Geospatial-based estimation of NMHC concentrations through an ensemble stacking Geo-AI algorithm to advance air...

Area of Science:

• Atmospheric Chemistry and Air Quality
• Environmental Science
• Geospatial Analysis

Background:

• Non-methane hydrocarbons (NMHCs) are crucial atmospheric volatile organic compounds (VOCs) influencing ozone and secondary organic aerosol formation.
• Limited geospatial data on NMHC concentrations hinders effective air quality management due to sparse monitoring networks.

Purpose of the Study:

• To develop and evaluate an enhanced geospatial framework for estimating NMHC concentrations in Taiwan.
• To compare the performance of various modeling approaches, including statistical, deep learning, and machine learning models.

Main Methods:

• Integrated daily data from 34 monitoring stations (2015-2021) in Taiwan.
• Evaluated four modeling approaches: Land Use Regression (LUR), Deep Neural Networks (DNN), single machine learning models (XGBR, GBR, LGBMR, CBR, RFR), and an ensemble stacking model.
• Utilized an ensemble stacking model integrating five single machine learning algorithms for optimal prediction.

Main Results:

• The ensemble stacking model achieved the highest predictive performance (R² = 0.820), significantly outperforming LUR (R² = 0.441) and DNN (R² = 0.708) models.
• Key predictors included residential areas, NO₂, O₃, water bodies, road networks, and meteorological variables.
• Geospatial maps identified NMHC hotspots in urban residential and mountainous areas, with lower levels in coastal/rural regions.

Conclusions:

• Ensemble stacking offers a robust and effective framework for improving NMHC geospatial estimation.
• The findings provide valuable insights for air quality management, environmental monitoring, and exposure assessment.
• The model demonstrated robustness by capturing temporal trends, even during the COVID-19 pandemic.