NEW PAPER: A coupled CFD and Wells-Riley model to predict COVID-19 infection probability for passengers on long-distance trains

FSEG have published a new journal paper concerning COVID-19 infection probability associated with long-distance train travel. The paper describes a methodology coupling Computational Fluid Dynamic (CFD) analysis, using the SMARTFIRE fire simulation software, with the Wells-Riley (WR) model, a well-known epidemiological model.?The coupling (WR-CFD) converts the CFD predicted concentration of respiratory aerosols to an infection probability (IP).??The CFD analysis makes use of a scalar approach to simulate the dispersion of respiratory aerosols. The WR-CFD model is validated using data collected from a study of known secondary infections resulting from known index patients travelling on Chinese long-distance trains (G-train). The analysis demonstrates that there is reasonable agreement between IP trends predicted by the model and those observed.?The study goes onto provide insight into the nature of respiratory aerosol dispersion within the ventilated saloon.?For example, the analysis demonstrates that the distribution of infectious aerosols is non-uniform and dependent on the nature of the ventilation.?The effectiveness of a range of mitigation strategies are also analysed including, ventilation rate, ventilation filtration, physical distancing, mask wearing and inoculation.?One of the most effective non-pharmaceutical measures is shown to be mask wearing.?Finally, as the approach is based on CFD it can be applied to a range of other indoor environments such as classrooms, office buildings, hospitals, etc.?

You can download the paper for free using the following link until 7 January 2022.?

https://authors.elsevier.com/a/1e5wL3IVV9nDvq

Make sure you also download the supplementary material associated with the paper.

The full citation for the paper is:

Zhaozhi Wang, Edwin R. Galea, Angus Grandison, John Ewer, Fuchen Jia,A coupled Computational Fluid Dynamics and Wells-Riley model to predict COVID-19 infection probability for passengers on long-distance trains, Safety Science, Vol 147, 2022, https://doi.org/10.1016/j.ssci.2021.105572.

This article is a summary of the key findings from the paper. Full details can be found in the cited work.

The Problem:

An issue of societal concern is the possibility of contracting COVID-19 while confined to the small volume of the passenger compartment of a train carriage (saloon), where passengers are seated in close proximity.?Of particular concern are long distance trains, where passenger journey times can vary from 1 to 8 hours.?Long distance trains are a popular and essential mode of public transport in many countries and so it is important to quantify the IP and the factors that may impact the risk. ?Thus, the main motivations of this work are to develop a modelling technique that can be used to quantify COVID-19 IP for susceptible occupants in a variety of indoor ventilated spaces, verify the model for rail applications and use the verified model to explore the efficacy of various mitigation strategies employed in long-distance train travel.

Typical internal configuration of a Chinese long distance train second class saloon

The Approach:

In this paper we present a methodology to numerically predict COVID-19 IPs for passengers travelling on long distance trains. The approach makes use of a research version of the SMARTFIRE V5.1 fire simulation software coupled to the WR model.?The WR model is modified to take into consideration that a proportion of the population may be wearing masks.?

The WR model is used by engineers and epidemiologists to estimate IP in confined spaces using ventilation and quanta generation rates.?Quanta, a term defined by Wells, is a representation of infectious dose, where inhalation of one quanta leads to an IP of 63%. The quanta concept provides a means of circumventing the knowledge gap associated with a new disease such as COVID-19; it avoids the need to specify unknown viral infectivity parameters such as, virion emission rates, the size distribution of respiratory particles, the number of virions carried by respiratory particles, aerosol deposition location within the respiratory tract and the virion dose required to cause infection.?Appropriate quanta generation rates are typically determined by back calculations of known infection events and are disease- and scenario- specific.?

While passive scalar- and particle tracking- approaches are both appropriate for modelling small particle bioaerosols, the CFD methodology adopted in this analysis represents aerosol concentration as an Eulerian scalar. As the exhaled aerosol particles are typically very small (of the order of 5 microns), drag forces will dominate gravitational forces, and so the aerosols are carried by the prevailing flow i.e., the aerosols do not require their own velocity description.?This is particularly appropriate for the scenarios investigated in this paper which are characterised by high ventilation rates which leads to air flow velocities orders of magnitude higher than the settling velocity of the aerosols. Thus, in the CFD analysis it is assumed that the respired aerosol cloud of droplets can be modelled using a simple scalar gas concentration release.?

In order to derive an estimation of absolute risk (rather than relative risk), the predicted scalar concentrations from the CFD analysis are first converted to quanta concentrations and IPs are then derived by applying the WR model. This methodology removes the limitations of using empirically derived probabilistic models and the WR model assumption of a well-mixed environment.?Furthermore, using the quanta approach means that it is not necessary to define respiratory aerosol particle size distributions, viral concentrations in droplets or viral dose required to cause infection.

The study first attempts to validate the model through comparison of the numerically predicted COVID-19 IPs for passengers travelling on long distance trains in China, reported by Hu et al. [1].?Hu et al. [1] collected a large amount of data concerning the risk of COVID-19 transmission on long distance trains in China – the so-called G-trains.?

The initial and boundary conditions used in the analysis are complex and can be found in the paper.?Here a brief summary of the key model setup parameters is presented.?As ventilation is one of the most critical factors that affect the transmission of respiratory diseases within confined spaces, the CFD scenarios are configured to the ventilation characteristics of G-trains. Two typical G-train ventilation scenarios are considered, Scenario 1, with an air change rate of 44 ACH (6200 m3/h including maximum 2120 m3/h fresh air) and Scenario 2 with an air exchange rate of 24 ACH (4400 m3/h including maximum 900-1800 m3/h fresh air). It should be noted that with up to 44 ACH, the G-trains have exceptionally high air change rates, more typical of passenger aircraft than passenger trains found in Europe and the UK.?For comparison, air change rates in some UK trains are typically about 8 ACH.??

Train geometry used in the simulations

The ventilation filters in the G-trains are believed to be equivalent to EN7779 grade 3 and above. For a grade 3 filter, the efficiency in capturing particles with size greater than 3 μm is no more than 20%. Considering that much of the aerosols from respiration are less than 3 μm, an overall filtration efficiency (FE) of 20% is applied for the G-trains in this study.

The heat generated by a human body will affect the airflow pattern and the transport of quanta inside the saloon. An average heat release rate of 50 W/m2 for resting people is applied for all the passengers with an effective body area of 1.2 m2.

Temperature distribution in a plane passing through the seated passengers

The quanta source from an index patient’s mouth is modelled as a CFD inlet (an area that releases respired air as a continuous out breath) with dimensions of 0.04 m wide and 0.05 m high at the location of the mouth of each index patient. Clearly, the quanta generation rate in each of the infection events investigated by Hu et al. [1], is unknown and is likely to have been different in each event.?In this study a single quanta generation rate of 14 quanta/h is applied in each scenario, which is assumed to be representative of infectious individuals in China at the time of their study.?Evidence justifying the quanta generation rate used in this analysis is presented in the paper.

While no mention of the use of face coverings was made by Hu et al. [1], it is likely that some people would have been wearing face coverings during the data collection period. Video footage of people queuing to board trains in Wuhan Rail station during the data collection period, show many (if not most) wearing face coverings.?Thus, as explained in the paper, in this study it is assumed that 40% of the population on the G-train were wearing face coverings. Furthermore, the efficiency of the face covering when worn by the index patient is assumed to be 50% and when worn by the susceptible 30%.

Once the model is validated, various mitigation strategies and the effects of key parameters that impact the spread of aerosols, are investigated. Study limitations are described in full within the paper.

The Main Study Findings:

(i) Validation

The WR-CFD model was able to reproduce, with reasonable agreement, trends in COVID-19 IPs observed in the Chinese study [1]. The model successfully predicts:

• the maximum IP (10.3% reported while predicted values were between 14.8% and 14.6% for the two ventilation scenarios simulated) and,
• the seat locations with the highest and lowest IP (seat B and F respectively).??
• the IPs, as a function of exposure time and distance from the index patient, with good agreement with the reported data at four of the five reported seating locations.??

Comparison of predicted and observed infection probabilities for passengers seated in the same seat row as the index patient

The differences between model predictions and observed values for IP are considered reasonable given the uncertainties in specifying actual conditions within the environment during the observation period.

The validated WR-CFD model was also used to explore the nature of respiratory aerosol dispersion within G-train saloons resulting from a single seated index patient and the efficacy of non-pharmaceutical interventions on reducing COVID-19 IP. The main observations of this analysis include:

?(ii) Non-uniform aerosol distribution

The dispersion of respiratory aerosol (quanta) within the ventilated passenger compartment is extremely non-uniform.

• The nature of the dispersion is unintuitive, with high aerosol concentration (and hence IP) within the seat row and side of the carriage containing the index patient and up to two seat rows behind the index patient.?In contrast, the seat rows ahead of the index patient have relatively low aerosol concentration.?
• Beyond the specific G-train application, this observation has implications for simpler analysis methods such as the WR model which assume that the space is well mixed, resulting in a uniform quanta distribution.?This is clearly inappropriate for complex geometries such as train carriages or situations where there are complex bulk/local airflows.?In these situations, the simplified approach may not only grossly underestimate IP, it may as a result fail to correctly assess the impact of mitigation strategies such as physical distancing.??

Non-uniform quanta distribution resulting from Index patient seated in seat 6C

(iii) Impact of ventilation rate

For long distance rail travel, the average IP is inversely related to the ventilation air change rate.??

• For an 8-hour exposure, there is an 85% increase in the expected average number of secondary infections if the air exchange rate is decreased from 44 ACH to 14 ACH.?Counter intuitively, higher air change rates also significantly increases the maximum and localised IPs of those seated close to the source.
• There is also a link between carriage ventilation filtration efficiency of the recirculated air and average IP.?If the ventilation filtration efficiency is increased from 20% to 100%, there is an up to 44% reduction in the average number of secondary infections.?However, counter intuitively, for those seated in the vicinity of the infected passenger, there is only a small reduction in IP.?

Infection probability distribution – number indicates infection probability (%) for Scenario 1 with index patient located in seat 6C

(iv) Impact of seat blocking and physical distancing

The impact of four seat blocking mitigation strategies implemented in Scenario 1 are investigated.?These involve seating passengers in the following configurations: the A, C and F seats; the odd seat rows; the A, C and F seats in odd seat rows; and a 2 m physical distancing strategy.?The physical distancing strategy is based on seating passengers only in odd rows with the A and F seats used in the first odd row followed by C seat in the next odd row; this pattern is then repeated throughout the saloon. These strategies are compared with the case in which all seats are occupied. The results are summarised in the Table below. The 2 m physical distancing strategy results in the greatest reduction in IP, reducing the average IP from 1.22% to 0.66%, a reduction of 46%.?

The 2-m physical distancing strategy (reducing seating capacity of a saloon from 85 to 14) results in a predicted reduction in average IP of 46% according to the WR-CFD model while the WR model suggests that there would be no reduction in IP.?

• Furthermore, the WR-CFD model suggests that there would be a 92% reduction in secondary infections while the WR model suggests that there would be an 84% reduction (due only to the reduced occupancy).
• Introducing inaccuracies of this type and magnitude can make the difference between accepting or rejecting a potential mitigation strategy.??

Concept of blocked seats in rail environments

(v) Impact of face coverings

Assuming 90% of the passengers correctly wear high efficiency face coverings (with a filtration efficiency of 90% for both index and susceptible) there is a 95% reduction in the average number of secondary infections compared to the case where 40% of the passengers wear low efficiency face coverings (with a filtration efficiency of 50% for index and 30% for susceptible).?

• This case is even superior to the seat blocking case with 2-m physical distancing (assuming 84 passengers in 6 saloons), producing approximately 42% fewer secondary infections than the extreme seat blocking case
• It is acknowledged that achieving such a high rate of correctly donned high efficiency face coverings is not without its challenges, especially for durations of up to 8 hours. Nevertheless, encouraging passengers to correctly wear face coverings at all times while travelling on long distance trains appears to be very effective in reducing secondary infections.?

The estimated average IP in the main analysis assumes that 40% of the passengers wear face coverings.?As a result, in the analysis it is not certain if the index patient is wearing a face covering.?For example, the average IP for an 8-hour exposure, for a susceptible in seat 6B in Scenario 1 with the index patient seated in 6C, is 25.3% or 3.2%/hr (See main paper Table 2c). ?In this analysis it is possible that the index patient and/or the susceptible are not wearing a face covering. However, it is informative to explore the IP for a susceptible passenger seated beside an index passenger, for deterministic face covering states.?

• There are four different deterministic face covering possibilities when estimating the IP of the susceptible passenger in seat 6B with index patient in seat 6C (see tables below). The IPs for the possible face covering combinations can be calculated using equations (4)-(7) in the paper, given the quanta concentration determined from the CFD simulation for seat 6B. These are determined for the case with 44 ACH (Scenario 1) and 14 ACH, a case with ventilation rate similar to that found on UK trains (see tables).

• These results demonstrate the importance of not only you as an individual wearing a face covering, but also the person seated next to you.?If you as the susceptible wear a face covering, but the person next to you, who is the index patient does not, your probability of infection is increased by 100% compared to the situation were both wear face coverings.?Conversely, if the person seated next to you wears a face covering (the index patient) and you don’t, your chance of infection in increased by 40%.?Clearly, it is better if both are wearing face coverings, but these results demonstrate that the mask status of the person seated next to you has a significantly greater impact on your IP than your own mask status.

?(vi) Impact of inoculation

In addition, the WR-CFD approach is also capable of exploring the impact of vaccination and changing population immunity through appropriate modifications of the IPs. As the pandemic progresses the effect of population vaccination programmes will also affect the transmission of the virus. For the purpose of demonstration, the current inoculation data for the UK (circa June 2021) is used.?For the B.1.617.2 (Delta) variant, PHE data suggests that one vaccination dose is 33.5% effective and two doses are 80.9% effective against infection.?Assuming 51% of the (adult) population have been partially vaccinated and 30% have been fully vaccinated, then the average IP for the vaccinated population is related to the previously calculated IP of the non-vaccinated population and leads to a relative decrease of 41.4%.??Another factor that could be included in the analysis is the natural immunity conferred by prior infection.?

Concluding Comments:

The importance of this work is not only that it validates, for the first time, the WR-CFD modelling approach for predicting COVID-19 IPs within a forced ventilated rail carriage with recirculated air environment, but that it provides invaluable insight into the nature of the aerosol dispersion within these complex and ventilated environments.?This enables the impact of proposed infection mitigation strategies for specific environments to be quantified, allowing regulatory authorities to identify the effectiveness and associated costs of competing strategies.

It is acknowledged that many of the specific findings are generally limited to the Chinese G-trains with their specific geometric arrangements and ventilation characteristics.?In particular, with up to 44 ACH, the G-trains have exceptionally high air change rates, especially compared to UK trains that typically operate at about 8 ACH.?However, the WR-CFD model can readily be applied to other scenarios, e.g. UK passenger trains, passenger aircraft, cruise and ferry ships and buildings, to allow detailed analysis of mitigation strategies, provided the environmental setup is sufficiently well described.

The following is the full citation for the paper:

Zhaozhi Wang, Edwin R. Galea, Angus Grandison, John Ewer, Fuchen Jia, A coupled Computational Fluid Dynamics and Wells-Riley model to predict COVID-19 infection probability for passengers on long-distance trains, Safety Science, Vol 147, 2022, https://doi.org/10.1016/j.ssci.2021.105572.

REFERENCES

[1] Hu M, Lin H, Wang J, et al., The risk of COVID-19 transmission in train passengers: an epidemiological and modelling study, Clinical Infectious Deseases, Vol 72, No. 4, 2021, pp 604-610. https://doi.org/10.1093/cid/ciaa1057