Observed decreases in on-road CO2 concentrations in Beijing during COVID-19

To prevent the spread of the COVID-19 epidemic, restrictions such as “lockdown”, were conducted globally, which led to significant reduction in fossil fuel emissions, especially in urban regions. However, CO2 concentrations in urban regions are affected by many factors, such as weather and background CO2 fluctuations. Thus, it is difficult to directly observe the reductions in CO2 concentrations with sparse ground observations. Here, we focus on urban ground transportation emissions, 20 which were dramatically affected by the prohibitions, to determine the reduction signals. We conducted six on-road CO2 observations in Beijing using mobile platforms before (BC), during (DC) and after COVID-19 prohibitions (AC). To reduce the weather and background impacts, we chose trips with the most similar weather as possible and calculated the enhancement, which mean the difference in the CO2 concentration between on-road and the “background” level measured at the Institute of Atmospheric Physics, Chinese Academy of Sciences (IAP) tower. The results showed that DC CO2 25 enhancement decreased by 41 parts per million (ppm) and 26 ppm compared to those during BC and AC, respectively, after eliminating the fluctuations in CO2 concentrations on polluted days. Detailed analysis showed that, during COVID, there was no difference between weekdays and weekends. The enhancements during rush hours were almost twice those during working hours, indicating that emissions during rush hours were much higher. Compared with DC and BC, the reductions in the enhancements during rush hours were much larger than those during working hours. Our findings showed a clear 30 decrease during COVID, which are consistent with the CO2 concentration and emissions reductions due to the pandemic. The enhancement way used in this study is an effective method to reduce the impacts of weather and background fluctuation and should be regularly and more frequently conducted in future work. https://doi.org/10.5194/acp-2020-966 Preprint. Discussion started: 12 October 2020 c © Author(s) 2020. CC BY 4.0 License.

fluctuations to analyze CO 2 concentration characteristics in urban areas and has been widely used for monitoring urban 80 carbon emissions and CO 2 concentration (Idso et al., 1998;Idso et al., 2002;George et al., 2007;Mitchell et al., 2018;Perez et al., 2009).
To determine the CO 2 concentration reduction "signal" due to ground transportation emissions decrease during COVID-19, we first chose the most similar weather condition as possible; second, we calculated enhancements by using on-road CO 2 85 observations minus the "baseline" IAP tower CO 2 concentration to reduce the influence from the fluctuation in the background CO 2 concentration due to weather. Our results may provide direct evidence of ground transportation emission reductions due to , and this method could be an appropriate tool to analyze the CO 2 concentration and emissions of urban ground transportation in the future works.  . (C) 95 China ground transport daily CO 2 emissions in the first five months of 2019 and 2020, data from .
conducted during the daytime, in which four of them were on weekdays and others were on Saturday. Four trips covered at least one rush hour.

105
To reduce the background fluctuations, we first chose the similar weather conditions. Three elements were considered: (1) reality photos collected from the IAP tower (photo available from: http://view.iap.ac.cn:8080/imageview/); (2) the PM 2.5 (atmospheric particulate matter that has a diameter of less than 2.5 µm ) concentration from the Olympic Sports Center Station (40.003°N, 116.407°E, 5 m height, purple square in Figure 2A), which is run by the Ministry of Ecology and 110 Environment of China (Zhang et al., 2015); (3) and wind speed data (available collected from: https://www.wunderground.com/history/daily/cn/beijing/ZBNY/date/2020-5-9).
Then, on-road CO 2 concentration enhancements were calculated by subtracting the simultaneous CO 2 concentrations from the IAP tower, which implies the "baseline" in Beijing city (Eq. 1).
Before 2020 (including the trip on 20 th February 2019), the CO 2 concentrations were measured at the lower and upper levels alternately for every 5 minutes, and each level lasted 5 minutes. After 2020 (including the other 5 trips), the CO 2 concentration was continuously measured at the surface level. To maintain consistency as much as possible, we used the 135 lower level CO 2 concentrations before 2020 and surface levels after 2020.
Three different CO 2 observation instruments were carried by vehicles for six on-road trips ( Table 2).
1) On 20 th February 2019, a Picarro G2401(Picarro, 2017) was adopted and installed on a vehicle; the air intake was set on 140 the roof of the vehicle to avoid potential direct plumes emitted from surrounding cars. The intake was linked/connected through a 2 m pipe with a particulate matter filter to Picarro ( Figure 3A and 3B). The instruments characteristics and precision have been described by Sun et al.(Sun et al., 2019). The CO 2 concentrations were collected every 2 seconds and then averaged into 1-minute intervals.
2) During COVID-19 (surveys on 13 th , 20 th , 21 st and 22 nd February 2020), a LI-COR LI-7810 CH 4 /CO 2 /H 2 O trace gas 145 analyzer was adopted, which uses optical feedback-cavity enhanced absorption spectroscopy technology(LI-COR, 2019). This instrument could obtain a CO 2 concentration with a precision of 3.5 ppm for 1 second and within 1 ppm after 1-minute averaging (lab testing). The observation platform of the LI-7810 was similar to that of Picarro. Before departure, the instrument was calibrated by using standard calibration gas (from MOC/CMA) to correct the drift.
3) On 9 th May 2020, a low-cost light sensor was adopted and installed on the front windshield of the vehicle ( Figure 3C).

150
The instrument mainly consisted of three non-dispersive infrared (NDIR) CO 2 measurement sensors (named K30), and one environment (temperature, humidity and pressure) sensor (named BME). Although the original precision of each K30 is ±30 ppm, after calibration and environmental correction in the lab, it was improved to within ±5 ppm comparing with Picarro (Martin et al., 2017;SenseAir, 2019). Here, we used three K30s in one instrument to recognize and eliminate data anomalies and used the averaged CO 2 concentrations from the three K30s for analysis. Figure 4 shows 155 the experiment conducted on 22 nd February 2020, which installed one low-cost light sensor and Picarro on the same vehicle for on-road monitoring. The results showed that the low-cost light sensor results are highly consistent with those of Picarro, with root mean square errors (RMSEs) less than 5 ppm.

170
Auxiliary data and analysis: The global positioning system (GPS) data during BC and DC were collected by a GPS receiver (BS-70DU) (Sun et al., 2019).
During AC, the data were collected by using mobile software (GPS Tracks), which provided time, longitude, latitude, speed and altitude at 1 second resolution. These geographic information data were averaged into 1-minute intervals and then 175 matched with CO 2 concentration data according to time.
Two remote sensing images were adopted (captured on 21 st February 2019 at 11:40:00 Local Standard Time (LST), from a Google Earth image, with 0.37 m spatial resolution; 19 th February 2020 at 10:20:08 LST, from a Beijing-2 remote sensing satellites panchromatic image, with 0.8 m spatial resolution). Considering the availability of data, we used the images from 180 the closest date and only part of the urban area. The comparison region covered 13.4 km of the 4 th Ring Road (accounting for 20.5 % of the whole road, for which the total length is 65.3 km) and 10 km of the 3 rd Ring Road (accounting for 20.7 % of the whole road, for which the length is 48.3 km). We used a visual interpretation method to obtain the numbers of vehicles on the 4 th and 3 rd Ring Roads before and during the COVID-19, respectively.

195
The CO 2 concentration maps of six on-road trips are shown in Figure 5. One trip was chosen as an example for BC (on 20 th February 2019), DC (21 st February 2020) and AC (9 th May 2020) (shown in Figure 5A, 5D and 5F). All three trips were conducted on clear days, and their trajectories were similar that from the outermost circle to the innermost circle and covered one (morning or evening) rush hour. The difference was that the BC and DC trips hit the evening rush hour on the innermost 200 circle road, whereas the AC trip hit the morning rush hour on the outermost circle. This difference explained the CO 2 concentration patterns ( Figure 5A, 5D and 5F). The comparison of the three trips indicated that the CO 2 concentration measured in Figure 5D was intuitively lower than that in Figure 5A and 5F, and the statistics show that the DC CO 2 mean was approximately 58 and 46 ppm lower than that of the BC and AC trips, respectively. In addition, the average CO 2 concentration observed by the IAP tower during the same periods was much lower than the on-road observations (Figure 2).

205
These concentration differences (gradients) also implied that ground transportation emissions were a major CO 2 source on urban roads.
The other three DC trips (on 13 th , 20 th and 22 nd February 2020) are shown in Figure 5B, 5C and 5E, with the averaged CO 2 concentrations of 508, 501 and 442 ppm. Due to background CO 2 concentration fluctuations (lightly polluted days), CO 2 210 concentrations on 13 th and 20 th February ( Figure 5B and C) were as high as those during the BC and AC trips. Statistically, without considering the variation in the background CO 2 concentration, the average of the four DC trips was 477 ppm, which was 36 and 24 ppm lower than that of the BC and AC trips, respectively.  Figure 6 shows the CO 2 enhancement maps of six trips, using the on-road CO 2 concentration minus those of IAP tower at the same time. The enhancements present a refined spatial distribution of the CO 2 concentration gradient, which implies ground transportation CO 2 emissions. As an example, Figures 6A, 6D and 6F presents the BC, DC and AC enhancement maps,

225
respectively. Statistics show that enhancement during the DC trip was 30 ppm, which is 35 and 20 ppm lower than that before and after this trip. The spatial distribution patterns of enhancement were similar to the CO 2 concentration maps, in which enhancements during rush hours were much higher for all trips. Compared to the CO 2 concentration maps, the enhancements showed important information in Figure 6B and 6C. The averaged CO 2 concentrations in these two trips were similar to those during BC and AC ( Figure 5B and 5C); however, the enhancements that extracted traffic emission signals 230 from the background, with averages of 33 and 16 ppm ( Figure 6B and C), were much lower than those of BC and DC. The enhancement maps also showed more useful information than the CO 2 concentration maps. For example, although the CO 2 concentration throughout the northern half of the 2 nd Ring Roads was high (550~600 ppm) ( Figure 5A), the enhancement extracted more specific variations induced by traffic emissions in the northwest ( Figure 6A). Generally, the statistical enhancement the average of the four DC trips was 24 ppm, which was 41 and 26 ppm lower than that of the BC and AC trips,

235
respectively. Because of the IAP tower Picarro calibration and measurement procedure (see Method Section), there were regular data gaps for the trip on 20 th February 2019 ( Figure 6A).

Diurnal variation analysis:
https://doi.org/10.5194/acp-2020-966 Preprint. 245 Figure 7 shows the diurnal variation from the IAP tower CO 2 concentrations, on-road CO 2 concentrations, enhancements and trajectories for all trips. In Figure 7A, the IAP tower CO 2 concentrations were relatively stable, and showed the difference between trips. The CO 2 concentrations during the two trips during COVID-19 (13 th and 20 th February 2020) were ~30 ppm higher than those during the BC and AC trips. However, the CO 2 concentrations during the other two trips (21 st and 22 nd February 2020) were ~20 ppm lower than those during the BC and AC trips. These "baseline" CO2 concentration 250 fluctuations make the on-road observations not comparable directly. In Figure 8B, the CO 2 concentrations show a "double-peak" pattern within the morning (7:00-9:00) and evening rush hours (17:00-20:00). During the rush hours, the CO 2 concentrations ranged from 500 to 600 ppm, which were approximately 100 ppm higher than the concentrations during working hours (9:00-17:00). The comparison of BC and AC indicates that the CO 2 concentrations measured on 13 th and 20 th February 2020 did not significantly decrease during 12:00-17:00. However, the CO 2 concentrations measured on 21 st and 255 22 nd February 2020 were much lower (~50 ppm) than those measured during the BC and AC trips. This difference is consistent with the spatial distribution mentioned before and is most likely due to the CO 2 concentration background fluctuations.
In Panel C, all DC CO 2 enhancements were generally lower than those of BC and AC. However, we still found very low 260 enhancements values for BC and AC; for example, AC enhancement at approximately 12:00 and 16:00 was almost the same as that of DC. With the help of trip routes (Panel D), we found that during that period, the on-road observation vehicle was https://doi.org/10.5194/acp-2020-966 Preprint. Discussion started: 12 October 2020 c Author(s) 2020. CC BY 4.0 License.
not driving on the main ring roads. Another example is BC at approximately 18:00, which indicates that the enhancement decreased in a stepwise manner, also because the vehicle drove on other roads (Panel D). In Panel C, all DC CO 2 enhancements were generally lower than those of BC and AC, and the statistics for different time periods are also listed in 265 Table 3. Table 3. Statistical analysis of CO 2 enhancement for six trips (ppm)

270
The average of CO 2 enhancement for the whole BC trip was 65 ppm, and the average for the evening rush hour (100 ppm) was two times that of the working hours (54 ppm). This result implies that the increase in vehicle volume in the evening rush hours leads to large traffic emissions and an increase in the on-road CO 2 concentration. For DC, all trips covered the working hours, with a low enhancement of approximately 20 ppm. There was not obvious difference between weekdays and weekends during this period, which indicated that there was no "week effect". The reason may because the government 275 encouraged people to work remotely at home. Therefore, even on weekdays, the commute was small. Among these four trips, two (13 th and 20 th February 2020) covered the evening rush hours with high averaged enhancements of 55 and 50 ppm.
Therefore, the total enhancement averages of these two trips were higher than those of the other two trips, which covered only working hours. For AC, on 9 th May 2020, although it was a Saturday, many residents chose to go out of town for weekends. The morning rush hours still existed, with a high enhancement of 80 ppm, and then during the working hours, the 280 enhancement decreased to 46 ppm.
The comparison of trips showed that the averaged CO 2 enhancement from 4 whole DC trips was 41 and 26 ppm lower than that from the BC and AC trips, respectively. Compared to the BC trip, the averaged AC enhancement was 15 ppm lower.
This difference may be caused by two factors: 1) "weekly effects", as previously mentioned; a previous study also suggested 285 that, compared to weekdays, the average daily traffic CO 2 emissions during weekends in the north part of the fifth Ring Road  ; 2) until 9 th May 2020, although there were approximately 30 days without increased COVID-19 cases in Beijing, the city was still under Level-2 response control; social life was recovering, but had not yet completely recovered.

290
Analysis of CO 2 enhancement on independent time periods and roads: According to the previous analysis, we found that enhancement exhibited a strong correlation with the time (rush or working hours) and road types. Therefore, we statistically analyze CO 2 enhancements according to road types and time periods, as shown in Figure 8. In Figure 8A, on 13 th and 20 th February 2020, the CO 2 concentrations on the other, 2 nd , 4 th Ring Roads 295 and all roads were at the same levels as those during the BC and AC trips. However, in Figure 8B, the enhancement showed that the four trips during COVID-19 were generally lower than those during AC and BC for all road types. Although on the 2 nd Ring Road, the DC trips on 13 th and 21 st February 2020 were almost the same as the BC and AC trips, the DC trips were during rush hours, whereas the AC and BC trips were during working hours. Some very high deviations also occurred (rush hours on the other roads, 2 nd and 5 th Ring Roads), which indicates the dispersion of CO 2 enhancement. The reason for this 300 difference is that we classified all roads excluding the ring roads as other roads, which may include arterial and residential roads, so the different road types may increase the deviation. For the 2 nd and 5 th Ring roads, high deviation occurred because during rush hour, traffic flow and transportation vary greatly and result in drastic changes to CO 2 enhancement, which also causes much higher deviations. We also calculated specific statistics, which are listed in Table 4.

Correlation analysis with traffic flow：
It was difficult to obtain a quantitative evaluation of the influence of COVID-19 on CO 2 emissions from traffic during our study period because there were limited data. In this study, we found one trip enhancement during DC (on 21 st February 2020, with the most similar weather and route as trips during BC and AC) was 30 ppm. The enhancement accounted for 46% 315 of that during BC (65 ppm), and the enhancement during AC (50 ppm) account 77% of that during BC. Here, we adopted four data and methods to explain our hypothesis that the decrease in traffic volume led to a reduction in on-road CO 2 emissions and concentration during COVID-19 control. First, according to "analysis of road traffic operation in Beijing during COVID-19 in 2020" published by the Beijing Transport Institute, during the first 8 weeks (from 1 st February to 31 st March, DC period in this study), Beijing ground transportation index (calculated based on ratio of congestion road length 320 and whole road length) decreased by 53% compared to normal days; whereas, during 1 st April to 31 st May, the index recovered to 92% (Zhang, 2020). The index implied that traffic flow of DC is dramatically decreased compared to that of BC, and AC recovered almost but not completely. This index variation is consistent with our observations results. Second, two remote sensing images from similar dates were adopted (Figure 9). According to statistics and estimations based on coverage area, we found that the BC traffic flows on the main roads of the 4 th and 3  information, road type and average speed for one-hour data from the Autonavi Open Platform (https://lbs.amap.com/). The data, although with low temporal and spatial resolution, could be used to show traffic conditions on roads, and then indicate the on-road traffic flow and emissions ( Figure 10). Fourth, the vehicle speed maps of six trips were also plotted ( Figure 11).
Overall, these maps reflect the spatial patterns of road traffic conditions during the surveys and could also reflect the 335 specifics on a single road. However, these maps are subject to subjective speed variations caused by drivers, such as when facing traffic lights.

355
Uncertainty analysis: The uncertainty of this research mainly existed in the following terms: (1) The IAP tower CO 2 concentration was used as the background in Beijing.

360
In this study, IAP tower data were adopted as the urban background CO 2 concentration in Beijing. However, IAP tower data were collected from different levels, as described in the method section. Generally, high-level data would have a large footprint and cover large regions. For example, Cheng et al. (Cheng et al., 2018) showed that 280 m height-level CO 2 data have an averaged fetch of ~17 km, which may cover a major part of the city; 80 m height-level data have an averaged fetch of ~8 km; and 8 m height-level data may have aa averaged fetch of only ~230 m, and the surface level (2 m) may be smaller.

365
Due to the data availability and comparison consistency, we chose the lower and surface level data. According to Cheng et al.(Cheng et al., 2018), the CO 2 concentration at the 80 m height level is ~15 ppm higher than that at 8 m. Therefore, adding this difference between the lower level and surface level, the BC enhancement would increase (~15 ppm), which means that the DC enhancement would be even lower (~56 ppm) than the BC enhancement. This result is consistent with our hypothesis.

370
(2) When data were collected, especially when switching between lower and upper levels, a large amount of data was lost.
However, because the data gaps were evenly distributed and the IAP tower CO 2 concentrations were relatively stable, we assumed that it would not affect the final statistical results.
(3) In this study, our on-road observations did not have a fixed route or beginning/ending time, which means that the observations on different dates represented different roads. Therefore, we analyzed a wide time range of observations (rush 375 hours, working hours or whole day), which may also cause uncertainty.

Conclusion
The CO 2 emission reduction caused by COVID-19 is an opportunity to test our ability to collect CO 2 observations in urban regions. In this study, aiming at traffic emissions, which the potentially represents the largest reduction source in urban areas due to COVID-19, we conducted six on-road observations in Beijing, China. The results showed that on-road CO 2 380 concentrations were strongly affected by traffic emissions and weather. However, the enhancement which was the difference of on-road CO 2 concentration and the city "background", reduced the impact of background CO 2 fluctuations. The results showed that during COVID-19, the total average CO 2 enhancements of the four trips were 41 ppm and 26 ppm lower than those before and after, respectively. Detailed analysis showed that this reduction commonly existed on all road types during the same time period (rush hours/working hours). During COVID-19, there was no significant difference between weekdays 385 and weekends. During rush hours, the enhancements were much higher than those during working hours, and compared with BC, the DC enhancements reduction during rush hours was most obvious. Our findings, which show a clear decrease during DC compared with those during BC and AC, are consistent with the COVID-19 control, which may be direct evidence of reductions in CO 2 concentrations and carbon emissions. On-road CO 2 observations are an effective way to understand and analyze urban carbon CO 2 concentration distribution and variation and should be regularly and more frequently conducted in 390 future work. With the development and successful application of miniaturized and low-cost CO 2 monitoring instruments used in this study (Khan et al., 2012;Shusterman et al., 2016;Martin et al., 2017;Mueller et al., 2020;Bao et al., 2020), these instruments will greatly help collect on-road observations and even high-density network observations and play a key role in future urban carbon observations.