The Potential of Low-Cost Tin-Oxide Sensors Combined with Machine Learning for Estimating Atmospheric <i>CH</i>₄ Variations around Background Concentration

• Main Authors: Rodrigo Rivera Martinez, Diego Santaren, Olivier Laurent, Ford Cropley, Cécile Mallet, Michel Ramonet, Christopher Caldow, Leonard Rivier, Gregoire Broquet, Caroline Bouchet, Catherine Juery, Philippe Ciais
• Format: Article
• Language: English
• Published: MDPI AG 2021-01-01
• Series: Atmosphere
• Subjects: low-cost sensors; artificial neural networks; methane; calibration
• Online Access: https://www.mdpi.com/2073-4433/12/1/107

Continued developments in instrumentation and modeling have driven progress in monitoring methane (<inline-formula><math display="inline"><semantics><msub><mi>CH</mi><mn>4</mn></msub></semantics></math></inline-formula>) emissions at a range of spatial scales. The sites that emit <inline-formula><math display="inline"><semantics><msub><mi>CH</mi><mn>4</mn></msub></semantics></math></inline-formula> such as landfills, oil and gas extraction or storage infrastructure, intensive livestock farms account for a large share of global emissions, and need to be monitored on a continuous basis to verify the effectiveness of reductions policies. Low cost sensors are valuable to monitor methane (<inline-formula><math display="inline"><semantics><msub><mi>CH</mi><mn>4</mn></msub></semantics></math></inline-formula>) around such facilities because they can be deployed in a large number to sample atmospheric plumes and retrieve emission rates using dispersion models. Here we present two tests of three different versions of Figaro^® TGS tin-oxide sensors for estimating <inline-formula><math display="inline"><semantics><msub><mi>CH</mi><mn>4</mn></msub></semantics></math></inline-formula> concentrations variations, at levels similar to current atmospheric values, with a sought accuracy of 0.1 to 0.2 ppm. In the first test, we characterize the variation of the resistance of the tin-oxide semi-conducting sensors to controlled levels of <inline-formula><math display="inline"><semantics><msub><mi>CH</mi><mn>4</mn></msub></semantics></math></inline-formula>, <inline-formula><math display="inline"><semantics><mrow><msub><mi mathvariant="normal">H</mi><mn>2</mn></msub><mi mathvariant="normal">O</mi></mrow></semantics></math></inline-formula> and <inline-formula><math display="inline"><semantics><mi>CO</mi></semantics></math></inline-formula> in the laboratory, to analyze cross-sensitivities. In the second test, we reconstruct observed <inline-formula><math display="inline"><semantics><msub><mi>CH</mi><mn>4</mn></msub></semantics></math></inline-formula> variations in a room, that ranged from 1.9 and 2.4 ppm during a three month experiment from observed time series of resistances and other variables. To do so, a machine learning model is trained against true <inline-formula><math display="inline"><semantics><msub><mi>CH</mi><mn>4</mn></msub></semantics></math></inline-formula> recorded by a high precision instrument. The machine-learning model using 30% of the data for training reconstructs <inline-formula><math display="inline"><semantics><msub><mi>CH</mi><mn>4</mn></msub></semantics></math></inline-formula> within the target accuracy of 0.1 ppm only if training variables are representative of conditions during the testing period. The model-derived sensitivities of the sensors resistance to <inline-formula><math display="inline"><semantics><mrow><msub><mi mathvariant="normal">H</mi><mn>2</mn></msub><mi mathvariant="normal">O</mi></mrow></semantics></math></inline-formula> compared to <inline-formula><math display="inline"><semantics><msub><mi>CH</mi><mn>4</mn></msub></semantics></math></inline-formula> are larger than those observed under controlled conditions, which deserves further characterization of all the factors influencing the resistance of the sensors.