the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 12 Sep 2025
High spatial resolution CO2 measurement using low-cost commercial sensors in Seoul megacity
Abstract. Carbon dioxide (CO2) is the most significant anthropogenic greenhouse gas. However, tracking CO2 levels can be challenging due to the uneven distribution of concentrations and the high cost of sensors. In this study, we explored several correction techniques to enable the large-scale use of affordable CO2 sensors, thereby enhancing the spatial resolution. We found that the low-cost CO2 sensor (HT-2000) closely aligned with the trends observed in data from a more accurate sensor (LI-840a). By applying multiple-point linear regression, we reduced the root mean square error (RMSE) to only 1–2 % of the measured value, which is accurate enough for urban monitoring at a local scale. Using a large network of low-cost sensors, we were able to map CO2 concentration in detail, capture fine spatial variations, and gain a clearer understanding of emission patterns at an urban road intersection and within a tunnel.
- Preprint
(1201 KB) - Metadata XML
-
Supplement
(366 KB) - BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2025-3775', Anonymous Referee #1, 02 Oct 2025
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2025/egusphere-2025-3775/egusphere-2025-3775-RC1-supplement.pdfCitation: https://doi.org/
10.5194/egusphere-2025-3775-RC1 -
RC2: 'Comment on egusphere-2025-3775', Anonymous Referee #2, 14 Oct 2025
The authors aim to address the spatial heterogeneity and temporal variability of urban CO2 by applying a multiple‑point linear regression to low‑cost sensor measurements. They present two experiments—one at a high‑traffic intersection (Bongcheon) for several months, and another in a tunnel over two days—and claim that high‑resolution CO2 maps can be generated with this approach.
I remain skeptical about the practicality of this linear‑regression‑based interpolation method. A linear model ignores the geometric constraints inherent to urban environments. Moreover, the paper provides insufficient detail on how the regression coefficients are learned, leaving open the possibility of overfitting. Finally, the authors do not discuss the spatial and temporal variability of those coefficients, which is essential for assessing the method’s real‑world applicability.
Please find below more details.
General comments
- Linear interpolation lacks physical realism. A purely linear model does not incorporate the geometric constraints that govern CO2 transport in cities. This shortcoming is evident in the presented concentration maps, which show little correlation with major roadways. Also, the manuscript does not explain that each spatial coordinate would require its own set of regression coefficients and time‑alignment parameters. A discussion of the spatio-temporal variability of these coefficients is necessary if the method is to be useful in practice.
- Training procedure for the multiple-point linear regression is under‑described. The authors do not specify how the data are split into training and test sets, nor whether any cross‑validation is employed to guard against over‑fitting. Without this information, the reported RMSE metric cannot be trusted.
- Calibration, correction, and interpolation are conflated. At times it is unclear whether the authors refer to sensor calibration, correction, or spatial interpolation. These three steps serve distinct purposes and should be clearly separated.
Specific comments
-
l. 17 – Clarify the local CO2 sources in Seoul (e.g. transportation, heating). When citing the city’s 9 % share of national electricity consumption, indicate whether the associated emissions stem from power plants within Seoul or elsewhere; otherwise the statement lacks context.
-
l. 21 – Explain why a high‑resolution CO₂ map would aid mitigation strategies. A brief illustrative example would help readers understand the use case.
-
l. 25 – Define “top‑down” and “bottom‑up” approaches.
-
l. 30 – How ground‑level CO2 measurements can improve eddy‑covariance flux estimates ?
-
l. 35 – Both sensors are using NDIR, maybe specify the exact sensor models used.
-
l. 43 – State the main contribution explicitly: the development of a spatial‑interpolation method for CO2 using low‑cost sensors.
-
l. 45 – Distinguish clearly among calibration, correction (post‑processing of raw measurements to remove spikes, drift, etc.), and interpolation/estimation (predicting CO2 at unmeasured locations, possibly after time‑lag alignment).
-
l. 60 – Provide the cost ratio between the two sensor types to understand the trade‑off between price and accuracy.
-
l. 64 – Consider reorganizing the methods section into three subsections: “Instrument description,” “Site description,” and “Interpolation method.”
-
l. 80 – Explain the nature of the reported time delays: are they instrument‑specific latencies, or delays relative to the target interpolation location? Discuss how the method would handle interpolation at a location lacking any ground‑truth measurement, and describe the optimization technique used for time‑lag correction.
-
l. 84–95 – Indicate which portion of the dataset is used to learn the regression coefficients and whether any cross‑validation scheme is applied to prevent over‑fitting.
-
l. 90 – Justify the inclusion of temperature and humidity as covariates in the interpolation model.
-
l. 114 – Clarify whether the displayed equation is for to sensor calibration or to the spatial interpolation itself.
-
l. 127 – The RMSE appears to compare CO2 values at two distinct sites simultaneously. One would expect a comparison between the interpolated value and the ground-truth.
-
l. 135 – Confirm whether the low‑cost sensor time-series are temporally aligned before regression.
-
l. 142 – If the interpolation described here is identical to the multiple‑point linear regression introduced at l. 94, state this explicitly and reference the MATLAB implementation in the “Interpolation method” subsection.
-
l. 143 – Expand the discussion of results: why is linear spatial interpolation justified in a complex urban environment? Address the apparent lack of correlation between major roads and elevated CO2 levels.
-
l. 153 – Specify whether the tunnel is two-way.
-
l. 168–176 – Separate statements about calibration from those about data interpolation to avoid confusion.
-
l. 193–205 – The reported decrease in error from the “time‑lag corrections” seem insignificant compared with the reading error (HT‑2000, LI‑870A). The observed variability (> 10 s) may stem from the optimization algorithm rather than true physical lag.
-
l. 194 – Describe the algorithmic procedure used to estimate the time lag.
-
l. 207–211 – As above, the error analysis does not appear meaningful when it is of the order of the sensor’s reading error.
-
l. 215–225 – I disagree with the notion of a universal set of regression coefficients for all locations. Coefficients inevitably depend on local geography and prevailing atmospheric conditions.
-
l. 262 – The concluding claim that the approach enables large‑scale, high‑resolution monitoring is premature. The manuscript does not examine the spatial or temporal stability of the regression coefficients, which is essential to validate the method.
Citation: https://doi.org/10.5194/egusphere-2025-3775-RC2
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 1,356 | 23 | 11 | 1,390 | 23 | 18 | 12 |
- HTML: 1,356
- PDF: 23
- XML: 11
- Total: 1,390
- Supplement: 23
- BibTeX: 18
- EndNote: 12
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1