Improving the Quality of Measurements Made by Alphasense NO2 Non-Reference Sensors Using the Mathematical Methods

Abstract

Conventional NO2 monitoring devices are relatively cumbersome, expensive, and have a relatively high-power consumption that limits their use to fixed sites. On the other hand, they offer high-quality measurements. In contrast, the low-cost NO2 sensors offer greater flexibility, are smaller, and allow greater coverage of the area with the measuring devices. However, their disadvantage is much lower accuracy. The main goal of this study was to investigate the measurement data quality of NO2-B43F Alphasense sensors. The measurement performance analysis of Alphasense NO2-B43F sensors was conducted in two research areas in Poland. Sensors were placed near fixed, professional air quality monitoring stations, carrying out measurements based on reference methods, in the following periods: July–November, and December–May. Results of the study show that without using sophisticated correction methods, the range of measured air pollution concentrations may be greater than their actual values in ambient air—measured in the field by fixed stations. In the case of summer months (with air temperature over 30 °C), the long-term mean absolute percentage error was over 150% and the sensors, using the methods recommended by the manufacturer, in the case of high temperatures could even show negative values. After applying the mathematical correction functions proposed in this article, it was possible to significantly reduce long-term errors (to 40–70% per month, regardless of the location of the measurements) and eliminate negative measurement values. The proposed method is based on the recalculation of the raw measurement, air temperature, and air RH and does not require the use of extensive analytical tools.

Unexpected Performance Improvements of Nitrogen Dioxide and Ozone Sensors by Including Carbon Monoxide Sensor Signal

“…Meanwhile, the value for NO 2 was between 0.18 and 0.30, with a maximum of 0.58. Furthermore, the performance of low-cost electrochemical NO 2 sensors is known to degrade when the ambient temperature exceeds 30 °C but less so for the CO sensor. ,− …”

“…Furthermore, the performance of low-cost electrochemical NO 2 sensors is known to degrade when the ambient temperature exceeds 30 °C but less so for the CO sensor. 31 , 44 − 46 …”

Sensors Network as an Added Value for the Characterization of Spatial and Temporal Air Quality Patterns at the Urban Scale

“…Most of the sensors stations have low correlations or show no correlation compared to the Aveiro reference station. The University sensors station had an extremely low correlation (0.11), probably because it is at a higher altitude in relation to the roads and may not receive such a direct and intense influence from road traffic peaks, typically visible in concentrations of NO 2 [ 4 ]. The sensors station at CETA was the one that presented the highest correlation (0.67) for NO 2 despite being at a great distance from the reference station, probably due to the high influence of the highway on the CETA station, which will likely have variations in road traffic closely related to traffic peaks close to the reference station.…”

“…Air pollution is recognized as the greatest environmental threat to human health [ 1 ], and risks such as an unhealthy diet and smoking have comparable impacts in relation to exposure to air pollutants [ 2 , 3 , 4 ]. This is particularly critical in urban areas, due to increased urbanization and multiple anthropogenic emission sources such as road traffic, residential combustion for heating, and industrial emissions [ 3 , 5 ].…”

Low-Cost CO Sensor Calibration Using One Dimensional Convolutional Neural Network

“…Many computational calibration approaches to improve the accuracy of low-cost sensors have been reported in the literature. Classic statistical regressions such as Multiple Linear Regression (MLR) are still being employed in recent works [ 6 , 27 , 28 , 29 , 30 , 31 , 32 , 33 ]. State-of-the-art calibration methods include supervised Machine Learning (ML) techniques such as Support Vector Regression (SVR) [ 34 , 35 , 36 , 37 , 38 , 39 ], ensemble ML techniques, such as Random Forest Regression (RFR) [ 8 , 34 , 36 , 40 , 41 , 42 , 43 ], and Neural Networks (NN) such as Multilayer Perceptron (MLP) [ 25 , 27 , 28 , 37 , 38 , 39 , 43 , 44 ]) and Recurrent Neural Networks (RNN) [ 37 , 38 , 39 , 40 , 45 , 46 ].…”