European primary emissions of criteria pollutants and greenhouse gases in 2020 modulated by the COVID-19 pandemic disruptions

. We present a European dataset of daily sector-, pollutant- and country-dependent emission adjustment factors associated with the COVID-19 mobility restrictions for the year 2020. We considered metrics traditionally used to estimate emissions, such as energy statistics or trafﬁc counts, as well as information derived from new mobility indicators and machine learning techniques. The resulting dataset covers a total of nine emission sectors, including road transport, the energy industry, the manufacturing industry, residential and commercial combustion, aviation, shipping, off-road transport, use of solvents, and fugitive emissions from transportation and distribution of fossil fuels. The dataset was produced to be combined with the Copernicus CAMS-REG_v5.1 2020 business-as-usual (BAU) inventory, which provides high-resolution (0 . 1 ◦ × 0 . 05 ◦ ) emission estimates for 2020 omitting the

The time span of the adjustment factors of the current dataset is from 21 February to 31 December 2020.
The beginning of the period corresponds to the date of the first localized lockdown in the region of Lombardy, Italy. The dataset covers: (i) the European first round of lockdowns, when mobility restrictions were at their maximum and remained almost unchanged for five weeks (mid-March until end of April), (ii) the transition period towards the post-lockdown conditions (beginning of May until end of 145 September), when national governments rolled-back COVID-19 measures, and (iii) the new round of lockdowns associated to the second pandemic wave in Europe (beginning of October until end of December), which forced governments back into mobility restrictions. In terms of spatial coverage, we included as many countries as possible that are covered by the CAMS-REG European working domain (30° W -60° E and 30° N -72°N), giving priority to EU27 + UK, Norway and Switzerland. The spatial 150 coverage of the adjustment factors constructed for each GNFR sector as well as a complete list of the countries considered is available in the Supplementary material (Table S1 and Figure S1). Table 1 summarizes the main sources of information used to compute the adjustment factors for each GNFR sector. For the GNFR_B, GNFR_C, GNFR_D, GNFR_E, GNFR_F2, GNFR_F4 and GNFR_I 155 categories, subsector adjustment factors were first computed to take into account the heterogenous impact of the COVID-19 restrictions across the different emission sources in some sectors (e.g., light duty vehicles versus heavy duty vehicles in GNFR_F2 and GNFR_F4). The lists of subsectors considered for each GNFR category are listed in Table 2. The adjustment factors computed for each subsector were later aggregated to the GNFR sector level by considering the relative contribution of each subcategory to total 160 GNFR emissions, as expressed by Eq. (1): subcategory N to total GNFR emissions for country c and pollutant p; being N the total number of subcategories considered for a given GNFR sector (e.g. 3 for GNFR_B, 4 for GNFR_C, according to Table 2). 170 As a result, pollutant-dependent adjustment factors were obtained for these seven GNFR sectors. The emission contributions from each subcategory to total GNFR emissions per country and pollutant (i.e., !%& ( , ) , !%' ( , ) ) were computed using emissions from the GNFR_B, GNFR_C, GNFR_D, GNFR_E, GNFR_F2, GNFR_F4 and GNFR_I sectors split following the subcategories listed in Table 2. 175 Figure 1 shows the resulting emission adjustment factors obtained per day, GNFR sector and selected pollutants. For all sectors except shipping, we show for illustrative purposes results for 6 European countries with different lockdown patterns (i.e., Italy, Spain, France, Germany, the United Kingdom and Sweden). Italy was the country where restrictions first started, followed by Spain and France, where 180 national lockdowns were imposed on 14 and 17 March, respectively. In contrast to Italy, where the transition from low to high stringency levels was gradual, these two countries experienced abruptly severe restrictions on movements, and commercial and industrial activities. A similar pattern occurred in Germany and the United Kingdom, where national lockdowns were imposed on the 20 and 23 March, respectively. Sweden, on the other hand, was one of the few European countries where no national 185 lockdown was implemented and only national recommendations (e.g., relatively soft social distancing measures) were provided to citizens.
The following subsections describe the data and methods for each sector along with the underlying assumptions. The resulting adjustment factors reported in Fig.1 are also discussed in the corresponding 190 subsection.

Public power industry
Changes in emissions from the public power sector (GNFR_A) were assumed to follow the changes observed in the electricity demand data reported by the European Network of Transmission System Operators for Electricity (ENTSO-E) transparency platform (Hirth et al., 2018;ENTSO-E, 2021). For 195 each country, we collected daily electricity demand data from January 2015 to December 2020. For Russia, Ukraine and Turkey we derived the electricity demand data from the corresponding national Transmission System Operators: SO-UPS (2021), UNEC (2021), and TEIAS (2021), respectively.
We first estimated the demand that would have occurred in the absence of COVID-19 under the same 200 meteorological conditions, hereafter referred to as BAU. To estimate the BAU electricity demand we used gradient boosting machine (GBM) models trained and tuned independently for each country using daily data from January 2015 to December 2019. As inputs, we considered the following features: countrylevel daily population-weighted temperature ( _ ( )), date index (number of days since 2015/01/01), Julian date, day of week and a Boolean feature indicating the country-specific bank holidays. The models 205 also consider bridge weekends, in the sense that when there is a holiday on Tuesday (resp. Thursday), the Monday (resp. Friday) is also set as a holiday. We replicated the GBM modelling and tuning strategy previously used in Guevara et al., (2021) with random search in the hyper-parameter space and rollingorigin cross-validation (appropriate for time series). (2) Where '5 ( , ) is the daily mean 2-meter outdoor temperature for grid cell x and day d [°C]; Pop(x) is 215 the amount of population included in grid cell x [nº of inhabitants] and n is the total number of grid cells that corresponds to a specific country. Outdoor temperature information was obtained from the ERA5 reanalysis dataset for the years 2015 to 2020 (C3S, 2017), while gridded population was derived from the Gridded Population of the World, Version 4 (GPWv4; CIESIN, 2016).

220
The difference between the daily BAU and measured 2020 electricity demand levels were used to derive country-dependent daily emission adjustment factors, as described in Eq. 2:  Spain, France, Germany, UK and Sweden). The resulting trends are consistent with the national lockdown calendars and levels of restriction implemented in each country. During the strictest period of the first lockdown, Italy experienced the largest reductions (-30%), followed by Spain (-25%) and France (-20%).
For Sweden, positive values are observed during the same period, in line with the results reported by Le Quéré et al. (2020). It is likely that in this country electricity demand from commercial services remained 235 unperturbed as no national lockdowns were enforced. We also hypothesize that a voluntary self-isolation of a fraction of the population may have increased household electricity consumption. When confinement was eased, electricity demand shows the first signs of recovering in all countries. This trend is confirmed in summer, as governments softened even more lockdown measures. Italy is where the recovery is more pronounced, reaching emissions above the BAU during August. A second significant drop of emissions 240 is observed in France and UK and, to a lesser extent, in Italy during November 2020 coinciding with the implementation of a second round of lockdowns. Emissions rebound sharply after that, and are back to BAU levels or even above during Christmas holidays.

Manufacturing industry
The adjustment factors for manufacturing industry (GNFR_B) are based on the monthly Industrial 245 Production Index (IPI) values reported by Eurostat (2021a). We considered the seasonally and calendar adjusted data. Note that for UK the IPI values for November and December 2020 were derived from ONS (2021)  subcategories listed in Table 2, according to the impacts of the COVID-19 restrictions observed on their 250 activity: • GNFR_B1: Manufacture of petroleum refining products. This industrial branch was considered to be essential and therefore was less affected than other industries during the full lockdown phase.
However, and due to the large decrease on the demand for finished petroleum products (e.g., jet 255 fuel, motor gasoline), the recovery of its activity has been lower than in other sectors during the lockdown exit process.
• GNFR_B2: Manufacture of pharmaceutical, chemistry, food and beverages products. These industrial branches were also considered to be essential during the full lockdown phase, but in contrast to the petroleum industry, the demand associated to their products barely decreased or 260 even increased during or after the lockdown, which is translated in a low decrease (slight increase) of their activity.
• GNFR_B3: Manufacture of other products (i.e., non-metallic mineral products, basic metals, paper and paper products and machinery and equipment). These industries were considered nonessential and therefore were heavily affected during the lockdown period as in the majority of 265 cases were forced to close. Nevertheless, a sharp recovery is observed with the easing of lockdowns.
For illustration, Fig. 2 shows the behaviour of the IPI monthly values from January 2019 until December 2020 for each of the three aforementioned manufacturing industrial subgroups for Germany, Spain, Italy 270 and the UK. The dashed black line represents the general IPI for the overall manufacturing industry. For the manufacturing industrial subcategories GNFR_B2 and GNFR_B3, the averaging was done considering the share of each industrial branch (i.e., pharmaceutical, chemistry, food and beverages products for GNFR_B2 and non-metallic mineral products, basic metals, paper and paper products and machinery and equipment for GNFR_B3) to the total fossil energy final consumption as reported by the 275 Eurostat (2021b) energy balances. For GNFR_B3, the manufacture of basic metals and non-metallic mineral products are the largest energy intensive activities (almost 70% of total energy consumption), whereas manufacturing of paper and machinery and equipment represent approximately 30% of total energy consumption (Fig. S2). Note that other industrial branches originally included in GNFR_B3 (i.e., manufacture of wood, textiles and leather) were not considered in the final calculations since the Eurostat 280 IPI statistics for these industrial categories are incomplete. It is expected that the removal of these industrial branches won't have a major impact on final results as their total fossil fuel consumption is not predominant (i.e., 12% in total according to Fig. S2).
For each manufacturing industry subgroup, we computed monthly and country-specific adjustment 285 factors from a baseline taken as the average value over the two months prior to the lockdown (January and February 2020). The computed monthly adjustment factors were translated into daily adjustment factors by considering the Oxford COVID-19 Government Response Tracker dataset (OxCGRT; Hale et al., 2021). The OxCGRT provides a systematic cross-national, cross-temporal measure to understand how government responses have evolved over the full period of the COVID-19 spread. We considered the 290 indicator "workplace closing", which records the closings of workplaces according to four different scales of intensity: 0 -no measures, 1 -recommended closing, 2 -require closing (or work from home) for some sectors or categories of workers and 3 -require closing (or work from home) all-but-essential workplaces. We assumed that changes on industrial emissions during March started to happen in each country once the corresponding indicator reached a value of 2 or more. 295 Daily emission adjustment factors were computed as a weighted average of the adjustment factors obtained for each industrial subcategory (Eq. (1)), taking into account their relative contribution to total GNFR_B emissions (Fig. 3). 300 Figure 1 illustrates the resulting adjustment factors proposed for NOx and NMVOC emissions, respectively. A common pattern is observed for the two pollutants, with the largest reductions occurring during April, when the restrictions were at their maximum and a large number of facilities were not allowed to operate. A pronounced recovery is observed from May onwards, coinciding with the easing of the lockdowns and the recovery of the industrial activity. For NOx, the computed reductions are larger 305 than for NMVOC, with Italy, France and Spain presenting the largest decrease (between -35% and -40% during April). Low reductions are observed for Sweden, where emissions never decreased more than -20%. Emission reductions reached levels close to BAU by the end of the year in almost all countries, as the new curfews adopted around October/November/December did not affect the manufacturing industry.
In the case of NMVOC, a general lower reduction than for NOx emissions is observed, with most 310 countries presenting a maximum decrease below -30% during April. It is worth noting that some countries even experienced an increase in emissions during the beginning of the first lockdowns (up to 10%). The adjustments computed for NMVOC are different relative to NOx as its emissions are related to food, beverage, pharmaceutical and chemical industry branches (Fig. 3), which were less affected by the COVID-19 restrictions or even had to increase their productivity due to an increase in demand. The largest 315 emissions reductions are reported for Italy, and the lowest ones for UK and Sweden, with the latter even showing emission values above BAU levels (i.e., up to 5%) during the second semester of 2020.

Other stationary combustion activities
This sector includes emissions from stationary combustion activities related to the commercial and institutional sector, the residential sector and other stationary sectors such as agriculture, forestry, fishing 320 and military sectors.
Our emission adjustment assume that the COVID-19 restrictions only affected the combustion activities in the commercial/institutional and residential sectors. In the first case, significant emission reductions are expected as a result of the closure of schools, universities, public buildings, restaurants, and other 325 non-essential businesses. In the second case, emission increases are expected due to the required household confinement during the lockdown period. Regarding the agriculture, forestry and fishing sectors, we assumed no changes occurred as they were considered to be essential. transit stations, retail and recreation, residential and workplaces) based on aggregated and anonymized sets of data from users who have turned on the Location History setting for their Google Account on their mobile devices. Reductions for each day are calculated by Google from a baseline taken as the median 335 value, for the corresponding day of the week, over a 5-week period prior to the lockdowns (3 January to 6 February). For this sector, we used the mobility trends reported for the following categories: • Retail and recreation: Mobility trends for places like restaurants, cafes, shopping centres, theme parks, museums, libraries, and movie theatres 340 • Grocery and pharmacy: Mobility trends for places like grocery markets, food warehouses, farmers markets, specialty food shops, drug stores, and pharmacies.
• Workplaces: Mobility trends for places of work.
• Residential: Mobility trends for places of residence 345 The mobility trends for retail and recreation, grocery and pharmacy and workplaces were used to derive an average trend for the commercial/institutional sector, while the mobility trends for places of residence were used for the residential sector.
These Google trends report changes in movements, which does not necessarily represent changes in 350 energy consumption (i.e., fossil fuels and biomass) and associated emissions. The increases in residential activity reported by Google are significantly larger than the ones reported in Le Quére et al. (2020), which indicates an average increase of 5%, and a maximum increase of 10% during the most restrictive lockdown phase. The results reported in Le Quére et al. (2020) inferred from UK smart meter data are consistent with the ones reported by the thermostat maker Tado (Tado, 2020), which indicates an average 355 increase of 14% in home heating consumption in Europe during March 2020 compared to March 2019.
Considering the aforementioned results, the original Google trend values for the residential sector were scaled down for countries to have a maximum daily relative change of 10%. Our approach is limited by data availability and further constraints will require more data on residential energy consumption. In the case of the commercial/institutional sector, we also adjusted the original daily decrease trends reported by Google making use of energy consumption statistics. We used information provided by IDAE (2018) on the energy consumption in the Spanish commercial/institutional sector. As shown in Table S2, Spanish commercial buildings represent more than 40% of the total energy consumption (fossil fuels and biomass) in the commercial/institutional sector, followed by workplaces (26.5%), hospitals (11.6%), other 365 buildings (8.8%, e.g., museums, public buildings and religious buildings), schools and universities (7.8%) and restaurants and hotels (4.3%). We hypothesized that the Spanish national lockdown restrictions implied a change in the energy consumption of: (i) -100% in schools and universities (all buildings were closed), (ii) -90% in hotels and restaurants (certain hotels were converted into temporary medical facilities), (iii) -80% in workplaces, commercial buildings and other buildings (supermarkets and other 370 grocery stores remained opened during the entire lockdown, as well as certain workplaces that were considered to be essential) and (iv) +50% in hospitals (due to the increase in the number of patients to attend). We combined the aforementioned information and derived an overall maximum reduction in energy consumption across Spanish commercial/institutional buildings of -66.9%. Following the approach applied for the residential sector, we scaled up the original Google trend values for the 375 commercial/institutional sector to set this minimum value.
Daily emission adjustment factors for the other stationary combustion sector were computed as a weighted average of the adjustment factors obtained for each GNFR_C subcategory (Eq. (1)), taking into account their relative contribution to total emissions (Fig. 3). 380 Figure 1 illustrates the resulting adjustment factors proposed for NOx and PM10 emissions, respectively.
For NOx, major reductions are observed for the United Kingdom, France and Italy. In these three countries, maximum reductions between -15% and -20% were reached during the strictest lockdown period. On the contrary, and despite being under similar lockdown measures, in Spain the maximum 385 relative reduction during the same period was only -10%. This is explained by the different contributions of agriculture, forestry and fishing subcategories (GNFR_C3) to the total GNFR_C NOx emissions.
While in Spain this category represents around 40% of total NOx emissions, in France, Italy and the United Kingdom the contribution is lower than 10% (Fig. 3). Assuming that this category was not affected by the COVID-19 restrictions implies a lower overall emission reduction in Spain. In the case of Sweden, 390 a slight emission increase is observed during the whole period of study. We hypothesize that this is a consequence of the likely small perturbation of the public and commercial service activity (i.e., nonessential businesses were not forced to close) and a slight increase of the residential activity as a consequence of a voluntary self-isolation of a fraction of the population. By the end of August most countries reached or were about to reach their BAU levels, except for the United Kingdom, where 395 emissions were still -10% below pre-lockdown values. A second significant drop in emissions is observed in France, United Kingdom and Italy during November, which is related to the forced closure of nonessential business under the second epidemic wave.
For PM10, an increase in the business-as-usual levels is observed for all selected countries. This is 400 explained by the fact that a majority of total emissions are driven by changes in the residential sector ( Fig.   3), which increased its activity due to the enforced confinement. Germany is the country that registered the lowest increase in total emissions (maximum increase of approximately 2.5%) compared to the other countries. This is again explained by the different contributions of subcategories to total GNFR_C emissions. In this particular case, the German commercial/service subcategory represents around 10% of 405 total emissions, while in the other countries the contribution for this subcategory is less than 5% (Fig. 3).
By the end of August, all countries were close to reach the BAU levels again, and in some countries like Italy emissions levels even reached values below BAU, as people started to spend more time outdoors. A slight increase in emissions is observed during November, coincident with the introduction of new additional mobility restrictions to curb the high incidence during the second wave of COVID-19 spread. sector were assumed to be unaffected by lockdowns and mobility restrictions.
The following sources of information were used to derive the adjustment factors: • GNFR_D1; coal mining and handling: monthly indigenous production of hard and brown coal per country reported by Eurostat (2021c). We computed monthly and country specific adjustment 420 factors from a baseline taken as the average value over the two months prior to the lockdown (January and February 2020). We then averaged the resulting monthly factors per month and country and derived daily adjustment factors using the "workplace closing" reported by OxCGRT, as detailed in Sect. 2.1.2.
• GNFR_D2; refining / storage & venting and flaring: monthly IPI related to the manufacture of 425 petroleum refining products (Eurostat, 2021a). For this subcategory, we used the same adjustment factors as for GNFR_B1 of the manufacturing industry (see Sect. 2.1.2).
• GNFR_D3; distribution of oil products (gasoline): we assumed that changes in this activity can be represented by changes in road fuel sales in filling stations, which at the same time can be linked to changes in road traffic activity. This hypothesis is illustrated in Fig. S3, which shows the 430 relationship between monthly/weekly changes in petrol sales and traffic activity for selected countries. In all cases the Pearson correlation coefficient (PCC) is larger than 0.9, the intensity in the drop of petrol sales during the lockdown periods fairly coinciding with the decrease in traffic activity. Considering these results, for this activity we used the same emission adjustment factors for road transport gasoline exhaust emissions (see Sect. 2.1.6). 435 GNFR sector-level daily emission adjustment factors were computed as a weighted average of the adjustment factors obtained for each subcategory (Eq. (1)), taking into account their relative contribution to total GNFR_D emissions (Fig. 3). the individual subcategory that dominates total emissions in each country, and to a lesser extent due to the different levels and types of restrictions implemented. For instance, in UK almost 40% of total NMVOC emissions come from refining activities (storage, flaring) and therefore the decrease in 445 emissions is largely driven by their decrease (Fig. 3). On the other hand, approximately 50% of total NMVOC emissions in France comes from the distribution of oil products, and subsequently the drop in emissions is similar to that of road traffic emissions, with two significant drops corresponding to the lockdowns implemented during the Spring and Fall epidemic waves.

Use of solvents 450
The GNFR_E category includes NMVOC emissions coming from the residential/commercial and industrial use of solvents. Our assumption for this sector is that the COVID-19 restrictions only affected certain industrial subcategories, including: (i) GNFR_E1: the use of organic solvents to remove grease, fats, oils, wax or soil from metal products and (ii) GNFR_E2: the use of inks in the printing industry.
Other industrial activities that involve the use of solvents (e.g., manufacturing of pharmaceutical products 455 or automobiles) could not be considered as they are not individually distinguished in the NFR reporting nomenclature, but rather reported as part of broader categories (e.g., 2.D.3.g: Chemical products, 2.D.3.i: Other solvent use, 2.G: Other product use). Emissions from domestic and commercial solvent use were assumed to remain constant due to the lack of specific activity data to compute the adjustment factors and the limited number of categories considered in the NFR nomenclature. We hypothesize that the potential 460 increase in the use of certain products containing solvents, such as cleaning products, was compensated by the potential decrease in the use of other products, such as car products or cosmetics for personal care.
We are aware that this hypothesis may be limited by the increased use of the so-called "pandemic products" triggered by the COVID-19 (Steinemann et al., 2021), which includes products intended to clean and disinfect, such as hand sanitizers or surface cleaners. However, the lack of specific information 465 does not allow us computing associated adjustment factors.
The adjustment factors for industrial solvent use are based on the monthly IPI values adjusted for seasonal and calendar effects (Eurostat, 2021a). As already mentioned in Sect. 2.1.2, for UK the IPI values for November and December 2020 were derived from ONS (2021). The "Manufacture of fabricated metal 470 products, except machinery and equipment" and "Manufacture of computer, electronic and optical products" on the one hand, and the "Printing and reproduction of recorded media" on the other hand were the industrial branches considered to quantify the impacts of restrictions on each of the two subcategories considered. For each subcategory, we computed monthly and country specific adjustment factors from a baseline taken as the average value over the two months prior to the lockdown (January and February 475 2020). The computed monthly adjustment factors were translated into daily adjustment factors by considering the "workplace closing" reported by the OxCGRT, as detailed in Sect. 2.1.2.
Daily emission adjustment factors for the use of solvents sector were computed as a weighted average of the adjustment factors obtained for each subcategory (Eq. (1)), taking into account their relative 480 contribution to total GNFR_E emissions (Fig. 3). Figure 1 illustrates the resulting adjustment factors proposed for NMVOC emissions. Decrease in emissions is generally low (i.e., below -10%) and mainly occurring during the Spring lockdowns. The small reductions are due to the limited contribution of metal cleaning and printing industrial activities to 485 the overall emissions from this sector (Fig. 3). A pronounced recovery is observed from May onwards, coinciding with the easing of the lockdowns and the recovery of the industrial activity.

Road transport
The emission adjustment factors considered for this sector are based on the Google COVID-19 Community Mobility Reports (Google LLC, 2021). We used the mobility trends reported for the transit 490 stations category, which includes places like public transport hubs such as subway, bus, and train stations.
We compared the Google movement trends against trends derived from measured traffic counts reported by 18 European national road administrations. Table A1 summarises the countries covered, sources of information and characteristics of the traffic count datasets considered, as well as the baseline considered to derive traffic activity trends.  On the other hand, a large discrepancy is observed between Google results and the HDV measured-based trends, the former presenting larger reductions. In the UK for instance, the average reduction for HDV was of -35.6% between March and 26 April, almost two times lower than the one reported by Google (-69%, respectively).
• COVID-19 lockdown-exit process (mid-May until end of September): Differences between LDV 510 and Google trends become larger, showing different rates of recovery. Google tends to underestimate the observed recovery of traffic activity. The discrepancies between measured trends and the Google dataset become larger with time. During summer (i.e., July, August), the LDV trends in the majority of countries are close or even above business-as-usual levels (e.g., Netherlands, Ireland), yet Google continues to report mobility values that are below business-as-515 usual levels. In the case of HDV trends, discrepancies with Google trends are reduced but still significant.
• Second COVID-19 lockdown period (beginning of October until end of December): Discrepancies between Google trends and LDV/HDV measured-based trends remain almost unchanged. Google trends are, qualitatively speaking, capable of reproducing the drops in traffic 520 activity observed in the LDV measured-based trends during November and the Christmas season, but not quantitively speaking, as reductions are systematically larger than the observed ones.  Figure 5 shows a comparison between averaged monthly adjustment factors for road traffic reported by Google LLC (2021) and LDV measured-based trends per each of the countries listed in Table A1. 525 Discrepancies between Google and measured-based trends started to increase with the easing of the restrictions (May) and reached a maximum difference in September. During this month, the average traffic reductions reported by the LDV measured-based trends are in a range between -20% and 0%, while in the case of Google reductions are between -40% and -10%. The overestimation of Google reductions when compared to measured-trends is somewhat reduced during November and December, coinciding 530 with the implementation of new lockdowns, but large differences are still observed. The plot also shows how the decrease of traffic activity during April (first lockdown) was much larger than during the second wave of restrictions (i.e., November): In April, maximum traffic reduction was -80% (Spain, ES), while in November the maximum drop was around -35% (France, FR).

535
The differences observed between measured-based trends and the Google trends are mainly related to the fact that Google data refers to mobility trends in public transport hubs. As a result of COVID-19, people are now avoiding public transport as these can be considered places where it might be difficult to avoid contact with other passengers (De Vos, 2020). The adjustment factors proposed by Google during the lockdown exit process are affected by this factor and therefore are underestimating the observed changes 540 in traffic activity during the lockdown exit process. This hypothesis is illustrated in Fig. S4, where the traffic movement trends obtained in Rome are compared to the evolution of the access to subway stations.
The recovery of mobility in the subway system during the lockdown exit process is very much in line with the Google trend and much lower than the one observed for the private transport sector. On the other hand, the lower reduction observed in HDV's activity when compared to Google is because these vehicles 545 supported the delivery of essential goods and products during the confinement (e.g., food, medical supplies), and subsequently their use decreased much less than that of LDV.
In order to overcome the identified limitations of the original Google trends, we used the LDV and HDV were computed as the ratio between the weekly average changes in traffic activity reported by the measured trends and the weekly average changes in mobility reported by Google. The resulting countrylevel weekly correction factors were then averaged to obtain a set of European weekly correction factors.
The countries considered to develop the European average weekly correction factors were the ones listed 555 in Table A1 except Poland and Estonia, as the number of traffic stations used to derive measured-based trends for these two countries was small.
The two sets of correction factors were applied to the original Google mobility trends in order to derive two new sets of adjustment factors for LDV and HDV emissions. Note that for those countries for which 560 we had daily traffic count datasets available (i.e., United Kingdom, Norway, France, Spain, Finland, Ireland, Netherlands and Switzerland) we directly substitute the original Google trends for the ones derived from traffic counts. Similarly, for countries with weekly and monthly traffic count datasets, adjustments of the original Google trends were done by considering only the correction factors of the corresponding country. 565 We applied the adjusted Google transit mobility trends with the LDV factors to the GNFR_F1 (exhaust gasoline) and GNFR_F3 (exhaust LPG gas) sectors, as the contribution of HDV to their emissions is null or almost residual. However, for the GNFR_F2 (exhaust diesel) and GNFR_F4 (non-exhaust) sectors, the final emission adjustment factors were computed as a weighted average of the adjustment factors obtained 570 for LDV (GNFR_F21 and GNFR_F41) and HDV (GNFR_F22 and GNFR_F42) vehicle categories following Eq. (1) and considering their relative contribution to total corresponding emissions (Fig. 3). to curb traffic activity in October. Strengthening measures caused a second significant drop in emission during November, although it was ~50% lower than that of April (e.g., UK, Italy). The first weeks of December were marked by a relaxation of the second lockdown measures and a subsequent recovery of the traffic emissions. However, a third drop in emissions was observed during the Christmas season, as additional measures were implemented to restrict social gatherings. 595

Aviation
We derived the adjustment factors related to air traffic emissions during Landing and Take-Off cycles (LTO) in airports from statistics provided by EUROCONTROL (2021), which reports daily arrivals and departures by airport from January 2016 to December 2020. We computed day and country-specific flight operation reductions from a baseline taken as the average value for the corresponding day of the week 600 (Monday to Sunday and National holidays) and month of the year from 2019. before the beginning of April. In contrast to road transport, the signs of recovery during May and June are very weak as the movements between countries were still restricted at that time. On the contrary, a general more pronounced recovery was observed during July and August as a consequence of the beginning of summer holidays and the lifting of restrictions to travel. This recovery was especially 610 significant in Spain and France. However, most of the countries still presented reductions larger than -50% during summer. Strengthening measures linked to the second wave of infections negatively impacted European air traffic in November, when new drops were observed, especially in the UK and France. The end of the year, however, was marked by a recovery in air traffic operations, similarly to the one observed during summer, that can be attributed to the Christmas holidays. were already close to BAU levels (e.g., BAS, -5%). Overall, MED and NOS were the sea regions presenting the largest (i.e., -17%) and lowest reductions (i.e., -3%), respectively. The contrast in the results obtained for these two sea regions is very much related to the different contribution of passenger ships to total shipping traffic, which is larger in the MED than in NOS. As reported by EMSA (2021), cruise ships and Ro-Ro/passenger ships were the ship types mostly affected by showing 635 reductions of 2020 ship calls in EU ports of -85% and -39% when compared to 2019 levels. These reductions were significantly larger than the ones reported for cargo ships (between -7% and -2%), which are dominant in NOS.

Off-road transport
The GNFR_I category reports emissions from non-road mobile machinery that is used in several sectors, 640 including: (i) commercial (e.g., transportable equipment), (ii) residential (e.g., gardening and handheld equipment), (iii) agriculture/forestry/fishing (e.g., harvesters, cultivators), (iv) manufacturing industries and construction (e.g., excavators, loaders, bulldozers) and (v) other categories including military, landbased railways and recreational boats. In the present work, the impact of COVID-19 restrictions was quantified for emissions related to mobile machines in the manufacturing industry and construction sector 645 (GNFR_I1), while emission from the other subcategories (GNFR_I2) were assumed to remain unaffected.
The adjustment factors are based on seasonally and calendar adjusted monthly IPI values reported by Eurostat (2021a). We considered the IPI values reported for the general manufacturing and construction categories. As for the manufacturing industry, monthly and country specific adjustment factors were 650 computed taking as a baseline the average value over January and February 2020. The translation from monthly to daily factors was done considering the evolution of the "workplace closing" indicator reported by OxCGRT. Figure 1 shows the emission adjustment factors for NOx emissions. The decrease in emissions is generally 655 low, with a maximum reduction of less than -15% in UK during April, and reductions between -2.5% and -5% in Germany and Spain during the same period. As showed in Fig. 3, the contribution of the manufacturing industry and construction machinery subcategory to total emissions is rather low (30% in average at the EU27 + UK level), which explains why reductions are not as large as the ones showed in e.g., the GNFR_B manufacturing industry sector. Emissions are reaching levels close to BAU by the end 660 of the year in almost all countries, as the new virus-related curfews adopted during the second wave did not affect the industrial manufacturing and construction activities.

Business-as-usual 2020 emissions
Our adjustment factors were designed to be applied to a gridded emission BAU inventory for 2020 developed based on the CAMS European regional emission inventory (CAMS-REG_v5.1) time series, 665 ranging from 2000 to 2018 (update from Kuenen et al., 2021). The CAMS-REG_v5.1 dataset makes use of official air pollutants and greenhouse emission inventories submitted by each country to EMEP, UNFCCC and the EU. Those country-level annual data form the basis of the emission inventory and are spatially disaggregated to a 0.1° × 0.05° grid for use in chemical transport models. Besides the 12 GNFR sectors for which the COVID-19 adjustment factors are prepared (Table 1), the inventory also includes 670 emissions from waste management (GNFR_J), livestock (GNFR_K) and other agricultural activities (GNFR_L). Additional sub-sectors are also defined, as explained before in Sect. 2. The methodology applied and sources of information used for the construction of the CAMS-REG emission inventory are described in detail in Kuenen et al. (2021) 675 The main disadvantage of the CAMS-REG_v5.1 gridded inventory is the 2-year lag in emission reporting.
To overcome this limitation a method was developed to estimate emissions for recent years (y-1), which makes use of sector-specific activity data. We have updated this methodology to make a BAU emission estimate for 2020 to be combined with the COVID-19 adjustment factors described in Sect. 2. The method follows three steps: 680 • Estimate the activity data (AD) per sector, country and year. For this we gathered data from a range of sources, which are listed in Table 3. If activity data are available for 2020 we use it directly. Otherwise, if activity data are available for previous years (time series cover between 7 to 21 years for the different data sources) we examine whether a significant trend exists (R 2 > 0.3) 685 and extrapolate that to 2020. Additionally, we have included the impact of the 0.5% sulphur cap on (international) shipping fuels as of January 1, 2020 (IMO, 2019). For the North Sea, Baltic Sea and English Channel we assume no impact of the sulphur cap, as these sea regions are part of the Sulphur emission control areas (SECA) and already showed strong reductions before (Kattner et al., 2015). For all other sea regions, we assume a 75% reduction in SO2 emissions compared to 2018. Also for PM we assume a 48% reduction compared to 700 2018 due to the reduction of SO4.
For the 2020 BAU emission estimates we ignore all AD that is impacted by the COVID-19 lockdowns and mobility restrictions. We still use the AD for trend analyses though, as a trend caused by, for example, technological progress will continue in 2020 and therefore be part of the BAU emission estimates. Note 705 that not all GNFR sectors are included in Table 3, for example due to absence of AD. In that case the emissions from 2018 are copied to 2020. weaker and also the COVID-19 impact on the manufacturing industry is less. Emissions from other 715 stationary combustion activities show an increase in 2020 in Italy (+5%), because it was a bit colder than in 2018. In Sweden, 2020 was warm compared to 2018 and the opposite effect is visible (-15%). This estimate is not affected by COVID-19, because it is purely based on the temperature (i.e., changes in the yearly degree days). Note that the estimate with COVID-19 is not comparable to the adjustment factors, as the AD used here do not necessarily capture the impact of the lockdowns. We merely use it to illustrate 720 that the BAU estimate indeed represents a situation without COVID-19. changes for each sector and country, as described in Guevara et al. (2021). The analysis of the results focuses on multiple aspects of the COVID-19 restrictions on emissions, including a description of the temporal evolution of emissions at the EU27 + UK level, per country, species and pollutant sector, as well as an analysis of the spatial distribution of the changes in total emissions. Figure 7 illustrates the COVID-related changes in the EU27 + UK daily emissions for criteria pollutants and greenhouse gases (GHGs) between January 1 st and December 31 st 2020 as compared to the BAU scenario. Dotted and solid lines represent the BAU and COVID-19 daily emissions, respectively, and differences between them are illustrated with the shaded areas. few exceptions during the Christmas holidays. It is important to note that the daily evolution of the emissions plotted in the charts is not only affected by the COVID-19 restrictions, but also by the inherent seasonality associated to emissions from each pollutant sector. For instance, emissions from other stationary combustion activities are mainly related to the combustion of fuels in households/commercial buildings for space heating, and therefore they decrease as winter ends and outdoor temperatures start to 755 be higher. This fact can be observed with the daily evolution of PM2.5 and CO2_bf emissions, as they are mainly driven by residential wood combustion emissions.

European and country-level analysis 735
In the aggregated, a reduction of -10.5% (-602 kt) was seen in NOx emissions, followed by a -7.8% (-260.2 Mt) in CO2_ff, -4.7% (-808.5 kt) in CO, -4.6% (-80kt) in SO2, -3.3% (-19.1 Mt) in CO2_bf, -3.0% 760 (-56.3 kt) in PM10, -2.5% (-173.3 kt) in NMVOC, -2.1% (-24.3 kt) in PM2.5, -0.9% (-156.1 kt) in CH4 and -0.2% (-8.6 kt) in NH3. The largest decline in European emissions was observed during the month of April for all pollutants, with an abrupt -32.8% and -25.5% decrease in total NOx and CO2_ff emissions, which corresponds to -157.3 kt and -70.2 Mt, respectively (Fig. 8). Around 25% of the total drop in emissions occurred in 2020 took place during the month of April. As mentioned before, emission levels 765 in September were already close to the pre-lockdown levels, although still presenting a slight decrease when compared to the BAU scenario (-4.8% and -3.9% for NOx and CO2_ff, respectively). The emission reductions observed during November and December (i.e., up to -10.5% for NOx and -6.5% for CO2_ff) were lower than those that occurred during the first epidemic wave because mobility restrictions implemented by governments were generally slower and softer (e.g., curfews, limited social gatherings, 770 early closing times for restaurants and bars) and only had to be toughened in those countries affected by a new and more contagious variant of the COVID-19 such as France, Germany, the UK and the Netherlands.
Results shown in Fig. 7 and Fig. 8 allow illustrating the heterogeneous impact of the COVID-19 775 restrictions on total emission changes across pollutants. Worth noting is the large contrast between decreases in NOx (-10.5%) and PM10/PM2.5 (-3% and -2.1%) emissions (see Sect. 4.1.2 for further discussions). The almost null reduction reported for NH3 and CH4 emissions is linked to the fact that the large majority of these emissions are related to agricultural and waste management activities (e.g., use of fertilizers, manure management and livestock), which in the present work were supposed to remain 780 unaffected during the COVID-19 restrictions. This assumption is in line with the results published by Elleby et al. (2020), which indicate that COVID-19 implied a reduction of direct GHGs from agriculture of only about 1% at the global scale. Figure 9 and 10 show the relative decline (%) in total emissions per country and species for criteria 785 pollutants and greenhouse gases, respectively. Vertical lines indicate the average relative declines computed at the EU27 + UK level for each species. Non-shaded marks highlight those countries/species where reductions were larger than the ones observed at the EU27 + UK level. A large variation in the relative declines of emissions is observed between countries due to 1) the heterogeneous levels and types of restrictions implemented across countries, and 2) the different contributions of each pollutant sector, 790 particularly of the road transport sector and other stationary combustion activities, to total emissions in each country.
The most pronounced declines occur for NOx and CO2 fossil fuel emissions, Italy being the country where these two pollutants suffered the largest relative reduction (i.e., -15.1% and -11.4%, respectively). On the 795 other hand, Malta presents the largest relative reductions of SO2, CO, NMVOC and CO2 biofuel emissions (between -17.2% and -6.8%). Despite not being among the countries where the strictest lockdowns and containment strategies took place, the contribution of road transport to total CO, NMVOC and CO2 biofuel emissions in this country is significantly larger than what it is reported at the EU27 + UK level (i.e., 54.1% versus 14.8%, 87% versus 21.1%, 40.3% versus 7.5% and 69% versus 10.5%, respectively). 800 A similar situation is observed in Cyprus, which presents the largest relative reduction of total PM2.5 emissions (-6.2%). This country reports the lowest fraction of PM2.5 emissions from other stationary combustion activities (4.9% versus 52.1% at the EU27 + UK level), a sector that suffered an increase in emissions during lockdown restrictions (see Sect. 4.1.2). For PM10 emissions, the largest relative drop occurs in the UK (-6.5%), which is among the countries that suffered the strictest restrictions. In the case 805 of CH4 the largest reduction is observed in Romania (-4.1%) mainly due to the decrease of emissions from coal mining activities. Finally, for NH3 most of the EU countries present relative reductions close to the average value and that are almost negligible (between -0.56% and -0.03%), as in all of them agricultural activities, which remained unaffected by COVID-19 restrictions, represent more than 90% of total NH3 emissions. Results also show that for certain countries and species, emissions not only decreased 810 but, in some cases, slightly increased due to the COVID-19 restrictions. This is the case, for instance, of PM2.5 emissions in Hungary and CO2 biofuel emissions in Croatia (i.e., 0.4% in both cases). The observed increase in these two countries is a direct consequence of the large contribution of the other stationary combustion activities to total PM2.5 and CO2 biofuel emissions, respectively. In Hungary, this sector represents 81.3% of total PM2.5 emissions, whereas in Croatia it represents 79.9% of total CO2 biofuel 815 emissions. These values are much larger than the average contribution observed at the EU27 + UK level, which is 52.1% and 39.1%, respectively. The aviation sector presents the largest drop among all sectors, with a reduction of between -51 and -56% in emissions during 2020. The second most affected sector is road transport, which presents a decline in emissions between -15.5% and -18.8%, depending on the pollutant. These two are by far the sectors 825 affected the most by the COVID-19 restrictions, with NOx emission declines reaching approximately -90% and -60%, respectively, during April (Fig. 13). Despite showing drops of similar intensity, the recovery of emissions differs significantly between these two sectors. For road transport emissions started to gradually and steadily recover during late April and almost reached again BAU levels by September For the manufacturing industry and other stationary combustion activity sectors, a heterogenous impact of the COVID-19 restrictions is observed across the different pollutants. For the first sector, a lower 840 reduction is observed for NMVOC and NH3 (between -2.8% and -3.5%) when compared to the other pollutants (between -6.8% and -7.2%). This is due to NMVOC and NH3 emissions being mostly driven by processes occurring in the food/beverage and chemistry industries, which were considered essential during the lockdown phase and were therefore less affected than other industry branches, such as the manufacturing of basic metals or mineral products (see Sect. 2.1.2). Similar to road transport, the largest 845 drop in industrial emissions was reported during April (i.e., -25% for NOx), when a significant number of facilities were not allowed to operate. However, emissions began to recover in late April and May, as industrial activities fully resumed in large part of Europe. As shown in Fig. 13, emissions from this sector quickly picked up again, approaching its pre-pandemic levels of activity during November (i.e., -1.1%

Sectoral analysis
for NOx). The reason for this rapid recovery is the fact that, unlike other sectors such as road transport For the other stationary combustion activities, the pollutants that are mainly related to residential wood combustion processes (i.e., PM10, PM2.5, NH3, NMVOC, CO, CO2_bf and CH4) experienced a slight 855 increase (between 1.1% and 1.7%), while the rest of pollutants (i.e., NOx, SO2 and CO2_ff) showed a modest decrease (between -0.4% and -2.9%). In both cases, the cumulative changes were not substantial and after the lockdowns in Spring, emissions were practically back to BAU levels by the end of July 2020 ( Fig. 13). A new decrease in emissions is observed during November and December, coinciding with the new round of restrictions and the closure or limitation of working hours of non-essential commercial 860 business such as restaurants or shopping stores.
In the public energy sector, the overall relative reduction in emissions during 2020 was approximately -3.3%. As for the previous sectors, large differences are observed between months: In September, public energy emissions in the COVID-19 scenario were only -2.5% lower than in the BAU scenario, compared 865 to being -12% lower in April. As in the case of the manufacturing industry sector, emissions were barely affected during the Fall restrictions and were almost back to BAU levels during December.
The shipping sector experienced a decrease in emissions of around -9.5% for all pollutants. The evolution of daily emissions in this sector indicates a slow recovery of the activity, which is partially linked to the 870 slow recovery of maritime passenger services. Decrease in emissions from off road transport emissions were between -3% and -1.8%. More than 50% of the total drop in emissions from this sector happened between April and May, when restrictions were at their maximum. After this period, a rapid recovery is observed, emissions being only -1% below BAU by the end of the year. Fugitive emissions from fossil fuel production and transportation show decreases of up to -10% for NMVOC and -6.7% for CH4. Finally, 875 the decrease of NMVOC emissions from use of solvents is very limited (-1.3%) as only metal cleaning and printing industrial activities were considered to be affected by COVID-19 restrictions. The comparison between the charts produced for NOx and PM2.5 allows understanding the heterogeneous 890 impact of COVID-19 across pollutants presented in Sect. 4.1.1. As shown in the charts, these differences are mainly due to the fact that total emission changes were primarily driven by changes in road transport and other stationary combustion activities and the contribution of these two sectors to total emissions of each pollutant. In the case of NOx, road transport is the largest contributor to total emissions and therefore the drop in total emissions is significant, while in the case of PM2.5 the main contributor to total emissions 895 is other stationary combustion activities, which were practically not affected by the COVID-19 restrictions. As a matter of fact, more than 70% of the total drop in NOx emissions occurred at the EU27 + UK level comes from the road transport sector.  -10% below and 10% above BAU levels. These results are in line with the fact that during this period mobility restrictions were lifted and traffic activity reached values above BAU levels due to the increase of domestic tourism. The drop in emissions observed during November and associated to the second round of nationwide COVID-19 restrictions show relative changes between ∼-26.5% and ∼-7.2%, which are approximately two times lower than the ones observed during the first round of lockdowns. In the case of 920 Germany, the relative changes during the lockdowns of Spring ranged approximately between −40 % (p95) and −10 % (p05). Similar to what is observed for Italy, during summer the relative decline of NOx emissions is considerably reduced, ranging between -10% and 5% below and above BAU levels, respectively. A second significant drop in emissions is observed during the second half of December, when Germany had to go into a new hard lockdown as the number of deaths and infections from COVID-925 19 reached record levels. During this period of time, average emission reductions reached values of between -4.5% (p05) and up to -24.5% (p95). As in the case of Italy, the reductions associated to the second round of restrictions is approximately two times lower than the ones observed during the Spring wave. domain between urban and rural areas was derived from the Global Human Settlement Layer (GHSL) project (Pesaresi et al., 2019). The decline of NOx urban emissions was in average 3.4 times larger than 935 the one obtained for NMVOC (i.e., -11.3% versus -3.3%). These results coincide with the general increase of O3 levels in urban areas observed during the spring COVID-19 lockdowns, which is attributed to the fact that the O3 production is largely VOC-sensitive across European urban areas (Grange et al., 2021;Querol et al., 2021). The largest differences between the NOx and NMVOC emission declines was found in Spain (-15.6% versus -3.1%) and Portugal (-17.1% versus -3.9%). These results are in line with the 940 relative changes in O3 concentrations in traffic stations reported by Grange et al. (2021), which show that the largest O3 increases occurred in Spain (61.9%) and Portugal (46.8%).

Data availability
Emission adjustment factors per country-, day of the year, sector and pollutant are provided in an Excel file through the CAMS Document Repository (https://doi.org/10.24380/k966-3957, Guevara et al., 2022). 945 The CAMS-REG_v5.1 BAU 2020 gridded emission inventory (https://doi.org/10.24380/eptm-kn40, Kuenen et al., 2022) is distributed as NetCDF (Network Common Data Format) files from the Emissions of atmospheric Compounds of Ancillary Data (ECCAD) system, which will be complemented with access through the ECMWF Atmosphere Data Store (ADS) as soon as this is technically feasible. For review purposes, ECCAD has set up an anonymous repository where a sample of the emission file can be 950 accessed directly (eccad.aeris-data.fr/essd-surf-emis-cams-reg/).

Conclusions
We present a dataset of daily sector-, country-and pollutant-dependent emission adjustment factors that allows quantifying the impact of the COVID-19 restrictions on European primary emissions of criteria pollutants and greenhouse gases for 2020. The dataset was constructed considering changes observed in 955 metrics traditionally used to estimate emissions, such as energy statistics or traffic counts, as well as information derived from new mobility indicators, meteorological data and machine learning techniques.
The resulting dataset allows reflecting the heterogeneous impact of COVID-19 restrictions across countries on air pollutants and greenhouse gases levels for a total of nine anthropogenic activity sectors, including road transport, energy industry, manufacturing industry, residential and commercial 960 combustion, aviation, shipping, off-road transport, use of solvents and fugitive emissions from transportation and distribution of fossil fuels. To the authors knowledge, this is currently the most comprehensive and complete European dataset for inferring changes in primary emissions derived from the COVID-19 restrictions. It is worth noting the intercomparison exercise performed between observed changes in traffic activity derived from governmental traffic flow data and from the Google mobility 965 trends, the latter being widely used in the current literature. Results indicate large deviations between novel Google mobility and traditional traffic flow data, which in the present work were reduced by constructing a set of adjustment factors to better reflect changes in emissions from light-duty and heavyduty vehicles.

970
We combined the resulting COVID-19 adjustment factors with the European CAMS-REG gridded (0.1 x 0.05 deg) emission inventory for 2020 following a business-as-usual (BAU) scenario, to spatially and temporally quantify reductions in emissions from both criteria pollutants and greenhouse gases. The main findings and conclusions are as follows:

975
• The largest decrease in European emissions in 2020 attributed to the COVID-19 lockdown measures were found for NOx (-10.5%) and CO2 fossil fuel (-7.8%) emissions. For these two pollutants, the most pronounced drop in emissions was found during April (-32.8% and -25.5%) when the mobility restrictions were at their maximum.
• By the end of summer, the effect of COVID-19 measures on emissions diminished as lockdown 980 restrictions relaxed, and emissions remained at values of -4.8% and -3.9% below business-asusual levels for NOx and CO2_ff.
• The emission reductions observed during the second epidemic wave (October, November and December) were between three and four times lower than those occurred during the Spring lockdowns, up to -10.5% for NOx and -6.5% for CO2 fossil fuel, since mobility restrictions were 985 https://doi.org/10.5194/essd-2022-31 generally softer and only had to be toughened in those countries affected by increasing rates of transmission such as France, Germany or the UK.
• Lower drops in emissions were found for PM10 and PM2.5 (-3.0% and -2.1%), as these were modulated by residential combustion activities, which slightly increased during the lockdowns.
• At the country level, the largest relative emission declines were reported for Italy, UK, Spain and France: between -15.1% and -13.5% for NOx and -11.4% and -10.4% for CO2 fossil fuel emissions.
• At the sectoral level, the largest emission declines were found for aviation (between -51 and -995 56%) and road transport (between -15.5% and -18.8%). A drop of similar intensity was observed for both sectors at the beginning of the pandemic. However, while aviation emissions remained almost unchanged, road transport started to gradually recover during late April and the beginning of May, and they reached values of around -5% below BAU by the end of September. A decrease ~50% lower than in April was observed during the second epidemic wave. 1000 • For the other stationary combustion activities, the pollutants that are mainly related to residential wood combustion processes (i.e., PM10, PM2.5, NH3, NMVOC, CO, CO2_bf and CH4) experienced a slight increase (between 1.1% and 1.7%), while the rest of pollutants (i.e., NOx, SO2 and CO2_ff) showed a modest decrease (between -0.4% and -2.9%). Similarly, for the manufacturing industry a heterogenous impact of the COVID-19 restrictions is observed across pollutants: a lower 1005 reduction is observed for NMVOC and NH3 (between -2.8% and -3.5%) when compared to the other pollutants (between -6.8% and -7.2%) as these two are mostly driven by processes occurring in the food/beverage and chemistry industries, which were considered to be essential during the • The largest contributions to the EU27 + UK decrease in emissions comes from the road transport sector for the majority of pollutants: up to 70.5% for NOx emissions.
• In terms of spatial analysis, the largest emissions reductions occurred in urban areas and main 1015 interurban roads. Isolated and significant emission drops were also observed where large point sources are located. The decline in NOx urban emissions was in average 3.4 times larger than the one obtained for NMVOC (-11.3% versus -3.3%).

Limitations of the dataset
The collection of COVID-19 emission adjustment factors and the CAMS-REG_v5.1 2020 BAU inventory 1020 have been produced using state-of-the-art information and methods in support of air quality modelling studies. There exist, however, some limitations associated with the current version of the datasets that users should be aware of: • The emission adjustment factors do not take into account potential variations within each country. 1025 This includes, for instance, the heterogeneous lockdown easing process across the different administration units, which may entail heterogeneous recovery rates of the road transport emissions. Similarly, within sea-regions the drop in passenger ship movements (e.g., cruise) during 2020 compared to 2019 has been significantly larger than the one observed for cargo ship movements. This fact implies that the COVID-19 impact on shipping emissions may not only vary 1030 per sea region, but also (and more significantly) per ship route.
• The current factors do not consider the potential impact on NMVOC emissions from residential use of solvents derived from the increase on the consumption of the so-called pandemic products such as hand sanitizers. In the present work, we only assessed the impact of COVID-19 on industrial use of solvents due to the lack of more detailed data. 1035 • The methodology developed to calculate CAMS-REG gridded emissions for recent years has been validated against reported emissions and shows good results for most sectors and pollutants. The activity data captures a lot of the year-to-year variability, except sudden changes due to, e.g., the closing of a power plant. However, to get a BAU inventory we altered the methodology by ignoring all activity data that may see an impact from the COVID-19 restrictions. This means that, 1040 besides the COVID-19 impact, part of the normal year-to-year variability may also be lacking.

Future perspective
Despite the aforementioned limitations, we believe that these emission datasets will allow to refine our understanding of concentration changes observed by satellite and in situ observations, and pinpoint the effect of COVID-19-related measures more precisely. It will also allow accurate estimates of how far 1045 these temporary concentration changes have improved air quality and lowered the related morbidity and mortality. The results reported by Badia et al., (2021), Barré et al., (2021), Guevara et al., (2021) and Schneider et al. (2021), among others, which have made use of previous versions of the emission adjustment factors dataset presented in this work, are proof of that. In this sense, future works will include using the resulting emission datasets to extend current air quality simulations to the whole year 2020. We 1050 also expect to perform intercomparisons of our estimated emission changes against results reported by other existing datasets (e.g., Doumbia et al., 2021;Liu et al., 2020;Forster et al., 2020) as well as the 2020 national official reported emissions when available. This intercomparison exercise will allow us, on the one hand, to assess the consistency between emission results and, on the other hand, to compare and contrast emission results derived from traditional estimation methodologies used for official reporting 1055 against new methods that make use of mobility data sets and other types of near-real time information.
Finally, future works will also investigate the potential temporal extension of the emission adjustments factors to 2021, to include the effect of the restrictions and hard lockdowns that were still in place in specific countries such as UK or Germany during Winter time and that may have an effect on main modes of transport including road traffic or aviation.   contributions from all co-authors.

Competing interests
The authors declare that they have no conflict of interest.

Acknowledgements
The research leading to these results has received funding from the Copernicus Atmosphere Monitoring 1080 Service (CAMS), which is implemented by the European Centre for Medium-Range Weather Forecasts (ECMWF) on behalf of the European Commission. We acknowledge support from the Ministerio de  Autostrady: Impact of SARS-CoV-2 and COVID-19 on the activity of Stalexport Autostrady S.A. Capital Group. Available at: https://www.stalexport-autostrady.pl/en/company/news (last accessed, May 2021)

Figure 4 Comparison of traffic movement trends derived from Google COVID-19 Community Mobility Reports (Google LLC, 2021)
and measured traffic counts for selected countries (see Table A1 for references), the latter one being distinguished by type of vehicle (i.e. heavy duty vehicles, HDV; light duty vehicles and cars, LDV), for the period 21 February to 31 December 2020.

1430
Mobility Reports (Google LLC, 2021) and light duty vehicles and cars (LDV) measured-based trends per country (see Table A1 for references).