Extending indoor airborne infection risk modelling to 3D, with local surface deposition

arXiv:2611.13372 · q-bio.PE, math.DS, stat.OT · Submitted 2026-11-09 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Genomics Radio. Generated commentary on the latest computational biology and genomics papers.

Ines: We're looking at the paper: "Quantifying the impact of 3D diffusion and local versus global removal by surface deposition on indoor airborne transmission".

Marcus: Here's the gist of the paper. The study extends the Viral Infection Risk Indoor Simulator (VIRIS) by modeling airborne transmission in three dimensions and replacing global removal with a local surface deposition condition to provide a more realistic assessment of infection risk and transmission rates Model Extensions The authors developed VIRIS-3D-L by extending the airborne transmission model in VIRIS (a reaction–diffusion equation) to three dimensions (3D) and replacing global removal with a Robin boundary condition representing localised removal due to surface deposition.

Ines: This extension addresses limitations in the original VIRIS model where the 2D formulation leads to underestimation of aerosol concentrations at individuals’ mouth level and cannot represent variations in aerosol transport due to differences in people and furniture heights.

Marcus: Specifically, they represented each agent as a point source or sink at a height of 1.8m, with (x, y) coordinates following the agent trajectory.

Ines: Furthermore, they removed the first-order global removal term, γ, from Eq.

Paper summary: Ines: To wrap up this work on "Quantifying the impact of three dee diffusion and local versus global removal by surface deposition on indoor airborne transmission," it's clear that moving from a two-dimensional model to a three-dimensional one with localized surface deposition is fundamentally changing the output <ref:2611.13372#pg1,Quantifying the impact of 3D diffusion and local versus global removal by>.

Marcus: The authors are showing that this extension addresses key limitations in how we used to calculate infection risk, specifically concerning the spatial variation of aerosols and concentrations at different heights within an indoor space >

Yuki: It seems like the biggest implication for us in population genetics and public health is that if our models don't account for vertical gradients, we're missing a lot about where people are actually getting infected in a building >

Ines: Right. And the authors conclude that VIRIS-three dee-L is preferable for prediction and assessment because it predicts higher breathing zone concentrations and greater spatial variation in risk > <ref:2611.13372#pg1>

Marcus: They also admit that while this model is more accurate, they acknowledge that the simulation time increases significantly, which means there's a trade-off between accuracy and speed >

Yuki: So what's the final word on this paper? It’s about using more complex physics to get a better picture of transmission dynamics in indoor settings >

Ines: Exactly. And they recommend VIRIS-three dee-L for infection risk prediction and intervention assessment, especially in situations with varying people and furniture heights or frequent movement > <ref:2611.13372#pg1>

Marcus: They suggest that VIRIS remains useful for rapid screening of simple single-storey buildings when lower computational cost is essential >

Conclusion: Ines: So we've been digging into how changing the model to three dimensions and focusing on surface removal actually shifts the numbers for infection risk in indoor settings, and now we're getting to what this paper is actually trying to tell us about that whole process.

Marcus: They’re looking at this model called VIRIS-three dee-L, which takes a 2D thing and makes it three dimensions, so they can see how aerosols move up and down in the space > <ref:2611.13372#pg1>

Yuki: I think what's interesting here is how they change the removal part. Instead of just removing all the germs globally at once, they make it local surface deposition, which is much more realistic for real life >

Ines: Exactly. It changes the math from a simple removal term to something that depends on where you are physically located in relation to surfaces >

Marcus: And they used this new model across three different buildings—a courtroom, a care home, and a supermarket—to see how much more accurate it is compared to the older VIRIS version >

Yuki: The results show that the old model really underestimates the actual risk by up to fifty percent in all those buildings >

Ines: That underestimation comes from assuming uniform aerosol concentrations everywhere, which isn't true when you have different people and furniture heights >

Marcus: And they used a statistical test, a Wilcoxon signed-rank test, and it confirmed that this systematic underestimation is real across all three building types >

Yuki: So what does this mean for us? It means if we only use the simpler model, we're severely miscalculating how dangerous certain indoor spaces are for transmission >

Ines: It tells us that when planning interventions, you need a model that captures that vertical movement and those surface effects to get a better picture of where risk is actually highest >

Marcus: But they also noted the trade-off. This more realistic model takes about four times the computation time compared to the original one, which is something we have to consider when we're trying to run these simulations frequently >

Yuki: Yeah, so while this new model gives us a much better picture of risk in terms of spatial variation, we still have that computational hurdle when we try to apply it widely across many different scenarios >

Cardiff University

q-bio.PE, math.DS, stat.OT

Submitted: 2026-11-09

Updated: 2026-11-13

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 92/100

The gist: The study extends the Viral Infection Risk Indoor Simulator (VIRIS) by modeling airborne transmission in three dimensions and replacing global removal with a local surface deposition condition to…

Key concepts

VIRIS-3D-L
This is an extended version of the original VIRIS model. It incorporates three dimensions (3D) and replaces global removal with a local surface deposition condition. This aims to create a more realistic simulation of airborne disease spread by accounting for spatial variations in air movement and removal.
Three-Dimensional Modeling
Instead of treating the environment as flat (2D), this approach models the indoor space in three dimensions. This allows the simulation to accurately capture how aerosols move and concentrate at different heights, which is crucial because aerosol concentration varies depending on where people and furniture are located.
Local Surface Deposition
This replaces the original model's global removal term with a condition that simulates local removal due to surfaces. This means aerosols are removed from the air specifically when they come into contact with surfaces, providing a more physically realistic way to account for how airborne particles are cleared from an indoor environment.

Terminology

Summary

The study extends the Viral Infection Risk Indoor Simulator (VIRIS) by modeling airborne transmission in three dimensions and replacing global removal with a local surface deposition condition to provide a more realistic assessment of infection risk and transmission rates

Model Extensions

The authors developed VIRIS-3D-L by extending the airborne transmission model in VIRIS (a reaction–diffusion equation) to three dimensions (3D) and replacing global removal with a Robin boundary condition representing localised removal due to surface deposition. This extension addresses limitations in the original VIRIS model where the 2D formulation leads to underestimation of aerosol concentrations at individuals’ mouth level and cannot represent variations in aerosol transport due to differences in people and furniture heights. p. 1

Where the paper says it

Consequently, the 2D formulation leads to underestimation of aerosol concentrations at individuals’ mouth level and cannot represent variations in aerosol transport due to differences in people and furniture heights. These limitations affect predicted spatial transmission dynamics and individual infection risks.

Specifically, they represented each agent as a point source or sink at a height of 1. p. 2
Where the paper says it

2) where 𝜂𝑖 is the mask efficiency of the 𝑖 th agent, ℎ is the room height, 𝐴𝑖 is aerosol emission rate of the 𝑖 th agent, 𝐹𝑣 is fraction of viral-laden aerosols, and 𝛿 is the Dirac Delta function representing a point source. The sink due to inhalation by all the agents 𝑛 in an indoor space, 𝑅(x, 𝑡) is defined as, 𝑅(x, 𝑡) = ∑︁𝑛 𝑖=1 (1 − 𝜂𝑖)𝜌𝑖𝛿(x − xi(𝑡)), (2.

8m, with (x, y) coordinates following the agent trajectory. p. 2
Where the paper says it

First, we represent each agent as a point source or sink at a height of 1. 8m, with (𝑥, 𝑦) coordinates following the agent trajectory.

Furthermore, they removed the first-order global removal term, γ, from Eq. p. 1
Where the paper says it

Further, in VIRIS, surface deposition of aerosols is represented as a first-order global removal term in the reaction-diffusion equation, as in several airborne transmission models [15, 17, 11]. In reality, aerosols deposited on surfaces such as walls, ceilings, floors, furniture, or other solid objects are required to be removed from the domain rather than homogeneously throughout the domain.

(2.1), changing it to β + λ and replacing the Neumann boundary condition with a Robin boundary condition representing local aerosol removal at surfaces. p. 8
Where the paper says it

: Computation time in minutes for three single-storey buildings using the VIRIS, and VIRIS-3D-L modelling frameworks. Conclusion In this paper, we extended VIRIS [30] by addressing its two assumptions: first, modelling airborne transmission using a reaction–diffusion equation in a 3D domain instead of a 2D domain, and second, replacing the global surface-deposition sink term with a Robin boundary condition representing local aerosol removal at surfaces (VIRIS-3D-L framework).

Comparative Results Across Buildings

The comparison of VIRIS and VIRIS-3D-L was conducted across three single-storey buildings: a courtroom, a care home, and a supermarket. The findings indicated that VIRIS underestimates the disease transmission rate and infection risks for individuals by up to 50% relative to VIRIS-3D-L in all three buildings. p. 0

Where the paper says it

We compare VIRIS and VIRIS-3D-L across three single-storey buildings with increasing architectural and people-movement complexity: a courtroom, a care home, and a supermarket. We find that VIRIS underestimates the disease transmission rate and infection risks for individuals by up to 50% relative to VIRIS-3D-L in all three buildings.

For instance, in the courtroom, the infection risk for P3 and P5 was 0. p. 1
Where the paper says it

In Sections 3. 2, respectively, we compare the infection risk predictions and computation times of VIRIS and VIRIS-3D-L for three single-storey buildings with increasing architectural and people movement complexity: a courtroom, a care home, and a supermarket.

5 using VIRIS and 0. p. 1
Where the paper says it

Next, we describe the VIRIS-3D-L framework. We refer to individuals as ‘agents’ in this paper; their behaviour is modelled using an agent-based model [30], and remains the same for VIRIS-3D-L.

6 using VIRIS-3D-L at the end of the simulation. p. 5
Where the paper says it

For example, at the end of the simulation, the infection risk for 𝑃3 and 𝑃5 is 0. 5 using VIRIS and 0.

Similarly, in the care home, infection risk for P4 and P6 was 47% and 50% relative to VIRIS-3D-L at the end of the simulation. p. 5
Where the paper says it

Further, VIRIS significantly underestimates individuals’ infection risk by 8% to 28% relative to VIRIS-3D-L. For example, at the end of the simulation, the infection risk for 𝑃3 and 𝑃5 is 0.

Statistical Validation

The statistical comparison using the one-sided Wilcoxon signed-rank test revealed that VIRIS significantly underestimates infection risk compared to VIRIS-3D-L for all three buildings. Hypothesis 3, which states that VIRIS underestimates, was supported by p-values below 0. p. 3

Where the paper says it

(𝑥𝑖(𝑡), 𝑦𝑖(𝑡), 1. Although we could use a custom height for each agent by providing the respective values as input to the agent-based model in VIRIS, we fix all heights to be equal for simplicity.

05 for the underestimation hypothesis. p. 1
Where the paper says it

in three spatial dimensions, resulting in vertical concentration gradients and higher concentrations of infectious aerosols in the breathing zone. Consequently, the 2D formulation leads to underestimation of aerosol concentrations at individuals’ mouth level and cannot represent variations in aerosol transport due to differences in people and furniture heights.

This systematic underestimation is attributed to the assumption of uniform aerosol concentrations over the building height and applying surface-deposition removal throughout the domain, which lowers infectious aerosol concentration. p. 8
Where the paper says it

This would overestimate safe exposure durations and lead to interventions that are insufficient to reduce infection risk during an outbreak. These differences arise because VIRIS assumes uniform aerosol concentrations over the building height and applies surface-deposition removal throughout the domain, including the breathing zone,

Computational Trade-offs

While VIRIS-3D-L provides a more realistic modelling of airborne transmission than VIRIS, it requires, on average, four times the computation time of VIRIS mainly because it extends the computational domain from a 2D surface to a 3D volume. The simulation time increases significantly depending on the building complexity; for example, total computation time increased by six minutes for the courtroom and 14 minutes for the care home. p. 8

Where the paper says it

1) takes longer in a 3D domain than in a 2D domain (Table 4 in Appendix D). Specifically, total computation time increases by six minutes for the courtroom, 14 minutes for the care home, and 1.

The highest increase in computation time was observed in the supermarket, reflecting its dynamic environment and complex layout. p. 8
Where the paper says it

The care home simulation involved 13 hours of moving residents, compared with three hours of static individuals in the courtroom. The increase in computation time for the supermarket is highest, reflecting its dynamic environment, 240 customers, and complex layout, compared with the care home’s six moderately dynamic residents and 10 static individuals in the courtroom.

Conclusion and Recommendation

Overall, VIRIS-3D-L is preferable for infection risk prediction and intervention assessment because it predicts higher breathing zone concentrations and greater spatial variation in risk, particularly in settings with varying people and furniture heights. Although only single-storey buildings are considered here, VIRIS-3D-L can be applied to multi-storey buildings without modification. p. 9

Where the paper says it

Extending VIRIS’s airborne transmission model to a 3D domain increases computation time by about fourfold on average. Although only single-storey buildings are considered here, VIRIS-3D-L can be applied to multi-storey buildings without modification.

The paper concludes that VIRIS remains useful for rapid screening of simple single-storey buildings when lower computational cost is essential. p. 9
Where the paper says it

Overall, VIRIS-3D-L is preferable for infection risk prediction and intervention assessment, particularly in settings with varying people and furniture heights or frequent people movement. VIRIS remains useful for rapid screening of simple single-storey buildings when lower computational cost is essential.

p. 9
Where the paper says it

Overall, VIRIS-3D-L is preferable for infection risk prediction and intervention assessment, particularly in settings with varying people and furniture heights or frequent people movement. VIRIS remains useful for rapid screening of simple single-storey buildings when lower computational cost is essential.

Future work could validate VIRIS-3D-L against measured aerosol concentrations and outbreak data. p. 9
Where the paper says it

Although only single-storey buildings are considered here, VIRIS-3D-L can be applied to multi-storey buildings without modification. Future work could validate VIRIS-3D-L against measured aerosol concentrations and outbreak data, investigate airflow-driven advection in well-ventilated spaces, and quantify how local and global representations of ventilation removal affect infection risk.

Computation Time Analysis

The difference in computation time between VIRIS and VIRIS-3D-L is provided in Table 4, separated by processing steps for three single-storey buildings. The reaction–diffusion solution component showed the largest difference in computation time between the two frameworks across all three buildings. p. 8

Where the paper says it

: Computation time in minutes for three single-storey buildings using the VIRIS, and VIRIS-3D-L modelling frameworks. Conclusion In this paper, we extended VIRIS [30] by addressing its two assumptions: first, modelling airborne transmission using a reaction–diffusion equation in a 3D domain instead of a 2D domain, and second, replacing the global surface-deposition sink term with a Robin boundary condition representing local aerosol removal at surfaces (VIRIS-3D-L framework).

For instance, the reaction-diffusion solution took 1. p. 0
Where the paper says it

Here, we extend VIRIS to model airborne transmission more realistically and compare the impact on infection risk predictions. We develop VIRIS-3D-L by extending the airborne transmission model in VIRIS (a reaction–diffusion equation) to three dimensions (3D) and replacing global removal with a Robin boundary condition representing localised removal due to surface deposition.

8 minutes for VIRIS and 170 minutes for VIRIS-3D-L in the supermarket. p. 8
Where the paper says it

5 Supermarket 25. : Computation time in minutes for three single-storey buildings using the VIRIS, and VIRIS-3D-L modelling frameworks.

The difference in computation time is provided in Table 4, separated by processing steps for three single-storey buildings. p. 3
Where the paper says it

The code is available on GitHub. Results Here, we compare individual infection risk predictions over time and computation times for VIRIS and VIRIS-3D-L across three single-storey buildings.

Concentration Comparison

The absolute difference between scaled spatio-temporal concentration values, M(x, y, t), was estimated to compare the two frameworks across the three buildings. Figure 7 plots the average and maximum values of these absolute differences in virus concentration between VIRIS and VIRIS-3D-L across the three buildings. p. 8

Where the paper says it

The variance in infection risk is generally higher (> 1) in VIRIS-3D-L than in VIRIS. Additionally, the maximum bounds of the box plots are lower in VIRIS than in VIRIS-3D-L across all four entrytime intervals, indicating significant underestimation in VIRIS, ranging from 29% to 45% relative to VIRIS-3D-L.

The maximum bounds of the box plots are lower in VIRIS than in VIRIS-3D-L across all four entrytime intervals, indicating significant underestimation in VIRIS. p. 8
Where the paper says it

The variance in infection risk is generally higher (> 1) in VIRIS-3D-L than in VIRIS. Additionally, the maximum bounds of the box plots are lower in VIRIS than in VIRIS-3D-L across all four entrytime intervals, indicating significant underestimation in VIRIS, ranging from 29% to 45% relative to VIRIS-3D-L.

Infection Risk Comparison

The Wilcoxon signed-rank test results showed that the two models produce systematically different infection-risk estimates for all three buildings. The one-sided test results presented in Table 3 showed that VIRIS systematically underestimates infection risk relative to VIRIS-3D-L for all three buildings. p. 8

Where the paper says it

Further, we use a one-sided Wilcoxon [29] signed-rank test (Appendix C) for statistical comparison of infection prediction between the two frameworks. We observe that VIRIS significantly underestimates infection risk compared with VIRIS-3D-L for all three buildings.

This systematic underestimation is a key finding supporting the recommendation to use VIRIS-3D-L for infection risk prediction. p. 8
Where the paper says it

2), we also find that the more realistic airborne transmission modelling in VIRIS-3D-L predicts significantly higher infection risk and a higher disease transmission rate. These results further strengthen the recommendation to use VIRIS-3D-L for infection risk prediction.

Summary Table of Results

Table 3 presents the Wilcoxon signed-rank test results for three single-storey buildings to compare VIRIS and VIRIS-3D-L. The two-sided p-values are below 0. p. 8

Where the paper says it

These results further strengthen the recommendation to use VIRIS-3D-L for infection risk prediction. Further, we use a one-sided Wilcoxon [29] signed-rank test (Appendix C) for statistical comparison of infection prediction between the two frameworks.

05 for all three buildings, indicating that the paired infection-risk estimates differ systematically between VIRIS and VIRIS-3D-L. p. 8
Where the paper says it

The variance in infection risk is generally higher (> 1) in VIRIS-3D-L than in VIRIS. Additionally, the maximum bounds of the box plots are lower in VIRIS than in VIRIS-3D-L across all four entrytime intervals, indicating significant underestimation in VIRIS, ranging from 29% to 45% relative to VIRIS-3D-L.

The one-sided test results presented in Table 3 showed that the corresponding p-values for the underestimation hypothesis are below 0. p. 8
Where the paper says it

These results further strengthen the recommendation to use VIRIS-3D-L for infection risk prediction. Further, we use a one-sided Wilcoxon [29] signed-rank test (Appendix C) for statistical comparison of infection prediction between the two frameworks.

05. This confirms that VIRIS systematically underestimates infection risk relative to VIRIS-3D-L for all three buildings. p. 0
Where the paper says it

We compare VIRIS and VIRIS-3D-L across three single-storey buildings with increasing architectural and people-movement complexity: a courtroom, a care home, and a supermarket. We find that VIRIS underestimates the disease transmission rate and infection risks for individuals by up to 50% relative to VIRIS-3D-L in all three buildings.

The median difference in infection risk was consistently negative across all buildings, indicating that VIRIS predicts higher infection risk estimates than VIRIS-3D-L. p. 8
Where the paper says it

The variance in infection risk is generally higher (> 1) in VIRIS-3D-L than in VIRIS. Additionally, the maximum bounds of the box plots are lower in VIRIS than in VIRIS-3D-L across all four entrytime intervals, indicating significant underestimation in VIRIS, ranging from 29% to 45% relative to VIRIS-3D-L.

The results further strengthen the recommendation to use VIRIS-3D-L to predict infection risk. p. 8
Where the paper says it

2), we also find that the more realistic airborne transmission modelling in VIRIS-3D-L predicts significantly higher infection risk and a higher disease transmission rate. These results further strengthen the recommendation to use VIRIS-3D-L for infection risk prediction.

Final Recommendation

Overall, VIRIS-3D-L is preferable for infection risk prediction and intervention assessment because it predicts higher breathing zone concentrations and greater spatial variation in risk. The lower estimates of infection risk and the disease transmission rate in VIRIS are unfavourable for informing policy during an outbreak. p. 8

Where the paper says it

Moreover, VIRIS underestimated the disease transmission rate compared to VIRIS-3D-L. The lower estimates of infection risk and the disease transmission rate in VIRIS are unfavourable for informing policy during an outbreak.

This would overestimate safe exposure durations and lead to interventions that are insufficient to reduce infection risk during an outbreak. p. 8
Where the paper says it

The lower estimates of infection risk and the disease transmission rate in VIRIS are unfavourable for informing policy during an outbreak. This would overestimate safe exposure durations and lead to interventions that are insufficient to reduce infection risk during an outbreak.

The paper concludes that VIRIS-3D-L is preferable for infection risk prediction and intervention assessment, particularly in settings with varying people and furniture heights or frequent people movement. p. 9
Where the paper says it

Future work could validate VIRIS-3D-L against measured aerosol concentrations and outbreak data, investigate airflow-driven advection in well-ventilated spaces, and quantify how local and global representations of ventilation removal affect infection risk. Overall, VIRIS-3D-L is preferable for infection risk prediction and intervention assessment, particularly in settings with varying people and furniture heights or frequent people movement.

VIRIS remains useful for rapid screening of simple single-storey buildings when lower computational cost is essential. The paper is organised as follows. p. 1
Where the paper says it

Additionally, the extension to VIRIS-3D-L enables the study of multi-storey buildings in which vertical heterogeneity and connections between floors can affect infection risk. The paper is organised as follows.

p. 1
Where the paper says it

Additionally, the extension to VIRIS-3D-L enables the study of multi-storey buildings in which vertical heterogeneity and connections between floors can affect infection risk. The paper is organised as follows.

The paper is organised as follows.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Improve airborne transmission modeling by extending VIRIS to a three-dimensional domain to account for vertical concentration gradients, which addresses limitations in predicting aerosol concentrations at individuals’ mouth level and variations due to differences in people and furniture heights.

  2. Enhance surface deposition modeling by replacing the first-order global removal term with a Robin boundary condition representing localized removal due to surface deposition, leading to a more physically realistic representation of aerosol removal.

  3. Increase predictive accuracy for infection risk by using the VIRIS-3D-L framework, as it is shown to predict significantly higher infection risk and disease transmission rates than VIRIS across all tested buildings.

  4. Enable study of multi-storey buildings by extending the reaction-diffusion equation to a 3D domain, allowing researchers to investigate how vertical heterogeneity and connections between floors can affect infection risk.

  5. Provide more realistic assessment tools by using the statistical comparison results, which show that VIRIS significantly underestimates infection risk compared with VIRIS-3D-L for all three buildings, guiding policy by avoiding insufficient disease control measures.

Abstract

The Viral Infection Risk Indoor Simulator (VIRIS) [Xue et al. 2024] combines architectural layout, people movement, and airborne transmission modelling to predict infection risk in indoor spaces. The modelling assumptions in VIRIS include: (i) uniform aerosol concentration over the building height, allowing airborne transmission to be modelled in a two-dimensional (2D) domain, and (ii) first-order global removal of aerosols to account for surface deposition. Both assumptions lead to underestimating the infectious aerosol concentration in the breathing zone. Here, we extend VIRIS to model airborne transmission more realistically and compare the impact on infection risk predictions. We develop VIRIS-3D-L by extending the airborne transmission model in VIRIS (a reaction-diffusion equation) to three dimensions (3D) and replacing global removal with a Robin boundary condition representing localised removal due to surface deposition. We compare VIRIS and VIRIS-3D-L across three single-storey buildings with increasing architectural and people-movement complexity: a courtroom, a care home, and a supermarket. We find that VIRIS underestimates the disease transmission rate and infection risks for individuals by up to 50% relative to VIRIS-3D-L in all three buildings. However, VIRIS-3D-L requires, on average, four times the computation time of VIRIS, mainly because it extends the computational domain from a 2D surface to a 3D volume. Overall, VIRIS-3D-L provides a more realistic modelling of airborne transmission than VIRIS and is particularly suited to settings with variations in people and furniture heights and frequent 3D movement. For simpler single-storey buildings, VIRIS is suitable because of its lower computational cost.

Related papers