COVID-19 induced lower-tropospheric ozone changes

The recent COVID-19 pandemic with its countermeasures, e.g. lock-downs, resulted in decreases in emissions of various trace gases. Here we investigate the changes of ozone over Europe associated with these emission reductions using a coupled global/regional chemistry climate model. We conducted and analysed a business as usual and a sensitivity (COVID19) simulation. A source apportionment (tagging) technique allows us to make a sector-wise attribution of these changes, e.g. to natural and anthropogenic sectors such as land transport. Our simulation results show a decrease of ozone of 8% over Europe in May 2020 due to the emission reductions. The simulated reductions are in line with observed changes in ground-level ozone. The source apportionment results show that this decrease is mainly due to the decreased ozone precursors from anthropogenic origin. Further, our results show that the ozone reduction is much smaller than the reduction of the total NO x emissions (around 20%), mainly caused by an increased ozone production efficiency. This means that more ozone is produced for each emitted NO x molecule. Hence, more ozone is formed from natural emissions and the ozone productivities of the remaining anthropogenic emissions increase. Our results show that politically induced emissions reductions cannot be transferred directly to ozone reductions, which needs to be considered when designing mitigation strategies.


Introduction
The COVID-19 pandemic has a strong socioeconomic impact [1]. As one consequence, in 2020 reduced carbon dioxide (CO 2 ) emissions from various sectors have been noted in many regions worldwide (e.g. [2,3]). Typically, such reductions of CO 2 emissions are expected to be related to air quality improvements through reduced co-emission of pollutants. Indeed, reductions of particulate matter and nitrogen-dioxide (NO 2 ) have been observed in northern China [4]. In the case of NO 2 a reduction has also been observed from space in various regions all over the world [5]. However, it was also noted that ozone surface levels have partly increased despite the decrease of emissions of the ozone precursor NO 2 (e.g. [4,6]). This increase is due to the complex (nonlinear) ozone chemistry, which explains that a reduction in ozone precursors can lead to increasing ozone production, if the ozone production takes place in the 'VOC-limited' regime (e.g. [1,[6][7][8][9]). The emission reduction during spring 2020 related to COVID-19 is a rare real-life experiment from a scientific viewpoint [1], similar to the eruption of Eyjafjallajökull in 2010, which halted air-traffic for a short period in the affected regions (e.g. [10,11]).
In this study we analyse the impact of strong emission reductions, observed during the first half of 2020, on the ozone chemistry by comparing the results of a business as usual (BAU) simulation and a simulation with strongly reduced emissions (COVID19). Additionally, we compare ground-level measurements during 2020 with measurements from previous years. For our simulations we employ a regional chemistry climate model (CCM) on-line nested into a global CCM, which is relaxed to operational meteorological analysis, to assess the impact of reduced emissions on boundary layer ozone in Europe. To complement previous results based on observations, which are subject to changes in both, meteorology and emissions, and hence more difficult to interpret ( [5,6]) we use the CCM in a so-called quasi-chemistry transport model (QCTM) mode to suppress feedback from chemistry on meteorology [12]. Furthermore, our model setup allows for an attribution of ozone to various emission sectors (e.g. land transport, aviation, shipping) via a tagging technique as described by [13][14][15].
This paper is structured as follows: section 2 contains the description of the atmospheric model(s) and simulation set-ups along with the emission scenarios and a short description of the measurement data employed. The results are presented in section 3. Finally, we discuss our results (section 4) and state our conclusions (section 5).

Atmospheric modelling and simulation set-up
For our analyses we employ the CCM MECO(n) (MESSy-fied ECHAM COSMO models nested ntimes) in a modified version of MESSy 2.54 ( [16,17]). Here, we use a MECO(1) setup which nests one instance of the regional CCM COSMO-CLM/MESSy [18] into the global CCM EMAC using the MESSy infrastructure ( [19,20]). COSMO-CLM is the community model of the German regional climate research community jointly further developed by the CLM-Community [21]. EMAC in turn uses the general circulation model ECHAM5 [22] as a base model. Our simulations cover the period 1 March-1 July 2020.
The global model, EMAC, was operated at a T42L90MA triangular spectral resolution, which corresponds to a quadratic Gaussian grid of approximately 2.8 • × 2.8 • (roughly 300 km) and 90 model levels in the vertical, which extend up to the middle atmosphere (∼0.01 hPa). Meteorological prognostic variables, i.e. divergence, vorticity, temperature (excluding mean temperature) and (logarithm of) surface pressure, have been 'nudged' by Newtonian relaxation to European Centre for Medium-Range Weather Forecasts (ECMWF's) operational analyses data for the simulation period with 6 h temporal resolution. The nudging coefficients are applied in a way that the large-scale synoptic patterns follow the ECMWF data, but the model can develop its own small-scale dynamics (for more details see [23]).
At the nesting boundaries, the global model passes boundary conditions with respect to dynamics and chemistry to the regional model with a high temporal resolution [17]. In this simulation, the regional COSMO refinement (nest) is centred over the North Atlantic and covers Europe, the North Atlantic and parts of North America with a resolution of approximately 50 km (see figure 1). The regional model is only forced by the boundary conditions and otherwise evolves freely.
Chemistry schemes for gas and aqueous phase chemistry are applied consistently in the global model and the regional refinement as described by [17]. For calculation of chemical kinetics, we use the MESSy submodel Module Efficiently Calculating the Chemistry of the Atmosphere (MECCA [24]). The chemical mechanism includes the chemistry of ozone, methane, and odd nitrogen. Alkynes and aromatics are not considered, but alkenes and alkanes are considered up to C 4 . The Mainz Isoprene Mechanism (MIM1 [25]) is applied for the chemistry of isoprene and some non-methane hydrocarbons. Scavenging of trace gases by clouds and precipitation is calculated by the submodel SCAV (scavenging of traces gases by clouds and precipitation [26]). Dry deposition is considered according to [27].
To avoid feedbacks of the chemistry on dynamics, the global and the regional models are operated in the so-called QCTM mode [12]. In this mode mixing ratios of greenhouse gases with respect to the calculation of radiative fluxes are prescribed from daily averaged values of a previous simulation. This previous simulation covers the same time period and uses the same set-up as the BAU simulation described below. Due to the usage of the QCTM-mode the emission sensitivities described in section 2.2 do not affect the meteorological situation. This means, that in both simulations the meteorology (i.e. wind, temperature, humidity etc.) is identical. The transport and processing of chemical constituents is, however, different due to the changed primary emissions.
Anthropogenic and natural emissions are prescribed by flux conditions at the lower boundary. In our reference or BAU simulation anthropogenic emissions are prescribed according to the EDGAR 4.3.1 emission inventory for the year 2010 [28]. From the emission sectors in EDGAR 4.3.1 we distinguish the emission sectors land transport (TRA), anthropogenic non-traffic fossil fuel use in industry and households (in the following referred to ANT emissions, ANT), and shipping. Emissions of agricultural waste burning, biomass burning and aviation are prescribed according to the RCP 8.5 emission inventory for the year 2020 [29,30].
Biogenic emissions of soil NO x and biogenic isoprene are interactively calculated by the global and the regional model according to the parameterisations of [31] and [32], respectively. NO x emissions from lightning are parameterised based on the cloudtop height [33] and scaled to global total emissions of ∼5 Tg(N)/a. The emissions of lightning-NO x are Table 1. Assumed emission reductions (in per cent) in the COVID19 sensitivity simulation compared to the BAU simulation. The total emissions are given in the supplement (section S3).

Sector/Region
Land transport (TRA) Anthropogenic non-traffic (ANT) Shipping (Ship) Aviation (Avia.) calculated in EMAC only and mapped to the finer resolved model instance (see [17]). The source attributions method (tagging) is an accounting system following the relevant reaction pathways and is based on the method introduced by [34]. This diagnostic method allows to completely decompose the budgets of considered chemical species into contributions of sources. For the source attribution, the source terms, e.g. emissions, of the considered chemical species (odd oxygen, NO y , CO, PAN, VOCs, OH and HO 2 ), are fully decomposed into N unique categories. In the present study we distinguish N = 16 different tagging categories (see section S4 in the supplement (available online at stacks.iop.org/ERL/16/064005/mmedia)). The details of the tagging method are described by [14] and [15] and the application in MECO(n) is demonstrated by [35]. The source attribution method allows us to separate the contributions of natural sources (tagging categories: stratosphere, CH 4 , biogenic, N 2 O, biomass burning and lightning) and anthropogenic emissions (all other sources, see section S4 in the supplement).
We have performed two simulations, which both use the same meteorology: BAU (BAU, with the reference emissions described above) and COVID19, which assumes emission reductions in the sectors land transport (TRA), ANT, shipping and aviation as described in section 2.2.

Assumptions on emissions during first half of 2020
For the COVID19 simulation we scale all emission sectors/species (CO, NO x , SO 2 , NH 3 , and VOC) contained in the EDGAR 4.3.1 inventory and the aviation NO x emissions of the RCP 8.5 emission inventory (table 1) in order to represent the reduced anthropogenic activities in individual regions of the globe. The reduction factors are constant over the whole simulation period (1 March-1 July 2020). Total emissions are given in the supplement (section S3).
BAU and COVID19 use the same initial conditions, but different emissions starting from 1 March 2020. Current estimates of rescaled emission factors for early to mid-2020 show large uncertainties and strong temporal variability (see figures 2-6 in [36]). Our study is highly idealized (e.g. by assuming time-independent reductions) and the assumed reductions should be seen as first-order estimates and were chosen during the early phase of the COVID-19 pandemic. Further, the qualitative conclusions (see next sections) are not critically dependent on the exact emission reductions.
A reduction of 30% has been assumed for ANT emissions (comprising industry and households) in Europe (EU), North America (NA) and East Asia (EA) and of 20% for the rest of the world (RoW). TRA emissions have been reduced by 30%, 20% and 20% in RoW, NA and EA, respectively, while a higher reduction factor of 50% has been assumed in EU. Our estimates of emissions in Europe are in line with emission reduction factors for individual countries in Europe estimated to be mostly in the order of 10%-30% for industry and 30%-80% for road transport [37]. Shipping emissions have been reduced by 20%, and aircraft emission by 90%, following ICAO movement data [38], which estimate a reduction of 94% in global aircraft movements.

Measurement data
As observational data (for the period 2017-2020) we used the air quality E1a & E2a data sets (formerly known as AirBase) available at [39]. We have set negative concentrations or unrealistic large concentrations to missing values for further analyses. In total this corresponds to less than 0.2% of all datapoints which are used in our analysis. As the resolution of our model does not account for localised effects [35], we use only data of stations which are classified as 'background'-stations (dataflag AirQuality StationType) in an area classified as either 'remote' or 'remote-rural' (dataflag AirQualityStationArea).
We chose the subset of stations (in total 273), which are available for the whole period for our analyses (section S7 in the supplement).
Similar to many comparable models (e.g. [40]), MECO(n) has deficits in simulating the night time ozone [35]. Usually, the simulated ozone levels during night are too high, while the model is able to capture the ozone peak during day. Therefore, the comparison of model results and measurements is restricted to the period 10-17 UTC.
To compare the model results with the measurements, we sample the one-hourly instantaneous model output at the lowest model layer at the positions of the respective observation stations. All datapoints, which are missing in the observational data for the year 2020, were set to missing values also for the model data.

Lower tropospheric ozone response to emission reductions
In our analyses we focus on the results of the regional model over the area 15 • W-25 • E to 35 • N-70 • N (as shown in figure 1), which is centred over Europe. Hence, we will refer to quantities over this domain as European values. Further, we focus on the period March-May 2020.
The simulated near-surface ozone under BAU conditions shows large temporal variability over Europe during the analysed period (figure 2 and figure S1 in the supplement). The lower tropospheric ozone column (LTC; from the surface up to 850 hPa) increases from around 5 DU in March to up to ∼7 DU in May 2020 due to increased ozone production arising from increased photochemical activity in the course of the year. We choose 850 hPa as upper boundary of the LTC as this typically includes the planetary boundary layer everywhere in Europe except over the alpine region. The peaks with large values of the ozone column especially mid of April are mainly related to events in which high pressure ridges transport ozone rich air masses from lower latitudes to Europe. The reduction of anthropogenic emissions due to COVID-19 continuously reduces the ozone production, resulting in a by 0.6 DU smaller increase of LTC by the end of May. Due to the same dynamics of both simulations the part of the variability driven by meteorology is aligned.
The ozone production efficiency (OPE), i.e. the net-ozone production per NO x molecule, however, increases in the COVID19 simulation compared to BAU (supplementary material figures S2 and S3 and section S2 for the definition). In addition, the commonly used indicator of the ozone production regime, the ratio of the production rate of H 2 O 2 to the production rate of HNO 3 [41], also increases everywhere (figures S4 and S5 in the supplement). This indicates a shift of ozone production from a NO xsaturated or intermediate to a NO x -limited regime, in line with previous findings e.g. by [9].
As a consequence, the contribution of natural emission sources to the ozone LTC increases by 0.  after roughly a month this increased ozone vanishes and lower ozone mixing ratios compared to the BAU simulation dominate over Europe in the COVID19 simulation (figure 1). Large reductions of ozone are found in Southern Europe, except for the polluted metropolitan areas (e.g. around Madrid, Barcelona and Rome) and areas like the Po-valley. In these regions the OPE is rather low and favours an increase in ozone productivity with reduced emissions, counteracting the ozone production decrease from the reduction in precursor emissions. Averaged over the period 15 March-31 May the decrease of surface NO x was between −10% and −40% over the different regions in Europe. At the same time, the reductions of ozone were only up to about −8% at maximum, while especially in Mid Europe ground-level ozone increased by up to 15% (see figure S6 in the supplement).
Our findings of lower ozone values in rural areas are largely supported by surface measurements  Also, the comparison of the difference between the daily ozone minimum and the daily ozone maximum showed similar biases (see figure S9 in the supplement). In general, however, the difference between the BAU and COVID19 simulation shows the same tendency as the difference between the measurements from 2020 and 2017-2019, respectively.
Of course, the difference between the measurements of 2017-2019 and 2020 is also influenced by the meteorological conditions, while the differences between our BAU and COVID19 simulations is caused by the emission reductions only. Nevertheless, the comparison of the measurement data indicates that a reduction of ground-level ozone due to the emission changes under COVID-19 conditions is very likely. This allows us to continue with our analysis of changes with respect to ozone sources related to the reduced emissions.

Attributing ozone reductions to emissions sectors
During May (1 May-30 May) the mean ozone LTC over the European domain (as depicted in figure 1) is roughly 6.1 and 5.6 DU for BAU and COVID19, respectively (see figure 2). The ozone decline in the COVID19 simulation stems from the reduction in anthropogenic emissions, which overcompensates the enhanced ozone productivity. This is caused by the reduction of ozone precursor emissions (mainly NO x and VOCs) and the corresponding ozone increase related to natural sources, 2) DU from BAU to COVID19. This translates to relative contributions of natural emissions to the mean LTC over Europe of roughly 37% and 43% in BAU and COVID19, respectively.
The relative contributions of almost all anthropogenic emissions sectors to the LTC decrease ( figure 4). The sectors with the largest contribution decrease to LTC ozone are the aviation sector (90% emission reduction) with a decrease of 2.7% points and the TRA sector in Europe (50% emission reduction) with a decrease of 1.6% points. This corresponds to an overall decrease of the contribution of anthropogenic emissions from roughly 63% to 57%. The relative contribution of the emission sectors TRA NA and shipping increase slightly, but their absolute contributions (section S5 in the supplement) decrease indicating that the ozone productivity of these two sectors increases slightly more than in the other emission sectors and regions.
Further, our results indicate that European emissions (i.e. TRA EU + ANT EU) contribute only around 15% (BAU) and 13% (COVID19) to lower tropospheric ozone in Europe during May 2020. All other anthropogenic emissions (i.e. shipping, aviation, non-EU TRA and non-EU ANT) contribute roughly 48% (BAU) and 44% (COVID19). This clearly indicates the well-known importance of longrange transport for ozone pollution (e.g. [42]). In addition, the change of the chemical regime implies an increase of the ozone lifetime, since a reduction of ozone leads to a reduction of OH and therefore HO x related ozone depletion rates in the troposphere (e.g. [43,44]). This can be seen by the increased contribution of stratospheric ozone to the LTC of ozone from 6.5% to 7.3% (figure 4). As stratospheric ozone is unperturbed in our COVID19 simulation the influx to the troposphere is almost unchanged. Therefore, the increase of the contribution of ozone from the stratosphere indicates an increase of the tropospheric ozone lifetime. The absolute contribution of stratospheric ozone to the LTC increases by around 2%, indicating an increase of the ozone lifetime of 2%.

Discussion
By design our study is highly idealized as we assume that the emission reductions take place world-wide at the same time and without temporal variability from March to June. There are first studies, which present more detailed emission modelling for Europe and Asia (e.g. [36,37,45,46]). Generally, our assumed reductions are in line with [36] however our estimates for EA ANT and shipping are slightly larger, whereas NA TRA reductions are somewhat low. However, the estimates presented in [36] show considerable uncertainties. Our results need to be interpreted while keeping these simplifications in mind.
Compared to, e.g. [45,47], our study analyses for the first time the impact of the emission reduction on ozone using a source apportionment method. This method allows a more detailed understanding of the changes of the ozone chemistry and is able to attribute the changes to certain emission sectors. Therefore, our study delivers important additional insights.
Even though our emission reductions in Europe of −50% and −30% of LT and ANT, respectively, are very large, we see only a rather small decrease of mean lower tropospheric ozone columns of around 8% during May 2020 over Europe (see figure 2). The main reason for this is the increasing OPE per NO x molecule (see definition of the OPE in the supplement). This leads to a small increase (∼0.1 DU) in the ozone values produced from natural emissions (figures 2 and 3). With respect to potential mitigation options this result demonstrates that detailed assessments are needed to judge, whether planned emissions reductions are sufficient to decrease tropospheric ozone burdens substantially. Indeed, some modelling studies (e.g. [45,47]) indicate a similarly low response of ozone. Some measurement studies (e.g. [4,6,[48][49][50]) even found an increase of ozone near city centres during March 2020 compared to previous years, probably due to decreased NO x emissions. However, also the role of meteorology needs to be considered [51].
Our study further highlights that due to the rather long lifetime of ozone, the emissions in other parts of the world strongly influence European ozone levels. Therefore, reducing emissions only in Europe will most likely not lead to envisaged ozone decreases in Europe.
Our simulation results show that around three months are needed until the difference of the ozone LTC between BAU and COVID19 (see figures 2 and S1) equilibrates. In most countries the strong emission reductions took place for some weeks, only (e.g. [37]). Therefore, the actual effect of the emission reductions during COVID-19 in spring 2020 is likely to be much smaller than the maximum signal of our idealized study.
Due to the uncertainty of the applied emission inventory for BAU and the non-linearity of the ozone production (e.g. 13) our simulated response of the emission reduction (figure 2) might still be overestimated, if our BAU emissions are underestimated. Indeed, the shift of ozone values between our BAU and COVID19 simulation is larger as the shift in the measurements from 2017 to 2019 and 2020. This could indicate an overestimated response, but this difference could also be caused by differences of the meteorological conditions in the previous years (2017-2019) compared to 2020.
The main focus of our study is on near groundlevel ozone, focusing mainly on-air quality related issues. Besides this, changes in ozone and other emissions influence also the climate. However, as already been shown by [3], the overall climate impact of the strong emission reductions during COVID-19 is small. According to [3], the decrease of ozone precursors leads to a short-term cooling, which is offset by a warming effect due to less aerosol.

Conclusion
We conducted a sensitivity experiment (COVID19) to analyse the processes occurring with respect to lower tropospheric ozone in a period of reduced anthropogenic emissions as during the recent COVID-19 pandemic compared to an emission scenario without the impact of the COVID-19 pandemic (BAU). Our simulations with a coupled global and regional CCM and a source attribution technique show: • the ozone LTC averaged over the European domain in the COVID19 simulation become continuously lower over time for around three months compared to the BAU simulation before a new equilibrium is reached. • there are large spatial inhomogeneities with respect to this overall trend in ozone LTCs, which are related to the ozone production regimes. • the overall shift towards smaller ozone LTCs in COVID19 and BAU is also found in measurement data from ground-based stations. • changes in anthropogenic emissions cause the changes in ozone LTCs and are to some degree compensated by enhanced ozone productivity from natural sources. Due to the increase of the OPE the reductions in ozone are much smaller than the emission reductions. In our case NO x at groundlevel is reduced by up to 40% in Europe, while ground-level ozone changes are in the range of −8% to +15% for Europe.
The results of our study are not only relevant for ozone changes related to the recent reduction in emissions due to the COVID19 pandemic, they also are a starting point for discussing mitigation strategies. In line with our model results, measurements during the first half of 2020 and first modelling studies show ozone responses which are much smaller than the emission reductions. This indicates that strong emission reductions are needed world-wide to achieve substantially reduced tropospheric ozone levels.

Data availability statement
The data that support the findings of this study are available upon reasonable request from the authors.