Event-Based Tracking of Spatiotemporally Contiguous PM_2.5 Pollution Events in China

Abstract

PM_2.5 pollution events evolve continuously through spatiotemporal diffusion. However, their three-dimensional spatiotemporal variation characteristics are often overlooked, and the interactions among key characteristics (e.g., duration, maximum concentration) have not yet been systematically analyzed. This study established a three-dimensional (longitude, latitude, and time) spatiotemporal framework for identifying contiguous PM_2.5 pollution events based on the high-resolution ChinaHighAirPollutants (CHAP) dataset (1 km spatial and 1-day temporal resolution). The framework applied the meteorological event tracking algorithm (i.e., the Forward-in-Time method) to track PM_2.5 pollution events. Based on this framework, we systematically tracked and characterized the spatiotemporal evolution of PM_2.5 events across China from 2013 to 2021, quantified the relationships among key event characteristics, and tracked their transport pathways. The results show that: (1) The combination of the FiT algorithm and CHAP dataset enables effective tracking and identification of the three-dimensional spatiotemporal evolution of PM_2.5 pollution events across China. (2) Event PM_2.5 totals, average totals per event and pollution events exhibit a distinct right-inclined “T”-shaped pattern, with hotspots located in Xinjiang, the Beijing-Tianjin-Hebei (BTH) region, Shandong, and Henan, where annual event frequency exceeds 15. (3) Event PM_2.5 totals show strong correlations with average duration per event and average maximum concentration per event, particularly in heavily polluted areas where the Pearson correlation coefficient is close to 1. (4) PM_2.5 pollution events are mainly characterized by short durations of 1 day or 2–3 days, accounting for over 80% of occurrences. Long-duration events are mostly concentrated in areas with severe pollution problems, and their persistence is closely linked to spatial coverage, terrain barrier effects, and meteorological conditions. (5) PM_2.5 pollution events consistently exhibit a west-to-east transport pattern. Short-duration events propagate slower across the inland northwest, whereas long-duration events show a pronounced increase in meridional transport speeds along the eastern coastal areas. This study elucidates the continuous spatiotemporal evolution and intrinsic drivers of PM_2.5 pollution events, offering scientific insights to support air quality improvement and the development of targeted management strategies.

1. Introduction

Spatiotemporal continuum characteristics of pollution events (e.g., event duration, average concentration, and maximum concentration) constitute critical determinants for environmental health risk quantification and emergency response protocol activation. These characteristics exhibit pronounced spatial heterogeneity and complex temporal evolution, as shaped by the interplay of regional emission sources, meteorological conditions, and atmospheric dynamics [1,2]. Consequently, understanding the spatiotemporal continuum characteristics of pollution events is essential for improving air quality and formulating effective management strategies. PM_2.5, defined as fine particulate matter with an aerodynamic diameter of ≤2.5 μm, is a key air pollutant known for its long-range transport, complex chemical composition, and alveolar deposition potential [3]. It is classified as a Group 1 carcinogen by the World Health Organization [4]. While the spatiotemporal distribution of PM_2.5 pollution is a well-established research domain, the continuum characteristics of PM_2.5 pollution events remain poorly characterized.

Rapid industrialization and urbanization have intensified PM_2.5 pollution. Many countries have implemented stringent emission regulations to mitigate localized contamination. However, PM_2.5 pollution continues to confront challenges such as transboundary transport and secondary pollution [5,6,7]. For instance, biomass burning emissions in Southeast Asia drive transboundary PM_2.5 transport, with significant impacts documented in Southwest China [8]; transcontinental dispersion delivers North American wildfire smoke to the eastern seaboard [9,10]. In this context, a growing consensus among scholars underscores the need to investigate transport pathways of PM_2.5 pollution events and to examine critical characteristics such as maximum concentration and event duration [11,12]. These studies have primarily focused on the multidimensional characterization of pollution events, including spatiotemporal pattern analyses, development of emission source inversion models, and investigations of how climate change and anthropogenic activities influence pollution processes. Amidst continuously strengthening air quality controls (e.g., “Blue Skies” type campaigns in China), annual mean PM_2.5 concentrations may stabilize or decline, while high-concentration, long-duration pollution events recur alarmingly frequently [13,14,15]. The intrinsic linkages among these event characteristics and their underlying mechanisms represent a critical research focus.

Conventional pollution event analysis relies on ground monitoring networks, while their inherent spatial sparsity prevents comprehensive characterization of pollution diffusion’s continuous spatial dynamics [16,17]. To overcome the spatial sparsity of ground monitoring networks, numerous studies employ spatial interpolation techniques to characterize the spatial variability of pollution events [18,19]. For example, to overcome the limitations of traditional spatial interpolation techniques, Ma et al. [20] introduced a novel spatial interpolation technique that leverages the Long Short-Term Memory neural network to produce the spatial distribution of air pollutant concentrations. Meng et al. [21] predicted PM_2.5 concentration changes in central and eastern China by integrating meteorological and air quality data via inverse distance interpolation and evaluating various machine learning models. They highlighted the effectiveness of spatial interpolation in capturing PM_2.5 trends. Moreover, the spatial sparsity of ground monitoring networks can introduce systematic representation biases [18], particularly in regions with complex terrain and atmospheric conditions [22].

Advancements in satellite remote sensing have greatly enhanced the large-scale monitoring and analysis of PM_2.5 pollution, with the fusion of multi-source remote sensing and ground-based observations markedly improving retrieval accuracy [23]. Yan et al. [24] utilized high-temporal-resolution satellite data to derive PM_2.5 concentrations and applied deep learning models for estimation, effectively overcoming the limitations of sparse ground monitoring networks. Kong et al. [25] provided a review of imaging lidar methods for detecting pollution sources, emphasizing the current applications and future potential of remote sensing data and technologies in air quality analysis. Furthermore, numerous studies have employed numerical models, including WRF-Chem, to simulate PM_2.5 spatial variability and transport dynamics [26,27,28,29]. Nonetheless, numerical modeling approaches frequently encounter challenges: substantial data requirements, computationally intensive simulations, and limited applicability beyond localized case studies. Although remote sensing datasets resolve macroscale PM_2.5 spatial heterogeneity, their inherent resolution constraints impede microscale pollution characterization. This limitation necessitates the synergistic integration of high-resolution ground monitoring networks for fine-scale spatial differentiation analysis [30]. For instance, the ChinaHighAirPollutants (CHAP) dataset integrates satellite remote sensing, ground observations, and model simulations to provide continuous, gap-free PM_2.5 data at 1 km spatial and daily temporal resolutions [31].

PM_2.5 pollution events are typically identified via ground stations or designated grid cells; however, their inherent spatiotemporal continuum characteristics remain fundamentally unaccounted for in conventional detection frameworks. To capture these spatiotemporal continuum characteristics, numerous methodologies and algorithms have been developed for identifying pollution events. Various methodologies—including FLEXPART [32], random forest-based machine learning techniques [33], Graph Convolutional Network-driven deep learning methods [34,35], Hermitized Models [36], and Gaussian atmospheric dispersion models [37]—have been implemented for pollution event identification and modeling, demonstrating proven effectiveness across diverse regions. Hart et al. [38] reviewed some methodologies for detecting air pollution events, critically assessed the strengths and limitations of each approach, and emphasized the critical role of pollution data in event tracking. Moreover, certain algorithms originally developed for tracking and identifying other types of events offer potential solutions for detecting and tracking air pollution events. For example, Object-based spatiotemporal analysis methods, such as the Forward-in-Time (FiT) algorithm, offer novel approaches to track air pollution events. Originally developed for meteorological event tracking (e.g., rainfall systems), the FiT algorithm identifies contiguous ‘objects’ and establishes temporal linkages to quantify event duration, spatial extent, and transport trajectories [39,40]. Despite its validated efficacy, the application of the FiT algorithm in PM_2.5 research remains limited, particularly in regions with complex terrain and heterogeneous emissions like China.

Despite substantial advances in understanding PM_2.5 pollution events, critical scientific questions remain unresolved, particularly regarding how to elucidate the continuous spatiotemporal evolution and underlying drivers of PM_2.5 pollution events from a high-resolution remote sensing dataset perspective, and specifically, how to incorporate dynamic behaviors including event splitting, merging, and migration. Furthermore, the interaction mechanisms among key characteristics of pollution events—such as event duration, maximum concentration, and event PM_2.5 totals—are not well understood, which leads to mitigation strategies that primarily focus on ‘end-of-pipe’ control rather than interrupting the pollution formation process. To address these challenges, this study investigates the spatiotemporal continuum characteristics of PM_2.5 pollution events in China from 2013 to 2021, utilizing the CHAP dataset and the FiT algorithm. The specific objectives are to: (i) characterize the spatiotemporal continuum characteristics of PM_2.5 pollution events, encompassing frequency, duration, intensity, and spatial coverage; (ii) quantify correlations among key event attributes (e.g., event duration, maximum concentration, event PM_2.5 totals) and identify dominant drivers; (iii) track the movement trajectories of pollution events to support refined pollution control strategies.

2. Methodology

2.1. Forward-in-Time Algorithm

Spatiotemporally contiguous air pollution events were detected from gridded daily average PM_2.5 data using the FiT algorithm. The FiT algorithm [39,40], which is developed based on the Object-based Diagnostic Evaluation method [41], is an object-based analysis approach, that is, it identifies different pollution events based on the temporal and spatial distribution of pollution objects. The tracked events are conceptualized as three-dimensional entities spanning the spatial (longitude, latitude) and temporal domains. The algorithm assumes that there is no interaction between spatial objects occurring in separate time intervals. Consequently, each time interval is treated as an independent scenario, and no attempts are made to establish correlations between objects in adjacent intervals. The process of spatiotemporally contiguous air pollution event detection involves two main steps: (1) identifying spatial objects at each time step, (2) establishing spatiotemporal linkages among the objects within the same event. More details of the FiT algorithm can be found in the literature [39,40].

(1)

Identifying spatial objects at each time step

The spatial object identification framework includes three key modules: (i) setting a threshold, (ii) performing convolutional processing, and (iii) identifying spatial objects.

(i) Setting a threshold

This study primarily investigates the continuous spatiotemporal variations of PM_2.5 pollution events. Additionally, to align with the standard stipulated in the “Ambient Air Quality Standards” (GB3095-2012) [42], where a daily average PM_2.5 concentration equal to or exceeding 75 μg/m³ is deemed as PM_2.5 pollution, the PM_2.5 concentration threshold was set at 75 μg/m³. This aligns with the “Ambient Air Quality Standards” (GB3095-2012) [42], which defines a daily average PM_2.5 concentration ≥ 75 μg/m³ as a pollution episode. When the daily average PM_2.5 concentration fell below this threshold, it was assigned a value of zero. Thus, days with PM_2.5 concentrations below this threshold were excluded from the analysis.

(ii) Performing convolutional processing

To obtain more realistic spatial extents for the identified PM_2.5 pollution events at each time step, a spatial convolution was applied to smooth the data. This convolution process involved computing a spatial moving average with a horizontal convolution radius Rs over the original PM_2.5 data. Specifically, the process calculates the mean PM_2.5 concentration for all grid cells within the specified radius Rs:

$$\tilde{Z}={\displaystyle \frac{1}{N_{ell}}}{\displaystyle \sum _{-Rs}^{Rs}{\displaystyle \sum _{-Rs}^{Rs}{Z}_{x,y}}}\times \left\{\begin{array}{c}1;\hspace{1em}if\hspace{1em}inside circle\\ 0;\hspace{1em}if\hspace{1em}outside circle\end{array}\right.$$

(1)

where $N_{ell}$ denotes the total number of grid cells within radius Rs; and $Z_{x,y}$ represents the PM_2.5 concentration at the grid cell with coordinates (x, y). Rs is set to 0.75° for this study, this value was chosen to effectively smooth the data and connect nearby pollution patches into coherent objects without excessively merging distinct features.

(iii) Identifying spatial objects

Following the convolution process described in step (ii), the original PM_2.5 concentration data are smoothed to suppress noisy grid cells. Subsequently, grid cells with PM_2.5 concentrations exceeding the threshold (i.e., ≥75 μg/m³) are identified, while those below the threshold are disregarded. The contiguous area formed by the identified grids is considered an object. Consequently, multiple objects can be identified simultaneously within a single time step. For each identified object, its centroid can be calculated based on its spatial configuration.

(2)

Establishing spatiotemporal linkages among the objects within the same event

During the identification of spatial objects at each time step, only the spatial extents of pollution objects are delineated. To establish spatiotemporal linkages of the objects across different time steps, the FiT algorithm employs a forward-in-time tracking mode. This mode permits a spatially contiguous pollution object to split into two or more discrete objects (i.e., fragmentation of a spatially continuous pollution entity) as well as the merging of multiple discrete objects into a single contiguous entity (i.e., amalgamation of spatially discontinuous entities). Critically, the algorithm operates strictly forward in time and does not incorporate backward tracking.

The FiT algorithm primarily assigns objects to the same pollution event by detecting spatial overlap between pollution objects at the current time step and the preceding time step. Figure 1 illustrates a schematic diagram of this assignment process. If an object at the current time step overlaps with an object from the previous time step, they are assigned to the same event. Similarly, if two distinct objects at the current time step both overlap with a single object from the previous time step, all three are assigned to the same event. It is important to note that the processes of splitting and merging are determined exclusively by the spatial overlap of objects between consecutive time steps. The PM_2.5 concentration values of the constituent grid cells remain unchanged from the original dataset. To identify pollution events within a specific period, tracking begins at the first time step. Therefore, for any identified pollution event, the object at its initial time step exhibits no overlap with objects from the preceding time step, and the object at its final time step exhibits no overlap with objects from the subsequent time step.

The duration of each PM_2.5 pollution event is calculated as the difference (in time steps) between its final and initial occurrences. Given the daily temporal resolution of the PM_2.5 data, the event duration in days is equivalent to the number of time steps it encompasses. For example, an event spanning a single time step has a duration of 1 day, and an event spanning two consecutive time steps has a duration of 2 days.

2.2. Pearson Correlation Coefficient

Pearson correlation coefficient (r) quantifies the relationship between two sets of interval variables. Its formula is expressed as:

$$r={\displaystyle \frac{{\displaystyle {\sum }_{i=1}^n(x_i-\overline{x})}(y_i-\overline{y})}{\sqrt{{{\displaystyle {\sum }_{i=1}^n(x_i-\overline{x})}}^2}\sqrt{{{\displaystyle {\sum }_{i=1}^n(y_i-\overline{y})}}^2}}}$$

A value of r = 0 indicates no linear correlation, while values approaching ±1 signify a strong correlation. The absolute value of r reflects the strength of the relationship. In this study, key characteristics of pollution events—such as frequency, mean concentration, and duration—are inherently interdependent. To quantitatively assess these complex interactions and elucidate their intrinsic relationships, we employed the Pearson correlation coefficient.

3. Study Area and Data

3.1. Study Area

This study investigates the continuous spatiotemporal evolution of PM_2.5 pollution events across China. China spans nearly 50° in latitude from north to south and 62° in longitude from east to west, exhibiting complex and diverse geographical and climatic features. The terrain descends in a three-step staircase pattern from west to east, resulting in dramatic elevation changes (Figure 2). This vast area contains diverse landforms, including mountains, plateaus, hills, basins, plains, and deserts.

Due to its vast geographical extent, China encompasses multiple climate zones: tropical monsoon, subtropical monsoon, temperate monsoon, temperate continental, and plateau climates. These diverse climate regimes result in pronounced spatiotemporal variations in precipitation and temperature. Annual precipitation decreases from the southeastern coast to the northwestern interior, with wet seasons receiving substantially more rainfall than dry seasons. Temperature disparities are equally significant, particularly in winter when northern regions experience severe cold compared to milder conditions in the south.

Regional economic development is highly uneven, shaped by heterogeneous natural resources and geographical constraints. The eastern coastal areas are characterized by advanced industrialization, in stark contrast to the less developed central and western regions, e.g., light manufacturing dominates southern China, while heavy industries prevail in the north. These differences directly impact PM_2.5 emission patterns and pollution distribution.

In recent years, China has prioritized environmental governance by implementing stringent policies to improve air quality. According to the Ministry of Ecology and Environment, the average proportion of “good-air-quality days” in 339 cities increased in 2023, and the annual mean PM_2.5 concentration declined compared to 2019. However, PM_2.5 pollution remains a critical environmental challenge, especially in northern China during winter when coal heating and industrial emissions peak. These regional economic disparities further complicate pollution control efforts; for instance, economically advanced areas face high emissions from concentrated industries, while less developed regions struggle with inadequate pollution mitigation infrastructure.

3.2. Data

This study utilizes PM_2.5 data obtained from the ChinaHighAirPollutants (CHAP) dataset, a high-resolution, high-quality near-surface air pollutant dataset for China, released by the National Earth System Science Data Center (https://www.geodata.cn/data/datadetails.html?dataguid=192965641360576&docid=131, accessed on 14 October 2024). The CHAP dataset is a seamless, long-term, full-coverage, high spatiotemporal resolution, and high-precision remote sensing dataset of ground-level PM_2.5 concentrations over China [31]. It was generated by integrating multi-source satellite remote sensing data. It assimilates extensive datasets including ground-based observations, atmospheric reanalysis, emission inventories, and model simulations. For this study, we employed the CHAP dataset covering China for the period 2013–2021. The dataset features a spatial resolution of 1 km and a temporal resolution of daily, with units of μg/m³. To align with the standard stipulated in the Ambient Air Quality Standards (GB3095–2012) [42], a daily average PM_2.5 concentration equal to or exceeding 75 μg/m³ is deemed as PM_2.5 pollution.

4. Results and Discussion

4.1. Spatiotemporal Continuum Characteristics of PM_2.5 Pollution Events

To investigate the spatiotemporal continuum characteristics of PM_2.5 pollution events across China, we defined a PM_2.5 pollution episode as any period with a 24-h average concentration ≥ 75 μg/m³. We then applied the FiT algorithm to track and identify these pollution events. Six indices (i.e., event PM_2.5 totals per year, PM_2.5 pollution events per year, average totals per event, average duration per event, average maximum concentration per event, average concentration per event) are selected to analyze the spatiotemporal continuum characteristics of PM_2.5 pollution events.

As shown in Figure 3a, the highest event PM_2.5 totals per year were predominantly found in provinces including Xinjiang, the BTH region (i.e., Beijing city, Tianjin city, and Hebei province), Shandong, and Henan. Moderate totals were observed in the northeastern provinces (Liaoning, Jilin, Heilongjiang), as well as in Hubei, Hunan, Guangxi, Anhui, Jiangsu, Shanxi, Shaanxi, and Sichuan, while the remaining provinces recorded relatively lower totals. This spatial pattern aligns with ground-based PM_2.5 air quality monitoring data [43]. The event PM_2.5 totals per year exhibited a distinct right-inclined “T”-shaped spatial distribution pattern. This distribution characteristic is closely linked to regional industrial composition and meteorological conditions [44,45,46]. Specifically, the monsoon corridor stretching from Xinjiang to East China provides dynamic conditions for PM_2.5 dispersion, while regional industrial composition serves as a primary source of PM_2.5 emissions.

Figure 3b reveals that the spatial distribution pattern of the PM_2.5 pollution events per year closely resembles that of the event PM_2.5 totals per year. That is, regions with higher event PM_2.5 totals also experienced a greater number of PM_2.5 pollution events. In southern Xinjiang, PM_2.5 pollution events generally exceeded 15 events/year, with the peak values located in the eastern part of southern Xinjiang, where the frequency exceeded 20 events/year. Provinces like Shandong, BTH, and Henan also had a high number of PM_2.5 pollution events, ranging between 10 and 30 events/year. Other provinces with relatively larger event PM_2.5 totals, primarily Jiangsu, Anhui, and Hubei, had PM_2.5 pollution event frequencies mainly between 7 and 20 events/year. The annual number of PM_2.5 pollution events in other provinces was relatively low, generally less than 10 events/year.

As evident from the spatial characteristics of PM_2.5 pollution events (Figure 3c–f), the spatial distribution patterns of average totals per event, average duration per event, average maximum concentration per event, and average concentration per event closely resembled those of the event PM_2.5 totals per year (Figure 3a) and PM_2.5 pollution events per year (Figure 3b). Therefore, the detailed spatial characteristics of these event-average indices are not reiterated here. It is noteworthy that the blank areas observed in parts of Tibet, Sichuan, and Qinghai resulted from the absence of qualifying pollution events in these regions; specifically, the daily mean PM_2.5 concentrations consistently remained below the 75 μg/m³ threshold, resulting in zero identified pollution events. Consequently, the corresponding characteristic values for these areas could not be calculated.

It is noteworthy that although Xinjiang, the BTH region, Shandong, and Henan all exhibit high frequencies of pollution episodes, their underlying mechanisms may differ substantially. Elevated PM_2.5 events in Xinjiang are largely associated with dust aerosols from the Taklamakan Desert and residential coal combustion for heating during winter. Their formation and persistence are strongly influenced by stagnant meteorological conditions and temperature inversions exacerbated by the basin’s topography. In contrast, pollution episodes in the BTH region and its vicinity primarily originate from intensive anthropogenic emissions related to population density, industrial activities, and transportation. Under unfavorable meteorological conditions—such as stagnant air, high humidity, and thermal inversions—these emissions facilitate rapid chemical transformation and accumulation of secondary particulate matter. Consequently, while these regions similarly display spatial hotspots of high PM_2.5 concentrations, the dominant drivers and physicochemical processes involved necessitate distinct interpretation and management strategies.

Since the Air Pollution Prevention and Control Action Plan was issued in 2013, air quality across China has significantly improved. To investigate the changes in the characteristics of PM_2.5 pollution events following the implementation of this plan, we conducted a comparative analysis of the event-based PM_2.5 air pollution indices between 2013 and 2021. Overall, the fundamental characteristics of PM_2.5 pollution events exhibited a general improvement trend nationwide (Figure 4). However, deterioration was still observed in certain regions compared to 2013. Event PM_2.5 totals per year (Figure 4a) and average concentration per event (Figure 4f) in 2021 showed an overall decrease compared to 2013, except in parts of Inner Mongolia and Xinjiang. The most substantial reduction in Event PM_2.5 totals per year occurred in regions that had the highest totals in 2013, namely the BTH region, Shandong, and Henan. Although Xinjiang also showed a decrease, the magnitude of reduction was relatively smaller than in the aforementioned provinces (Figure 4a). Average concentration per event generally decreased, with the spatial pattern of this reduction being relatively uniform, except for increases observed in parts of Inner Mongolia and Xinjiang (Figure 4f).

It is important to note that the comparative analysis between 2013 and 2021 provides a snapshot of the net change over this period but does not capture the interannual variability. Furthermore, the year 2021 was still subject to the influence of the COVID-19 pandemic, which led to temporary reductions in anthropogenic emissions in certain regions and sectors. While this may have contributed to the observed improvements in some indices, the overall declining trend from 2013 to 2021 is consistent with the long-term effectiveness of China’s air pollution control policies.

The spatial patterns of change for the other four indices (i.e., PM_2.5 pollution events per year, average totals per event, average duration per event, and average maximum concentration per event) between 2013 and 2021 were broadly similar (Figure 4b–e). The spatial patterns of change in the remaining four indices—annual pollution event frequency, average total per event, average duration, and average maximum concentration—from 2013 to 2021 were broadly consistent (Figure 4b–e). While most areas showed improvement, increases were observed in the PM_2.5 pollution events per year, average totals per event, and average duration per event in some regions, including parts of Xinjiang, the BTH region, and Henan (Figure 4b–e).

4.2. Correlation Analysis of the Patiotemporal Continuum Characteristics of PM_2.5 Pollution Events

In air quality management, the spatiotemporal continuum characteristics of PM_2.5 pollution events (e.g., event PM_2.5 totals, event duration, and maximum concentration) have garnered significant attention. Quantifying the relationships among these indices is therefore essential. This study used Pearson correlation analysis to quantify the relationships between key event characteristics, specifically between: event PM_2.5 totals per year and average duration per event, event PM_2.5 totals per year and average maximum concentration per event, event PM_2.5 totals per year and average concentration per event, average maximum concentration per event and average concentration per event. To ensure the statistical robustness of the correlation analysis, we only considered regions where the total number of pollution events during 2013–2021 exceeded five.

As shown in Figure 5, regions with more severe PM_2.5 pollution problems (e.g., Xinjiang, BTH, Shandong) exhibited strong correlations between event PM_2.5 totals and event duration, with Pearson correlation coefficients approaching 1. Conversely, in regions with less severe pollution (e.g., Fujian, Guangdong, Zhejiang), the correlation between these two characteristics was weaker, with coefficients generally below 0.3. In areas with high event PM_2.5 totals, pollution concentrations remained elevated throughout most of the event period, leading to a pronounced increase in total pollutant mass with longer durations (Figure 5a). The correlation between event PM_2.5 totals and average maximum concentration per event exhibited a spatial distribution pattern similar to that between event PM_2.5 totals and event duration. However, the Pearson correlation coefficients in heavily polluted areas were relatively lower, ranging between 0.7 and 1. This indicates that the correlation between event totals and maximum concentration was generally weaker than the correlation between event totals and duration (Figure 5b). The correlation between event PM_2.5 totals and average concentration per event (Figure 5c) was weaker than the correlations between event totals and either duration or maximum concentration. The correlation coefficients for this pair were generally below 0.8, with higher values typically found in regions experiencing higher pollution concentrations and lower values in less polluted regions. Similarly, the correlation between average maximum concentration per event and average concentration per event (Figure 5d) also showed a pattern where coefficients were generally higher (between 0.7 and 1) in more polluted areas. It is worth noting that the Pearson correlation coefficients approaching 1 observed in severe pollution hotspots should be interpreted as evidence of a highly constrained and deterministic relationship in these regions, primarily resulting from a data structure with low internal variability around a strong linear trend.

Comparative analysis of Figure 5a–c reveals that in regions with high event PM_2.5 totals, both event duration and average maximum concentration per event were strongly correlated with event PM_2.5 totals. However, in some high-total regions (e.g., parts of Xinjiang), the correlation coefficient between event PM_2.5 totals and average concentration per event was considerably weaker (0.4 to 0.6). This weaker correlation indicates greater temporal variability in PM_2.5 concentrations during events in these areas, which are often characterized by particularly high maximum concentrations.

4.3. Characteristics of PM_2.5 Pollution Events with Different Durations

Event duration is a key factor for understanding regional PM₂.₅ pollution dynamics and formulating effective control strategies. To analyze the spatial variation patterns of PM_2.5 pollution events across different regions, this study calculated the proportion of pollution events falling into various duration categories relative to the total number of events affecting each 1 km × 1 km grid cell (Figure 6). It is important to note that within a single grid cell, a specific PM_2.5 pollution event is counted only once. However, the same pollution event is calculated separately in each grid cell.

Overall, PM_2.5 pollution events across China were predominantly short-duration events, specifically those lasting 1 day or 2–3 days. In regions with less severe PM_2.5 pollution (e.g., Inner Mongolia, Gansu, Qinghai, Guangdong, Jiangxi, Fujian, Zhejiang), the proportion of 1-day duration events was generally greater than 0.6 (Figure 6a), significantly higher than in regions with more severe pollution (e.g., BTH, Shandong, Shanxi). The proportion of events lasting 2–3 days was generally above 0.2 (excluding provinces like Tibet and Sichuan with virtually no qualifying events), and its spatial distribution was relatively uniform (Figure 6b). This proportion remained fairly consistent between areas of higher and lower pollution severity, exhibiting significantly less spatial variation than the proportion of 1-day events. In less polluted regions, the combined proportion of 1-day and 2–3 day events typically exceeded 0.9. Conversely, in some heavily polluted areas (e.g., BTH), the combined proportion of events lasting less than 3 days was below 0.1.

Regions experiencing more severe PM_2.5 pollution generally exhibited higher proportions of long-duration events. As shown in Figure 6c–f, the spatial extent affected by long-duration PM_2.5 events progressively decreased as the event duration increased. Long-duration events were primarily concentrated in regions with more prominent pollution problems. Events with durations of 4–5 days and 6–8 days were still distributed across various provinces, but their proportions were generally less than 0.3 and 0.2, respectively. In regions with relatively higher PM_2.5 pollution levels (e.g., BTH, Shandong, Shanxi, and over ten other provinces), the proportions of events lasting 9–12 days and >12 days ranged approximately between 0.05–0.15 and 0.10–0.15, respectively. Notably, the geographical area experiencing events lasting longer than 12 days was relatively limited.

4.4. Spatial Variation Patterns in Transport Pathways of PM_2.5 Pollution Events

The transport pathways of PM_2.5 pollution are critical for controlling regional atmospheric compound pollution. This study analyzed the movement and spatial variation patterns of PM_2.5 pollution events with different durations across China, taking the transport pathways observed in 2021 as an illustrative example. Overall, the duration of PM_2.5 pollution events in China is intrinsically linked to their spatial coverage, terrain barrier effects, and meteorological conditions.

The duration of PM_2.5 pollution events significantly affects their spatial transport range. Events with shorter durations (2–3 days) were predominantly concentrated in Northwest to North China (Figure 7a), exhibiting local-scale spatial coverage. These events typically result from rapid accumulation of regional high-emission sources (e.g., coal heating, industrial emissions) under weak wind conditions [47]. Enclosed terrain in northwestern regions (e.g., Xinjiang) further exacerbates pollution stagnation, increasing the frequency of short-duration events. Events lasting 4–5 days and 6–8 days showed progressively eastward-extending trajectories (Figure 7b,c). These were frequently associated with persistent stagnant weather, where reduced boundary layer height and temperature inversions limited vertical dispersion, while horizontal transport was governed by weak westerly disturbances. Longer-duration events demonstrated trans-regional transport characteristics, with trajectories reaching East and Southeast China. Such events are often coupled with large-scale weather systems (e.g., the Siberian High in winter or subtropical high margins in summer), facilitating long-range transport via the westerly jet stream or monsoon circulation.

As shown in Figure 7, all PM_2.5 pollution events exhibited a consistent west-to-east transport trend, aligning with China’s prevailing winter westerlies and cold air outbreak paths. Short-duration events in inland Northwest China (e.g., Xinjiang, Gansu) showed slower propagation (<5° E/day meridional displacement), while long-duration events accelerated significantly along the eastern coast (>8° E/day), likely due to periodic intensification of the East Asian trough and jet stream acceleration. Additionally, the high trajectory density observed east of 110.3° E indicates that the superposition of intensive anthropogenic emissions in the eastern economic corridor with regionally transported pollutants likely amplifies secondary aerosol formation [42].

High-frequency PM_2.5 pollution corridors clustered in mid-to-high latitudes (32.3° N–53.7° N), where winter heating-driven coal consumption surged and stable meteorological conditions amplified PM_2.5 concentrations. Conversely, sparse trajectories in low latitudes (18.0° N–25.1° N) resulted from efficient wet deposition and horizontal dispersion under frequent South China Sea monsoon and tropical cyclones, enhanced by marine air mass cleansing [48].

It is noteworthy that the primary objective of this was to identify and characterize the role of PM_2.5 pollution event migration across China, rather than to diagnose the underlying physical and chemical drivers in detail. The “movement” or continuity of events, as tracked by the FiT algorithm, represents the systematic progression of high-concentration areas in a three-dimensional space-time continuum. We acknowledge that the specific mechanisms driving these pathways—such as the complex interplay between long-range transport of pollutants versus local emission accumulation, the role of topographic forcing, and the breakdown of wintertime inversions by specific meteorological events—are undoubtedly critical and vary significantly across regions. A rigorous, mechanistic attribution of these pathways requires a dedicated investigation that integrates detailed emission inventories, high-resolution meteorological modeling, and chemical transport analysis, which falls beyond the scope of this methodological and macro-scale characterization study. Therefore, the pathways presented herein should be interpreted as the emergent, large-scale spatiotemporal signatures of PM_2.5 pollution events.

5. Conclusions

PM_2.5 pollution events pose severe health and environmental risks across China, yet their continuous evolution in three-dimensional space (longitude–latitude–time) is often overlooked, and the intrinsic linkages among key event characteristics (e.g., duration, maximum concentration, event PM_2.5 totals) remain poorly understood. Traditional methods based on ground stations or low-resolution data fail to capture dynamic behaviors such as event splitting, merging, and migration, thereby limiting the development of refined air pollution control strategies. By integrating the meteorological event tracking algorithm (i.e., the FiT algorithm) with the CHAP dataset (1-day temporal resolution, 1 km spatial resolution), this study developed a novel three-dimensional spatiotemporal framework for identifying contiguous PM_2.5 pollution events. This framework enabled a systematic characterization of the spatiotemporal evolution of PM_2.5 events across China (2013–2021), a quantification of correlations among key event indices, and the tracking of transport pathways. The principal findings are as follows:

(1)

An innovative framework applying the FiT algorithm to track PM_2.5 pollution events is proposed. Results demonstrate that combining the FiT algorithm with the CHAP dataset effectively captures the three-dimensional (space-time) continuum characteristics of PM_2.5 events in China.

(2)

The annual event PM_2.5 totals and pollution event frequency both exhibit a distinct right-skewed “T”-shaped spatial pattern. Hotspots concentrate in Xinjiang, the BTH region, Shandong, and Henan, where the annual event frequency exceeds 15. Provinces such as Jiangsu, Anhui, and Hubei report 7–20 events/year, whereas most others reported fewer than 10 events/year.

(3)

Strong positive correlations exist between event PM_2.5 totals and both event duration and maximum concentration, particularly in heavily polluted areas where Pearson’s values approach 1. This indicates that in these regions, longer-duration events accumulate greater pollutant mass and maintain higher peak concentrations.

(4)

Event duration exhibited significant regional heterogeneity. Short-duration events (1 day or 2–3 days) dominate (>80% of all events), while long-duration events cluster in heavily polluted areas. In less polluted regions, 1-day events account for >60% of occurrences, contrasting with lower proportions in severely polluted areas. The proportion of 2–3-day events is spatially uniform (>20%). In pollution hotspots (e.g., BTH), events lasting 9–12 days and >12 days constitute 5–15% and 10–15% of occurrences, respectively.

(5)

Short-duration events (2–3 days) were typically confined to local areas in Northwest–North China. Medium-duration events (4–8 days) extended further eastward, and long-duration events demonstrated trans-regional transport, reaching the southeastern coast. All events generally show a consistent west-to-east transport trend. Short-duration events propagate more slowly across inland Northwest China (<5° E/day), whereas long-duration events accelerate significantly along the eastern coast (>8° E/day).

It is noteworthy that the field lacks a universal standard for defining PM_2.5 pollution events. Variations in key parameters—such as the concentration threshold, spatial criteria, and tracking algorithms—can significantly influence the identified characteristics of events (e.g., frequency, duration, and transport pathways). The findings of this study are based on a specific framework employing a 75 μg/m³ threshold and the FiT algorithm, ensuring internal consistency and policy-relevance within China. Consequently, direct comparisons with studies using different criteria require caution. Future work should systematically evaluate the sensitivity of event characteristics to these definitional parameters to foster more harmonized understanding and robust comparisons across studies.