Ricardo: an analysis of Covid-19 lockdown on UK local air pollution

David Carslaw, air quality knowledge leader at Ricardo, writes about how they are understanding changes in air pollution during these unprecedented times.

There has been widespread media coverage of how air pollution has changed due to the COVID-19 pandemic. The satellite measurements have been especially compelling, showing the before and after situation for many of the world’s cities, see here for example. We do of course have hundreds of ground-based continuous air pollution monitors across the UK that can also be investigated to better understand the changes in air pollution at a local level.

Winds of change

A perennial problem when analysing air pollution measurements and looking to see a change due to some intervention is the effect of the weather. Indeed, changes in the weather can easily falsely mask or emphasise changes in pollutant concentrations. This is a particularly tricky problem when looking to identify the impact that pollution reduction interventions have had – ideally, we need the counterfactual i.e. what would have occurred had the intervention not taken place.

This year has been extraordinary from a weather perspective. We have had a series of Atlantic storms hit the UK throughout February … storm Ciara, storm Dennis, storm Jorge. As a cyclist, these storms are etched on my memory. These storms have had a large effect on monthly average wind speeds. Figure 1 for example, shows the monthly average wind speed at Heathrow over the past couple of years – and it is easy to spot just how high wind speeds were in February 2020 – a trend seen across the UK.

Such large changes in our weather do of course have a direct and potentially large effect on air pollution, which in this case would mostly act to reduce most pollutant concentrations. However, it also makes the analysis of air pollution data more challenging – is a change due to the weather or underlying activity?

Figure 1: Monthly average wind speeds at London Heathrow from January 2018 to late March 2020.

Unmasking the changes in air pollution
If we want to understand changes in concentrations due to reductions in emissions, it is useful if we can ‘account’ for or ‘remove’ the concentration variation due to the weather. Air pollution analysis would be a lot easier if we had the same weather every day (although preferably not February 2020 weather)!

Helpfully, there are ways that we can analyse air pollution data to estimate what trends we would have seen if the weather had been ‘average’ every day. Over the years we have developed a range of statistical models to remove the variation in concentration trends due to the weather (Carslaw and Taylor, 2009, Carslaw et al., 2012; Grange and Carslaw, 2019). These models use sophisticated statistical techniques to predict concentrations using meteorological and other data as input. Once a model is developed, it can be used to predict concentrations over a wide range of meteorological conditions.

Before we look into analysing the data with statistical models, it is useful to see what the trends in NOx have been so far in March 2020, noting that these data are provisional. Figure 2 shows the trends in NOx at a wide range of sites with potentially diverse influences: from the rural site at Chilbolton, to busy roadsides in England, Scotland and Wales, to a site that is close to Heathrow Airport. All the data shown in Figure 2 are from the UK Government’s Automatic Urban and Rural Network.

Figure 2: Daily mean concentrations of NOx at a range of air pollution monitoring sites across the UK throughout March 2020. The blue shaded rectangle shows the period from 16th March, when social distancing was first recommended.  The pink shaded rectangle shows the period from the start of the ‘lockdown’ that began on the 23rd March.

Looking at the concentrations in March 2020, Figure 2, it is hard to see any obvious reduction in NOx that could be due to responses to COVID-19. The shaded rectangles shown in Figure 2 shows the period when social distancing was first recommended and then the subsequent, stricter, ’lockdown’  when Boris Johnson advised everyone in the UK against ‘non-essential’ travel and contact with others, to work from home if possible and ordered closure of pubs, clubs and theatres.

So, is there an effect on NOx concentrations?

We have developed statistical models for hourly NOx concentrations at all the sites shown in Figure 2 from January 2018 to February 2020. These models use as their input local meteorological data such as wind speed, wind direction, ambient temperature as well as other temporal variables such as hour of the day. These models are then used to predict the expected pollution concentrations throughout March, providing a ‘business as usual’ (our counterfactual).

An example of the output from the models is shown for the York Fishergate site (my home city). The plot shows the difference in predicted NOx compared with measured NOx. The green colour shows where concentrations were lower than business as usual and the pink the opposite. The first two thirds of the timeline show there is a close correspondence between measured and predicted concentrations (shown by small periods of green or pink shading). In other words, the model does a good job of predicting NOx in the early part of March, before COVID measures were introduced, even on an hourly basis. However, towards the end of the plot, there are clear periods where measured concentrations were less than the business as usual, as shown by the increased green shading – a possible effect of reduced emissions as a result of a reduction in pollution generating activity.

Figure 3: Concentrations of NOx at the York Fishergate roadside site. Trend areas shaded green indicate that concentrations of NOx were estimated to be lower than that expected through business as usual. The pink trend shading shows where concentrations of NOx were estimated to be higher than business as usual. The blue shaded rectangle shows the period recommended social distancing from 16th March, the pink shaded rectangle shows the period from the start of the lockdown starting 23rd March.

It’s easy to ‘eyeball’ plots like Figure 3 and conclude there seems to be a difference between measured and business as usual concentrations – but is there a better way to see if changes occurred at a similar time across sites?

One straightforward approach to understand changes is to plot a cusum chart. The idea behind a cusum plot is simple – plot the accumulated difference in concentrations from one time point to the next point in time. By accumulating these differences, the effect is to amplify the changes and also provide an indication of when changes may have occurred. In this case, we consider the differences in concentrations between measured and business as usual and consider how they diverge.

Cusum plots are shown for the same range of sites as shown in Figure 2. Taking the York Fishergate example first (top-left of Figure 4), it can be seen the cusum values did not change until around March 16th i.e. when there was Government advice to reduce unnecessary travel. The first period of no change means that the business as usual NOx and measured NOx were very similar i.e. things behaved as ‘normal’. The steep decrease in the cusum line after the 16th March indicates that there has been a shift to much lower concentrations – and a clear divergence from business as usual.

Most other sites show a similar pattern, although there seems to be a time delay at some sites such as those in Glasgow and Manchester. The London Harlington site close to Heathrow seems to show evidence of a reduction in NOx ahead of the 16th March, which might be caused by reduced air travel ahead of the 16th March – something that should be investigated further. The anomalous site seems to be the rural Chilbolton site, which seems to show increased in concentration after the 16th March.

Figure 4: The cumulative sum (or cusum) of measured minus business as usual NOx at a range of air pollution monitoring sites across the UK. The blue shaded rectangle shows the period recommended social distancing from 16th March, the pink shaded rectangle shows the period from the start of the lockdown starting 23rd March.

The measured and BAU concentrations can be used to estimate the changes in NOx due to the COVID-19 pandemic. These provisional estimates are shown in Figure 5 and show that NOx concentrations are predicted to have typically reduced by about 30 to 40% depending on the site (except Chilbolton, which showed an increase). As more data becomes available, the robustness of these estimated changes should increase. However, it is already clear that there has been a dramatic and mostly consistent decrease in NOx across a wide range of sites.

Figure 5: Measured and estimated business as usual NOx concentrations by site. The numbers show the percentage change in concentration relative to business as usual.

What next?

This analysis is a first look at some potential changes in air pollutions due to COVID-19. We will aim to update and extend the analysis on a regular basis and hopefully say more about a wider range of pollutants. We might also expect an accelerated reduction in many air pollutants from 23rd March onwards after the Prime Minister’s statement on television, which essentially enforces a ‘lockdown’. We also want to understand the impact on regional air pollution from reduced emissions in the rest of Europe and how that affects the UK’s air pollution climate.


Carslaw, D.C. and P.J. Taylor (2009). Analysis of air pollution data at a mixed source location using boosted regression trees. Atmospheric Environment.  Vol. 43, pp. 3563–3570.

Carslaw, D.C., Williams, M.L. and B. Barratt A short-term intervention study — impact of airport closure on near-field air quality due to the eruption of Eyjafjallajökull. (2012) Atmospheric Environment, Vol. 54, 328–336.

Grange, S. K. and Carslaw, D. C. (2019) ‘Using meteorological normalisation to detect interventions in air quality time series’, Science of The Total Environment. 653, pp. 578–588. doi: 10.1016/j.scitotenv.2018.10.344.

Leave a Reply

Notify of