Scholarlink

<p>Plant responses to air pollution are complex under dynamic environments, requiring advanced integrated approaches for capturing interactions among pollutant exposure, meteorology, and plant physio-biochemical responses in real-world conditions. This study developed an integrated remote sensing and machine learning framework to predict species-level plant air pollution tolerance from satellite-derived pollutant exposures, reanalysis meteorology, and plant deciduous-state indicator, using laboratory-derived Air Pollution Tolerance Index (APTI) as the response variable. APTI was calculated from pH, relative water content, ascorbic acid concentration, and total chlorophyll content as indicators of plant response to air pollution stress. Sentinel-5P (NO<sub>2</sub>, SO<sub>2</sub>, O<sub>3</sub>), MODIS (PM<sub>2.5</sub>-derived), and satellite/reanalysis datasets, CHIRPS (rainfall) and ERA5-Land (temperature, humidity) data were processed in Google Earth Engine. Leaf samples of <i>Pterocarpus indicus</i> Willd. were collected across multiple provinces in Thailand under varying air pollutant and meteorological conditions, and laboratory analysis was conducted to calculate APTI. Random Forest (RF), Gradient Boosting Regressor (GBR), Extreme Gradient Boosting (XGB), Support Vector Regression (SVR), and Multiple Linear Regression (MLR) algorithms were evaluated for APTI prediction. Model performance was evaluated using R<sup>2</sup>, RMSE, and MAE, and robustness was assessed with cross-validation (CV) and bootstrap resampling. XGB and GBR showed the highest predictive performance, with test R² values of 0.814 and 0.794 and CV R² values of 0.691 and 0.698, respectively. In contrast, SVR and MLR underperformed in capturing the highly complex and nonlinear patterns. The findings demonstrate the potential of multi-source geospatial data fusion for species-level assessment of plant air pollution tolerance under heterogeneous environmental conditions, especially where ground monitoring is limited. The proposed framework provides a useful basis for future plant selection and urban green infrastructure planning, subject to further validation across broader spatiotemporal settings.</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

An integrated remote sensing-machine learning framework for assessing plant air pollution tolerance using satellite-derived air quality and reanalysis meteorological data

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月19日

查看全文
  • Chamila Priyankara Hetti Arachchige,
  • Vacharaporn Soonsin,
  • Emlyn David Qivitoq Witt

摘要

Plant responses to air pollution are complex under dynamic environments, requiring advanced integrated approaches for capturing interactions among pollutant exposure, meteorology, and plant physio-biochemical responses in real-world conditions. This study developed an integrated remote sensing and machine learning framework to predict species-level plant air pollution tolerance from satellite-derived pollutant exposures, reanalysis meteorology, and plant deciduous-state indicator, using laboratory-derived Air Pollution Tolerance Index (APTI) as the response variable. APTI was calculated from pH, relative water content, ascorbic acid concentration, and total chlorophyll content as indicators of plant response to air pollution stress. Sentinel-5P (NO2, SO2, O3), MODIS (PM2.5-derived), and satellite/reanalysis datasets, CHIRPS (rainfall) and ERA5-Land (temperature, humidity) data were processed in Google Earth Engine. Leaf samples of Pterocarpus indicus Willd. were collected across multiple provinces in Thailand under varying air pollutant and meteorological conditions, and laboratory analysis was conducted to calculate APTI. Random Forest (RF), Gradient Boosting Regressor (GBR), Extreme Gradient Boosting (XGB), Support Vector Regression (SVR), and Multiple Linear Regression (MLR) algorithms were evaluated for APTI prediction. Model performance was evaluated using R2, RMSE, and MAE, and robustness was assessed with cross-validation (CV) and bootstrap resampling. XGB and GBR showed the highest predictive performance, with test R² values of 0.814 and 0.794 and CV R² values of 0.691 and 0.698, respectively. In contrast, SVR and MLR underperformed in capturing the highly complex and nonlinear patterns. The findings demonstrate the potential of multi-source geospatial data fusion for species-level assessment of plant air pollution tolerance under heterogeneous environmental conditions, especially where ground monitoring is limited. The proposed framework provides a useful basis for future plant selection and urban green infrastructure planning, subject to further validation across broader spatiotemporal settings.