the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 13 Feb 2025
Satellite Detection of NO2 Distributions and Comparison with Ground-Based Concentrations
Abstract. In this study we assess the capability of current-generation satellites to capture the variability of near-surface nitrogen dioxide (NO2) monitoring data, with the goal of supporting health and regulatory applications. We consider NO2 vertical column densities (VCD) over the United States from two satellite instruments, the Tropospheric Monitoring Instrument (TROPOMI), and Tropospheric Emissions: Monitoring of Pollution (TEMPO), and compare with ground-based concentrations as measured by the EPA’s Air Quality System (AQS) monitors. While TROPOMI provides a longer-term record of assessment (2019–2023), TEMPO informs diurnal patterns relevant to evaluating peak NO2. We analyze frequency distributions and quantify their similarity using the Jensen-Shannon Divergence (JSD), where smaller values indicate better agreement. Satellite and ground monitor NO2 distributions are most similar away from major roads, as indicated by the JSD of 0.008 calculated for TROPOMI and ground monitors at non-roadways, compared with a JSD near interstates of 0.158 and a JSD near highways of 0.095. Seasonal analysis shows the most similarity in distributions in winter, with a JSD of 0.010, and the most difference in summer, with a JSD of 0.035. Across seasons and monitor locations, TEMPO consistently has a lower or similar JSD as TROPOMI, with TEMPO JSDs ranging from 0.005 to 0.151 and TROPOMI JSDs ranging from 0.012 to 0.265. TEMPO’s agreement with monitors in both December 2023 and July 2024 is found to be best around midday, with non-road monitors’ JSD in July as low as 0.008 at 16 UTC (~11 am LT).
- Preprint
(1646 KB) - Metadata XML
-
Supplement
(426 KB) - BibTeX
- EndNote
Status: open (until 24 Apr 2025)
-
RC1: 'Comment on egusphere-2025-226', Anonymous Referee #1, 13 Feb 2025
reply
Overall, this is an interesting area of study and the manuscript could become stronger if my concerns are addressed.
Abstract: The authors should consult the “guidelines for authors”: https://www.atmospheric-chemistry-and-physics.net/policies/guidelines_for_authors.html. For instance, the title is not specific and the abstract lacks specific conclusions on the important implications of the work (i.e., just JSD values are given).
Introduction: The introduction is fairly well written, but verbose. Please consider condensing text given that the audience for this journal article will already be familiar with much of the information provided. The paragraphs on advanced methods, frequency distributions and JSD could be moved to “data and methods”.
Section 2: How are the data averages binned in time? Seasons? Months?
Sections 2.2-2.3: It would be worth discussing the number of satellite pixels that are required for comparison to your distance definitions in Section 2.4. That is, do most of the distances fall within one relative-coarse satellite pixel? Two pixels? Is this the reason there is little variation in the satellite results presented in Figure 1? Could you address this issue with oversampling or some other similar technique?
Line 214: mixing meters and miles. Is that how distances are reported for the AQS monitors? Why not meters and kilometers or feet and miles?
Section 2.4: Concerning the way that you classified the monitors, would it make sense to create categories such as “rural interstates”, “urban interstates”, “industrial interstates”, etc. The location of nearby non-interstate roads and industry will certainly affect your results, especially in the pixel size of a satellite instrument. Development typically occurs along major roadways. You could create such bins based on population density or some other metric that indicates how developed an area is.
Figure 1: The time period isn’t specified. A summer month? An annual average? Could you show these relationships for four seasons? Regions?
Figure 3: Are the TEMPO observations filtered for the same time of day as the TROPOMI overpass to make this more of an apples-to-apples comparison? I see now in the paragraph beginning on line 451 that the TEMPO data are for all hours.
Section 3: You present a lot of statistics, which can be overwhelming to a reader. Could you begin each Section 3 subsection with a statement like “In this subsection, we show that our data analysis indicates [insert high level conclusion here].” That is, tell the reader up front what the take-away messages are.
Section 4 last paragraph: Can you elaborate on this with an example or two of how your study may help? I’m worried that people involved in these activities may not know how to translate your results into action.
Citation: https://doi.org/10.5194/egusphere-2025-226-RC1 -
AC1: 'Reply on RC1', Summer Acker, 17 Mar 2025
reply
Thank you so much for your helpful feedback! We have carefully revised our manuscript to address all of your comments to the best of our ability and believe these changes have strengthened the clarity and impact of our work. Below, we detail how each comment has been addressed in the revised manuscript.
Abstract: The authors should consult the “guidelines for authors”: https://www.atmospheric-chemistry-and-physics.net/policies/guidelines_for_authors.html. For instance, the title is not specific and the abstract lacks specific conclusions on the important implications of the work (i.e., just JSD values are given).
We consulted the “guidelines for authors” and have updated the title and abstract to be more specific. The abstract now includes the broader implications for our findings.
Introduction: The introduction is fairly well written, but verbose. Please consider condensing text given that the audience for this journal article will already be familiar with much of the information provided. The paragraphs on advanced methods, frequency distributions and JSD could be moved to “data and methods”.
We have condensed the Introduction by removing background information that may be familiar to the journal’s audience. Additionally, we removed the detailed discussion of frequency distributions and moved the Jensen-Shannon Divergence (JSD) metric to the Data and Methods section (section 2.5), as suggested.
Section 2: How are the data averages binned in time? Seasons? Months?
We have updated methods section 2.5 to explain how the daily data is binned for the distribution and Jensen-Shannon Divergence (JSD) calculations as well as how different analyses aggregate data by distance from roads, seasons, and specific months.
Sections 2.2-2.3: It would be worth discussing the number of satellite pixels that are required for comparison to your distance definitions in Section 2.4. That is, do most of the distances fall within one relative-coarse satellite pixel? Two pixels? Is this the reason there is little variation in the satellite results presented in Figure 1? Could you address this issue with oversampling or some other similar technique?
We analyzed the number of NO2 monitors falling within each TROPOMI pixel by performing a spatial join between monitor locations and the oversampled 4 km × 4 km TROPOMI grid. Our analysis shows that 97% of TROPOMI pixels contain only one monitor, with only 2.7% containing more than one, indicating that Figure 1 is not dominated by a small subset of TROPOMI pixels. Since the monitors are spatially distributed across the U.S., most are assigned to distinct TROPOMI pixels rather than repeatedly falling within the same ones, ensuring that a broad range of TROPOMI grid cells is represented. We discuss this in Section 2.2 and have added a new supplemental figure (Figure S1) illustrating the spatial distribution of monitor counts per TROPOMI pixel and valid TROPOMI retrieval days. Oversampling to a finer grid would require us to average the satellite data on a monthly or annual basis, but this study focuses on daily measurements. Future work may explore 1 km x 1 km satellite data on a monthly or annual average temporal scale.
Line 214: mixing meters and miles. Is that how distances are reported for the AQS monitors? Why not meters and kilometers or feet and miles?
We have updated Figure 1 to use meters and kilometers to maintain consistency.
Section 2.4: Concerning the way that you classified the monitors, would it make sense to create categories such as “rural interstates”, “urban interstates”, “industrial interstates”, etc. The location of nearby non-interstate roads and industry will certainly affect your results, especially in the pixel size of a satellite instrument. Development typically occurs along major roadways. You could create such bins based on population density or some other metric that indicates how developed an area is.
We appreciate this suggestion and agree that land-use and urban vs rural areas influence NO2 levels. We further classified interstate monitors as urban or rural interstates by spatially joining the monitor Point data with the Census Bureau’s 2020 Urban Areas Tiger/Line Shapefile. Since we only found one interstate monitor that was in a rural area, we decided not to do further analysis, but have mentioned this in in section 2.4.
Figure 1: The time period isn’t specified. A summer month? An annual average? Could you show these relationships for four seasons? Regions?
We have clarified that the time period used in Figure 1 is 2019 to 2023 within the figure caption and main text, specifying that the distributions represent daily observations. We also added a brief paragraph and a supplemental figure (Figure S4) on the seasonal relationships.
Figure 3: Are the TEMPO observations filtered for the same time of day as the TROPOMI overpass to make this more of an apples-to-apples comparison? I see now in the paragraph beginning on line 451 that the TEMPO data are for all hours.
For the comparison of TEMPO with TROPOMI (section 3.2), the UTC equivalents of 1 pm and 2 pm LT were determined for each time zone based on the latitude and longitude of each monitor location. TEMPO NO2 values corresponding to these calculated UTC hours were averaged to align with the TROPOMI overpass time (~13:30 LST). Similarly, for ground-based measurements, the monitor data were filtered to include only values corresponding to 1 pm and 2 pm LT and then averaged. We have clarified this throughout the text. Section 3.3 uses hourly TEMPO data.
Section 3: You present a lot of statistics, which can be overwhelming to a reader. Could you begin each Section 3 subsection with a statement like “In this subsection, we show that our data analysis indicates [insert high level conclusion here].” That is, tell the reader up front what the take-away messages are.
We have added a high-level conclusion statement to the beginning of each results subsection for more clarity.
Section 4 last paragraph: Can you elaborate on this with an example or two of how your study may help? I’m worried that people involved in these activities may not know how to translate your results into action.
We have removed the final paragraph to avoid speculation on ways our results could be applied by public health managers.
Citation: https://doi.org/10.5194/egusphere-2025-226-AC1
-
AC1: 'Reply on RC1', Summer Acker, 17 Mar 2025
reply
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 171 | 41 | 7 | 219 | 13 | 4 | 3 |
- HTML: 171
- PDF: 41
- XML: 7
- Total: 219
- Supplement: 13
- BibTeX: 4
- EndNote: 3
Viewed (geographical distribution)
| Country | # | Views | % |
|---|---|---|---|
| United States of America | 1 | 143 | 62 |
| China | 2 | 14 | 6 |
| France | 3 | 7 | 3 |
| India | 4 | 7 | 3 |
| Netherlands | 5 | 6 | 2 |
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
- 143