Development and Evaluation of Ensemble Learning-based Environmental Methane Detection and Intensity Prediction...

Area of Science:

• Environmental Science
• Machine Learning
• Atmospheric Chemistry

Background:

• Global warming concerns necessitate advanced methane (CH4) detection technologies.
• Fugitive CH4 emissions pose significant environmental risks.
• Existing detection methods require enhancement for speed and accuracy.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for detecting fugitive methane (CH4).
• To predict the intensity of methane (CH4) emissions using ML.
• To integrate meteorological data for improved CH4 monitoring.

Main Methods:

• Utilized ensemble learning to build weighted ML models from weaker models.
• Trained models for CH4 detection (classification) and intensity prediction (regression).
• Incorporated meteorological variables: wind speed, temperature, pressure, humidity, water vapor, heat flux.

Main Results:

• Classification model achieved 97.2% accuracy, 0.972 F1 score, 0.945 MCC, and 0.995 AUC ROC for CH4 detection.
• Regression model achieved an R² score of 0.858 for predicting CH4 intensity.
• Ensemble ML approach demonstrated high efficacy in both detection and intensity prediction.

Conclusions:

• Developed ML models offer a robust solution for fugitive CH4 detection and intensity assessment.
• Models are adaptable for deployment with fixed ground sensors or mobile UAV-mounted systems.
• This technology can significantly aid in mitigating environmental impacts of CH4 emissions.