the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Transferability of ML-based Global Calibration Models for NO2 and NO Low-Cost Sensors
Abstract. It is essential to accurately assess and verify the effects of air pollution on human health and the environment in order to develop effective mitigation strategies. More accurate analysis of air pollution can be achieved by utilizing a higherdensity sensor network. In recent studies, the implementation of low-cost sensors has demonstrated their capability to quantify air pollution at a high spatial resolution, alleviating the problem of coarse spatial measurements associated with conventional monitoring stations. However, the reliability of such sensors is in question due to concerns about the quality and accuracy of their data. In response to these concerns, active research efforts have focused on leveraging machine learning (ML) techniques in the calibration process of low-cost sensors. These efforts demonstrate promising results for automatic calibration, which would significantly reduce the efforts and costs of traditional calibration methods and boost the low-cost sensors’ performance.
As a contribution to this promising research field, this study aims to investigate the calibration transferability between identical low-cost sensor units (SUs) for NO2 and NO using ML-based global models. Global models would further reduce calibration efforts and costs by eliminating the need for individual calibrations, especially when utilizing networks of tens or hundreds of low-cost sensors. This study employed a dataset acquired from four SUs that were located across three distinct locations within Switzerland. We also propose utilizing O3 measurements obtained from available nearby reference stations to address the cross-sensitivity effect. This strategy aims to enhance model accuracy as most electrochemical NO2 and NO sensors are extremely cross-sensitive to O3. The results of this study show excellent calibration transferability between SUs located at the same site (Case A), with the average model performance being of R2 = 0.90 ± 0.05 and RMSE = 3.4 ± 0.9 ppb for NO2, and R2 = 0.97 ± 0.02 and RMSE = 3.1 ± 0.8 ppb for NO. There is also relatively good transferability between SUs deployed at different sites (Case B), with the average performance for NO2 being R2 = 0.65 ± 0.08 and RMSE = 5.5 ± 0.4 ppb, and R2 = 0.82 ± 0.05 and RMSE = 5.8 ± 0.8 ppb for NO. Interestingly, the results illustrate a substantial improvement in the calibration models when integrating O3 measurements, which is more pronounced when SUs are situated in regions characterized by elevated O3 concentrations. Although the findings of this study are based on a specific type of sensor and sensor model, the methodology is flexible and can be applied to other low-cost sensors with different target pollutants and sensing technologies. Furthermore, this study highlights the significance of leveraging publicly available data sources to promote the reliability of low-cost air quality sensors.
- Preprint
(2622 KB) - Metadata XML
-
Supplement
(2507 KB) - BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on amt-2023-261', Anonymous Referee #1, 30 Jan 2024
“Transferability of ML-based Global Calibration Models for NO2 and NO Low-Cost Sensors”
General Comments
The manuscript presents research work looking at the application of machine learning (ML) techniques in developing a transferable calibration methodology for a network of low-cost sensor units (SU) focusing on NO and NO2 pollutants. This study assessed the performance for collocated and non-collated networks of low-cost sensors in different urban environments in Switzerland and Italy. They incorporated several commonly used variables (raw sensor signals, RH, temperature) in their algorithm but emphasised the key role ozone plays as an input in their model, concluding that the best results were obtained in studies involving co-located networks which have ozone as part of the input variable.
Specific comments
The authors have used low-cost SU that have a pair of electrochemical sensors for the two species of interest (NO, NO2). The reviewer found it odd that the pair of signals were used in the model setup for each species. For instance in modelling the corrected NO2, both the ‘NO2_A’ and ‘NO2_B’ are used but I would expect these two signals to be very correlated as summarised in Table 2. I would have thought one of the pairs should be sufficient, particularly the one with the best R value in Table 3.
While the reviewer agree with the authors on the inclusion of O3 for as input variable for the NO2 calibration (there are ample literature evidence for this), there are very little evidence for the same for NO, thus questioning if this could lead to overfitting/training dependence for NO on O3 and potentially result in additional error in transferability of this method to regions where O3 is high but low NOX.
Technical corrections
Figures 1 & S2, the caption describes the central line of box plots to mean the median but the median are not shown in these figures.
P.11, line 232 & 234, units missing for the RMSE values. Autor need to correct instance of this in the whole manuscript
- 17, line 311-312: the statement “Moreover, this study advocated enhancing global calibration models by incorporating O3 measurements from available nearby monitoring stations.” This statement is too generic, it implies that O3 needs to be considered for all low-cost SU network calibration like CO, PM, CO2 and NO (see reviewers general comment about this species above) etc.
- 18 line 333, the statement “The utilization of multiple electrochemical cells within each SU targeting the same pollutant to enhance data reliability” needs to be revised in context of the reviewers general comments above for: 1) NO species and 2) overfitting by incorporating pair sensor of reading for same species.
Citation: https://doi.org/10.5194/amt-2023-261-RC1 -
AC1: 'Reply on RC1', Ayah Abu Hani, 16 Apr 2024
The comment was uploaded in the form of a supplement: https://amt.copernicus.org/preprints/amt-2023-261/amt-2023-261-AC1-supplement.pdf
-
AC3: 'Reply on AC1', Ayah Abu Hani, 16 Apr 2024
The comment was uploaded in the form of a supplement: https://amt.copernicus.org/preprints/amt-2023-261/amt-2023-261-AC3-supplement.pdf
-
AC3: 'Reply on AC1', Ayah Abu Hani, 16 Apr 2024
-
RC2: 'Comment on amt-2023-261', Anonymous Referee #2, 21 Mar 2024
Transferability of ML-based Global Calibration Models for NO2 and NO Low-Cost Sensors by Abu-Hani et al
The manuscript presents a framework for the global machine learning (ML)-based calibration models for NO2 and NO electrochemical cells using data from low-cost sensor units (SUs) utilized in a previous study by Bigi et al. (2018). This study mainly focuses on calibration transferability among SUs when deployed at the same location (or with the same environmental conditions) and different locations (or with different environmental conditions), given that no explicit overlap exists between the training and testing data distributions. This approach uses a simple standardization to account for sensor-to-sensor variations. In addition, the author claims that a potential improvement in model transferability was achieved by using O3 from nearby regulatory air quality monitoring stations.
Minor comments:
Figures 1 and S2: Where are the central (median) lines in the Box plots? Please include the median line and mean (with a symbol) in the figures.
Figure 4: What does the negative sensor voltage convey? No mention of this in the manuscript.
Appendix A and Line 222: Although MAE was mentioned as one of the measures for quantifying the deviation between the calibrated values and their corresponding reference values, it was never discussed in the main manuscript. Tables S1-S6 and Figures S3-S6 are not referred to in the main manuscript.
Line 232: RMSE units are missing.
Figure 9: In the caption, please include how the RMSE relative improvement (%) was estimated.
Altogether, the manuscript is well written and should be considered for publication as this study demonstrates the capability of ML models to generalize calibration models that can be adopted to improve the reliability of low-cost sensors.Citation: https://doi.org/10.5194/amt-2023-261-RC2 -
AC2: 'Reply on RC2', Ayah Abu Hani, 16 Apr 2024
The comment was uploaded in the form of a supplement: https://amt.copernicus.org/preprints/amt-2023-261/amt-2023-261-AC2-supplement.pdf
-
AC2: 'Reply on RC2', Ayah Abu Hani, 16 Apr 2024
Viewed
HTML | XML | Total | Supplement | BibTeX | EndNote | |
---|---|---|---|---|---|---|
278 | 69 | 18 | 365 | 27 | 9 | 6 |
- HTML: 278
- PDF: 69
- XML: 18
- Total: 365
- Supplement: 27
- BibTeX: 9
- EndNote: 6
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1