by combining remote sensing with advanced machine learning for accurate water quality monitoring.</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

Improving Satellite-Based Dissolved Oxygen Prediction for River Management Using a Stepwise Regression–Neural Network Hybrid in Pahang River, Malaysia

  • 期刊论文

  • 发表日期：

    2026年4月6日

查看全文
  • Nurshahida Azreen Mohd Jais,
  • Ahmad Fikri Abdullah,
  • Muhamad Saufi Mohd Kassim,
  • Nur ‘Atirah Muhadi,
  • Zoran Vojinovic,
  • Izwaharyanie Ibrahim

摘要

Dissolved oxygen (DO) is a crucial indicator of river water quality, directly influencing aquatic ecosystem health and biogeochemical processes. Accurate estimation of DO in tropical river systems remains challenging due to limited in situ observations, high environmental variability, and persistent cloud cover that constrains satellite-based monitoring. This study addresses this gap by proposing a hybrid modelling approach that combines stepwise regression (SR) with two ANN architectures, Backpropagation Neural Network (BPNN) and Radial Basis Function Neural Network (RBFNN), to enhance DO prediction in the Pahang River, Malaysia, using Landsat 8 imagery. Stepwise regression was first applied to identify optimal predictor variables, reducing model complexity and mitigating overfitting before ANN training. The selected features were then used to develop and evaluate standalone ANN and hybrid SR–ANN models. Performance assessment using the coefficient of determination (R-squared) and root mean square error indicated that the hybrid SR–RBFNN achieved the best results, achieving an R-squared value of 0.991 and a root mean square error of 0.045 mg/L on the testing set. Analysis of hidden layer configurations further revealed that excessive neuron numbers may lead to overestimation, underscoring the importance of balanced network design. Spatial distribution maps of DO were generated for upstream, midstream, and downstream river segments, revealing distinct spatial variability influenced by salinity gradients and cloud contamination. Overall, the hybrid SR–ANN framework offers a cost-effective and robust solution for satellite-based DO estimation in data-scarce tropical regions. The approach demonstrates the value of combining feature selection with neural networks and supports digital decision-making for sustainable river basin management.

Graphical abstract

Based on the graphical snapshot, this study aimed to enhance the estimation of dissolved oxygen (DO) in the Pahang River, Malaysia, using Landsat 8 imagery and a hybrid machine learning approach. Secondary data were obtained from monitoring stations operated by the Department of Environment (DOE) Malaysia, providing comprehensive records of water quality parameters. A total of 25 sampling points along the river, surveyed in 2018, 2020, and 2021, were used to extract water quality information through remote sensing. Parameters were selected based on their availability and retrievability from Landsat 8 imagery. The proposed method integrates stepwise regression (SR) for optimal feature selection with artificial neural networks (ANNs), specifically Backpropagation Neural Network (BPNN) and Radial Basis Function Neural Network (RBFNN), to improve DO predictions. SR reduces model complexity and mitigates overfitting, while the hybrid SR–ANN models leverage the selected predictors to achieve enhanced accuracy. Performance evaluation demonstrated that the SR–RBFNN hybrid outperformed other models, achieving an R² of 0.991 and RMSE of 0.045 mg/L. Analysis of hidden layer configurations revealed that excessive neuron numbers may lead to overestimation, highlighting the importance of balanced network design. Spatial DO maps were generated for upstream, midstream, and downstream river segments, showing distinct variability influenced by salinity gradients and cloud cover. Overall, this hybrid SR–ANN framework offers a cost-effective, reliable, and robust tool for satellite-based DO estimation in data-scarce tropical rivers. The approach supports informed and sustainable river management by combining remote sensing with advanced machine learning for accurate water quality monitoring.