Air pollution has been broadly acknowledged as a significant contributor to various health issues (Brunekreef and Holgate, 2002; Makri and Stilianakis, 2008). Physiologically, air pollution has been found to be associated with respiratory diseases, cardiovascular problems, and even adverse effects on the nervous system. The inhalation of pollutants such as particulate matter and toxic gases can lead to lung inflammation, reduced lung function, and increased susceptibility to respiratory infections (Franklin et al., 2015; Pryor et al., 2022). Prolonged exposure to air pollution has also been linked to an increased risk of developing conditions like asthma, bronchitis, and even lung cancer. Psychologically, air pollution has been unveiled to be correlated to mental health issues such as stress, anxiety, and depression (Bakolis et al., 2021; Buoli et al., 2018). Living in areas with high levels of pollution can contribute to exacerbated feelings of helplessness and frustration among residents. Exposure to air pollution can also impact cognitive function and overall well-being, affecting productivity and quality of life (Boudier et al., 2022; Liu et al., 2023; Thompson et al., 2023).

The health implications of air pollution exhibit varying degrees of impact contingent upon several factors, including pollution levels within different geographical locations (Habibi et al., 2017; Qin et al., 2020) and temporal dimensions including times of the day, days of the week, and seasons of the year (Chen et al., 2015; Vecchi et al., 2007). Specific population groups, including children, the elderly, and individuals with pre-existing health conditions, are also more susceptible to the detrimental effects of air pollution (Bentayeb et al., 2012; Goldizen et al., 2016). Furthermore, communities located in high-density urban centers or close to industrial regions frequently face an unequal distribution of pollution-related health risks owing to elevated exposure levels within these environments (Dai et al., 2013; Popescu, 2016).

Conventionally, the level of air pollution has been quantified by different factors, including ambient respirable suspended particulates (RSPs) such as PM2.5, and PM10, as well as pollutants like sulfur dioxide (SO2), nitrogen oxides (NOx), and carbon monoxide (CO). While pollutants like SO2, NOx, and CO are commonly produced from vehicular engine combustion and exhaust emissions, RSPs, especially PM2.5, emerge as a critical component influencing air pollution levels due to their minuscule size and capacity to penetrate deep into the respiratory system, posing severe health risks. PM2.5 can originate from diverse sources including vehicular traffic and industrial activities (Andrade et al., 2012; Dai et al., 2013; Lin et al., 2020) and has been widely recognized for its association with adverse psychological and physiological health effects (Feng et al., 2016; Sharma et al., 2020). In terms of physiological health, PM2.5 primarily targets the lungs, causing airway inflammation, reduced lung function, and increasing susceptibility to asthma, chronic obstructive pulmonary disease, and infections (Habre et al., 2014; Huang et al., 2019; Yang et al., 2020). PM2.5 induces oxidative stress and inflammation, leading to cellular damage and impaired immune responses, which exacerbate respiratory diseases (Gualtieri et al., 2011; Zhao et al., 2024). Simultaneously, PM2.5 affects cardiovascular health by hampering cardiac autonomic functions, decreasing heart rate variability, and elevating the risks of hypertension, myocardial infarction and atherosclerosis (Lin et al., 2017; Madrigano et al., 2013; Tian et al., 2021). In terms of psychological health, PM2.5 exposure has been linked to adverse mental health outcomes, including depression and stress-related conditions, due to its impact on pollution perception (Li et al., 2021; Roberts et al., 2019). These multifaceted physiological and psychological impacts underscore the adverse health effects posed by PM2.5 air pollution.

Existing measurements of air pollution have primarily relied on fixed monitoring stations located across cities. For instance, in Hong Kong, the Environmental Protection Department operates 18 fixed general and roadside monitoring stations, while in Singapore, the National Environment Agency maintains 22 fixed air quality monitoring stations. However, the geographical distribution of these fixed stations is sparse, failing to capture the intricate nuances of air pollution exposure at a fine spatial scale within smaller neighborhood units. This sparsity issue becomes particularly evident in districts with vast land areas, such as Tuen Mun District in Hong Kong, which houses just one fixed monitoring station. This inadequacy makes it impossible to uncover the intra-district air pollution exposure inequality, highlighting the lack of insight into variations in pollution levels among different neighborhood units within the same district.

The advent of portable environmental sensors equipped with Global Positioning System (GPS) technology represents a paradigm shift in exposure assessment, moving beyond the limitations of sparse, static monitoring networks. While fixed stations provide valuable data, they are often limited in spatial resolution and fail to capture the hyperlocal variations in pollution exposure (Russell et al., 2024; Wang et al., 2023a,b; Yuan et al., 2024). Mobile sensing technology directly addresses this key gap by enabling high-resolution and individual-level data collection. This method facilitates real-time monitoring of individual exposure as individuals move, delineating dynamic exposure patterns in micro-environments (e.g., traffic corridors, parks, residential areas) that are challenging for static monitoring systems to capture (Geng et al., 2022; Ma et al., 2021a; Song and Kwan, 2023; Song et al., 2024; Wang et al., 2024). These portable devices, easily transported by individuals or mounted on vehicles, generate rich spatiotemporal datasets that reveal pollution distributions across fine-grained neighborhood units and multiple temporal scales, including daytime/nighttime, weekday/weekend, and across seasons (Chatzidiakou et al., 2022; Xu et al., 2022). By integrating precise GPS tracking with portable sensor devices, this approach provides a nuanced understanding of real-time fine-grained exposure levels, effectively quantifying the exposure heterogeneity that traditional methods often miss (Park, 2022; Wang et al., 2021). Consequently, mobile sensing does not merely offer more data points; it substantially facilitates the investigation of the complex interactions between human mobility, urban infrastructure, and environmental exposure, thereby providing critical insights into spatial and temporal variations at a granular spatiotemporal scale previously unattainable.

Many existing studies have employed real-time mobile sensing to enhance the accuracy of spatial and temporal air pollution exposure measurements. For example, Zhao et al. (2021) equipped electric vehicles in a ride-hailing fleet in Beijing, China with low-cost sensors to collect real-time and spatial-resolved data on PM2.5 concentrations and found that concentrations exhibited significant fine-scale spatiotemporal variability, with over 20 % of adjacent 1 km grids differing by more than 10 μg/m3 and a persistent pollution hotspot in the southeastern part of the city. Kaivonen and Ngai (2020) deployed wireless sensors on public transport vehicles in Uppsala, Sweden, and unveiled localized pollution peaks that were likely influenced by the bus's own systems, which also showed a slight correlation with temperature. Zhang and Woo (2020) examined real-time air quality patterns in Songdo, South Korea, by placing IoT sensor devices on patrolling vehicles and revealed a distinct spatial pattern with higher pollution concentrations near factories in the northeastern area and lower levels in the central green and residential zones. Ma et al. (2021b) recruited 117 residents in the Meiheyuan residential community in Beijing, equipping them with portable air pollutant sensors over a three-month period during the winter of 2017–2018 to measure real-time exposure They discovered that spatiotemporal exposure varied significantly by mode of transportation, with subway travel and walking recording the highest real-time PM2.5 concentrations, while car travel had the lowest. They also found that travel satisfaction for walking and cycling peaked in the evening (19:00–21:00) despite higher pollution exposures. Park et al. (2023) recruited 44 participants in eastern North Carolina in the fall of 2021 to conduct personal exposure monitoring. They found that PM2.5 exposure exhibited significant spatiotemporal heterogeneity, with peak concentrations occurring during specific indoor activities. Meanwhile, social and religious venues showed the highest mean exposure, and transportation generally yielded the lowest exposure despite occasional spikes during traffic idling or neighborhood burning events.

Despite the contributions made by existing literature utilizing real-time mobile sensors to advance spatiotemporal assessment of air pollution exposure, several major empirical and methodological gaps remain. First, while existing studies have focused on examining individual-level exposure patterns or hotspot identification, few have managed to translate these fine-grained data into a scalable methodology for population-level exposure modeling that adequately accounts for microenvironmental heterogeneity. Second, while past studies have predominantly quantified exposure at a single spatial scale (e.g., a community) or a limited temporal scale (e.g., daytime), few studies have integrated an analysis across multiple spatiotemporal scales to provide a comprehensive view of exposure dynamics. Finally, although existing research has documented inequalities across coarser administrative units like districts, the measurement of inequality at a finer intra-district scale and its manifestation across different times of day and year remains particularly underdeveloped. Collectively, these gaps have limited an accurate and dynamic understanding of how air pollution exposure is inequitably distributed across urban populations.

The objective of this study is to contribute to addressing these knowledge gaps by conducting a spatiotemporal assessment of air pollution exposure distribution and inequality using fine-grained data collected by GPS-enabled portable sensors in Hong Kong. Our proposed methodological framework integrates rigorous machine learning-enhanced data calibration with population-weighted exposure and Gini coefficient analysis at the fine-grained neighborhood level. Specifically, this research aims to translate the air pollution exposure measured by individual residents into the identification of population-level spatiotemporal exposure patterns. We intend to analyze temporal variations in air pollution exposure distribution across different regions and districts at a fine neighborhood unit level within Hong Kong, considering multiple temporal levels, including daytime and nighttime, weekdays and weekends, and the four seasons. Additionally, we aim to uncover intra-district air pollution exposure inequality by examining variations in air pollution exposure across fine neighborhood units within the same district across the aforementioned temporal dimensions. The findings and insights generated by this research can guide policymakers and urban planners in developing accurate and targeted interventions to address air pollution exposure injustice and improve overall air quality.

The subsequent sections of this article are organized as follows. The second section introduces the datasets and methods employed in this research. The third section presents the results and findings derived from the analysis. The fourth section delves into the major implications, policy recommendations, and limitations based on the study's outcomes. The fifth and final section concludes the article by highlighting the key takeaways from this research.