|Revised submission-- “Mobile monitoring of urban air quality at high spatial resolution by low-cost sensors: Impacts of COVID-19 pandemic lockdown”|
The authors have considered responses to each of the points raised by the reviewers. However, there are key points at the base of earlier comments that are still quite unclear. These include: 1) specifics on the sensors used as basis for data source and 2) the implications of sensor choice, and 3) the implications of sensor calibration protocols on the data and its use in a traffic impacted environment.
A final, perhaps less important general concern is with the title of this paper. The title clearly shows this was a study of the impacts of COVID lockdowns. This event was a “nice” add-on, but it is not at all a major part of the effort. I suggest a title that better reflects the nature of the study.
The overall concern is that while the modelling and allocation of pollutants to roadways are interesting demonstrations of mobile sensor-based monitoring it is unclear that actual and accurate pollutant data represents air quality on roadways. And while the calibration data from a Nanjing University campus show good performance it is not clear that this translates to accurate data from on road measurements. And without further information on sensors and the data they produced for this study it is not possible to evaluate or accept the modelling results.
1. Specifics on sensors
The current version now adds a few words to identify the sensors as electrochemical cells, but there are many possible sources of cells of this type, and they differ. Please provide the make and models of the sensors. And fully describe any special characteristics that apply to the sensor calibration and validation in this study. For example, do they include any chemical filtration?
2. Implications of sensor choice
Electrochemical sensors are not fully chemically specific in response. This complicates their use in complex atmospheres—such as near roadways due to established interactions between ozone and nitrogen dioxide. Most “ozone” electrochemical sensors are “oxidant” sensors since they respond to oxidants (especially ozone and NO2). Data from them should not be simply viewed as ozone, as is the case in this study, since it is performed on NO2-rich roadways where ozone is likely to be disproportionately lower than in ambient air. NO2 electrochemical sensors may also have interference issues. Without clear identification of the sensors and a description of any features or data correction steps taken, it is not possible to understand the nature of data from either the “ozone” or NO2 sensors. Use of data from these sensors in subsequent plotting and modeling are open to considerable uncertainty. The reviewer is concerned of the use of sensor data as an off-the-shelf product without sufficient quality control and attention to details.
The data collected while on road may be quite different than those from ambient community sites. This could complicate ozone reported. The authors should describe how ozone data was produced from the output of sensors and whether they considered interferences due to NO2. And depending on the NO2 sensor, they should address interferences that may occur for that sensor in the atmosphere under study. Because the accuracy of these two data streams is essential to the overall study, the details presented at adequate depth to inform the reader of what was done.
3. Calibration issues
Calibration is a key activity to assure good data from sensors. In this case the calibrations were performed using periodic co-location at an air monitoring station located at a campus of Nanjing University that may be distant from the urban center. This is possibly an acceptable approach, however some studies have shown the calibration results may not be transferrable especially crossing different concentration ranges or different environments. Expanded consideration of environmental conditions included in the calibration and how the change of environmental conditions can affect the are needed to help ensure credibility of data when sensors are deployed in the field. Again, back to sensor selection and data issues—The mix of ozone and NO2 are likely to be quite different between the roadways and the fixed site. Are there near-road ambient air monitoring sites in Nanjing that can show NO2, NO an O3 data levels that can be compared to the university site (which seems quite far from busy roadways)? How do they differ? How might these differences impact the utility of the calibrations performed?
It is important in sensor-based papers, that calibration as part of robust quality control and assurance is evident to ensure sufficient confidence for data interpretation or data fusion with modelling results. Overall, these topics are weakly considered in the manuscript.
One further point is identified that is separate from the above point regarding sensors and sensor data. That is with regards to the use of data from these two taxis to represent an entire urban area road grid. It seems that in many of the observed roadway segment data were collected perhaps once or twice over a one- or two-month interval while in other cases several observations may have occurred on a single day. The reviewer is not convinced this small data can be representative enough to be compared with modeling results to draw meaning conclusion. How might the nature of the temporal (even seasonal) nature of data quality impact the representation of the city roadways? Are two mobile monitors enough for a city the size of Nanjing?
Line 77—states that taxis with natural gas and electric power were used. It it unclear how many taxis , but line 67 says that 2 were used in this study. If this is correct, may we assume this means one each?
Section 2.2, line 90—the authors list important characteristics to be considered with use of sensors. The list does not include specificity and interference issues or adequate details/citations on these matters. Overall, section 2.2 needs considerable editing and clarification. Calibration is a key point for this study.
It is stated that NO2 comparisons with the fixed station calibrations, which was only fair (R2=0.67 for sensor 2), were improved by training and inclusion of t and rh. But the resulting improvements are not shown in this section.
Line 107—added text is unclear and perhaps is important. “The success of supervised model training with target labels (i.e. co-located with SORPES, Figure 2a) does not guarantee for its predicting power for conditions without labels (i.e. on road or co-located with SORPES but not feeding the station data to the algorithm, Figure 2b)”. Perhaps this is related to key point 2. However, it should be rewritten and clarified. If it is related to point 2 then it should be considerably expanded.
Line 140—it is unclear how the overall data reductions employed highly time and spatially resolved data and generated hourly average data. What was this data used for?
Line 143—what was the accuracy and data completeness for GPS in the urban areas. Others have found such data difficult to collect reliable complete files in urban canyon conditions.
Line 163—“background” data are generated by finding the minimum value of all the stations in the Nanjing area. Is it clear that they represent the urban “background”? Overall, this text is unclear. Are any of these remote from the city or roadside stations? If so, how might these impact any determinations of background? Roadside station, for example might have very low ozone levels due to reaction between NOx and ozone. But this might not be urban “background”. Please expand discussion of background assumptions.
Line 185—reference to the Apte study may or may not be valuable or applicable here. It was performed in a small area of a city—16km2 and employed considerable resampling of roadways over a prolonged period. Does it apply well in this much larger city/region?
Line 239/246 and table 1—how were specific sources of pollutants, such as cooking identified, as source contributors to hotspots? Only one cooking establishment is actually reported as a hotspot contributor. Basically, a structured assessment of hotspots vs. sources is valuable, but this paper does not present any robust information on what allowed the identification of contributions beyond visual sightings or general proximity.