Dry deposition is an important process of removal of various airborne substances from the atmospheric boundary layer. In many applications it is convenient to assume that the deposition flux of a substance is proportional to the near-surface concentration, and that the proportionality coefficient does not depend on particle concentration. This assumption is based on the idea of a constant-flux layer between the reference height and the surface, and holds for substances that have no sources/sinks in the layer. The deposition velocity concept is a core part of dry deposition schemes of atmospheric transport models.

We address large discrepancies between field and wind-tunnel measurements of deposition velocities of aerosols with aerodynamic diameter between approximately 0.1µm and 2µm. In seemingly similar conditions, deposition velocities derived from field measurements are in range of 1-10 cm/s, while wind-tunnel measurements show a fraction of a millimeter per second. This difference translates to the discrepancy in dry deposition parametrizations.

SILAM chemistry transport model features a dry deposition scheme for particles by Kouznetsov and Sofiev (2012, https://doi.org/10.1029/2011JD016366) that predicts 'low' deposition velocities. With such a scheme, simulations that explicitly account for aerosol transformations are able to reproduce the ambient observed fluxes and agree well with the 'high' apparent deposition velocity. A regional simulation covering the period of the Gallagher (2007, https://doi.org/10.1016/S1352-2310(96)00057-X) field campaign was capable of reproducing both magnitude and temporal evolution of aerosol fluxes measured over a forest.

We demonstrate that the conservation of aerosol mass in the immediate vicinity of the surface is not fulfilled for ambient aerosols when the aerosols include a fraction of ammonium nitrate. For such a mixture the fluxes of ambient aerosols are not controlled by particle deposition but rather by gas-particle partitioning in the vicinity of the surface and by the deposition flux of nitric acid. The particle flux does not depend on particle concentrations in quite a wide concentration range. For such a mixture the entire concept of deposition velocity is inapplicable.

Simulations of atmospheric aerosol composition show that the presence of ammonium nitrate as a part of aerosol is rather common in many places of the world. Moreover, we are not aware of any publication that demonstrates a linear dependency between the flux and concentration for ambient accumulation-mode aerosols. Based on our findings and the results of wind tunnel measurements we suggest that field campaigns could observe detectable fluxes of aerosol only if the fluxes were caused by aerosol processes in air. Therefore, such measurements cannot be used to directly infer particle deposition velocities, and the measurements with known conservative particles should be used instead. Parametrizations of deposition velocities that are based on the field-measured fluxes do not predict flux-concentration relation for particles if ammonium nitrate is present, and strongly over-deposit conservative aerosols. Therefore, the parametrizations based on wind-tunnel measurements with calibrated particles should be used instead, despite high-vegetation cases are not covered by such experiments.

How to cite: Kouznetsov, R., Sofiev, M., Uppstu, A., and Hänninnen, R.: On the applicability of the deposition velocity concept for ambient aerosols, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-11407, https://doi.org/10.5194/egusphere-egu25-11407, 2025.

Benzo[α] pyrene (BaP), a polycyclic aromatic hydrocarbon (PAHs), is a ubiquitous environmental contaminant. Since 1998, PAHs have been listed in the Convention on Long-Range Transboundary Air Pollution (CLRTAP) Protocol on Persistent Organic Pollutants. This pollutant is of great public concern because of its toxicity and potential carcinogenicity. Emissions of BaP occurred as early in 1970s, increased till 1990, then decreased before spiking again from 2000 onwards. This shift in emissions from developed to developing countries is largely attributed due to shifting of BaP emitting industries, with China and India being the largest emitters of PAHs. In light of environmental significance, it is important to know the emission scenario of BaP. Results indicated that Asia has the highest regional emissions (1.73 х 10⁸ kg), while Australia (1.03 х 10⁶ kg) has the lowest. In the present study, have used the BETR-Global model to understand the BaP scenario at global scale. Here, we will highlight the long-term trends (1970 – 2018) of BaP transboundary seasonal depositions and seasonal inflows across Asia, Europe, Africa, North America, South America, Australia, Arctic and, Antarctica. This research underscores the importance of understanding the shifting dynamics of BaP emissions for effective environmental management and policy development.

How to cite: Yadav, P. and Qureshi, A.: Source attribution of seasonal continental deposition and trans-continental fluxes of benzo [α] pyrene, 1970-2018, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-998, https://doi.org/10.5194/egusphere-egu25-998, 2025.

In this study, the chemical transport model COSMO-MUSCAT (Wolke et al., 2012) is used to investigate the sources of particulate matter (PM). Model results are compared with observational data from winter and summer campaigns conducted at one site in Germany and two sites in the Czech Republic. These sites are located in a central European transition zone with a gradient from highly polluted to less polluted regions.

A non-reactive tagging approach was used to track primary organic matter (OM) and black carbon (BC) emissions by sector and country of origin at a high spatial resolution of about 2 km. In addition, sensitivity analyses were performed to assess the impact of volatile organic compound (VOC) emissions and associated secondary organic aerosol (SOA) formation.

Source attribution showed that residential heating is a major contributor to primary particulate matter (PM2.5) in winter. Sensitivity tests indicated that the model likely underestimates SOA production from AVOCs emitted during wood and coal combustion. By adjusting the SOA yields and emission rates for these combustion sources, modeled OM concentrations increased by up to 40% on average at the monitoring sites.

The findings underscore the significant role of AVOC precursors in the SOA budget, which is currently underrepresented in the model. Comparison with summer campaign data provides further insights into model performance and highlights seasonal variations in PM composition and sources across this critical region of Central Europe.

How to cite: Wiedenhaus, H., Schroedner, R., Wolke, R., Arora, S., Poulain, L., and Lhotka, R.: Importance of Anthropogenic Sources for Seasonal and Spatial Variability of Primary and Secondary Particulate Matter in Central Europe, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-8756, https://doi.org/10.5194/egusphere-egu25-8756, 2025.

Despite mitigation efforts, ozone pollution in Europe remains a significant issue. As part of the EMEP program, the Task Force on Measurement and Modelling (TFMM) conducted a measurement campaign from 12-19 July 2022 to evaluate the impact of various VOC species on ozone concentration levels and variability. This campaign coincided with adverse thermal conditions – a strong heatwave moving from west to east across Europe, enhancing biogenic VOC emissions and ozone production.
To interpret the campaign results, TFMM launched an air quality modelling exercise involving 11 models that reproduced the variability of chemical tracer concentrations in July 2022. Apart from the experiments to reproduce the evolution of concentrations, additional scenarios aimed at assessing the contribution of anthropogenic vs. biogenic VOC emissions were undertaken. The importance of the dry deposition of ozone was also evaluated. We will show preliminary results from the study focusing on model performance during the peak of the episode and the spread between individual models based on measurements taken at EMEP stations and during the campaign. The average contribution of biogenic emissions to ozone and its precursors will also be assessed.

How to cite: Struzewska, J., Przybyła, T., Norowski, A., Kaminski, J., and Jeleniewicz, G.: Impact of Biogenic Emissions on Ozone Episode Evolution During the July 2022 Heatwave: A TFMM Modelling Exercise, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19789, https://doi.org/10.5194/egusphere-egu25-19789, 2025.

Aerosol acidity is an essential property of atmospheric particles that affects not only atmospheric processes such as cloud formation, oxidation capacity, climate, and gas-particle phase partitioning, but also the Earth system, such as nutrient availability in terrestrial and marine ecosystems, and human health. The global distribution of aerosol acidity exhibits distinct spatial and temporal patterns, driven by variability in aerosol chemical composition, aerosol abundance, and local meteorological parameters. Due to the implementation of related clean air policies, a substantial reduction in aerosol abundance and a significant shift in chemical composition have been observed in recent times (Tsimpidi et al., 2024). However, the response of aerosol acidity remains modest and depends on the combined effect of aerosol changes and meteorology (Karydis et al., 2021). The contribution of each driving factor is debated, and the decadal trend of aerosol acidity is not well understood. In this study, we present a decadal simulation of global aerosol acidity using the EMAC atmospheric chemistry-climate model. The simulation is evaluated with results derived from field measurements over the continents of North America, Europe, and Asia. A one-at-a-time approach is employed to quantify the contributions of key driving factors, including temperature, relative humidity, and the availability of sulfate, total (gas and aerosol) nitrate, ammonium, and chloride, and nonvolatile cations  (sum of Na⁺, Ca²⁺, K⁺, and Mg²⁺) to annual and seasonal trends in aerosol acidity. Compared to field measurements, our simulation accurately reproduces temperature and relative humidity and shows good agreement of aerosol acidity with field measurements in Europe and the Pearl River Delta. We find that the underestimation of acidic ions, particularly sulfate, is the main reason for the low bias in simulated aerosol acidity in North America, and the underestimation of alkaline nonvolatile cations leads to high bias in aerosol acidity in the North China Plain. These findings highlight the nuanced interplay between chemical composition and meteorological factors in shaping global aerosol acidity trends and emphasize the importance of regional analyses in understanding long-term changes.

References

Karydis, V.A., Tsimpidi, A.P., Pozzer, A., Lelieveld, J., 2021. How alkaline compounds control atmospheric aerosol particle acidity. Atmospheric Chemistry and Physics 21, 14983-15001.

Tsimpidi, A.P., Scholz, S.M.C., Milousis, A., Mihalopoulos, N., Karydis, V.A., 2024. Aerosol Composition Trends during 2000-2020: In depth insights from model predictions and multiple worldwide observation datasets. EGUsphere 2024, 1-66.

How to cite: Wang, X., Tsimpidi, A. P., and Karydis, V. A.: Decadal trends and drivers of global aerosol acidity: insights from model simulations and observational data, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9571, https://doi.org/10.5194/egusphere-egu25-9571, 2025.

Secondary air pollution, especially ozone (O₃) and secondary aerosols, are emerging air quality challenges confronting China. Nitrous acid (HONO), as the predominant source of hydroxyl radicals (OH), are acknowledged to be essential for secondary pollution. However, HONO concentrations are usually underestimated by current air quality models due to the inadequate representations of its sources. In the present study, we revised the Weather Research and Forecasting & Chemistry (WRF-Chem) model by incorporating additional HONO sources, including primary emissions, photo-/dark oxidation of NO_x, heterogeneous uptake of NO₂ on surfaces, and nitrate photolysis. By combining in-situ measurements in the Yangtze River Delta (YRD) region, we found the improved model show much better performance on HONO simulation and is capable of reproducing observed high concentrations. The source-oriented method is employed to quantitatively understand the relative importance of various processes, which showed that heterogeneous NO₂ uptake on the ground surface was the major contributor to HONO formation in urban areas. Comparatively, photo-oxidation of NO_x is a main contributor in rural areas. The introduction of multiple sources of HONO led to an apparent increase in OH and hydroperoxyl (HO₂) radicals. The promoted HO₂ levels further increased diurnal O₃ concentration by 4.5–12.9 ppb, while secondary inorganic and organic concentrations were also increased by 14–32% during a typical secondary pollution event. The improved description of HONO emission and formation in the model substantially narrowed the gaps between simulations and observations, highlighting the great importance in understanding and numerical representations of HONO in secondary pollution study.

How to cite: Zhang, H. and Huang, X.: Improving HONO Simulations and Evaluating its Impacts on Secondary Pollution in the Yangtze River Delta Region, China, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-3372, https://doi.org/10.5194/egusphere-egu25-3372, 2025.

Ground-level ozone (O₃) is a secondary air pollutant and one of the major air pollutants that shape the atmospheric chemistry and influence many chemical reactions in the atmosphere. O₃ is the second most significant air pollutant after particulate matter contributing mortality world-wide. To explore the O₃ (a criteria pollutant) concentration variability influenced by primary air pollutants, especially criteria air pollutants such as volatile organic compounds (VOC), carbon monoxide (CO), nitrogen dioxide (NO₂), and sulfur dioxide (SO₂), we conducted a field campaign at Kharagpur city in India on April 2024. Sample air was measured by the USEPA approved Serinus 10 ozone analyser, Serinus 40 NO_x analyser, Serius 500 Portable Air Quality Monitor (with swappable sensor heads) to get O₃, NO_x (NO+NO₂), total VOC (TVOC), CO, and SO₂ concentrations, respectively. The measurement was carried out on National Highway 49 (NH-49) (22.379128°N, 87.361647°E) in Kharagpur. Results showed a strong negative correlation between O₃ and NO (r = -0.82), a weak positive correlation with NO₂ (r = 0.19), moderate negative correlations with TVOC (r = -0.54) and CO (r = -0.49), and a very weak positive correlation with SO₂ (r = 0.11). All correlations are statistically significant at the p < 0.01 level. We applied Quantile Regression Model (QRM) to explore a robust framework for analyzing the relationships between dependent (O₃) and independent variables (NO, NO₂, TVOC, CO, SO₂) across different points of the data distribution by capturing conditional quantiles. Analysis revealed nonlinear distribution of O₃ concentration across all the quantiles (τ) with a strong performance at the median quantile (τ = 0.5) that explained 76.06% of variability in O₃ concentration (R¹(τ)_0.5) = 0.7606) with high accuracy and low predicting errors (MAE = 14.19, RMSE = 17.86). The local measure of goodness of fit, R¹(τ) were diminished at lower (below τ = 0.1) and higher quantiles (above τ = 0.95) and the Quantile Loss of 7.10 confirms effective handling of O₃ variability. The standardized coefficients for NO were negative across all quantiles and became less negative at higher quantiles (0.8-1.0) that indicated a weaker adverse effect as O₃ increased. NO₂ showed positive coefficients that peaked at the 0.4 quantile and declined at higher quantiles and suggesting a stronger influence at moderate O₃ levels. TVOC consistently exhibited negative coefficients with a stronger effect at lower quantiles (0.1-0.2) that stabilizes at higher quantiles. CO and SO₂ coefficients fluctuate around zero and shows minimal and inconsistent influence. Monte Carlo simulation of health risk assessment showed that O₃ could significantly pose the development of non-carcinogenic health risks (Hazard Quotient (HQ) > 1). Sensitivity analysis revealed the variance in the O₃ Hazard Index (HI) where NO significantly contributed the most (60.4%), followed by NO₂ (23.3%), TVOC (12.1%), SO₂ (4.0%), and CO (0.2%). The Air Quality Index (AQI) analysis categorized Kharagpur as a ‘Moderately Polluted’ region. Overall, NO_x and TVOC are the two types of major gaseous pollutants that contributed majorly in O₃ concentration variability, thus O₃ pollution levels. Targeted policies to reduce VOC and NO_x emissions are essential.

How to cite: Santra, S.: Gaseous Pollutants Driven Ozone Variability, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6582, https://doi.org/10.5194/egusphere-egu25-6582, 2025.

Volatile organic compounds (VOCs) are crucial for atmospheric radical recycling and ozone formation. Despite significant reductions in other air pollutants in China since 2013, ozone and VOC levels remain persistently high, shifting air quality management toward VOC control. However, limited and short-term speciated VOC measurements hinder our understanding of regional VOC characteristics and effective emission reduction strategies for ozone mitigation in many Chinese cities. Therefore, in this study, we leveraged year-round routine VOC measurements in Hong Kong, together with field campaign and spaceborne TROPOMI data, to explore regional VOC characteristics and their relationships with ozone in the CMAQ chemical transport model. Results show that non-methane hydrocarbons (NMHCs) had higher concentrations in colder months and lower levels in warmer months, while oxygenated VOCs (OVOCs) peaked in September, coinciding with the annual ozone maximum and indicating strong photochemical activity in late summer. Notably, HCHO demonstrated a strong temporal correlation with total measured VOCs (R = 0.72–0.85) and ozone (R = 0.7). Among all measured VOC species, many are unaccounted for in the model, resulting in the model capturing only 30% of the total observed concentrations for NMHCs and 26% for OVOCs, as well as 14% of the ozone formation potential for NMHCs and 25% for OVOCs. This underrepresentation led to an overestimation of VOC sensitivity in ozone formation, classifying more areas as VOC-limited in the model. The findings provide valuable insights into regional VOC characteristics, aiding VOC-related model development and informing ozone air quality management strategies in VOC-limited urban environments.

How to cite: Liu, X., Huang, Y., Wang, Z., Chen, Y., Feng, X., Xu, Y., Chen, Y., Gu, D., Sun, H., Ning, Z., Yu, J., Chow, B., Lin, C., Xiang, Y., Zhang, T., and Fung, J.: Ambient volatile organic compounds and their impact on ozone pollution regulation: insights from multi-platform observations and model representations from the 2021-2022 HKEPD-HKUST field campaign in Hong Kong, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-10060, https://doi.org/10.5194/egusphere-egu25-10060, 2025.

Massive computing resources are nowadays required by current chemical transport models (CTMs) operating at global and/or regional scale to solve the system of ordinary differential equations associated with chemical kinetics. The sheer complexity of our atmosphere (in terms of the number of different constituents and reactions) together with the orders of magnitude differing between the chemical and the transport time scales, hinder the use of comprehensive mechanisms in large-scale 3D models. The rapid advancements in the field of machine learning (ML), alongside with the latest improvements in parallel computing, supplied new and powerful tools to the equipment of present-day atmospheric modelers. A notable example is given by physics-informed ML, where specialized network architectures are designed to satisfy the physical constraints of the system under investigation, leading to promising results in the emulation of physical processes. Indeed, physics may be introduced in the ML architecture at different stages, therefore determining the type of constraint (hard vs soft) embedded into the model. 

In this work, the baseline performance is defined on a fully-connected multilayer perceptron (fc-MLP) trained to predict the concentration change using the composition vector at a given time as input. The dataset is generated using Sobol sampling of different initial conditions within a specified concentration range to ensure comprehensive and efficient coverage of the input space. As a first attempt of including physics in the model architecture, we introduce the mech-MLP model, obtained by exploiting an array of MLPs—one per each chemical reaction present in the mechanism—whose outputs (i.e., the change in composition) are aggregated together to determine the total change to each chemical species. Furthermore, chemical and physical soft constraints are introduced also via the use of custom loss functions by imposing penalty terms for un-physical predictions (e.g., negative concentration or divergence from stoichiometry). The trade-off between dataset size, creation cost, and training efficiency, the inductive biases arising from the architecture choice, and the reliability of the model when tested on unseen conditions will be presented for two study cases: an explanatory mechanism involving 3 species and 2 reactions, and a simple, stiff air pollution mechanism (POLLU, doi.org/10.1137/0915076) composed by 20 species and 25 reactions.

How to cite: Melli, A., Mouchel-Vallon, C., Petetin, H., and Jorba Casellas, O.: Emulating tropospheric chemistry with physics-informed machine learning, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-5628, https://doi.org/10.5194/egusphere-egu25-5628, 2025.

Air pollution poses significant risks to human health and the environment. Nitrogen dioxide (NO₂) is a key air pollutant with well-documented adverse health effects and a precursor to ozone (O₃). Generally, particulate matter (PM) is a major global cause of mortality, with both short-term and long-term exposure linked to severe health outcomes (WHO, 2021). High-resolution maps of near-surface air pollutant concentrations are essential to assess these impacts effectively.

We will present daily gridded maps of near-surface NO₂ and O₃, as well as PM_2.5 and PM₁₀ (particles with an aerodynamic diameter equal to or less 2.5 and 10 µm respectively) concentrations across the Italian territory, generated for 2021–2023 at a spatial resolution of 0.01° × 0.01° (~1 km). Machine learning (ML) models are trained using a combination of spatial and spatiotemporal predictors, informed by in-situ ground measurements from over 300 monitoring stations sourced from the European Air Quality Portal (AQP), managed by the European Environmental Agency (EEA) (EEA, 2024). Key spatial predictors include CORINE Land Cover, which provides 44 thematic classes at 100 m resolution; the Global Human Settlement Layer, reflecting human presence; NASA’s Digital Elevation Model, offering topographical information; and the ESA Plant Functional Type dataset (Harper et al., 2023). Spatiotemporal predictors integrate meteorological fields from ERA5-Land and aerosol optical depth (AOD) data from the Modern-Era Retrospective Analysis for Research and Applications, Version 2 (MERRA-2).

We will evaluate various supervised ML models (Mansour et al., 2024b), including neural networks, regression ensembles, and regression trees across various landscapes (urban, suburban, and rural) to identify the optimal approach for each context. Additionally, explainable AI techniques (e.g., partial dependence analysis and Shapley additive exPlanations) and statistical analysis (e.g., clustering and empirical orthogonal functions) will be employed to elucidate the relationships between predictors and aerosol spatiotemporal distributions (Mansour et al., 2023; Mansour et al., 2024a), providing novel insights into the dynamics of air quality across regions. These advancements contribute to a more refined understanding of air pollution patterns and their underlying drivers.

Funding:

Project funded under the National Recovery and Resilience Plan (NRRP), Mission 04 Component 2 Investment 1.5 – NextGenerationEU, Call for tender n. 3277 dated 30/12/2021. Award Number: 0001052 dated 23/06/2022 (ECS_00000033_ECOSISTER).

References:

EEA: Europe’s air quality status (2024), https://www.eea.europa.eu//publications/europes-air-quality-status-2024.

Harper, et al. (2023), Earth System Science Data, 15, 1465-1499, 10.5194/essd-15-1465-2023.

Mansour, et al. (2023), Science of The Total Environment, 871, 10.1016/j.scitotenv.2023.162123.

Mansour, et al. (2024a), npj Climate and Atmospheric Science, 7, 10.1038/s41612-024-00830-y.

Mansour, et al. (2024b), Earth System Science Data, 16, 2717–2740, 10.5194/essd-16-2717-2024.

WHO (2021), World Health Organization, https://iris.who.int/handle/10665/345329.

How to cite: Mansour, K., Rinaldi, M., Decesari, S., Paglione, M., and Landi, T. C.: Machine learning for high-resolution mapping of air pollutants over Italy (2021–2023), EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-16433, https://doi.org/10.5194/egusphere-egu25-16433, 2025.

High-resolution modelling of air pollutants such as NO₂ and PM_2.5 is an essential step in the quantification of the impacts on human health, especially in urban areas. Often, such modelling uses relatively coarse-resolution chemistry transport models (CTMs), which exhibit biases when compared to measurements and cannot consider the heterogenity of urban pollutant concentrations.

This study develops a machine learning (ML) framework to downscale CAMS regional air quality reanalyses for PM_2.5 and NO₂ from approximately 10×10 km² (0.1 degrees) to 1×1 km² resolution, enabling more detailed urban air quality assessments across Europe.

The downscaling methodology integrates meteorological, land-use, and spatial predictors to bridge the resolution gap. Key steps include: (1) interpolating CAMS outputs to a 1×1 km² grid, (2) constructing a training dataset by pairing interpolated CAMS data with ground-based measurements, (3) applying XGBoost (a gradient-boosted decision tree algorithm) and Gaussian Processes to model pollutant concentrations at 1×1 km² resolution, and (4) validating model performance using independent measurement data and FAIRMODE evaluation principles (e.g. Model Quality Objective, MQO). Predictor variables encompass meteorological inputs (e.g., daily temperature extremes, surface pressure, boundary layer height), geographical features (e.g., terrain height, proximity to roads, and coastlines), temporal indicators (e.g., year, month, date), and land-use data (e.g., Corine Land Cover and urban bounding boxes).

Preliminary results demonstrate the ability of the downscaling approach to capture fine-scale spatial patterns in urban air quality for a range of cities in Europe, with improved alignment to ground-based measurements compared to CAMS reanalyses. The high-resolution (1×1 km²) predictions reveal urban-level detail, enabling better inference on pollutant distribution in urban environments. Adherence to FAIRMODE principles ensures transparency and quality of results.

Future work will refine the ML framework, extend its application to other pollutants, and explore spatial and temporal scalability, ultimately aiming to deliver a transferable tool for high-resolution air quality modeling in any urban area across Europe.

How to cite: Ramacher, M. O. P. and Keil, P.: Machine Learning Downscaling of CAMS Regional Air Quality Reanalyses: High-Resolution Urban Concentrations of PM2.5 and NO2 Across Europe, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-9157, https://doi.org/10.5194/egusphere-egu25-9157, 2025.

Mobile monitoring campaigns using Land Use Regression (LUR) models effectively capture fine-scale spatial variations in urban air pollution. While traditional LUR models rely on land-use and demographic features, integrating micro-environmental information from Google Street View (GSV) images offers the potential to further enhance the model performance.

We developed LUVR, a framework that integrates vision-transformer-based (ViT) object detection and semantic segmentation features derived from GSV images into LUR models. Using 5.7 million mobile air pollution measurements and 0.37 million GSV images collected in Amsterdam, we modeled nitrogen dioxide (NO₂), black carbon (BC), and ultrafine particles (UFP) in 50m road segments. Three temporal image selection strategies—specific year, most nearby year, and season-weighted—were tested with stepwise linear regression and random forest models.

We found that adding GSV-derived features improved model performance, increasing R² by 0.01–0.05 and reducing errors by 0.7%–10.3%. The most-nearby-year strategy performed the best for NO₂, while BC and UFP benefited slightly more from the season-weighted strategy. This result suggests that for air pollution modeling, GSV-derived built environment features remain relatively stable across years. Using an open-vocabulary object detection module, we detected customized objects described in natural language in a zero-shot fashion, revealing previously unrecognized predictors such as chimneys, traffic lights, and shops. Combined with segmentation-derived features like walls, roads, and grass, visual features contributed 8%–18% to the overall model prediction.

This study demonstrates the potential of integrating visual features into LUR models to enhance hyperlocal air pollution monitoring and exposure assessment. Future research should optimize feature selection and expand applications to broader urban and environmental health studies.

How to cite: Yuan, Z., Kerckhoffs, J., Hoek, G., and Vermeulen, R.: LUVR: An interpretable Land Use and Visual Regression model embedding Street View images in air pollution modeling with mobile monitoring, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-18163, https://doi.org/10.5194/egusphere-egu25-18163, 2025.

Atmospheric pollutants affect human health, disrupt ecosystems, and impact the economy. Spatial prediction of atmospheric pollutants using data from ground monitoring stations (GMS) is vital for informed decision-making and sustainable ecosystem management. To estimate atmospheric pollutants, this study introduces the Deep-Pollutant-Spatial-Operator-Network (DPSON) framework that combines GMS data with multi-source spatial covariates in order to produce precise predictions at a 1 km × 1 km grid across Delhi (Indian capital city). Pollutant data from 40  Central Pollution Control Board, India (CPCB) monitored GMS locations (January 2021–December 2022) were used for this purpose. The PM_2.5 and PM₁₀ datasets are available at a 3-hour resolution, while O₃ and NO₂ at a 1-hour resolution.

Normalized static spatial covariates, such as population density, waterbody concentration, road-length concentration, green cover, and Land Use Land Cover (LULC), were included to improve the DPSON model’s accuracy. To improve the dataset's generalization in relation to spatial covariate variations, additional samples were generated using the Sequential Gaussian Simulation (SGS) algorithm, randomly simulating pollutant observations at 100 grid locations on a 1 km² spatial grid for each timestamp and pollutant species, based on the pollutant concentrations observed at 40 GMS locations. These SGS-generated and GMS-observed datasets were combined for developing the DPSON model.

A specially crafted reference Distance-Assisted Location Embedding (DALE) approach was utilized to provide accurate spatial scaling and embedding of the locations within the DPSON network. The approach utilizes cosine and sine transformations of latitude and longitude, combined with a sine transformation of the distance from a reference point, to create suitable spatial embeddings for the network. The model architecture comprises two parameterized networks: (1) the Branch Network and (2) the Trunk Network. The Branch Network is responsible for embedding the pollutant data observed by GMS along with the static spatial covariates of the corresponding locations and their DALE. The Trunk network uses the DALE of unsampled locations, their static spatial covariates to estimate the pollutant concentration at those locations. The DPSON network’s reconstruction error (i.e.: Trunk network output) on the CPCB locations were considered for checking the model capability. The DPSON model was eventually compared with other baseline models. The proposed DPSON model achieved the following performance metrics: for PM_2.5, RMSE of 31.91 µg/m³, MAE of 18.35 µg/m³, and R² of 0.88; for PM₁₀, RMSE of 49.95 µg/m³, MAE of 32.02 µg/m³, and R² of 0.87; for O₃, RMSE of 11.75 µg/m³, MAE of 7.27 µg/m³, and R² of 0.85; and for NO₂, RMSE of 12.05 µg/m³, MAE of 7.67 µg/m³, and R² of 0.88. The proposed DPSON model outperforms all the baseline models for each of the pollutants and is adept at managing various types of spatial covariates, accommodating complex GMS observation distributions, while also providing a computationally efficient framework for the spatial estimation of pollutants.

Acknowledgement: We gratefully acknowledge that this study was supported by the “Indo-German Joint Research Collaboration” grant (DST/INT/DAAD/P-23/2023 (G)) from the Department of Science and Technology (DST), Government of India and DAAD, Germany

How to cite: Mandal, S., Thakur, M., Balamurugan, V., Chen, J., and Roy, A.: A Deep-Pollutant-Spatial-Operator-Network (DPSON) for spatial estimation of PM2.5, PM10, O3 and NO2, case study at Delhi, India, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-20076, https://doi.org/10.5194/egusphere-egu25-20076, 2025.

How to cite: Sánchez-Jiménez, F., Raluy-López, E., Segado-Moreno, L. C., García-Fernández, E., Jiménez-Guerrero, P., and Montávez, J. P.:  Reconstruction, Regionalization, and Prediction of Tropospheric Pollution in the Mediterranean Basin: A Machine Learning Approach, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-19339, https://doi.org/10.5194/egusphere-egu25-19339, 2025.

The rapid decline in air quality across Southeast and Western Pacific Asia is occurring at an accelerated pace due to population growth and industrial development. The region’s Meteorological factors, including the monsoon seasonality, exert a significant influence on air pollution levels, particularly PM_2.5 concentrations. In this study, we employ a statistical modeling approach to derive daily PM_2.5 levels from meteorological parameters in five major polluted cities: Lahore (Pakistan), Delhi (India), Dhaka (Bangladesh), Hanoi (Vietnam), and Shanghai (China). The incorporated meteorological parameters are wind speed, barometric pressure, temperature, and rainfall, which are known to affect air pollution levels from 2020 to 2022. The statistical modeling was based on the comparative analysis of 35 different machine learning (ML) regression techniques with the purpose of selecting the algorithms most efficient for reconstructing and predicting PM_2.5 levels from meteorological variables alone. Specifically, each ML regression model was trained to reconstruct daily PM_2.5 levels in 2020–2021, and then used to reconstruct both missing daily PM2.5 levels in 2020–2021 and forecast the whole of 2022 using only the 2022 meteorological records. The results indicated that most of the daily and seasonal variability in daily PM_2.5 levels could be reconstructed from meteorological conditions. However, the performance of the various ML models (as assessed by Root Mean Square Error tests) exhibited considerable variability. Among the tested models, the Ensembles Boosted Tree ML method demonstrated optimal efficiency during the training period (the first 2 years, 2020 and 2021) and it also was highly efficient in predicting the third year (2022) using only meteorological data. Additionaly, the Trilayer Neural Network ML method was found the most effective at reconstructing the data after 3 years of training and may therefore be preferred to fill in short periods of missing PM_2.5 data. In contrast, our comparative analyses showed that the traditional multi-linear regression models under-performed in both constructing and predicting PM_2.5 data. This study demonstrates the necessity and usefulness of assessing multiple ML regression methodologies for selecting which ones better perform for reconstructing the data of interest (in our case PM_2.5 records) from their hypothesized constructors (in our case meteorological parameters). In particular, this study has highlighted the utility of using ML regression techniques for forecasting air quality and reconstructing missing pollution data, which is crucial for policy-making across South-East and Western-Pacific Asia regions, where only limited pollution monitoring infrastructure are available.

How to cite: Shafi, S. and Scafetta, N.: Optimal machine learning techniques for meteorological modeling of PM2.5 concentration in five major polluted cities of South-East Asia, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-3240, https://doi.org/10.5194/egusphere-egu25-3240, 2025.

In urban areas, pollutant deposition leads to the accumulation of particles on surfaces. These particles include emissions from traffic, heavy metals, micro-plastics and other debris. In the right meteorological conditions, these pollutants can be detached from the surface and resuspended, adversely affecting air quality near the ground and directly exposing city inhabitants. Soil resuspension was found to be a lingering cause of lead exposure for children in several US cities [1]. With the advent of electrical vehicles, indirect sources such as resuspension will become greater contributors to air pollution.

However, quantifying the contribution of resuspension to pollutant exposure remains challenging. Field studies often rely on indirect measurement methods, and wind tunnel experiments use simplified topologies. Computational fluid dynamics tools (CFD) have been employed in only a few studies, with even fewer utilizing the Large Eddy Simulation (LES) framework.

Here we present the coupling of a particle resuspension model to PALM, an open-source LES code. Resuspension is simulated according to the Rock’n’Roll model [2], a probability based approach estimating the resuspension rate from macroscopic properties. The originality of the coupling is that the distribution of adhesion forces is discretized. This allows resuspension to interact with deposition, which is crucial to apply the Rock’n’Roll model to urban air quality studies.

The coupling has been validated against experimental data from [2]. Further validation is planned against more recent datasets, paving the way for the expansion of the Rock’n’Roll model to include effects such as surface roughness [3], flow acceleration [4] and non-spherical particles. We will discuss preliminary results obtained in the case of a street canyon, offering insights into resuspension dynamics in urban environments. Our work aims to provide guidelines to create healthier urban environments and understand how evolving transportation technologies will shape pollutant exposure patterns.

[1] M. Laidlaw, G. Filippelli, Resuspension of urban soils as a persistent source of lead poisoning in children: A review and new directions, Applied Geochemistry, Volume 23, Issue 8,  2021-2039, (2008).

[2] M.W. Reeks, D. Hall, Kinetic models for particle resuspension in turbulent flows: theory and measurement, Journal of Aerosol Science, Volume 32, Issue 1, 1-31, (2001).

[3] S. Peillon et al., Adhesion forces of radioactive particles measured by the Aerodynamic Method–Validation with Atomic Force Microscopy and comparison with adhesion models, Journal of Aerosol Science, Volume 165, (2022).

[4] C. Cazes, Resuspension of microparticles in the air induced by transient events in the flow, experimental approach, Ecole nationale supérieure Mines-Telecom Atlantique, (2023)

How to cite: Bourgin, V., Sellam, M., and Feiz, A.: Implementation of a particle resuspension model in a Large Eddy Simulation code, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-17198, https://doi.org/10.5194/egusphere-egu25-17198, 2025.

To help guide advisories and various societal decision processes aimed at reducing humanity’s detrimental exposure and risks tied to poor air quality, NOAA predicts ozone (O3), fine particulate matter (PM2.5), and other harmful pollutants daily. Unfortunately, air quality forecasts still suffer from errors emanating from the driving datasets, inaccurate emissions, and an incomplete understanding of air quality processes. With increasingly intense western North American wildfires producing expansive and harmful smoke plumes that impact millions of people downstream, it is important to improve predictions. This work aims to design a dynamical ensemble based on the NOAA’s Online Community Multiscale Air Quality (Online CMAQ) embedded within the UFS. The ensemble is based on perturbations of (a) meteorological and chemical initial and lateral boundary conditions, (b) anthropogenic, biogenic, and biomass burning emissions, (c) secondary organic aerosol response to temperature changes and solubility of semi-volatile organic compounds (SVOCs), and (d) removal processes including the hygroscopicity of aerosols and dry deposition velocities of O3, precursors, and SVOCs. Such a perturbation strategy leads to >50 ensemble members. In this presentation, the ensemble is evaluated against AirNOW observations of O3 and PM2.5 in the summer of 2020 when historic western US wildfires generated extensive smoke plumes. The ensemble validation and analysis of the uncertainty will be the central focus of this presentation. The project's ultimate goal is to develop down-selection techniques with calibration to reduce the ensemble size to ~10 members such that the majority of skill and ensemble quality is retained. This will provide a cost-effective air quality ensemble for NOAA’s operational air quality forecasting.

How to cite: Alessandrini, S., Kumar, R., Rozoff, C., Lee, J. A., McCarthy, P., and Tang, W.: A dynamical ensemble approach to characterizing uncertainties in the prediction of air quality downstream of massive Western US wildfires in 2020, EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-14560, https://doi.org/10.5194/egusphere-egu25-14560, 2025.

A new long-term emission inventory called the Inversed Emission Inventory for Chinese Air Quality (CAQIEI) was developed in this study by assimilating surface observations from the China National Environmental Monitoring Centre (CNEMC) using an ensemble Kalman filter (EnKF) and the Nested Air Quality Prediction Modeling System. This inventory contains the constrained monthly emissions of NOx , SO2 , CO, primary PM2.5, primary PM10, and non-methane volatile organic compounds (NMVOCs) in China from 2013 to 2020, with a horizontal resolution of 15 km × 15 km. This paper documents detailed descriptions of the assimilation system and the evaluation results for the emission inventory. The results suggest that CAQIEI can effectively reduce the biases in the a priori emission inventory, with the normalized mean biases ranging from −9.1 % to 9.5 % in the a posteriori simulation, which are significantly reduced from the biases in the a priori simulations (−45.6 % to 93.8 %). The calculated root-mean-square errors (RMSEs) and correlation coefficients were also improved from the a priori simulations, demonstrating good performance of the data assimilation system. Based on CAQIEI, the total emissions from 2015 to 2020  decreased by 54.1 % for SO2, 44.4 % for PM2.5, 33.6 % for PM10, 35.7 % for CO, and 15.1 % for NOx but increased by 21.0 % for NMVOCs. It is also estimated that the emission reductions were larger during 2018–2020 (from −26.6 % to −4.5 %) than during 2015–2017  (from −23.8 % to 27.6 %) for most of the species. In particular, the total Chinese NOx and NMVOC emissions were shown to increase during 2015–2017, especially over the Fenwei Plain area (FW), where the emissions of particulate matter (PM) also increased. The situation changed during 2018–2020, when the upward trends were contained and reversed to downward trends for the total emissions of both NOx and NMVOCs and the PM emissions over FW. This suggests that the emission control policies may be improved in the 2018–2020 action plan. We also compared CAQIEI with other air pollutant emission inventories in China. CAQIEI suggested higher CO emissions in China, with CO emissions estimated by CAQIEI (426.8 Tg) being more than twice the amounts in previous inventories (120.7–237.7 Tg). CAQIEI suggested higher NMVOC emissions than previous emission inventories by about 30.4 %–81.4 % over the North China Plain (NCP) but suggested lower NMVOC emissions by about 27.6 %–0.0 % over southeastern China (SE). CAQIEI suggested lower emission reduction rates during 2015–2018 than previous emission inventories for most species, except for CO. In particular, China’s NMVOC emissions were shown to have increased by 26.6 % from 2015 to 2018, especially over NCP (by 38.0 %), northeastern China (by 38.3 %), and central China (60.0 %). These results provide us with new insights into the complex variations in air pollutant emissions in China during two recent clean-air actions. All of the datasets are available at https://doi.org/10.57760/sciencedb.13151.

How to cite: Tang, X., Kong, L., Wang, Z., Zhu, J., Li, J., Wu, H., Wu, Q., Chen, H., Zhu, L., Wang, W., Liu, B., Wang, Q., Chen, D., Pan, Y., Li, J., Wu, L., and Carmichael, G.: Changes in air pollutant emissions in China during two clean-air action periods derived from the newly developed Inversed Emission Inventory for Chinese Air Quality (CAQIEI) , EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-20377, https://doi.org/10.5194/egusphere-egu25-20377, 2025.