Impact of wind-induced resuspension on urban air quality: a CFD study with air quality data comparison

Impact of wind-induced resuspension on urban air quality: a CFD study with air quality data comparison

Introduction

Air pollution in urbanized areas caused by particulate matter (PM) is an ongoing problem with increasing severity. WHO has classified the air pollution as the most significant environmental risk of the century, causing 4.2 million premature deaths in 20191. Based on United Nations report, 99% of people breathes PM concentrations, which do not satisfy the WHO recommended threshold values2. The origin of the pollution is mostly intense human activity3, mostly related to the burning of fossil fuels. On average, 45% of the PM particles in cities comes from traffic and residential heating4. Due to climate change policies the transition to electromobility and shift away from the reliance on solid fuels for heating are being promoted today5. However, those will not solve the entire issue with traffic emissions, as non-exhaust sources of particles will remain or even increase6. These sources, including wear particles from traffic are problematic to be reduced by mitigation measures (Amato et al.7) including street cleaning or building green infrastructure8,9. In addition, what is often overlooked is the natural source of particles, in particular resuspension caused by wind (sometimes wind induced resuspension is considered as anthropogenic source of particles because in cities it lift up particles, previously created by human activity, new European directive PE-88-2024-REV-1 defines resuspension as natural source of PM). Resuspension happens when previously deposited particles are lifted from the surface due to strong flows. One particle can repeatedly be airborne and cause air pollution10. Under suitable conditions, it can be responsible for up to 18% of total PM10 concentration4. The continuous production of particulates in urban environments, combined with the fact that climate models predict potentially more frequent and intense drought episodes due to higher temperatures, could increase the share of this source from the overall particle emissions.

There are two major approaches to resolve wind-induced resuspension in urban environments. In the first approach, long-term data series from ambient air quality monitoring stations are analyzed to create a characteristic U-shaped PM concentration curve as the function of wind speed, see Fig. 1. Grundström et al., Harrison et al., Kassomenos et al., Linda et al.11,12,13,14 pointed out, that this function has two independent PM components. One component decreases with increasing wind speed due to intense turbulence mixing the air and the other increases, with particle resuspension being the source. From this relationship, the wind speed at which resuspension arises and its contribution to the pollution can be described by extracting the part of the relationship where PM concentration rises after previous decline–resuspension curve. This curve has its own characteristics, namely threshold wind speed causing resuspension (TWSR), maximal concentration increase (resuspension intensity), and average slope. It should be noted that the impact of traffic is accounted for in these studies through correlation with NOx concentrations. Therefore, the concentration increase can be attributed to wind-induced resuspension.

Fig. 1: Wind speed vs. PM concentration relationship and extracted resuspension curve.
Impact of wind-induced resuspension on urban air quality: a CFD study with air quality data comparison

The properties of resuspension curve are depicted too. The threshold wind speed causing resuspension – TWSR, extracted at the point when concentration starts to rise, average slope of the curve and maximal concentration increase evident from the data.

Full size image

The second approach uses computational fluid dynamics (CFD) to describe how resuspension happens10,15. CFD can compute the wind flow field within the city and especially near the surface to resolve the parameter called wall shear stress (WSS). WSS directly determines whether resuspension occurs at a particular surface16. Banari et al.10 found that resuspension would happen in a complex city environment, when wind speed outside the city exceeds 7 m/s at 40 m height above the ground. Surfaces prompt to resuspension are near flow acceleration or flow separation and reattachment zones. Similar results were reported by Linda et al.15. Both studies agree that resuspension spreads the particles over significant distances, polluting large surrounding areas. CFD is preferred air quality model to address the pollutant dispersion in the cities17. Its greatest privilege in the context of resuspension is its ability to distinguish the flow in great detail, even close to the wall15. There are many studies summarizing the results of using CFD to address air pollution problems in cities18,19. A review of detailed methods used for modeling of the pollution dispersion in cities using CFD is summarized in chapter 1.3 of the study Linda et al.15.

Both approaches identifying resuspension has some shortcomings. When resuspension is analyzed using air quality data, a heterogeneous time averaging step is used. Some datasets contain hourly averaged data12, some one-minute averaged data11. Even though available studies related to resuspension accounts for background concentrations, worth mentioning is the fact that also background concentration has high wind speed and wind direction variability20. Subsequent dispersion of particles within the city with multiple PM sources can create heterogeneous concentration levels, variable with wind speed21. Also, in some periods in Europe, desert dust can lead to high PM concentrations22. Therefore, wind induced resuspension is not the only source variable with wind speed. Also, vehicle-induced resuspension and traffic in general can scatter the data. Numerical modeling studies Banari et al. and Linda et al.10,15 provide only information weather resuspension happen or not on a specific surface. These studies use Lagrangian approach to disperse the resuspended particles but does not contain emission factor of resuspension, therefore provides information about particle motion, not concentration levels. The two existing approaches represent a top-down and bottom-up ways to resolve resuspension while both having factors that may scatter their results and their results were never compared. Therefore, the question is whether both of these approaches sufficiently describe resuspension. It is not clear how much PM pollution can wind-induced resuspension cause and how long resuspended particles remain in the air. The purpose of this study is to use the combination of analytical resuspension model and CFD flow model to predict concentration levels and its variability in a complex urban environment. CFD model can resolve areas where resuspension happens in detail, while analytical model of resuspension add emission factor of resuspension using custom functions within CFD to compute the dispersion and concentration levels. Resuspension model requires detailed information of the flow, which CFD can resolve to produce concentration levels caused by resuspension. Through this approach, numerical model can be compared with the analysis of resuspension from air quality data.

The contribution of wind-induced resuspension to the concentration levels was only studied through urban air quality data analysis. This approach can be burdened with error, as many sources contribute to the overall pollution level. Only two studies resolve resuspension within the city using CFD. However, they only deal with particle motion after resuspension, so concentration levels are not computed. Resuspension occurs only in specific areas and is strongly time dependent. It is not clearly defined how much PM pollution can wind-induced resuspension cause. In this study, the developed model combines a CFD model with a resuspension model to calculate the wind flow field in urban environments. It predicts areas where resuspension occurs and uses an adjusted emission factor for resuspension, which includes the silt load value. This approach defines the total resuspended mass, allowing for the computation of concentration levels, which can be directly compared with air quality data.

Because resuspension is a multidisciplinary subject, the theoretical background about the process itself needs to be highlighted before the results section. These paragraphs explain the process in its micro-scales, mention previous approaches to model the resuspension and discuss the input parameters to the model. Resuspension itself happens at a micro-meter size range at the surface. A particle is exposed to hydrodynamic forces resulting from the air flow and by adhesion forces resulting from particle-surface interaction23. Hydrodynamic forces are often generalized into a parameter called friction velocity ({u}^{* }) [m/s], representing the intensity of the flow near the surface. Particle-surface interaction is described by threshold friction velocity ({u}_{{th}}^{* }), when exceeded, particle break the bond24. Breaking the bond does not mean that particle enters the main flow, it can undergo complex motion near the wall or even redeposit. Surface roughness is the most important factor determining the ({u}_{{th}}^{* }) and the near wall motion25. Yet it is not implemented in micro-scaled models for complex surfaces. It was found that surface roughness with size equal to a size of particle can double the ({u}_{{th}}^{* }) values26. Particles can also form multilayer deposits, where the adhesive forces increase and where collisions can increase the number of resuspended particles27.

Friction velocity ({u}^{* }) [m/s] is defined as ({u}_{* }=sqrt{frac{{tau }_{w}}{rho }}) 28, where (rho) [kg/m3] is fluid density and ({tau }_{w}) [Pa] is WSS. WSS represents friction against the wall and is equal to the velocity gradient near the surface and a viscosity (mu) [Pa.s], ({tau }_{w}=mu frac{{du}}{{dy}})28. On the other hand, ({u}_{{th}}^{* }) depends on the properties of fluid medium, particles, and surface.

There are several studies dedicated to resuspension modeling in micro-scale and are summarized in Henry et al.23. These models are based on definition of lift, drag and adhesion force. Turbulence and surface roughness are distinguished through statistical distributions. A variety of particle separation mechanisms are used (sliding, rolling, lift off). The result of the micro-scale resuspension models are functions defining the number of separated particles in time at certain flow velocity. The prediction and experimental observation agree that particles undergo resuspension in tenths or units of seconds if flow exceeds ({u}_{{th}}^{* })29. The smaller the particle and the bigger the surface roughness, the more time is needed to detach substantial fraction of particles30. However, these models have limitations regarding surface roughness size which should be smaller than the particle23. Therefore, they cannot be applied to dispersion models in real environments, where surface roughness have both macro and micro features. Also, these models require the definition of a wide range of parameters and their distributions that are not precisely known in real environments and cannot be used in CFD models due to the high level of detail.

Micro-scale models can only be used if ({u}_{{th}}^{* }) is exceeded and particle breaks the bond with the surface. However, particle motion can lead to immediate redeposition, which is not considered by these models. Perhaps the most applicable resuspension model is the empirical model from Loosmore31, derived from studies which measure resuspension in a wind tunnel. Particle concentration data measured at the end of the tunnel are used in this study. Therefore, only particles which enters the main flow are captured by the measurement. The final function incorporates particle diameter (d) [m], friction velocity ({u}^{* }) [m/s], particle density ({rho }_{p}) [kg/m3], surface roughness height ({z}_{0}) [m] and time (t) [s] with resuspension rate (varLambda) [1/s]. Resuspension rate is a flux of particles from surface, divided by the initial surface concentration and is defined as31:

$$varLambda =0.42frac{{u}_{* }^{2.13}{d}^{0.17}}{{t}^{0.92}{z}_{0}^{0.32}{rho }_{p}^{0.76}},[1/{rm{s}}],$$
(1)

Probably the most important factor affecting resuspension is the amount of the material being resuspended and its properties. Previously deposited particles in cities form what is often called road dust. Road dust has high size and shape variability, ranging from 132 to 1000 μm33. The density varies between 1.3 – 1.9 g/cm3 34,35. With decreasing particle size, the density also decreases35, for PM10 a value of 1.3 g/cm3 is the most probable. Smaller fraction of road dust which is often studied related to traffic emissions is called slit load (sL). US-AP-42 defines sL as the amount of particles with size below 75 μm extracted from one square meter of a road surface. However, it is believed that only particles with a size below 10 μm are subjected to resuspension process36. In size fraction of sL below 10 μm (sL10), there is less particle material on the surface. The value of sL is highly variable, it decreases with increasing traffic intensity37 and is dependent on meteorological conditions38 and surface type37. High values of sL are observed in dryer climates, on roads with relatively high aggregate size (surface roughness) and near a source of significant pollution, eg. construction site36. Amato et al.37 found the sL10 value ranging from 0.85-12.3 mg/m2 on roads with around 1000 vehicles per day. Similar variable range 0.49-50 mg/m2 of sL10 was found by39. The value can reach almost 100 mg/m2 or more for some areas40. In some extreme cases (unpaved roads), sL can even reach few grams per square meter41. All studies agree, that sL value for less used roads can be significantly higher. No study reports the sL10 values for surfaces like parking lots or sidewalks, or other urban surfaces, where sL may be significantly higher. Besides the fact, that rougher surface can accumulate more particles (even creating multilayer deposits)37, there are other factors that could be considered. However a list of factors affecting resuspension could be very long, but is summarized well in Henry et al. and Linda et al.15,42.

Results

Evaluation of the flow field

After 20,000 iterations, the solution achieved stability and convergence across all cases involving three geometries and four different flow regimes. The highest residuals were observed at ({u}_{{ref}}=15) m/s, with the intermittency for the case H/W = 0.25 being the largest, stabilizing at value of 2.10-4. Other residuals reached stability at values of Sdr 10-12, Tke 10-7, X-momentum 10-7 and Y-momentum 10-7. Residuals, excluding intermittency, were consistent across flow regimes and varied only slightly across geometrical configurations, remaining below 10-6. The variability in intermittency can be attributed to vortex shedding at the building edges, which increases with increasing wind speeds43. The structured mesh, with the finest settings at the bottom of the domain, resulted in the lowest y+ values, favorable for resuspension modeling. For ({u}_{{ref}}=8.75) and 10 m/s, y+ values were below 5 at the bottom of the street canyon. For ({u}_{{ref}}=12.5) and 15 m/s, a small area beneath the vortex center has y+ slightly higher than 5. Most cells in the first mesh layer are within the viscous sublayer, and for cells with y + > 5, the wall treatment ensures accurate wall functions calculations. Surfaces within the street canyon array, such as building sides and roofs, generally have y+ values below 10. In this study the mesh was constructed similarly as in Linda et al.15, grid sensitivity analysis was provides there.

Due to relatively high inlet wind speeds, a canyon vortex forms in all cases, with no other flow regimes observed44. The flow within the street canyons becomes steady after the fourth canyon. The first three canyons are strongly influenced by upstream buildings, creating unstable flow regimes. However, turbulence production and transfer cause wall shear stress (WSS) to increase with each subsequent canyon up to the seventh canyon, where WSS stabilizes. The values of friction velocity ({u}^{* }) across the street canyon width matches previous studies (Linda et al.15) for similar ({u}_{{ref}}) validating the flow state. Even though ({u}_{{ref}}=15) m/s was not used in that study, it aligns well with the results. It is visible from Fig. 2 that the increase of ({u}^{* }) with rising ({u}_{{ref}}) is nonlinear, with the peak located beneath the vortex center due to high velocity gradients near the surface. In the surfaces further from the vortex center, the main flow detaches from the surface and ({u}^{* }) decreases. As the canyon width increases, the vortex center shifts toward the windward side. Foe each case the main vortex is off-center, and smaller vortexes are formed near the canyon edges (for H/W = 1, 0.5 two vortexes are present and only one for H/W = 0.25). These smaller vortexes create local extremes of ({u}^{* }), while the lowest ({u}^{* }) values occur between the main and secondary vortexes. Only the main vortex generates ({u}^{* }) > ({u}_{{th}}^{* }).

Fig. 2: Flow field and corresponding friction velocity profile.
figure 2

For each configuration, vector velocity field inside the steet canyon for ({u}_{ref} = 10) m/s is depicted. Corresponding friction velocity profile inside the last canyon of street canyon array is displayed for each of the inlet wind speeds.

Full size image

To quantify the resuspension process, the ({u}_{{th}}^{* }=0.142{m}/s) was applied. In prior studies (Linda et al.15) a minimum ({u}_{{ref}}=7.5) m/s was required for ({u}^{* }) > ({u}_{{th}}^{* }) in a few surface cells. Consequently, the lowest ({u}_{{ref}}=8.75) m/s was used here, where the surface area with ({u}^{* }) > ({u}_{{th}}^{* }) is more substantial. In each geometry configuration, ({u}_{{ref}}=10) m/s was needed to expose approximately 50% of the surface to resuspension. As ({u}_{{ref}}) increased more surfaces particularly on the windward side are exposed. At ({u}_{{ref}}=15) m/s over 60% of the surface is subject to resuspension. For surface cells, where ({u}^{* }) > ({u}_{{th}}^{* }) particles are introduced to the bottom surface of the last street canyon (9th for H/W = 1, 0.5 and 7th for H/W = 0.25) with corresponding emission factor.

Particle dispersion and deposition

After the particles were injected into a domain, with the emission factor defined by ({u}^{* }), model functionality was assessed. The largest number of particles with constant diameter and density was released from the surfaces near the center of the street canyon vortex, leading to the emission of the largest mass parcels. These parcels, each with a unique ID, allowed tracking of their origin, and their mass was calculated using Eq. 5. Simulation results aligned with manual calculations. Particle behavior near surfaces was also monitored. For regions where ({u}^{* }) > ({u}_{{th}}^{* }), particles rebounded, although some, with lower velocities, reattached to the surface. These particles may resuspend again in real conditions, but this effect is not captured in steady-state simulations. Similarly to Linda et al. and Roy et al.15,45, early deposition occurred for particles originating from low ({u}^{* }) areas on the leeward side of buildings. The total particle dispersion was completed in approximately 1-2 minutes.

Resuspension creates unique concentration fields, visible from Fig. 3. Particles travel to the leeward side of the canyon, resulting in higher concentrations ({C}_{{PM}10},[mu g/{m}^{3}]), consistent with prior studies46,47 on street canyon pollution. Some particles exit the canyon, while those returning often hit the windward wall, causing lower concentrations and deposition. This poses a hazard, as redeposited particles can be resuspended over and over with each wind shifts. Although the concentration field shape remains similar as ({u}_{{ref}}) increases, ({C}_{{PM}10},[mu g/{m}^{3}]) intensifies. In the H/W = 1 case, particles predominantly travel along canyon edges due to strong velocity gradients near the vortex center. As the H/W ratio decreases, particles disperse more towards the center, particularly at H/W = 0.25, where lower velocity gradients and longer residence times lead to a more homogeneous concentration field. At least ({u}_{{ref}}=15) m/s was needed to reach ({C}_{{PM}10}) > 1 μg/m3 for heights under 2 m in all monitoring areas and geometry cases. In case with H/W = 0.25, ({C}_{{PM}10}) > 1 μg/m3 even in case with ({u}_{{ref}}=12.5{m}/s). Concentrations were lower in other cases starting at around 0.1 μg/m3 for ({u}_{{ref}}=8.75) m/s. Computed concentrations are valid for an sL value of 250 mg/m², with values scaling proportionally for different sL. Resuspension can contribute significantly to dust levels, with long-range particle transport observed, see Fig. 4. Most notable for H/W = 0.5. In contrast, in H/W = 1, particles circulate longer, and at H/W = 0.25, early deposition dominates.

Fig. 3: Concentration field values for every case studied.
figure 3

Because concentration has very different values in different places, the logarithmic scal is used, this better portrays the shape and elements of the concentration field.

Full size image

Fig. 4: Long range transport of the particles after resuspension event in cases with ({u}_{{ref}}=15) m/s.
figure 4

Note the difference between concentration fields inside and outside the canyon, meaning only a fraction of particles undergo long range transport.

Full size image

Comparison between CFD results and air quality analysis results

To compare the model results with studies analyzing resuspension from air quality data, resuspension curves were extracted and analyzed from studies Harrison et al., Kassomenos at al., Linda et al.11,12,13. While many curves were available, attempts to generate an average resuspension curve for a single location and year were unsuccessful. Full datasets would be needed to create such curves not to avoid losing unique details. Figure 5 shows how the process of averaging affects the curves. The whole process of data evaluation is provided in supplementary material. In addition, few things come up in the consideration when resuspension properties were studied. A key issue in these studies was the spatial mismatch between meteorological and air quality measurements, with distances ranging from a few meters to kilometers. Given the variability in concentration and flow fields, these results were excluded for comparison purposes, though they still highlight general relationships relevant to resuspension. These insights help explain long-range particle transport from urban resuspension. Only results from Linda et al.11 allowed matching concentration and wind speed at a single location, aligning with the model results. The resuspension curves from this study were steeper than those in others, beginning at similar wind speeds but peaking at lower wind speeds, likely due to differences in averaging times (10-minute averages vs. hourly averaged data) and methodologies.

Fig. 5: Resuspension curves extracted from11.
figure 5

Resuspension curves valid for different day periods are displayed on the left and average resuspension curve for each location in the right.

Full size image

Resuspension curves were derived from simulations as the average concentration over a 1 × 1 m area spanning the street canyon width, at heights of 1.5–2.5 m and 2.5–3.5 m. This yielded 20, 40, and 80 resuspension curves for H/W = 1, 0.5, and 0.25, respectively, as shown in Fig. 6. For H/W = 1 and 0.5, the curves were consistent in shape but shifted relative to one another. However, significant variation was observed at the canyon edges due to high boundary layer concentrations computed in very small cells. The shape of the curves is convex for H/W = 1 and concave for H/W = 0.5. For H/W = 0.25 the curves are convex at the leeward side and windward side and concave for canyon center. Except for the leeward side in the H/W = 0.25 case, the resuspension curves had TWSR > 1m/s. High concentrations were computed on the canyon’s leeward side, besides that the maximum concentration increase remained below 1 μg/m³. There were no significant differences between the upper and lower monitoring areas, except for H/W = 0.25, where wind speeds and concentration values were lower in the upper areas.

Fig. 6: Position and indexing of the monitoring areas within street canyon.
figure 6

For each of these areas, the concentrations are extracted.

Full size image

The resuspension curves along the street canyon width were similar for H/W = 1 and 0.5, while H/W = 0.25 showed distinct differences. Based on the canyon flow field, the lower monitoring areas exhibited higher TWSR in most cases, with TWSR decreasing from the windward side, peaking near the canyon vortex center, and decreasing again towards the leeward side. The resuspension curve slopes followed a similar pattern, decreasing from the windward side and increasing near the leeward side. Concentration increases were also higher in the lower row, reflecting particle distribution during resuspension, with fewer particles entering the vortex center. This trend was most evident in the H/W = 0.25 case, while in other cases, large and abrupt changes in the flow field masked differences across monitoring areas, see Fig. 7.

Fig. 7: Selected resuspension curves and their properties.
figure 7

Note that H/W = 1 has 10 curves, H/W = 0.5 have 20 curves and H/W = 0.25 have 40 curves. The difference between the lower and upper row of monitoring areas is visible and is related to the nature of the flow near the center of the canyon vortex. The biggest difference between the values occurred in the monitoring areas very close to the wall, where particle concentration reaches is peak values. U*th stands for threshold wind speed, slope for average slope of the curve and C for concentration increase due to resuspension.

Full size image

The model values were compared with air quality data analysis to assess their similarity. Since standard validation methods could not be applied, box plots were used to visualize multiple statistics and compare value distributions, including potential outliers. Extreme values can occur in measurements, in a non-standard situation but also in the model often due to high concentrations in the boundary layer cells. As shown in Fig. 8 by comparing mean values, the model underpredicts resuspension properties, including TWSR by 30%, slope by 70%, and concentration increase by 64.5%. Such variation is typical in resuspension studies, whether from measurements, models, or theoretical predictions. Although differences exist, the overlap between model and measurement results is significant, especially for higher TWSR, steeper slopes, and greater concentration increases usually measured near the leeward side, see Fig. 7. Under standard circumstances, measuring stations are located near buildings, not in the middle of the canyon. Here the model-measurement agreement improves. However, on the windward side, lower values for slope, TWSR, and concentration increase are evident. These types of curves are also visible in Fig. 5. In real situations, the influence of a more frequently occurring wind direction will dominate, generating more data points and ultimately shaping the resuspension curve.

Fig. 8
figure 8

Comparison of the resuspension curves properties predicted by the model and by the measurements.

Full size image

There are many factors which can explain the great discrepancy between the model and the measurements. In terms of TWSR, complex flow features around the station can affect the obtained values, also meteorological parameters like humidity can significantly increase the TWSR. In terms of slope and concentrations increase, these arise from the amount of dust deposited on the ground. In the study, sL = 250 mg/m2 was used, which is perhaps still lower than in real situations. More measurement would be needed to address distribution of sL values in the variety of surfaces in the city. The wind-induced resuspension is a phenomenon, which can create and array of results. The specific shape of the resuspension curve will then depend on the location of the measurement site in the city and the meteorological conditions, as well as the amount of settled dust.

Generalization of results

Resuspension is a complex phenomenon, challenging to predict in terms of its occurrence and concentration increase. Previous studies have demonstrated that concentration levels accountable to resuspension and the shape of resuspension curves are influenced by numerous factors, particularly the measurement location and the amount of dust settled on the ground. Real-world measurements of air quality are not fully capturable by models, and precise values of sL on variety of surfaces are not available. Even though a great uncertainty, both experimental and model data provide values that can help generalize the shape of resuspension curves. When comparing the results of studies and models, it is evident that these values fall within a specific area. Resuspension initiates within wind speeds of 0.75 to 3.5 m/s, with average slopes ranging from 0.25 to 6.4. The area forms a trapezoidal shape, with the shorter side at the top, which in extreme cases may resemble a 1/x curve shape. This suggests that curves with steeper slopes are associated with lower TWSR values, and vice versa. The concentration increase is estimated to range from a few tenths to lower tens of μg/m³, with an average of 3.99 μg/m³. Despite its local nature, resuspension patterns can be summarized to predict future behavior in similar conditions, as illustrated in Fig. 9. The overlapping area indicates potential points at which additional measurements would be displayed or the results of a more complex 3D model would predict resuspension curves. Factors like monitoring area location and flow features nearby affect both position on x and y axis of the Fig. 9. Climate of the location affects the postion on x axis and the value of sL affects the y xis position.

Fig. 9: Combined results of all data currently available on urban wind-induced resuspension.
figure 9

The points indicate the exact results of the previous measurements or model, while shaded area indicated the range of values, where other measurements of results of the model might fall in.

Full size image

Discussion

Pollutant dispersion in street canyons is a well-researched area in CFD, with model settings carefully studied in detail48. Both 2D and 3D geometries have been explored, revealing areas of high particle accumulation49. However, modeling resuspension requires a different approach. A fine mesh was designed to meet y + < 5 criteria, although in some cases, particularly with ({u}_{{ref}}) = 15m/s lightly exceeded this value. Higher y+ values on roof edges, where no particle sources exist, do not affect the results. The fine mesh, where the first cell lies within the viscous sublayer, captures boundary layer turbulence without relying on approximations50. Increasing turbulence variables across each consecutive canyon establish favorable conditions for resuspension. The WSS profile along the canyon width, shaped by velocity and turbulence, stabilizes in the seventh canyon and reaches peak values15. The results of flow field rely mostly on turbulence model. Alternative models, such as the Lattice Boltzmann Method combined with near wall flow corrections, have been used in specific studies to adjust near-wall velocity values10. But the SST k-ω turbulence model with standard wall functions still remains suitable for studies of wind flow in cities. Although lower wind speeds would produce a different flow regime, the wind speeds in this study were sufficient to form the vortex and trigger resuspension by overcoming the threshold friction velocity44. No thermal effect are considered, which may change a flow field a bit and perhaps disperse the particles from the surface faster due to the upward flow51. Heated surface can also affect adhesion forces, but there is no study confirming this related to resuspension in cities. Interestingly, the highest ({u}^{* }) values can be found in H/W = 0.5 configuration, which aligns with the ideal formation of a canyon vortex. These ({u}^{* }) peaks increase nonlinearly with wind speed, explaining the near-exponential shape of the resuspension curves12. This follows the established relationship where the resuspension emission factor (EF) depends on ({u}^{* }) with an exponent of 2.1331. In 3D geometries, resuspension is also expected near building edges, where flow acceleration occurs10,45,52. Resuspension model configuration is crucial with ({u}_{{th}}^{* }), which was set to 0.142 m/s based on an average of studies on resuspension from air quality data11. While analytical expressions, such as those used for volcanic dust properties53, could be applied, these were not directly suitable due to differences in particle characteristics such as shape and density36, which are most important factors affecting adhesion forces54.Experiment in wind tunnels also reports ({u}_{{th}}^{* }) = 0.155, for particles on the concrete surfaces. Slight changes in ({u}_{{th}}^{* }) would not significantly affect the area for resuspension, given the steepness of the ({u}^{* }) profile across the street canyon width. It would also not affect concentrations much because most of the particles are produced in the areas with high ({u}^{* }) values. In terms of emission factor (EF) expression, there exists no better equation than that reported by Loosmore31. The EF model used here, although not perfect, remains one of the best available for CFD. Many other EFs either overlook dust properties or are too complex for CFD integration. The most important factor in determining the magnitude of resuspension is the amount of deposited material sL, which varies widely across urban areas41. While milligram-level deposits are typical on busy roads56, sL can be much higher on less-used roads40, not mentioning the situation on the sidewalks etc. More research needs to be done to address the sL distribution in the city. In this study, the uniform distribution was considered, but if the distribution of sL across the street canyon would be known, it can be incorporated in the model. Importantly, sL doesn’t affect the threshold wind speed causing resuspension (TWSR) values but does influence the resuspension curve’s steepness and concentration levels. This highlights the risks associated with high-dust activities like construction57,58.

Resuspension-induced concentration fields share similarities with those from traffic or other sources59,60. Concentration is always higher in the leeward side of the canyon with peak values located in the side vortex near the bottom61. However, minor differences remain. Concentration fields from resuspension show a higher value along the surface, as particles tend to stay on the vortex’s edges and frequently contact the walls. It takes substantial time (several seconds) for the particles to move from the boundary layer into the mainstream. Accumulation along canyon sides, also observed in traffic dispersion cases46, indicates that low sL values measured on roads do not capture the full picture, reinforcing the need for surface measurements beyond roadways. Several studies have attempted to predict resuspension curves, but their variability, even within a single site, makes it difficult to generalize. Seasonal and annual resuspension curves13, when averaged, show differences due to the methods used to fit mathematical functions to measured data. Not all measured data accurately reflect the resuspension process, which complicates curve generation15. Starting with raw measured data, rather than derived functions, would yield more accurate curves, as these show very low R2 and do not describe the data very well14. The resuspension curves from Harrison et al. and Kassomenos et al.12,13 show concentrations at higher velocities than Linda et al.11, which may be since the concentration and meteorological parameters were measured at different locations. Therefore, these results were not used in the comparison. However, the characteristics of the curves do not differ significantly between the studies, as shown in Fig. 9. Resuspension curves derived from air quality studies can be scattered by a few factors. Background concentrations tend to change with wind direction and wind speed and with changing conditions, the contribution in specific site can vary, affecting the shape of resuspension curves. Also, vehicle-induced resuspension and other sources can have a wind speed dependence, which is not considered in these studies. More precise measurement would need to be done to address this issue. The resuspension curves in this model, although variable, align with those derived from air quality data. Moving from the windward to the leeward side, TWSR decreases, while concentration and slope increase. However, concentration increases are less reliable due to their dependence on wind speed. TWSR and slope provide more stable indicators. The model predicts a broad range of resuspension behaviors, influenced by monitoring area locations. Extreme values are more common in the model, potentially because it covers the entire street width, unlike typical monitoring stations located near buildings or roadsides in the real world. TWSR results show the closest alignment with real-world data, while deviations in slope and concentration may be influenced by the variability in sL. Barari et al. and Roy et al.10,45 have shown that resuspension affects not only road surfaces but also areas outside of roads, where higher particle deposition occurs.

The main question of the article was to answer how much pollution can resuspension cause and how long can particles remain in the air by comparing the model to air quality data results. Resuspension is a complex process evaluated primarily through air quality measurements, which introduce complications when applying inverse models like CFD. Despite these challenges, CFD remains a useful tool for capturing resuspension, and the similarities between model predictions and measurements are promising. Though both approaches have limitations, they provide complementary insights into the phenomenon, enabling a better understanding of resuspension dynamics. Because resuspension is such a complex and variable process, dependent on the local factors, the concentration contribution of resuspension is not limited to a single number. The opposite is true, the answer to the question of when resuspension happens and how much pollution it produces is an array of values. Based on the model results compared to the air quality data, the contribution can reach almost 20 μg/m3 in extreme cases. However, most of the results from both the model and measurements falls below 6 μg/m3. This array of results is dependent on the value of sL, position of monitoring station and climate as reflected in Fig. 9. The results of this study represent only specific parts of this array. The highlighted area in Fig. 9 shows where results from further studies, either more complex 3D CFD models or other measurements, would show up. Time dependence, following a 1/t decline, was factored into the EF, with values averaged over a 60-second period to match real-world data42. If the model were also compared with the values from Harrison et al. and Kassomenos at al.12,13, it would have to be adjusted so that the EF is averaged per hour, not per minute as it is now. Linda et al.11 using minute data predicts steeper resuspension curves more often, which may just be due to the shorter measurement periods and significant time decline of resuspended particles. However, the comparison between all the results even with hourly average data in Fig. 9 shows, that resuspension curves properties does not differ that much between each other. This could indicate, that particles originated in resuspension stays in the air for significat amount of time. The gradual inclusion of surfaces due to random occurrence of turbulence in large cities may explain why resuspension does not occur in a short time span, but contributes to pollution relatively long time.

Resuspension is difficult to assess due to its dependence on highly variable environmental factors. Air quality data, which rely on long-term measurements, are subject to seasonal changes in particle properties62,63. Rainfall38, wind direction and other variables complicate resuspension predictions64,65. These variabilities are reflected in resuspension curves, making them difficult to standardize. In cities, other particle sources can also influence concentration, further distorting resuspension data64,66. The primary limitation of the current model is its computational complexity, which restricts it to 2D. A 3D model would better capture the variety of flow regimes and improve the accuracy of resuspension predictions10,45. Differences between 2D models and real-world conditions, such as edge effects along buildings or longitudinal street effects, remain unaccounted for47. Strong turbulent structures present in 3D model can affect both TWSR and concentration increase. Also, in larger 3D geometries, particles can be brought to monitoring areas from greater distances affecting the shape of resuspension curves4. While resuspension is still not fully explored, particularly regarding emission factors, the EF used in this study provided satisfactory results. The main limitation of the study is a lack of data from cities which identifies resuspension from air quality data. More cities should be involved in those studies reflecting differences in climate. Also, the lowest possible data time step should be used to address the resuspension dynamics. The discrepancy between the results and the models can also arise from the estimate of the sL value, far more measurements in different types of surfaces should be done to reveal the distribution od sL values in the city. Resuspension is a complex multiphysical process, analytical model of resuspension can be refined, if more research is done in the laboratory conditions, pairing flow parameters to the particle properties and their concentrations. These measurements could also be done in changing conditions, with a heated surface for example. Complex models of resuspension should also involve dynamics, so they would not be done in steady conditions. All these limitations affect the comparison between the model and the measurements. This study reveals how much can resuspension contribute to the total concentration in cities. These results can be directly used in the new European air quality directive PE-CONS 88/1/24 REV 1, which defines that concentration of natural sources of PM can be subtracted from total concentration in case of exceedance of air quality standard. Figure 9 can be used for this purpose. Also, cities could take measures aiming to reduce the amount of deposited PM, which would significantly reduce resuspension impact. Measures can also be directed towards more general measures reducing the particle production.

Methods

This study builds upon our previous work Linda et al.15, with similar CFD settings (geometry, mesh, turbulence model and other CFD flow model settings). Grid sensitivity analysis and indirect validation of the model was carried out in this study (direct validation wasn’t possible due to much higher inlet wind speeds and different turbulence modelused, for which data are not available for comparison). The scope was expanded to include additional cases with higher wind speed. The resuspension model was refined by adding emission factor of resuspension and calculation of concentration levels using Lagrangian method of dispersion. These additional functions implemented in CFD allows the model to be compared with the studies analyzing the resuspension from air quality data. CFD settings are briefly summarized here, with detailed descriptions available in Linda et al.15. Also, because the air flow models which are used here are standardized and were not changed by any kind, a detailed description of those is not provided but can be found in the manual of the STAR CCM+ software.

The model is composed of three parts. First is the computational model, which involves standard CFD methods to compute the air flow field, and the crucial parameter wall shear stress. WSS in the form of friction velocity then enters the second part, resuspension model, which defines where and by with what emission particles are released to the domain. Afterwards, the third part, multiphase model define the trajectory of the particles and concentration levels. From concentrations, resuspension curves are created and compared to available studies. Conceptual diagram of the model can be found in the supplementary material.

Computational model

Wind in urban geometries can create complex flows67, but a well-known case in CFD is the street canyon vortex, which forms when wind flows perpendicular to street between two buildings48. A 2D model was developed to simulate detailed flow and particle dispersion, similar to Linda et al.15. For street canyons with length B [m], width W [m], and building height H [m], 3D edge effects can be neglected if B/W > 20 for H/W = 1 and B/W > 50 for H/W = 260. Three configurations H/W = 1, 0.5, 0.25 with H = 10 m (common values) were used to maintain a consistent particle-to-environment size ratio. Although the H/W = 2 configuration was studied in Linda et al.15 wind flow wasn’t strong enough to initiated resuspension. The computational domain dimensions were 50 x H in length and 10 x H in height (due to software domain limitations 500 m in STAR CCM + ). Studies agree that modeling at least four street canyons is required to avoid the influence of upstream building affecting the vortex formation15,52,68. nine street canyons were modeled for H/W = 1 and 0.5, and seven for H/W = 0.25. Geometry of the domain for each configuration is depicted in the supplementary material, and for H/W = 0.5 in Fig. 10.

Fig. 10: Geometry of the domain for H/W = 0.5 configuration together with the mesh details.
figure 10

The figure on lower left describes the mesh structure in the street canyon (the structure is governed by the parameter “cell growth rate”, which enlarges the cell size with distance from the wall). The structure of the near wall mesh is visible on the lower right figure (boundary layer mesh includes 8 prismatic layers with variable thickness). Note that the first near wall cell has size of 0.008 × 0.0008 m to achieve very low y+values.

Full size image

With high wind speeds that causes resuspension, an intense turbulence and flow separation and re-entrainment zones are expected within a street canyon50. Since the particle motion during resuspension is influenced by the wall-adjacent region, a fine mesh is required in these areas. Following Linda et al.15 an unstructured polygonal mesh was generated near the surfaces, with a very fine mesh applied close to the walls to capture the boundary layer accurately. The objective was to achieve y + < 5 in regions prone to resuspension, requiring multiple mesh refinements. The first cell is then located within viscous sublayer and mesh captures the flow and velocity gradient in the boundary layer. Then the calculation of wall functions and particle motion near the wall are resolved adequately69. Mesh parameters included a first-cell height of – 0.8 mm, eight prismatic layers with a total thickness of 0.024 m, a minimum surface size of 0.004 m, and a cell growth rate of 1.05. These parameters define the mesh in the near wall region, in other regions, mesh is constructed as the extension of this fine mesh. Across all cases, identical mesh settings were applied, with approximately 1.3 to 2 million cells used per case.

Surface roughness significantly influences the resuspension process. In this study, the surfaces were modeled as smooth, no-slip walls, with roughness effects incorporated into the resuspension model through the parameter ({u}_{{th}}^{* }). A zero-gradient condition was applied at the outlet boundary, while the upper boundary was treated as a symmetry plane. The inlet velocity profile can be uniform, follow a power law, or logarithmic. Based on previous research on street canyon flows and pollutant dispersion, a power law profile was selected60,70.

$$u={u}_{{ref}}{left(frac{z}{{z}_{{ref}}}right)}^{n},left[frac{m}{{rm{s}}}right]$$
(2)

where ({u}_{{ref}}) [m/s] is the flow velocity in the height of ({z}_{{ref}}=H) [m] with an exponent n = 0.21 following experimental investigation in Tominaga & Stathopoulos71. Turbulence intensity was set at 1.8%, corresponding to open atmospheric flow conditions28. As demonstrated in Banari et al.10, resuspension occurs under severe wind conditions, meaning ({u}_{{ref}},) cannot be limited to values below 5 m/s. In the article multiple inlet wind speeds were tested, namely ({u}_{{ref}}=5,,7.5,,8.75,,10,,12.5,,15) m/s. At least ({u}_{{ref}}=7.5) m/s was needed to overcome ({u}_{{th}}^{* }) in very small areas, releasing only few particles disrupting the dispersion characteristics. Because of this, the lowest ({u}_{{ref}}) needs to be higher. In this study, ({u}_{{ref}}=8.75,,10,,12.5,,15) m/s are used for each geometry case. In more complex 3D geometries, lower inlet wind speeds could still induce resuspension due to high vorticity and flow acceleration in specific zones of the building layout, namely near the edges of the buildings45. No thermal effects are considered affecting flow features. Atmospheric stability can change the formation of eddies in street canyon leading particles to disperse to a wider area51.

The simulations were conducted using STAR CCM+ (commercial version 2023 17.06.008-R8) with multiple user-defined field functions. To accurately capture the resuspension process, the flow was resolved down to the viscous sublayer using a suitable turbulence model. While the RANS approaches combined with k-ε model is commonly used in street canyon pollutant dispersion59, it has been shown that the complex LES approach aligns better with experimental data50. Given the simplicity of the geometry and flow features in this, however, RANS approach combined with the SST k-ω two-equation eddy-viscosity turbulence model was employed. SST k-ω model combines the advantages of the k-omega model near walls and the k-epsilon model in the free-stream region. It introduces a blending function to smoothly transition between these two models, improving accuracy for boundary layer flows. A key feature is its inclusion of the shear stress transport term, which enhances the prediction of flow separation compared to standard models. Given the high inlet wind speeds and expected flow separation, the SST k-ω model was selected for its ability to better predict wall shear stress72,73. Additional settings included the SIMPLE algorithm for pressure-velocity coupling, second-order interpolation and discretization schemes, and all y+ wall treatment. The all y+ wall treatment dynamically selected the most accurate wall function based on local y+ values, which varied throughout the domain due to the structured mesh near surfaces. The simulation ran for 20,000 iterations to achieve stable and minimal residual values.

Resuspension model

As discussed before, detailed resuspension models and general dispersion equations are not directly applicable in CFD simulations. Instead, a hybrid approach that incorporates both the threshold friction velocity ({{rm{u}}}_{{rm{th}}}^{* }) and an emission factor for resuspension should be employed, particularly for particles such as road dust. Although road dust exhibits considerable variability in size and density, PM was treated as solid spherical particles with constant density of ({rho }_{p}=1.3frac{g}{{m}^{3}}) and a diameter of ({d}_{p}=10,mu m) (as mentioned in Section 1.2). While no precise formula exists for calculating ({u}_{{th}}^{* }), it can be estimated through various methods, including wind tunnel experiments. In these experiments, ({u}_{{th}}^{* }) can be derived from the logarithmic wind profile at the point where resuspension initiates:

$${U}_{z}=frac{{u}_{* {th}}}{kappa }.,{ln}left(frac{z-{d}_{h}}{{z}_{0}}right)[{rm{m}}/{rm{s}}],$$
(3)

where ({U}_{z}) [m/s] is the flow velocity at a certain height above the surface, (kappa) [-] is the Karman constant, (Z) [m] is the height above the surface, ({d}_{h}) [m] is the height of the surrounding obstacles, and ({z}_{0}) [m] is the surface roughness height. Nicholson55 studied PM10 resuspension from concrete surfaces and reported ({u}_{{th}}^{* }=0.1{m}/s), grassy surface was studied in Giess et al.74 and reports ({u}_{{th}}^{* }=0.3{m}/s), ({u}_{{th}}^{* }=0.8{m}/s) was found by Garland75 investigating resuspension from bare soil. Studies focusing on outdoor environments, such as those examining volcanic dust with varying particle sizes and humidity levels, found ({u}_{{th}}^{* }=0.3-0.38frac{m}{s})76. Cornelis & Gabriels77 experiments on sand dunes dunes reported ({u}_{{th}}^{* }=0.3-0.483{m}/s). Similar methodologies have been applied to urban environments using air quality measurement data, which suggest ({u}_{{th}}^{* }=0.11-0.21{m}/s). While some studies have developed semi-empirical relationships for calculating ({u}_{{th}}^{* }), their applicability is limited to the specific conditions (e.g., particle type, surface, and environment) for which they were derived, such as volcanic dust resuspension in open landscapes78,79,80. The most relevant values of ({u}_{{th}}^{* }) are those obtained in urban environments or wind tunnels with corresponding surface conditions, with an average value of ({u}_{{th}}^{* }=0.142{m}/s), as used in other studies15,45.

Emission factor of resuspension could be defined by resuspension rate calculated by31 with initial number of particles on the surface:

$$E={sL}* S* 0.42frac{{u}_{* }^{2.13}{d}_{p}^{0.17}}{{t}^{0.92}{z}_{0}^{0.32}{rho }_{p}^{0.76}},left[frac{{kg}}{s.{m}^{2}}right]$$
(4)

where sL is silt load [g/m2] on surface from which particles are released, dp is particle diameter [m], ({u}_{* }) is friction velocity [m/s], ({rho }_{p}) is particle density [kg/m3] and ({z}_{0}) is surface roughness size [m] (proportional to 0.3 mm for concrete surface according to Loosmore31. The value of sL significantly influences the total emission from resuspension. Previous studies mainly focused on collecting particles from road surfaces37,81, though wind can also lift particles from other paved areas like sidewalks and parking lots. Urban paved surfaces tend to accumulate more particles37. It has been found that only a few vehicle passes can resuspend a significant proportion of road dust resulting in lower sL values when measured directly from roads. After resuspension, most particles settle near the road15, creating higher values of sL on surfaces like sidewalks etc. Limited data exist on sL variability from non-road surfaces36,82. Resuspension’s contribution to concentration levels varies throughout the day and seasons, likely due to PM availability13. Although non road surfaces may accumulate more particles than traffic studies suggest, no definitive conclusions can be drawn. Based on estimates from various studies, a value of sL = 250 mg/m² is used uniformly across the computational domain, as discussed before. In steady-state simulations, particle concentration is directly proportional to sL, meaning halving sL (e.g., to 125 mg/m²) would halve the concentration.

Since air quality monitoring data are reported in one-minute intervals, the steady-state emission factor will be calculated for the same period. It can be determined by equating the total mass of particles released over a 60-second interval from the unsteady emission factor. Thus, the emission factor can be defined as:

$$begin{array}{c}mathop{int }limits_{0}^{60}{{rm{E}}}_{60}{rm{dt}}=mathop{int }limits_{0}^{60}{rm{sL}}* {rm{S}}* 0.42frac{{{rm{u}}}_{* }^{2.13}{{rm{d}}}^{0.17}}{{{rm{t}}}^{0.92}{{rm{z}}}_{0}^{0.32}{{rm{rho }}}_{{rm{p}}}^{0.76}}{rm{dt}}\ {E}_{60}=frac{{rm{sL}},*, {rm{S}},*, 0.42frac{{{rm{u}}}_{* }^{2.13}{{rm{d}}}^{0.17}}{{{rm{z}}}_{0}^{0.32}{{rm{rho }}}_{{rm{p}}}^{0.76}}mathop{int }nolimits_{0}^{60}frac{1}{{{rm{t}}}^{0.92}}{rm{dt}}}{60}=2,4713.{10}^{-7}{{rm{u}}}_{* }^{2.13}[frac{{rm{kg}}}{{rm{s}}.{{rm{m}}}^{2}}]end{array}$$
(5)

Description of multiphase flow

A Lagrangian approach was used to track motion of individual particles in the flow field. This approach also allows to model particle–surface interactions. Elements known as parcels are used to model the particles. Each parcel, with a mass determined by the emission factor, represents a group of particles with identical properties. This significantly improves computational efficiency by reducing the number of elements needed to represent the total particle population. To ensure that the results are independent of the number of parcels, a sensitivity analysis was conducted. Starting with 100,000 parcels and gradually increasing to 1,500,000, particle concentrations at specific points were tracked until convergence was achieved. Consequently, 2,000,000 parcels were used in each simulation to ensure a correct description of particle dispersion.

Ziskind24 suggests that the initial velocity of particles can be estimated as the ({u}^{* }) value, with their initial position set at one particle diameter above the ground. Since CFD cannot resolve such small scales, particles are inserted into the domain in a single sub-step through a set of points, positioned at the centroids of the first cell, 0.4 mm above the surface where ({u}_{{th}}^{* }) was exceeded. The last street canyon is selected as the particle source due to stable, fully developed flow15. The total mass released from each point is determined by the emission factor from Eq. 5. The number of points range from a few hundred to several thousand, evenly distributes the parcels. Since the emission factor only accounts for particles entering the mainstream from the surface, detailed near-wall particle motion does not need to be resolved in the simulation. Processes like particle detachment and near-wall motion are not fully applicable in CFD30.

To accurately model particle behavior, distinct physical properties must be defined in the CFD model. During the particles motion, Schiller drag force and Saffman lift force influence their trajectories. Since particle concentrations were not expected to significantly affect the flow, two-way coupling was not used. Wall interactions were modeled in two ways: when ({u}^{* }) exceeded the threshold, particles were allowed to rebound from the wall using a custom function; otherwise, they adhered to the surface, with their positions recorded. Due to the RANS approach’s inability to resolve all turbulent eddies, a turbulent dispersion model was applied, introducing an eddy velocity fluctuation ({u}^{,}) is to account for local disturbances to the mean RANS flow field83. This method is widely used for pollutant dispersion modeling in urban environments45,59,84,85. Given the significant time required for particles to move from their initial positions, the maximum residence time was set at 100,000 s. The iterative nature of steady-state particle movement calculations required a maximum of 200,000 substeps, and a 2nd-order tracking integration method was employed. Because standardized models are used for the multiphase flow, details with equations are not provided but can be found in Linda et al.15.

After calculating particle trajectories, custom functions were used to derive key variables, particularly concentration values. In Lagrangian approaches, concentrations are not computed by default; instead, a 2D volume fraction is calculated as the ratio of area occupied by a given phase to the total cell area. For consistency with measurements (3D environment), surface length was used in place of surface area in Eq. 5, assuming homogeneity in particle distribution in third dimension. With a structured mesh near the surface, each cell had an 8 mm length. Finally, mass concentration was calculated as ({c}_{p}={V}_{f}.{rho }_{p},)[kg/m3], where ({V}_{f}) [-] is volume fraction of particles (in in 2D calculated as proportion of the area occupied by a particular phase relative to the total area of the cell in the grid).

The resuspension model is therefore defined by I. areas, where resuspension happen by defined by ({{u}}_{{th}}^{* }), II. by emission factor to defining the total resuspended mass and III. by calculation of concentration from Lagrangian particle tracking.

$${boldsymbol{I}}{boldsymbol{.}},{u}_{{th}}^{* }=0.142left[frac{m}{s}right],{boldsymbol{II}}{boldsymbol{.}}{E}_{60}=2,4713.{10}^{-7}{{rm{u}}}_{* }^{2.13}left[frac{{kg}}{s.{m}^{2}}right]{boldsymbol{III}}{boldsymbol{.}},{c}_{p}={V}_{f}.{rho }_{p}$$
(6)

Comparison with the results of other studies

The comparison between the model and studies focused on resuspension identified from air quality data may be imprecise for several reasons. One key difference lies in the use of a 2D model compared to the complex 3D real-world environment, where varying flow patterns, wind directions, and speeds affect particle behavior. Additionally, the value of sL fluctuates throughout the day and across seasons, and background concentrations are influenced by changing wind conditions. Other sources of PM within the city may also contribute to total PM concentrations. Moreover, studies differ in their data properties and the timescales of their results, ranging from seasonal to annual assessments, making direct comparisons with the model challenging. Attempts were made to run unsteady simulations in 2D, but these resulted in several weeks of computation per simulation despite significant simplifications. Similarly, efforts to conduct 3D simulations across various geometries led to such long computational times that the simulations had to be terminated after weeks of processing.

There is a way to compare the model results by analyzing the dispersion of PM resulting from resuspension. When wind speed is sufficiently high, particles resuspend, leading to specific concentration patterns at a given location. These concentrations can be measured to establish a relationship between wind speed and PM concentration, forming a well-known resuspension curve. In the process of dispersion, the information about where the particles are separated from at certain value of ({u}^{* }) is lost (in laboratory experiments31, where resuspension is measured, the emission from a particular surface can be directly related to the resulting concentration). For each geometry, four inlet wind speeds ({u}_{{ref}}=8.75,) (10,,12.5,,15) m/s were simulated, resulting in four distinct concentration patterns. Resuspension curves were then generated by extracting concentration data from specific points. The concentration values were measured from the height of 2 m in Linda et al.11 and 3,3 m in Harrison et al. and Kassomenos et al.12,13, while in CFD, concentrations were calculated for each grid cell. Due to potential variations between adjacent cells, concentrations were averaged over regions of 1 × 1 m, termed “monitoring areas,” located in the last street canyon. Two rows of monitoring areas were placed along the entire width of the street canyon at heights of 1.5 to 2.5 m and 2.5 to 3.5 m, which aligns with typical PM measurement heights. These areas were indexed systematically: “E” for the edges, “C” for the center, “L” for left side and “R” for right side of the canyon with the upper row having an additional number in its index, see Fig. 6. The number of areas between the edge and center varied depending on the height-to-width (H/W) ratio of the canyon: 1-3 for H/W = 1, 1-8 for H/W = 0.5, and 1–18 for H/W = 0.25. For each monitoring area, geometry, and wind speed, PM concentrations were extracted from the simulations, resulting in resuspension curves based on four points per monitoring area.

The resuspension curve properties including TWSR, average slope, and maximum concentration increase (as shown in Fig. 1), from the simulation are then compared to those derived from air quality measurements. Since the simulation does not provide a direct TWSR value (because the first point of the curve at ({u}_{{ref}}=8.75) m/s already has a concentration value), TWSR is determined by linearly extending the first two points of the curve to where the concentration reaches zero. All the available resuspension curves are used from the studies11,12,13, while their properties extracted. Efforts were made to group up the seasonal resuspension curves to annual ones, but with no satisfactory results. The whole process of data extraction is described in Supplementary material.

Related Articles

The decreasing housing utilization efficiency in China’s cities

‘Ghost cities’ are a well-known phenomenon of (almost) complete vacancy of urban living space in China. Underutilization of urban living space, however, is far more common than complete vacancy. Here we propose the concept of housing utilization efficiency (HUE) and present the following findings: (1) the overall HUE in China’s highly urbanized areas decreased from 84% in 2010 to 78% in 2020, (2) the HUE in central, old urban areas was generally lower than that in the outer layers of urban areas and declined more from 2010 to 2020 and (3) four development types are found to represent different patterns of urban population movement, urban housing growth and HUE change at the intraurban level. These findings provide comprehensive insight into the discrepancies between urban housing supply and demand in China and highlight their connections to the country’s particular urbanization characteristics and policies, which are crucial for future housing development and planning.

The multidimensional relationship between renewable energy deployment and carbon dioxide emissions in high-income nations

Cross-national research has found that the decarbonization effect of renewable energy development is relatively weak in high-income nations. It is crucial to identify effective points of intervention to enhance renewables’ decarbonization effect. Using a multidimensional analytical framework, this study examines whether certain structural components of high-income nations’ CO2 emissions are particularly susceptible to barriers to decarbonization and therefore are less effectively mitigated by renewable energy development. Analyzing a panel dataset covering 33 high-income nations from 1996 to 2019, I identify a pattern of uneven decarbonization. Renewable energy development has mitigated production-based emissions with increasing effectiveness over time; however, the mitigation effect has been largely confined to emissions from domestic-oriented supply chain activities. Meanwhile, renewables’ inability to mitigate emissions embodied in exports and direct end-user emissions has largely persisted over time. Additionally, developing renewable energy has not spurred growth in emissions in imports, indicating that it has not intensified carbon leakage.

Unequal roles of cities in the intercity healthcare system

Cities are increasingly interdependent regarding healthcare provision and demand. However, the intercity healthcare system (IHS) behind the nationwide patient mobility remains insufficiently understood. Here, leveraging human mobility big data, we reveal cities’ roles in providing and demanding quality healthcare within the IHS of China. We find that 8% of Chinese cities are national and regional hubs that address the healthcare shortage of cities deprived of quality healthcare, while 63% of the cities that are unnoticed compensate for migrant workers being denied healthcare rights in megacities. The IHS generates new structural inequalities in healthcare access exhibiting a Matthew effect. The few cities (12%) that are already rich in healthcare resources benefit more and can strengthen their advantages in providing healthcare to local populations (32% of China’s total population). The many cities (35%), while facing healthcare shortages, are further disadvantaged in ensuring adequate healthcare for their local populations (26% of China’s total population).

Machine learning map of climate policy literature reveals disparities between scientific attention, policy density, and emissions

Current climate mitigation policies are not sufficient to meet the Paris temperature target, and ramping up efforts will require rapid learning from the scientific literature on climate policies. This literature is vast and widely dispersed, as well as hard to define and categorise, hampering systematic efforts to learn from it. We use a machine learning pipeline using transformer-based language models to systematically map the relevant scientific literature on climate policies at scale and in real-time. Our “living systematic map” of climate policy research features a set of 84,990 papers, and classifies each of them by policy instrument type, sector, and geography. We explore how the distribution of these papers varies across countries, and compare this to the distribution of emissions and enacted climate policies. Results suggests a potential stark under-representation of industry sector policies, as well as diverging attention between science and policy with respect to economic and regulatory instruments.

What, how and where: an assessment of multi-level European climate mitigation policies

The European Green Deal’s goal of making Europe the first climate-neutral continent by 2050 requires an adequate mix of policies. This paper analyses three decades of climate policy from a historical perspective to provide insights into the multi-level policy framework within the EU and its Member States. Based on the Climate Policy Dataset, the paper develops an assessment guided by three key perspectives: policy density, sectoral focus, and policy instruments. Two new indexes are proposed for policy evaluation: the emissions coverage indicator, which assesses the sectoral application of policies, and the Policy Mix Thickness Index, which measures the complexity of the policy packages in terms of instruments employed. The results indicate that different strategies have been adopted at the EU and national levels in terms of policy instruments and targeted sectors. EU-level policies tend to complement Member States actions by providing long-term strategies and addressing sectors with limited national-level initiatives.

Responses

Your email address will not be published. Required fields are marked *