RI-URBANS is an EU-funded project aimed at reshaping how urban air quality is measured. It focuses on monitoring emerging hazardous pollutants, including ultrafine particles and black carbon, to better protect public health. The project has developed over 20 innovative measurement protocols, modelling tools, and emission inventories to track previously overlooked, non-regulated pollutants. These methods have been tested in several European pilot cities to demonstrate how complementary data can be integrated into routine monitoring networks. RI-URBANS maps urban exposure and investigates how specific atmospheric nanoparticles affect human morbidity and mortality, directly supporting EU efforts to implement stricter air quality standards. The initiative builds on existing research infrastructures, including ACTRIS, and aligns with World Health Organization guidelines.

While the RI-URBANS project addressed multiple themes (e.g. measurements, modelling and emission inventories of emerging pollutants, source apportionment, health effect assessment; see https://riurbans.eu/project/, last access: 20 August 2026), the present paper focuses on the urban mapping and pollution hotspot identification, discussing the pilots undertaken in RI-URBANS from several European research groups (Fig. 1). In general, there were two main approaches to the AQ mapping and pollution hotspot identification pilots. Firstly, the use of previously tested and novel modelling methodologies was deployed on pre-existing data, for detailed AQ mapping of areas. In several cases, two monitoring and modelling approaches were combined. Secondly, novel measurement approaches were tested, which in some cases involved citizen scientists from the local community. The data collected from these studies were interpreted using AQ mapping and various statistical analyses.

Different data analysis approaches were used depending on the envisaged research question and/or use case, either data-only or modelling approaches. While the data-only approaches generate AQ maps purely relying on the collected monitoring data, the model approach relies on modelling techniques to extrapolate air quality outside of the spatiotemporal monitoring window. Thus, the modelling techniques can also incorporate data from external sources (remote sensing, traffic intensity, emission factors, etc.) to estimate pollution levels over locations without measurement data, the identification of air pollution sources and the dispersion of their emissions.

Modelling can provide AQ mapping with or without the use of field measurements using information from emission inventories. Dispersion models using emission inventories or Land-Use Regression (LUR) models using emission inventories were developed and were also extensively tested in many studies as part of the RI-URBANS project to predict pollution concentrations across the area studied from the underlying statistical relationships. Data from emission inventories, while offering an alternative when field data are not available, can lead to great discrepancies in areas where localised emission factors or activity data are not available, reducing the reliability of estimated concentrations (Holicki, 2011). Regardless, advanced modelling techniques, particularly LUR, source apportionment methodologies, AI-based algorithms and machine learning approaches, which allow meaningful interpretation of heterogenous datasets (Hofman et al., 2022b; Yuan et al.,  2022) were also tested by the RI-URBANS pilots. Table 1 lists the models used and the pollutants assessed from each pilot country.

The pilots in France mainly used the CHIMERE model, an open-source multi-scale chemistry transport model, which can forecast pollutant concentrations and make long-term simulations for emission control scenarios (LMD, 2025). The pilots in the UK focused on the ADMS (Atmospheric Dispersion Modelling System) model, an advanced dispersion model used to model the air quality impact of existing and proposed industrial installations. Specifically, the ADMS-Urban model can model the dispersion of particulate matter and chemical substances in the urban environment (CERC, 2026). The Environment – High Resolution Limited Area Model (Enviro-HIRLAM), used by the Finnish group, is developed as a fully online integrated numerical weather prediction and atmospheric chemical transport model for research and forecasting of joint meteorological, chemical and biological weather (Baklanov et al., 2017). The pilot cities in Greece used several models. The CAMS (Copernicus Atmosphere Monitoring Service) atmospheric model, an operational global forecasting system that tracks and predicts atmospheric composition, including air quality, greenhouse gases, and aerosols (ECMWF, 2026). Another model used was the Particulate Matter Comprehensive Air Quality Model with Extensions (PMCAMx), which is a state-of-the-art three-dimensional chemical transport model developed to simulate the atmospheric processing, mass concentration, and chemical composition of particulate matter over regional and urban scales (Gaydos et al., 2007). Finally, the group from Patras in Greece developed the SmartAQ system, which incorporates input from models including the PMCAMx, WRF (for meteorological data), MEGAN3 (for biogenic emissions), and outputs concentration forecasts for several gaseous pollutants and PM. The SmartAQ model (1 × 1 km²) provides advanced treatment of OA volatility chemistry, and uses an updated emission inventory, including biomass burning emissions and can forecast not only pollutant concentrations but also the source contributions for them (Siouti et al., 2022).

These models can handle large, complex datasets and uncover patterns that may not be evident through traditional statistical techniques, providing high-resolution mapping, exposure assessment, and policy evaluation. Moreover, emission factors from inventories can be optimized based on data assimilation techniques (Nguyen and Soulhac, 2021).

Novel measurement techniques involved the setup of stationary networks, mobile measurement approaches or remote sensing to complement the existing regulatory AQMN. Apart from novel setups of research grade instruments (e.g. on vans) low-cost sensors (LCS) were also extensively tested. These LCS can measure a variety of pollutants, including PM_2.5, PM₁₀, NO₂, CO, O₃, and in some cases black carbon (BC), ultrafine particles (UFP) and organic aerosols (OA). Table 2 lists the campaigns using novel data collection techniques from the RI-URBANS core and partner countries. The Alphasense OPC-N3 was the primary sensor used by the UK group for PM measurements. This is an optical particle counter sensor measuring PM in the range 0.35 to 40 µm, providing PNSD in 24 size bins and using that information to estimate PM₁, PM_2.5 and PM₁₀ concentrations. For UFP measurements the Testo DISCmini, a hand-held ultrafine particle counter measuring the number and average diameter of nanoparticles in the diameter range from 10 to 700 nm, was used. For BC measurements, several sensors were tested, for example the single wavelength microAeth AE51 and Observair and the more advanced MA200, MA350 capable of measuring in multiple wavelengths providing additional information about the carbon content of the samples (Elomaa et al., 2025). Finally, the Kunak Air Mobile, a modular mobile IoT sensor system measuring PM, NO₂ and O₃, was deployed by some of the campaigns.

Remote sensing approaches were also tested for air pollution mapping and hotspot identification in several RI-URBANS sub-projects, in particular the use of satellite and LIDAR data. These approaches are expected to play an increasingly important role in future AQ monitoring, providing data collection with greater spatial coverage.

The combination of all these novel approaches, in combination with the data obtained from the existing AQMN, can deliver valuable insights and help guide urban AQ policies. The combination of different methodologies as well as the use of LCS require robust QA/QC, which within RI-URBANS was achieved through campaign specific AQMN co-location campaigns, data cleaning, calibration, curation and normalisation procedures, assisting in the consistency and accuracy of the data collected. In most cases the data collected were openly available in data banks, such as the ARGOS platform (RI-URBANS, 2025), further promoting transparency and increasing awareness.

Finally, the involvement of citizen scientists in the monitoring process was used in several pilot cities in the RI-URBANS project. Citizen science projects face several challenges. These, together with the measures adopted to overcome them and the lessons learned from the campaigns, are discussed in this study to inform future citizen science applications.

Several models and approaches were tested by different groups within RI-URBANS. For the Paris pilot, techniques based on deterministic modelling provided high resolution outdoor exposure city maps. The CHIMERE model (Menut et al., 2024) was used and coupled with the MUNICH street scale model (Sartelet et al., 2020). This model was used to successfully estimate the human exposure to NO₂, BC and PM₁ and PM_2.5 (exposure scaling factors 1.04–1.26) (Park et al., 2025), the variability of O₃ and PM with different meteorological conditions (while ozone was efficiently predicted, PM_2.5 was overestimated by a factor of 1.5–2) (Di Antonio et al., 2025a), as well as the oxidative potential of atmospheric particles (r=0.35–0.60) (Vida et al., 2025), a metric which can be used for AQ risk assessment. The results from these studies were also evaluated against observations from the local AQMN. Maison et al. (2024) using the CHIMERE-MUNICH model found a mixed effect of the urban trees upon the AQ on the streets of Paris, finding a substantial increase in the organic particle concentrations while an opposing effect on gas and particle concentrations was found from the dry deposition on the leaves. Furthermore, Vida et al. (2024) modelled the fungal spore concentrations (r=0.60 for the estimation of monthly concentrations), which can make a significant portion of the total PM₁₀ in the atmosphere, but are not considered in many air quality models. While the model was successful in estimating the concentrations and seasonal changes in the northern and eastern parts of France, it was not so successful for the southern parts of France, mainly due to the limited dataset availability there. The CHIMERE model was also used for modelling the effect of large-scale events, such as the wildfires in Canada in 2023, for which the dispersion of the BC and Brown Carbon (BrC) emissions were estimated, with the effect reaching up to Eastern Europe, a result which was confirmed when the model estimations were compared against satellite observations (r=0.65–0.79) (Tuccella et al., 2025).

Another commonly used model for the dispersion of the pollutants is the ADMS, which was tested thoroughly in several case studies by the Birmingham pilot. The ADMS was used to estimate the dispersion of road and regional emissions for PM, BC and NO₂. For the PM (r=0.47–0.49, for the different size ranges), it was found that while road transport is a major source, the rural background plays a decisive role for ultrafine particle concentrations in the urban environment (Zhong et al.,  2023). On a street-scale simulation over Birmingham, UK, it was found that reduction of traffic, while resulting in signficant NO₂ concentration reduction, had a limited effect on PM_2.5 (r=0.49–0.79, for the different pollutants) (Zhong et al., 2024). Furthermore, the ADMS-Urban model was capable of modelling BC concentrations in different scenarios, though adjustments to emission factors were suggested for improved estimation of the traffic contributions when compared to observations from the AQ monitoring network (r=0.46–0.61, for the different sites) (Zhong et al., 2025).

For the Athens pilot, the concentrations of inorganic secondary aerosols was assessed using the multi-scale numerical atmospheric model system CAMS/WRF/EPISODE-CityChem (Myriokefalitakis et al., 2024). The simulated concentrations, within 100 m cells, showed that for the three inorganic secondary aerosols simulated, the mean concentrations compare well with observations satisfying the model performance criteria and goal of Boylan and Russell (2006) in most cases (r=0.46–0.61, for the different aerosols). Another group from the Greek pilot in Patras used two models for the prediction and mapping of the air quality. The PMCAMx was the base for the estimation of the concentrations of the gases and aerosols. Using this model the effect of biomass burning on PM_2.5 (Siouti et al., 2023) as well as that of the different mechanisms of nucleation on the UFP (Patoulias et al., 2025) were estimated throughout Europe. Furthermore, using the Particulate Source Apportionment Technology (PSAT) algorithm (Wagstrom et al., 2008), biomass burning was found to be the most significant contributor to the PM_2.5 during the winter months (up to 70 %). In both cases the results were evaluated against field measurements, which for the first study were derived from measurements from the city's LCS network (Fractional error up to 0.85 for hourly estimations for the first study and normalised mean error up to 61 % for the second). The PMCAMx model was also successfully tested to predict the PM_2.5 concentrations and its sources, as well as exposure in a 1 × 1 km² resolution grid in Athens (Siouti et al., 2025). The performance of the model was good for both PM_2.5 and OA estimation according to the criteria set by Morris et al. (2005) (Fractional error 0.38–0.57, for the different sites). The group from Patras also developed and tested the SmartAQ system for one month for each season in 2021–2022 in the city of Patras for PM_2.5 and NOx. The system performed excellently for PM_2.5 during the summer and winter times, while its performance was good during autumn for the city centre and average for the suburbs, due to overestimation of long-range transport. Similar was the performance for NOx, with an underestimation of the concentrations during the daytime against the observed values (Siouti et al., 2024).

Other models were also tested in specific cases. For example, the Enviro-HIRLAM modelling framework was used to estimate the contribution of forest fires on the BC concentrations over Ukraine (Savenets et al., 2022). The model estimated a contribution of 10 %–20 % of BC to the total aerosol mass near the wildfires in the lowest 2 km layer, while BC emissions from the wildfires were found in the accumulation and coarse modes in distances up to 2000 km.

Finally, several machine learning methodologies were tested by Fung et al. (2024) for the mapping of BC concentrations. For this work, the machine learning models were trained in Barcelona and then tested with datasets from urban and traffic sites across Europe. It was found that BC concentrations correlate well with PNC of the accumulation mode and NO₂, which was consistent in other European sites. Overall, the tested ML model gave an acceptable performance (R2=0.75–0.83, for the different models), highlighting the transfer possibility of these models across space and time.

The pilot cities tested novel methods for data collection for air quality mapping. For the Birmingham pilot, monitoring included citizen-involved approaches using LCS for static and mobile measurements, including walking and cycling in Selly Oak, a populous neighbourhood next to the University of Birmingham. The main sensor used in all these studies was the Alphansense OPC-N3. The air quality assessment at neighbourhood scale included monitoring at 6 static measuring points (including at the Birmingham Air Quality Supersite, as a background site) which also allowed evaluation of long-term LCS performance vs. research grade instruments for air pollution monitoring. The street level air quality assessment and pollution source apportionment included stationary monitoring and both walking and cycling sessions with the involvement of citizen scientists. For this additional equipment was used, including a Aethlabs microAeth AE51 for BC measurements and a Testo Discmini for PNC and Lung Deposited Surface Area (LDSA), fitted into a backpack for easy transport. This allowed the air quality mapping and identification of PM hotspots (Damayanti et al., 2026) and the estimation of the effect of both regional and local sources at high spatial resolution (100 × 100 m) (Bousiotis et al., 2024), thereby providing valuable information on hyperlocal sources of pollution (Fig. 2). Furthermore, PM_2.5 concentrations, using a combination of collected data (mobile and static), with additional traffic, topography and demographic data were processed to create a machine learning model to fill data gaps and predict PM concentrations when measured data is not available (Baruah et al., 2024).

Finally, in one of the few indoor air quality studies among the pilots, with the involvement of students at the University, LCS were installed in three student houses, within the same Selly Oak region as the mobile measurements, to assess the differences in PM sources and concentrations within indoor environments (Rathbone et al., 2025).

In the Rotterdam pilot, citizen-based mobile monitoring and car-based measurements were performed. The citizen science mobile monitoring was performed following methods developed by Van Den Bossche (2015). In total, 38 participants collected data whilst commuting (to/from work) with portable instruments for BC. The data was used to map exposure and derive representative long-term average AQ maps during commuting hours in winter and summer. Like the Birmingham pilot, the microAeth AE51 was used for data collection, after confirming the comparability against the reference (AE33) by means of an initial co-location. The variability reported by the mobile sensors was similar to that of the reference monitor, and representative spatial maps of BC exposure were derived. Moreover, a sensitivity (subsampling) analysis revealed that up to 15 and 29 sampling repeats were needed to obtain within-season and multi-season representative AQ results, after post-processing the mobile data by means of winsorizing (process of limiting the extreme pollution measurements so they do not disproportionally affect the analysis) and background normalization (Fig. 3).

Car-based mobile monitoring was also implemented In Rotterdam, measuring the ambient concentrations of NO₂ (CAPS, Aerodyne Research Inc., USA), BC (Magee Scientific AE33) and UFP (TSi EPC 3783). The collected measurements were used to develop a mixed-effect model following methods described in Kerckhoffs et al. (2022). Predictions represent long-term average air pollution concentration maps, and where possible, to investigate the impact of industrial sources (mainly port activities) on the total concentration values. The pilot was successful in producing maps of the individual pollutants, but less successful in the interpretation based upon monitoring data directly. The ratios between the pollutants offered new insights into the source contribution of the pollutants. UFP was often elevated near airports, whereas BC and NO₂ were more confined to road traffic sources. This dataset was also used to elucidate the effect of street trees on the pollutants measured (Fry et al., 2025). Concentration of NO₂ and BC were higher during the summertime (when trees had more leaves) in areas with many trees due to pollutant trapping (Vos et al., 2013), while PM_2.5 was lower, highlighting that this pollutant is usually transported from other areas into street canyons. Finally, these data were combined with data from mobile campaigns from Copenhagen and were used to develop LUR models for Amsterdam, for which mobile data was also available for evaluation (Yuan et al., 2024). Testing of the models highlighted the ability for successful hyperlocal air pollution mapping for a city even without any pollution measurements, with Pearson correlations reaching up to 0.92 for NO₂ and 0.90 for UFP.

For the Bucharest pilot, mobile measurements campaigns without citizens' involvement were carried out for UFP, particulate matter fractions (PM₁, PM_2.5, and PM₁₀), BC and NO₂ using LCS on specified routes (including heavy traffic roads, inside the city, residential, industrial, commercial, sub-urban areas), along with the ESCAPE LUR model, the PyLUR tool and QGIS. Maps based on car measurements showed that the UFP sources seem to be widely distributed during summer, while winter is characterized by more homogeneous sources (Talianu et al., 2025). Significantly elevated concentrations were found mainly in the industrial area and urban agglomerations, but also on some important traffic routes. Overall, while the model performed better during the cold season, NO₂ values were overestimated, and PM₁₀ levels were slightly underestimated during the warm season, resulting from the assumptions made and the uncertainty of the data used for training the models. Similar was the variance found for BC, using mobile data collected by a car in the city of Cluj-Napoca during periods within and out of the COVID-19 lockdown periods (Van Poppel et al., 2023). The BC concentrations on roads with high traffic intensities were found to be up to four times greater compared to those with reduced traffic. Furthermore, about halved BC concentrations were found during the lockdown periods highlighting the effect of limiting the traffic activity during these periods. Additionally, using mobile measurements and the LUR model developed from the Rotterdam pilot for pollutant concentration prediction, the effect of power plants on the nearby residential areas in the city of Bucharest was assessed (Nicolae et al., 2025). It was found that despite the initial estimations, the power plant in the western side of the city was not the main source of pollutants for the nearby residential area. Such studies provide crucial information on the assessment of the environmental impact of industrial activities, which in many cases can be misinterpreted, thus leading to erroneous actions which may increase the environmental action cost without providing the anticipated impact.

Helsinki measurements includes the Kumpula campus campaign measurements of BC (Elomaa et al., 2025) and a mobile bike-based measurement campaign (Kleemola et al., 2024). In the urban area, on the Kumpula campus the measurements were conducted at the SMEAR-III station as an urban background environment (Järvi et al., 2009) and at the Mäkelänkatu supersite as the urban street canyon (Barreira et al., 2021). A network of four types of small-scale filter-based BC sensors (AE51, MA200, MA350, Observair) was deployed with the objective to evaluate these sensors for monitoring ambient BC concentrations and to study variations in high resolution data. Sporadic and transient high values were observed both with sensors and with the reference instruments indicating spatially and temporally varying BC sources in the area. For this campaign the sensor data correlated relatively well against the reference (Pearson corelation 0.78–0.84). Mobile bike-based UFP, PM_2.5, BC measurements were also performed in Helsinki, where instrumentation was constructed on a bike and connected to cloud-services for data access. Sampling was done on a route that connected areas with high variability in aerosol number concentration and two AQ supersites, namely SMEAR-III and the Mäkelänkatu supersite. However, results showed that the correlation was modest. As expected, the highest concentrations were observed near traffic and considerably lower concentrations were observed in the park areas. Apart from bike campaigns, mobile measurements were also collected using the ATMo-Lab (Aerosol and Trace Gas Mobile Laboratory), a van carrying instruments targeted on the assessment of traffic derived UFP. Measurements collected using the van were combined with measurements from the Traffic and UB AQ stations in Helsinki for a source apportionment study to identify organic factors connected to different particulate sources (Teinilä et al., 2025). It was found that local traffic emissions increased the PNC, especially for particle sizes < 10 nm, while long range transport increased the PM mass concentration and particle size. A different van, the “Sniffer”, was used to identify pollution hotspots in the city centre of Helsinki (Järvi et al., 2023). In this campaign, with the use of the mobile data along with data from the local monitoring station, both the horizontal and vertical pollutant variation was studied. For the vertical measurements a drone collecting LDSA data (Naneos Partector) was used. Using this data the mechanisms causing pollution hotspots in street canyons were explored, pointing to different drivers for the warm and cold periods of the year, highlighting the role of thermal processes within the street canyon as an important factor during the winter.

Finally, analysis of mobile measurements was also implemented in Belgium as part of the RI-URBANS, though the datasets were collected in other studies. Two approaches were tested, involving citizens in the collection of BC data using bicycle rides in a targeted approach, and the use of postal service vehicles for PM, NO₂ and O₃ data in an opportunistic approach. For the first project, the measurements were collected by citizen volunteers in Mechelen, Belgium who participated in a local citizen observatory, as part of the GroundTruth2.0 project (Van Poppel et al., 2024). For this campaign the citizen scientists, under the guidance from members of the research partner VITO and the city council, collected BC data using the AE51 in repeated rides on four predefined routes within the city in periods from all seasons. As repeated measurements are needed for adequate assessment of the BC concentrations, for each route and campaign 25 repetitions were done. Strong seasonality of the BC concentrations was observed, with the highest concentrations found in the late autumn campaign and the lowest in the summer campaign, while the diurnal variability was more complex and was found to depend on the proximity to sources, dilution and dispersal dynamics. The mobile measurements were also used to evaluate modelled results from the ATMO-Street model (using multiple data sources including the AQ network). The mobile measurements presented a fair correlation with the modelled data. Higher correlations were found in the measurements collected in the suburban areas, where the cyclists were not exposed to traffic emissions. The model underestimated the peak concentrations observed in areas with heavier traffic. For the second study, the Kunak Air Mobile was used (Hofman et al., 2023). Twenty of these systems were installed on postal vehicles operating in Antwerp. A total of 945 km of road was covered by the vehicles in the 7-month campaign, providing about 8 million datapoints scattered throughout the city. NO₂ measurements properly reflected the observed exposure range as measured by the AQMN in the city and pollution hotspots were mapped throughout the city (Fig. 4). This opportunistic data collection method was proven to be a valid approach for pollution exposure assessments and a valuable source of air quality data for cities with a limited monitoring network. The same mobile mapping approach was also used in a project in Cluj-Napoca, Romania, with the the participation of citizens, where the impact of traffic on BC concentrations during the COVID-19 lockdown was assessed highlighting that background corrections are needed for better assessment of the traffic impact (Van Poppel et al., 2023).

A number of alternative approaches were also tested in the RI-URBANS project from both the pilot and partner cities. An interesting approach for air mapping data was used by a group from Barcelona. In this study, sand and soil was collected from 23 playgrounds and parks around Barcelona (González-Romero et al., 2025). The mineral content and trace element content of the samples were determined offline at the laboratory using two types of spectrometry. The sand samples from several parks exhibited significant enrichment factors for elements associated with road emissions (brake or tyre wear), while others exhibited elevated content of Pb and Sb which are most likely associated with industrial and port activities. Using this methodology improves the understanding of the urban PM dynamics while pollution hotspots can be identified. However, caution is necessary in data interpretation due to the long residence time of many trace elements in soil and associated legacy effects.

Air quality mapping with the use of the Aerosol Optical Depth (AOD) to monitor the aerosol load and distribution in the atmosphere is a tested approach which was also used by some pilot groups. The group for the Paris pilot studied the variation in Paris and its surrounding area Using the variations of the different wavelengths between the urban and sub-urban environments (Di Antonio et al., 2025b). In this study, the organic aerosols contributed up to 50 % of the aerosol mass with variations observed between urban and forested areas. Similarly, the AOD was used from the pilot in Athens for the estimation of the impact of the wildfires in Athens, Greece in August 2021 on local AQ (Kaskaoutis et al., 2024). The biomass burning tracers used in this study, like nss-K⁺ (from PM_2.5 filter samples) presented high correlations with BC, OC and EC, as well as with specific scattering and absorption coefficients, highlighting the ability of this approach to provide AQ data without the need for traditional monitoring instruments. Using the same methodology, a group of institutions in Italy formed a network of automated lidar ceilometers in 2015, the ALICENET. The ALICENET, comprising of 22 stations throughout Italy (in urban, industrial, coastal and mountainous sites), allows the monitoring of vertical aerosol profiles over a wide range of environmental and atmospheric conditions (Bellini et al., 2024). Apart from routine monitoring the network monitors all kinds of events, such as Saharan dust events, short and long range transport of biomass burning emissions, and the emissions from the volcanic activity in the South of the country. The data provided by the ALICENET network were tested as part of the RI-URBANS project against the ERA5 dataset and CAMS model with good agreement on variables such as the Boundary Layer Height and PM₁₀ (Bellini et al., 2025), providing an alternative and cost-effective way for fine spatiotemporal air quality data collection. In a similar manner but using satellite lidar data instead, the pilot group from Helsinki collaborated with the University of Jordan in one of the first tests of simultaneous use of ground based and satellite observations for air quality mapping in the city of Amman, Jordan (Panahifar et al., 2023). Analysing the data from the space-borne Cloud-Aerosol Lidar with Orthogonal Polarisation (CALIOP) three main groups of aerosols were found over Amman, the coarse mode dust, the fine mode dust (polluted dust) and non-dust aerosols (pollution). The vertical aerosol profile over Amman was mapped and using the trajectory analysis, the sources of the incoming pollutants were distinguished (Fig. 5).

Finally, a novel approach on air quality mapping using street images was tested as part of the RI-URBANS. For this, 2800 high resolution street images were captured in 4 locations in Germany (Augsburg, Neubiberg, Warnemünde) and the Czech Republic (Zelezna Ruda). Using three ML methodologies on the luminance of the red, blue and green colours found in each pixel of the images the PM concentrations was assessed. To assist the model construction, sampling campaigns were conducted in the study areas. These sampling campaigns included walks around the study areas in sunny and cloudy days, for the collection of PM₁, PM_2.5, PM₁₀, BC, BrC, PNC and LDSA data. The models tested demonstrated adequate performance and satisfactory generalisation capabilities in both temporal and spatial dimensions, indicating that with proper calibration this approach can be used in different areas and seasons (Liu et al., 2024) (Fig. 6).

The preceding sections outlined the multi-faceted strategy adopted by the RI-URBANS project to advance urban AQ monitoring and pollution hotspot identification. The need for spatially dense datasets and on demand campaigns led to the considerations of more flexible monitoring methodologies. The LCS can provide on demand air quality measurements for a wide range of atmospheric variables with an affordable cost and great flexibility. Stationary sensor networks offer cost-effective means of expanding long-term monitoring coverage, especially in underrepresented or vulnerable communities (Shabbir et al., 2025), while mobile measurements can substantially expand the spatial coverage. Standardized QA/QC protocols across municipalities would enhance trust and enable cross-comparative analyses (Bousiotis et al., 2025). The integration of stationary and mobile measurements allows for real-time assessment of pollution hotspots and population exposure across diverse urban environments. These platforms are especially valuable for identifying local sources of pollution, validating traffic-related emission controls, and assessing the effectiveness of urban interventions. For example, Hofman et al. (2022b) showed the significant air quality impacts of temporary traffic restrictions in a school street using a stationary sensor network.

The pilot studies focused on modelling, and mobile monitoring with reference grade monitors or portable LCS which have been shown to compare well with reference grade monitors. Dispersion modelling was useful and agreed well with measurement data from the often-few regulatory monitoring stations. The mobile approaches also worked well to develop maps across the city, useful for studies which could benefit from finer resolution, for example epidemiological studies. Figure 7 illustrates the difference between mobile and stationary monitoring approaches and associated considerations in terms of monitoring set-up and device requirements. For the identification of specific hotspots, more focus on data-only approaches is needed, as was the case on the approach adopted in Rotterdam. Monitoring data are useful to identify locations where current models under- or overestimate concentrations. Furthermore, citizen involvement increases the awareness and enriched interpretation as citizens were able to provide feedback on observed spikes and hotspot locations in their commuting environment. Identification of hotspot locations, and assessment of personal exposure during commuting are additional merits of mobile monitoring.

In general, modelling is easier to implement since it requires a fewer number of repeat measurements, and can provide sufficient air pollution mapping, if prior expertise already exists and input data on emissions are available. However, contrary to data-only approaches, models only provide AQ predictions and the robustness of these model predictions relies on underlying model assumptions. Challenges may be raised by the requirements of regulatory modelling, limiting the flexibility of model choice. This applies less to unregulated pollutants such as UFP. Novel measurement techniques though, can assist in expanding current AQ mapping capabilities. LCS monitoring, stationary or mobile, is promising to refine spatial resolution, assist in hotspot identification and evaluate models' output. As one of the main targets of the RI-URBANS project was to expand the adoption of both mobile measurements and citizen involvement, the lessons learnt from these applications will be discussed on the next sections.