University of the Aegean
Synergy of machine learning models, statistical analysis and geoinformatic tools to assess population exposure to air pollution and related health impacts
Abstract
dc:descriptionAs growing environmental pressures challenge urban resilience, sustainability, and human well-being, the availability of high-resolution geospatial information, down to the intra-urban scale, remains a critical gap for effective, tailored decision-making. Addressing this gap in the field of air quality and its health impacts, the central objective of this PhD study is the development of methodologies for the spatial representation of atmospheric emissions, pollutant concentrations, and population exposure at very high resolution, down to several hundred meters. Additionally, based on the developed methodologies, the relationships between air pollution—its levels and spatiotemporal dynamics—and health outcomes, in conjunction with relevant socio-economic factors, are investigated. Specifically, a methodology for creating high-resolution emission inventories across Europe, starting from the Copernicus Atmosphere Monitoring Service (CAMS), at a base resolution of 0.1° × 0.05°, is developed. This top-down approach, built upon the CAMS-REG-AP database for anthropogenic emissions, generates value-added products for road transport emissions (line sources), industrial point-source emissions, and area sources, redistributed to a 1 km2 grid annually. The spatial disaggregation leveraged publicly available, high-resolution datasets tailored to each source category, including population density from the Global Human Settlement (GHS), land-use classifications from the Copernicus Land Monitoring Service (CLC 2018), the OpenStreetMap (OSM) road network, and marine traffic density from the European Marine Observation and Data Network (EMODNet). This methodology is conducted through UrbEm, a geospatial tool that enables the reliable and scalable generation of high-resolution emission inventories for any European urban center or region. Within the scope of this PhD study, the methodology is implemented and evaluated in Athens, with detailed pollutant—and sector-specific assessments. In parallel, annual emission inventories are developed for 12 European cities with varying geographic and meteorological characteristics, allowing comparative analysis of the methodology’s performance in capturing urban-scale pollution. Athens’ high-resolution inventory is integrated into an urban-scale air pollution Chemistry Transport Model capable of simulating the transport and physicochemical transformation of pollutants at the street level. The model's outputs are evaluated against ground-based observations and satellite-derived data, offering insights into both horizontal and vertical pollutant distributions. Furthermore, the advancement from annual to monthly emission high-resolution inventories is a significant enhancement. Combining the CAMS-REG database with CAMS-TEMPO spatiotemporal profiles results in the first high-resolution monthly emissions inventory for PM2.5 and NOx in Greece, a dataset suitable for developing long-term exposure estimates across the whole country for the first time. The lack of high-resolution, long-term PM2.5 observations in Greece hinders the development of spatial models, crucial for providing representative exposure estimates to health studies. This PhD presents contributions and improvements along the whole chain of a spatial modeling approach to address this gap and assess PM2.5 spatial variability in Greece, by integrating in-situ observations, meteorological parameters, emission inventories, and satellite AOD data, among others. A high-resolution (1 km²) gridded dataset of PM2.5 concentrations across Greece from 2015 to 2022 has been developed, and seven statistical, machine learning, and hybrid models have been evaluated under different prediction scenarios. The Random Forest (RF) model demonstrates superior performance (R² = 0.73, MAE = 2.2 μg m-3), validated against ground-based measurements. Winter months consistently show the highest PM2.5 levels, averaging 16.8 μg m-3, across the domain, primarily due to residential biomass burning (BB) and limited atmospheric dispersion. Summer months have the lowest concentrations, averaging 10.3 μg m-3, while substantial decreases nationwide were observed during the 2020 COVID-19 lockdown. Population exposure analysis indicates that the entire Greek population was exposed to long-term PM2.5 concentrations exceeding the WHO air quality guideline (AQG) of 5 μg m-3. Moreover, the dataset reveals elevated PM2.5 levels across several regions of mainland Greece. Notably, 70% to 90% of the population experience levels exceeding 10 μg m-3 in Central and Northern regions of continental Greece like Thessaly, Central Macedonia, and Ioannina. The Ioannina region, which is severely impacted by residential BB, recorded pollution levels up to five times the WHO AQG highlighting the urgent need for targeted interventions. The high-resolution RF model’s superior performance for monthly average concentrations, compared to the CAMS European air quality reanalysis dataset, renders it a reliable tool for long-term PM2.5 assessment in Greece that can support air quality management and health studies. In addition, the RF‘s high-resolution (1 km2) dataset of PM2.5 concentrations is integrated with health data and socioeconomic parameters. The aim is to support evidence-based and targeted decision-making across all levels of administrative governance. A cause-specific mortality analysis is conducted to further explore the impacts of air pollution, as well as other environmental and social stressors, on health.National-level assessments of the health effects of air pollution in Greece remain limited. While large-scale cohort studies have advanced our understanding of these impacts, a key constraint persists: the lack of publicly available, high-resolution spatiotemporal mortality data. To address this gap, the present study investigates PM2.5-related mortality from 2015 to 2020, considering the combined influence of environmental and socioeconomic stressors across six cause-specific mortality categories, 333 municipalities, and 13 administrative regions. Using a Generalized Linear Mixed Model (GLMM) with both a Poisson and a Negative Binomial distribution, we identify significant associations between PM2.5 exposure and increased mortality from respiratory (RESP), cardiovascular (CVD), lung cancer (LUNG), and genitourinary diseases (GU) diseases, highlighting the substantial public health burden of air pollution. In contrast, mental and behavioral disorders (MBD) and diabetes mellitus (DM) show weaker associations, suggesting the influence of other unmeasured factors. PM2.5 exposure is linked to an approximate 4% increase in mortality for RESP, CVD, and LUNG. The interaction of air pollution with temperature, relative humidity, and poverty influence mortality patterns, suggesting complex relationships between environmental and socioeconomic determinants. Spatial disparities are evident, with insular regions facing the highest mortality risks, while Attica exhibits up to 70% lower risks, likely due to better healthcare infrastructure. Socioeconomic factors amplify disparities, with unemployment increasing risks up to eightfold, while lower educational attainment is surprisingly linked to reduced mortality. Occupational groups in high-stress jobs show elevated risks, underscoring the role of occupational exposures and psychosocial stress.
Degree
thesis:*- Grantor dc:publisher
- University of the Aegean
- Year dc:date
- 2025
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Kakouri, Anastasia
- Κακούρη, Αναστασία
Subjects
dc:subject × 26- Ατμόσφαιρα, Ποιότητα
- Αέρια ρύπανση
- Αιωρούμενα σωματίδια
- Παράγοντες θνησιμότητας
- Τεχνιτή νοημοσύνη
- Μηχανική μάθηση
- 'Εξυπνες πόλεις
- Εκπομπές αέριων ρύπων
- Ατμοσφαιρικά αερολύματα
- Atmospheric aerosols
- Air quality
- Air pollution
- Particulate matter
- Mortality Greece
- Artificial intelligence
- Machine learning
- Smart cities
- Emission inventories
- Γεωεπιστήμες και Επιστήμες Περιβάλλοντος
- Φυσικές Επιστήμες
- Μετεωρολογία και Επιστήμες της ατμόσφαιρας
- Επιστήμες περιβάλλοντος
- Earth and Related Environmental Sciences
- Natural Sciences
- Meteorology and Atmospheric Sciences
- Environmental Sciences
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/60879