key: cord-0935367-dtqtd8fp authors: Hörmann, Siegfried; Jammoul, Fatima; Kuenzer, Thomas; Stadlober, Ernst title: Separating the impact of gradual lockdown measures on air pollutants from seasonal variability date: 2020-11-02 journal: Atmos Pollut Res DOI: 10.1016/j.apr.2020.10.011 sha: b6bda9b03293f9ce7a274d4ab6402dfa5be058c6 doc_id: 935367 cord_uid: dtqtd8fp Analysis of near-surface measurements at several measuring points in Graz, Austria, reveals the impact of restrictive measures during the COVID-19 pandemic on the emission of atmospheric pollutants. We quantify the effects at traffic hotspots, industrial and residential areas. Using historical data collected over several years, we are able to account for meteorological and seasonal confounders. Our analysis is based on daily means as well as intraday pollution level curves. Nitrogen dioxide (NO(2)) has decreased drastically while the levels of particulate matter PM(10) and carbon monoxide (CO) mostly exhibit little change. Traffic data shows that the decrease in traffic frequency is parallel to the decline in the levels of NO(2) and NO. The coronavirus disease COVID-19 has spread around the world since the end of 2019. 22 In an effort to halt this infectious disease, many countries imposed restrictions on public In this paper, we focus on the impact of the lockdown in Graz, the second largest city 35 in Austria, where pollution data has been diligently recorded for a long time. We can 36 build upon a long expertise in pollution level monitoring and forecasting. Our aim is to 37 investigate how the economic shutdown in Graz affected the levels of local near-surface 38 pollution with nitrogen dioxide (NO 2 ), nitric oxide (NO), particulate matter (PM 10 ) and 39 carbon monoxide (CO) across different types of stations (traffic, industrial, residential). 40 Similar studies have been carried out in different geographical regions. Bauwens et al. 41 [2020] show that tropospheric NO 2 levels fell by 40% over Chinese cities and by 20-38% 42 in Europe and North America. Kanniah et al. [2020] found a drop of around 30% over 43 urban areas in Malaysia. The effect was even stronger for near-surface levels of NO 2 . 44 Lee et al. [2020] found reductions of 42% in urban areas in the UK, Baldasano [2020] 45 found reductions of 50-62% in Spain, Kanniah et al. [2020] found reductions of 63-64% 46 in Malaysia. In contrast, the reduction of particulate matter PM 10 was less pronounced 47 and amounted to 26-31%. [Kanniah et al., 2020] . 48 When pollution levels are strongly impacted by a seasonal trend or meteorological 49 variables (as is the case in Graz), we need to account for these effects and can assess 50 2 J o u r n a l P r e -p r o o f changes in the concentration only after eliminating these influences by means of a sta-51 tistical model of some kind. The above mentioned papers do not take meteorological 52 effects into account. In contrast, Bao and Zhang [2020] analysed data of near-surface 53 measurements from 44 Chinese cities during the lockdown period, taking local weather 54 conditions into account. They found that under lockdown, the levels of NO 2 , PM 10 , and 55 CO were reduced by 25%, 14%, and 5%, respectively. Using the CHIMERE multi-scale 56 model, Menut et al. [2020] carried out simulations for the month of March 2020. A 57 reference scenario with emissions as usual was compared to a scenario that took into ac-58 count the reduction of pollutant emissions due to lockdown measures. This reduction in 59 emissions was estimated for different sectors in each country. For example, they found a 60 reduction in surface concentration of 37% for NO 2 and 10% for particulate matter PM 2.5 61 in Austria. and trend before and after the lockdown. As a result, they found a significant reduction 67 only for NO 2 at one of the above mentioned stations, which is located in a heavy traffic 68 zone. A survey by Myllyvirta and Thieriot [2020] assessed the weather-corrected reduction 70 of air pollutants in Europe during the time of the lockdown and found population-71 weighted average decreases of 17-58% for NO 2 and up to 55% for PM 10 , depending on the 72 country. Two countries even showed a slight increase in PM 10 levels. Ordóñez et al. [2020] 73 investigated 1-h daily maximum NO 2 levels as well as maximum daily 8-h average ozone 74 levels (O 3 ). Taking meteorological effects into account, they fit generalized additive 75 models for 1331 rural and urban sites in Europe to obtain a model for the "business- Our approach is a statistical one, and we intend to quantify the impact of the lock-89 down based on measured pollution concentrations. We use pollutant measurements and 90 meteorological data from several ground-based measuring sites within the city borders 91 for our analysis. While Bao and Zhang [2020] restricted their analysis to data sampled 92 in 2020, we include historic data going back to 2015 for building our statistical models. A unique feature of our data is their 30 minutes resolution, which means that we have 94 access to 48 half-hourly mean values for each day. This allows us to capture intraday 95 effects. The dataset is part of a long-going campaign to measure atmospheric pollution 96 in Graz in a major effort to meet EU pollution standards. Graz has struggled with 97 increased PM 10 levels for decades, but has seen the pollutant levels steadily decreasing In addition, we also compare our findings on the pollution levels with the traffic count 102 data that is available in Graz for the lockdown period. The data we consider is for the pollutants NO 2 , NO, CO and PM 10 . All raw mea- . 157 We will hence empirically explore these residuals in order to quantify a potential effect 158 and then test for significant changes, by contrasting Phases 1-3 to the pre-lockdown 159 Phase 0, which serves as our control group. We process the meteorological data before using them as input for the regression model. The air temperature (temp), relative humidity (hum) and wind speed (wisp) are summa- Lastly, we also included control variables that are independent of the meteorology. Some details on average pollution levels at the different sites can be seen in Table 1 . We analyse the data on two levels. In a first step, we investigate the effect of the 215 lockdown measures on the daily mean values of the pollutants. Secondly, we take a 216 closer look to determine the intraday effects. Following on from this, we fit two regression models for each station and pollutant: one multiple linear regression model for the daily means and one scalar-on-function 219 regression model for the daily curves (i.e. 48 half-hourly measurements per day). In 220 both regression approaches, we use the variables mentioned in Section 2.4 as predictors. In addition to the meteorological variables from the same day, we also include lag-1 222 differences of the meteorological variables. Furthermore, we use two-way interactions 223 and quadratic effects. To account for skewness, heavy tails and heterogeneous variances, it is common to 225 apply a log-transform to pollutants. We use a logarithmic transformation for NO and 226 CO. However, we noted that a square-root transformation is better suited for NO 2 and where Y t is the pollutant concentration, f is a transformation function, X t,i are the 235 predictors and ε t is a random noise that represents the unexplained fluctuations and the 236 deviation from the linear model. The models of the daily means are multiple linear regression models (lm) that we fit The models that we fitted in Section 2.5 are used to predict the pollution levels for A summary of the prediction errors in 2020 on the transformed scale along with 303 a standard deviation and color code for the p-values can be found in Table 2 . The 304 statistical testing procedure described in Section 2.6 reveals that there has been a highly 305 significant difference between Phase 0 and Phase 1 in all five sites for the pollutants NO 2 306 and NO. For NO 2 this deviation is still present in Phase 2 for all stations, while for NO 307 the site IN2 no longer shows a significant deviation from Phase 0. In Phase 3, we see a 308 mixed picture, but with a trend back to normal levels. The pollutants CO and PM 10 show a different structure. While for CO we can detect The typical intraday pattern is fundamentally different for working days and Satur-325 days, Sundays and holidays. We will therefore restrict this section to analysis of working 326 days. When looking at these plots, it is important to take the scale into account. The strongest reduction at all sites is typically seen around 9pm. The typical daily pattern of NO is dominated by a very high peak in the morning. Depending on the season, a secondary, very minor peak emerges in the evening. Ac- Here the numbers in brackets are the corresponding p-values. 360 We conclude that the intraday analysis reveals further deviations in the shape of the 361 pollution curves, which are not reflected in a daily average. We would point out here that we have only limited traffic data from before the shut-378 down, which is why we did not include this as a separate variable in our regression 379 models. 5 Summary and discussion 381 We have analyzed the effect of the lockdown measures on air pollution levels in the city 382 of Graz, Austria. Using data from the preceding five years, we were able to provide 383 well-founded models that account for meteorological influences on the pollution. The 384 deviation from the meteorological models showed an abrupt change of NO 2 and NO, 385 while we observed no significant effect on CO and no effect on PM 10 that was consistent 386 across all stations. In particular, we found daily mean NO 2 levels reduced by 35-41%, COVID-19 lockdown effects on air quality by NO2 in the cities of 433 Does lockdown reduce air pollution? Evidence from 44 cities 436 in northern China Impact of coronavirus outbreak on NO2 pollution assessed using TROPOMI and 441 OMI observations The effect of corona virus lockdown on air pollution: Evidence from the 444 city of Brescia in Lombardia region (Italy) Applied Multivariate Statistical Analysis Principal Component Analysis COVID-19's 450 impact on the atmospheric environment in the Southeast Asia region UK surface 456 NO2 levels dropped by 42 % during the COVID-19 lockdown: impact on surface 457 O3. Atmospheric Chemistry and Physics Discussions Impact of 460 lockdown measures to combat Covid-19 on air quality over Western Europe 11,000 air pollution-related deaths avoided in Europe as 464 coal, oil consumption plummet Early spring near-surface 467 ozone in europe during the COVID-19 shutdown: Meteorological effects outweigh 468 emission changes Meteorology-normalized impact of COVID-19 lockdown upon NO2 472 pollution in Spain R: A Language and Environment for Statistical Computing. R Foundation 475 for Statistical Computing Methods for scalar-on-478 function regression Saisonbereinigte PM10-Mittelwerte der Winterhal-481 bjahre Quality and performance of a PM10 daily 483 forecasting model This research did not receive any specific grant from funding agencies in the public, 426 commercial, or not-for-profit sectors. The authors wish to thank the city of Graz for 427 kindly providing the traffic count data. Declaration of competing interest 429 The authors declare that they have no known competing financial interests or personal 430 relationships that could have appeared to influence the work reported in this paper.