Instruments for measuring fluxes of urban air pollutants and greenhouse gases are situated in a small laboratory atop the BT Tower located in central London, UK (51∘31′17.4′′ N, 0∘8′20.04′′ W). The measurement height is 190 m above street level, with a mean building height of 8.8±3.0 m in the 10 km radius surrounding the tower . The gas inlet and ultrasonic anemometer are attached to a solid mast that extends 3 m above the top of the tower. Air is pumped down a 45 m Teflon tube (3/8′′ OD) in a turbulent flow of 20–25 L min-1 to the gas instruments, which are situated in a small air-conditioned room inside the tower on the 35th floor.

Long-term measurements of NO and NO2 fluxes began in September 2020 with data presented here up to September 2021. Data are compared to previous measurements made by from March–August 2017. Both chemical species were measured using a dual-channel chemiluminescence analyser (Air Quality Design Inc., Boulder Colorado, USA; 5 Hz), as described previously by . The number of photons measured by the photomultiplier tube was converted into a part-per-trillion (ppt) mixing ratio using a five-point calibration curve produced through dilutions of a 5 ppm NO in N2 calibration standard (BOC Ltd., UK; traceable to the scale of the UK National Physical Laboratory, NPL) into NOx-free air (generated from an external Sofnofil and activated charcoal trap). NO2 was calculated by conversion of NO2 into NO using a photolytic blue-light converter (BLC). Here, both NO and NO2 were measured, from which NO2 can be quantified by subtracting the NO mixing ratio and applying a correction factor for the conversion efficiency of the BLC. The instrument was calibrated every 37 h in addition to an hourly zero measurement to subtract the temperature-dependent background signal of each channel. The uncertainty of the NO measurement is given as ±3 %, resulting from uncertainties in the sample mass flow controller, calibration gas mass flow controller and calibration gas certification. The uncertainty for the NO2 measurement is given as ±4.7 % due to the additional uncertainty in the conversion efficiency calculation, determined in the laboratory via variation in repeated tests. The precision for each channel is calculated as 53 and 184 ppt for NO and NO2 respectively from the standard deviation in all the hourly zeros during the measurement period.

Long-term measurements of CO2 have been ongoing at the BT Tower since 2011 as part of UKCEH's National Capability programme. Dry mass fractions (1σ precision < 300 ppb) were measured initially using a cavity ringdown spectrometer (Model 1301-f, Picarro Inc., Santa Clara, California, USA; 10 Hz) as described by . Unfortunately, instrumental failure means data are not available between February and June 2021, after which a closed-path infrared gas analyser (Li-7000, LI-COR Environmental, Lincoln, Nebraska, USA; 10 Hz) took over the CO2 flux measurement.

Meteorological measurements were made at the BT Tower as described by . Wind speed, wind direction and sonic temperature were measured using an ultrasonic anemometer (Gill R3-50, Gill Instruments, Lymington, UK; 20 Hz) along with pressure and relative humidity measurements using a weather station (WXT520, Vaisala Corp. Helsinki, Finland; 1 Hz). For ease of processing, each chemical species is logged separately at its maximum measurement frequency into a file with the sonic-anemometer data averaged to the same measurement frequency.

The flux, F, is defined in this context as the vertical transport of a chemical species per unit area per unit time. Hourly fluxes were calculated using eddy covariance theory as described by Eq. (), where F is equal to the covariance between the instantaneous change in vertical wind speed, w′, and the instantaneous change in species concentration, c′, averaged over the hour. 1F=w′c′‾ Eddy covariance calculations were performed using the modular software packages in eddy4R, adopting the same processing settings described in , as adapted from . This was to allow a direct comparison to be made to the previous measurements made in 2017. The lag time correction was determined by maximisation of the cross-covariance between the pollutant concentration and the vertical wind component, with an additional application of a high-pass filter which improves the precision of the determined lag time by an order of magnitude . This resulted in median lag times of 7.2 s for NO, 7.6 s for NO2 and 21 s for CO2. Data were filtered such that the friction velocity (u*) is >0.2 to ensure sufficiently developed turbulence and using eddy4R's quality control flagging scheme. The QA/QC process is described in detail by and . Data are flagged as either valid or invalid based on the combination of individual flags for input data validation, homogeneity and stationarity, and development of turbulence.

A parameterised version of the backwards Lagrangian stochastic particle dispersion model implemented in eddy4R was used to estimate the footprint for each hourly flux measurement at the BT Tower. The model is described by and has been parameterised for a range of meteorological conditions and receptor heights to reduce the computational expense of running it. The original model aims to produce a cross-wind-integrated footprint function as a function of its along-wind distance, which has now been further extended into two dimensions using a Gaussian distribution driven by the standard deviation in the cross-wind component . Meteorology statistics from the eddy covariance calculations are used in combination with modelled boundary layer height from ERA5 and a surface roughness length of 1.1 m to produce a weighted matrix of 100×100 m grid cells. Each output-weighted matrix was then scaled and aligned to an appropriate coordinate reference system to allow each matrix to be plotted onto a map.

Hourly traffic loads surrounding the BT Tower were calculated by summing the traffic load from each of the 24 automatic traffic counters (ATCs) within the flux footprint, as shown in Fig. . This gave an indication of the magnitude of traffic load for both measurement periods and allowed relative changes to be studied between the two years. In addition, daily vehicle length breakdown was examined, from which vehicles were separated into three length classes: <5.2 m, indicating the number of passenger cars; 5.2–12 m, indicating the number of vans and rigid lorries; and >12 m, indicating the number of buses and articulated lorries. As the LAEI estimates that almost all lorry emissions in central London are due to the rigid class, the >12 m class is assumed to solely be made up of buses. Data were provided by the Operational Analysis Department, Transport for London (TFL) via a freedom-of-information request .

The London Atmospheric Emissions Inventory (LAEI) is a spatially disaggregated 1 km2 gridded map of the annual emissions of various air pollutants for the London area up to the M25 motorway ring road . Annual emissions are estimated using emission factors and activity factors for the different sources. For example, emissions from domestic combustion in tonnes per year would be calculated as gas consumption (GW h per year) × emission factor (tonnes of pollutant per GW h). The inventory is produced roughly every 3 years by TFL and the Greater London Authority (GLA). At the time of writing, there were no inventory estimates for the pandemic-affected years of 2020 and 2021. We therefore use the most recent version produced in 2016, which relates to a “normal” year unaffected by lockdowns, to understand how emissions have changed since the 2017 measurements.

Estimated emissions from the London Atmospheric Emissions Inventory (LAEI) for the measurement footprint were calculated to aid understanding of these observations. The hourly-footprint-weighted matrix output from eddy4R was used to select the relevant areas of the LAEI. The theoretical contribution to the flux was extracted from each footprint grid cell and scaled for hour of day, day of week and month of year for each emissions sector using a set of anthropogenic scaling factors described by . An excellent agreement between the diurnal profiles and measurement footprint (shown in Fig. ) for the 2017 and 2020–2021 measurement periods was seen. This gave us confidence that any changes in emissions were not to do with sampling in different times of year or sampling in different areas of central London. The source breakdown of the 2017 inventory-generated time series for both CO2 and NOx is shown in Fig. . Emissions of both species are almost entirely made up from combustion of fossil fuels to generate heat and power in the domestic, commercial and industrial sectors and the transport sector, which is dominated by various forms of road transport. However, each species has a significantly different relative contribution from each sector. A total of 75 % of CO2 emissions are estimated to arise from heat and power generation, but only 42 % of NOx emissions are estimated to arise from the same source.

The inventory breakdown for each species and the different percentage reductions in measured emissions since 2017 were used to disentangle changes in emissions of each sector. This was done simultaneously using a number of assumptions.

highlight the excellent agreement of CO2 emissions measured at the BT Tower with those estimated by the LAEI, and thus, the source contributions of 75 % from heat and power generation and 25 % from transport for CO2 are taken as accurate here. On the other hand, previous observations have shown a significant underestimation of NOx emissions in central London . This is most likely due to an underrepresentation of road transport NOx emissions, in line with a poor representation of diesel vehicle emissions and/or congestion. Therefore, rather than using the inventory-predicted 42:58 split for heat and power generation:transport, the relative contributions were varied. Labelled as α:β, different scenarios between 50:50 and 0:100 (where α+β=100 %) were chosen to represent all possible levels of NOx underestimation. The percentage reduction in heat and power generation emissions for both species is labelled as x with the assumption that any reduction in this sector's emissions would have the same reduction in measured flux for both species. With minimal legislation for the sector introduced between 2017 and 2020–2021 and a failure to address NOx emissions from boilers in the UK's Clean Air Strategy, this assumption is considered reasonable . However, this is likely to be untrue for transport. Policy implemented between the two measurement periods specifically targeted NOx emissions, and NOx emissions are disproportionately higher in higher traffic loads due to the ineffectiveness of exhaust treatment systems in that environment. Additionally, the modernisation of the vehicle fleet will have introduced more vehicles with lower NOx/CO2 emission ratios. Therefore, the relative change in the emissions of NOx and CO2 from traffic sources may not have been the same, and different values are given here as y and y′. This information is all summarised in Table , with the two independent constraints displayed in Eq. () for CO2 and Eq. () for NOx: 7δCO2=0.75x+0.25y[32100]=20%,8δNOx=αx+βy[73100]′=73%. The different scenarios are visualised in Fig.  to aid understanding of the possible solutions. Two bounding conditions drawn as dashed lines are applied to constrain the solutions. These are as follows: (1) a reduction in transport emissions greater than 100 % is not possible, and (2) CO2 emissions from transport must have decreased by at least 32 % in line with the 32 % reduction in traffic load. In reality, CO2 emissions will have decreased by more than 32 % as a result of the fleet modernisation which has lead to a decrease in the average CO2 emissions per new vehicle registration .

Highlighted in green are all the resulting possible solutions where, crucially, to achieve the observed reductions in NOx flux, there must have been a 73 %–100 % reduction in transport NOx emissions, with transport contributing >70 % to total NOx emissions. This transport contribution percentage demonstrates the underestimation in the inventory of transport emissions, in agreement with and . However, the most interesting observation is that a 73 %–100 % decrease in transport NOx emissions is seen for only a 32 % decrease in traffic load since 2017.

When compared to concentrations, we found that NOx concentrations at Marylebone Road, a kerbside monitoring site within the flux footprint, had declined by 62 % between the two periods (248 µg m-3 vs. 95 µg m-3). This is increased to 69 % (214 µg m-3 vs. 67 µg m-3) when looking at the roadside increment concentration (roadside – urban background) as determined from Marylebone Road and London North Kensington monitoring sites. Whilst this was slightly lower in magnitude than the measured change in flux, the concentration data will have been heavily influenced by meteorology, and so some disagreement was expected. There are a number of plausible explanations for the large decrease in NOx fluxes. The introduction of the ULEZ is thought to have resulted in a pre-pandemic 31 % reduction in NOx emissions from road transport in central London ; this is likely to be an upper estimate for our measurements due to the fact of a significant proportion of our flux footprint being situated outside of the ULEZ zone. With average traffic loads between April and November 2019 after the ULEZ was introduced only being down 1.8 % on 2018 levels for the same period, the vast majority of the reduction is due to a clean-up of the fleet, which reduces the emissions per vehicle per unit distance. The remaining emissions are further reduced by 32 % due to there being 32 % less vehicles on the roads surrounding the BT tower during the pandemic. This leaves ∼20 %–45 % of unaccounted-for emissions reduction. A small portion of this unaccounted-for reduction may be due to a 40 % reduction in the number of buses on the roads surrounding the BT Tower during 2020–2021. The >12 m class of the vehicle length breakdown in Fig.  represents the bus classification. With buses making up 17 % of the total NOx emissions from road transport, the decrease in relative proportion between 2017 and 2020–2021 could result in a maximum of 7 % extra reduction in NOx emissions. It is likely to be less than 7 % due to the small increase in the relative proportion of the 5.2–12 m class.

Examining how the NOx flux correlated with traffic load for both measurement time periods gives further insight into the unaccounted emissions reduction. Figure  generally shows significantly enhanced NOx emissions in 2017 above 25 000 vehicles per hour. With road transport being the dominant source during these measurements, it is highlighting what is thought to be the effect of congestion on NOx emissions.

During periods of high congestion, increased emissions are expected due to increased length in journey time, a greater number of accelerations in the stop–start nature of traffic and the reduced effectiveness of exhaust gas NOx treatment systems in diesel vehicles at low engine temperatures . The effect of congestion on NOx emissions is highly dependent on several variables, including the fleet composition, type of exhaust treatment system and the actual level of congestion . It is thought that, for individual roads, excess emissions from congestion can be anything up to 75 % greater than those from non-congested roads . Therefore, it is thought that reducing the peak-traffic load below 25 000 vehicles per hour has had a large impact on traffic NOx emissions, more than accounting for the remaining emissions reduction.

This change in emissions is clearly seen in the spatial mapping of the NOx fluxes in Fig. . assigned the heightened emissions to the northeast of the BT Tower in Fig. a to Euston station, including not only the train station but also the large Euston bus station, taxi ranks and busy roads feeding the station. In addition, the high fluxes measured to the southwest are assigned to highly congested streets such as Oxford Street, Regent Street and Piccadilly. Both areas are associated with high traffic volumes and congestion and support the notion that road transport emissions dominated in central London in 2017. However, these areas have almost an order of magnitude smaller emissions and are barely visible for 2020–2021 when shown on the same scale. This adds additional support to the conclusion that reduced traffic load and thus reduced congestion in 2020–2021 have been a major cause of the reduced NOx flux.

The spatial map for 2020–2021 in Fig. c) on its own scale identified a shift in the dominant NOx emissions source between 2017 and 2020–2021. Whilst Euston station and the previously congested southwesterly area are still noticeable, the major stand-out area is to the east. Depicted as a white box is the outline of the University of London point source as documented by the National Atmospheric Emissions Inventory, a similar inventory to the LAEI but for the whole of the UK . The University of London is the largest university in the UK, and its Bloomsbury Campus appears directly under the heightened emissions area. This site is made up of much of the University College London (UCL) central administration, the UCL hospital and Bloomsbury Heat and Power, a number of combined heat and power (CHP) sites to power the university. CHP systems simultaneously generate heat and electrical power from a single source of energy. By capturing and utilising the heat that is generated as a byproduct of the electricity generation process, efficiency is increased, which can reduce carbon emissions by up to 30 % compared to conventional separate generation . However, the requirement for CHP to be in urban areas risks an increase in air pollution. Indeed, it has been shown that CHP can “substantially” impact air quality due to NOx, the highest criteria judged by Environment Protection UK and the Institute of Air Quality Management . Here, the heat and power generation source stands out and dominates over transport but is only seen due to the drastic reduction in transport NOx emissions during the period of pandemic-reduced mobility. The Spearman correlation coefficients presented in Fig.  give further evidence that the dominant source between the two periods has changed. Correlations between NOx flux and traffic load are reduced in 2020–2021, particularly in the easterly direction. Here, the lowest correlation is observed, and high NOx fluxes are seen even at low traffic loads. These observations are in agreement with the spatial-mapping interpretation in that heat and power generation is the dominant source from this direction.