Scholarlink

<p>Air pollution is a major environmental issue of concern. Previous studies lack in utilizing multi-band approaches for the accurate prediction of pollutants. This study focuses on utilizing Sentinel-5P and Sentinel-2 satellite data to monitor and predict tropospheric nitrogen dioxide (NO<sub>2</sub>) concentrations in Kolkata. The methodology combines differential optical absorption spectroscopy (DOAS) for estimating NO<sub>2</sub> concentrations from Sentinel-5P and surface reflectance analysis from Sentinel-2. The correlation between NO<sub>2</sub> and Sentinel-2 spectral bands was explored, revealing that shortwave infrared (SWIR) bands B11 and B12 have the highest correlation, suggesting their sensitivity to atmospheric aerosols and gases. A Random Forest regression model was applied to predict NO₂ concentrations based on the spectral bands of Sentinel-2, with the dataset divided into training and test sets. The model achieved the best performance with 50 trees with maximum depth attained at 19, showing the lowest Test RMSE of 7.8656 × 10⁻⁶, Test MSE of 6.1867 × 10⁻¹¹, Test Error of 5.73%, and the highest t-statistic of 4.22 (<i>p</i> = 2.95 × 10⁻⁵). The error distribution at 50 trees shows low absolute variability but high relative variation, with a near-normal, slightly right-skewed (skewness = 0.49) and platykurtic (kurtosis = − 0.36) distribution. Results highlight the importance of multi-band approaches for improved predictability. This study demonstrates the potential of combining satellite-based remote sensing and machine learning techniques to better understand and predict air pollution dynamics at a smaller scale, providing valuable insights for environmental monitoring and urban planning.</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

Prediction of Tropospheric Nitrogen Dioxide in Kolkata Using Sentinel-5p and Sentinel-2 Multispectral Data and Machine Learning Algorithm

  • 期刊论文

  • 发表日期：

    2025年11月12日

查看全文
  • Aman Srivastava,
  • Aditya Kumar Thakur,
  • Rahul Dev Garg,
  • Pradeep Kumar Garg

摘要

Air pollution is a major environmental issue of concern. Previous studies lack in utilizing multi-band approaches for the accurate prediction of pollutants. This study focuses on utilizing Sentinel-5P and Sentinel-2 satellite data to monitor and predict tropospheric nitrogen dioxide (NO2) concentrations in Kolkata. The methodology combines differential optical absorption spectroscopy (DOAS) for estimating NO2 concentrations from Sentinel-5P and surface reflectance analysis from Sentinel-2. The correlation between NO2 and Sentinel-2 spectral bands was explored, revealing that shortwave infrared (SWIR) bands B11 and B12 have the highest correlation, suggesting their sensitivity to atmospheric aerosols and gases. A Random Forest regression model was applied to predict NO₂ concentrations based on the spectral bands of Sentinel-2, with the dataset divided into training and test sets. The model achieved the best performance with 50 trees with maximum depth attained at 19, showing the lowest Test RMSE of 7.8656 × 10⁻⁶, Test MSE of 6.1867 × 10⁻¹¹, Test Error of 5.73%, and the highest t-statistic of 4.22 (p = 2.95 × 10⁻⁵). The error distribution at 50 trees shows low absolute variability but high relative variation, with a near-normal, slightly right-skewed (skewness = 0.49) and platykurtic (kurtosis = − 0.36) distribution. Results highlight the importance of multi-band approaches for improved predictability. This study demonstrates the potential of combining satellite-based remote sensing and machine learning techniques to better understand and predict air pollution dynamics at a smaller scale, providing valuable insights for environmental monitoring and urban planning.