Next Article in Journal
Atmosphere-Transported Emerging and Persistent Contaminants (EPCs) in Rainfall and Throughfall: Insights from a Rural Site in Northern Thailand
Previous Article in Journal
Meteorological and Hydrological Drought Risks under Future Climate and Land-Use-Change Scenarios in the Yellow River Basin
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Atmosphere
Volume 14
Issue 11
10.3390/atmos14111602
atmosphere-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles thumb_up
...
Endorse textsms
...
Comment
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Comparison of NO2 and BC Predictions Estimated Using Google Street View-Based and Conventional European-Wide LUR Models in Copenhagen, Denmark

by
Shali Tayebi
1,
Jules Kerckhoffs
2,
Jibran Khan
3,4,
Kees de Hoogh
5,6,
Jie Chen
2,7,
Seyed Mahmood Taghavi-Shahri
1,
Marie L. Bergmann
1,
Thomas Cole-Hunter
1,
Youn-Hee Lim
1,
Laust H. Mortensen
1,8,
Ole Hertel
9,
Rasmus Reeh
10,
Joel Schwartz
7,
Gerard Hoek
2,
Roel Vermeulen
2,
Zorana Jovanovic Andersen
1,
Steffen Loft
1,* and
Heresh Amini
11,12,*
1
Department of Public Health, University of Copenhagen, 1172 Copenhagen, Denmark
2
Institute for Risk Assessment Sciences (IRAS), Utrecht University, 3584 CS Utrecht, The Netherlands
3
Department of Environmental Science, Aarhus University, 8000 Roskilde, Denmark
4
Danish Big Data Centre for Environment and Health (BERTHA), Aarhus University, 8000 Roskilde, Denmark
5
Swiss Tropical and Public Health Institute (Swiss TPH), 4123 Allschwil, Switzerland
6
University of Basel, 4001 Basel, Switzerland
7
Department of Environmental Health, Harvard TH Chan School of Public Health, Boston, MA 02115, USA
8
Statistics Denmark, 2100 Copenhagen, Denmark
9
Faculty of Technical Sciences, Aarhus University, 8000 Roskilde, Denmark
10
UrbanDigital.dk, 2500 Valby, Denmark
11
Department of Environmental Medicine and Public Health, Icahn School of Medicine at Mount Sinai, New York, NY 10029, USA
12
Institute for Exposomic Research, Icahn School of Medicine at Mount Sinai, New York, NY 10029, USA
 
add Show full affiliation list
 
remove Hide full affiliation list
*
Authors to whom correspondence should be addressed.
Atmosphere 2023, 14(11), 1602; https://doi.org/10.3390/atmos14111602
Submission received: 10 August 2023 / Revised: 28 September 2023 / Accepted: 5 October 2023 / Published: 26 October 2023
(This article belongs to the Section Atmospheric Techniques, Instruments, and Modeling)
Browse Figures
Review Reports Versions Notes

Abstract

:
A widely used method for estimating fine scale long-term spatial variation in air pollution, especially for epidemiology studies, is land use regression (LUR) modeling using fixed off-road monitors. More recently, LUR models have been developed using data from mobile monitors that repeatedly measure road pollutants and mixed-effects modeling. Here, nitrogen dioxide (NO2) and black carbon (BC) predictions from two independent models were compared across streets (defined as 30–60 m road segments) (N = 30,312) and residences (N = 76,752) in Copenhagen, Denmark. The first model was Google Street View (GSV)-based mixed-effects LUR models (Google-MM) that predicted 2019 mean NO2 and BC levels, and the second was European-wide (EUW) LUR models that predicted annual mean 2010 levels at 100 m spatial resolution. Across street segments, the Spearman correlation coefficient between the 2019 NO2 from Google-MM-LUR and 2010 NO2 from EUW-LUR was 0.66, while at residences, this was 0.60. For BC, these were 0.51 across street segments and 0.40 at the residential level. The ratio of percentile 97.5 to 2.5 for NO2 across the study area streets using Google-MM NO2 was 4.5, while using EUW-LUR, this was 2.1. These NO2 ratios at residences were 3.1 using Google-MM LUR, and 1.7 using EUW-LUR. Such ratios for BC across street segments were 3.4 using Google-MM LUR and 2.3 using EUW-LUR, while at the residential level, they were 2.4 and 1.9, respectively. In conclusion, Google-MM-LUR NO2 for 2019 was moderately correlated with EUW-LUR NO2 developed in 2010 across Copenhagen street segments and residences. For BC, while Google-MM-LUR was moderately correlated with EUW-LUR across Copenhagen streets, the correlation was lower at the residential level. Overall, Google-MM-LUR revealed larger spatial contrasts than EUW-LUR.

1. Introduction

Air pollution, with a mixture of particles and gases, is a major contributor to disease and death worldwide [1]. According to global estimates, ambient and household air pollution (primarily particulate matter with aerodynamic diameter ≤ 2.5 µm (PM2.5)) caused about 7 million annual premature deaths, mostly related to non-communicable diseases, such as cardiovascular and respiratory diseases [2,3]. Other pollutants, such as nitrogen dioxide (NO2) have also been reported to be associated with extra health burdens independent of other pollutants, such as all-cause mortality and incidence of acute coronary heart disease, stroke, and asthma [4,5,6,7,8]. Black carbon (BC) is reported to be associated with mortality [7] and morbidity, including dementia, at higher effect sizes compared to other pollutants [9], yet no guideline has been set for regulating BC in the latest World Health Organization (WHO) 2021 revision [10]. Nonetheless, WHO recommends systematically measuring BC and/or elemental carbon (EC) and developing standards (or targets) for ambient BC/EC concentrations when appropriate [10,11].
Exposure assessment models for NO2,, and more recently BC, are well developed using off-road monitored data, such as the European wide land use regression (EUW-LUR) model [12]. Such models are used in large-scale studies, such as the multi-cohort ELAPSE (Effects of Low-Level Air Pollution: A Study in Europe) project [13,14,15,16]. Previous studies have demonstrated that Google Street View (GSV) cars equipped with fast-response instruments can measure on-road air pollution levels with a high spatial resolution in urban areas [17]. These outputs have been used to develop land use regression (LUR) models in Amsterdam, the Netherlands, and in Copenhagen, Denmark (hereafter referred to as Google-MM) that predict the hyperlocal variation of air pollutants such as NO2, BC, and ultrafine particles [18,19]. However, what remains uncertain is the applicability of Google-MM LUR models for long-term exposure assessment in epidemiological studies, especially for residential locations that are often away from main roads. One possible way to evaluate the Google-MM LUR models (in the absence of enough independent long-term measurements across diverse locations) is to compare them with existing European-wide LUR models developed to assess long-term exposure using fixed-site standard monitors and other predictors. Although the EUW-LUR models are based on data sources that are older or different from the Google-MM LUR models, previous research has suggested that the long-term spatial patterns of air pollution in developed cities are relatively stable over time [20,21,22]. Therefore, a comparison of the Google-MM and EUW-LUR models could provide useful insights into their performance and limitations. In this study, we aimed to compare the long-term NO2 and BC concentrations estimated by Google-MM and EUW-LUR models at the street and residential levels in Copenhagen, Denmark.

2. Materials and Methods

2.1. Study Area

The area of this study is the Danish capital, Copenhagen (latitude: 55.676098, and longitude: 12.568337). This area spans about 150 km2 [23] with a total population of more than 800,000 people [24]. The city’s climate is mild, and the annual average temperature is ~9 °C. It receives significant rainfall, with the annual average reaching ~700 mm [25].

2.2. Air Pollution Data

2.2.1. Dataset 1: GSV-Based Mixed-Effects Model LUR Predictions (Google-MM)

The data consisted of predicted concentrations of NO2 and BC in every street (total 30,312 street segments) using a mixed-effects model [18,19]. The repeated measurements used to create the mixed-effects model were gathered from Monday through Friday, between 8:00 a.m. and 10:00 p.m. (with most measurements having been taken between 10:00 a.m. and 4:00 p.m.), from 15 October 2018 until 15 March 2020 by GSV cars. Street-by-street levels of air pollutants were measured and averaged over 30–60 m road segments (defined by OpenStreetMap) using fast-response laboratory-grade instruments driving with the flow of traffic at regular speeds. The measurements are thus an approximation of on-road air pollution concentrations throughout the city in 2019, while also containing data from a few months prior and subsequent to 2019. The number of drive days per street segment ranged from 1 to 126 (mean: 7 drive days, median: 4 drive days). NO2 was measured with a Cavity Attenuated Phase Shift (CAPS) monitor, which measures NO2 at 450 nm in the blue section of the electromagnetic spectrum; and BC was recorded by an AE33 aethalometer. Subsequently, the daily average of each pollutant on each road segment was calculated, and a LUR model was built based on predictors such as traffic-related variables and land use. The final predictors were used as fixed effects for the mixed-effects model while having a random intercept for each road segment [18,19,26].
For NO2, the predictors of the LUR model were: traffic intensity on all roads in a 50-m buffer; length of major roads in a 500-m buffer; traffic intensity on the nearest road; average building height in a 100-m buffer; area of transportation services in a 1000-m buffer; area of water in a 1000-m buffer; and length of major roads in a 100-m buffer. The p-value of all these predictors in the NO2 model were <0.001 and the R2 for NO2 LUR model was 0.54.
For the BC, the predictors of the LUR model were: traffic intensity on all roads in a 50-m buffer; area of transportation services in a 1000-m buffer; traffic intensity on the nearest road; length of major roads in a 100-m buffer; area of industry in a 5000-m buffer; traffic intensity on all roads in a 300-m buffer; area of airports in 5000-m buffer; and population density in a 5000-m buffer. The p-value of all these predictors in the BC model were <0.001 and the R2 for the BC LUR model was 0.34 [19].

2.2.2. Dataset 2: Air Pollution Data from European-Wide ELAPSE Project LUR Models (EUW-LUR)

The exposure estimates of the annual mean of NO2 and BC concentrations at 100 × 100 m grids for the year 2010 were obtained from a Western European LUR model (hereafter called EUW-LUR10)—a development as part of the ELAPSE project [12]. The EUW-LUR10 for NO2 was developed using data from the AirBase monitoring network and predictions from satellite observations, dispersion model estimates, and land use amongst others. AirBase monitors across Europe predominantly use chemiluminescence methods for measuring NO2. This EPA-regulated standard procedure involves the generation of chemiluminescence using ozone. In this process, NO2 is initially converted to nitric oxide (NO). NO then interacts with O3, resulting in the excited state of NO2 which emits light as it reverts to its ground state. The emitted light’s intensity is directly proportional to the concentration of NO2. Previous studies have documented the CAPS monitor readings (the monitor used in dataset 1) to show a high correlation (R2 > 0.99) with those from the chemiluminescence-based monitor. The absolute concordance between the CAPS and the chemiluminescence analyzer falls within an anticipated statistical noise range (±0.4 ppb with 60 s averaging) [27].
As the EUW-LUR10 NO2 model estimated data for the year 2010, and the Google-MM predicted data for the year 2019, a prediction model was used for NO2 in order to make the datasets more comparable. The 2010 and 2019 annual summary published by the Danish Air Quality Monitoring program was used to create such a model [28] where measurements of NO2 from three monitoring stations placed around Copenhagen: two at Jagtvej and H.C. Andersens Boulevard, which measured street-level air quality, and one placed on the rooftop of the H.C Ørsted Institute which measured urban background air quality. The mean changes of NO2 measurements from 2010 to 2019 gave a ratio to predict the 2019 concentrations for EUW-LUR10 (hereafter referred to as EUW-LUR19). The ratios indicated that between 2010 and 2019, NO2 has reduced by a factor of 1.63. The EUW-LUR10 for BC was developed using monitored data from the European Study of Cohorts for Air Pollution Effects (ESCAPE) project where PM2.5 absorbance based on reflectance measurements of filters between 2009 to 2010 at 20 project sites. Due to the lack of monitored data for BC in Copenhagen by the regulatory network, we were unable to estimate BC EUW-LUR19. The R2 for NO2 LUR model in the five-fold hold-out-validation was 0.57, and for BC LUR model, it was 0.51.

2.3. Statistical Analyses

The Google-MM polyline format allowed NO2 and BC data to be visualized on Copenhagen’s street segments. However, the EUW-LUR data were in raster format (cell size: 100 m × 100 m). Therefore, to compare data of Google-MM LUR with EUW-LUR across the 30,312 street segments, each street segment (length range: 15–60 m) was assigned one mid-street point and the data of EUW-LUR were extracted and merged with data of Google-MM at each of these points. For the 76,752 residences in the study area, a near analysis was conducted to calculate the distance of each residence to the nearest Google-MM LUR street and to identify the nearest road segment. Next, the air pollution data from Google-MM LUR were assigned to the nearest residence. The summary statistics for each model’s pollutant in various neighborhoods were conducted using R version 4.1.1 and ArcGIS version 10.8.2.
To assess the correlations of pollutants within and between the models across street segments and residences, we calculated Spearman’s correlation coefficients.
Additionally, Bland–Altman (BA) plots were created to further investigate the agreement between model predictions by the Google-MM and EUW-LUR models for street segments and residences. The limits of agreement are defined as the average difference ± 1.96 standard deviation of the difference, which would indicate how far apart the measurements from the two models are likely to be for most locations.

3. Results

According to the Google-MM predictions, the on-road weekday daytime mean (standard deviation, SD) for NO2 was 17.3 (8.0) μg/m3 across 30,312 street segments and 15.0 (5.0) μg/m3 across 76,752 residences, and for BC, it was 1.1 (0.4) μg/m3 across street segments and 0.99 (0.26) μg/m3 across residences (Table 1). According to the EUW-LUR19 predictions, the yearly average for NO2 was 20.9 (3.4) μg/m3 across street segments and 20.4 (2.6) μg/m3 across residences, and 1.6 (0.3) μg/m3 across street segments, and for BC, it was 1.6 (0.24) μg/m3 across residences based on EUW-LUR10 across the total study area (Table 1).
In 90% of Copenhagen’s roads, the annual average of NO2 levels, as predicted by the Google-MM predictions, exceeded the WHO Air Quality Guideline (AQG) threshold of 10 μg/m3. Spatial hotspots for NO2 between the two models were consistent with high concentration values observed near major roads and highways. However, the Google-MM LUR revealed lower values of NO2 around the airport in the southeastern part of the city (Figure 1). For BC, high concentrations were consistently found near major roads and highways, while the EUW-LUR model also identified significant hotspots within the inner city of Copenhagen.
The Spearman correlation coefficient, which measured the relationship between street segments, indicated a value of 0.66 for the 2019 NO2 data obtained from Google-MM-LUR and 2010 NO2 data from EUW-LUR (Figure 2), while this was 0.60 across residences (Figure 3). For BC, these were 0.51 across street segments (Figure 2), and 0.40 across residences (Figure 3).
Across 30,312 street segments, the ratio of percentile 97.5 to 2.5 for NO2 using Google-MM NO2 was 4.5, while using EUW-LUR, this was 2.1 (Table 1). Across 76,752 residences, these NO2 ratios were 3.1 using Google-MM LUR and 1.7 using EUW-LUR. Such ratios for BC across street segments were 3.4 using Google-MM LUR and 2.3 using EUW-LUR, whereas on the residence-level, they were 2.4 and 1.9, respectively (Table 1).
The BA plot for NO2 (based on Google-MM and EUW-LUR19 predictions) across Copenhagen street segments and residences suggested a decreasing agreement between the two model’s estimates as mean concentrations increased, showing an upward trend. The differences were particularly above the limits of agreement when the mean of NO2 concentrations from the two models were above ~20 μg/m3. This, for BC (based on Google-MM and EUW-LUR10 predictions), suggested a similar pattern (Figure 4).

4. Discussion

By using two independent data sources, we were able to describe and compare the estimates of air pollution concentration levels in Copenhagen, Denmark. We found generally consistent spatial patterns and moderate correlations for the air pollution data between the two used LUR models employed. The correlations of pollutants estimated by the two models were generally lower at residences, especially for BC. Notably, Google-MM LUR reported a larger variation and contrast in air pollutants across Copenhagen street segments and residences compared to EUW-LUR, particularly in regions with low NO2 levels near the airport. Such differences were more pronounced at locations with higher mean values, as revealed by Bland–Altman plots.
The NO2 concentrations are generally dominated by local emission sources, such as street traffic (emissions from cars, buses, or trucks), and diesel-powered heavy construction equipment or other movable engines [29]. The low NO2 near the airport area as determined by Google-MM LUR (values lower than 11 μg/m3) suggests that Kastrup Airport may not substantially contribute to this. This finding contradicts the prediction of the EUW-LUR model, which indicated that the annual mean NO2 concentration of about 30 μg/m in residential areas near the airport, warranting further investigation. Even though the NO2 concentrations in Copenhagen during the daytime are generally low relative to many parts of the world, they still exceed levels of public health concern (10 μg/m3) compared to the air quality guidelines (AQGs) published by the WHO [10] at more than 90% of the street segments, based on the Google-MM predictions. It has been recommended to adopt the WHO AQGs 2021 or the interim targets as national air quality standards in order to implement mitigation efforts and reduce air pollution concentrations [30,31,32].
The elevated levels of NO2 and BC observed on major roads and highways are consistent with previous findings in other European countries [33]. The relatively large hotspot within Copenhagen, as reported by the EUW-LUR model for BC (values above 1.84 μg/m3), could possibly result from high residential land use, which was a significant predictor of EUW-LUR model, given that the area has a large residential area and high population density [12]. Further investigation is needed to confirm the high values around this area compared to other parts of the city. The WHO guidelines did not provide sufficient quantitative evidence to establish AQG levels for BC. However, they did formulate a statement of good practice. As a result, one of the WHO recommendations for countries and regional authorities is to reduce BC emissions and develop their own standards or targets for ambient concentrations.
We observed substantially greater spatial variations using Google-MM model compared to the EUW-LUR, as shown by the ratio of percentile 97.5 to 2.5 for estimated NO2 and BC in the area. For NO2 concentrations, the Google-MM yielded a ratio of 4.5 across street segments, while for EUW-LUR, this ratio was 2.1. BC concentrations exhibited a 3.4-fold variation across street segments according to Google-MM, while using EUW-LUR data, this was 2.3-fold. In Amsterdam, these ratios for NO2 and BC with the Google-MM model were 3.2 and 3.3 across street segments, respectively, highlighting greater spatial disparities compared to Copenhagen. The discrepancies between Google-MM LUR and EUW-LUR can be attributed to differences in measurement techniques used (GSV used on-road measurements, whereas the EUW model utilized more background and off-road monitors) or the detection of intense spatial variations for smaller scales by the Google-MM. Notably, the Google-MM method made predictions for road segments in a vector format, whereas the EUW model predicted concentrations in a raster format (100 × 100 m). The midpoint of the grid cells of the raster dataset rarely overlapped exactly with the midpoints of the road segments, resulting in generally smaller coefficients for traffic related predictors are generally smaller for the EUW model. Also, as the EUW model’s measurements were not done on the road, its predictions tended to be lower on average than the Google-MM predictions. As shown in Figure 4, the agreement between the two models diminished as mean concentrations increased, reflecting the Google-MM’s ability to predict larger spatial variations at fine scale.
We found moderate correlation between NO2 predictions from the Google-MM and the EUW-LUR19 predictions across both streets (0.66) and residences (0.60). Conversely, absolute NO2 concentrations have reduced considerably from 2010 to recent years. The moderate spatial and temporal stability of NO2 from 2010 to 2019 holds promise for epidemiological studies employing both models. A study in Vancouver, Canada, reported temporal stability of LUR models for estimating long-term spatial variation of NO2. The correlation between 2003 and 2010 NO2 measurements at the same sites was 0.74 [22]. Similarly, an Italian study in Rome also reported relatively good agreement between measured and predicted NO2 values from LUR models based on samples collected 12 years apart. NO2 measurements from 1995 and 2007 exhibited a correlation of 0.79, and the correlation of predictions across residences by LUR models was 0.96 [20]. A Dutch study by Eeftens et al. (2011) yielded comparable results, with NO2 measurements conducted in 2007 aligning well with NO2 measurements taken in 1999–2000 at the same locations (R2 = 0.86) [21]. Nonetheless, in our study, the NO2 correlations between the two models were lower for residences. This warrants further considerations when utilizing Google-MM LUR predictions for air pollution exposure assessment in residential areas.
There was also a low correlation for BC between the two models across the street segments (0.51). This can be explained by the differences in measurement instruments, modeling techniques, or city developments from 2010 to 2019. The Google-MM study utilized an AE33 aethalometer to record the pollutant during the daytime hours, while the EUW-LUR study measured BC using PM2.5 absorbance based on reflectance measurements of the filters collected during the ESCAPE project campaigns between 2009 to 2010 (including both day and night measurements) [12], resulting in a greater number of data points at each location. These two methodologies for measuring BC differ. A previous study has reported high correlation (R2 > 0.94) between PM2.5 absorbance and elemental carbon (EC) measurements using the German reference method VDI 2465, which employs a thermal measurement method [34]. A Canadian study has also reported a high correlation between BC measurements using thermal methods and light absorption [35]. Therefore, the different BC measurement methods employed by Google-MM and EUW-LUR models may account part of the discrepancy. While both models for BC share predictors related to traffic emissions, there are some differences. For instance, the EUW-LUR model includes urban green and residential land use variables, which are absent in the Google-MM LUR model. This could potentially result in different BC spatial patterns.
The increase in number of private cars from 2010 to 2019, changes in policy regulations, the establishment of designated corridors for lorries, the significant growth in delivery services in residential areas, shifts in traffic patterns possibly attributed to the use of optimized routing from navigation apps, and alterations in bus routes accompanied by the adaptation to electrified motors across the years are just a few examples of urban developments that have occurred. Less is known about the stability of BC LUR models over the years, and our study demonstrates relatively good agreement between street-level predictions for 2010 and 2019 by two independent models. Like NO2, BC correlations at residential locations show lower consistency between the two models (Figure 2 and Figure 3).
The input data for both the Google-MM and EUW-LUR models, as well as the predictions generated by these models, differ significantly. Google-MM’s input data were collected through local on-road measurements conducted street by street during daytime weekdays within the city of Copenhagen. In Google-MM, conventional LUR model predictors serve as fixed effects within a mixed-effects model, with a random intercept assigned to each road segment [18,19]. Consequently, extrapolating to unmeasured street segments and off-road locations requires additional consideration. Nonetheless, the fixed-effect component of the mixed model can be employed as a conventional LUR model to make predictions for such locations. In contrast, the EUW-LUR monitored data for NO2 relied on long-term mean measurements from regulatory network monitoring stations in several western Europe cities [12,36]. Subsequently, a conventional land use regression model was developed using various satellite and land use covariates to predict the variability of monitored air pollutants. It is important to note that such a model relies on generalization and extrapolated data from European data that have been modeled to a local scale.
The study presented in this paper provides valuable insights into the spatial variation of NO2 and BC in Copenhagen, Denmark, using two different LUR modeling approaches. The atmospheric implications of this study are twofold. First, it demonstrates the potential of using Google Street View data and mixed-effects modeling to capture the fine-scale spatial variability of air pollutants, especially in urban areas with complex street configurations and traffic patterns. This can improve the exposure assessment of air pollution for epidemiological studies and inform the design and implementation of effective air quality management strategies. Second, it highlights the temporal changes of air pollution levels and spatial patterns over a decade, which may reflect the impacts of various emission sources, meteorological conditions, and policy interventions. This can enhance the understanding of the long-term trends and drivers of air pollution, as well as the evaluation of the effectiveness of air quality policies and regulations.
Both the Google-MM LUR and EUW-LUR predictions are both valuable tools for addressing air pollution. These models consistently identify the areas of long-term exposure to air pollution in Copenhagen. Gehl architects in Copenhagen have recently utilized air pollution exposure estimates from the Google-MM LUR to design healthier environments for childcare institutions and urban areas. Their recommendations include removing parking spaces on specific streets with high air pollution levels and instead introduce green buffer zones as pollution barriers (https://gehlpeople.com/projects/air-quality-copenhagen/ accessed on 10 August 2023). Such initiatives and interventions have the potential to benefit not only children but also to the general population.

Strength and Limitations

The Google-MM LUR has the advantage of being able to capture the influence of local sources in small areas, which is challenging for a European scale model as the data about that specific source may not be available for all monitored locations in the EUW-LUR model. Therefore, the Google-MM LUR provides a broader range of concentrations as indicated by the ratio of percentile 97.5/2.5 values.
However, this work has limitations. Firstly, we could not predict BC concentrations in 2019 because there were no BC measurements from regulatory network monitoring stations in 2010. Therefore, it may not be suitable to directly compare the absolute BC values between the two models. Secondly, the datasets used for building the models differed and were separated by approximately 10 years. Therefore, there may be inherent changes in spatial and temporal variation of NO2 and BC in Copenhagen due to new developments and policies in the city, which may explain part of the discrepancies between the predictions of the two models.

5. Conclusions

In this study, we compared two different methods for estimating the spatial variation of NO2 and BC in Copenhagen, Denmark. We utilized Google Street View-based mixed-effects LUR models and European-wide LUR models to predict the long-term mean levels of these pollutants at both street and residential scales. Our results indicated that Google-MM LUR and EUW-LUR models show moderate associations for long-term NO2 levels along streets and at residential locations in Copenhagen. This suggests that Google-MM LUR, based on mobile monitoring, can effectively capture the long-term spatial patterns of NO2 in Copenhagen at both street and residential scales. Furthermore, the 2010 EUW-LUR for NO2 also appears to be a reliable model for estimating the long-term spatial patterns of this pollutant in Copenhagen over several years. However, for BC, the correlation between Google-MM LUR and EUW-LUR was moderate along streets but lower at residential locations in Copenhagen. Google-MM LUR also reported higher spatial contrasts than EUW-LUR for both pollutants along streets and at homes, which requires further investigation to verify the existence of such large spatial variations.

Author Contributions

Conceptualization: S.T., S.L. and H.A.; Formal analysis: S.T. and H.A.; Resources: R.V., J.K. (Jules Kerckhoffs), J.K. (Jibran Khan), K.d.H., J.C. and G.H.; Methodology: all co-authors; Writing—original draft: S.T. and H.A.; Writing—review and editing: all co-authors; Project administration and funding acquisition: H.A.; Supervision: S.L. and H.A. All authors have read and agreed to the published version of the manuscript.

Funding

This work was supported by Health Effects Institute (HEI) (#4982-RFA19-2/21-5) and Novo Nordisk Foundation Challenge Programme (NNF17OC0027812). HEI is an organization jointly funded by the United States Environmental Protection Agency (EPA) (Assistance Award CR 83998101) and certain motor vehicle and engine manufacturers. The contents of this article do not necessarily reflect the views of HEI, or its sponsors, nor do they necessarily reflect the views and policies of the EPA or motor vehicle and engine manufacturers. Heresh Amini was supported by the US National Institute of Health (grant numbers: P30ES023515 and UL1TR004419).

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

Data of dataset 1 is publicly available (see reference [36]). Data of dataset 2 can be obtained from Kees de Hoogh ([email protected]) or Gerard Hoek ([email protected]).

Conflicts of Interest

The authors declare no conflict of interest. Mr. Rasmus Reeh is owner of UrbanDigital. The paper reflects the views of the scientists, and not the company.

References

  1. Murray, C.J.; Aravkin, A.Y.; Zheng, P.; Abbafati, C.; Abbas, K.M.; Abbasi-Kangevari, M.; Abd-Allah, F.; Abdelalim, A.; Abdollahi, M.; Abdollahpour, I. Global burden of 87 risk factors in 204 countries and territories, 1990–2019, a systematic analysis for the Global Burden of Disease Study 2019. Lancet 2020, 396, 1223–1249. [Google Scholar] [CrossRef]
  2. Burnett, R.; Chen, H.; Szyszkowicz, M.; Fann, N.; Hubbell, B.; Pope, C.A., 3rd; Apte, J.S.; Brauer, M.; Cohen, A.; Weichenthal, S.; et al. Global estimates of mortality associated with long-term exposure to outdoor fine particulate matter. Proc. Natl. Acad. Sci. USA 2018, 115, 9592–9597. [Google Scholar] [CrossRef]
  3. World Health Organization. Burden of Disease from Ambient Air Pollution for 2016, Version 2. 2018. Available online: https://cdn.who.int/media/docs/default-source/air-pollution-documents/air-quality-and-health/aap_bod_results_may2018_final.pdf (accessed on 9 August 2023).
  4. Brunekreef, B.; Strak, M.; Chen, J.; Andersen, Z.J.; Atkinson, R.; Bauwelinck, M.; Bellander, T.; Boutron, M.-C.; Brandt, J.; Carey, I.; et al. Mortality and Morbidity Effects of Long-Term Exposure to Low-Level PM2.5, BC, NO2, and O3, An Analysis of European Cohorts in the ELAPSE Project. Res. Rep. Health Eff. Inst. 2021, 2021, 1–127. [Google Scholar] [PubMed]
  5. Huang, S.; Li, H.; Wang, M.; Qian, Y.; Steenland, K.; Caudle, W.M.; Liu, Y.; Sarnat, J.; Papatheodorou, S.; Shi, L. Long-term exposure to nitrogen dioxide and mortality: A systematic review and meta-analysis. Sci. Total Environ. 2021, 776, 145968. [Google Scholar] [CrossRef]
  6. So, R.; Andersen, Z.J.; Chen, J.; Stafoggia, M.; de Hoogh, K.; Katsouyanni, K.; Vienneau, D.; Rodopoulou, S.; Samoli, E.; Lim, Y.-H.; et al. Long-term exposure to air pollution and mortality in a Danish nationwide administrative cohort study: Beyond mortality from cardiopulmonary disease and lung cancer. Environ. Int. 2022, 164, 107241. [Google Scholar] [CrossRef] [PubMed]
  7. Stafoggia, M.; Oftedal, B.; Chen, J.; Rodopoulou, S.; Renzi, M.; Atkinson, R.W.; Bauwelinck, M.; Klompmaker, J.O.; Mehta, A.; Vienneau, D.; et al. Long-term exposure to low ambient air pollution concentrations and mortality among 28 million people: Results from seven large European cohorts within the ELAPSE project. Lancet Planet. Health 2022, 6, e9–e18. [Google Scholar] [CrossRef] [PubMed]
  8. Zhang, Y.; Li, Z.; Wei, J.; Zhan, Y.; Liu, L.; Yang, Z.; Zhang, Y.; Liu, R.; Ma, Z. Long-term exposure to ambient NO2 and adult mortality: A nationwide cohort study in China. J. Adv. Res. 2022, 41, 13–22. [Google Scholar] [CrossRef] [PubMed]
  9. Shi, L.; Zhu, Q.; Wang, Y.; Hao, H.; Zhang, H.; Schwartz, J.; Amini, H.; van Donkelaar, A.; Martin, R.V.; Steenland, K.; et al. Incident dementia and long-term exposure to constituents of fine particle air pollution: A national cohort study in the United States. Proc. Natl. Acad. Sci. USA 2023, 120, e2211282119. [Google Scholar] [CrossRef]
  10. World Health Organization. WHO Global Air Quality Guidelines: Particulate Matter (PM2.5 and PM10), Ozone, Nitrogen Dioxide, Sulfur Dioxide and Carbon Monoxide. 2021. Available online: https://www.who.int/publications/i/item/9789240034228 (accessed on 9 August 2023).
  11. Goshua, A.; Akdis, C.A.; Nadeau, K.C. World Health Organization global air quality guideline recommendations: Executive summary. Allergy 2022, 77, 1955–1960. [Google Scholar] [CrossRef]
  12. De Hoogh, K.; Chen, J.; Gulliver, J.; Hoffmann, B.; Hertel, O.; Ketzel, M.; Bauwelinck, M.; van Donkelaar, A.; Hvidtfeldt, U.A.; Katsouyanni, K.; et al. Spatial PM2.5, NO2, O3 and BC models for Western Europe—Evaluation of spatiotemporal stability. Environ. Int. 2018, 120, 81–92. [Google Scholar] [CrossRef]
  13. Liu, S.; Jørgensen, J.T.; Ljungman, P.; Pershagen, G.; Bellander, T.; Leander, K.; Magnusson, P.K.E.; Rizzuto, D.; Hvidtfeldt, U.A.; Raaschou-Nielsen, O.; et al. Long-term exposure to low-level air pollution and incidence of asthma: The ELAPSE project. Eur. Respir. J. 2021, 57, 2003099. [Google Scholar] [CrossRef]
  14. Liu, S.; Jørgensen, J.T.; Ljungman, P.; Pershagen, G.; Bellander, T.; Leander, K.; Magnusson, P.K.; Rizzuto, D.; Hvidtfeldt, U.A.; Raaschou-Nielsen, O.; et al. Long-term exposure to low-level air pollution and incidence of chronic obstructive pulmonary disease: The ELAPSE project. Environ. Int. 2021, 146, 106267. [Google Scholar] [CrossRef] [PubMed]
  15. Strak, M.; Weinmayr, G.; Rodopolou, S.; Tewis, M.; de Hoogh, K.; Forastiere, F.; Andersen, Z.; Brunekreef, B.; Hoek, G.; Katsouyanni, K. Natural and cause-specific mortality and low-level air pollution in a pooled cohort of 392,826 participants in Europe: The ELAPSE project. Environ. Epidemiol. 2019, 3, 383–384. [Google Scholar]
  16. Strak, M.; Weinmayr, G.; Rodopoulou, S.; Chen, J.; de Hoogh, K.; Andersen, Z.J.; Atkinson, R.; Bauwelinck, M.; Bekkevold, T.; Bellander, T.; et al. Long term exposure to low level air pollution and mortality in eight European cohorts within the ELAPSE project: Pooled analysis. BMJ 2021, 374, n1904. [Google Scholar] [CrossRef] [PubMed]
  17. Apte, J.S.; Messier, K.P.; Gani, S.; Brauer, M.; Kirchstetter, T.W.; Lunden, M.M.; Marshall, J.D.; Portier, C.J.; Vermeulen, R.C.; Hamburg, S.P. High-Resolution Air Pollution Mapping with Google Street View Cars: Exploiting Big Data. Env. Environ. Sci. Technol. 2017, 51, 6999–7008. [Google Scholar] [CrossRef] [PubMed]
  18. Kerckhoffs, J.; Khan, J.; Hoek, G.; Yuan, Z.; Ellermann, T.; Hertel, O.; Ketzel, M.; Jensen, S.S.; Meliefste, K.; Vermeulen, R. Mixed-Effects Modeling Framework for Amsterdam and Copenhagen for Outdoor NO2 Concentrations Using Measurements Sampled with Google Street View Cars. Environ. Sci. Technol. 2022, 56, 7174–7184. [Google Scholar] [CrossRef] [PubMed]
  19. Kerckhoffs, J.; Khan, J.; Hoek, G.; Yuan, Z.; Hertel, O.; Ketzel, M.; Jensen, S.S.; Al Hasan, F.; Meliefste, K.; Vermeulen, R. Hyperlocal Variation of Nitrogen Dioxide, Black Carbon, and Ultrafine Particles measured with Google Street View Cars in Amsterdam and Copenhagen. Environ. Int. 2022, 170, 107575. [Google Scholar] [CrossRef] [PubMed]
  20. Cesaroni, G.; Porta, D.; Badaloni, C.; Stafoggia, M.; Eeftens, M.; Meliefste, K.; Forastiere, F. Nitrogen dioxide levels estimated from land use regression models several years apart and association with mortality in a large cohort study. Environ. Health 2012, 11, 48. [Google Scholar] [CrossRef]
  21. Eeftens, M.; Beelen, R.; Fischer, P.; Brunekreef, B.; Meliefste, K.; Hoek, G. Stability of measured and modelled spatial contrasts in NO2 over time. Occup. Environ. Med. 2011, 68, 765. [Google Scholar] [CrossRef]
  22. Wang, R.; Henderson, S.B.; Sbihi, H.; Allen, R.W.; Brauer, M. Temporal stability of land use regression models for traffic-related air pollution. Atmos. Environ. 2013, 64, 312–319. [Google Scholar] [CrossRef]
  23. Statistics Denmark Documentation of Statistics for Land Accounts 2018. 2018. Available online: https://www.dst.dk/Site/Dst/Udgivelser/GetPubFile.aspx?id=27468&sid=gnatuk (accessed on 10 August 2023).
  24. Worldpop. Worldpop: Population Counts. 2020. Available online: https://www.worldpop.org/project/categories?id=3 (accessed on 11 January 2023).
  25. Liu, L.; Jensen, M.B. Climate resilience strategies of Beijing and Copenhagen and their links to sustainability. Water Policy 2017, 19, 997–1013. [Google Scholar] [CrossRef]
  26. Kerckhoffs, J.; Hoek, G.; Messier, K.P.; Brunekreef, B.; Meliefste, K.; Klompmaker, J.O.; Vermeulen, R. Comparison of Ultrafine Particle and Black Carbon Concentration Predictions from a Mobile and Short-Term Stationary Land-Use Regression Model. Env. Environ. Sci. Technol. 2016, 50, 12894–12902. [Google Scholar] [CrossRef] [PubMed]
  27. Kebabian, P.L.; Wood, E.C.; Herndon, S.C.; Freedman, A. A practical alternative to chemiluminescence-based detection of nitrogen dioxide: Cavity attenuated phase shift spectroscopy. Env. Environ. Sci. Technol. 2008, 42, 6040–6045. [Google Scholar] [CrossRef] [PubMed]
  28. Ellermann, T.; Khan, J.; Ketzel, M.; Jensen, S.S.; Hertel, O. Air View Data on the Spatial Distribution of Air Pollution in Copenhagen; Aarhus University, DCE—Danish Centre for Environment and Energy: Aarhus, Denmark, 2021; p. 27s. [Google Scholar]
  29. Jensen, S.S. Kildebidrag til Sundhedsskadelig Luftforurening i København; DCE-Nationalt Center for Miljø og Energi: Aarhus, Denmark, 2013. [Google Scholar]
  30. Amini, H. WHO Air Quality Guidelines Need to be Adopted. Int. J. Public Health 2021, 66, 1604483. [Google Scholar] [CrossRef]
  31. Andersen, Z.J.; Gehring, U.; De Matteis, S.; Melen, E.; Vicedo-Cabrera, A.M.; Katsouyanni, K.; Yorgancioglu, A.; Ulrik, C.S.; Medina, S.; Hansen, K.; et al. Clean air for healthy lungs—An urgent call to action: European Respiratory Society position on the launch of the WHO 2021 Air Quality Guidelines. Eur. Respir. J. 2021, 58, 2102447. [Google Scholar] [CrossRef] [PubMed]
  32. Hoffmann, B.; Boogaard, H.; de Nazelle, A.; Andersen, Z.J.; Abramson, M.; Brauer, M.; Brunekreef, B.; Forastiere, F.; Huang, W.; Kan, H.; et al. WHO Air Quality Guidelines 2021—Aiming for Healthier Air for all: A Joint Statement by Medical, Public Health, Scientific Societies and Patient Representative Organisations. Int. J. Public Health 2021, 66, 1604465. [Google Scholar] [CrossRef] [PubMed]
  33. Eeftens, M.; Beelen, R.; de Hoogh, K.; Bellander, T.; Cesaroni, G.; Cirach, M.; Declercq, C.; Dėdelė, A.; Dons, E.; de Nazelle, A.; et al. Development of Land Use Regression Models for PM2.5, PM2.5 Absorbance, PM10 and PMcoarse in 20 European Study Areas; Results of the ESCAPE Project. Environ. Sci. Technol. 2012, 46, 11195–11205. [Google Scholar] [CrossRef]
  34. Cyrys, J.; Heinrich, J.; Hoek, G.; Meliefste, K.; Lewné, M.; Gehring, U.; Bellander, T.; Fischer, P.; van Vliet, P.; Brauer, M.; et al. Comparison between different traffic-related particle indicators: Elemental carbon (EC), PM2.5 mass, and absorbance. J. Expo. Sci. Environ. Epidemiol. 2003, 13, 134–143. [Google Scholar] [CrossRef]
  35. Sharma, S.; Brook, J.R.; Cachier, H.; Chow, J.; Gaudenzi, A.; Lu, G. Light absorption and thermal measurements of black carbon in different regions of Canada. J. Geophys. Res. Atmos. 2002, 107, AAC 11-1–AAC 11-11. [Google Scholar] [CrossRef]
  36. Utrecht University; Google. Google Environmental Insights Explorer. 2021. Available online: https://insights.sustainability.google/labs/airquality (accessed on 10 August 2023).
Atmosphere 14 01602 g001
Figure 1. The long-term mean 2019 Google-MM NO2 and BC (top), and 2010 EUW-LUR NO2 and BC (bottom) in the Greater Copenhagen Area, Denmark (latitude: 55.676098, and longitude: 12.568337). The units for all pollutants are (µg/m3). The breaks in the legend are based on deciles. Note that Google-MM LUR used an aethalometer for BC measurements while the EUW-LUR study measured BC by PM2.5 absorbance method. Differences in measurement methods could explain part of the BC discrepancy.
Figure 1. The long-term mean 2019 Google-MM NO2 and BC (top), and 2010 EUW-LUR NO2 and BC (bottom) in the Greater Copenhagen Area, Denmark (latitude: 55.676098, and longitude: 12.568337). The units for all pollutants are (µg/m3). The breaks in the legend are based on deciles. Note that Google-MM LUR used an aethalometer for BC measurements while the EUW-LUR study measured BC by PM2.5 absorbance method. Differences in measurement methods could explain part of the BC discrepancy.
Atmosphere 14 01602 g001
Atmosphere 14 01602 g002
Figure 2. Street-level scatter-matrix, histogram, and Spearman rank correlation matrix for NO2 and BC by Google-MM and EUW-LUR over the Greater Copenhagen Area. The figure is based on data across 30,312 road segments. Note that Google-MM LUR used an aethalometer for BC measurements while the EUW-LUR study measured BC by PM2.5 absorbance method. Differences in measurement methods could explain part of the BC discrepancy.
Figure 2. Street-level scatter-matrix, histogram, and Spearman rank correlation matrix for NO2 and BC by Google-MM and EUW-LUR over the Greater Copenhagen Area. The figure is based on data across 30,312 road segments. Note that Google-MM LUR used an aethalometer for BC measurements while the EUW-LUR study measured BC by PM2.5 absorbance method. Differences in measurement methods could explain part of the BC discrepancy.
Atmosphere 14 01602 g002
Atmosphere 14 01602 g003
Figure 3. Residential-level scatter-matrix, histogram, and Spearman rank correlation matrix for NO2 and BC by Google-MM and EUW-LUR over the Greater Copenhagen Area. The figure is based on data for nearest road data from Google-MM LUR and grid cell centroid from EUW-LUR across 76,752 residences. Note that Google-MM LUR used an aethalometer for BC measurements while the EUW-LUR study measured BC by PM2.5 absorbance method. Differences in measurement methods could explain part of the BC discrepancy.
Figure 3. Residential-level scatter-matrix, histogram, and Spearman rank correlation matrix for NO2 and BC by Google-MM and EUW-LUR over the Greater Copenhagen Area. The figure is based on data for nearest road data from Google-MM LUR and grid cell centroid from EUW-LUR across 76,752 residences. Note that Google-MM LUR used an aethalometer for BC measurements while the EUW-LUR study measured BC by PM2.5 absorbance method. Differences in measurement methods could explain part of the BC discrepancy.
Atmosphere 14 01602 g003
Atmosphere 14 01602 g004
Figure 4. Bland–Altman plots for street level (top) and residential (bottom) NO2 (estimated by Google-MM and EUW-LUR19) and BC (estimated by Google-MM and EUW-LUR10). The middle dash line in each plot indicates mean difference between the Google-MM and EUW-LUR predictions while the upper and lower dash lines are limits of agreement defined as average difference ±1.96 standard deviation of the difference. Note that Google-MM LUR used an aethalometer for BC measurements while the EUW-LUR study measured BC by PM2.5 absorbance method. Differences in measurement methods could explain part of the BC discrepancy.
Figure 4. Bland–Altman plots for street level (top) and residential (bottom) NO2 (estimated by Google-MM and EUW-LUR19) and BC (estimated by Google-MM and EUW-LUR10). The middle dash line in each plot indicates mean difference between the Google-MM and EUW-LUR predictions while the upper and lower dash lines are limits of agreement defined as average difference ±1.96 standard deviation of the difference. Note that Google-MM LUR used an aethalometer for BC measurements while the EUW-LUR study measured BC by PM2.5 absorbance method. Differences in measurement methods could explain part of the BC discrepancy.
Atmosphere 14 01602 g004
Table 1. Descriptive statistics of Google-MM (during daytime) and EUW-LUR19 (during both daytime and nighttime) NO2 (μg/m3) and Google-MM (during daytime) and EUW-LUR10 BC (μg/m3) (during both daytime and nighttime) concentrations across 30,312 street segments and 76,752 residences in Copenhagen, Denmark.
Table 1. Descriptive statistics of Google-MM (during daytime) and EUW-LUR19 (during both daytime and nighttime) NO2 (μg/m3) and Google-MM (during daytime) and EUW-LUR10 BC (μg/m3) (during both daytime and nighttime) concentrations across 30,312 street segments and 76,752 residences in Copenhagen, Denmark.
PollutantLevel# Of ObservationsMinimum25th50thMean75th90thMaximumRatio of Percentile 97.5/2.5
Model-Google-MMEUW-LUR19Google-MMEUW-LUR19Google-MMEUW-LUR19Google-MMEUW-LUR19Google-MMEUW-LUR19Google-MMEUW-LUR19Google-MMEUW-LUR19Google-MMEUW-LUR19
NO2Street Level30,31288.11218.81520.817.320.91922.62824.96239.44.52.1
NO2Residential76,75288.51218.71420.615.020.41722.02123.35236.43.11.7
BCStreet Level30,3120.60.70.81.40.91.61.11.61.21.81.71.93.42.93.42.3
BCResidential76,7520.60.80.81.40.91.60.991.61.11.71.31.93.42.82.41.9
Abbreviations: Google-MM (Google Air View-based land use regression); EUW (Effects of Low-Level AP: A Study in Europe); NO2 (nitrogen dioxide); BC (black carbon).
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Tayebi, S.; Kerckhoffs, J.; Khan, J.; de Hoogh, K.; Chen, J.; Taghavi-Shahri, S.M.; Bergmann, M.L.; Cole-Hunter, T.; Lim, Y.-H.; Mortensen, L.H.; et al. Comparison of NO2 and BC Predictions Estimated Using Google Street View-Based and Conventional European-Wide LUR Models in Copenhagen, Denmark. Atmosphere 2023, 14, 1602. https://doi.org/10.3390/atmos14111602

AMA Style

Tayebi S, Kerckhoffs J, Khan J, de Hoogh K, Chen J, Taghavi-Shahri SM, Bergmann ML, Cole-Hunter T, Lim Y-H, Mortensen LH, et al. Comparison of NO2 and BC Predictions Estimated Using Google Street View-Based and Conventional European-Wide LUR Models in Copenhagen, Denmark. Atmosphere. 2023; 14(11):1602. https://doi.org/10.3390/atmos14111602

Chicago/Turabian Style

Tayebi, Shali, Jules Kerckhoffs, Jibran Khan, Kees de Hoogh, Jie Chen, Seyed Mahmood Taghavi-Shahri, Marie L. Bergmann, Thomas Cole-Hunter, Youn-Hee Lim, Laust H. Mortensen, and et al. 2023. "Comparison of NO2 and BC Predictions Estimated Using Google Street View-Based and Conventional European-Wide LUR Models in Copenhagen, Denmark" Atmosphere 14, no. 11: 1602. https://doi.org/10.3390/atmos14111602

APA Style

Tayebi, S., Kerckhoffs, J., Khan, J., de Hoogh, K., Chen, J., Taghavi-Shahri, S. M., Bergmann, M. L., Cole-Hunter, T., Lim, Y. -H., Mortensen, L. H., Hertel, O., Reeh, R., Schwartz, J., Hoek, G., Vermeulen, R., Jovanovic Andersen, Z., Loft, S., & Amini, H. (2023). Comparison of NO2 and BC Predictions Estimated Using Google Street View-Based and Conventional European-Wide LUR Models in Copenhagen, Denmark. Atmosphere, 14(11), 1602. https://doi.org/10.3390/atmos14111602

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop