Calibration and assessment of electrochemical air quality sensors by co-location with regulatory-grade instruments

Abstract

Abstract. The use of low-cost air quality sensors for air pollution research has outpaced our understanding of their capabilities and limitations under real-world conditions, and there is thus a critical need for understanding and optimizing the performance of such sensors in the field. Here we describe the deployment, calibration, and evaluation of electrochemical sensors on the island of Hawai'i, which is an ideal test bed for characterizing such sensors due to its large and variable sulfur dioxide (SO 2 ) levels and lack of other co-pollutants. Nine custom-built SO 2 sensors were colocated with two Hawaii Department of Health Air Quality stations over the course of 5 months, enabling comparison of sensor output with regulatory-grade instruments under a range of realistic environmental conditions. Calibration using a nonparametric algorithm (k nearest neighbors) was found to have excellent performance (RMSE < 7 ppb, MAE < 4 ppb, r 2 > 0.997) across a wide dynamic range in SO 2 (< 1 ppb, > 2 ppm). However, since nonparametric algorithms generally cannot extrapolate to conditions beyond those outside the training set, we introduce a new hybrid linear-nonparametric algorithm, enabling accurate measurements even when pollutant levels are higher than encountered during calibration. We find no significant change in instrument sensitivity toward SO 2 after 18 weeks and demonstrate that calibration accuracy remains high when a sensor is calibrated at one location and then moved to another. The performance of electrochemical SO 2 sensors is also strong at lower SO 2 mixing ratios (< 25 ppb), for which they exhibit an error of less than 2.5 ppb. While some specific results of this study (calibration accuracy, performance of the various algorithms, etc.) may differ for measurements of other pollutant species in other areas (e.g., polluted urban regions), the calibration and validation approaches described here should be widely applicable to a range of pollutants, sensors, and environments.

Development of a general calibration model and long-term performance evaluation of low-cost sensors for air pollutant gas monitoring

“…The clustering model presented here seeks to estimate the outputs corresponding to new inputs by searching for input-output pairs in the training data for which the distance (by a predefined distance metric in a potentially high-dimensional space) between the new input and the training inputs is minimized and using the average of several outputs corresponding to these nearby inputs (the nearest neighbors). In a traditional k-nearest-neighbor approach, such as that used in previous work (Hagan et al, 2018), every input-output pair from the training data is stored for comparison to new inputs. Although this provides the best possible estimation performance via this approach, storing these data and performing these comparisons are computation-and memory-intensive.…”

“…Neural networks have been applied to a large number of problems, including the calibration of low-cost gas sensors (Spinelle et al, 2015). Neural networks represent an extremely versatile framework and are able to capture nearly any nonlinear input-output relationship (Hornik, 1991). Unfortunately, to do so may require vast numbers of training data, which it is not always practical to obtain.…”

“…The use of this approach for RAMP data was suggested by Zimmerman et al (2018). Furthermore, it is similar to the approach of Hagan et al (2018), who hybridize nearest-neighbor and linear regression models. In this modified model, a random forest is applied to new data to estimate the concentrations of various measured pollutants.…”

“…Field calibrations of the sensors are thus necessary, with the sensors being collocated with highly accurate regulatory-grade instruments. Various calibration methods that have been explored include the determination of sensor calibrations from physical and chemical principles (Masson et al, 2015), higher-dimensional models to capture nonlinear interactions (Cross et al, 2017), and nonparametric approaches including artificial neural networks (Spinelle et al, 2015) and k-nearest neighbors (Hagan et al, 2018). Recent work by our group compared lab-based linear calibration models with multiple linear regression and nonparametric random forest algorithms based on ambient collocations (Zimmerman et al, 2018).…”

An improved low-power measurement of ambient NO&lt;sub&gt;2&lt;/sub&gt; and O&lt;sub&gt;3&lt;/sub&gt; combining electrochemical sensor clusters and machine learning

“…ML techniques (methods ii-iv) are powerful tools for identifying relationships between variables and have been shown to support improved concentration estimates that correct interferences in low-cost sensors (Geron, 2017;Zimmerman et al, 2018;Lin et al, 2018;Esposito et al, 2016;Hagan et al, 2018).…”