Scholarlink

<p>Airborne<?tk 4?> particulate matter (PM<sub>10</sub>) has emerged as a critical indicator of deteriorating air quality, with significant implications for human health, urban sustainability, and climate across the globe. Given its strong association with respiratory and cardiovascular diseases, predicting PM<sub>10</sub> concentrations becomes vital for safeguarding both ecological balance and human well-being. In this context, this study attempts to predict and analyze the PM₁₀ concentrations in Delhi, India, for the year 2023 with the aid of remote sensing and supervised and unsupervised machine learning techniques. The dataset utilized for the analysis comprises meteorological parameters, land use data, environmental factors, and daily station-specific datasets, incorporating PM<sub>10</sub> concentrations and land surface temperature (LST). The selected parameters include ground-based measurements collected from the 39 monitoring stations of the Central Pollution Control Board (CPCB) across Delhi, aerosol optical depth (AOD), fine aerosol optical depth (FAOD), aerosol index (AI), and air humidity (AH), along with gaseous pollutants (O₃, NO₂, SO₂, CO), and wind speed. Moreover, the spectral signatures derived in the form of normalized difference vegetation index (NDVI), soil adjusted vegetation index (SAVI), built-up index (BUI), and urban thermal field variance index (UTFVI) were used for the predictive model development. The efficacy of six regressors, multiple linear regression (MLR), support vector regressor (SVR), random forest (RFR), stochastic gradient boosting regressor (SGB), extreme gradient boosting regressor (XGB), and categorical boosting regressor (CBR) was assessed for predicting PM₁₀ concentrations. The CBR model demonstrated an impressive performance (R² = 0.94; RMSE = 4.77&#xa0;µg/m<sup>3</sup>), surpassing the other regressors. The CBR model shows prominence in ordered boosting and capturing complex nonlinear relationships more effectively than conventional regression models. Further, the K‑Means clustering technique was used to delineate pollution hotspots across the city. The resulting maps reveal persistently elevated PM₁₀ in dense residential and industrial zones, offering actionable inputs for local air‑quality management and urban planning in Delhi. This study reinforces the importance of evidence-based approaches that align public health priorities with sustainable environmental policies.<?tk 0?></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

Analyzing and predicting PM₁₀ using remote sensing and machine learning: A case study of Delhi, India

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年10月28日

查看全文
  • Masud Karim,
  • Saakshi Tomar,
  • Sahil K. Shah,
  • T. P. Singh

摘要

Airborne particulate matter (PM10) has emerged as a critical indicator of deteriorating air quality, with significant implications for human health, urban sustainability, and climate across the globe. Given its strong association with respiratory and cardiovascular diseases, predicting PM10 concentrations becomes vital for safeguarding both ecological balance and human well-being. In this context, this study attempts to predict and analyze the PM₁₀ concentrations in Delhi, India, for the year 2023 with the aid of remote sensing and supervised and unsupervised machine learning techniques. The dataset utilized for the analysis comprises meteorological parameters, land use data, environmental factors, and daily station-specific datasets, incorporating PM10 concentrations and land surface temperature (LST). The selected parameters include ground-based measurements collected from the 39 monitoring stations of the Central Pollution Control Board (CPCB) across Delhi, aerosol optical depth (AOD), fine aerosol optical depth (FAOD), aerosol index (AI), and air humidity (AH), along with gaseous pollutants (O₃, NO₂, SO₂, CO), and wind speed. Moreover, the spectral signatures derived in the form of normalized difference vegetation index (NDVI), soil adjusted vegetation index (SAVI), built-up index (BUI), and urban thermal field variance index (UTFVI) were used for the predictive model development. The efficacy of six regressors, multiple linear regression (MLR), support vector regressor (SVR), random forest (RFR), stochastic gradient boosting regressor (SGB), extreme gradient boosting regressor (XGB), and categorical boosting regressor (CBR) was assessed for predicting PM₁₀ concentrations. The CBR model demonstrated an impressive performance (R² = 0.94; RMSE = 4.77 µg/m3), surpassing the other regressors. The CBR model shows prominence in ordered boosting and capturing complex nonlinear relationships more effectively than conventional regression models. Further, the K‑Means clustering technique was used to delineate pollution hotspots across the city. The resulting maps reveal persistently elevated PM₁₀ in dense residential and industrial zones, offering actionable inputs for local air‑quality management and urban planning in Delhi. This study reinforces the importance of evidence-based approaches that align public health priorities with sustainable environmental policies.