Numerical Simulations and Assessment of the Effect of Low-Emission Zones in Sofia, Bulgaria

Abstract

Bulgaria continues to face serious challenges related to air quality. To mitigate traffic-related air pollution and in line with the European regulations, the Metropolitan Municipal Council of Sofia has adopted and introduced low-emission zones (LEZs) in the city centre. The goal of this study is to address the specific needs of urban planning in the city in support of local decision-making. A bespoke emission inventory was developed for the LEZs in Sofia, and high-resolution numerical simulations (100 m resolution) were carried out to assess the effect of the measures implemented to reduce emissions in the central part of the city. The results show a decrease in nitrogen dioxide concentrations along major roads and intersections, but projected concentrations will still be high. No significant improvement is expected for particulate matter pollution due to the limitations of this study. High-resolution (100 m) emission inventories of domestic heating, minor roads, and bare soil surfaces, the major sources of particulate matter pollution, are not included in this study. An integrated model is needed to analyse and compare different scenarios for the development of the transport system, and the gradual introduction of LEZs must be accompanied by a number of other additional measures and actions.

1. Introduction

Globalisation has long positioned cities as drivers of societal progress, but city living comes with major trade-offs between opportunities for prosperity and access to services, on one hand, and exposure to environmental stressors such as population density and traffic congestion, on the other [1]. Urbanised areas are the largest source of greenhouse gases and air pollutants as the concentration of human activities in the cities leads to emissions that modify the thermal conditions and chemical composition of the urban atmosphere [2]. Finding a balance between these benefits and harms is critical for sustainable and healthy urban development and calls for multisectoral collaboration and research-informed urban policies.

Air pollution has ranked second among the leading risk factors for premature mortality on a global scale (8.1 million cases) in 2021, and it is also one of the main contributors to the global burden of disease [3]. Although many countries in Southeast Europe have made air quality improvements over the last decade, exposures have remained higher than those seen broadly across Europe [4]. In Bulgaria, the country with the highest air pollution burden in the European Union (EU) [5], fine particulate matter with an aerodynamic diameter of 2.5 micrometres or smaller (PM_2.5) is associated with at least 9000 premature deaths every year [4]. A recent health impact assessment study estimated that in the capital city of Sofia alone, exposure to PM_2.5 and nitrogen dioxide (NO₂) above the World Health Organization (WHO) air quality guidelines leads to over 2000 annual deaths or 16% of the annual mortality, and to many more non-fatal cases of respiratory and cardiometabolic diseases [6]. However, PM_2.5 is monitored at only eight government stations in Bulgaria, and most annual exceedances are registered for the more widely monitored particulate matter with an aerodynamic diameter of 10 micrometres or smaller (PM₁₀) [7]. With 48 stationary monitoring stations in total, located in 34 settlements, the spatial coverage in the country is limited.

Like other sources of pollution, traffic, specifically the high-temperature combustion process, leads to the formation of a wide range of pollutants, such as nitrogen oxides (NO_x), carbon monoxide (CO), particulate matter (PM), benzo-α-pyrene and others [8]. NO₂ often serves as a proxy for traffic-related air pollution (TRAP), which is a major health risk factor [9]. However, it is also insufficiently covered by the monitoring network, even though the number of cars in Bulgaria has increased significantly in recent years [10] and has intensified the TRAP-related impact of NO_x and PM on urban smog and human health. Furthermore, it is a well-known fact that NOx is a major precursor to ground-level ozone (O₃), which has a highly harmful acid effect on both human health and vegetation. In 2022, transport emitted 38% of the total amount of NO_x in Bulgaria [7]. Sofia is the largest and the most developed complex transport hub in the country. The average age of cars in the city is 16 years, with almost 70% of all vehicles having a standard of performance below Euro 4, and more than half (56.5%) with no Euro category at all [11]. This outlines the need for local measures, tailored to the specific exposure situation, to address air pollution. The limited density of gold-standard air pollution monitoring stations, and the coarse resolution of existing numerical models for predicting intra-city variability in air pollutants are seen as major setbacks for local research on air pollution and health impact. This further hinders the development of evidence-based recommendations for urban design and transport planning [12].

In an attempt to mitigate the TRAP-related burden, low-emission zones (LEZs) have been established in many European cities. LEZs are designated urban areas where restrictions are imposed on vehicles that emit high levels of harmful pollutants. These zones are typically implemented in city centres or areas with high traffic congestion and poor air quality [13,14,15,16,17,18]. They work by restricting access for vehicles that do not meet specific emission standards, such as those based on the European emission standards. And yet, the establishment of LEZ is not straightforward and local idiosyncrasies of the area, including urban form, alternative transport modes, and societal acceptance, should be taken into consideration [19,20,21,22,23,24]. While many studies show marked reductions in NO₂ and PM levels following the introduction of LEZs [25,26], others have noted increased pressure on adjacent areas which receive extra traffic [25,27,28], and general dissatisfaction of residents whose daily mobility may be compromised in the absence of viable alternatives.

Air quality modelling can be an indispensable tool to better understand the local situation, explore counterfactual scenarios, and identify the most acceptable approach to establishing LEZs for different stakeholders and population groups. This is especially true in the absence of suitable monitoring data, as is the case of Sofia. In accordance with European regulations and a national air quality plan, the Metropolitan Municipal Council of Sofia has adopted LEZs on the territory of the city [29]. LEZs were provisionally introduced in the previous air quality program for the Sofia metropolitan municipality around 2016 [30,31]. Early preparation and parallel studies by the local government and non-governmental organisations started in the next few years after that envisaged introduction, but the plans were postponed due to COVID-19 and the local elections in the late 2023. The National Program for Improving Ambient Air Quality (2018–2024) [32] and the actual Comprehensive program for improving the quality of atmospheric air of Sofia Municipality for the period of 2021–2026 [33] made a transport-oriented LEZ for Sofia a more explicit intervention and it was officially introduced in December 2023, gradually taking effect with improvements in enforcement.

This work addresses the specific needs of urban planning in cities in support of local decision-making. There is a lack of observations and integrated modelling studies combining traffic emissions and meteorology at high resolution in Bulgarian cities. This study aimed to develop a bespoke emission inventory for the LEZs of Sofia and conduct high resolution numerical simulations for the Sofia municipality area. It builds on prior numerical experiments conducted for the same area [34,35,36]. This study also aimed to assess the effect of implemented emission reduction measures in the central part of the city of Sofia.

2. Materials and Methods

2.1. Study Area

This study focuses on the city of Sofia, a challenging and complex urban system due to its geographical location and contrasting modes of urban development, which currently combine in a rich variety of patterns. The municipality of Sofia is located in the Sofia Valley, reaching the foothills of the Stara Planina and lower parts of the Vitosha, Plana, Lozen and Rila mountains. The area includes 38 settlements, and the complex terrain and meteorological conditions (a large number of inversions, especially in winter, which prevent turbulent mixing and dispersion) are favourable to high-pollution events in the area [30,31,33].

The morphology and diversity of the city’s forms create certain conditions for the concentration and absorption of diverse doses of pollution in different parts of the city, with corresponding profiles of street–neighbourhood structures and functional differences in the activities carried out during certain parts of the day, week, and seasons.

The location of the study area is shown in Figure 1.

2.2. Models Used in This Study

Several models were used in this study, applying a new methodology for linking regional and local models.

The Urban Air Quality Management System (ADMS-Urban v. 5.1) is local-scale comprehensive software, established and supported by Cambridge Environmental Research Consultants (CERC) Ltd. (Cambridge, UK), suitable for modelling air quality in urban areas [38,39]. More details regarding the model are available in [40]. The Comprehensive Emissions Inventory Toolkit—EMIT [41] is used for manipulating and assessing emission data from a variety of sources considered explicitly (major roads, railways, and industrial sources), and average emissions on a regular grid for minor roads, commercial, and domestic sources. EMIT holds the emission data from all sources and provides it as input data for air dispersion modelling software such as ADMS-Urban. These models have been successfully applied to study air quality in Sofia in previous studies [34,35].

The regional-scale Community Multiscale Air Quality Modelling System (CMAQ v. 5.0.1) is a comprehensive computer software package developed by the U.S. Environmental Protection Agency (EPA, Washington, DC, USA) that addresses air pollution, taking into account transport, deposition and chemical transformations [42]. The Sparse Matrix Operator Kernel Emissions (SMOKE v. 4.5) system [43] is applied for emission processing, designed to produce hourly values disaggregated by pollutants in a selected calculation grid to be used as input to CMAQ. It should be noted that only individual SMOKE modules are used in this study due to the difference between the Selected Nomenclature for air pollution categories in the USA and the EU, which necessitated the development of our own emission processing codes [44].

The Weather Research and Forecasting model (WRF v. 3.9.1) is used for modelling thermo-hydrodynamics in the atmosphere and provides necessary meteorological conditions for regional air quality modelling [45]. The model was developed as a collaborative partnership of several government institutions and features a wide range of applications on scales ranging from tens of metres to thousands of kilometres.

The same coupled regional models (WRF/SMOKE/CMAQ) were used in numerous research studies for the region of Bulgaria and the city of Sofia [46,47]. The Bulgarian Chemical Weather Forecast System (BCWFS), based on the three models described above, has been running operationally since 2012 [44,48].

The Multi-Model Air Quality System (MAQS v. 1.2), developed and maintained by CERC, is an automated system for coupling the high-resolution ADMS-Urban with a regional air quality model with an hourly concentration output such as CMAQ [49]. Meteorological data from the WRF meso-scale model is used for both regional and the local modelling. MAQS was recently implemented for the Sofia metropolitan area [36].

2.3. Data Sources for Validation

Validation is essential because it ensures the accuracy and reliability of numerical models. Validated models enhance confidence in predictions and decision-making based on their outputs. Two types of data were used for comparison with the modelling results: in situ data from observation sites and data obtained from the EXPANSE (EXposome Powered tools for healthy living in urbAN SEttings) project [50].

In situ data was obtained from the National Automated System for Environmental Monitoring (NASEM), operated by the Executive Environment Agency, Ministry of Environment and Water (Sofia, Bulgaria) [7]. Five air quality stations (AQSs) are located in the city of Sofia (Figure 2b)—three are urban background stations, Druzhba (42.67, 23.40), Nadezhda (42.73, 23.31), and Hipodruma (42.68, 23.30), and two are traffic-oriented stations—Pavlovo (42.67, 23.27), Mladost (42.66, 23.38). These stations provide hourly air quality and meteorological data. More stations measure the main pollutants NO₂, PM₁₀ and O₃, but only one (AQS Hipodruma) collects data for PM₂.₅. The data provided by the Sofia Municipality from the mobile air quality station delivers additional information on pollution and meteorological parameters in the central part of the city (latitude 42.69, longitude 23.335). The dataset covers the period from December 2021 to April 2025 and is used to compare air pollution levels before and during the implementation of the LEZ.

EXPANSE mapped the mean annual concentrations of NO₂, PM_2.5, PM₁₀, and O₃ Europe-wide for each year from 2000 to 2019 [51]. The maps are available in raster format with an original resolution of 25 × 25 m and aggregated to 100 × 100 m or 1 × 1 km. Unlike our dispersion modelling approach, EXPANSE used a stochastic approach, developing land use regression models (LUR) that provided highly granular exposure surfaces suitable for environmental health research [6]. These LUR models were developed using a geographically weighted regression of measured air pollutants at monitoring stations across Europe against a predictor set that included GIS indicators of land use and road infrastructure, and measured pollutant levels from spectral analysis of satellite imagery and chemical transport model data. Five-fold cross-validation of the models for individual pollutants derived R² values of 0.66 for NO₂, 0.77 for PM_2.5, 0.62 for PM₁₀, and 0.58 for O₃ [51].

A standardised protocol for validating air quality prediction applications was previously proposed as a common framework for comparative analysis for model developers and users, supporting policy development in accordance with European directives on ambient air quality, which is a major achievement [52]. But due to the limited number of AQSs in Sofia (five for PM₁₀ and NO₂, one for PM_2.5), this type of in situ validation is not very accurate in providing information on models’ capabilities and potential to predict pollution. Satellite data can be used as another option for identifying areas with high concentrations of pollutants, with all the limitations of data availability depending on the temporal resolution associated with the orbit that satellites follow and on the satellite technology and extraction algorithm (spatial resolution can vary from 1 km to ~100 km). The best validation method is to use a reanalysis that combines model data and assimilated fields of air pollutant concentrations based on current observation data. The regional air quality product of Copernicus Atmosphere Monitoring Service (CAMS), based on an ensemble of 11 state-of-the-art numerical air quality models developed in Europe, has often been used in different studies, but the spatial resolution is approximately 10 km for the region of Bulgaria. Data from the EXPANSE project was selected for model validation in this study due to its high resolution of 25 m. The data was aggregated to a 1 km cell to correspond to the regional model grid.

2.4. Study Settings

High-resolution dispersion modelling (with 100 m grid) was performed for the urban area of Sofia municipality using the ADMS-Urban system. The MAQS system was employed to provide the necessary input conditions for local-scale modelling. The results of offline regional meteorological (WRF), emission (SMOKE) and dispersion (CMAQ) modelling were used for meteorological conditions, emission sources not explicitly included at a higher resolution, and background concentrations of various pollutants. MAQS combines both regional- and local-scale dispersion and chemical modelling without double counting local emissions, downscaling meteorological and background concentrations calculated with a resolution of 1 km into local grid cells (100 m). This modelling system was installed and is running on the Nestum cluster [53] at the HPC laboratory, Sofia Tech Park.

A system of three nested domains for regional modelling was used with spatial resolutions of 9, 3, and 1 km (Figure 2a). Parent domain 1 covers the Balkan Peninsula, domain 2 covers Bulgaria, and domain 3 covers the Sofia municipality area. These domains represent 51 vertical levels above ground, with 26 levels below 1 km altitude. The anthropogenic emissions for Bulgaria, used in the CMAQ simulations, were derived from the National Atmospheric Emissions inventory for the year 2019 [10]. For all countries neighbouring Bulgaria in South-Eastern continental Europe, the 2019 emission inventory was taken from the Copernicus Atmosphere Monitoring Service [54] product developed by Netherlands Organization for Applied Scientific Research (TNO) [55]. The annual totals of the two inventories are distributed over time according to monthly, weekly, and hourly variations using variable time profiles [56]. The vertical distribution of anthropogenic emissions follows CAMS-TNO profiles. Biogenic emissions were obtained with the SMOKE system, and the same system was used to combine different types of emission sources. The ADMS-Urban domain is nested in domain 3 of the CMAQ model (Figure 2b).

We developed a local road transport emission inventory for Sofia municipality, covering major streets and national roads with heavy traffic, for the selected baseline year 2018 and for LEZ scenarios for 2022 and 2026. The methodology applied for the baseline 2018 simulation, which includes traffic data preparation and emission calculation using EMIT, and the comparison between results from the CMAQ and MAQS systems are described in detail in [36]. The newly implemented MAQS system provides a more reliable approach and procedure for studying urban-scale pollution in the city of Sofia.

In this study, two numerical experiments were conducted using the same meteorology and background concentrations, but with new emission inventories corresponding to the small ring LEZ and the official signals for its introduction with the ordinance approved by Sofia municipality (scenario of 2022) and the large ring LEZ, which enters into effect in late 2025 (scenario of 2026). The differences between the results related to the LEZ and the baseline 2018 allow for an assessment of the outcome of the implementation of the transport-related measures in the local air quality program with its time horizon until 2026. Only emissions from major roads are defined explicitly; all other sources are not explicitly included, but their contribution is taken into account through the emissions implemented in the CMAQ model.

2.5. Development of New Local Emission Inventory for LEZs

This study developed a detailed inventory of traffic emissions for Sofia along the major streets and roads, using fusion of diverse datasets based on previous experiments [35,36] with new adjustments. Emissions were calculated using the EMIT tool. The overall pathway to the inventory included several separate steps for the pre-processing and processing of the EMIT input dataset. The output from EMIT was then used as the local emission inventory input for ADMS-Urban dispersion modelling (Figure 3).

The traffic volume was derived from an array of traffic-related variables and an ensemble of machine learning models [57]. Due to the challenges posed by fragmented and inconsistent traffic volume data, the approach succeeded in synthesising input data from various national, municipal, and proprietary datasets. This includes automatic, manual and hybrid-controlled counts from various campaigns in 2017 and 2018 at junctions, as well as in 2021 and 2022 at sections of secondary roads, together with secondary traffic models, street and road characteristics such as capacity, typology, and other parameters, among them the integration and choice spatial syntax parameters, population density and point-of-interest data, as well as noise and NO₂ measurements, etc. The datasets were harmonised through preprocessing based on QGIS (https://qgis.org/) and Python (https://www.python.org/). Ensemble regression models such as Random Forest, XGBoost, and CatBoost were employed and stacked to infer information about traffic activity across the entire street network.

The resulting dataset for average annual daily traffic (AADT) was calibrated for 2018. The vehicle fleet was categorised into motorcycles, light, and heavy-duty vehicles, further disaggregated by weight specification intervals, engine volume, Euro standards, fuel types, and propulsion technologies, as well as conditions of the pollutant removal devices (catalyst converters and particle traps).

The vehicle fleet categorisation was based on comparison and analytical hierarchical process of crossing and calculation of the shares of the fleet subcategories (n = 546) in MS Excel based on the separately available unlinked datasets with shares of fuel types, propulsion technologies, or Euro standards. Weight categories, especially in the heavy vehicles’ domain, reflect the zonal restrictions (Centre zone, Zone 1, and the rest of the city).

The street segment share of heavy vehicles, light vehicles and motorcycles was calculated on the basis of separately processed IDW interpolation among the different road and street types owing to the point data obtained from the Strategic noise map and Noise action plan counts in Sofia prepared under the Environmental Noise Directive regulations for the agglomeration of the city.

Then, the emission factors from the EMEP/EEA air pollutant emission inventory guidebook 2019 were chosen. Although it is well understood that exhaust emissions, tyre, brake and road wear, and resuspension are usually higher on Bulgarian streets and in Sofia compared to the EU average and due to poor conditions of the fleet, the infrastructure and other urban surfaces, the team chose a more conservative approach due to the high level of uncertainties and knowledge gaps. It is important to note that the heavy vehicles are not affected by the regulations, but, at the same time, the heavier ones are restricted in the relevant zones which strongly overlap with the small and large rings.

The street and road gradients were calculated on the basis of a detailed municipal digital elevation model in the QGIS 3.10.10 environment. The overall canyon height was summarised from calculations on the basis of the CERC Street Canyon Tool, part of the ADMS-Urban model, in the ArcGIS 9.3 environment. At a later stage, the more detailed urban street canyon parameters calculated through this tool were added as part of the ADMS-Urban input.

The emission inventory formed the basis for the baseline and several future scenarios in Sofia. These scenarios, modelled using EMIT, differ in terms of time frame (2018–2022–2026) and inclusion in LEZ zones (small ring, large ring, rest of the city), taking into account the provisions for heavy-traffic access to the city within the three other zones different from the LEZ (Centre zone, Zone 1, and the rest of the city). Overall, one baseline with three distributions of heavy goods’ fleet traffic, as well as two scenarios differentiated into sub-zones according to the scope of the LEZ (the whole city for 2022, as well as the small and large rings and the rest of the city for 2026) and the corresponding access regulations for heavy goods’ traffic, which stay in place, were subsequently used as combined annual input data to simulate pollutant concentrations (PM₁₀, PM_2.5, NO_x, NO₂) for the years 2018, 2022, and 2026.

The major assumptions made were in the following directions:

• A shift from pre-Euro, Euro 1 and Euro 2 categories to the Euro 3, 4 and 5 with relatively small growth of the Euro 6 and zero-emission vehicles, which usually, as newly purchased vehicles, have a small proportion in relation to the second-hand ones that dominate the scene, even though Bulgaria and Sofia have experienced some of the highest growth rates of these but from a very low position;
• The gradual enlargement of the restrictions for entry from the small to the large ring with the expectation for the first in 2022 and the already introduced larger ring in 2026 is assumed to influence the decisions of more and more owners and users of the lower-performance fleet with more significant impact for the fleet within the small and large rings while considering that there are exemptions for the local residents and disabled persons, which have a relatively high share in the city, mostly due to ageing and poor physical health, so that the changes are more moderate;
• The diminishing share of failed catalysts and particle traps reflecting the poor condition of the fleet at the baseline 2018 year in comparison with the one in the EU and the UK, as well as the gradual tightening of national technical regulations and control from 2022 and especially towards the 2026 scenario.

A more detailed description of the emission factors used, data sources and two LEZs established—a small ring and a large ring—is presented in Appendix A and in the Supplementary Materials.

3. Results

3.1. Validation of the Models

The results obtained from MAQS and CMAQ were validated against the official NASEM data for the base year 2018. Standard statistical metrics (including mean bias, mean absolute error, root mean square error, Pearson’s correlation coefficient and index of agreement), along with the maximum values calculated from both the models and the measurements, are presented in

Table A1. MAQS validation for NO₂, PM₁₀ and PM_2.5 at the air quality stations (AQSs) for the base year 2018.

NO2
AQS	MB, µg/m3	ME, µg/m3	RMSE, µg/m3	IA	r	Missing data, %	Max obs/mod, μg/m3
Druzhba	−16.95	17.10	23.28	0.55	0.61	1.97	131.68/101.82
Hipodruma	−22.19	22.73	28.29	0.58	0.61	1.19	155.08/127.84
Mladost	−5.76	10.45	22.94	0.74	0.41	26.37	160.96/192.18
Nadezhda	−15.26	16.44	22.36	0.57	0.50	11.84	122.40/107.71
Pavlovo	−20.78	21.32	29.74	0.58	0.61	14.44	176.14/122.46
PM10
AQS	MB, µg/m3	ME, µg/m3	RMSE, µg/m3	IA	r	Missing data, %	Max obs/mod, μg/m3
Druzhba	−16.92	17.37	23.54	0.55	0.56	4.20	180.62/138.03
Hipodruma	−29.03	29.25	44.93	0.49	0.64	1.24	386.76/132.11
Mladost	−24.84	25.02	37.21	0.49	0.61	4.67	301.36/129.62
Nadezhda	−33.86	34.01	45.82	0.46	0.57	1.58	326.91/128.38
Pavlovo	−30.75	30.87	43.63	0.49	0.71	9.09	331.79/117.32
PM2.5
AQS	MB, µg/m3	ME, µg/m3	RMSE, µg/m3	IA	r	Missing data, %	Max obs/mod, μg/m3
Hipodruma	−18.73	19.69	37.72	0.50	0.55	31.35	253.04/126.92

and

Table A2. CMAQ validation for NO₂, PM₁₀ and PM_2.5 at the air quality stations (AQSs) for the base year 2018.

NO2
AQS	MB, µg/m3	ME, µg/m3	RMSE, µg/m3	IA	r	Missing data, %	Max obs/mod, μg/m3
Druzhba	−20.27	20.41	27.22	0.46	0.38	1.97	131.68/47.39
Hipodruma	−24.80	25.66	32.40	0.48	0.34	1.19	155.08/84.79
Mladost	−16.64	17.73	26.98	0.43	0.25	26.37	160.96/49.24
Nadezhda	−19.52	19.85	26.30	0.47	0.38	11.84	122.40/50.55
Pavlovo	−26.01	26.33	35.76	0.47	0.39	14.44	176.14/62.91
PM10
AQS	MB, µg/m3	ME, µg/m3	RMSE, µg/m3	IA	r	Missing data, %	Max obs/mod, μg/m3
Druzhba	−17.45	18.02	24.46	0.51	0.49	4.20	180.62/142.66
Hipodruma	−28.92	29.27	45.50	0.47	0.58	1.24	386.76/124.87
Mladost	−26.05	26.34	37.53	0.48	0.58	4.67	301.36/127.21
Nadezhda	−33.60	33.80	46.02	0.45	0.51	1.58	326.91/128.28
Pavlovo	−30.45	30.76	44.15	0.49	0.61	9.09	331.79/122.35
PM2.5
AQS	MB, µg/m3	ME, µg/m3	RMSE, µg/m3	IA	r	Missing data, %	Max obs/mod, μg/m3
Hipodruma	−18.74	19.92	38.56	0.48	0.49	31.35	253.04/118.42

(Appendix B). The statistics show that the modelled concentrations are underestimated at all locations, more for PM₁₀ than for NO₂. This is expected, as other sources, such as the minor road network and domestic heating, which have a significant contribution [34], are not explicitly included in this study. The concentrations and emissions produced by all additional sources are calculated with the CMAQ model, and it is well known that the regional model underestimates pollutant concentrations due to the 1 km grid resolution limitations.

The maps showing the calculated deviations between the model results (CMAQ and MAQS) and the EXPANSE data visualise the spatial distribution and show an underestimation of concentrations (negative values) in some problematic grid cells (Figure 4). The explicitly implemented new local transport emission inventory, covering major roads with heavy traffic in MAQS, shows a reduction in deviations. Areas with calculated lower NO₂ concentrations from the model are industrial zones in the northern part of the city, where the Sofia Thermal Power Plant is located, and in the southeastern part, where the East Thermal Power Plant is situated. The results indicate an underestimation of the contribution of point sources in the CMAQ regional model. The largest deviations from the EXPENSE data are observed in the outskirts of the city, where domestic heating sources contribute 100% of the total PM_2.5 concentration [34]. There is an insignificant reduction in deviations from the EXPENSES data in the city in terms of PM₁₀ concentration. The largest discrepancies are in a few areas where additional local emissions may not have been correctly assessed—the industrial zone between the villages of Kazichene and Elin Pelin, open bare soil surfaces near Sofia Technology Park, industrial zones close to Poduene, and industrial zones in the northern part of the city in Nadezhda, Orlandovtsi, and Vrabnitsa.

Both models yield lower values than the EXPANSE model. Despite some discrepancies with the EXPENSE data, the use of MAQS shows a significant improvement compared to the regional CMAQ model. Maps showing the differences in pollutant concentration fields between the two models (MAQS–CMAQ) are presented in Appendix C for the cold (months with domestic heating included) and warm periods (without domestic heating sources) of the baseline scenario. These maps show a significant increase in the concentration of NO₂ over the main roads—up to 45 μg m⁻³ and approximately 10 μg m⁻³ over the entire city during the cold period. The increase in concentration of PM₁₀ and PM_2.5 is much smaller, reaching a maximum of about 8 μg m⁻³ and 5 μg m⁻³, respectively, at intersections with heavy traffic, and the differences are more pronounced during the cold period, when inversion conditions are typically observed.

The coefficient of determination calculated between MAQS results and EXPANCE data for individual pollutants derived R² values of 0.60 for NO₂, 0.72 for PM₁₀, and 0.40 for PM_2.5. It should be noted that EXPANSE data is also modelled based on land use regression models and is used as only a proxy method for assessing the performance of our deterministic model. Furthermore, only the emission inventory for transport from major streets and national roads with heavy traffic is included explicitly, while emissions from other sources such as domestic heating are not modelled with such high resolution. This is likely the main reason for the low correlation between both datasets for PM_2.5.

3.2. Emission Modelling

Two numerical experiments were conducted to model small and large LEZ rings corresponding to the developed emission inventories for 2022 and 2026 described in Section 2.5. Detailed data on the distribution of the total number of vehicles by various categories (Pre-Euro 1, Euro 1–6, and zero-emission vehicles) and their sub-categories is presented in the Supplementary Materials (Scenarios_en_applied.xlsx).

The 2022 scenario affects the small ring and the entire part of the city, following a change in categories based on trends in the replacement of old vehicles, and differs slightly from the emission inventory for the base year 2018. The 2026 scenario applies different weights for the small ring, large ring, and the rest of the city, with strict restrictions for the pre-Euro 1, Euro 1, and Euro 2 categories in the central part of the city and slight changes in other categories based on trends in the replacement of old vehicles. This scenario corresponds to the expected measures introduced by the Sofia municipality, which will come into force in the winter of 2027. Maps with calculated emissions are shown in Figure 4, Figure 5, Figure 6 and Figure 7.

3.3. Air Quality Modelling

The MAQS results from both experiments were compared with the concentrations calculated for the base year 2018 to assess the effect of the implemented measures. The same meteorological conditions, background concentrations and emissions from other sources were used, with only emissions from major road sources being changed. The differences between the concentration maximums for the base year, small ring LEZ and large ring LEZ are shown in

Table 1. Annual, cold and warm period mean maximum modelled (with MAQS) values in μg/m³.

NO2
Simulation	Annual mean	Cold period	Warm period
Base year 2018	99.67	110.01	92.29
Small ring 2022	75.73	84.02	69.81
Large ring 2026	65.18	73.11	59.51
PM2.5
Simulation	Annual mean	Cold period	Warm period
Base year 2018	14.63	19.31	11.29
Small ring 2022	13.98	18.70	10.61
Large ring 2026	13.87	18.57	10.52
PM10
Simulation	Annual mean	Cold period	Warm period
Base year 2018	23.14	29.58	18.54
Small ring 2022	22.34	28.63	17.85
Large ring 2026	22.25	28.52	17.77

and

Table 2. LEZ implementation period mean, and monthly mean maximum modelled (with MAQS) values in μg/m³.

NO2
Simulation	Mean	January	February	December
Base year 2018	111.42	128.87	93.70	111.68
Small ring 2022	85.26	99.31	70.80	85.68
Large ring 2026	74.36	87.19	60.98	74.93
PM2.5
Simulation	Mean	January	February	December
Base year 2018	21.01	24.99	15.48	22.56
Small ring 2022	20.06	23.87	14.69	21.62
Large ring 2026	19.94	23.72	14.59	21.50
PM10
Simulation	Mean	January	February	December
Base year 2018	31.50	37.37	24.04	33.08
Small ring 2022	30.54	36.23	23.24	32.15
Large ring 2026	30.42	36.09	23.13	32.05

.

Simulations were performed for the entire year, and Table 1 shows the differences in the maximum annual mean concentrations and separately for the cold (November–March) and for the warm periods (April–October). All modelled monthly mean maximum values are estimated at the intersection of Prof. Alexander Tanev street (Mladost residential area) and Okolovrasten Pat boulevard. The simulation results for the different scenarios indicate a reduction in the maximum annual average values of approximately 25% (2022) and 35% (2026) for NO₂, 4% (2022) and 5% (2026) for PM_2.5, 3% (2022) and 4% (2026) for PM₁₀.

Since the main air pollution problems in Sofia are recorded during the months of December to February, the LEZ implementation was first applied for these three months. Table 2 summarises the average monthly maximum values for the period of LEZ implementation. The same tendency can be observed: a decrease in the maximum concentration of NO₂ to less than the annual maximum value—23% (2022) and 33% (2026)—and similarly for PM_2.5—4% (2022) and 5% (2026)—and for PM₁₀—3% (2022) and 3.4% (2026). The highest concentrations were calculated for January.

Data from the mobile station, provided by the Sofia municipality, which operated for the period of December 2021–April 2025, was used to estimate the difference between the three months corresponding to the LEZ implementation. The average daily data is divided into the LEZ implementation period (1 December 2024– 28 February 2025) and the remaining available data covering the same period but for the previous years: 2021–2022; 2022–2023; 2023–2024.

Table 3. Average daily concentrations of the main pollutants in the city centre for the LEZ, established from 1 December 2024, to 28 February 2025, and the same months of previous years.

Pollutant Name	PM10	PM2.5	NO2	CO *	O3
Previous periods	37.49	26.61	44.55	4.25	68.17
LEZ	48.08	35.98	46.35	3.61	77.71

* Concentrations are in μg/m³, except for CO, which is given in mg/m³.

shows the calculated daily average concentration values of the various pollutants based on the measurement data.

A comparison of the two periods shows a 15% reduction in CO concentration. A slight increase (4%) was observed in NO₂ concentration, though there was a much greater increase in O₃ (14%). The PM_2.5 concentration increased by 35% and PM₁₀ by 32% based on on-site measurements in the city centre. 34 exceedances of the daily average threshold 50 μg/m³ for PM₁₀ were recorded in just the three months covering the implementation of the LEZ. Of course, it should be noted that the averaging period for the implementation of the LEZ is only 3 months, compared to 9 months covered for the same months in the previous three years. A much longer set of measurements over several years and further data analysis are needed to provide a reliable assessment of the effect of LEZ implementation in Sofia, which is planned for future work. The data in this work is used for additional assessment of the effect of the applied measures.

Maps with calculated average concentrations for the months of LEZ operation are shown in Figure 8, Figure 9 and Figure 10. The implementation of the LEZ leads to a significant reduction in NO₂ concentrations, especially in the 2026 scenario (Figure 8). The reduction is up to 30 μg/m³ on major boulevards with heavy traffic. Changes in PM_2.5 and PM₁₀ concentrations are not significant, as emissions from transport are lower than those from other sources (Figure 9 and Figure 10). The largest reductions in concentrations are on busy streets and at intersections where traffic is heavier.

4. Discussion

A bespoke emission inventory for the LEZs in Sofia was developed and high-resolution numerical simulations (with a 100 m resolution) were carried out to assess the effect of the measures implemented to reduce emissions in the central part of the city of Sofia. The implementation of LEZ reduces NO₂ concentration by approximately 25% (small ring, scenario 2022) and 35% (large ring, scenario 2026) on major roads and at intersections with heavy traffic, and this finding is consistent with other studies reporting reductions in NO₂ [25,26,58,59,60,61]. PM concentrations in Sofia have not changed since the introduction of the LEZs. Furthermore, measurements from a mobile station located in the city centre show an increase in PM concentration during the implementation period from 1 December 2024, to 28 February 2025. Many factors influence pollution concentrations through complex nonlinear relationships. Weather conditions and domestic heating, which provide background concentrations of PM_2.5, can vary significantly from year to year; massive construction works in the area add additional emissions. It should be noted that traffic patterns were significantly altered by changes in residents’ behaviour during the COVID-19-related restrictions, as well as in the subsequent years. It is more likely that a combination of all these factors is responsible for the estimated increase in concentrations of some pollutants during the implementation of the LEZ. Despite the reduction in NO₂ concentration for the large ring LEZ, the simulated maximum concentrations, averaged over the month, remain high, with values over 60 μ/m³ at specific locations in the city, close to busy boulevards and affected by mesoscale circulation characteristics leading to accumulation of pollution. A previous study, which conducted numerical modelling for a small area in the central part of the city, covering parts of G. S. Rakovski street with very high resolution (5 m), also highlights the need for additional traffic reduction to lower pollution, in addition to restricting the entry of the most polluting types of vehicles into the city centre [35].

A recent study [62] presented a comprehensive framework for analysing urban transport dynamics and travel behaviour related to private vehicles, taking into account various temporal and spatial dimensions in Sofia, Bulgaria. The published maps showing the distribution of the number of vehicles indicate high traffic density in central business districts and major arterial routes—Sofia Ring Road and Tsarigradsko Shose Blvd. The most congested traffic locations presented in this study correspond to the hot spots identified here by numerical modelling—Tsarigradsko shoes, Sofia Ring Road near the intersection with Bulgaria Blvd, Peyo Yavorov Blvd between the intersection with Tsarigradsko shose and Dragan Tsankov Blvd, the intersection between Sofia Ring Road and Bulgaria Blvd., Sofia Ring Road near the road connection with the Kazichene district, etc. On weekdays, the highest traffic volumes in Sofia occur between 17:00 and 18:00 at major intersections and boulevards, coinciding with peak periods of commuting from work and school, during which the main thoroughfares experience significant congestion.

Air pollution levels and related health impacts are significantly higher in Eastern Europe compared to Western Europe, creating significant inequalities [12]. This disparity is driven by factors such as economic differences, outdated infrastructure, and historical dependence on polluting industries. Research on South-Eastern Europe remains limited, and to our knowledge, no studies have investigated the impact of LEZs on traffic-related pollution in Bulgaria. Bulgaria continues to face serious challenges related to air quality. In 2022, it recorded some of the highest relative health impacts from PM_2.5 pollution among European Union (EU) countries [63]. In addition, in January 2024, the European Commission brought an action against Bulgaria before the Court of Justice of the EU for failure to transpose Directive (EU) 2019/1161 [64] on public procurement of clean vehicles, which risks further delays in the transition to low-emission transport. These developments highlight the urgent need for harmonised, evidence-based tools to support effective local policy actions.

Despite a significant reduction in concentrations at measuring stations over the past five years, PM₁₀ continues to be a major problem for air quality in Bulgaria. Annual average exceedances of PM₁₀ were recorded at 12 of 30 monitoring stations, and 41% of the population was exposed to PM₁₀ concentrations above the EU annual standard (40 μg/m³). Concentrations of fine PM_2.5 and polycyclic aromatic hydrocarbons (PAHs) are high in most municipalities where monitors are located. According to the latest report [7], the population was not exposed to PM_2.5 above the annual average exposure limit (20 μg/m³). The average exposure indicator (3-year running mean) for PM_2.5 for 2022 was 16.23 μg/m³. No issues with NO₂ pollution have been reported by the official AQSs, and the concentration has been sporadically elevated only in the city of Plovdiv.

In Bulgaria, TRAP has become a serious problem, mainly in larger cities, and the lack of observations makes monitoring and assessing the extent of NO_x pollution very difficult. Of all the AQSs that are part of NASEM, only five are urban transport oriented. Understanding the impact of transport on air quality in Sofia is based on data from two sites, neither of which is currently designed to monitor the air directly adjacent to areas with more intense and constant urban traffic. According to Bulgarian standards, traffic monitoring stations should be placed in close proximity (less than 10–15 m) to a single major road. However, in Sofia, for example, AQS Pavlovo is located 25 m away from the main road, and the AQS Mladost is 70 m away. Although official monitoring does not show a significant problem with NO₂ concentration, independent measurements with diffusion tubes conducted by non-profit civil society organisations show high monthly averaged values [65]. Experts from the Ecological Association “Za Zemiata” concluded after analysing the data that Sofia has “a problem with nitrogen dioxide concentrations above the limits, which due to the specific location of the official stations is not reported as such” [66]. A subsequent measurement campaign using the same method in 2023 and 2024 found that NO₂ levels remained relatively constant throughout the year during and after the end of the operation of the newly introduced LEZ in Sofia (1 December 2024–28 February 2025), remaining around and above 50 μg/m³, which potentially poses a risk to human health [67]. The latest available National Report on the State and Protection of the Environment in 2022 reports a 1.4% increase in gross domestic fuel and energy consumption in the country compared to 2021, and a 3.7% increase compared to the base year 2018 [7].

The trends in the development of Sofia over the last 30 years have been unfavourable in terms of air quality. The main reasons are the increasing density and proportion of impervious surfaces [57], the formation of canyons along city streets, and the discontinuity of so-called “green wedges,” which play an important role in the ventilation of the city. In addition, transport infrastructure construction has been intensified, which favours individual travel by car. Several studies [68,69] show statistically and experimentally that fragmented but compact development is associated with higher concentrations of NO₂ and PM₁₀ and that the neighbourhood and street-level characteristics are very important for pollutant exposure levels. More recent exposure studies, comparative and case-specific, show a significant health burden for Sofia, the major part of which is attributable to impacts from traffic, among them the PM_2.5 and NO₂ being the most significant [6,70]. However, there are many gaps in knowledge and information, particularly due to limited data collection on energy-related behaviour, traffic, air pollution measurements, and higher-resolution modelling and simulations for the Sofia municipality area.

The World Health Organization (WHO) has revised its Air Quality Guidelines (AQGs) since 2021 [71]. The updated guidelines are based on evidence from studies around the world and offer evidence-based public health recommendations and guidance on air quality. The old EU DIRECTIVE has been substantially amended by the new EU DIRECTIVE 2024/2881 of October 2024 [72]. The EU introduced stricter annual air quality standards and targets for key pollutants, which are to come into force by 2030. These new standards are more ambitious than the current ones, aiming to significantly reduce the impact of air pollution on human health. The daily average values have also been significantly updated, with a threshold not to be exceeded more than 18 times per calendar year.

The MAQS system used in this study is a new methodology implemented for the city of Sofia, which combines the complementary strengths of regional and local models, improving the prediction of pollutant concentrations. Regional (usually Eulerian) models contain complex chemistry mechanisms that operate over long spatial and temporal scales and can model the accumulation of concentrations in very-low-wind-speed conditions. Local (usually Gaussian-type plume) models can represent the detailed concentration gradients from explicitly defined sources, but generally only account for simplified chemical mechanisms and spatially homogeneous meteorological data, which limits their applicability to receptors distant from the source (e.g., more than 50 km). Coupling local and regional models allows for both the resolution of high concentration gradients near the source and the accurate representation of transport and chemistry on large spatial and temporal scales.

Although this work demonstrates the potential of numerical modelling to help decision makers to address pressing social problems in our society, it has some limitations. A significant obstacle to a comprehensive and precise analysis and validation of the models remains the insufficient availability of high-quality data from air-quality-monitoring stations and activity data needed to develop emission inventories in Bulgarian cities. Emission inventories from domestic heating, minor roads, and bare soil surfaces, the major sources of particulate matter pollution, are not included at a high resolution (100 m). The aim of our study was to assess the applicability of the proposed modelling pipeline as a tool to support the impact assessment of the LEZ in Sofia. It is essential to explore sophisticated and area-specific model inputs and specifications to ensure the accuracy of the modelling results. Given the limitations of monitoring data and its susceptibility to meteorological confounding, which often hinders the detection of subtle changes in air quality, emission modelling is recommended to assess the cost effectiveness of introducing a LEZ prior to implementation [59]. However, it was beyond the scope of this study to model environmental justice aspects of LEZs. Nevertheless, these issues are of high interest, as there are ongoing concerns in public discourse about the potential adverse social impacts of Sofia’s LEZ. Those concerns are exacerbated by the significant socioeconomic inequalities within Bulgarian society, pervasive distrust, and general scepticism toward authorities [23]. The implementation of LEZs is often accompanied by perceived drawbacks and controversial arguments, which raise debate over the complex environmental, financial, and social trade-offs LEZs entail [24]. Critics of the LEZ have noted its potential to impose an economic burden on disadvantaged population groups, people with mobility issues, and large families. These groups are heavily reliant on personal vehicles, but they are unable to switch to cleaner vehicles. Local residents, or those residing in nearby areas, may experience spillover air pollution and traffic congestion as traffic is displaced to adjacent streets. However, some findings challenge this claim [13,17]. Individuals lacking the financial means to acquire LEZ-compliant vehicles may encounter restrictions on accessing essential services in the area. Small business owners within the LEZ have expressed concerns about supply chain disruptions and customer attrition due to restricted vehicular access. The general perception is that insufficient incentives have been offered to affected citizens and businesses in Sofia to encourage transitioning to alternative transport modes and investment in cleaner vehicles, and exemptions should be broadened [22,23]. Furthermore, LEZ interventions should be integrated with complementary measures to limit private vehicle usage. These measures may include alternating restriction schemes, parking restrictions, and short-term restrictions during periods of high pollution [24,59].

While the findings have been mixed, there is evidence that cardiovascular health outcomes decrease following the implementation of a LEZ [21]. However, in this instance, we did not assess the impact of LEZ implementation scenarios on population health. We recently conducted a quantitative health impact assessment for Sofia, focusing on spatial variation in the burden associated with high PM_2.5 and NO₂ exposure. This study identified uneven distribution of the burden across Sofia, with central areas characterised by high socioeconomic status experiencing significant exposure and health impacts. Less privileged populations in the northern districts are also adversely affected [6]. Some authors have argued that LEZs are most effective when they cover a wide area and are stricter, as opposed to establishing rings in city centres, as is the case in Sofia [13,17]. The health benefits can also be extended beyond a limited geographic area by scaling up LEZs [16].

5. Conclusions

The modelling results show a reduction in NO₂ along major roads and at intersections with heavy traffic, but the expected concentrations are still high, especially according to the new EU Directive that will come into force by 2030. PM levels in Sofia have not changed significantly since the introduction of the LEZs. The Bulgarian government and local authorities have to take immediate steps to reduce pollution in the coming years in order to meet the new air quality standards. Public transport must become the primary way of transportation in the city for longer distances. However, this requires high-quality working environment and convenient conditions for passengers to arrive, wait, transfer, and be transported. The promotion of other alternative modes of transport (including walking, cycling, shared vehicles, motorbikes/scooters) is necessary to limit car use to an appropriate level. The LEZ should cover a larger area than the large ring, following a preliminary assessment of the results of traffic displacement to other urban areas.

This study showed the potential of numerical modelling in the process of selecting different measures and providing recommendations to decision-making institutions. An integrated model is needed to analyse and compare different scenarios for the development of the transport system, and the gradual introduction of LEZs must be accompanied by a number of other additional measures and actions.