the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 01 Dec 2023
An intercomparison of satellite, airborne, and ground-level observations with WRF-CAMx simulations of NO2 columns over Houston, TX during the September 2021 TRACER-AQ campaign
Abstract. Nitrogen dioxide (NO2) is a precursor of ozone (O3) and fine particulate matter (PM2.5) – two pollutants that are above regulatory guidelines in many cities. Bringing urban areas into compliance of these regulatory standards motivates an understanding of the distribution and sources of NO2 through observations and simulations. The TRACER-AQ campaign, conducted in Houston, TX in September 2021, provided a unique opportunity to compare observed NO2 columns from ground-, airborne-, and satellite-based spectrometers. In this study, we investigate how these observational datasets compare, and simulate column NO2 using WRF-CAMx with fine resolution (444 x 444 m2) comparable to the airborne column measurements. We find that observations from the GEOCAPE Airborne Simulator (GCAS) instrument were strongly correlated (r2=0.80) to observations from Pandora spectrometers with a negligible bias (NMB=0.1 %). Remote-sensing observations from the TROPOMI instrument were generally well correlated with Pandora observations (r2=0.73) with a negative bias (NMB=-22.8 %). We intercompare different versions of TROPOMI data and find similar correlations across three versions but slightly different biases (from -22.8 % in v2.4.0 to -18.2 % in the NASA MINDS product). Compared to Pandora observations, the WRF-CAMx simulation had reduced correlation (r2=0.34) and a low bias (-25.5 %) over the entire study region. We find particularly poor agreement between simulated NO2 columns and GCAS-observed NO2 columns in downtown Houston an area of high population and roadway densities. These findings point to a potential underestimate of vehicle NOX emissions in the WRF-CAMx simulation driven by the Texas state inventory; and further investigation is recommended.
-
Notice on discussion status
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
-
Preprint
(12187 KB)
-
Supplement
(21785 KB)
-
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(12187 KB) - Metadata XML
-
Supplement
(21785 KB) - BibTeX
- EndNote
- Final revised paper
Journal article(s) based on this preprint
Interactive discussion
Status: closed
-
CC1: 'Comment on egusphere-2023-2844', Steven Compernolle, 11 Dec 2023
With interest I read your manuscript where three different types of measurement are combined.
I have the following comments and questions:
Line 20. Define which correlation statistic is used here. Is it Pearson-R?
Line 21. NMB abbreviation is used in the abstract but is not defined. Does it mean 'normalized mean bias' or 'normalized median bias'
Line 114-115. "Pan #59 at Aldine, #61 at La Porte, and #25 and University of Houston were chosen..."
59 is not in Table 1. 61 is not at La Porte but at Aldine, according to Table 1.Table 1. I would include here also the altitude, as it is a reason for not using P188.
Line 172. "We download the publicly available data (https://data-portal.s5p-pal.com/products/no2.html) "
In fact, only data processed with processor version 2.3.1, with data up to November 2021, can be found here. The latest reprocessed version 2.4.0 (and 2.5.0 for measurements since March 2023) are currently available from the Copernicus Data Space Ecosystem (https://dataspace.copernicus.eu/)Line 186. "A comparison between TROPOMI version 2.4.0 and a MAX-DOAS network found that in moderately polluted locations, TROPOMI had a median bias of -35% (Lambert et al., 2023). "
Not only MAX-DOAS, but also Pandora measurements from PGN are included in this quarterly updated validation report. Given that Pandora measurements are used in the current paper, it is logical to mention these results as well.Line 189-190. What kind of regridding approach is applied?
Line 196-198. More detail is needed here. Which TROPOMI version is used? How exactly are the total and tropospheric values calculated? By calculating new total and tropospheric AMF and in this way recalculating from the available tropospheric and total VCD available in the TROPOMI file?
Note also that in the operational product, there are two total column variables. Normally only the 'summed column' is recommended to be used.
Also, there is already a stratospheric NO2 available in the TROPOMI data files. Why was this not used to correct the GCAS and WRF-CAMx data?Line 309. r^2 and NMB are used but not introduced. Is Pearson-R used? What is the exact mathematical definition of NMB?
I see later in the text 'normalized mean bias' appearing, but that does not yet explain how exactly it is normalized (using only reference data or e.g., the mean of reference and satellite data), and if the normalization happens before or after the averaging.When using relative measures, I recommend using median rather than the mean, to prevent small numbers in the denominator to dominate.
I recommend also to add a statistic on the dispersion. E.g., the standard deviation of the differences.
Fig. 2. MBE, NMB are used in the figures but are not defined.
Also there is a typo in the unit: this should be molecules cm-2 10^16, not molecules cm-2 10^(-16). I recommend to add a dispersion measure like the standard deviation of the differences.Line 312. "but there was a negative bias (NMB=-22.8%) in v2.4.0. "
Here, it is relevant to compare with the TROPOMI vs Pandora results on the validation server (https://mpc-vdaf-server.tropomi.eu).Results using Pandora #25 (university of Houston; PGN location name HoustonTX) are at
https://mpc-vdaf-server.tropomi.eu/no2/no2-offl-pandora/houston-txResults using Pandora #58 are at https://mpc-vdaf-server.tropomi.eu/no2/no2-offl-pandora/la-porte-tx-gsfc058
While it is not fully clear to me what is the meaning of MBE and NMB, the numbers seem broadly consistent with the bias statistics obtained from the validation server, despite the fact that the sampling is not the same (as on the validation server no subsampling to the airborne measurements is done). This indicates that the sample taken in the current paper is a representative set.
Note that on the server, the filter qa_value>75% is not applied yet (only qa_value>50%), so you'll first have to download the data and apply this more strict filter yourself to obtain exactly the numbers below. This setting was taken to have more pixels also for the sites where the stratospheric contribution is dominant.
La Porte TX P58:
mean difference=-2.5 10^15 molec/cm2
median difference =-2.2 10^15 molec/cm2
mean relative difference (i.e., mean of (SAT-REF)/REF)=-21%
median relative difference (i.e., median of (SAT-REF)/REF)=-24%Houston TX P25:
mean difference=-2.3 10^15 molec/cm2
median difference =-1.9 10^15 molec/cm2
mean relative difference =-21%
median relative difference=-23%Finally, also the comparison with La Porte TX P63 (mentioned in Table 1 but not used in the paper) is available on the validation server. However, in this case the bias is quite different, due to Pandora values being lower (about 2 Pmolec cm-2 lower when calculating on an overlapping time range 2021-10-07 to 2022-01-22).
In general, it would be interesting if the authors added a consistency check of the Pandora instruments that are at the same location, perhaps in section 2.1.Line 607. 'both outperform version 2.4.0'. The bias of NASA MINDS and v2.3.1 is indeed slightly better, but the correlation of NASA MINDS is lower than that of 2.4.1. Also, it would be interesting to look at a dispersion measure (e.g., standard deviation of the differences).
Line 643. "It is common to consider TROPOMI measurements as accurate representation of NO 2 column concentrations; however, if we had done so in this study, we would have failed to identify the substantial negative bias"
Here, it is useful to remind that a negative bias has been mentioned in earlier literature (e.g., Verhoelst et al, 2021) and in the quarterly issued operational validation reports (available at https://mpc-vdaf.tropomi.eu/). This information is therefore already available to the user.Line 648. In the acknowledgments, there should as a minimum appear: 'contains modified Sentinel data'.
An example sentence could be 'This work contains modified Copernicus Sentinel-5 Precursor
data processed by KNMI and post-processed by ... (ending with the name of your research institute).Citation: https://doi.org/10.5194/egusphere-2023-2844-CC1 -
RC1: 'Comment on egusphere-2023-2844', Anonymous Referee #1, 19 Dec 2023
This work conducted inter-comparisons of NO2 columns between GCAS, TROPOMI, Pandora, and WRF-CAMx during TRACER-AQ campaign over Houston, TX. Using Pandora measurements as reference, authors found good agreements in GCAS, and systematical underestimation in TROPOMI and WRF-CAMx. The analyses are generally sound and comprehensive. My major concern is that WRF-CAMx is performed at a very high resolution (444 x 444 m2) but is lack of enough meteorology evaluations, which is very important for column comparisons. In this case, it’s difficult to attribute the model bias to emission inventory, and arbitrary to conclude the potential underestimation of vehicular NOx emissions from the Texas state inventory. Also, authors should clarify if the NO2 columns are tropospheric or the total columns, to avoid misunderstanding.
My detailed comments are listed below.
- Line 21: please give the full name of TROPOMI in the first instance.
- Line 27 – 29: I doubt this conclusion because there are no meteorology evaluations of the model.
- Line 102: please specify if the column is tropospheric, or the total.
- Line 113 – 114: not sure what this sentence means. Please rephrase it.
- Line 164 – 166: in TROPOMI, the separation of tropospheric and stratospheric column is also another big source of uncertainty. Using the stratospheric NO2 columns derived from TROPOMI can bring in unpredicted bias as well. Can you give analyses on the stratospheric NO2 columns here? What’s the data range and spatial distributions?
- Sect. 2.4: at such high spatial resolution, how does meteorology work?
- Line 248 – 250: what’s the resolution of the emission inventory? For CEMS, they are point sources with exact geophysical locations. How about other sources? The in-compatible spatial resolution between model and inventory can be a big issue.
- Line 277 – 279: TROPOMI and GCAS AMF calculation not only include geometry, but also on a priori information from models. I’m just curious how does Pandora calculate the air mass factor without the a priori NO2 vertical profiles? Can you elaborate more on this?
- Line 613 – 614: This, further emphasize the needs of meteorology evaluations in the model.
Citation: https://doi.org/10.5194/egusphere-2023-2844-RC1 - RC2: 'Comment on egusphere-2023-2844', Anonymous Referee #2, 21 Dec 2023
- AC1: 'Response to the reviewers of egusphere-2023-2844', Omar Nawaz, 01 Mar 2024
Interactive discussion
Status: closed
-
CC1: 'Comment on egusphere-2023-2844', Steven Compernolle, 11 Dec 2023
With interest I read your manuscript where three different types of measurement are combined.
I have the following comments and questions:
Line 20. Define which correlation statistic is used here. Is it Pearson-R?
Line 21. NMB abbreviation is used in the abstract but is not defined. Does it mean 'normalized mean bias' or 'normalized median bias'
Line 114-115. "Pan #59 at Aldine, #61 at La Porte, and #25 and University of Houston were chosen..."
59 is not in Table 1. 61 is not at La Porte but at Aldine, according to Table 1.Table 1. I would include here also the altitude, as it is a reason for not using P188.
Line 172. "We download the publicly available data (https://data-portal.s5p-pal.com/products/no2.html) "
In fact, only data processed with processor version 2.3.1, with data up to November 2021, can be found here. The latest reprocessed version 2.4.0 (and 2.5.0 for measurements since March 2023) are currently available from the Copernicus Data Space Ecosystem (https://dataspace.copernicus.eu/)Line 186. "A comparison between TROPOMI version 2.4.0 and a MAX-DOAS network found that in moderately polluted locations, TROPOMI had a median bias of -35% (Lambert et al., 2023). "
Not only MAX-DOAS, but also Pandora measurements from PGN are included in this quarterly updated validation report. Given that Pandora measurements are used in the current paper, it is logical to mention these results as well.Line 189-190. What kind of regridding approach is applied?
Line 196-198. More detail is needed here. Which TROPOMI version is used? How exactly are the total and tropospheric values calculated? By calculating new total and tropospheric AMF and in this way recalculating from the available tropospheric and total VCD available in the TROPOMI file?
Note also that in the operational product, there are two total column variables. Normally only the 'summed column' is recommended to be used.
Also, there is already a stratospheric NO2 available in the TROPOMI data files. Why was this not used to correct the GCAS and WRF-CAMx data?Line 309. r^2 and NMB are used but not introduced. Is Pearson-R used? What is the exact mathematical definition of NMB?
I see later in the text 'normalized mean bias' appearing, but that does not yet explain how exactly it is normalized (using only reference data or e.g., the mean of reference and satellite data), and if the normalization happens before or after the averaging.When using relative measures, I recommend using median rather than the mean, to prevent small numbers in the denominator to dominate.
I recommend also to add a statistic on the dispersion. E.g., the standard deviation of the differences.
Fig. 2. MBE, NMB are used in the figures but are not defined.
Also there is a typo in the unit: this should be molecules cm-2 10^16, not molecules cm-2 10^(-16). I recommend to add a dispersion measure like the standard deviation of the differences.Line 312. "but there was a negative bias (NMB=-22.8%) in v2.4.0. "
Here, it is relevant to compare with the TROPOMI vs Pandora results on the validation server (https://mpc-vdaf-server.tropomi.eu).Results using Pandora #25 (university of Houston; PGN location name HoustonTX) are at
https://mpc-vdaf-server.tropomi.eu/no2/no2-offl-pandora/houston-txResults using Pandora #58 are at https://mpc-vdaf-server.tropomi.eu/no2/no2-offl-pandora/la-porte-tx-gsfc058
While it is not fully clear to me what is the meaning of MBE and NMB, the numbers seem broadly consistent with the bias statistics obtained from the validation server, despite the fact that the sampling is not the same (as on the validation server no subsampling to the airborne measurements is done). This indicates that the sample taken in the current paper is a representative set.
Note that on the server, the filter qa_value>75% is not applied yet (only qa_value>50%), so you'll first have to download the data and apply this more strict filter yourself to obtain exactly the numbers below. This setting was taken to have more pixels also for the sites where the stratospheric contribution is dominant.
La Porte TX P58:
mean difference=-2.5 10^15 molec/cm2
median difference =-2.2 10^15 molec/cm2
mean relative difference (i.e., mean of (SAT-REF)/REF)=-21%
median relative difference (i.e., median of (SAT-REF)/REF)=-24%Houston TX P25:
mean difference=-2.3 10^15 molec/cm2
median difference =-1.9 10^15 molec/cm2
mean relative difference =-21%
median relative difference=-23%Finally, also the comparison with La Porte TX P63 (mentioned in Table 1 but not used in the paper) is available on the validation server. However, in this case the bias is quite different, due to Pandora values being lower (about 2 Pmolec cm-2 lower when calculating on an overlapping time range 2021-10-07 to 2022-01-22).
In general, it would be interesting if the authors added a consistency check of the Pandora instruments that are at the same location, perhaps in section 2.1.Line 607. 'both outperform version 2.4.0'. The bias of NASA MINDS and v2.3.1 is indeed slightly better, but the correlation of NASA MINDS is lower than that of 2.4.1. Also, it would be interesting to look at a dispersion measure (e.g., standard deviation of the differences).
Line 643. "It is common to consider TROPOMI measurements as accurate representation of NO 2 column concentrations; however, if we had done so in this study, we would have failed to identify the substantial negative bias"
Here, it is useful to remind that a negative bias has been mentioned in earlier literature (e.g., Verhoelst et al, 2021) and in the quarterly issued operational validation reports (available at https://mpc-vdaf.tropomi.eu/). This information is therefore already available to the user.Line 648. In the acknowledgments, there should as a minimum appear: 'contains modified Sentinel data'.
An example sentence could be 'This work contains modified Copernicus Sentinel-5 Precursor
data processed by KNMI and post-processed by ... (ending with the name of your research institute).Citation: https://doi.org/10.5194/egusphere-2023-2844-CC1 -
RC1: 'Comment on egusphere-2023-2844', Anonymous Referee #1, 19 Dec 2023
This work conducted inter-comparisons of NO2 columns between GCAS, TROPOMI, Pandora, and WRF-CAMx during TRACER-AQ campaign over Houston, TX. Using Pandora measurements as reference, authors found good agreements in GCAS, and systematical underestimation in TROPOMI and WRF-CAMx. The analyses are generally sound and comprehensive. My major concern is that WRF-CAMx is performed at a very high resolution (444 x 444 m2) but is lack of enough meteorology evaluations, which is very important for column comparisons. In this case, it’s difficult to attribute the model bias to emission inventory, and arbitrary to conclude the potential underestimation of vehicular NOx emissions from the Texas state inventory. Also, authors should clarify if the NO2 columns are tropospheric or the total columns, to avoid misunderstanding.
My detailed comments are listed below.
- Line 21: please give the full name of TROPOMI in the first instance.
- Line 27 – 29: I doubt this conclusion because there are no meteorology evaluations of the model.
- Line 102: please specify if the column is tropospheric, or the total.
- Line 113 – 114: not sure what this sentence means. Please rephrase it.
- Line 164 – 166: in TROPOMI, the separation of tropospheric and stratospheric column is also another big source of uncertainty. Using the stratospheric NO2 columns derived from TROPOMI can bring in unpredicted bias as well. Can you give analyses on the stratospheric NO2 columns here? What’s the data range and spatial distributions?
- Sect. 2.4: at such high spatial resolution, how does meteorology work?
- Line 248 – 250: what’s the resolution of the emission inventory? For CEMS, they are point sources with exact geophysical locations. How about other sources? The in-compatible spatial resolution between model and inventory can be a big issue.
- Line 277 – 279: TROPOMI and GCAS AMF calculation not only include geometry, but also on a priori information from models. I’m just curious how does Pandora calculate the air mass factor without the a priori NO2 vertical profiles? Can you elaborate more on this?
- Line 613 – 614: This, further emphasize the needs of meteorology evaluations in the model.
Citation: https://doi.org/10.5194/egusphere-2023-2844-RC1 - RC2: 'Comment on egusphere-2023-2844', Anonymous Referee #2, 21 Dec 2023
- AC1: 'Response to the reviewers of egusphere-2023-2844', Omar Nawaz, 01 Mar 2024
Peer review completion
Journal article(s) based on this preprint
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 321 | 142 | 22 | 485 | 43 | 12 | 15 |
- HTML: 321
- PDF: 142
- XML: 22
- Total: 485
- Supplement: 43
- BibTeX: 12
- EndNote: 15
Viewed (geographical distribution)
| Country | # | Views | % |
|---|---|---|---|
| United States of America | 1 | 253 | 53 |
| China | 2 | 36 | 7 |
| United Kingdom | 3 | 32 | 6 |
| Netherlands | 4 | 24 | 5 |
| India | 5 | 21 | 4 |
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
- 253
M. Omar Nawaz
Jeremiah Johnson
Greg Yarwood
Benjamin de Foy
Laura M. Judd
Daniel L. Goldberg
The requested preprint has a corresponding peer-reviewed final revised paper. You are encouraged to refer to the final revised version.
- Preprint
(12187 KB) - Metadata XML
-
Supplement
(21785 KB) - BibTeX
- EndNote
- Final revised paper