key: cord-1020360-6yu55guc authors: Nicolini, Giacomo; Antoniella, Gabriele; Carotenuto, Federico; Christen, Andreas; Ciais, Philippe; Feigenwinter, Christian; Gioli, Beniamino; Stagakis, Stavros; Velasco, Erik; Vogt, Roland; Ward, Helen C.; Barlow, Janet; Chrysoulakis, Nektarios; Duce, Pierpaolo; Graus, Martin; Helfter, Carole; Heusinkveld, Bert; Järvi, Leena; Karl, Thomas; Marras, Serena; Masson, Valéry; Matthews, Bradley; Meier, Fred; Nemitz, Eiko; Sabbatini, Simone; Scherer, Dieter; Schume, Helmut; Sirca, Costantino; Steeneveld, Gert-Jan; Vagnoli, Carolina; Wang, Yilong; Zaldei, Alessandro; Zheng, Bo; Papale, Dario title: Direct observations of CO2 emission reductions due to COVID-19 lockdown across European urban districts date: 2022-03-19 journal: Sci Total Environ DOI: 10.1016/j.scitotenv.2022.154662 sha: e0e38d26c1d54b80b0685484db9e38dbb7ec23de doc_id: 1020360 cord_uid: 6yu55guc The measures taken to contain the spread of COVID-19 in 2020 included restrictions of people's mobility and reductions in economic activities. These drastic changes in daily life, enforced through national lockdowns, led to abrupt reductions of anthropogenic CO2 emissions in urbanized areas all over the world. To examine the effect of social restrictions on local emissions of CO2, we analysed district level CO2 fluxes measured by the eddy-covariance technique from 13 stations in 11 European cities. The data span several years before the pandemic until October 2020 (six months after the pandemic began in Europe). All sites showed a reduction in CO2 emissions during the national lockdowns. The magnitude of these reductions varies in time and space, from city to city as well as between different areas of the same city. We found that, during the first lockdowns, urban CO2 emissions were cut with respect to the same period in previous years by 5% to 87% across the analysed districts, mainly as a result of limitations on mobility. However, as the restrictions were lifted in the following months, emissions quickly rebounded to their pre-COVID levels in the majority of sites. EC has been widely used over natural ecosystems to investigate biosphere responses to environmental and biological factors (Baldocchi, 2014) , while in urban ecosystems its application has grown steadily over the past 15 years (e.g. Helfter et al., 2016; Nordbo et al., 2012; Pérez-Ruiz et al., 2020; Roth et al., 2017; Salgueiro et al., 2020; Stagakis et al., 2019; Vogt et al., 2006; Ward et al., 2015) . Urban EC measurements have been shown to be valuable for detecting short-and long-term changes in fluxes, and for studying the drivers of local CO 2 emissions and 'urban metabolism' leading to a better understanding of the carbon cycle in cities Crawford et al., 2011; Feigenwinter et al., 2012; Velasco and Roth, 2010) . The EC method is based on simultaneous high-frequency (e.g. 10-20 Hz) measurements of the vertical wind velocity and CO 2 concentration, or any other scalar entity such as heat, moisture, trace gases, and aerosols, allowing for the estimation of the vertical exchange of such scalars through turbulent motions (eddies) within the mean air flow. Conventionally, positive values represent upward fluxes, i.e. net emissions to the atmosphere, and negative values represent downward fluxes, i.e. net uptake by the underlying surface. Measurements are continuous and representative of a target source area (footprint) typically of the size of a city district (i.e. 10 4 -10 8 m 2 ) depending on the measurement height with respect to average building height. For a complete description of the EC flux method see, for example, Aubinet et al. (2012) . The onset of the COVID-19 pandemic in Europe in early spring 2020, caused drastic changes to people's lives and socio-economic activity. The governmental actions taken to break the chain of disease transmission included the closure of schools and non-essential businesses, banning social gatherings, and enforcing home confinement. Such measures reduced mobility and economic activity, and inevitably impacted energy use and anthropogenic CO 2 emissions. It also displaced daytime populations from the work-place to residential areas, which likely impacted the spatial distribution of emissions associated with building energy use, as well as affecting emissions from transport. Assessments of national activity reductions combined with empirical relations to predict emissions suggested that CO 2 emissions of individual countries fell by up to 30% during the peak of the lockdowns in J o u r n a l P r e -p r o o f spring 2020 (Forster et al., 2020; Le Quéré et al., 2020; Liu et al., 2020) . Although associated CO 2 emissions reductions at the city scale are to be expected, the magnitude and variability of these reductions cannot be simply determined from national-level changes. Quantitative estimates of urban emission reductions due to COVID-19 restrictions based on atmospheric measurements have so far only been estimated for a few cities worldwide (Gualtieri et al., 2020; Lamprecht et al., 2020; Sugawara et al., 2021; Velasco, 2021; Yadav et al., 2021) . In this study we present CO 2 fluxes measured by a network of 13 EC stations in 11 European cities. These datasets span several years before the pandemic until October 2020. Urban EC stations operating before and during 2020 present a unique opportunity to investigate how the drastic perturbations in human activity caused by the COVID-19 pandemic have impacted local CO 2 emissions. CO 2 flux data at half-hourly resolution allows for temporal changes in CO 2 emissions to be tracked both during the initial lockdown period and during the subsequent recovery phase when economic activities and mobility gradually resumed. Our analysis focuses on the following questions: • Do direct flux measurements confirm the emissions reduction predicted by coarsescale inventory models? • What was the magnitude of the reduction in emissions at the district scale, and how does this vary from place to place? • Were reductions in emissions related to the stringency of the restriction measures? • Were there more substantial reductions for certain hours of the day or days of the week? • Did emissions return to previous levels after the restrictions were lifted? • Was this recovery dependent on urban features (e.g. land use type, density of the road network, amount of vegetation)? J o u r n a l P r e -p r o o f Using micrometeorological data from 13 urban EC stations in 11 cities across Europe (Tab. 1 and supplementary "Study sites" Section), we evaluated district-scale changes in urban CO 2 fluxes between 2020 and previous years. These changes were analysed in relation to the stringency of the local lockdown rules, taking into consideration the characteristics of each site in terms of local citizens activities (e.g. commuting, economic activities, domestic heating) and urban features. Each EC station (Tab. 1) was equipped with a 3D ultrasonic anemometer, a gas analyser for measuring CO 2 concentrations, and meteorological sensors measuring air temperature and humidity, air pressure and solar radiation. All systems collected data at 10 or 20 Hz, which were processed by the researchers in charge of each station according to commonly accepted procedures (Aubinet et al., 2012) to obtain the half-hourly CO 2 fluxes used in the analysis. Data quality assurance was assessed by each group following standard quality control and filtering procedures, albeit with allowances for site-to-site variations (details can be found in the main references for each site which are given in Tab. 1). For this study, we additionally excluded CO 2 flux data beyond the physically plausible range of -50 to 200 µmol m -2 s -1 , those from wind sectors prone to flow disturbance by physical obstacles (e.g. measurement tower structure) and those outside a site-specific quantile range calculated over a 3-weeks moving window using the 0.5% and the 99.9% probability were excluded. This quantile range was taken asymmetrically because in urban environments sporadic large negative fluxes (sinks) are far less likely than large positive fluxes (emissions). Data were not gap-filled as it was not required for the type of analysis performed. Flux observations from previous years up to and including 2019 were used as a reference to compare changes before, during, and after the lockdown periods with related restrictions that affected each city district. Pre-2020 records span between 2 and 14 years depending on the station (Tab. S1). We focused our analysis on four distinct periods defined following the Oxford COVID-19 Government Response Tracker (OxCGRT) Stringency Index (SI) (Hale et J o u r n a l P r e -p r o o f Journal Pre-proof al., 2020a, 2020b) . The SI quantifies policies that governments have taken to respond to the pandemic; it is calculated as the average of a set of macro-indicators of containment and closure policies including closure of schools, universities and workplaces, cancelling of public events, limits on private gatherings, shutting-down public transport, orders to "shelterin-place" or otherwise confine to homes, restrictions on internal movement between cities/regions, and restrictions on international travel. The SI ranges between 0 (no restrictions) and 100 (maximum level of restrictions). Although the contributing factors to this index vary between countries and cities, the SI has been shown to provide global insight into the pandemic's evolution and implementation of measures (Cross et al., 2020) . We set a minimum threshold of 65 (64.3 is the 60 th quantile of the SI values for the analysed cities over the period January -October 2020) to define the lockdown period (LOCK) in each city. The length of this period (and the maximum value of SI reached) varied from 19 days in Helsinki to 75 days in Basel and Amsterdam (Tab. S2). Then, a pre-pandemic period (PRE) was defined lasting from January 1 st to the beginning of the lockdown in each location, and two subsequent periods of 60 days each, POST1 and POST2, were identified after LOCK to evaluate the emissions recovery (Tab. S2). The anomalies in CO 2 fluxes during each of these periods were quantified in terms of the relative flux changes (RFC, %) computed as: where x 2020 and x base are the average fluxes observed for each period in 2020 and for the corresponding period in previous years (considered as the baseline period), respectively. The computation of RFC was based on daily means (Sect. 3.1 and 3.2), diel cycles (Sect. We also calculated the relative air temperature change in a similar way (RTC, %) to evaluate the potential effect of temperature anomalies on the observed fluxes. We assumed that CO 2 J o u r n a l P r e -p r o o f emissions from commercial and domestic heating become relevant when the daily mean air temperature is below 15 °C. This threshold is considered as the temperature at which heating in Europe is expected to be switched on (Matzarakis and Balafoutis, 2004; Pigeon et al., 2007) . Uncertainties in CO 2 flux averages, both daily and half-hourly in case of diel cycles, were calculated as the standard error over the single half-hourly values. Non-parametric statistical metrics (Spearman's rank correlation, Kruskal-Wallis test by rank and Wilcoxon rank-sum test) were performed to evaluate the significance of differences in the CO 2 fluxes, and the correlation of RFC with RTC and SI. To evaluate the contributions to RFC from different land cover categories and emission sources within the footprint of each tower, CO 2 fluxes were also analysed by wind sectors (see also the supplementary "Spatial analysis" Section). The land use and land cover (LULC) information was extracted from the European Urban Atlas (UA) 201 2 database (Montero et al., 2014) , cropping circular areas centred at each flux tower. The radii of these areas were set equal to the median distance of the 70 th percentile of the estimated contribution of the cumulative flux footprint (Kljun et al., 2015) , irrespective of wind direction. This length represents the distance within which 70% of the measured flux is estimated to originate. It was calculated for 8 sites at which footprint estimates were available whereas, for the other 5 sites, it was estimated according to an average ratio between the calculated distances and measurement heights (Tab. S1). To facilitate the interpretation of results, similar UA-LULC classes were combined into broader classes, the main four of which were: (i) predominantly residential areas (RES), consisting mainly of residential structures, but also including downtown areas and city centres (higher storeys of buildings are mostly residential), with various degrees of soil sealing; (ii) non-residential areas (nRES), consisting in industrial and commercial areas, as well as schools and military units; (iii) areas dominated by roads and railway networks (ROD); and (iv) green urban areas (GUA), that include pervious surfaces with vegetation e.g. lawns, parks, greenbelts, farmland, urban forests. We used Google Earth Imagery (© 2020 J o u r n a l P r e -p r o o f Google), Sentinel-2 satellite Normalised Difference Vegetation Index (NDVI), and local knowledge of each site, to improve the characterization of districts. At each site, relatively homogeneous wind sectors were associated with single LULC class (supplementary "Spatial analysis" section). The NDVI data were used to determine the extent of green spaces and to track the temporal dynamics in vegetation activity, to minimise the risk of misinterpreting trends in RFC arising from year-to-year variations (supplementary "Vegetation analysis" section). Two independent datasets of city-scale activity, the Carbon Monitor emission inventory (CM, https://carbonmonitor.org/) and Google COVID-19 Community Mobility Reports (©2021Google), were used to assist interpretation of the RFC at each site. We estimated CO 2 daily emissions from road transportation at the city scale for 2019 and 2020 following the CM methodology (Liu et al., 2020) : a sigmoid function describing the relationship between daily mean congestion level (as reported by TomTom's traffic report, https://www.tomtom.com) and mean traffic volume data for the city of Paris was used as a proxy to estimate the road-traffic related CO 2 emissions from the other cities using the TomTom's reports for each city. Then, we compared the changes between measured and predicted emissions for residential and non-residential sectors with a substantial presence of roads (20% as minimum). Community mobility data released by Google during the pandemic (Google LLC., 2020) were used to track changes in people's mobility, specifically as a measure of the time spent by the population at places of residence. This data source has already been used to estimate changes in CO 2 emissions at the neighbourhood scale (Velasco, 2021) . Mobility levels during the LOCK, POST1 and POST2 periods were compared to baseline mobility values calculated as median values of each day of the week over the five weeks between January 3rd and February 6th 2020. J o u r n a l P r e -p r o o f 3. Results For all sites, we found a clear reduction in CO 2 emissions coinciding with COVID-19 restrictions ( Fig. 1) , with daily RFC values mostly spanning between -%5 to -87%. Compared to the same period in previous years, the observed reduction during the LOCK period in which the most restrictive measures were applied, was statistically significant at all sites The dynamics of the RFC roughly followed those of the SI at all sites, with stronger emission reductions (negative RFC) observed during more stringent confinement periods. Over the whole monitored period, the correlation between RFC and SI was statistically significant (see Tab Since the lockdown period began in early spring, CO 2 emissions from building heating contributed to total CO 2 fluxes at most sites. As a result, synoptic variations in weather patterns and associated temperature changes impact CO 2 emissions and are responsible for some of the variability in the RFC. For relatively cold days (mean air temperature < 15°C) without strong restrictions (SI < 40), a negative correlation was found between RFC and RTC, in particular at CH-Basel-A, DE-Berlin-ROTH and NL-Amsterdam (Spearman ⍴ correlation coefficient between -0.33 and -0.44, ɑ = 0.05), probably due to a reduction in heating-related emissions. This aspect is further analysed in Section 3.3. All cities had a significant reduction in emissions during the most restrictive measures (LOCK, SI > 65), and several cities also reported reduced emissions in the subsequent months (POST1 and POST2) during which some restrictive measures remained in place. The relation between RFC and SI is examined in more detail in figure 2 (the statistical significance is shown in Fig. S9 ). In almost all cities the RFC reached its minimum (< -50%) for the highest values (SI > 70-80), when the most restrictive measures were in place. There were however two districts where emissions were only slightly reduced even under When analysing all districts together (Fig. 3) , a consistent and significant correlation between RFC and SI was found (slope = -0.75, p < 0.001). The average RFC during the LOCK period across all sites was -36%, and decreased to -17% and -7% in the POST1 and POST2 periods, respectively (star symbols in Fig. 3 ). In the period before the restriction or with SI < 20 (PRE, grey dots and stars in Fig. 3) , the average RFC was not significantly different from zero. Journal Pre-proof Fig. S1 ). This likely caused a decrease in the use of domestic heating of which about 70% is by fossil fuel in Innsbruck and 25% in Basel. As a result, CO 2 emissions were reduced, in particular in the afternoon (Fig. 4 , PRE column). In Heraklion, the temperature during this period was close to normal (supplementary "Air temperature analysis" section) and heating for this district is mainly from electricity. The reduction seen in early 2020 affected both the morning and afternoon peaks (Fig. 4 LOCK column) , and this could be due to changes in the traffic regulations imposed by the city authorities in the city centre (Politakos et al., 2020) . By comparing emissions from paired EC observations in residential and non-residential areas of the same city, it is possible to infer qualitative information on the dominant driver. For most of these urban sites, the observed CO 2 fluxes are positive throughout the day and across the January to October period considered (Fig. 4) . However, net CO 2 uptake was periods (PRE, LOCK, POST1, POST2). As with RFC data, the daily city-scale emission estimates were smoothed using a 7-day rolling mean. CO 2 emissions did not increase significantly in residential areas (with less than 20% of road cover) during the LOCK period, despite a mean 20% increase in time spent at home according to Google mobility data, and some instances of net reductions in emissions were recorded. This suggests that vehicular traffic is the main factor driving CO 2 fluxes in the monitored districts, with emissions from heating, cooking and human respiration playing a lesser role. The negative RFC in the three districts with less than 10% of roads during the PRE period (Fig. 6e , largest dots) could instead respond to a reduction in heating emissions due to anomaly warmer weather in these cities (e.g. Basel, Berlin, Innsbruck, see also Fig.1 red lines and Fig. S1 ). Direct CO 2 flux measurements from 13 eddy covariance (EC) stations across Europe reveal the effect of the COVID-19 pandemic restrictions on district level CO 2 emissions. At all sites, CO 2 emissions were significantly reduced during the strictest lockdown measures. In contrast to other approaches, the fine temporal resolution of EC data allows the evolution of the emissions to be analysed (at sub-daily to seasonal time-scales) and the changing response to restrictions to be quantified. For most sites CO 2 emissions returned to prepandemic levels by autumn of 2020. The emission reductions occurred mainly during daytime, principally as a consequence of limitations on mobility, and particularly reductions in vehicular traffic. In contrast, emissions related to home confinement (heating, cooking, human metabolism) did not increase enough to compensate for the reductions in emissions J o u r n a l P r e -p r o o f Journal Pre-proof from road traffic; this was true in all neighbourhoods studied, even in the more residential ones where the workforce was displaced during lockdown periods. The substantial emission reduction recorded through the first COVID-19 pandemic wave was temporary in most of the city districts and emissions rebounded to previous levels once restrictions were eased in the following months. The speed and extent of the emission recovery varied from district to district, with the fastest and most complete recovery seen mainly in the non-residential areas and attributed to re-established vehicular traffic. This study demonstrates that the EC method is a valuable tool for monitoring continuously and almost in real-time the short-and long-term changes in urban trace gas emissions, and, potentially, for assessing the effectiveness of climate change mitigation policies (for example limiting traffic emissions versus reducing building heating demand). Of great importance to this aim is the availability of auxiliary data such as detailed traffic data, inventories of emission sources related to human activities, and data on city urban features. This study highlights the additional advantages of monitoring networks, where data collected from individual stations can be synergistically combined for long-term monitoring activities such as the Integrated Carbon Observation System (ICOS, www.icos-ri.eu). Our results demonstrate that altering human behaviour has a direct, immediate and significant effect on the reduction and recovery of urban CO 2 emissions. The temporary nature of the observed emission reductions emphasises the need to implement systemic changes in the city ecosystem and people's lifestyles to achieve effective and sustained climate change mitigation. To reach the target of climate-neutrality in 2050, cities need to take action across multiple sectors but, according to our data, this must include interventions on private and public mobility aimed at reducing associated emissions. Other data that support the findings of this study are available at the following links: J o u r n a l P r e -p r o o f (Ny, excluding the 2020 and years with no data for the period of concern). Average RFC values are those displayed in Fig. 3 . RFC values are ranked by colours: from red to green going from higher emission reductions (negative RFC) to higher emission increases (positive RFC). Eddy Covariance: a practical guide to measurement and data analysis Measuring fluxes of trace gases and energy between ecosystems and the atmosphere -the state and future of the eddy covariance method Validation of modeled carbon-dioxide emissions from an urban neighborhood with direct eddy-covariance measurements Five years of carbon dioxide fluxes measurements in a highly vegetated suburban area Trading health for wealth: The effect of COVID-19 response stringency Eddy Covariance Measurements Over Urban Areas, in: Eddy Covariance: A Practical Guide to Measurement and Data Analysis Current and future global climate impacts resulting from COVID-19 Methane and carbon dioxide fluxes and source partitioning in urban areas: the case study of Inclusion of CO2 flux modelling in an urban canopy layer model and an evaluation over an old European city centre Quantifying road traffic impact on air quality in urban areas: A Covid19-induced lockdown analysis in Italy The Hestia fossil fuel CO2 emissions data product for the Los Angeles megacity (Hestia-LA) Variation in government responses to COVID-19, Version 9 Controls of carbon dioxide concentrations and fluxes above central London Spatial and temporal variability of urban fluxes of methane, carbon monoxide and carbon dioxide above London The urban measurement station SMEAR II: Continuous monitoring of air pollution and surface-atmosphere interactions in Helsinki Seasonal and annual variation of carbon dioxide surface fluxes in Helsinki A simple two-dimensional parameterisation for Flux Footprint Predictions FFP Decoupling of urban CO2 and air pollutant emission reductions during the European SARS-CoV2 lockdown Temporary reduction in daily global CO2 emissions during the COVID-19 forced confinement Variability of CO2 concentrations and fluxes in and above an urban street canyon On the controlling factors for the variability of carbon dioxide flux in a heterogeneous urban environment Near-real-time monitoring of global CO2 emissions reveals the effects of the COVID-19 pandemic Carbon Dioxide Emissions of the City Tall tower eddy covariance measurements of CO2 fluxes in Vienna Heating degree-days over Greece as an index of energy consumption Land Use and Land Cover Mapping in Europe. Remote Sensing and Digital Image Processing Revised eddy covariance flux calculation methodologies -effect on urban energy balance Urban Climates Urban land cover type determines the sensitivity of carbon dioxide fluxes to precipitation in Phoenix, Arizona Anthropogenic heat release in an old European agglomeration Inter-annual variability of Eddy Covariance CO₂ flux measurements in the city center of Heraklion R: A language and environment for statistical computing. R Foundation for Statistical Computing Effects of changing population or density on urban carbon dioxide emissions Multi-year energy balance and carbon dioxide fluxes over a residential neighbourhood in a tropical city RStudio: Integrated Development Environment for Annual and seasonal variability of greenhouse gases fluxes over coastal urban and suburban areas in Portugal: Measurements and source partitioning Eddy Covariance measurements and source partitioning of CO2 emissions in an urban J o u r n a l P r e -p r o o f Journal Pre-proof environment: Application for Heraklion The first 1-year-long estimate of the Paris region fossil fuel CO2 emissions based on atmospheric inversion Observing the surface radiation and energy balance, carbon dioxide and methane fluxes over the city centre of Amsterdam Anthropogenic CO2 Emissions Changes in an Urban Area of Tokyo, Japan, Due to the COVID-19 Pandemic: A Case Study During the State of Emergency Impact of Singapore's COVID-19 confinement on atmospheric CO2 fluxes at neighborhood scale. Urban Clim Cities as net sources of CO2: Review of atmospheric CO2 exchange in urban environments measured by eddy covariance technique Temporal dynamics of CO2 fluxes and profiles over a Modeling urban evapotranspiration using remote sensing, flux footprints, and artificial intelligence Effects of urban density on carbon dioxide exchanges: Observations of dense urban, suburban and woodland areas of southern England The Impact of COVID-19 on CO2 Emissions in the Los Angeles and Washington DC/Baltimore Metropolitan Areas USA-1 Greece GR-Heraklion 35 EC system instruments: CSAT3, 3-D sonic anemometer EC155, closed-path /H2O gas analyser WindMaster Pro 3-axis anemometer (Gill Instruments Limited) LI-7500 open path CO2/H2O analyser HS-100, HS-100 horizontal-head ultrasonic anemometer (Gill Instruments Limited) Irgason integrated CO2/H2O open-path gas analyser and 3-D sonic anemometer; USA-1, USA-1 ultrasonic anemometer LI-7200 enclosed-path CO2/H2O gas analyser HS-100, HS-100 horizontal-head ultrasonic anemometer (Gill Instruments Limited) Picarro 1301-f flux CO2, CH4, and H2O gas concentration analyser Code availability Data in this study were analysed using R (R Core Team, 2019) version 3.6.2 (2019-12-12, Dark and Stormy Night) and RStudio (RStudio Team, 2021) version 1.4.1717 (Juliet Rose).The custom code used for the analysis is available at: https://github.com/Giacomo-Nicolini/UrbanFluxes_COVID19. Table 1 . Summary of the cities and respective EC stations involved in the study. The station ID is the naming used in this analysis, containing the names of the respective cities, z is the measurement height (m), z/zh is the ratio between z and mean building height (zh). Station