key: cord-0812107-e2i427fh authors: Cai, Wan-Jin; Wang, Hong-Wei; Wu, Cui-Lin; Lu, Kai-Fa; Peng, Zhong-Ren; He, Hong-Di title: Characterizing the interruption-recovery patterns of urban air pollution under the COVID-19 lockdown in China date: 2021-08-11 journal: Build Environ DOI: 10.1016/j.buildenv.2021.108231 sha: a10d3f0b3ab2b5e8b623f4f8b5914049fbfd0991 doc_id: 812107 cord_uid: e2i427fh The COVID-19 pandemic provides an opportunity to study the effects of urban lockdown policies on the variation in pollutant concentrations and to characterize the recovery patterns of urban air pollution under the interruption of COVID-19 lockdown policies. In this paper, interruption-recovery models and regression discontinuity design were developed to characterize air pollution interruption-recovery patterns and analyze environmental impacts of the COVID-19 lockdown, using air pollution data from four Chinese metropolises (i.e., Shanghai, Wuhan, Tianjin, and Guangzhou). The results revealed the air pollutant interruption-recovery curve represented by the three lockdown response periods (Level I, Level II and Level III) during COVID-19. The curve decreased during Level I (A 25.3%–48.8% drop in the concentration of NO(2) has been observed in the four metropolises compared with the same period in 2018–2019.), then recovered around reopening, but decreased again during Level III. Moreover, the interruption-recovery curve of the year-on-year air pollution difference suggests a process of first decreasing during Level I and gradually recovering to a new equilibrium during Level III (e.g., the unit cumulative difference of NO(2) mass concentrations in Shanghai was 21.7, 22.5, 11.3 (μg/m(3)) during Level I, II, and III and other metropolises shared similar results). Our findings reveal general trends in the air quality externality of different lockdown policies, hence could provide valuable insights into air pollutant interruption-recovery patterns and clear scientific guides for policymakers to estimate the effect of different lockdown policies on urban air quality. The outbreak of the Corona Virus Disease 2019 (COVID-19) pandemic has detrimentally 24 impacted urban development and human health. Confirmed cases of the new coronavirus reached 25 nearly 79 million worldwide, with 1.7 million deaths, by the end of 2020 (WHO, 2020). To curb 26 the spread of COVID-19, lockdown policies varying in degree were implemented across cities in 27 China and urban traffic and industrial production were strictly limited. Nonetheless, this response 28 also provides an opportunity to characterize the urban air pollution recovery pattern under the 29 interruption of pandemic lockdown policies. Additionally, the response can isolate the major 30 pollution sources, which makes it the best natural experiment for assessing the impact of 31 lockdown policies. The concept of 'interruption-recovery' is introduced here that has been 32 commonly used to describe the process of a system to revert, or 'bounce back' to a normal level 33 after its original state has been affected by a disruptive event. To our knowledge, existing 34 research rarely focuses on the interruption-recovery pattern of urban air pollution during COVID-35 19. A good characterization of such interruption-recovery patterns not only can often assist 36 policymakers in better understanding the dynamics of urban air pollution and has potential for 37 analyze the differential impact of various lockdown policies on urban air quality during the 42 COVID-19 pandemic. 43 As reported in previous studies, the emergence and fast spread of COVID-19 has 44 significantly improved urban air quality compared with the same period in years before, and the 45 concentrations of regulatory air pollutants (for example NO x , SO 2 , PM 2.5 , PM 10 , CO) in urban 46 areas have also declined substantially Lal et al., 2020; Xu 47 et al., 2020) , particularly those of PM 10 and NO 2 due to sharp declines in traffic volume (Bao & 48 Zhang, 2020). The lowered PM 2.5 , SO 2 , and CO concentrations were mainly caused by the 49 shutdown of the industrial sector (Wang & Su, 2020) . By However, studies conducted to clearly reveal the interruption-recovery pattern of urban air 70 pollution for the entire period of COVID-19 lockdown policies are limited in number (He, Pan, & 71 Tanaka, 2020; . In this study, the air pollution interruption-recovery pattern has 72 been revealed and can provide clear scientific guides for policymakers to estimate the effect of 73 different lockdown policies on urban air quality. 74 In China, the pandemic response measures span different urban lockdown policies at Level 75 I, Level II, and Level III. Fig. 1 To quantify the variation in the air quality at different response stages during the pandemic, 169 we used RDD to quantify the environmental effects of the lockdown policies. The hourly 170 pollutant data were used in the RDD model to further analyze the effectiveness of different 171 lockdown policies on improving urban air quality. The basic idea of this method is that a 172 continuous variable X is developed to determine the probability of an individual variable Y 173 receiving a policy intervention on both sides of a critical point, the latte usually called a 'cutoff'. 174 An RDD model focuses on discontinuity at this cutoff, which could be discerned by a 'jump' 175 between its two sides; if this discontinuity is tested and found to be statistically significant, the 176 policy intervention can be inferred to have a causal effect upon that outcome (result). In this 177 study, the individual variable Y is the urban air pollution and the continuous variable X is the 178 timeline of the response lockdown policies which corresponds to the policy intervention. The term, also the auto-correlated term on day t. 208 After developing the RDD model, a robustness test is needed to verify the model's results. 210 As revealed by previous studies, a robustness test could verify the causality among factors 211 derived from RDD and the validity of the RDD model by considering the support obtainable from 212 these three tests (Bloom, 2012) : 213 (1) Covariate continuity test: This is also called the pseudo-outcome test. Specifically, 214 covariates (i.e., meteorological factors) are used as pseudo-outcomes to test whether the 215 corresponding RDD estimates are significant. If the estimates are indeed significant, the 216 covariates violate the continuity assumption. 217 (2) Test for continuity of reference variable distribution: If the reference variable conforms 218 to a continuous distribution, an individual variable cannot accurately manipulate the reference 219 variable at the cutoff. 220 (3) Pseudo-cutoff point test: In other positions of the reference variable (e.g., the midpoints 221 on the left and right side of the cutoff, as pseudo-cutoffs), the same method is adopted to 222 calculate the RDD estimates. If these pseudo-cutoffs are found significant, the RDD model is not 223 correct, that is under influence of other observational factors; hence, causal effects are mainly 224 driven by other mixed jumps (albeit unobserved) rather than an intervention's influence per se 225 (Hausman & Rapson, 2018) . 226 In this paper, we defined the cumulative difference of pollutant mass concentrations between 228 the COVID-19 period in 2020 and the same period in the prior years as Loss of Performance 229 (LoP), as shown in Equation (2): 230 where, pmc tp is pollutant mass concentration of NO 2 (µg/m 3 ), PM 2.5 (µg/m 3 ), PM 10 (µg/m 3 ), SO 2 232 (µg/m 3 ), O 3 (µg/m 3 ), or CO (mg/m 3 ), on day t during the pandemic in 2020; pmc tr is the 233 pollutant mass concentration on day t during the same time in the prior years; i is a given 234 response stage; t ib is the start date of response stage i, and t ie is the end date of response stage i. 235 The basic urban air pollution interruption-recovery pattern of Fig. 2 294 As Fig. 6 shows, although being two typical pollutants, NO 2 and PM 2.5 had different trends 295 during the pandemic. However, a typical loss and recovery process was observed for both NO 2 296 and PM 2.5 , consisting of an evident decrease at the cutoff of Level I and a notable recovery 297 around the cutoff of Level II and Reopen (i.e., reopening). However, these results are based on 298 observation data and the actual inflection point was not yet confirmed (i.e., whether it occurred at 299 the cutoff of Level II or Reopen). 300 In summary, the general trends of air pollutants except for O 3 all showed a gradual decrease, 301 but this did not conform to the basic interruption-recovery pattern depicted in Fig. 2(a) . The 302 seasonal factors are one of the main reasons for this phenomenon, which would be discussed in 303 detail in Section 3.2. Nevertheless, we are able to observe the process of loss and recovery, when 304 going from Level I to Level II. 305 To uncover the predominant interruption-recovery pattern of urban air pollution, the RDD 308 Table 4 350 and Table 5, the lockdown policies at Level II and Level III had little effect on air pollutant 351 concentrations. These results indicate the Level II and Level III lockdown policies hardly brought 352 any improvement to urban air quality, and the inflection point between their respective loss and 353 recovery was not the time cutoff of Level II. 354 Table 6 The three air pollutants (i.e., NO 2 , PM 2.5 , and O 3 ) were selected for further study, due to the 387 typical and differing characteristics in the variation of concentrations of these three air pollutants. 388 Several meaning findings are summarized as follow and presented in Fig. 7 (2) and Equation (3). The results for Unit LoP and Total LoP of six regulatory air pollutants 416 during peak and non-peak periods across the four megacities are presented in Table 7 . 417 In Table 7 , the Unit LoP of air pollutants except for O 3 showed a decreasing trend with the 420 changed lockdown policies against the COVID-19. The difference between the general trend in 421 the variation of a given pollutant's concentration during COVID-19 and its original equilibrium 422 was gradually reduced. This result also reveals the interruption-recovery pattern of the difference 423 in pollutant concentrations during the same period of 2020 vs. 2018-2019, resembling Fig. 2(c) , 424 in that the difference of pollutant concentrations first showed a significant decline at the cutoff of 425 Level I but then gradually recovered to a new equilibrium. Additionally, for NO 2 and O 3 , the 426 differences in their Unit LoP between peak hours and non-peak hours were notably larger than 427 those of other pollutants. These results indicate that NO 2 and O 3 concentration levels are greatly 428 affected by traffic activities . Furthermore, the effects of 429 different COVID-19 lockdown policies on urban air quality during peak hours were more 430 pronounced than those during the non-peak hours. To better understand the dynamics of pollutant concentrations during the lockdown period in 432 2020 vis-à-vis the same period in 2018-2019, the corresponding percentage changes of pollutant 433 concentrations were also calculated and compared. The results are presented in Fig. 9 and Table 8 . In Fig. 9 , the curve of percentage changes in the pollutant concentrations showed a similar 440 trend to the basic interruption-recovery pattern corresponding to Fig. 2(a) I. The main difference in the interruption-recovery pattern as revealed in Fig. 2(b) and Fig. 2(c) is 464 that the latter's pattern entails a gradual process of recovery. Additionally, that no inflection point 465 was found between the recovery process and the new equilibrium indicates the NO 2 466 concentrations at either Level II or Level III exhibited less significant variation. Therefore, the 467 interruption-recovery pattern of the year-on-year air pollution difference presented a similar 468 pattern as the basic loss-recovery process but it also harbored a gradual process of recovery. 469 initially decreased significantly at the cutoff of Level I and then recovered to a new equilibrium 502 state. For example, in Shanghai, the NO 2 concentration decreased by 11.47%, 33.68%, 17.99%, -503 6.84% during Pre-lockdown, Level I, Level II, Level III respectively, which further confirms the 504 interruption-recovery curve of air pollution difference. 505 (5) The two different air pollution interruption-recovery patterns highlighted a process of 506 loss and recovery between Level I and Reopen. However, the major difference between the two 507 interruption-recovery patterns was that the recovery pattern of urban air pollution under the 508 interruption of COVID-19 in 2020 decreased again after the recovery process, showing clear 509 cutoffs (i.e., Level I and the start date of Reopen). Further, the recovery process of air pollution 510 difference between the lockdown period in 2020 and the same time in 2018-2019 presented a 511 gradual trend, with no clear inflection points found in the interruption-recovery curve. 512 The main novelty and contributions of this research are as follows: 513 (1) We found and explored the interruption-recovery patterns of urban air pollution during 514 the COVID-19 lockdown period ( Fig. 2(b) ), and the interruption-recovery patterns of air 515 pollution difference between 2020 and 2018-2019 (Fig. 2(c) ). The pattern can assist 516 policymakers in better understanding the dynamics of urban air pollution and optimizing the 517 decision-making of lockdown policies. 518 (2) This study quantified the effects of different lockdown policies against COVID-19 upon 519 urban air quality. The results can assist policymakers in anticipating the potential outcomes of 520 differing lockdown policies for alleviating urban air pollution. 521 (3) The RDD model was developed and fitted to the empirical data in this paper. These 522 results demonstrated that the RDD method was more effective and accurate in analyzing the 523 effects of time-based policy under the impact of confounding short-term factors (i.e., 524 meteorological factors). 525 (4) The concentrations of pollutants significantly decreased but were not fully eliminated, 526 even under the most severe lockdown policies. Therefore, in addition to a tailored lockdown 527 policy, some parallel strategies-e.g., the adjustment of industry structures and the adoption of 528 clean energy-should also be employed to further improve urban air quality. 529 One limitation of this paper is that our findings focused on metropolises in China, while 530 other global cities may present different interruption-recovery patterns of urban air pollution. 531 Future studies could consider other cities or regions worldwide, to further explore the 532 interruption-recovery patterns of urban air pollution and elucidate inherent mechanisms of the 533 variation in urban air quality during the COVID-19 pandemic. 534 The authors confirm their respective contribution to the paper as follows: study conception 536 and design: Wan-Jin Cai; data collection: Wan-Jin Cai, Cui-Lin Wu; analysis and interpretation 537 of results: Wan-Jin Cai; draft manuscript preparation: Wan-Jin Cai; Hong-Wei Wang, Kai-Fa Lu, 538 Zhong-Ren Peng, and Hong-Di He co-wrote the paper. All the authors reviewed the results and 539 approved the final version of the manuscript. 540 The authors declare that they have no known competing financial interests or personal 542 relationships that could have appeared to influence the work reported in this paper. 543 Air quality status during 2020 Malaysia Movement Control Order (MCO) due to 2019 novel 550 coronavirus (2019-nCoV) pandemic A new look at the statistical model identification Model Selection and Multi-model Inference Does lockdown reduce air pollution? Evidence from 44 cities in northern China Regression discontinuity design with many thresholds Modern regression discontinuity analysis Covid-19 lockdown Regression discontinuity: review with extensions Personal 572 exposure monitoring of PM2.5 among US diplomats in Kathmandu during the COVID-19 lockdown The Status of Air Quality in the United 575 States During the COVID-19 Pandemic: A Remote Sensing Perspective Recovery time and 578 state change of terrestrial carbon cycle after disturbance Why high-order polynomials should not be used in regression discontinuity 581 designs Temporary driving restrictions, air pollution, and contemporaneous health: 584 Evidence from China The short-term impacts of COVID-19 lockdown on urban air 587 pollution in China Regression Discontinuity in Time: Considerations for Empirical 589 Particulate matter exposure at a densely populated urban traffic intersection and 592 crosswalk Generic metrics and quantitative approaches for system resilience as A theory of pulse dynamics and disturbance in ecology Modeling the impact of 605 COVID-19 on air quality in southern California: implications for future control policies. Atmospheric 606 Chemistry and Physics Regression discontinuity designs in economics Review on difference in differences Air quality changes during the 616 COVID-19 lockdown over the Yangtze River Delta Region: An insight into the impact of human 617 activity pattern changes on air pollution variation Spatiotemporal impacts 620 of COVID-19 on air pollution in California Characterizing temporal and vertical 623 distribution patterns of traffic-emitted pollutants near an elevated expressway in urban residential 624 areas Effect of lockdown amid COVID-19 pandemic on air quality of the A spatial and temporal 633 analysis of forest dynamics using Landsat time-series Air pollution impacts from COVID-19 pandemic control strategies in 636 Impacts of the COVID-19 lockdown on air pollution at regional and urban background sites in 639 northern Italy Lessons from the COVID-19 air pollution decrease in Spain: Now what? Atmospheric Chemistry and Physics: From Air Pollution to Climate 645 COVID-19 outbreak, lockdown, and air 647 quality: fresh insights from New York City Amplified 650 ozone pollution in cities during the COVID-19 lockdown Ground-level ozone over time: An observation-based global overview. Current Opinion in 653 Regression-discontinuity analysis -an alternative to the ex-post-655 facto experiment Air pollution and COVID-19 lockdown in a large South American city Distrubance-660 recovery experiments to assess resilience of ecosystems along a stress gradient Identifying Post-Fire Recovery Trajectories and 662 Driving Factors Using Landsat Time Series in Fire-Prone Mediterranean Pine Forests Regional prediction of ground-665 level ozone using a hybrid sequence-to-sequence deep learning approach Evaluation and 668 prediction of transportation resilience under extreme weather events: A diffusion graph convolutional approach Heterogeneous effects of COVID-19 lockdown 672 measures on air quality in Northern China Severe air pollution events not avoided by reduced 675 anthropogenic activities during COVID-19 outbreak. Resources, Conservation and Recycling A preliminary assessment of the impact of COVID-19 on environment ? A case study 678 of China Responses of 680 decline in air pollution and recovery associated with COVID-19 lockdown in the Pearl River Delta Changes in air quality related to the 683 control of coronavirus in China: Implications for traffic and industrial emissions Association between NO2 concentrations and spatial configuration: a study of the impacts of COVID-687 19 lockdowns in 54 US cities Building and Environment Impacts of the COVID-19 692 responses on traffic-related air pollution in a Northwestern US city Possible environmental 695 effects on the spread of COVID-19 in China Identification of the disturbance 698 and trajectory types in mining areas using multitemporal remote sensing images Some pseudo-cutoffs may also exist when performing an RDD analysis. These pseudo-720 cutoffs usually show a significant jump in the outcome variables with the continuous variable X 721 changing and could lead to an the pseudo-cutoff test were both selected. Specifically, the pollutants' data from Wuhan is selected in this experiment and the NO 2 concentrations here exhibit the most significant 729 change in the pseudo-cutoff test. As Fig. A.1 shows, no significant jumps in the outcome 730 variables were observed at all three pseudo-cutoffs A regression discontinuity analysis should include the robustness tests of the model results 703 (Bloom, 2012) . Only with such a robustness test are the conclusions derived from the RDD 704 model meaningful. This paper used the covariate (i.e., meteorological factors) as the outcome 705 variable to perform the RDD analysis and determine whether the covariate has an actual cutoff (a 706 'jump') at the original cutoff. If there found to be a jump at the cutoff, the results of the RDD 707 analysis are interfered and hence are meaningless. The meteorological factors (temperature, wind 708 speed, and wind direction) were used as pseudo-result variables to perform the RDD analysis. 709The order selection was determined using the same method as above, this being set to 2. This 710 resulted in a coefficient of cutoff d that equaled 0.04, which is very small. The p-value was much 711