Kriging-Based Land-Use Regression Models That Use Machine Learning Algorithms to Estimate the Monthly BTEX...

Area of Science:

• Environmental Science
• Geospatial Analysis
• Machine Learning Applications

Background:

• Volatile organic compounds like BTEX (benzene, toluene, ethylbenzene, and xylenes) pose environmental and health risks.
• Understanding the spatial-temporal variation of BTEX is crucial for effective pollution control.
• Traditional Land-use Regression (LUR) models have limitations in capturing complex environmental variations.

Purpose of the Study:

• To refine Land-use Regression (LUR) models for estimating BTEX concentrations in Kaohsiung, Taiwan.
• To assess the spatial-temporal variation of BTEX using advanced modeling techniques.
• To compare the performance of different hybrid modeling approaches incorporating machine learning.

Main Methods:

• Development of a new LUR model using Taiwanese EPA BTEX data (2015-2018).
• Application of Hybrid Kriging-LUR, geographically weighted regression (GWR), Random Forest (RF), and Extreme Gradient Boosting (XGBoost).
• Validation of model reliability using 2019 external data.

Main Results:

• Initial Hybrid Kriging-LUR models explained 37%-52% of BTEX variation.
• Machine learning, specifically XGBoost, enhanced model explanatory power to 61%-79%.
• The combination of Hybrid Kriging-LUR and XGBoost demonstrated superior performance over other integrated methods.

Conclusions:

• Machine learning algorithms, particularly XGBoost, significantly improve the accuracy of BTEX concentration prediction.
• Hybrid Kriging-LUR integrated with XGBoost offers a robust approach for estimating spatiotemporal BTEX variations.
• The findings provide valuable insights for environmental monitoring and policy development in urban areas.