Envisaging a world with greener cities.

These visualisations provide insights into the degree of mixing along the height and width of the saloon, and will be useful to compare with our CFD simulations to ensure that our simulated flow patterns are representative.

Particle dispersion experiment: A nebuliser was used to release puffs of aerosol particles, positioned on a seat within the saloon. Particulate matter sensors were placed at several locations within the saloon. These sensors measured the concentration of particles at each location, showing the degree of exposure at each location to the released particles. Both fine ( ) and coarse ( ) particles were released in order to investigate the importance of particle mass on dispersion. This is a key question when looking at virus transmission as the size of virus-laden droplets can vary significantly.

Results of the experiments are currently being analysed, with results to be published shortly.


Figure 1: NOx and NO2 concentrations time series for a 20 minute period of the September field study.


The SARS-CoV-2 virus has so far infected more than 2 million people around the world, and its impact is being felt by all. Patients with airborne diseases such as Covid-19 should ideally be treated in negative pressure isolation rooms. However, due to the overwhelming demand for hospital beds, patients are being treated in general wards, hospital corridors, and makeshift hospitals.

Adequate building ventilation in hospitals and public spaces is a crucial to contain the spread of disease, exit from the lockdown situations and reduce the chances of subsequent waves of outbreaks. 

The MAGIC team is looking at how to use displacement ventilation (either mechanical or natural ventilation), where air intakes are at low level and extracts are at high level, as a viable alternative to negative pressure isolation rooms, which are often not available on site in hospital wards and makeshift hospitals. Displacement ventilation produces negative pressure at the occupant level, which draws fresh air from outdoor, and positive pressure near the ceiling, which expels the hot and contaminated air out. The team will look at whether a lack of openings, in some settings, can be supplemented by installing extraction fans.

Our aim is to provide guidelines for such mechanically assisted-naturally ventilated makeshift hospitals, and for public spaces such as supermarkets and restaurants. Our results will be published on this page. 

The researchers explored a range of different modes of exhalation: nasal breathing, speaking and laughing, each both with and without a mask. By imaging the heat associated with the exhaled breath, they could see how it moves through the space in each case. If the person was moving around the room, the distribution of exhaled breath was markedly different as it became captured in their wake.

The results show that room flows are turbulent and can change dramatically depending on the movement of the occupants, the type of ventilation, the opening and closing of doors and, for naturally ventilated spaces, changes in outdoor conditions. Masks were found to be effective at reducing the spread of exhaled breath, and therefore droplets.

“One thing we could clearly see is that one of the ways that masks work is by stopping the breath’s momentum,” says MAGIC PI, Paul Linden. “While pretty much all masks will have a certain amount of leakage through the top and sides, it doesn’t matter that much, because slowing the momentum of any exhaled contaminants reduces the chance of any direct exchange of aerosols and droplets as the breath remains in the body’s thermal plume and is carried upwards towards the ceiling. Additionally, masks stop larger droplets, and a three-layered mask decreases the amount of those contaminants which are recirculated through the room by ventilation.”

The researchers found that laughing, in particular, creates a large disturbance, suggesting that if an infected person without a mask was laughing indoors, it would greatly increase the risk of transmission.

For more information please see our papers in the Journal of Fluid Mechanics and Royal Society Open Science. ​

Figure 2: Snapshot of video footage showing cement mixer passing the roadside sensor

LONDON TEST SITE (September, 2019):

In September 2019 we ran a 2-week intensive field study at London Road next to London South Bank University (LSBU). Our aim for this study was to collect data for calibration and validation of the different MAGIC models.

Originally, the idea was to replicate the 2017 study which involved various low-cost sensors developed by the MAGIC project. In 2017 these sensors were based along a stretch of London Road just South of St Georges Circus and also inside a room in the LSBU Clarence Centre. This time we decided to also add a camera to obtain traffic information with the hope to be able to determine the effect of traffic on pollution levels.

As planning progressed, we were lucky to have TfL join us. TfL agreed to change the signal timings at the Garden Row/London Road junction to see whether doubling the cycle time would lead to a decrease in emissions. It became clear very quickly that we would need high precision and high time resolution sensors to really be able to understand the effects of the changes in signal timings. As these high cost sensors are quite power intensive and supposed to be run indoors on mains power, we had to design various bespoke battery systems to power these instruments and find boxes to protect the sensors against rain.

We relied heavily on the help and support of the team at LSBU, including the security staff and lab technicians who enabled us to move and store our equipment safely, and Elsa Aristodemou, MAGIC PI, who coordinated the LSBU team. Thank you for all your help!

To ready more about the study click here

Figure 2: Cross-section of plume 95% percentile at half a building height downwind of the building for (a) the normal facing building and (b) the pitched roof at 45 degrees.Type your paragraph here.


A novel methodology based on a Weak Constraint Gaussian Process (WCGP) has been successfully developed in order to advise on the optimal placement of pollutant sensors in a complex urban environment.

The question of optimal sensor placement is usually addressed using a Gaussian Process (GP) approach. However, the main limitations of traditional GP regression are its sensitivity to noisy input and computational complexity. Indeed the vast amount of real-life noisy data usually available makes the prediction quality in most applications non-optimal at best. The novel WCGP method developed in the context of the MAGIC project overcomes the computational complexity issues by decomposing the domain and therefore enabling parallel computations, while optimally dealing with noisy data by using a data assimilation approach that incorporates noisy input in a time window. The scientific paper, written by the MAGIC team members at Imperial College London and accepted recently in Journal of Computational Science, exposes in details the mathematical formulation of such an approach as well as the technical challenges and solutions adopted.

This innovative approach has been tested and has proved its efficiency on the LSBU area, i.e. MAGIC test site, under a westerly wind, in order to advise on the optimal sensor location. The sensors are optimally positioned such that they will provide the most “useful” data able to correct our predicted model, i.e. Fluidity (open-source CFD software). In other words, the sensors do not necessarily aim to capture peaks of high pollutant concentration. Instead, for example, they aim to give the data that will greatly improve the accuracy of the pollutant concentration predictions. In the picture opposite, as an example, the optimal location of the sensors as advised by a traditional GP (blue spheres) and our novel WCGP (red spheres) are shown when the LSBU site is subject to a westerly wind. The time it took to output the sensors locations was about 8 times faster with our WCGP (~11H) than using the traditional GP (~100H). Moreover, the parallel implementation of the algorithm allows to speed up the process up to 13 times when run on 16 cores.


One of the interesting things coming out of the September field study at the LSBU site is the very high NOx and NO2 concentration spikes seen when high emitting vehicles pass the roadside sensor. This includes heavy goods vehicles, bin lorries, skip lorries or coaches.

Below (Figure 1) is a time series plot of the NO2 (blue) and NOx (red) concentrations measured at the roadside at the field study junction in Elephant and Castle. Also shown are the number of vehicles waiting at the lights in the lane nearest the sensor. It can be seen that each concentration peak corresponds to a number of vehicles waiting at the lights, with a lorry or a coach usually present.

At roughly 13:06 there is a particularly high spike in NOx, with concentrations reaching >2000ppb. This spike corresponded to the passing of a cement mixer lorry (Figure 2). The cement mixer lorry has a secondary engine used to power the cement mixer. Secondary engines are regulated to a much lower standard than main vehicle engines. As well as cement mixers, lorries with refrigeration units have secondary engines. The big spike in NOx seen at 13:06 is likely due to the secondary engine, which are known to have much higher NOx and PM emissions than allowed under modern Euro emission standards. These engines use “red diesel”, used by Non-Road Mobile Machinery (NRMM) which is taxed at a much lower rate than road diesel. Campaigners have long called for this tax loophole to be closed to encourage the use of cleaner alternatives such as liquid nitrogen or electric.

These peak concentrations at the roadside are missed by static sensor networks. Most air quality models will also miss these peaks due to insufficient spatial and temporal resolution. Further, the epidemiological estimates of the impact of pollution exposure on health depend on these lower resolution models and do not account for the full variability of street level concentrations.

On the MAGIC project we plan to use Fluidity’s ability to simulate high resolution dispersion at the microscale to investigate to what degree these peak concentrations contribute to overall pedestrian exposure, and what measures can be taken to reduce exposure within urban areas.

THE WATER FLUME (August, 2019):

Megan and her team have been carrying out experiments investigating the effect of temperature differences on cross-ventilation. The experiments are performed in a large flume, with a cross-sectional area of 2 m x 1 m. Into the flume is placed a model room, which is a 0.5 m cube.

Before the start of an experiment, the water inside the model room is heated and dyed. The flume is switched on, to provide a flow past the model room, modelling the effect of wind. Two windows are opened on opposite walls of the model room, one at the front, facing the wind, and one at the back. 

Recent experiments have examined the effect of temperature on cross-ventilation. Results show that the major effect of the initial temperature difference between the room and ambient is to create a two-layer stratification, reducing the effective volume of the room. New theory indicates that the temperature difference does not significantly impact the ventilation flow rate through the windows, but the change in effective room volume alters the timescale over which the room is ventilated. These results are currently being written up for publication. You can read a summary of the findings here

Experiments performed by Megan Davies Wykes and Elkhansaa Chahour.

WIND TUNNEL (May, 2019):

In the video, Will explains his work recreating the London test site in the Enflo  wind tunnel at the University of Surrey. Will is now building a scaled-down replica of our Cambridge test site for the next phase of wind-tunnel experiments, and he has already created a drawing of each of the 167 model buildings.

Makeshift hospital in the US

Optimal sensor placement when using a traditional Gaussian Process (GP) approach (blue spheres) or a Weak Constraint Gaussian Process (WCGP) approach (red spheres). Credit: Journal of Computational Science.


Ventilation systems in many modern office buildings, which are designed to keep temperatures comfortable and increase energy efficiency, may increase the risk of exposure to the coronavirus, particularly during the coming winter, according to research published by MAGIC researchers.

Widely-used mixing ventilation systems, which are designed to keep conditions uniform in all parts of the room, disperse airborne contaminants evenly throughout the space. These contaminants may include droplets and aerosols, potentially containing viruses. The research has highlighted the importance of good ventilation and mask-wearing in keeping the contaminant concentration to a minimum level and hence mitigating the risk of transmission of SARS-CoV-2, the virus that causes COVID-10.

The evidence increasingly indicates that the virus is spread primarily through larger droplets and smaller aerosols, which are expelled when we cough, sneeze, laugh, talk or breathe. In addition, the data available so far indicate that indoor transmission is far more common than outdoor transmission, which is likely due to increased exposure times and decreased dispersion rates for droplets and aerosols.

As winter approaches in the northern hemisphere, and people start spending more time inside, understanding the role of different types of ventilation is critical to estimating the risk of contracting the virus and helping slow its spread. While direct monitoring of droplets and aerosols in indoor spaces is difficult, we exhale carbon dioxide that can easily be measured and used as an indicator of the risk of infection. Small respiratory aerosols containing the virus are transported along with the carbon dioxide produced by breathing, and are carried around a room by ventilation flows. Insufficient ventilation can lead to high carbon dioxide concentration, which in turn could increase the risk of exposure to the virus.



Public transport poses a unique challenge when evaluating the risk of possible airborne transmission of Covid-19 and when seeking to establish suitable mitigation strategies.

The airflow patterns within the vehicle are driven by both the ventilation and by thermal plumes driven by the body heat of passengers. So, these airflow patterns are likely to depend on the number and location of occupants in addition to the ventilation configuration. The provision of outdoor air when doors are opened to allow passengers to board and alight should also be considered, including with vehicles which are otherwise entirely mechanically ventilated. In order to understand the risk of viral transmission on public transport we must first understand the airflow patterns on these vehicles.  

The MAGIC team have therefore conducted three separate experiments on a static inter-city train carriage. The aim was to determine the viability of each experiment in improving our understanding of the airflow patterns within the carriage and its relevance to the risk of airborne transmission. This was done via three experimental set ups, as follows:

CO2 experiments. These involved the generation of CO2 by six volunteer "passengers'' sat in the carriage saloon while the ventilation was turned off, allowing CO2 concentrations to rise, before turning the ventilation on and measuring the decay rate of CO2 at several locations. The objective was to explore the feasibility of using CO2 generated by exhaled breath and low-cost sensors to resolve concentration differences within the saloon in addition to variations in decay rates.

Airflow visualisation experiments: The objective here was to visualise the airflow patterns across a cross-section of the carriage. The saloon of the carriage was filled with smoke from a fog machine while a cross-section of the carriage was illuminated by lasers, allowing us to see the smoke’s movement. 

Flow visualisation in a train carriage

Pressure distribution, and neutral height inside a naturally ventilated building.

Figure 1: Instantaneous plume (green) dispersion past (a) a normal facing cube and (b) 45 degree building with pitched roof.Type your paragraph here.


This project, funded by the Atmospheric Dispersion Modelling Liaison Committee (ADMLC) and in partnership with CERC, makes use of the advanced computational fluid dynamics (CFD) and wind tunnel capabilities within MAGIC to assess the performance of Gaussian plume models at short time and length scales. Gaussian plume models were initially developed 40 years ago for application to nuclear power facilities and accidental releases, and have since been developed for use in regulation of stack releases, odour releases and urban air quality analysis. These models are popular with regulators and designers as they give good estimates for long term averages and are very quick to run. However, these models are not designed for problems over length scales of meters, or for problems such as puff or instantaneous releases, where short time scales are relevant. CERC’s Atmospheric Dispersion Model System 5 (ADMS5) is an advanced Gaussian plume model. ADMS5 possesses features such as a building model and fluctuations model. Comparisons were made between the ADMS5 building model and Fluidity CFD simulations validated against wind tunnel experiments undertaken the University of Surrey.

The ADMS5 building model simplifies buildings with roofs or at oblique angles to the prevailing wind direction into a single, normal facing building with an equivalent frontal area. The flow features around a normal facing building are well understood and therefore the model can then make assumptions on the nature of the dispersion of releases from the rooftop or from upwind locations. Over the short length scales considered in this project, building orientation and the shape of roofs may have a significant effect on plume dispersion. Figure 1a shows a snapshot of a Fluidity simulation of the instantaneous dispersion of a plume past a cube building, normal facing to a turbulent flow. Figure 1b shows an equivalent simulation for a  building set at a 45 degree angle to the approaching wind direction, with a pitched roof. The effect of the different building geometry on the downwind dispersion is clear, with a much increased dispersion for the pitched roof case.Type your paragraph here.

This increased turbulence due to the pitched roof leads to increased mixing downwind of the building, and increases the entrainment of the plume down into the building wake. This can be seen from Figure 2 which shows the plume 95% percentile concentrations at a distance of H/2 downwind of the building face, where H is the height of the normal facing building. Much lower concentrations are seen at the plume centreline for the pitched roof case due to the increased mixing, while higher concentrations are seen lower down in the building wake for this case compared to those for the normal facing building.

Comparisons such as these serve to inform our understanding of microscale dispersion and to evaluate the limitations of simpler models used day-to-day by regulators at such short time and length scales.  

One of our Raspberry- Pi Cameras used on site