Long-term calibration models to estimate ozone concentrations with a metal oxide sensor

Sayahi, Tofigh; Garff, Alicia; Quah, Timothy; Lê, Katrina; Becnel, Thomas; Powell, Kody M.; Gaillardon, Pierre-Emmanuel; Butterfield, Anthony; Kelly, Kerry E.

doi:10.1016/j.envpol.2020.115363

Cited by 21 publications

(10 citation statements)

References 33 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Random forest regression (RFR) is one of the most widely used non-linear machine learning algorithms (Breiman and Friedman, 1997;Breiman, 2001), and it has already found applications in air pollution sensor calibration as well as in other aspects of atmospheric chemistry (Keller and Evans, 2019;Nowack et al, 2018Nowack et al, , 2019Sherwen et al, 2019;Zimmerman et al, 2018;Malings et al, 2019). It follows the idea of ensemble learning where multiple machine learning models together make more reliable predictions than the individual models.…”

Section: Random Forest Regressionmentioning

confidence: 99%

Machine learning calibration of low-cost NO2 and PM10 sensors: non-linear algorithms and their impact on site transferability

Nowack

Konstantinovskiy²,

Gardiner³

et al. 2021

Atmos. Meas. Tech.

View full text Add to dashboard Cite

Abstract. Low-cost air pollution sensors often fail to attain sufficient performance compared with state-of-the-art measurement stations, and they typically require expensive laboratory-based calibration procedures. A repeatedly proposed strategy to overcome these limitations is calibration through co-location with public measurement stations. Here we test the idea of using machine learning algorithms for such calibration tasks using hourly-averaged co-location data for nitrogen dioxide (NO2) and particulate matter of particle sizes smaller than 10 µm (PM10) at three different locations in the urban area of London, UK. We compare the performance of ridge regression, a linear statistical learning algorithm, to two non-linear algorithms in the form of random forest regression (RFR) and Gaussian process regression (GPR). We further benchmark the performance of all three machine learning methods relative to the more common multiple linear regression (MLR). We obtain very good out-of-sample R2 scores (coefficient of determination) >0.7, frequently exceeding 0.8, for the machine learning calibrated low-cost sensors. In contrast, the performance of MLR is more dependent on random variations in the sensor hardware and co-located signals, and it is also more sensitive to the length of the co-location period. We find that, subject to certain conditions, GPR is typically the best-performing method in our calibration setting, followed by ridge regression and RFR. We also highlight several key limitations of the machine learning methods, which will be crucial to consider in any co-location calibration. In particular, all methods are fundamentally limited in how well they can reproduce pollution levels that lie outside those encountered at training stage. We find, however, that the linear ridge regression outperforms the non-linear methods in extrapolation settings. GPR can allow for a small degree of extrapolation, whereas RFR can only predict values within the training range. This algorithm-dependent ability to extrapolate is one of the key limiting factors when the calibrated sensors are deployed away from the co-location site itself. Consequently, we find that ridge regression is often performing as good as or even better than GPR after sensor relocation. Our results highlight the potential of co-location approaches paired with machine learning calibration techniques to reduce costs of air pollution measurements, subject to careful consideration of the co-location training conditions, the choice of calibration variables and the features of the calibration algorithm.

show abstract

Section: Random Forest Regressionmentioning

confidence: 99%

Machine learning calibration of low-cost NO2 and PM10 sensors: non-linear algorithms and their impact on site transferability

Nowack

Konstantinovskiy²,

Gardiner³

et al. 2021

Atmos. Meas. Tech.

View full text Add to dashboard Cite

show abstract

“…The SM50 is factory calibrated and the outputs can be analogue or digital. Aeroqual devices have been used in many studies, showing good agreement with reference instruments [54][55][56][57] and demonstrating usefulness in air-quality monitoring [33,44,45,58]. However, none of the studies that were conducted lasted more than 12 months and were carried out under very harsh meteorological conditions.…”

Section: O 3 Sensors and Measurement Setupmentioning

confidence: 99%

“…Although SM50 devices use metal oxide sensors, which are sensitive to temperature and humidity [57,71], the influence of environmental factors on their operation is very limited due to the compensation algorithms used. For this reason, the use of temperature and RH in the calibration models for SM50 was not profitable.…”

Section: Influence Of Environmental Factors On Sensors Calibrationmentioning

confidence: 99%

Low- and Medium-Cost Sensors for Tropospheric Ozone Monitoring—Results of an Evaluation Study in Wrocław, Poland

Badura

Batog²,

Drzeniecka-Osiadacz

et al. 2022

Atmosphere

View full text Add to dashboard Cite

The paper presents the results of a 1.5-year evaluation study of low- and medium-cost ozone sensors. The tests covered electrochemical sensors: SensoriC O3 3E 1 (City Technology) and semiconductor gas sensors: SM50 OZU (Aeroqual), SP3-61-00 (FIS) and MQ131 (Winsen). Three copies of each sensor were enclosed in a measurement box and placed near the reference analyser (MLU 400). In the case of SensoriC O3 3E 1 sensors, the R2 values for the 1-h data were above 0.90 for the first 9 months of deployment, but a performance deterioration was observed in the subsequent months (R2 ≈ 0.6), due to sensor ageing processes. High linear relationships were observed for the SM50 devices (R2 > 0.94), but some periodic data offsets were reported, making regular checking and recalibration necessary. Power-law functions were used in the case of SP3-61-00 (R2 = 0.6–0.7) and MQ131 (R2 = 0.4–0.7). Improvements in the fittings were observed for models that included temperature and relative humidity data. In the case of SP3-61-00, the R2 values increased to above 0.82, while for MQ131 they increased to above 0.86. The study also showed that the measurement uncertainty of tested sensors meets the EU Directive 2008/50/EC requirements for indicative measurements and, in some cases, even for fixed measurements.

show abstract

“…One recent study found the differences in R 2 between metal oxide and electrochemical ozone sensors to be negligible, although the metal oxide sensors exhibited lower concentration limits than the electrochemical sensors, making them ideal for ambient concentrations lacking abnormally large spikes [6]. Other works have similarly shown small net gains and losses in R 2 and RMSE among sensor types [2,5,[7][8][9], proving both electrochemical and metal oxide sensors to be effective in quantifying ozone depending on the application. Based on these previous findings, we use metal oxide sensors to quantify ozone and methane in this work.…”

Section: Introduction 1previous Gas-phase Sensor Quantification Workmentioning

confidence: 99%

Improving Air Pollutant Metal Oxide Sensor Quantification Practices through: An Exploration of Sensor Signal Normalization, Multi-Sensor and Universal Calibration Model Generation, and Physical Factors Such as Co-Location Duration and Sensor Age

Okorn

Hannigan

2021

Atmosphere

View full text Add to dashboard Cite

As low-cost sensors have become ubiquitous in air quality measurements, there is a need for more efficient calibration and quantification practices. Here, we deploy stationary low-cost monitors in Colorado and Southern California near oil and gas facilities, focusing our analysis on methane and ozone concentration measurement using metal oxide sensors. In comparing different sensor signal normalization techniques, we propose a z-scoring standardization approach to normalize all sensor signals, making our calibration results more easily transferable among sensor packages. We also attempt several different physical co-location schemes, and explore several calibration models in which only one sensor system needs to be co-located with a reference instrument, and can be used to calibrate the rest of the fleet of sensor systems. This approach greatly reduces the time and effort involved in field normalization without compromising goodness of fit of the calibration model to a significant extent. We also explore other factors affecting the performance of the sensor system quantification method, including the use of different reference instruments, duration of co-location, time averaging, transferability between different physical environments, and the age of metal oxide sensors. Our focus on methane and stationary monitors, in addition to the z-scoring standardization approach, has broad applications in low-cost sensor calibration and utility.

show abstract

Long-term calibration models to estimate ozone concentrations with a metal oxide sensor

Cited by 21 publications

References 33 publications

Machine learning calibration of low-cost NO<sub>2</sub> and PM<sub>10</sub> sensors: non-linear algorithms and their impact on site transferability

Machine learning calibration of low-cost NO<sub>2</sub> and PM<sub>10</sub> sensors: non-linear algorithms and their impact on site transferability

Low- and Medium-Cost Sensors for Tropospheric Ozone Monitoring—Results of an Evaluation Study in Wrocław, Poland

Improving Air Pollutant Metal Oxide Sensor Quantification Practices through: An Exploration of Sensor Signal Normalization, Multi-Sensor and Universal Calibration Model Generation, and Physical Factors Such as Co-Location Duration and Sensor Age

Contact Info

Product

Resources

About

Long-term calibration models to estimate ozone concentrations with a metal oxide sensor

Cited by 21 publications

References 33 publications

Machine learning calibration of low-cost NO&lt;sub&gt;2&lt;/sub&gt; and PM&lt;sub&gt;10&lt;/sub&gt; sensors: non-linear algorithms and their impact on site transferability

Machine learning calibration of low-cost NO&lt;sub&gt;2&lt;/sub&gt; and PM&lt;sub&gt;10&lt;/sub&gt; sensors: non-linear algorithms and their impact on site transferability

Low- and Medium-Cost Sensors for Tropospheric Ozone Monitoring—Results of an Evaluation Study in Wrocław, Poland

Improving Air Pollutant Metal Oxide Sensor Quantification Practices through: An Exploration of Sensor Signal Normalization, Multi-Sensor and Universal Calibration Model Generation, and Physical Factors Such as Co-Location Duration and Sensor Age

Contact Info

Product

Resources

About

Machine learning calibration of low-cost NO<sub>2</sub> and PM<sub>10</sub> sensors: non-linear algorithms and their impact on site transferability

Machine learning calibration of low-cost NO<sub>2</sub> and PM<sub>10</sub> sensors: non-linear algorithms and their impact on site transferability