Calibration of a Low-Cost Methane Sensor Using Machine Learning.

Area of Science:

• Environmental Science
• Sensor Technology
• Machine Learning

Background:

• Greenhouse gas emissions require understanding their sources for effective mitigation.
• Low-cost wireless sensors offer a method for environmental monitoring in remote locations, such as peatlands.
• The Figaro NGM2611-E13 methane sensor has limited calibration data for low concentrations typical in outdoor environments.

Purpose of the Study:

• To investigate the feasibility of calibrating the Figaro NGM2611-E13 methane sensor for low concentration measurements (0-200 ppm) using machine learning.
• To develop and evaluate machine learning models that account for environmental variables like temperature and humidity.

Main Methods:

• Trained machine learning models on over 50,000 calibration data points.
• Included methane concentrations from 0-200 ppm, temperatures from 5-30 °C, and relative humidity from 40-80%.
• Incorporated interaction terms between variables to enhance model performance.

Main Results:

• A final machine learning model achieved a root-mean-square error (RMSE) of 5.1 ppm.
• The model demonstrated a high coefficient of determination (R²), reaching 0.997.
• The results indicate successful calibration for methane concentrations below 200 ppm.

Conclusions:

• The Figaro NGM2611-E13 sensor, when calibrated with machine learning, shows significant potential for measuring low methane concentrations.
• This approach can enhance environmental monitoring of greenhouse gas emissions in challenging settings.
• Accurate low-concentration methane detection is crucial for understanding and combating climate change.