Skip to main content
NCBI home page
ACS AuthorChoice logoLink to ACS AuthorChoice
. 2021 Oct 25;55(22):15505–15518. doi: 10.1021/acs.est.1c03748

Substantial Reduction in Particulate Matter Air Pollution across Europe during 2006–2019: A Spatiotemporal Modeling Analysis

Anton Beloconi †,‡, Penelope Vounatsou †,‡,*
PMCID: PMC8600664  PMID: 34694135

Abstract

graphic file with name es1c03748_0005.jpg

Air pollution poses the largest environmental health risk in Europe. Particulate matter (PM) concentrations are the most harmful pollutants representing the main air quality indicator in the Sustainable Development Goals (SDGs). The air quality surveillance in Europe is based on a monitoring network that is too coarse for a comprehensive evaluation of the air pollution burden. We link raw pollutant data with remotely sensed products using Bayesian geostatistical models and for the first time estimate pan-European near-surface concentrations of both fine (PM2.5) and coarse (PM10) particles at 1 km2 spatial resolution during 2006–2019. We evaluate the compliance with the air quality thresholds set by the World Health Organization (WHO) and the European Union (EU) and assess country-wise trends. The results show that during the last 14 years, PM10 and PM2.5 concentrations declined by 36.5% (95% credible interval: 30.3, 41.9%) and 39.1% (26.6, 50.5%), respectively. The number of people exposed to PM10 levels above the WHO thresholds decreased from 78.3% (52.6, 91.8%) in 2006 to 28.4% (16.2, 43.7%) in 2019; for PM2.5, the decrease was smaller: from 91.0% (61.3, 99.1%) exposed in 2006 to 53.6% (33.5, 76.3%) in 2019. Although there is a clear improvement in the overall picture, stricter measures are needed to ensure compliance with the WHO guidelines.

Keywords: PM10, PM2.5, long-term dynamics, air quality guidelines, population exposure, Gaussian processes, Bayesian inference, geostatistics

Short abstract

Our methodology quantifies the reduction in the particulate matter across Europe and provides evidence to guide locally adapted environmental protection strategies.

Introduction

According to the European Commission (EC), air pollution is currently the most important environmental risk to human health in Europe and is perceived as the second biggest environmental concern for Europeans, after climate change.1 The European Union (EU) has been working for decades to improve air quality by controlling emissions of harmful substances into the atmosphere, improving fuel quality, and integrating environmental protection requirements into the transport, industrial, and energy sectors.2

Ground-level particulate matter (PM) concentrations represent the main air quality indicator in several global agendas, including the Sustainable Development Goals (SDGs) related to health, welfare, and urbanization.3 The physical and chemical characteristics of PM affect its relevance to human health, as demonstrated by the observed differences in the behavior, the composition, and the health impacts for fine (PM2.5) and coarse (PM10) particles.4 The 2017 global burden of disease study estimated that long-term exposure to only PM2.5 caused around 3 million deaths, mainly due to ischemic heart disease, chronic obstructive pulmonary disease, lower respiratory infections, as well as tracheal, bronchus, and lung cancers.5

The European directives currently regulating ambient air concentrations are designed to avoid, prevent, or reduce the harmful effects of air pollutants on human health and the environment by implementing limit or target values for ambient concentrations. One of the main policy instrument within the EU is the Ambient Air Quality Directive,6 which sets the ambient air quality standards for several pollutants. In terms of long-term exposure to PM10, the average annual limit value is 40 μg/m3, and the deadline for Member States to meet this value was January 1, 2005. The deadline for meeting the annual target value of 25 μg/m3 for PM2.5 concentration was 2010, while the deadline for meeting the exposure concentration obligation of 20 μg/m3 (based on the average exposure indicator assessed as a 3 year running annual PM2.5 mean) was 2015. The Ambient Air Quality Directive also requires Member States to assess air pollution in all of their territories, to adopt and implement plans to improve air quality where standards are not met and to maintain it where the air quality is good.

The World Health Organization (WHO) cautioned that limit levels set in the EU directive were not sufficient to adequately protect human health7 and suggested to adhere to the stricter air quality guidelines (AQGs), recommended already in the global update of 2005,8 with annual limit values of 20 and 10 μg/m3 for PM10 and PM2.5, respectively. Since no threshold for PM has been identified below which no damage to health is observed,9 the recommended AQGs are considered as an acceptable and attainable goal that will minimize health effects; the aim, however, is to achieve the lowest concentrations possible.

For tracking progress in air quality improvements and for an accurate assessment of the environmental policies, reliable estimates of the spatiotemporal distribution of the pollutant concentrations are needed. In the last two decades, a substantial number of methods were developed, which assess long-term air pollution exposure and provide location-specific (e.g., residence or pixel level) estimates of PM concentration across large spatial domains.10−12 Many studies have used location-specific geographic characteristics, such as proximity to roads, to describe the small-scale variation in air pollutant levels, i.e., land-use regression (LUR) models.13,14 Other works modeled long-term PM averages using spatial modeling alone15,16 or in combination with LUR.17,18 Spatiotemporal models that incorporated land-use/cover and climatic data to predict PM levels at different temporal scales have also been developed.19,20 Many studies included observations of satellite-based aerosol optical depth (AOD) to predict PM concentrations over small21−23 and large24−26 domains.

In our earlier work,24 we used Bayesian geostatistical regression (GR) models to estimate pan-European concentrations of PM10 and PM2.5 in 2016 at a high spatial resolution of 1 km2. We have shown that these models not only correct for the bias in the effects of covariates (by modeling explicitly the spatial correlation structure in the data) but also outperform (in terms of predictive ability) the LUR and geographically weighted regression (GWR) models. Furthermore, in contrast to the methods commonly used in environmental epidemiology, the Bayesian formulation enabled quantification of the prediction uncertainty and probabilistic inferences on the exceedance of AQGs. Linking these estimates to the gridded population data allowed determining the number of people living in regions that exceed international thresholds.

Predictions of pollutant levels at high spatial resolution over large areas using Bayesian GR models are computationally complex.27 In fact, due to a high computational burden, previous GR models have provided gridded PM concentrations either for smaller areas of investigation or for one specific year.24,25,28 Pan-European estimates of the reduction in pollutant concentrations and of changes in population exposure, which take into account the geographical distribution of pollutants at high spatial resolution, are not available.

Here, we apply the Bayesian GR methodology to estimate for the first time PM10 and PM2.5 exposure in Europe at 1 km2 spatial resolution during 2006–2019, integrating information from monitoring stations, high-resolution satellite-derived products, land-use/cover variables, and meteorological data. We use a rigorous model selection procedure to evaluate which products contribute most to a more accurate prediction of the pollutants at an annual scale. We evaluate the compliance with the air quality thresholds set by the WHO and the EU’s Ambient Air Quality Directive and assess the exposure burden by linking the estimates to the gridded population data.

Materials and Methods

Study Area and Data

The PM10 and PM2.5 stations’ measurements were obtained from the AirBase29 and the Air Quality e-Reporting30 repositories, for the periods 2005–2012 and 2013–2019, respectively. Both databases are maintained by the European environment information and observation network (Eionet). The monitoring network covers up to 38 European countries, including 28 European Union (EU) member states and 33 member countries of the European Environment Agency (EEA). The analyses were based on the yearly-averaged data (reported in μg/m3) at stations with ≥75% data capture for each year. The number of stations available for both pollutants is typically increasing from 1 year to another, with a higher improvement observed in the PM2.5 coverage. The location of the stations for each analyzed year is illustrated in the Supporting Information (SI), Figures S1−S2, and S3−S4 for PM10 and PM2.5, respectively.

For the aerosol optical depth (AOD), the Moderate Resolution Imaging Spectroradiometer (MODIS) Terra and Aqua combined multiangle implementation of atmospheric correction (MAIAC)31 land AOD-gridded Level 2 product was used (MCD19A2 V6). The data set, comprised of daily observations at 1 km2 spatial resolution, was preprocessed using the Google Earth Engine (GEE) API.32 GEE makes it possible to rapidly process a vast amount of satellite imagery on a global scale with the power of Google’s cloud computing. GEE was used to calculate and extract the annual AOD averages for the years 2006–2019.

To better assess the spatial variability of pollutants’ concentrations across Europe, a number of additional products were used as predictors. To this end, a very large spatiotemporal covariate-data set was acquired. The focus was on satellite-derived products with continental or global coverage, comprising data related to land-use/cover, urban mapping, climate, and meteorology. References to data sources and preprocessing steps performed for each product are provided in the Supporting Information (SI); Table S1 summarizes all of the covariates used in the models.

For external validation, we compared our estimates with the surface PM10 and PM2.5 simulations obtained from the Ensemble of regional CTMs (median value of seven state-of-the-art European numerical air quality models—ENSEMBLE) provided by the Copernicus atmosphere monitoring service (CAMS).33 In particular, we used the reanalysis data set of the atmospheric composition produced by the European Centre for Medium-Range Weather Forecasts (ECMWF)34 available at 10 km2 spatial resolution for the years 2014–2018.

High-resolution gridded population data were obtained from the Gridded Population of the World, Version 4 (GPWv4) database. In particular, the population density (adjusted to match 2015 revision UN WPP country totals) data set,35 available at 30 arc-second (∼1 km2) spatial resolution for the years 2005, 2010, 2015, and 2020 was used to calculate the number of people exposed to elevated levels of air pollution in the years 2006, 2010, 2015, and 2019, respectively. The country borders were defined using the European administrative country boundaries’ shapefile from Eurostat’s GISCO service.36

Bayesian Geostatistical Regression

In the geostatistical framework, spatial correlation is modeled by location-specific random effects through a Gaussian process. The covariance matrix of this process assumes a correlation decay, which is a function of the distance between locations. Let y(sj) denote the realization of the spatial process Y(·) that represents the log-observed yearly-averaged PM10 or PM2.5 concentration for a particular year (between 2006 and 2019) at a station Inline graphic (location sj). We assumed a stationary, isotropic geostatistical regression (GR) model

graphic file with name es1c03748_m002.jpg 1

where β0 is the intercept, z(sj) = (z1(sj), ..., zp(sj))T is the vector of the p covariates at site sj, β = (β1, ..., βp)T is the vector of the corresponding coefficients, ω(sj) represents the spatial random effect, and ϵ(sj) is the random error.

We assumed that the spatial random effect ws = (w1, ..., wS)T arises from a multivariate normal distribution: Inline graphic with mean 0S, an S ×1 vector of zeros and spatial process variance, σω2. Rω is the S ×S dense correlation matrix with elements Inline graphic and Inline graphic is the Matern function given by: Inline graphic, where dij is the distance between stations i and j, κ is a scaling parameter, ν is a smoothing parameter (fixed to 1 in our application), and Kν is the modified Bessel function of second kind. This specification implies that the range r (the distance at which the spatial variance becomes less than 10%) is given by Inline graphic. Furthermore, we assumed that the ϵ(sj) is i.i.d. from a normal distribution Inline graphic.

The Bayesian model formulation is completed by specifying prior distributions for parameters and hyperparameters. The log-gamma priors were chosen for σϵ–2, σω, and r parametrized on the log scale, i.e., log(σϵ–2), log(σω)∼log Ga(1, 5 × 10–5), and log(r)∼log Ga(1,102). Normal priors Inline graphic~ were assigned to regression coefficients and a vague normal one for the intercept.

Model Fit and Predictions

All of the continuous covariates were standardized by subtracting the mean and dividing by the standard deviation (calculated using the yearly-averaged measurements from all of the monitoring stations). For the estimation of model parameters, covariates were extracted at the locations of stations, while for the prediction at unknown locations, each covariate was aggregated within a fixed 1 km2 grid using bilinear or nearest neighbor interpolation methods (for continuous and categorical data, respectively).

Model fit was done using the stochastic partial differential equation (SPDE) method and the integrated nested Laplace approximation (INLA) algorithm37,38 for fast approximation of marginal posterior distributions. In the SPDE/INLA approach, the spatial process is approximated by a Gaussian Markov random field (GMRF) with zero mean and a symmetric positive definite precision matrix Q (defined as the inverse of Σω = σω2Rω). First, a GMRF representation of the Matern field is constructed on a set of nonintersecting triangles partitioning the domain of the study area.38 Subsequently, the INLA algorithm estimates the posterior distribution of the latent Gaussian process and hyperparameters using the Laplace approximation.37 More details regarding this methodology are provided elsewhere.39 All of the models were fitted on an Intel Xeon E5-2697 CPU machine (2 × 2.60 GHz, 128 GB RAM) using the R-INLA package40 available within the R software.41

For each year, we fitted all possible combinations of covariates, i.e., 16 384 (=214) distinct models and ordered them according to Bayesian model comparison criteria. Particularly, models with the best predictive performance were selected based on the lowest logarithmic score (logscore)—a measure of the predictive ability of an individual model,42 given by: LCV = −∑s = 1S log CPOs, where the leave-one-out conditional predictive ordinate (CPO) is based on cross-validatory predictive densities π(Ys, Y–s) and is given by CPOs = π(Ys,Y–s) for each excluded location s. Subsequently, model performance was evaluated using the five-fold-cross-validation method; each data set was randomly divided five times in 80% (training set) and 20% (validation set) splits of the total number of PM10 or PM2.5 sites and the following performance metrics were examined for each fold: mean absolute error (MAE), mean absolute prediction error (MAPE), root mean squared error (RMSE), and the coefficient of determination (R2).

The models with the best predictive ability (for each particular year) were used to predict PM10 and PM2.5 concentrations over a gridded surface of 1 km2 spatial resolution covering the study area. To this end, 1000 samples from the posterior predictive distribution were drawn at the centroids of each grid cell (approximately 5.8 million pixels). The sample-based means and medians of these distributions were used for mapping. The prediction uncertainty was quantified using the sample-based standard deviations and half of the length of 95% credible intervals.43 For external validation, we compared our estimated surface PM10 and PM2.5 concentrations with the CAMS Ensemble CTM simulations (available at 10 km2 resolution and covering the years 2014–2018) extracted at the locations of monitoring sites.

The Bayesian framework allowed us to make probabilistic statements about areas exceeding the international AQGs. The probability of exceeding threshold limits set by the EU and the WHO was calculated at pixel level by the proportion of samples drawn from the predictive posterior distributions of the pollutant’s concentration with pollution levels above the threshold. We furthermore estimated the total number of people exposed to elevated PM10 and PM2.5 levels. In particular, we overlayed the gridded population data at 1 km2 spatial resolution with threshold maps and summed up, for each country, the population in pixels exceeding the threshold. Repeating it for each of the 1000 samples drawn from the posterior predictive distribution allowed us to estimate the number of exposed populations per country together with the prediction uncertainty.

Results and Discussion

Models with Highest Predictive Ability

Rigorous model selection indicated that although there is variability in the combination of covariates leading to the highest predictive ability between the years, for both PM10 and PM2.5 concentration, some predictors appear more often than others in optimal models (Tables S2 and S3 in the Supporting Information). Thus, for more than half of the years, we found an important positive association of PM10 and PM2.5 concentration with impervious surfaces (IMP) and night-time light intensity (NTL), a negative association of PM10 with elevation (DEM), normalized difference vegetation index (NDVI) and surface humidity (SHUM), and an important negative association of PM2.5 with elevation (DEM) and wind speed (WINDSP). Additionally, the highest levels of PM10 and PM2.5 concentration were estimated in urban and industrial areas (i.e., over land cover categories LC1 and LC2) followed by agricultural (LC3) and forest (LC4) areas (Tables S2 and S3). The AOD covariate was included (with a statistical important positive effect) in the best models for around half of the years.

Despite the fact that the number (N) and the location of monitoring stations change from one year to another, the estimated regression coefficients for each covariate appear to be similar between the years; however, the estimated range parameters (r) and the variances of the spatial process (σω2) vary between the years. The five-fold-cross-validation method indicated that the predictive ability is also comparable between the years, as indicated by the MAE, MAPE, RMSE, and R2 metrics. The increased number of monitoring stations, in general, leads to a better out-of-sample predictive ability of GR models.

Model-Based Maps of Pollutant Concentrations

The best resulting models for each year (i.e., the ones with the covariate combination giving the lowest logscore) were used to predict PM10 and PM2.5 concentrations across the continent. The estimates (i.e., the median of the posterior predictive distribution based on 1000 samples) for the years 2006, 2010, 2015, and 2019 are depicted in Figure 1 (left column) and Figure 2 (left column), for PM10 and PM2.5, respectively. Predictions and their uncertainty (expressed as the standard deviation of the posterior predictive distribution and the half of the length of 95% credible intervals) for each year during 2006–2019 are presented in Figures S1–S4 in the Supporting Information (SI).

Figure 1.

Figure 1

Open in a new tab

PM10 concentration. Predicted yearly-averaged PM10 concentration (i.e., median of the posterior predictive distribution) at 1 km2 spatial resolution in Europe (left column), probabilities that it exceeds the EU Air Quality Directive’s limit of 40 μg/m3 (middle) and WHO’s threshold of 20 μg/m3 (right), respectively, for the years 2006 (a–c), 2010 (d–f), 2015 (g–i), and 2019 (j–l).

Figure 2.

Figure 2

Open in a new tab

PM2. 5 concentration. Predicted yearly-averaged PM2.5 concentration (i.e., median of the posterior predictive distribution) at 1 km2 spatial resolution in Europe (left column), probabilities that it exceeds the EU Air Quality Directive’s limit of 25 μg/m3 (middle) and WHO’s threshold of 10 μg/m3 (right), respectively, for the years 2006 (a–c), 2010 (d–f), 2015 (g–i), and 2019 (j–l).

Comparisons between the medians of the posterior predictive distribution estimated using Bayesian GR models (at 1 km2 spatial resolution) and CTM simulations from the ENSEMBLE (at ∼10 km2 resolution) are shown in Figures S5 and S6 (in the SI). External validation indicates a good agreement between the estimates obtained using the two approaches and suggests that GR models can better capture measurements at monitoring stations, especially at locations with higher pollution levels (as indicated by the lower MAE and RMSE and the higher R2 values of the Bayesian GR models in each compared year).

Exceedance Probability Maps

The Bayesian framework allowed us to make probabilistic statements about areas exceeding the international air quality thresholds. We evaluated the compliance with the AQGs for annual averages, rather than for short-term or episodic exposure to pollutants. The results based on the EU Ambient Air Quality Directive, with annual limit values of 40 and 25 μg/m3 for PM10 and PM2.5, respectively, are shown in the middle columns of Figures 1 and 2. Estimates based on the stricter WHO recommendations, with thresholds of 20 and 10 μg/m3 for PM10 and PM2.5, are shown in the right columns of the same figures (Figures 1 and 2). Although the vast majority of the continent meets the requirements of EU AQG threshold standards, especially in the latest years, the exceedance probabilities based on WHO thresholds indicate that there are large parts of the continent where these limits are still to be reached, primarily for PM2.5 concentrations. Nonetheless, one can note a clear decrease in exceedance probabilities between 2006 and 2019 for both pollutants, especially in the central, the northern, and the western parts of Europe.

The historical estimates can also be used to evaluate the compliance of EEA member states in meeting the exposure concentration obligation target of 20 μg/m3 for PM2.5 concentration. Initially, this target is based on the average exposure indicator (AEI), calculated at a national level as an average of concentration levels over a 3 year period (i.e., the current one and the two preceding years), measured at urban background stations (representative of general urban population exposure) selected for this purpose by every national authority. Our models evaluate this target at 1 km2 spatial resolution. The left column of Figure 3 depicts the AEI estimated at a continental scale and the right column shows the probability of exceeding the corresponding threshold. It should be noted that, in this case, all of the monitoring stations (and not only the urban background stations) are taken into consideration.

Figure 3.

Figure 3

Open in a new tab

EU Average exposure indicator (AEI) for PM2.5. Predicted AEI (average of PM2.5 concentration levels over a 3 year period) at 1 km2 spatial resolution in Europe (left column) and probabilities that it exceeds the EU exposure concentration obligation of 20 μg/m3 (right), respectively, for the years 2010 (a, b), 2015 (c, d), and 2019 (e, f).

Particulate Matter Declines

The posterior median maps (Figures 1 and 2, left columns) were used to assess country-wide changes in PM10 and PM2.5 concentration between 2019 and the years 2015, 2010, and 2006. The estimated relative changes (and 95% Bayesian credible intervals) per country are shown in Tables 1 and 2 for PM10 and PM2.5, respectively. The highest relative reduction in PM10 between 2006 and 2019, with more than 48% decrease, was found in Iceland, Norway, Romania, and North Macedonia. In terms of PM2.5, countries with the highest reduction rate during the last 14 years are Iceland, Sweden, Slovenia, and Croatia. At continental scale, the results indicated a decrease of about 36.5% (30.3%, 41.9%) in PM10 and 39.1% (26.6%, 50.5%) in PM2.5 concentration between 2006 and 2019. It should be pointed out that the number of monitoring stations highly increased from 2006 to 2019 (especially the PM2.5 station coverage) and therefore the uncertainty of estimates is much higher in earlier years when compared to more recent years (see Figures S1, S2, S3, and S4 in the Supporting Information). Consequently, comparisons between the latest years (i.e., 2019 vs 2015 and 2015 vs 2010) presented in Tables 1 and 2 are generally more accurate.

Table 1. Average PM10 Concentrations (in μg/m3) and Relative Reduction (RR) (in %) between 2019 and the Years 2015 (RR2015–2019), 2010 (RR2010–2019), and 2006 (RR2006–2019)a,b.

  PM10 posterior median (95% Bayesian credible intervals)
country 2019 2015 RR2015–2019 (%) 2010 RR2010–2019 (%) 2006 RR2006–2019 (%)
(AL) Albania 17.7 (14.6, 21.9) 26.4 (22.6, 30.9) 32.7 (12.9, 48.0) 28.4 (22.9, 35.9) 37.3 (15.7, 53.8) 31.7 (24.3, 41.5) 44.2 (22.0, 59.7)
(AT) Austria 10.6 (9.9, 11.3) 12.4 (11.3, 13.6) 14.6 (4.5, 23.4) 15.6 (14.4, 16.9) 32.2 (24.7, 38.9) 18.2 (16.3, 20.5) 42.1 (33.9, 49.1)
(BA) Bosnia and Herzegovina 22.2 (19.8, 24.7) 31.0 (25.4, 38.6) 28.3 (9.1, 44.8) 23.6 (19.4, 28.7) 6.2 (−17.6, 25.4) 29.1 (23.3, 36.5) 23.9 (2.1, 41.0)
(BE) Belgium 15.6 (14.7, 16.3) 17.5 (16.3, 18.8) 11.0 (3.2, 18.5) 22.1 (20.8, 23.5) 29.5 (23.8, 34.6) 24.5 (21.9, 27.2) 36.5 (28.2, 43.3)
(BG) Bulgaria 20.4 (19.1, 21.9) 26.2 (24.3, 28.2) 22.4 (13.8, 29.9) 31.2 (28.6, 34.0) 34.6 (27.4, 41.8) 30.2 (27.2, 33.9) 32.7 (23.5, 40.9)
(CH) Switzerland 8.2 (7.6, 8.8) 9.4 (8.3, 10.5) 12.6 (−0.8, 24.0) 11.8 (10.8, 13.1) 30.9 (21.9, 39.1) 14.1 (12.4, 16.1) 42.2 (32.3, 50.3)
(CY) Cyprus 18.7 (15.1, 23.1) 27.2 (21.5, 34.4) 31.7 (4.8, 49.5) 34.4 (27.4, 45.4) 46.0 (24.3, 61.4) 35.9 (25.9, 49.8) 48.0 (23.2, 64.7)
(CZ) Czech Republic 15.6 (14.7, 16.5) 18.8 (15.2, 22.8) 17.0 (−3.1, 32.7) 22.1 (20.6, 23.6) 29.3 (22.7, 35.5) 25.8 (21.3, 30.9) 39.7 (26.7, 49.7)
(DE) Germany 13.0 (12.4, 13.6) 15.4 (14.4, 16.7) 15.6 (8.2, 22.6) 18.4 (17.3, 19.5) 29.4 (23.9, 34.2) 21.0 (19.5, 22.7) 38.2 (32.5, 43.5)
(DK) Denmark 14.0 (12.0, 16.4) 15.9 (12.8, 19.9) 12.0 (−14.4, 32.7) 16.4 (13.7, 19.5) 14.1 (−7.5, 32.8) 21.7 (17.3, 27.2) 35.3 (16.9, 50.8)
(EE) Estonia 10.7 (9.1, 12.6) 12.0 (9.9, 14.9) 10.8 (−13.4, 31.4) 13.3 (11.2, 15.9) 20.7 (−2.0, 35.3) 19.9 (14.2, 26.8) 46.0 (22.6, 62.1)
(EL) Greece 19.1 (17.0, 21.3) 22.6 (19.0, 27.0) 15.2 (−2.8, 31.2) 27.3 (23.3, 32.2) 29.9 (15.0, 42.9) 28.9 (23.4, 36.4) 33.7 (16.0, 50.0)
(ES) Spain 13.6 (12.7, 14.5) 15.3 (14.3, 16.4) 10.9 (1.2, 19.0) 16.1 (14.7, 17.7) 15.0 (5.1, 25.4) 22.1 (19.7, 25.0) 38.3 (29.7, 46.4)
(FI) Finland 7.7 (6.6, 9.2) 8.3 (7.0, 9.8) 6.3 (−18.0, 24.2) 11.3 (9.1, 13.9) 31.3 (10.2, 46.4) 13.8 (10.2, 18.6) 43.9 (22.3, 59.9)
(FR) France 11.8 (11.3, 12.4) 14.1 (13.4, 14.8) 16.0 (10.0, 21.7) 16.6 (15.7, 17.6) 28.8 (23.3, 34.1) 14.4 (13.5, 15.6) 18.1 (10.3, 25.1)
(HR) Croatia 18.3 (16.7, 20.2) 24.5 (21.9, 27.2) 25.0 (12.9, 34.8) 21.9 (19.1, 25.0) 16.4 (0.3, 30.0) 27.9 (24.0, 32.7) 34.7 (21.8, 45.7)
(HU) Hungary 20.3 (18.6, 22.1) 23.3 (19.0, 29.4) 12.9 (−8.8, 31.0) 24.6 (22.5, 26.9) 17.7 (6.7, 28.0) 28.8 (23.0, 35.8) 30.1 (10.6, 44.4)
(IE) Ireland 11.3 (10.3, 12.5) 10.8 (8.9, 13.8) –4.0 (−30.6, 21.5) 13.2 (11.2, 15.5) 14.6 (−3.8, 29.5) 14.0 (10.4, 18.7) 19.2 (−10.6, 41.3)
(IS) Iceland 7.2 (5.6, 9.2) 15.7 (7.0, 35.9) 54.0 (0.5, 80.9) 13.8 (10.5, 18.4) 48.1 (24.5, 63.7) 16.0 (10.0, 25.5) 55.3 (27.2, 73.1)
(IT) Italy 16.0 (15.2, 16.9) 18.3 (16.7, 20.1) 12.2 (2.2, 21.4) 18.5 (17.3, 19.9) 13.5 (6.0, 20.8) 23.1 (20.4, 26.2) 30.5 (20.6, 39.6)
(LT) Lithuania 17.8 (16.1, 19.6) 18.6 (16.6, 20.7) 4.2 (−11.6, 17.5) 20.9 (18.4, 23.8) 15.1 (−0.3, 27.5) 21.1 (17.7, 25.1) 15.7 (−3.1, 30.6)
(LU) Luxembourg 11.7 (10.6, 12.8) 14.4 (12.8, 16.1) 19.2 (5.9, 30.2) 16.7 (15.0, 18.4) 30.1 (19.0, 39.7) 17.7 (15.3, 20.4) 34.1 (21.4, 44.7)
(LV) Latvia 16.0 (13.7, 18.6) 17.3 (14.8, 20.3) 8.2 (−12.9, 25.4) 17.9 (15.0, 21.3) 11.5 (−12.6, 28.7) 19.6 (14.1, 27.8) 18.8 (−15.9, 44.0)
(ME) Montenegro 17.0 (13.5, 21.2) 26.9 (22.5, 31.7) 36.4 (16.2, 52.5) 24.6 (20.9, 29.5) 31.0 (8.4, 49.0) 26.8 (18.3, 37.5) 36.4 (1.4, 58.5)
(MK) North Macedonia 22.6 (19.8, 25.9) 35.2 (31.7, 39.3) 35.9 (23.1, 45.4) 36.7 (31.6, 42.3) 38.2 (24.5, 49.7) 43.8 (38.2, 50.7) 48.5 (36.2, 58.8)
(NL) Netherlands 16.5 (15.5, 17.6) 17.7 (15.9, 19.7) 6.6 (−5.9, 17.9) 23.0 (21.5, 24.9) 28.4 (20.9, 34.9) 27.0 (23.4, 30.8) 38.8 (29.0, 47.0)
(NO) Norway 7.8 (6.7, 9.1) 9.3 (7.7, 11.3) 16.1 (−6.1, 33.8) 11.3 (9.2, 13.7) 29.8 (9.4, 45.9) 15.2 (11.0, 20.1) 48.3 (27.0, 61.5)
(PL) Poland 20.3 (19.3, 21.3) 25.6 (23.0, 28.6) 20.5 (11.0, 29.9) 27.8 (25.9, 29.8) 26.9 (20.6, 33.1) 29.6 (26.4, 33.1) 31.3 (22.6, 39.3)
(PT) Portugal 12.9 (11.8, 14.2) 15.5 (13.9, 17.5) 17.3 (3.7, 28.6) 17.7 (15.8, 20.1) 27.3 (15.6, 37.4) 23.5 (20.2, 27.7) 45.0 (34.4, 55.0)
(RO) Romania 18.0 (16.9, 19.2) 20.0 (17.4, 23.1) 10.1 (−4.1, 23.2) 20.6 (19.0, 22.4) 12.8 (3.1, 21.5) 35.5 (30.1, 41.9) 49.4 (38.8, 57.6)
(RS) Serbia 25.2 (23.2, 27.4) 41.2 (35.1, 49.5) 38.9 (26.2, 49.7) 32.6 (27.8, 38.7) 22.8 (7.0, 36.7) 39.7 (32.1, 49.9) 36.7 (19.5, 50.2)
(SE) Sweden 9.2 (8.1, 10.4) 9.9 (8.2, 11.9) 7.1 (−14.4, 25.6) 10.5 (8.7, 12.7) 12.2 (−11.3, 29.0) 15.8 (11.8, 20.9) 41.4 (21.4, 56.8)
(SI) Slovenia 13.9 (12.8, 15.1) 17.4 (15.9, 18.9) 20.0 (9.6, 29.1) 18.6 (17.0, 20.2) 25.2 (15.7, 33.4) 21.6 (19.2, 24.4) 35.6 (25.0, 44.7)
(SK) Slovakia 17.5 (16.3, 18.8) 21.7 (16.3, 29.0) 19.5 (−9.5, 39.9) 26.3 (24.2, 28.6) 33.4 (25.2, 40.0) 28.0 (21.6, 36.0) 37.3 (17.9, 52.2)
(UK) United Kingdom 11.3 (10.6, 12.1) 11.4 (10.0, 12.9) 0.5 (−15.0, 14.1) 13.4 (12.1, 15.0) 15.3 (5.1, 25.8) 17.1 (14.5, 20.0) 33.6 (21.4, 44.2)
Whole study area 13.6 (13.0, 14.2) 16.3 (15.4, 17.2) 16.4 (10.1, 21.9) 17.9 (16.8, 19.0) 23.7 (17.5, 29.2) 21.5 (19.8, 23.4) 36.5 (30.3, 41.9)
Open in a new tab
a

The medians and the 95% Bayesian credible intervals of the posterior distributions are presented. No statistically important reduction is highlighted in bold.

b

Relative reduction (%) 2015–2019 is calculated as: Inline graphic using the 1000 posterior samples. The medians and the 95% Bayesian credible intervals of the resulting RR are shown in the table. The RR2010–2019 and RR2006–2019 are calculated similarly.

Table 2. Average PM2.5 Concentrations (in μg/m3) and Relative Reduction (RR) (in %) between 2019 and the years 2015 (RR2015–2019), 2010 (RR2010–2019), and 2006 (RR2006–2019)a,b.

  PM2.5 posterior median (95% Bayesian credible intervals)
country 2019 2015 RR2015-2019 (%) 2010 RR2010-2019 (%) 2006 RR2006-2019 (%)
(AL) Albania 11.4 (9.3, 14.3) 18.7 (14.5, 23.2) 38.6 (13.6, 54.6) 14.7 (10.2, 20.6) 22.4 (−15.0, 47.4) 20.8 (11.5, 37.7) 45.2 (−2.9, 70.3)
(AT) Austria 7.5 (6.9, 8.3) 9.3 (8.1, 10.6) 18.9 (4.5, 30.6) 12.1 (10.6, 14.0) 38.1 (27.8, 47.2) 13.8 (11.0, 17.4) 46.0 (31.0, 57.3)
(BA) Bosnia and Herzegovina 16.3 (13.9, 19.1) 9.6 (7.4, 12.4) –68.6 (−136.0, −24.6) 27.0 (21.1, 34.8) 39.7 (18.7, 54.9) 24.8 (17.6, 35.7) 35.1 (3.7, 55.1)
(BE) Belgium 9.3 (8.6, 10.0) 11.0 (10.2, 12.0) 15.6 (5.7, 24.4) 15.4 (14.2, 16.8) 39.8 (32.7, 45.9) 13.3 (10.9, 16.0) 30.0 (12.9, 42.8)
(BG) Bulgaria 14.1 (12.2, 16.4) 16.1 (13.6, 19.1) 12.9 (−10.4, 30.4) 18.0 (15.4, 20.8) 21.8 (1.7, 36.9) 22.3 (16.2, 30.1) 37.0 (9.1, 54.5)
(CH) Switzerland 5.7 (5.2, 6.3) 7.0 (6.0, 8.4) 18.7 (2.2, 33.7) 9.0 (7.7, 10.6) 36.1 (23.4, 47.6) 10.4 (8.2, 13.3) 45.5 (29.3, 57.6)
(CY) Cyprus 10.7 (8.4, 13.6) 12.6 (10.5, 15.1) 15.5 (−13.4, 37.6) 16.1 (13.3, 19.4) 33.7 (9.5, 51.5) 13.3 (5.1, 32.2) 20.4 (−116.3, 68.1)
(CZ) Czech Republic 11.6 (10.8, 12.5) 13.8 (11.1, 17.6) 16.3 (−5.8, 34.5) 17.5 (15.7, 19.4) 33.5 (24.3, 41.8) 19.1 (15.8, 22.8) 39.0 (25.3, 49.9)
(DE) Germany 8.7 (8.2, 9.3) 10.6 (9.7, 11.8) 18.0 (7.8, 27.8) 14.4 (13.1, 15.7) 39.6 (32.6, 46.3) 14.4 (11.9, 17.3) 39.6 (26.0, 50.2)
(DK) Denmark 8.1 (6.9, 9.4) 8.7 (7.1, 10.8) 7.2 (−21.1, 28.7) 11.7 (9.7, 13.8) 30.6 (13.0, 45.3) 10.8 (7.6, 15.3) 25.0 (−13.1, 48.3)
(EE) Estonia 4.8 (4.0, 5.8) 6.4 (5.3, 7.7) 24.3 (1.7, 41.3) 7.9 (6.5, 9.4) 38.7 (20.7, 52.7) 9.0 (5.6, 14.5) 47.0 (10.8, 67.5)
(EL) Greece 11.5 (9.9, 13.6) 17.1 (13.3, 22.6) 32.9 (9.9, 51.8) 13.6 (9.7, 20.0) 15.4 (−23.1, 43.7) 17.1 (9.8, 29.4) 31.9 (−18.1, 62.2)
(ES) Spain 7.6 (6.9, 8.4) 8.4 (7.5, 9.4) 9.8 (−6.2, 22.7) 8.7 (7.6, 10.1) 13.1 (−3.6, 27.9) 10.3 (8.2, 13.0) 26.6 (7.1, 43.2)
(FI) Finland 3.9 (2.9, 5.4) 5.2 (4.1, 6.6) 24.5 (−13.6, 48.2) 6.6 (5.2, 8.6) 41.0 (14.0, 60.2) 7.0 (4.9, 9.9) 44.0 (10.3, 64.8)
(FR) France 7.0 (6.6, 7.5) 9.6 (8.9, 10.3) 26.7 (18.7, 34.3) 12.8 (11.6, 14.1) 45.3 (39.0, 51.6) 9.6 (7.9, 11.7) 27.0 (10.2, 41.1)
(HR) Croatia 12.5 (11.1, 14.2) 12.6 (10.7, 14.7) 0.5 (−23.0, 17.2) 23.6 (18.9, 29.4) 46.9 (31.8, 59.3) 23.9 (17.3, 33.6) 47.7 (25.7, 63.5)
(HU) Hungary 13.2 (11.6, 15.1) 14.8 (10.8, 20.3) 11.1 (−25.3, 37.0) 18.5 (15.6, 21.9) 28.6 (11.8, 43.0) 20.7 (15.6, 27.9) 36.4 (12.3, 53.6)
(IE) Ireland 6.9 (5.9, 8.2) 6.2 (5.0, 7.7) –11.2 (−42.1, 15.4) 9.5 (7.5, 12.2) 27.0 (3.4, 46.1) 9.5 (5.0, 19.6) 27.1 (−43.0, 65.6)
(IS) Iceland 1.8 (1.4, 2.4) 7.0 (3.0, 16.6) 73.9 (37.6, 88.9) 5.5 (3.4, 9.2) 66.8 (44.4, 81.2) 12.7 (7.3, 21.7) 85.6 (73.9, 92.1)
(IT) Italy 10.1 (9.3, 10.9) 12.5 (11.3, 13.9) 19.5 (9.4, 29.6) 12.9 (11.4, 14.6) 22.1 (9.4, 32.3) 18.5 (14.7, 23.9) 45.8 (30.5, 58.6)
(LT) Lithuania 10.5 (9.3, 12.1) 13.1 (11.2, 15.6) 19.7 (1.3, 36.2) 14.4 (11.8, 17.4) 26.4 (7.7, 42.2) 15.7 (8.4, 29.0) 31.3 (−25.2, 64.3)
(LU) Luxembourg 6.7 (6.1, 7.5) 9.3 (8.1, 10.6) 27.5 (13.5, 38.6) 12.8 (11.2, 14.3) 47.1 (37.9, 54.8) 10.5 (7.7, 14.0) 35.8 (10.4, 52.9)
(LV) Latvia 8.2 (7.0, 9.8) 11.9 (9.9, 14.5) 30.4 (10.8, 46.3) 12.0 (9.9, 14.8) 31.6 (12.4, 46.9) 11.8 (6.5, 21.0) 30.4 (−28.7, 62.3)
(ME) Montenegro 13.3 (10.3, 17.3) 9.8 (7.8, 12.3) –36.4 (−92.7, 4.0) 18.3 (12.3, 27.0) 27.1 (−20.3, 55.0) 20.9 (12.5, 35.4) 36.4 (−11.6, 65.2)
(MK) North Macedonia 12.9 (10.5, 16.2) 24.4 (19.8, 30.9) 46.8 (26.5, 61.7) 16.4 (11.6, 24.5) 21.3 (−24.5, 48.3) 23.2 (14.2, 37.5) 43.7 (5.5, 66.7)
(NL) Netherlands 9.8 (8.9, 10.7) 11.3 (9.9, 12.9) 13.8 (−2.6, 26.6) 15.8 (14.2, 17.4) 38.4 (28.9, 45.9) 15.3 (10.9, 21.6) 36.4 (9.7, 56.1)
(NO) Norway 3.5 (2.9, 4.4) 4.3 (3.6, 5.4) 18.4 (−8.5, 38.4) 6.1 (4.8, 8.3) 42.4 (19.8, 59.5) 6.3 (4.5, 8.6) 43.7 (17.2, 61.2)
(PL) Poland 15.0 (14.0, 16.0) 18.4 (16.3, 21.1) 18.8 (6.0, 30.6) 23.0 (20.8, 25.4) 34.8 (26.7, 42.4) 22.8 (16.5, 34.2) 34.4 (6.6, 55.8)
(PT) Portugal 6.0 (5.3, 6.8) 8.0 (7.1, 9.0) 25.0 (11.3, 37.2) 8.0 (6.9, 9.3) 24.8 (9.2, 38.8) 11.1 (9.1, 13.9) 46.2 (31.4, 58.3)
(RO) Romania 12.5 (10.4, 15.2) 13.4 (10.8, 16.9) 7.0 (−25.4, 30.4) 12.8 (11.2, 14.6) 2.3 (−21.2, 22.6) 22.2 (15.3, 32.2) 43.5 (15.1, 62.9)
(RS) Serbia 19.9 (16.1, 25.0) 16.5 (12.4, 21.4) –20.7 (−71.4, 13.0) 24.4 (19.4, 31.2) 18.5 (−11.6, 40.1) 32.1 (22.4, 47.0) 38.0 (5.3, 59.8)
(SE) Sweden 3.4 (3.0, 4.0) 4.5 (3.5, 5.7) 24.0 (0.2, 43.0) 4.9 (4.0, 6.0) 31.1 (10.6, 46.3) 7.6 (5.6, 10.2) 55.0 (37.9, 67.6)
(SI) Slovenia 9.5 (8.5, 10.6) 12.6 (11.0, 14.2) 25.0 (10.0, 36.2) 15.5 (13.1, 18.1) 38.7 (25.4, 49.9) 18.7 (13.5, 26.1) 49.8 (27.3, 63.8)
(SK) Slovakia 12.7 (11.6, 13.9) 15.0 (11.0, 21.0) 15.7 (−18.4, 40.6) 18.1 (16.2, 20.3) 29.9 (19.1, 39.7) 18.2 (14.1, 24.6) 30.5 (8.4, 48.8)
(UK) United Kingdom 7.0 (6.4, 7.6) 7.2 (6.2, 8.3) 2.3 (−16.1, 17.3) 9.4 (8.6, 10.4) 25.3 (14.0, 34.5) 10.3 (7.5, 14.1) 31.2 (5.5, 50.4)
Whole study area 8.3 (7.9, 8.9) 10.1 (9.3, 10.9) 17.0 (8.0, 25.1) 12.1 (11.0, 13.2) 30.9 (22.5, 38.1) 13.7 (11.5, 16.7) 39.1 (26.6, 50.5)
Open in a new tab
a

The medians and the 95% Bayesian credible intervals of the posterior distributions are presented. No statistically important reduction is highlighted in bold.

b

Relative reduction (%) 2015–2019 is calculated as: Inline graphic using the 1000 posterior samples. The medians and the 95% Bayesian credible intervals of the resulting RR are shown in the table. The RR2010–2019 and RR2006–2019 are calculated similarly.

Population Living in Areas Exceeding WHO AQGs

The exceedance maps (Figures 1 and 2, right columns) were used to estimate the number of people exposed to PM levels above the WHO air quality guidelines, by overlaying the gridded population data at 1 km2 spatial resolution with the resulting threshold maps. The result shows that, over the whole study area, the population exposed to PM10 levels above the WHO thresholds decreased from 78.3% (52.6%, 91.8%) in 2006 to 28.4% (16.2%, 43.7%) in 2019. For PM2.5, the decrease in the number of exposed population was smaller with 91.0% (61.3%, 99.1%) exposed in 2006 compared to 53.6% (33.5%, 76.3%) in 2019. The country-level population exceedance is presented in Tables 3 and 4 for PM10 and PM2.5 concentrations, respectively. The highest relative reduction in population exposure to PM10 between 2006 and 2019 was estimated in Ireland, Luxembourg, Estonia, Norway, Iceland, Finland, Switzerland, Sweden, Germany, Denmark, Austria, United Kingdom, Portugal, and Netherlands. In fact, in all of these countries, the number of people exposed to elevated PM10 levels decreased by more than 85%, with some countries achieving almost 100% reduction. The lowest decrease was found in Cyprus, Serbia, Greece, Bosnia and Herzegovina, and Albania (less than 25% relative reduction). In terms of PM2.5, the countries with the highest decrease of population exposure are similar to the ones mentioned above for PM10. Thus, more than 85% relative reduction is found in Iceland, Luxembourg, Estonia, Norway, Finland, Sweden, Ireland, Portugal, and Switzerland. However, the relative reduction in countries such as Germany, Denmark, and United Kingdom were less pronounced (between 50 and 65%). The lowest reduction in the number of people exposed to elevated PM2.5 levels was estimated in Poland, Slovakia, Hungary, Serbia, Greece, Czech Republic and Bosnia, and Herzegovina (less than 10% reduction). Cyprus has shown an increase in PM2.5 exposure between 2006 and 2019; however, as already mentioned in the previous subsection, the estimates for 2006 are based on rather low number of stations and, therefore, have higher uncertainty (as also reflected in the large BCIs of the estimated reduction). The comparisons are more reliable for countries and years with higher coverage of the monitoring stations.

Table 3. Population (pop.) of Each Country (in Millions), Percentage of People within Each Country Exposed to PM10 Levels above the WHO Threshold, and Relative Reduction (RR) in the Number of People Exposed (in %) between 2019 and the Years 2015 (RR2015–2019), 2010 (RR2010–2019), and 2006 (RR2006–2019)a.

country pop. 2019 exposed 2019 (%) pop. 2015 exposed 2015 (%) RR2015–2019 (%) pop. 2010 exposed 2010 (%) RR2010–2019 (%) pop. 2006 exposed 2006 (%) RR2006–2019 (%)
(AL) Albania 2.9 77.3 (34.7, 91.9) 2.9 95.6 (80.2, 98.5) 17.3 2.9 97.2 (91.9, 99.3) 19.1 3.1 97.4 (91.5, 99.6) 24.3
(AT) Austria 8.8 1.4 (0.1, 6.0) 8.7 25.2 (6.3, 41.7) 94.1 8.5 61.4 (52.7, 69.5) 97.6 8.3 75.0 (64.0, 83.5) 98.0
(BA) Bosnia and Herzegovina 3.8 73.3 (64.2, 82.5) 3.8 87.9 (74.4, 95.5) 17.2 3.8 82.0 (58.6, 96.1) 11.9 3.8 94.6 (81.6, 99.7) 23.4
(BE) Belgium 11.7 27.8 (14.5, 42.0) 11.4 52.2 (21.7, 73.3) 45.9 11.0 94.1 (88.4, 96.6) 68.4 10.7 95.2 (89.5, 97.4) 67.7
(BG) Bulgaria 6.9 60.3 (49.1, 71.8) 7.2 86.9 (77.7, 91.8) 32.6 7.4 96.6 (94.1, 98.6) 41.9 7.7 89.6 (77.5, 96.7) 39.2
(CH) Switzerland 9.0 0.1 (0.0, 0.3) 8.7 3.4 (1.5, 12.2) 97.3 8.2 21.1 (12.3, 34.0) 99.5 7.7 60.9 (37.0, 78.1) 99.8
(CY) Cyprus 1.2 74.8 (30.6, 94.5) 1.2 96.2 (85.2, 98.1) 17.6 1.1 99.4 (97.4, 99.8) 17.1 1.0 97.8 (93.6, 98.4) 9.5
(CZ) Czech Republic 10.7 23.8 (16.1, 34.3) 10.6 63.0 (19.4, 88.5) 60.3 10.6 84.8 (76.5, 92.9) 71.7 10.3 95.1 (80.8, 99.0) 73.7
(DE) Germany 81.7 1.0 (0.1, 3.0) 82.0 15.7 (8.8, 29.6) 93.9 81.7 53.3 (43.7, 64.4) 98.2 82.5 76.4 (61.5, 87.3) 98.7
(DK) Denmark 5.8 1.2 (0.0, 12.3) 5.7 15.3 (0.0, 65.0) 90.6 5.6 13.6 (0.8, 45.7) 90.0 5.5 85.9 (35.9, 99.4) 98.3
(EE) Estonia 1.3 0.0 (0.0, 2.4) 1.3 0.2 (0.0, 14.0) 100.0 1.4 0.9 (0.0, 16.0) 100.0 1.4 69.9 (10.7, 98.9) 100.0
(EL) Greece 10.9 79.2 (69.2, 87.3) 11.0 85.8 (73.9, 94.1) 8.7 11.2 94.7 (87.3, 98.3) 18.8 11.1 94.6 (85.1, 98.6) 17.6
(ES) Spain 44.2 27.4 (20.8, 33.7) 44.2 41.8 (35.3, 49.0) 34.6 44.7 55.6 (42.3, 67.0) 51.2 42.1 85.4 (74.1, 93.1) 66.3
(FI) Finland 5.9 0.0 (0.0, 0.0) 5.9 0.0 (0.0, 0.2) 100.0 5.7 0.0 (0.0, 0.8) 100.0 5.6 9.5 (0.0, 27.9) 100.0
(FR) France 66.2 14.8 (9.5, 19.0) 64.9 30.8 (25.3, 37.5) 51.5 63.4 59.0 (52.1, 67.0) 73.9 61.7 33.2 (25.1, 41.5) 52.7
(HR) Croatia 4.2 53.7 (38.4, 65.8) 4.2 73.4 (64.3, 84.3) 28.4 4.3 82.9 (67.7, 91.5) 36.7 4.4 92.7 (83.5, 97.6) 44.5
(HU) Hungary 9.8 66.0 (48.7, 77.7) 10.0 88.9 (44.6, 99.0) 25.6 10.2 97.1 (88.1, 99.8) 33.7 10.3 98.2 (86.1, 99.9) 35.0
(IE) Ireland 4.9 0.0(0.0, 1.1) 4.7 0.0 (0.0, 13.7) 100.0 4.7 3.1 (0.0, 16.6) 100.0 4.2 5.2 (0.0, 38.4) 100.0
(IS) Iceland 0.3 0.0 (0.0, 0.1) 0.3 30.6 (0.0, 99.6) 100.0 0.3 3.2 (0.3, 11.2) 100.0 0.3 25.9 (0.1, 83.9) 100.0
(IT) Italy 60.5 66.5 (59.7, 73.0) 60.5 81.8 (71.8, 88.8) 18.4 60.3 83.4 (77.7, 88.1) 20.0 59.4 91.3 (82.2, 96.0) 25.7
(LT) Lithuania 2.9 28.8 (12.6, 55.3) 2.9 45.8 (22.0, 71.0) 36.9 3.2 76.2 (44.7, 96.7) 65.4 3.4 72.6 (31.1, 98.6) 65.9
(LU) Luxembourg 0.6 0.0 (0.0, 0.0) 0.6 0.0 (0.0, 14.5) 100.0 0.5 17.4 (0.8, 46.3) 100.0 0.5 24.1 (2.5, 71.6) 100.0
(LV) Latvia 2.1 41.5 (8.5, 65.6) 2.1 45.2 (7.9, 75.1) 10.8 2.2 64.0 (37.0, 85.6) 40.6 2.4 63.1 (2.5, 99.7) 43.5
(ME) Montenegro 0.6 34.1 (9.9, 60.7) 0.6 73.1 (54.7, 89.0) 52.8 0.6 76.7 (56.3, 92.6) 54.9 0.6 83.3 (43.1, 99.7) 56.1
(MK) North Macedonia 2.1 65.0 (53.4, 76.7) 2.1 94.0 (88.7, 97.6) 30.5 2.1 94.7 (89.1, 98.7) 30.5 2.1 96.5 (91.3, 99.7) 31.2
(NL) Netherlands 17.7 10.6 (2.8, 23.7) 17.4 26.3 (3.4, 58.9) 58.0 17.1 95.9 (86.7, 99.7) 88.5 16.8 99.8 (96.9, 99.8) 88.8
(NO) Norway 5.6 0.0 (0.0, 0.2) 5.3 0.3 (0.0, 5.6) 100.0 5.0 12.6 (1.0, 36.2) 100.0 4.7 32.7 (5.9, 67.2) 100.0
(PL) Poland 38.9 71.8 (64.5, 78.6) 39.1 90.9 (86.1, 95.4) 21.3 39.1 96.1 (91.2, 99.2) 25.5 39.0 98.8 (94.6, 99.8) 27.3
(PT) Portugal 9.6 7.7 (1.2, 24.5) 9.7 42.3 (11.1, 69.1) 80.8 10.0 77.7 (64.1, 88.6) 90.4 9.9 95.7 (87.2, 99.1) 92.1
(RO) Romania 19.2 71.5 (38.5, 79.4) 19.8 70.2 (54.5, 84.3) 4.5 20.6 67.9 (55.9, 80.0) 4.1 21.7 98.3 (94.6, 99.8) 35.5
(RS) Serbia 8.8 93.7 (88.0, 96.5) 9.0 99.4 (98.3, 99.8) 7.6 9.2 99.0 (97.7, 99.7) 9.5 9.3 99.5 (97.9, 100.0) 11.2
(SE) Sweden 10.6 0.2 (0.0, 1.2) 10.2 0.6 (0.0, 8.0) 79.1 9.8 6.0 (0.4, 23.8) 96.9 9.4 45.1 (12.6, 78.4) 99.5
(SI) Slovenia 2.1 17.1 (6.0, 34.5) 2.1 67.1 (49.4, 79.2) 73.9 2.1 76.3 (66.2, 82.8) 77.1 2.0 85.4 (74.3, 93.2) 79.1
(SK) Slovakia 5.5 34.8 (20.9, 52.7) 5.5 80.4 (17.5, 97.7) 53.7 5.5 98.1 (94.7, 99.6) 64.1 5.4 98.1 (80.7, 99.9) 63.6
(UK) United Kingdom 67.3 2.9 (0.3, 11.3) 65.3 1.0 (0.0, 14.0) -163.5 63.3 18.1 (3.9, 38.4) 82.3 60.8 70.8 (41.9, 90.1) 95.3
Whole study area 545.3 28.4 (16.2, 43.7) 542.0 42.2 (25.1, 64.5) 13.8 538.3 61.7 (43.3, 82.9) 33.3 530.1 78.3 (52.6, 91.8) 49.9
Open in a new tab
a

The medians and the 95% Bayesian credible intervals of the posterior distributions are presented.

Table 4. Population (Pop.) of Each Country (in Millions), Percentage of People within Each Country Exposed to PM2.5 Levels above the WHO Threshold, and Relative Reduction (RR) in the Number of People Exposed (in %) between 2019 and the Years 2015 (RR2015–2019), 2010 (RR2010–2019), and 2006 (RR2006–2019)a.

country pop. 2019 exposed 2019 (%) pop. 2015 exposed 2015 (%) RR2015–2019 (%) pop. 2010 exposed 2010 (%) RR2010–2019 (%) pop. 2006 exposed 2006 (%) RR2006–2019 (%)
(AL) Albania 2.9 92.1 (77.5, 98.0) 2.9 99.4 (97.5, 99.9) 5.9 2.9 95.5 (81.2, 99.7) 2.0 3.1 99.7 (85.4, 100.0) 11.1
(AT) Austria 8.8 51.1 (40.8, 62.3) 8.7 77.0 (62.5, 85.1) 31.8 8.5 88.4 (83.0, 93.2) 40.1 8.3 91.1 (82.9, 97.8) 40.8
(BA) Bosnia and Herzegovina 3.8 91.8 (82.0, 97.4) 3.8 60.6 (25.7, 83.4) -47.5 3.8 100.0 (98.6, 100.0) 10.0 3.8 100.0 (95.0, 100.0) 9.6
(BE) Belgium 11.7 68.2 (54.2, 80.3) 11.4 91.1 (81.9, 95.6) 22.5 11.0 99.8 (98.6, 100.0) 27.3 10.7 97.0 (87.6, 100.0) 22.2
(BG) Bulgaria 6.9 88.6 (77.5, 95.1) 7.2 95.5 (77.5, 99.5) 10.1 7.4 97.7 (91.2, 99.8) 15.2 7.7 97.8 (87.7, 100.0) 18.3
(CH) Switzerland 9.0 8.1 (3.6, 18.1) 8.7 50.9 (20.7, 81.1) 82.7 8.2 85.0 (68.6, 92.1) 89.3 7.7 88.4 (75.9, 95.8) 89.1
(CY) Cyprus 1.2 88.9 (46.3, 98.7) 1.2 95.6 (85.8, 98.0) 2.3 1.1 99.6 (96.8, 100.0) 1.5 1.0 96.2 (0.0, 100.0) -12.2
(CZ) Czech Republic 10.7 89.3 (81.6, 95.1) 10.6 97.8 (84.9, 99.9) 7.5 10.6 99.9 (98.9, 100.0) 10.0 10.3 99.9 (98.1, 100.0) 7.4
(DE) Germany 81.7 33.7 (23.1, 48.0) 82.0 80.6 (59.6, 92.6) 57.5 81.7 98.7 (95.2, 99.9) 65.8 82.5 95.2 (79.4, 99.8) 64.3
(DK) Denmark 5.8 31.0 (3.0, 60.4) 5.7 44.4 (6.1, 82.7) 30.0 5.6 92.0 (57.4, 99.4) 63.7 5.5 83.7 (29.6, 100.0) 58.2
(EE) Estonia 1.3 0.0 (0.0, 0.0) 1.3 1.5 (0.0, 14.2) 100.0 1.4 14.0 (0.3, 33.9) 100.0 1.4 45.5 (0.0, 98.2) 100.0
(EL) Greece 10.9 89.4 (78.2, 95.8) 11.0 94.6 (84.9, 99.1) 6.7 11.2 93.6 (45.7, 99.4) 6.8 11.1 95.3 (65.4, 100.0) 7.3
(ES) Spain 44.2 43.7 (31.6, 55.3) 44.2 53.8 (39.4, 67.3) 18.2 44.7 66.7 (52.2, 79.5) 34.8 42.1 83.5 (68.6, 95.1) 44.7
(FI) Finland 5.9 0.0 (0.0, 0.2) 5.9 0.0 (0.0, 1.2) 100.0 5.7 1.3 (0.0, 14.6) 100.0 5.6 12.4 (0.1, 49.4) 100.0
(FR) France 66.2 31.0 (24.8, 37.4) 64.9 74.2 (65.1, 82.2) 57.4 63.4 96.5 (92.7, 98.6) 66.3 61.7 71.4 (54.7, 88.6) 53.5
(HR) Croatia 4.2 76.0 (68.8, 83.6) 4.2 84.3 (74.7, 92.6) 11.4 4.3 99.7 (98.3, 100.0) 26.4 4.4 99.8 (96.3, 100.0) 27.4
(HU) Hungary 9.8 97.9 (89.0, 99.8) 10.0 99.6 (63.3, 100.0) 2.8 10.2 100.0 (98.3, 100.0) 5.1 10.3 100.0 (97.5, 100.0) 5.8
(IE) Ireland 4.9 1.8 (0.0, 13.9) 4.7 0.0 (0.0, 9.8) -127.7 4.7 57.3 (9.3, 95.2) 96.4 4.2 59.4 (0.0, 98.5) 95.3
(IS) Iceland 0.3 0.0 (0.0, 0.0) 0.3 15.2 (0.0, 99.2) 100.0 0.3 0.4 (0.0, 10.7) 100.0 0.3 93.6 (45.5, 99.6) 100.0
(IT) Italy 60.5 86.3 (79.4, 91.4) 60.5 94.5 (88.8, 97.3) 8.5 60.3 94.5 (85.4, 98.6) 8.2 59.4 99.0 (91.7, 100.0) 10.7
(LT) Lithuania 2.9 77.8 (44.2, 97.6) 2.9 98.3 (77.1, 100.0) 20.6 3.2 99.8 (86.5, 100.0) 28.7 3.4 99.4 (26.3, 99.7) 30.3
(LU) Luxembourg 0.6 0.0 (0.0, 0.0) 0.6 58.2 (15.0, 89.4) 100.0 0.5 100.0 (92.1, 100.0) 100.0 0.5 80.0 (5.8, 100.0) 100.0
(LV) Latvia 2.1 40.6 (3.6, 74.9) 2.1 90.4 (64.3, 98.9) 55.0 2.2 93.0 (75.7, 100.0) 59.0 2.4 88.9 (3.7, 100.0) 55.4
(ME) Montenegro 0.6 83.5 (51.4, 98.5) 0.6 55.0 (35.3, 79.3) -48.0 0.6 99.0 (75.6, 100.0) 12.6 0.6 100.0 (71.7, 100.0) 13.5
(MK) North Macedonia 2.1 79.0 (61.3, 92.1) 2.1 99.1 (94.8, 100.0) 19.5 2.1 90.3 (66.2, 99.9) 10.0 2.1 99.2 (79.1, 100.0) 16.4
(NL) Netherlands 17.7 64.0 (38.0, 87.3) 17.4 94.8 (70.5, 100.0) 29.7 17.1 100.0 (100.0, 100.0) 33.9 16.8 100.0 (81.0, 100.0) 31.1
(NO) Norway 5.6 0.0 (0.0, 0.5) 5.3 0.1 (0.0, 4.3) 100.0 5.0 21.7 (8.0, 39.3) 100.0 4.7 31.6 (11.1, 64.3) 100.0
(PL) Poland 38.9 97.8 (94.8, 99.7) 39.1 99.8 (98.2, 100.0) 2.4 39.1 100.0 (99.9, 100.0) 2.6 39.0 99.9 (97.1, 100.0) 2.1
(PT) Portugal 9.6 7.1 (0.0, 36.8) 9.7 41.1 (23.9, 67.3) 83.0 10.0 31.8 (10.8, 51.3) 77.1 9.9 94.5 (81.8, 99.5) 92.7
(RO) Romania 19.2 92.3 (80.1, 97.8) 19.8 91.3 (73.7, 98.3) 2.1 20.6 85.7 (75.5, 92.8) -0.1 21.7 98.2 (84.7, 100.0) 16.3
(RS) Serbia 8.8 98.9 (97.7, 99.5) 9.0 98.8 (86.4, 99.9) 2.0 9.2 99.5 (97.5, 100.0) 5.0 9.3 100.0 (98.8, 100.0) 6.7
(SE) Sweden 10.6 0.3 (0.0, 5.4) 10.2 3.7 (0.0, 13.1) 90.0 9.8 9.1 (2.1, 22.0) 96.5 9.4 58.5 (28.7, 87.4) 99.5
(SI) Slovenia 2.1 75.0 (52.2, 86.8) 2.1 95.1 (88.2, 98.4) 21.0 2.1 98.8 (95.8, 99.8) 23.0 2.0 99.5 (95.1, 100.0) 21.6
(SK) Slovakia 5.5 95.4 (88.6, 98.9) 5.5 99.1 (80.7, 99.9) 2.9 5.5 100.0 (99.3, 100.0) 4.0 5.4 99.2 (90.0, 100.0) 2.4
(UK) United Kingdom 67.3 37.0 (17.5, 56.7) 65.3 40.2 (7.9, 66.2) 4.4 63.3 87.0 (79.0, 92.9) 54.6 60.8 88.3 (46.2, 99.2) 51.6
Whole study area 545.3 53.6 (33.5, 76.3) 542.0 74.2 (45.3, 90.2) 20.6 538.3 88.6 (72.1, 95.5) 35.0 530.1 91.0 (61.3, 99.1) 37.4
Open in a new tab
a

The medians and the 95% Bayesian credible intervals of the posterior distributions are presented.

The changes in the averaged PM concentration (Tables 1 and 2) and in the population exposure (Tables 3 and 4) estimates are complementary to each other. In fact, reducing the actual concentration, is as important as reaching the AQGs. Thus, some countries, despite not reaching the limit values suggested by the WHO, have highly decreased the actual concentration levels within the studied time period.

Discussion

Our work is the first to estimate the pan-European spatiotemporal PM10 and PM2.5 dynamics at 1 km2 spatial resolution during the period 2006–2019. We have quantified the relative reduction in the air pollution burden and identified, at a fine geographical scale, areas where the thresholds defined by the international air quality guidelines were exceeded. We employed validated, state-of-the-art Bayesian methodology incorporating data from monitoring stations and high-resolution remotely sensed products. The results have shown that during the last 14 years, PM10 and PM2.5 concentrations in Europe declined by 36.5 and 39.1%, respectively. The number of people exposed to PM10 levels above the WHO thresholds decreased from 78.3% in 2006 to 28.4% in 2019; for PM2.5, the decrease was smaller: from 91.0% exposed in 2006 to 53.6% in 2019.

Rigorous variable selection, performed separately for each year, identified predictors that contribute most to a more accurate exposure estimation. The optimal set of covariates differs between the years; however, estimated regression coefficients appear to be similar (Tables S2 and S3 in the Supporting Information (SI)). This finding suggests that the effect of each covariate on the annually averaged PM10 and PM2.5 concentration levels are rather stable over time and therefore advocates valid inferences regarding the important (negative or positive) associations with both pollutants.

External validation of our estimates with simulations from the state-of-the-art chemical transport models (provided by CAMS during the years 2014–2018) revealed a good agreement between the values extracted at the locations of the monitoring stations, especially for PM2.5 concentrations (Figures S5 and S6 in the SI). In fact, predictions obtained using GR models (at 1 km2 spatial resolution) were closer to the measurements than the ENSEMBLE CTM simulations (∼ at 10 km2 resolution), especially at elevated pollution levels. This result implies that the higher spatial resolution of the estimates leads to a smaller underestimation of the exposure.

The Bayesian geostatistical framework allowed us to make probabilistic statements about areas exceeding the international air quality guidelines thresholds and evaluate the compliance with the European and global policies currently in place. We estimated that in some parts of the continent, notably in either less economically developed regions in Eastern Europe, like Bulgaria, Serbia, Romania, and North Macedonia or in industrial regions, like southern Poland or Po Valley in Northern Italy, the target values set by the EU Ambient Air Quality Directive were not met by the suggested deadlines for both PM10 and PM2.5 concentrations. Furthermore, in some smaller parts of the continent, predominantly in Eastern Europe, the PM2.5 limit values and average exposure obligation targets were not met even by 2019. In terms of stricter WHO AQG thresholds, we estimated a clear decrease in the exceedance probabilities of both pollutants between 2006 and 2019, especially in Western and Northern Europe; however, in most of the countries, the recommended limit values have not yet been reached.

The high-resolution predictions were used to evaluate country-wise trends in air pollution dynamics. By linking the estimated exceedance probability maps with gridded population data, we were able to assess the number of people exposed to elevated PM levels. We estimated that for some countries, such as Denmark, Germany, and the United Kingdom, the relative reduction in a number of people exposed to elevated PM10 concentration, during the last 14 years, was much higher than in those exposed to PM2.5. On the other hand, some countries, including Serbia, Bosnia and Herzegovina, and Poland, have shown a relatively low reduction in the population exposure to both pollutants. The lowest reduction in the number of people exposed to elevated PM2.5 was found in many southern countries, including Italy, Greece, Cyprus, Montenegro, and Albania. This result may partly reflect the influence of the Saharan dust episodes reaching Europe rather than the effect of the environmental policies.

At the continental scale, the results indicated a decrease of about 50 and 37% (between 2006 and 2019) in the number of people exposed to PM10 and PM2.5 concentrations above the WHO thresholds, respectively. Although there is an improvement in the overall picture, stricter measures are needed to ensure that compliance with the WHO guidelines is achieved. In a global study that modeled PM2.5 exposure at ∼10 km2 spatial resolution,26 it was estimated that 94% of the population in Western Europe and 88% of the population in Eastern Europe were in excess of WHO AQGs of 10 μg/m3 between 2001 and 2006. Findings put forth in a recent WHO report,44 also based on global estimates at ∼10 km2 spatial resolution, indicated that in 2014 just 1% of the population in low- and middle-income45 European countries and only 18% of the population in high-income countries breathed clean (below the WHO AQGs limit values) air, with values increasing to 6 and 24%, respectively, in the updated report for the year 2016.46 Our high-spatial-resolution (1 km2) Bayesian geostatistical analysis estimates (53.6% exposed to PM2.5 in 2019) are comparable to the ones reported in the previous studies and indicate a further decrease in PM2.5 exposure in Europe in 2019.

The main limitation of the data used in these analyses is the changing number of monitoring stations between the years. In particular, the number of sites monitoring PM2.5 septupled from 2006 to 2019. As for any statistical model, predictions rely on appropriate input data, and therefore, the accuracy of the estimates also varies between the years. Several empirical gap-filling methods have been previously used to increase the number of PM2.5 samples from the colocated PM10 measurements.47,48 Bayesian models allow similar analyses that can incorporate the additional prediction uncertainty within a single hierarchical formulation.49 However, this approach is computationally very intensive for the scale of analysis of our work. On the other hand, the maps of the prediction uncertainty, which is high in areas with few or no stations, identify potential locations for installing new monitoring stations, to reduce the prediction uncertainty.

To evaluate the impact of the small number of sampling PM locations in some countries and years, on the predictive ability of the models, we carried out a sensitivity analysis. In particular, for each of the years 2017–2019, we refitted the models using only ∼270 stations measuring PM2.5 concentration during the years 2006–2008 and compared the resulting pan-European estimates with those obtained using the full number of monitors. The results (Figure S7) suggested that the median of the posterior predictive distribution is not highly affected by the number of the stations used to fit the models; however, the uncertainty of the predictions is much larger when we have less observations, especially in the areas with a small number, or absence of stations.

Our model-based, high-resolution air pollution exposure estimates can further contribute to human and ecosystem health research in Europe. To our knowledge, this study is the first to compute and compare continental exceedance maps of PM10 and PM2.5 using the EU and the WHO air quality guidelines for the 2006–2019 period. The fine-scale predictions of the air pollution dynamics can be used in studies evaluating country-specific or regional environmental policies. Furthermore, the estimated trends could enable a more accurate assessment of the disease burden attributable to the changes in PM exposure during the last decade. The developed methodology provides data-driven evidence, which can support decision makers to evaluate the compliance to policies that are already in place, as well as to develop locally adapted environment protection and public health strategies.

Acknowledgments

The authors thank the financial support of the European Research Council (ERC) Advanced Grant (Project no. 323180).

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.est.1c03748.

  • Detailed information on the data sources, retrieval, and processing; results of the model selection for each pollutant and year; model-based maps of PM10 and PM2.5 concentrations for every year between 2006 and 2019; comparison between the Bayesian GR model estimates and CAMS Ensemble simulations for the years 2014–2018; and sensitivity analysis (PDF)

The authors declare no competing financial interest.

Supplementary Material

es1c03748_si_001.pdf (68.6MB, pdf)

References

  1. European Commission. Special Eurobarometer 468: Attitudes of European citizens towards the environment, 2017. http://data.europa.eu/euodp/en/data/dataset/S2156_88_1_468_ENG (accessed March 1, 2021).
  2. European Environment Agency. Air quality in Europe - 2019 report. EEA Report No. 10/2019, 2019. https://www.eea.europa.eu/publications/air-quality-in-europe-2019 (accessed 1 March 2021).
  3. United Nations. Sustainable Development Goals, 17 goals to transform our world. 2015, https://www.un.org/sustainabledevelopment/sustainable-development-goals (accessed March 1, 2021).
  4. Anderson J. O.; Thundiyil J. G.; Stolbach A. Clearing the air: a review of the effects of particulate matter air pollution on human health. J. of Med. Toxicol. 2012, 8, 166–175. 10.1007/s13181-011-0203-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Stanaway J. D.; Afshin A.; Gakidou E.; Lim S. S.; Abate D.; Abate K. H.; Abdela J.; et al. Global, regional, and national comparative risk assessment of 84 behavioural, environmental and occupational, and metabolic risks or clusters of risks for 195 countries and territories, 1990-2017: a systematic analysis for the Global Burden of Disease Study 2017. Lancet 2018, 392, 1923–1994. 10.1016/S0140-6736(18)32225-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. EU. Directive 2008/50/EC of the European Parliament and of the Council of 21 May 2008 on ambient air quality and cleaner air for Europe, OJ L 152, 11.6.2008, 2008; pp 1–44. https://eur-lex.europa.eu/legal-content/EN/TXT/?uri/CELEX3A32008L0050 (accessed 1 March 2021).
  7. World Health Organization. Review of evidence on health aspects of air pollution–REVIHAAP project, Technical Report; World Health Organization, Regional Office for Europe: Copenhagen, Denmark, 2013. https://www.euro.who.int/__data/assets/pdf_file/0004/193108/REVIHAAP-Final-technical-report-final-version.pdf (accessed March 1, 2021).
  8. World Health Organization, Regional Office for Europe. Air quality guidelines: global update 2005: particulate matter, ozone, nitrogen dioxide and sulfur dioxide; World Health Organization. Regional Office for Europe, 2006. https://apps.who.int/iris/handle/10665/107823 (accessed March 1, 2021).
  9. World Health Organization. Ambient (outdoor) air quality and health, Fact sheet Nr. 313, Updated March 2014; World Health Organization, 2014. http://www.who.int/mediacentre/factsheets/fs313/en/ (accessed March 1, 2021).
  10. Paciorek C. J.; Yanosky J. D.; Puett R. C.; Laden F.; Suh H. H. Practical large-scale spatio-temporal modeling of particulate matter concentrations. Ann. Appl. Stat. 2009, 3, 370–397. 10.1214/08-AOAS204. [DOI] [Google Scholar]
  11. Yanosky J. D.; Paciorek C. J.; Suh H. H. Predicting chronic fine and coarse particulate exposures using spatiotemporal models for the Northeastern and Midwestern United States. Environ. Health Perspect. 2009, 117, 522–529. 10.1289/ehp.11692. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Yanosky J. D.; Paciorek C. J.; Laden F.; Hart J. E.; Puett R. C.; Liao D.; Suh H. H. Spatio-temporal modeling of particulate air pollution in the conterminous United States using geographic and meteorological predictors. Environ. Health 2014, 13, 63 10.1186/1476-069X-13-63. [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Beelen R.; Hoek G.; Pebesma E.; Vienneau D.; de Hoogh K.; Briggs D. J. Mapping of background air pollution at a fine spatial scale across the European Union. Sci. Total Environ. 2009, 407, 1852–1867. 10.1016/j.scitotenv.2008.11.048. [DOI] [PubMed] [Google Scholar]
  14. Vienneau D.; de Hoogh K.; Bechle M. J.; Beelen R.; van Donkelaar A.; Martin R. V.; Millet D. B.; Hoek G.; Marshall J. D. Western European land use regression incorporating satellite-and ground-based measurements of NO2 and PM10. Environ. Sci. Technol. 2013, 47, 13555–13564. 10.1021/es403089q. [DOI] [PubMed] [Google Scholar]
  15. Jerrett M.; Burnett R. T.; Ma R.; Pope C. A. III; Krewski D.; Newbold K. B.; Thurston G.; Shi Y.; Finkelstein N.; Calle E. E.; Thun M. J. Spatial analysis of air pollution and mortality in Los Angeles. Epidemiology 2005, 16, 727–736. 10.1097/01.ede.0000181630.15826.7d. [DOI] [PubMed] [Google Scholar]
  16. Liao D.; Peuquet D. J.; Duan Y.; Whitsel E. A.; Dou J.; Smith R. L.; Lin H. M.; Chen J. C.; Heiss G. GIS approaches for the estimation of residential-level ambient PM concentrations. Environ Health Perspect. 2006, 114, 1374–1380. 10.1289/ehp.9169. [DOI] [PMC free article] [PubMed] [Google Scholar]
  17. Diez Roux A. V.; Auchincloss A. H.; Franklin T. G.; Raghunathan T.; Barr R. G.; Kaufman J.; Astor B.; Keeler J. Long-term exposure to ambient particulate matter and prevalence of subclinical atherosclerosis in the Multi-Ethnic Study of Atherosclerosis. Am. J. Epidemiol. 2007, 167, 667–675. 10.1093/aje/kwm359. [DOI] [PubMed] [Google Scholar]
  18. Sampson P. D.; Richards M.; Szpiro A. A.; Bergen S.; Sheppard L.; Larson T. V.; Kaufman J. D. A regionalized national universal kriging model using Partial Least Squares regression for estimating annual PM2.5 concentrations in epidemiology. Atmos. Environ. 2013, 75, 383–392. 10.1016/j.atmosenv.2013.04.015. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Beloconi A.; Kamarianakis Y.; Chrysoulakis N. Estimating urban PM10 and PM2.5 concentrations, based on synergistic MERIS/AATSR aerosol observations, land cover and morphology data. Remote Sens. Environ. 2016, 172, 148–164. 10.1016/j.rse.2015.10.017. [DOI] [Google Scholar]
  20. Szpiro A. A.; Sampson P. D.; Sheppard L.; Lumley T.; Adar S. D.; Kaufman J. Predicting Intra-Urban Variation in Air Pollution Concentrations with Complex Spatio-Temporal Dependencies. Environmetrics 2009, 21, 606–631. 10.1002/env.1014. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Benas N.; Beloconi A.; Chrysoulakis N. Estimation of urban PM10 concentration, based on MODIS and MERIS/AATSR synergistic observations. Atmos. Environ. 2013, 79, 448–454. 10.1016/j.atmosenv.2013.07.012. [DOI] [Google Scholar]
  22. Emili E.; Popp C.; Petitta M.; Riffler M.; Wunderle S.; Zebisch M. PM10 remote sensing from geostationary SEVIRI and polar-orbiting MODIS sensors over the complex terrain of the European Alpine region. Remote Sens. Environ. 2010, 114, 2485–2499. 10.1016/j.rse.2010.05.024. [DOI] [Google Scholar]
  23. Paciorek C. J.; Liu Y. Limitations of remotely sensed aerosol as a spatial proxy for fine particulate matter. Environ. Health Perspec. 2009, 117, 904–909. 10.1289/ehp.0800360. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Beloconi A.; Chrysoulakis N.; Lyapustin A.; Utzinger J.; Vounatsou P. Bayesian geostatistical modelling of PM10 and PM2.5 surface level concentrations in Europe using high-resolution satellite-derived products. Environ. Int. 2018, 121, 57–70. 10.1016/j.envint.2018.08.041. [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Shaddick G.; Thomas M. L.; Amini H.; Broday D.; Cohen A.; Frostad J.; Green A.; Gumy S.; Liu Y.; Martin R. V.; Pruss-Ustun A.; Simpson D.; van Donkelaar A.; Brauer M. Data Integration for the Assessment of Population Exposure to Ambient Air Pollution for Global Burden of Disease Assessment. Environ. Sci. Technol. 2018, 52, 9069–9078. 10.1021/acs.est.8b02864. [DOI] [PubMed] [Google Scholar]
  26. van Donkelaar A.; Martin R. V.; Brauer M.; Kahn R. A.; Levy R. C.; Verduzco C.; Villeneuve P. J. Global estimates of ambient fine particulate matter concentrations from satellite-based aerosol optical depth: development and application. Environ. Health Perspect. 2010, 118, 847–855. 10.1289/ehp.0901623. [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Shaddick G.; Yan H.; Salway R.; Vienneau D.; Kounali D.; Briggs D. Large-scale Bayesian spatial modelling of air pollution for policy support. J. Appl. Stat. 2013, 40, 777–794. 10.1080/02664763.2012.754851. [DOI] [Google Scholar]
  28. Fioravanti G.; Martino S.; Cameletti M.; Cattani G. Spatio-temporal modelling of PM10 daily concentrations in Italy using the SPDE approach. Atmos. Environ. 2021, 248, 118192 10.1016/j.atmosenv.2021.118192. [DOI] [Google Scholar]
  29. European Environment Agency. Air Base V.8 - The European air quality database, 2014. https://www.eea.europa.eu/data-and-maps/data/airbase-the-european-air-quality-database-8 (accessed March 1, 2021).
  30. European Environment Agency. Air Quality e-Reporting - The European air quality database, 2020. https://www.eea.europa.eu/data-and-maps/data/aqereporting-8 (accessed March 1, 2021).
  31. Lyapustin A.; Wang Y.; Laszlo I.; Kahn R.; Korkin S.; Remer L.; Reid J. S.. Multiangle implementation of atmospheric correction (MAIAC): 2. Aerosol algorithm J. Geophys. Res.: Atmos. 2011, 116 (D3), 10.1029/2010JD014986. [DOI] [PMC free article] [PubMed]
  32. Gorelick N.; Hancher M.; Dixon M.; Ilyushchenko S.; Thau D.; Moore R. Google Earth Engine: Planetary-scale geospatial analysis for everyone. Remote Sens. Environ. 2017, 202, 18–27. 10.1016/j.rse.2017.06.031. [DOI] [Google Scholar]
  33. Copernicus Atmosphere Monitoring Service (CAMS). Regional air quality, 2021. http://www.regional.atmosphere.copernicus.eu (accessed March 1, 2021).
  34. Inness A.; Ades M.; Agustí-Panareda A.; Barré J.; Benedictow A.; Blechschmidt A.-M.; Dominguez J. J.; Engelen R.; Eskes H.; Flemming J.; Huijnen V.; Jones L.; Kipling Z.; Massart S.; Parrington M.; Peuch V.-H.; Razinger M.; Remy S.; Schulz M.; Suttie M. The CAMS reanalysis of atmospheric composition. Atmos. Chem. Phys. 2019, 19, 3515–3556. 10.5194/acp-19-3515-2019. [DOI] [Google Scholar]
  35. CIESIN (Center for International Earth Science Information Network) - Columbia University. Gridded Population of the World, Version 4 (GPWv4): Population Density Adjusted to Match 2015 Revision UN WPP Country Totals, Revision 11; NASA Socioeconomic Data and Applications Center (SEDAC): Palisades, NY, 2018. (accessed March 1, 2021).
  36. EuroStat–GISCO service. Eurogeographics for the administrative boundaries, 2016. https://ec.europa.eu/eurostat/web/gisco/geodata/reference-data/administrative-units-statistical-units (accessed March 1, 2021).
  37. Rue H.; Martino S.; Chopin N. Approximate Bayesian inference for latent Gaussian models by using integrated nested Laplace approximations. J. R. Stat. Soc., Ser. B Stat. Methodol. 2009, 71, 319–392. 10.1111/j.1467-9868.2008.00700.x. [DOI] [Google Scholar]
  38. Lindgren F.; Rue H.; Lindström J. An explicit link between Gaussian fields and Gaussian Markov random fields: The stochastic partial differential equation approach. J. R. Stat. Soc., Ser. B Stat. Methodol. 2011, 73, 423–498. 10.1111/j.1467-9868.2011.00777.x. [DOI] [Google Scholar]
  39. Blangiardo M.; Cameletti M.. Spatial and spatio-temporal Bayesian models with R - INLA; Wiley, 2015. [Google Scholar]
  40. Rue H.; Martino S.; Lindgren F.; Simpson D.; Riebler A.. R-INLA: Approximate Bayesian inference using integrated nested Laplace approximations; Trondheim, Norway, 2013. http://www.r-inla.org (accessed 1 March 2021).
  41. R Core Team, R: A Language and Environment for Statistical Computing; R Foundation: Vienna, Austria, http://www.R-project.org/, 2015. (accessed March 1, 2021).
  42. Ntzoufras I.Bayesian Modeling Using WinBugs; Wiley, 2008. [Google Scholar]
  43. Shaddick G.; Thomas M. L.; Green A.; Brauer M.; van Donkelaar A.; Burnett R.; Chang H. H.; Cohen A.; Dingenen R. V.; Dora C.; Gumy S.; Liu Y.; Martin R.; Waller L. A.; West J.; Zidek J. V.; Prüss-Ustün A. Data integration model for air quality: a hierarchical approach to the global estimation of exposures to ambient air pollution. J. R. Stat. Soc. Ser. C: Appl. Stat. 2018, 67, 231–253. 10.1111/rssc.12227. [DOI] [Google Scholar]
  44. World Health Organization. Ambient air pollution: a global assessment of exposure and burden of disease, 2016. https://apps.who.int/iris/handle/10665/250141 (accessed March 1, 2021).
  45. World Bank, Country Classification , 2016. https://datahelpdesk.worldbank.org/knowledgebase/topics/19280-country-classification (accessed March 1, 2021).
  46. World Health Organization. Exposure to ambient air pollution from particulate matter for 2016, 2018. https://www.who.int/airpollution/data/AAP_exposure_Apr2018_final.pdf (accessed March 1, 2021).
  47. Stafoggia M.; Bellander T.; Bucci S.; Davoli M.; de Hoogh K.; de’ Donato F.; Gariazzo C.; Lyapustin A.; Michelozzi P.; Renzi M.; Scortichini M.; Shtein A.; Viegi G.; Kloog I.; Schwartz J. Estimation of daily PM10 and PM2.5 concentrations in Italy, 2013015, using a spatiotemporal land-use random-forest model. Environ. Int. 2019, 124, 170–179. 10.1016/j.envint.2019.01.016. [DOI] [PubMed] [Google Scholar]
  48. Shtein A.; Kloog I.; Schwartz J.; Silibello C.; Michelozzi P.; Gariazzo C.; Viegi G.; Forastiere F.; Karnieli A.; Just A. C.; Stafoggia M. Estimating Daily PM2.5 and PM10 over Italy Using an Ensemble Model. Environ. Sci. Technol. 2020, 54, 120–128. 10.1021/acs.est.9b04279. [DOI] [PubMed] [Google Scholar]
  49. Muff S.; Riebler A.; Held L.; Rue H.; Saner P. Bayesian analysis of measurement error models using integrated nested Laplace approximations. J. R. Stat. Soc. Ser. C: Appl. Stat. 2015, 231–252. 10.1111/rssc.12069. [DOI] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

es1c03748_si_001.pdf (68.6MB, pdf)

Articles from Environmental Science & Technology are provided here courtesy of American Chemical Society

RESOURCES