Roof dust accumulation characteristics and influencing factors in typical urban areas of Beijing

Abstract

Understanding roof dust and its environmental drivers is essential for interpreting urban particulate dynamics. We collected 159 roof- and ground-dust samples across a university campus in central Beijing (December 2020–November 2021) as a representative urban setting to examine vertical and seasonal patterns. Dust mass, PM₁₀ and PM_2.5 concentrations were analyzed against meteorological conditions. Clear seasonality emerged, with dust mass peaking in February–March and reaching minima in summer–autumn. PM₁₀ correlated strongly with total dust mass (r = 0.773), indicating a prominent coarse-particle contribution to rooftop deposition. Sectoral winds modulated accumulation: southeast and south-southwest flows enhanced dust loading, whereas north-northwest winds suppressed it. A wind-speed threshold was observed, with rooftop loads increasing above ~ 2.5 m s^–1 and strongest accumulation at > 5 m s^–1. Precipitation frequency and relative humidity were inversely related to dust mass (r = − 0.755 and − 0.773), consistent with wet removal and reduced resuspension. A multivariate synthesis (PCA/PCR) further shows that a high-RH, warmer and low-wind composite state is significantly and negatively associated with roof dust, supporting denser roof sampling and cleaning during dry, windier periods and highlighting roof deposition as a complementary urban monitoring metric. Roof dust accumulation generally decreased with increasing building height, although unexpectedly high dust loads were observed on mid-rise buildings, likely due to higher building density and restricted local air circulation. These findings demonstrate that roof dust accumulation is strongly modulated by seasonal meteorology, wind regimes, and urban morphology. The vertical distribution patterns identified here provide a basis for integrating roof sampling into urban air quality monitoring and for refining models of particulate dispersion in dense cityscapes.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-27439-2.

Introduction

With the acceleration of urbanization and the intensification of activities such as building construction, transportation development, coal combustion, and waste incineration, the emission of urban dust has continued to rise. This increase has led to excessive concentrations of atmospheric particulate matter (PM), posing significant environmental challenges in urban areas^1,2. As a major air pollutant, PM is not only a key indicator of air pollution levels³, but also has well-documented adverse effects on both ecosystems and human health^4,5. Although air quality in cities such as Beijing has improved over the past decade⁶, it still lags behind that of developed countries, with PM₂.₅ and O₃ posing serious compound pollution threats⁷. PM_2.5 and PM₁₀ pollution remain a persistent global environmental concern and continues to pose a substantial risk to public health^8–10.

Controlling PM emissions remains central to air pollution mitigation, and many studies have focused on near-surface PM monitoring or the characterization of surface dust, such as road dust. For example, Zhang et al.¹¹ found that open sources—such as soil, road dust, and cement—contributed up to 50% of PM₁₀ in Chinese cities, occasionally reaching as high as 70%. A study by Farah et al.¹² in Beirut, Lebanon, highlighted the notable impact of dust deposition on PM₁₀ and PM_2.5. However, relatively little attention has been paid to particulate matter accumulation in vertical urban spaces, such as building roof.

Roof, as elevated and relatively undisturbed deposition surfaces in urban environments, can serve as indicators of long-term atmospheric PM accumulation and migration patterns. They offer significant potential for evaluating environmental risks and resuspension behavior. Existing research has demonstrated that building height plays an important role in the vertical distribution of air pollutants. Ma et al.¹³ observed pollutant accumulation at altitudes exceeding 100 m due to anthropogenic activities. Similarly, Zou et al.¹⁴ reported that elevation gradients of underlying surfaces affect the geochemical cycling of urban particulates and may influence both surface runoff and atmospheric pollution patterns.

Roof dust, acting as a passive sampler of atmospheric PM, can reflect the temporal integration of regional pollution through its chemical composition (e.g., heavy metals, carbon fractions). While rainfall can temporarily improve air quality, roof dust—once disturbed by strong winds or precipitation—can be resuspended or transformed into secondary aerosols, leading to delayed or indirect contributions to pollution. This process may exacerbate urban air pollution and pose health risks through inhalation exposure^15,16. The mechanisms governing roof dust pollution are complex and strongly influenced by meteorological conditions that affect its accumulation, dispersion, and transpor¹⁷. Li et al.¹⁸ further demonstrated the impact of climate change on dust levels in arid regions of the Middle East, highlighting broader atmospheric influences.

However, comprehensive research on the spatiotemporal variation, meteorological drivers, and vertical distribution patterns of roof dust remains limited, especially in northern Chinese cities. Long-term, multi-level observational datasets are lacking, and most existing studies are confined to single-height or short-term samples, which cannot fully capture the dynamics of vertical particulate transport and deposition in complex urban settings.

To address the existing knowledge gaps in urban dust accumulation dynamics, this study conducted a comprehensive year-long sampling campaign of roof and ground-level dust at 15 sites spanning various building heights (low-, mid-, and high-rise) within a representative urban campus in central Beijing. While the spatial extent of sampling is confined to a single campus, this area typifies the urban environment of Beijing’s central districts in terms of building density, morphology, and meteorological conditions. By integrating meteorological data and PM concentration indices, we employed statistical, graphical, and physical analyses to investigate the relationships between roof dust and its influencing factors. Our findings aim to provide a novel perspective for understanding urban vertical dust processes and offer theoretical support for managing potential resuspension risks of roof particulate pollution.

Materials and methods

Sample collection and data source

Beijing, the capital of China and the focus of this study, experiences a temperate, sub-humid continental monsoon climate characterized by hot, humid summers, cold, dry winters, and short transitional autumns¹⁹. Due to its location in northern China, where arid and semi-arid regions with sparse vegetation are prevalent, Beijing is also occasionally affected by dust storms, especially in spring²⁰. The unique geographical location, complex climate conditions, and human activities have contributed to the complexity of dust storms in Beijing²¹. The dust accumulation sampling points (Fig. 1) were located on the main campus of Capital Normal University, situated on the West Third Ring Road in the Haidian District, Beijing (39° 57’N, 116° 17’E). Although sampling was confined to a single campus, the site typifies the urban core of Beijing in terms of its dense surrounding traffic network, mixed residential–commercial land use, and building morphology. These features make it a representative microcosm of the city’s central urban environment. Sampling locations included roof of low-, mid-, and high-rise buildings, as well as ground-level points. At each site, a 1 × 1 m wooden frame was placed, with its base lined with an antistatic, non-adhesive polyolefin (PO) film. The PO film, primarily composed of polyethylene, has a smooth surface that resists particle adhesion. This setup allowed the dust to settle naturally within the frame, facilitating uncontaminated sample collection. The frames were secured with concrete bricks to safeguard them from potential disturbances caused by strong winds or adverse weather conditions. The dust that accumulated on the PO film was meticulously collected using a plastic brush and stored in envelopes made of sulfuric acid paper for analysis. From December 2020 to November 2021, a total of 159 dust samples were collected from 15 locations in the study area (comprising 14 roof sites at various heights and one ground level site). The samples included measurements from all seasons: 44 in spring, 43 in summer, 44 in autumn, and 28 in winter. The winter samples were fewer compared to other seasons due to the inability to collect them in time during the Spring Festival holiday and the COVID-19 pandemic. Therefore, dust accumulation from February and March was collected together. Corresponding meteorological data, including daily average concentrations of PM_2.5 and PM₁₀, were obtained for the same period (2020–2021) from the Haidian Wanliu Environmental Quality Monitoring Station in Beijing (http://www.bjmemc.com.cn). These data, recorded hourly from 00:00 to 23:00 each day, were processed using clustering algorithms to remove outliers, and the remaining values were averaged to represent daily concentrations. Subsequently, the data were processed and analyzed on a monthly timescale to ensure comprehensive results.

Fig. 1.

Locator maps (upper) and campus sampling locations (lower). Maps produced by the authors in ArcGIS Desktop 10.8 (Esri; https://www.esri.com/arcgis/). Administrative boundaries from GADM v4.1 (CC BY 4.0; https://gadm.org). The lower panel uses OpenStreetMap vector data (© OpenStreetMap contributors, ODbL; https://www.openstreetmap.org) with no proprietary basemap; symbols and labels were added by the authors.

Sample handling and methods

Sample handling

All samples were collected under dry-weather conditions (no precipitation for at least 3 days before sampling). After collecting the dust samples, they were meticulously cleaned using only plastic tweezers to remove extraneous materials, such as plant tissues, stone fragments, insect remains, and other debris. Subsequently, larger particles were filtered through a 200-mesh sieve. The sieved samples, based on previous studies and standard experimental methods, were placed in an oven set at 105 °C for 2 h to ensure completely dehydration. After drying, the samples were preserved in a desiccator to maintain a dry state. For accurate quantification, each sample was weighed twice using an electronic balance with a precision of 1000th of 1 g. The two measurements were validated to ensure they fell within a ± 0.001 g error margin. All weighing instruments were calibrated with certified reference weights before each sampling campaign, and paired blank samples were used for quality assurance. This ensures that the results are more accurate and lays a solid foundation for future research, such as the analysis of heavy metal sources in dust accumulation. The consistent weight between the two readings was recorded as the final sample weight.

Dust weighing

Sampling frequency was once per month, with 3–5 days of exposure at each site to capture representative deposition. The weight of the dust collection samples on the roof was determined using a unit square meter and specific sampling days. The formula is as follows:

1

where D represents the dust accumulation,, W₁ denotes the weight of the aluminum box after sweeping the dust accumulation (g), W₀ designates the weight before the aluminum box is not placed in the dust accumulation (g), and n comprises the sampling day, accurate to 0.1 day.

Results and discussion

Characteristics of roof dust accumulation in Beijing

The collected dust samples were processed and weighed using the Dust Deposition Gravimetric Method. This method allowed the calculation of roof dust deposition levels across the study area for each month from December 2020 to November 2021. The detailed data are presented in Fig. 2. According to the figure, the monthly average dust accumulation on low-rise buildings (1–5 floors) varied from 0.10 to 3.38 t/(km²·30 day), whereas for mid-rise buildings (6–9 floors), it ranged from 0.12 to 5.49 t/(km²·30 day). For high-rise buildings (10–14 floors), the accumulations spanned from 0.06 to 2.18 t/(km²·30 day). The annual average dust accumulation was 1.58 t/(km²·30 day), with notable monthly variations.

Fig. 2.

Comparison of dust accumulation at different heights in different months.

Due to the impact of the Spring Festival and the COVID-19 pandemic, only one month’s dust accumulation was recorded for February and March. Meanwhile, February to March also coincided with the peak dust accumulation on the roof, which was 3.27 t/(km²·30 day), coinciding with Beijing’s sandstorm season. Historical meteorological data revealed that in March, Beijing experienced 17 d of wind speeds above Level 2 (1.5–3.3 m/s), stirring up considerable amounts of dust. The most intense dust storm over the past decade, according to the National Satellite Meteorological Center’s (https://www.cma.gov.cn/) meteorological satellite dust detection map and monitoring from the FengYun-4 Series of meteorological satellites, occurred between March 14 and 15, 2021. Originating from northern Mongolia’s interior, this dust storm transported large quantities of dust to the Beijing area, where it settled with the airflow and rain, resulting in a higher roof dust accumulation in March and spring than in other seasons, this observation is consistent with the conclusion made by Gao et al.²², who noted that sandstorms in Beijing primarily occur in spring (March to June). Recent studies have further reinforced these findings. Jin et al.²³ highlighted the intensity of the 2021 dust storms, documenting the dust emissions originating from northern Mongolia and transported to Beijing. Gui et al.²⁴ noted the extreme meteorological conditions that enhanced dust accumulation, while Liang et al.²⁵ described the long-range transport of dust during these events. He et al.²⁶ confirmed the significant impact of the dust storms on air quality across East Asia, underscoring the importance of spring dust storms. Additionally, Kong et al.²⁷ introduced the E20 dry deposition scheme, which has been shown to improve the simulation of dust deposition and PM₁₀ concentrations, helping to reduce uncertainties in dust storm modeling. The lowest dust deposition value of the year, recorded in July, was 0.16 t/(km²·30 day), likely due to Beijing’s high rainfall during this month²⁸.

At the same time, the ground dust (ground dust collection points are the Ground points in Fig. 1) accumulation data from December 2020 to November 2021 in the study area were also recorded. The average dust accumulation per unit area was 16.54 t/(km²·30 day).In terms of spatial distribution, the average roof dust accumulation in Beijing was lower than the ground-level dust accumulation.This is also consistent with Zvi et al.²⁹ research findings, which indicate that the dust accumulation in the atmosphere always decreases with increasing altitude. Although previous studies generally suggest that dust accumulation in the atmosphere decreases with height, our research found that dust accumulation in mid-rise buildings showed greater variation and, in some cases, was even higher than in low-rise buildings. This counterintuitive phenomenon may be related to the higher density of mid-rise buildings in the study area. The concentration of such buildings could reduce local wind speeds, leading to increased dust deposition.Citywide building-height statistics derived from remote sensing show that mid-rise buildings (12–24 m) dominate Beijing by number (~ 50%) and plan area (~ 49%) (Table S1). This city-scale prevalence, together with higher local plan density and stronger frontal area toward the SE or SSW sectors around our sites, is consistent with reduced near-roof wind exposure and roughness-sublayer recirculation, providing data-supported context for the observed mid-rise roof dust anomaly.

In addition to changes in wind speed, the structure of the buildings themselves may also disrupt local wind fields and affect the redistribution of dust. Li et al.³⁰, through large-eddy simulations, found that the aspect ratio of urban street canyons and ground heating conditions significantly altered the airflow and pollutant diffusion within the canyons. This indicates that building morphology, by regulating airflow patterns, influences particle transport and retention at different heights, indirectly contributing to varying dust accumulation levels at different building tiers. Therefore, the higher dust accumulation in mid-rise buildings could result from the combined effects of building density, restricted ventilation, and increased surface roughness.

Seasonal characteristics of roof dust accumulation in Beijing

The average roof dust accumulation across the different seasons was observed as follows: 1.10 t/(km²·30 day) in spring (March to May), 0.19 t/(km²·30 day) in summer (June to August), 0.26 t/(km²·30 day) in autumn (September to November), and 0.88 t/(km²·30 day) in winter (December to February). The average roof dust accumulation is highest in spring, with low-rise buildings having 1.40 t/(km²·30 day), mid-rise buildings having 1.46 t/(km²·30 day), and high-rise buildings having 0.67 t/(km²·30 day), respectively. Generally, the seasonal trend in dust accumulation followed the order spring > winter > autumn > summer, with the dust accumulation in spring can be up to seven times higher than that in the other three seasons. Annually, the roof dust accumulation exhibited a U-shaped fluctuation pattern, peaking in spring, declining from spring to summer, increasing from summer to autumn, and continuing to increase from autumn to winter. This seasonal pattern was consistent with the findings of previous studies.

According to Lin³¹, sandstorms are a typical spring phenomenon in northern China, and their seasonal characteristics are closely related to climatic conditions. Further observational data from Kou et al.³² indicate that in March, due to the dry and windy climate of spring, windblown sand activities are intensified, resulting in the peak dust deposition period of the year. In contrast, June (the wet monsoon period) and December (the surface freezing period) experience the lowest dust deposition, as increased precipitation and surface freezing suppress dust lifting. Thus, the observations in this study were consistent with those of other domestic studies.

Correlation analysis of roof dust accumulation and ambient air PM_2.5 and PM₁₀ in Beijing

In this study, monthly roof dust accumulation exhibited a significant correlation with ambient PM_2.5 and PM₁₀ concentrations. As depicted in Figs. 3 and 4, the monthly variation trends of roof dust accumulation closely align with those of ambient air PM_2.5 and PM₁₀. During January to March 2021, ambient air concentrations of PM_2.5 and PM₁₀ escalated, reaching their yearly peaks in March at 241 and 2600 µg/m², respectively. These concentrations subsequently decreased to lower levels of approximately 179 µg/m² for PM_2.5 and 438 µg/m³ for PM₁₀ by June 2021, remaining stable until October. By November, the concentrations had again risen to 442 µg/m² for PM_2.5 and 738 µg/m³ for PM₁₀. This pattern is in accordance with the findings of Fan et al.³, who analyzed air pollution trends in China and reported low winter and summer PM_2.5, high winter and spring PM₁₀, and low levels in summer and autumn. The roof dust accumulation trend mirrored the ambient air PM_2.5 and PM₁₀ trends. From the observed “U-shaped” fluctuating pattern of dust accumulation from December 2020 to November 2021, it can be concluded that the trends of PM2.5 and PM10 concentrations in the ambient air corresponded closely with the variations in roof dust accumulation. However, the roof dust accumulation and the trends of PM_2.5 and PM₁₀ showed some differences during the months of May and October, possibly attributable to increased dust fallout owing to human activity, such as the increased population movement and other factors related to the national public holidays in May and October. A correlation was noted between ambient air PM_2.5 and the roof dust accumulation, although it was not as strong as the correlation between PM₁₀ and the roof dust accumulation.

Fig. 3.

Relationship between dust accumulation content and PM_2.5 concentration.

Fig. 4.

Relationship between dust accumulation content and PM₁₀ concentration.

Table 1 presents the correlations between the PM_2.5 and PM₁₀ concentrations as well as the roof dust accumulation to demonstrate the relationships between these elements. The data revealed that the roof dust accumulation was most strongly correlated with PM₁₀, followed by PM_2.5. Both correlations passed the 99% significance test and exhibited positive relationships. This result suggests that PM₁₀ and roof dust likely share similar sources and that their content variations are closely aligned.

Table 1.

Correlation coefficient of roof dust accumulation with PM_2.5 and PM₁₀.

Element	PM2.5	PM10
Correlation coefficient	0.773**	0.955**

Note: * * denotes passing the 99% significance test.

To assess external consistency beyond the campus, we aligned monthly rooftop dust with citywide monthly PM₁₀/PM_2.5 (Figs. S1–S2). Dust correlates more strongly with PM₁₀ than with PM_2.5 (Spearman ρ = 0.955 vs. 0.773, n = 11), and the peak occurs at lag k = 0 months (Fig. S2), consistent with this paper’s conclusions on seasonal peaks and wind dependence; thus, the consistency check provides external evidence for generalizability.

Correlation analysis of dust accumulation content and wind characteristics

Wind speed and direction distribution

Figure 5 presents monthly wind rose plots from December 2021 to November 2022, illustrating wind-direction and wind-speed distributions alongside rooftop dust accumulation. February and March are combined in a single panel due to data integration.

Fig. 5.

Monthly wind direction and speed distribution and corresponding rooftop dust accumulation.

The data show that in most months, winds predominantly originated from the south to southwest (S–SW) sectors, accompanied by a broad range of wind speeds. In February–March and October, winds from the southern to southwestern sectors (S–SW) were relatively frequent. In May, southerly winds (S) were clearly dominant, while in July, wind directions were more broadly distributed from east to south (E–S), without a distinct prevailing direction.

Most months experienced moderate winds (2–6 m/s) and gusts exceeding 6 m/s were generally rare. These seasonal fluctuations in wind speed and direction provide essential context for interpreting monthly rooftop dust-accumulation trends.

Correlation between dust accumulation and wind direction

Wind direction strongly modulates rooftop dust accumulation³³. In Fig. 5, February–March shows the highest monthly averages, whereas other months are lower. During this peak period, winds from the SE–S sectors were most frequent. May, another relatively high month, was likewise dominated by southerlies and exhibited elevated dust mass. In contrast, July–August experienced broadly distributed E–S winds and correspondingly lower dust loads. April showed intermediate levels, coincident with moderate southerly and westerly frequencies but remaining below the February–March maximum.

A sector analysis further indicates that SE–SSW inflow tends to enhance rooftop dust, whereas NNW inflow tends to suppress it, consistent with the seasonal peaks above. This sectoral enhancement or suppression matches expectations from advection-dominated urban dispersion behavior: inflow aligned with upwind source sectors elevates near-surface particle fluxes. The observed wind-speed threshold (onset around ~ 2.5 m s^–1, strongest effects at > 5 m s^–1) also accords with parameterizations in which friction velocity u_∗ drives resuspension and mechanical turbulence. Altogether, these patterns suggest that directional wind components can promote rooftop deposition under specific regimes, although the effect is neither uniform nor exclusive and depends on concurrent meteorological conditions.

Correlation between dust accumulation and wind speed

The relationships between wind speed, dust transport and deposition, and the influence of topography on surface wind have been thoroughly established³⁴. Figure 5 uses color bands to indicate wind-speed bins in each monthly rose plot. A correlation analysis was conducted between the varying wind speed frequencies and monthly dust accumulation to identify the wind speed that had the strongest correlation with roof dust accumulation. The resulting data, presented in Table 2, indicate that the correlation coefficient was highest for wind speeds above 5 m/s. Wind speeds above 3 m/s also showed a strong correlation with dust accumulation, whereas speeds below 2.5 m/s were negatively correlated. These results align with the frequency distributions in Fig. 5: the combined February–March period—dominated by moderate winds (3–6 m/s)—coincided with the highest dust accumulation, whereas July and September—dominated by low-speed events (≤ 2 m/s)—showed the lowest accumulation.

Table 2.

Correlation coefficient between dust accumulation content and wind speed.

Wind speed	Less than 2.5 m/s	More than 2.5 m/s	More than 3 m/s	More than 5 m/s	More than 7 m/s
Correlation coefficient	-0.288	0.363	0.696*	0.772**	0.516

Note: *indicates passing the 99% significance test, * indicates passing the 95% significance test.

These findings suggest a potential wind-speed threshold effect, with 2.5 m/s potentially marking the onset of effective dust transport and deposition. Above this threshold, dust accumulation increased with wind speed, suggesting that moderate-to-strong winds play a more significant role in promoting rooftop dust deposition, although this relationship is modulated by wind direction and other environmental factors. These findings are consistent with previous work³⁵ and further validate the combined role of wind direction and wind speed in controlling rooftop dust accumulation.

Correlation between roof dust accumulation and other meteorological elements

After calculating the correlation between dust accumulation, precipitation frequency, relative humidity, and temperature, and passing the 99% significance test, the correlation coefficients were found to be -0.755, -0.773, and − 0.682, respectively these results indicate a strong negative correlation of roof dust with other meteorological elements. Li et al.³⁶ study employed 18 meteorological factors were used to perform a linear fitting on the standardized spring dust storm frequency at 97 monitoring stations in northern China. It was found that factors such as vapor pressure, relative humidity, and minimum temperature were negatively correlated with dust storms, showing a high degree of consistency with our finddings. The detailed correlation with each meteorological factor is analyzed sequentially in the following sections.

Correlation between roof dust accumulation and precipitation frequency

To detail the relationship between the monthly precipitation frequency and roof dust accumulation, we depicted the correlation between the number of precipitation days per month and the corresponding monthly dust accumulation (Fig. 6). The roof dust accumulation peaked when there were only 4 d of precipitation in a month and reached its minimum when the precipitation reached 20 d. Furthermore, the correlation coefficient between roof dust accumulation and precipitation frequency was − 0.755, which passed the 99% significance test. This result demonstrates a strong negative correlation between the roof dust accumulation and monthly precipitation frequency, which is consistent with the findings of Liu and Li³⁷.

Fig. 6.

Relationship between precipitation frequency and roof dust accumulation content.

Correlation between roof dust accumulation and relative humidity

Relative humidity significantly influences dust emission and deposition processes³⁸. As illustrated in Fig. 7, roof dust accumulation reached its peak when relative humidity was between 35% and 40%, and declined to a minimum near 80%. An evident decreasing trend in dust accumulation was observed with increasing humidity.

Fig. 7.

Dust content related to relative humidity and daily and monthly average humidity.

From December 2020 to May 2021, relative humidity remained relatively low (25–50%), coinciding with elevated dust accumulation. Beginning in June, rising temperatures introduced warmer, more humid air and precipitation, leading to increased humidity and a substantial decrease in dust accumulation. During the period from July to October, humidity levels stabilized between 50% and 75%, with dust accumulation remaining consistently low.

Correlation analysis revealed a strong negative relationship between roof dust accumulation and relative humidity (r = -0.773, p < 0.01). This pattern aligns with the observations of Csavina et al.³⁹, who found that dust concentrations peaked near 25% relative humidity and declined at higher levels, likely due to enhanced particle cohesion and increased wet deposition under moist conditions.

Correlation between roof dust accumulation and temperature

Temperature is an indirect but important factor influencing dust processes by affecting relative humidity, aerosol particle properties, and precipitation patterns^39,40. Figure 8 shows the relationship between roof dust content and temperature.

Fig. 8.

Relationship between roof dust content and temperature.

From December to March, roof dust accumulation increased as temperatures rose. This trend likely results from enhanced drying of surface particles and reduced snow cover on rooftops, which facilitate dust resuspension under windy conditions. Even small temperature increases during cold months can promote moisture evaporation, loosening dust particles and increasing their availability for transport.

Between May and October, dust accumulation exhibited a U-shaped pattern, with an initial decline followed by a rebound in October. During this period, higher temperatures were often accompanied by increased humidity and precipitation, which enhance particle cohesion and wet deposition, reducing dust accumulation on rooftops despite elevated temperatures.

Although a moderate negative correlation was observed between air temperature and roof dust accumulation (r = -0.682), this relationship is complex and influenced by seasonal variations in humidity and precipitation. Temperature acts more as an indicator of changing environmental conditions rather than a direct driver of dust accumulation. These findings are consistent with those reported by Liu et al.⁴¹.

The above subsections describe single-predictor relationships for wind speed, relative humidity, and temperature. Given their seasonal co-variation and multicollinearity, we next apply a PCA on the standardized predictors and a principal component regression (PCR) of ln(Dust) on the PC scores to disentangle relative contributions.

Multivariate synthesis of meteorological drivers (PCA/PCR)

Relative humidity, wind, and temperature often co-vary seasonally, making bivariate correlations insufficient to isolate their relative roles. To address this, we performed a principal component analysis (PCA) on standardized Wind Speed, Temperature, and RH and used the leading components as composite predictors of roof dust.

As illustrated in Fig. 9, PC1 loads positively on RH and Temperature and negatively on Wind Speed, representing a high-RH, warmer and low-wind composite state (PC1 and PC2 explain 75.3% and ~ 21.5% of the variance, respectively). Consistent with the humidity-driven suppression described above, principal component regression (PCR) revealed a significant negative association between PC1 and roof dust (β ≈ −0.37, p ≈ 0.05; adjusted R² ≈ 0.26; Table 3), whereas PC2 was not significant. This pattern indicates that more humid and warmer conditions with weaker winds correspond to reduced roof dust accumulation, in line with the expectation enhancement of particle cohesion, wet removal and diminished resuspension under such conditions.

Fig. 9.

PCA loadings for standardized predictors.

Table 3.

Principal component regression (PCR) of ln(dust) on PC1–PC2, with standardized OLS comparisons.

Predictor	β*(A)	p(A)	Partial R²(A)	VIF(A)	β*(B)	p(B)	Partial R²(B)	VIF(B)
Wind speed	0.198	0.723	0.019	5.51	-0.08	0.869	0.004	5.82
Temperature	-0.084	0.818	0.008	4.3	-0.082	0.828	0.006	3.99
RH	-0.327	0.623	0.036	10.25	-0.638	0.304	0.131	10.56

Note: Panel A drops February; Panel B linearly interpolates February.

To aid visual interpretation of these multivariate results, we provide a standardized monthly heatmap in the Supplement (Fig. S3), which shows that months with enhanced rooftop dust co-occur with lower humidity and lower temperature. The result was robust to the handling of the February gap (exclusion vs. linear interpolation), yielding comparable estimates (Table 3). For visualization, the biplot in Fig. S4 shows monthly scores distributed along PC1 in a manner consistent with the seasonal decline in dust during wetter months, providing a compact, multivariate counterpart to the single-factor relationships reported above.

Implications for monitoring and environmental impact

Roof dust provides a complementary, deposition-integrated indicator for urban particle burden. Based on our results, we recommend prioritizing sampling during dry and windier periods (onset around ~ 2.5 m s⁻¹, strongest effects > 5 m s⁻¹) and when SE–SSW winds are frequent, while deprioritizing weeks with persistent rainfall or high humidity. A minimal network should include at least one low-, mid-, and high-rise roof; mid-rise districts deserve emphasis given their prevalence and the observed variability. Each site should have a fixed sampling patch, clear access, and basic safety measures (two-person entry, fall protection, no work under adverse weather). QA/QC includes paired blanks/duplicates (~ 10%), recording exposure days, concurrent wind sector and speed, precipitation/humidity, and local maintenance/cleaning events. As an order-of-magnitude guide, per-visit consumables and access logistics typically fall in the USD 15–45 per site range (excluding personnel and laboratory analysis). These practices enable routine monthly sampling in winter and monthly to bi-monthly in spring (with post-event sampling 24–72 h after dust/wind episodes), and quarterly in summer–autumn. While roof dust does not replace fixed-site PM monitors, its sensitivity to wind direction, speed, and episodic transport makes it a practical addition to urban networks.

Beyond monitoring applications, roof dust also has important environmental implications. Roof dust, once mobilized by wind or rain, can enter urban runoff systems, potentially affecting rooftop drainage and downstream water quality. Its composition reflects long-term atmospheric deposition, serving as a valuable indicator for urban particulate pollution and potential human exposure through resuspension or contact. Roof dust, once mobilized by wind or rain, can enter urban runoff systems, potentially affecting rooftop drainage and downstream water quality. Its composition reflects long-term atmospheric deposition, serving as a valuable indicator for urban particulate pollution and potential human exposure through resuspension or contact.

Conclusion

This study combined statistical, graphical, and physical reasoning to examine roof dust dynamics using 159 samples (Dec 2020–Nov 2021). The main findings are:

1. Seasonality and magnitude. Roof dust followed a U-shaped annual pattern, peaking in February–March and reaching minima in summer–autumn (spring > winter > autumn > summer).

2. Particle-size linkage. Roof dust accumulation had the strongest correlation with PM₁₀, followed by PM_2.5. This result suggests that PM₁₀ and roof dust originated from similar sources and exhibit similar variations in content.

3. Wind regime controls. Sectoral winds from SE–SSW enhanced accumulation, whereas NNW winds suppressed it. A wind-speed threshold was observed: loads increased above ~ 2.5 m s^–1, with the strongest accumulation at > 5 m s^–1; speeds < 2.5 m s^–1 were negatively correlated.

4. Moisture effects and temperature. Monthly precipitation frequency and RH were strongly negatively correlated with dust, consistent with wet removal and reduced resuspension. Temperature showed a weak, seasonally reversing association (positive in Dec–Mar; negative in Apr–Oct), yielding a slight overall negative tendency.

5. Multivariate synthesis and implications. A PCA/PCR analysis shows that a high-RH/warmer/low-wind composite state is significantly and negatively associated with rooftop dust, supporting denser rooftop sampling/cleaning during dry, windier periods and highlighting rooftop deposition as a complementary urban monitoring metric.

6. Vertical pattern and urban form. Dust generally decreased with building height, with unexpectedly high loads on mid-rise rooftops, likely reflecting higher local building density and restricted air circulation.

Limitations and outlook: The one-year record and single-campus setting limit statistical power and generalizability. Results provide directional evidence that should be verified with longer multi-site datasets and explicit urban-canopy or LES modeling to test roughness-sublayer recirculation and dispersion mechanisms. Future work will expand sampling to multiple urban districts to further test the representativeness of the observed patterns.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Chen Su: Conceptualization, Methodology, Software Wenji Zhao: Supervision Jie Dong: Data Curation Xing Yan: Writing - Review & Editing Yixue Zhong: Data Curation Zhiqiang Yang: Investigation.All authors reviewed the manuscript.

Funding

This research was supported by the Fundamental Research Funds of Capital Normal University (Project No. 42071422), titled “Remote sensing identification of building debris dumps and simulation of their atmospheric dispersion characteristics”.

Data availability

The datasets used and/or analysed during the current study available from thecorresponding author on reasonable request.

Competing interests

The authors declare no competing interests.