Spatiotemporal Evolution and Influencing Factors of Air Pollutants in the Three Major Urban Agglomerations of the Yellow River Basin

Abstract

Against the backdrop of the ongoing advancement of China’s dual-carbon goals and the coordinated strategy for ecological protection and high-quality development in the Yellow River Basin (YRB), it is important to clarify the spatiotemporal dynamics of air pollution in the densely populated urban agglomerations of the mid–lower YRB. Using station-based daily observations from 2015 to 2024, this study examines six major air pollutants (PM_2.5, PM₁₀, CO, NO₂, O₃ and SO₂) across the Shandong Peninsula, Central Plains, and Guanzhong Plain urban agglomerations. Sen’s slope estimator and the Mann–Kendall test are applied to quantify long-term trends, while partial correlation analysis and the GeoDetector model are used to diagnose pollutant co-variations and the drivers of spatial heterogeneity. Results indicate that while PM_2.5, PM₁₀, NO₂, SO₂, and CO concentrations significantly decreased, O₃ exhibited a statistically significant upward trend (Z = 2.32, p = 0.02), particularly with pronounced summer maxima. PM_2.5 shows clear seasonal variation, with elevated levels during winter and reduced levels during summer. Marked spatial contrasts are also observed: elevated particulate matter and CO are concentrated in the northern part of the Central Plains, while higher O₃ levels are more evident in coastal areas, particularly within the Shandong Peninsula urban agglomeration. In terms of inter-pollutant relationships, particulate matter and CO are positively associated with SO₂, whereas O₃ is negatively correlated with NO₂. GeoDetector results further suggest that air temperature, wind speed, and topography are the key factors associated with the spatial differentiation of pollutant levels; notably, the interaction between wind speed and temperature provides the greatest explanatory power, with effects that vary seasonally. These findings provide a scientific basis for region-specific air-pollution control and for advancing the co-benefits of carbon reduction and pollution mitigation in the YRB.

1. Introduction

Air pollution is one of the primary global environmental challenges and has been widely proven to be closely associated with the occurrence of various respiratory, cardiovascular, and cerebrovascular diseases [1]. Since the Industrial Revolution, the issue of atmospheric pollution has spread globally alongside the processes of urbanization and industrialization. Developed nations were the first to experience several iconic pollution events, such as the photochemical smog in London, UK, and the photochemical pollution in Silicon Valley, USA [2,3]. Entering the 21st century, developing countries like India and Brazil have also frequently suffered from severe haze and ozone pollution episodes [4,5,6]. As the world’s largest developing country, China has witnessed a series of regional heavy pollution events during its rapid industrialization and urbanization [7]. To address this severe situation, China has successively implemented a series of environmental governance policies. Since 2013, the country has built and refined the National Ambient Air Quality Monitoring Network. In 2015, PM_2.5 monitoring was fully promoted in cities at or above the prefecture level. By deploying air quality monitoring stations in the urban areas of each prefecture-level city, the Ministry of Ecology and Environment has significantly enhanced the monitoring capacity and spatiotemporal resolution of multiple pollutants [8,9,10].

Acquiring pollutant data is merely a prerequisite; more importantly, it is essential to conduct comprehensive analyses of long-term time-series data [11]. Currently, atmospheric pollutant monitoring primarily relies on three technical approaches: ground-based station observations [12], satellite remote sensing retrieval [13], and numerical simulation [14]. The national automated ambient air quality monitoring stations now substantially cover prefecture-level cities and above across China, providing hourly observational data for multiple pollutants. This enables the accurate analysis of daily, monthly, seasonal, and multi-year variation trends [15,16]. In time-series trend analysis based on station data, parametric methods such as linear regression, cumulative anomaly, and sliding t-tests are commonly employed to describe fluctuations in pollutant concentrations [17,18,19]. However, considering that atmospheric environmental data often do not follow a strict normal distribution and exhibit seasonal fluctuations, research utilizing non-parametric statistical methods—such as Sen’s slope estimator and the Mann–Kendall (MK) trend test—has gradually become mainstream in recent years. Compared with linear regression, it is more suitable for pollutants such as PM_2.5 and O₃, which show strong seasonality and non-normal data distributions [20,21]. Kernel density estimation and the global Moran’s I index can visually show the spatial distribution of pollutants such as PM_2.5 and O₃ across the three urban agglomerations. By introducing trend analysis, the ten-year change trends of pollutants such as PM_2.5 and O₃ in the three urban agglomerations can be obtained, and the cities with the fastest declining trends within the study area can be identified, so that targeted regulation can be carried out for key cities.

Regarding analyses of factors influencing air pollutants, geographically weighted regression (GWR) and geographically and temporally weighted regression (GTWR) introduce weighting schemes that vary with geographic location and time, allowing regression coefficients to change across space and time, thereby describing how the influence intensity of different driving factors varies among regions and periods [22,23,24]. GeoDetector is often used to quantify spatial stratified heterogeneity and to compare the explanatory power of candidate drivers for observed pollutant patterns [25,26]. Explanatory variables have also expanded from meteorological conditions alone to include socio-economic indicators, context-specific human activities and interventions (e.g., short-term emission changes during the Spring Festival), and abrupt events such as the COVID-19 pandemic. Looking ahead, global warming is expected to intensify heatwaves [27,28,29]. Compound heatwaves not only exacerbate O₃ pollution but also increase the risk of heart disease and other ailments for populations experiencing concurrent heat and pollution exposure [30,31]. Accurately identifying heatwaves is a prerequisite for addressing crises caused by global warming [32,33] and represents one of the future directions for atmospheric pollution research.

In terms of research scales, the spatiotemporal distribution and evolution patterns of air pollution can be explored across national, regional, urban, and even station-level scales. Domestically, most studies have focused on key urban agglomerations with high population densities and intensive emissions [34,35]. Regarding the Yellow River Basin (YRB), research has progressively shifted from macro-level analyses of individual cities or the entire basin toward a focus on the urban agglomeration scale. Studies have confirmed that high-pollutant zones within the basin are concentrated in the Central Plains, Shandong Peninsula, and Guanzhong Plain urban agglomerations in the mid–lower reaches [36,37,38]. Furthermore, while concentrations of pollutants such as SO₂ and NO₂ have exhibited a significant downward trend following the implementation of the “Air Pollution Prevention and Control Action Plan,” the seasonal prominence of O₃ pollution has emerged as a new governance challenge [39,40].

While existing studies have yielded fruitful results regarding the spatiotemporal evolution and driving mechanisms of multi-pollutants at the urban agglomeration scale, providing essential references for this study, several critical knowledge gaps remain. First, existing studies mainly focus on coastal urban agglomerations including Beijing–Tianjin–Hebei, Yangtze River Delta, and Pearl River Delta. Comparative analyses among urban agglomerations and policy-driven differences are insufficient, especially systematic research on the three major agglomerations in the mid–lower Yellow River Basin. Second, most works focus on PM_2.5 or a limited number of pollutants; however, research on the synergistic evolution, correlation structures, and compound pollution characteristics of all six criteria pollutants (PM_2.5, PM₁₀, NO₂, O₃, CO, and SO₂) remains insufficient. Third, this study innovatively combines partial correlation analysis and the GeoDetector to integrate linear and nonlinear methods, yielding more scientific results. This study reveals the spatiotemporal evolution patterns and influencing factors of six pollutants across the three major urban agglomerations, providing empirical support for designing differentiated pollution reduction strategies targeting PM_2.5- or O₃-dominated pollution regimes in different cities (from coastal to inland regions) and across different seasons.

2. Materials and Methods

2.1. Study Area

The three major urban agglomerations in the middle and lower reaches of the Yellow River Basin (YRB), namely the Shandong Peninsula Urban Agglomeration (SPUA), the Central Plains Urban Agglomeration (CPUA), and the Guanzhong Plain Urban Agglomeration (GZPUA), represent three different types of geographical environments from the coastal area to the inland plain, and have unique meteorological and topographic conditions related to pollutant accumulation, transport, and photochemical formation. SPUA covers all 16 prefecture-level cities in Shandong Province. The CPUA includes 29 cities. While the GZPUA is centered on Xi’an and includes 11 prefecture-level cities. The specific city names are shown in Figure 1.

The SPUA is characterized by a complex topographical system comprising low hills, fault basins, and coastal plains. Bordering the Yellow Sea and the Bohai Sea, it possesses a temperate marine climate, with an average annual temperature of 11–14 °C and annual precipitation of 650–850 mm. The region is marked by a high level of urbanization, high population density, and a solid foundation for both industrial and agricultural development. The CPUA is located in the transition zone from the warm temperate to the subtropical zone. The northern part is dominated by a temperate monsoon climate, while the southern part exhibits subtropical monsoon climate characteristics, featuring distinct seasons and synchronous heat and rain. It is one of the primary grain-producing regions in China. The GPUA is located in the climatic transition zone between northern and southern China. The core area belongs to a temperate continental climate, with an annual mean temperature of about 6–13 °C and annual precipitation of about 500–900 mm. In terms of topography, it is bordered by the Qinling Mountains to the south and the Loess Plateau to the north, and the middle part is a lowland corridor formed by the Weihe alluvial plain. This topography is prone to pollutant accumulation under stable meteorological conditions, and it can also transport pollutants to other regions through monsoon circulation.

2.2. Data Source

The PM_2.5 and other five criteria pollutant datasets used in this study were obtained from the China National Environmental Monitoring Centre (CNEMC) (https://www.cnemc.cn/). In accordance with the Ambient Air Quality Standards (GB 3095-2012) [41], the data were recorded hourly. These monitoring stations are primarily distributed in urban areas and have been widely utilized in atmospheric pollutant monitoring across China. Data preprocessing included the removal of outliers and null values. Daily, monthly, seasonal, and annual averages were subsequently calculated using the arithmetic mean method. Data analysis was performed using Python (v3.13) and R (v4.3.3).

To analyze the factors influencing the spatiotemporal distribution of pollutants, multiple datasets were collected. Meteorological factors, including air temperature, atmospheric pressure, precipitation, and relative humidity, were derived from the China High-resolution Multi-element Meteorological Forcing Product (China Met). This dataset was developed by integrating multi-source remote sensing data, reanalysis products, and observational data from over 2000 meteorological stations, offering a spatial resolution of 1 km and temporal resolutions at daily, monthly, and annual scales [42]. In this study, annual data were averaged to obtain multi-year mean values.

Digital Elevation Model (DEM) data were obtained from the Copernicus DEM, which provides a 30 m spatial resolution and is recognized as one of the highest-quality open-source global DEM products. In this study, the DEM for the research area was extracted via mask extraction based on the shapefile (shp) boundaries of the three major urban agglomerations in the YRB. Slope data were subsequently derived from the DEM.

Population data were sourced from the WorldPop 1 km resolution global dataset, which was resampled and mosaicked from the 100 m resolution population count products, covering the period from 2015 to 2030. Solar radiation (Srad) data were obtained from the China Meteorological Forcing Dataset (CMFD). The CMFD 2.0 product features a temporal resolution of 3 h and a horizontal spatial resolution of 0.1°. It integrates the European Centre for Medium-Range Weather Forecasts (ECMWF) ERA5 reanalysis data with surface station observations, specifically incorporating artificial intelligence techniques for radiation and precipitation products.

2.3. Analytical Methods

2.3.1. Sen’s Slope Estimator and Mann–Kendall Test for Trend Analysis

Sen’s slope estimator is a non-parametric method characterized by high computational efficiency and robustness against outliers. It is commonly integrated with the Mann–Kendall (MK) non-parametric test to detect trends in long-term time series. The calculation formula is presented in Equation (1). In this formula, $\beta$ represents the six criterion pollutants, while ${{\displaystyle x}}_i$ and ${{\displaystyle x}}_{\mathrm{j}}$ denote the pollutant concentration values in the $\mathrm{i}$th and $\mathrm{j}$th years, respectively ($\mathrm{i}$, $\mathrm{j}$ = 1, 2, 3, …, n and $\mathrm{i}$ > $\mathrm{j}$). The “Median” refers to the median function. A value of $\beta$ > 0 indicates an increasing trend in pollutant concentrations within the study area, $\beta$ = 0 signifies that the concentrations remained relatively stable during the study period, and $\beta$ < 0 denotes a decreasing trend.

$$\begin{array}{ccc}\beta & =\mathrm{Median}\left({\displaystyle \frac{x_j-x_i}{\mathrm{j}-\mathrm{i}}}\right)& \forall \mathrm{j}>\mathrm{i}\end{array}$$

(1)

The Mann–Kendall (MK) significance test is widely employed to evaluate the statistical significance of trends in time-series variables [21]. As a non-parametric approach, it does not require the assumption of normality for data variance, thereby effectively mitigating the influence of outliers on the final results. The MK test is calculated as follows:

$$S={\displaystyle \sum _{i=1}^{n-1}{\displaystyle \sum _{j=i+1}^n\mathrm{sgn}}}(x_j-x_i),$$

(2)

$$\mathrm{sgn}(x_j-x_i)=\left\{\begin{array}{ll}+1\hfill & x_j-x_i>0\hfill \\ 0\hfill & x_j-x_i=0\hfill \\ -1\hfill & x_j-x_i<0\hfill \end{array}\right.,$$

(3)

$$Z=\left\{\begin{array}{ll}{\displaystyle \frac{S-1}{\sqrt{\mathrm{VAR}(S)}}},\hfill & S>0,\hfill \\ \hspace{1em}\hspace{1em}0\hspace{1em}\hspace{1em},\hfill & S=0,\hfill \\ {\displaystyle \frac{S+1}{\sqrt{\mathrm{VAR}(S)}}},\hfill & S<0.\hfill \end{array}\right.,$$

(4)

For a given significance level $\alpha$, a trend is considered statistically significant when $\left|Z\right|$ > ${{\displaystyle \mathrm{z}}}_{1-\alpha /2}$; otherwise, the change is regarded as slight or non-significant. Based on the $\mathrm{Z}$-scores derived from the Mann–Kendall test, the variation trends of atmospheric pollutants are categorized into four levels: significant decrease, non-significant decrease, non-significant increase, and significant increase. Specifically, a value of |$Z$| > 1.96 indicates a significant change at the 95% confidence level ($\alpha$ = 0.05). When 0 < |$Z$| ≤ 1.96, the trend is identified as non-significant.

2.3.2. Partial Correlation Coefficient

The partial correlation coefficient is an indicator used to measure the “pure degree of association” between two variables after eliminating the interference of all other potential variables. The value of this coefficient ranges from −1 to 1; the closer the absolute value is to 1, the stronger the correlation, whereas a value closer to 0 indicates a weaker correlation. Common methods for calculating partial correlation coefficients include the residual method and the inverse correlation matrix method.

Given that this study requires the calculation of pairwise correlations among six distinct pollutants—involving a large number of variables—the residual method would be computationally inefficient as it necessitates step-by-step calculations. Therefore, the inverse correlation matrix method is adopted for batch computation. Let ${{\displaystyle r}}_{XY}$ represent the Pearson correlation coefficient between $\mathrm{X}$ and $\mathrm{Y}$; ${{\displaystyle r}}_{XZ}$ between $\mathrm{X}$ and $\mathrm{Z}$; and ${{\displaystyle r}}_{YZ}$ between $\mathrm{Y}$ and $\mathrm{Z}$. Then, ${{\displaystyle r}}_{XY\cdot Z}$ denotes the partial correlation coefficient between $\mathrm{X}$ and $\mathrm{Y}$ while controlling for variable $\mathrm{Z}$. The formula for the inverse correlation matrix is as shown in Equation (5):

$$r_{XY\cdot Z}={\displaystyle \frac{r_{XY}-r_{XZ}\cdot r_{YZ}}{\sqrt{(1-r_{XZ}^2)\cdot (1-r_{YZ}^2)}}},$$

(5)

2.3.3. GeoDetector

GeoDetector is a statistical method designed to detect spatial differentiation of geographical phenomena and reveal their underlying driving mechanisms. In this study, the factor detector and interaction detector modules of GeoDetector were employed to quantify the influence of individual environmental factors on the spatial differentiation of pollutant concentrations and to evaluate the effects of interactions between these factors. To mitigate the uncertainty associated with single-year detection results, this study assessed the driving intensity of various factors regarding the spatial differentiation of PM_2.5 concentrations comprehensively across the period from 2015 to 2024.

The q-statistic for the factor detector is calculated as shown in Equation (6):

$$q^2=1^2-\left({\displaystyle \sum _{h=1}^L{N}_h}{\sigma }_h^2\right)/N{\sigma }^2,$$

(6)

$\mathrm{q}$ represents the explanatory power (influence) of a driving factor on the spatial differentiation of PM_2.5 concentrations. Its value ranges within [0, 1], where a higher value indicates a stronger influence of the factor. $\mathrm{h}$ = 1, 2, …, $\mathrm{L}$ denotes the strata (sub-regions) of the explanatory variable $\mathrm{X}$. $\mathrm{L}$ denotes the number of strata or classifications for the variable. $\mathrm{N}$ and ${{\displaystyle N}}_h$ represent the total number of samples in the entire study area and the number of samples in stratum $\mathrm{h}$, respectively. ${\sigma }^2$ and $N{\sigma }^2$ represent the total variance of the PM_2.5 concentrations across the entire area and the variance within stratum $\mathrm{h}$, respectively.

3. Results

3.1. Temporal Evolution and Trend Analysis

During the 2015–2024 period, the air quality across the three major urban agglomerations of the YRB improved substantially. As illustrated in Figure 2, five out of the six criterion pollutants exhibited significant downward trends, aligning with the coordinated implementation of regional ecological protection and “Dual Carbon” strategies.

3.1.1. Inter-Annual Evolution and Decoupling Trends

Statistical summaries of the inter-annual trends are provided in

Table 1. Mean concentrations in 2015 and 2024 (averaged across monitoring sites) and corresponding long-term trend statistics (Sen’s slope; Mann–Kendall Z and p) for six criterion air pollutants (2015 to 2024).

Pollutant	Mean (2015)	Mean (2024)	Sen’s Slope	M-K(Z)	p-Value
PM2.5 (µg/m3)	67.27	39.29	−0.25	−4.21	<0.001
SO2 (µg/m3)	37.38	8.11	−0.17	−5.13	<0.001
O3 (µg/m3)	58.61	77.97	0.12	2.32	0.02
CO (mg/m3)	1.40	0.65	−0.01	−3.95	<0.001
NO2 (µg/m3)	38.20	24.07	−0.15	−4.08	<0.001

. PM_2.5, PM₁₀, NO₂, SO₂, and CO all demonstrated significant monotonic decreasing trends (p < 0.001). Notably, SO₂ concentrations exhibited the most dramatic decline, with a Mann–Kendall Z-score of −10.28 and a reduction exceeding 80% over the decade. This plummet in SO₂ is primarily attributed to the stringent implementation of “coal-to-gas” conversion projects and the widespread upgrading of industrial desulfurization facilities across the mid–lower reaches of the basin. Similarly, PM_2.5 dropped from an annual average of 67.27 µg/m³ in 2015 to 39.29 µg/m³ in 2024 (slope = −0.25). During the period from 2015 to 2024, O₃ concentrations exhibited an upward trend (slope = 0.12, Z = 2.32, p = 0.02). The highest O₃ levels occurred in summer throughout the year. In the Mann–Kendall test, the Z-value was slightly below 1.96 and the p-value slightly above 0.05, which undermines the persuasiveness of interpreting the trend as statistically significant.

3.1.2. Seasonal Variation Patterns

Figure 2 shows pronounced seasonality for PM_2.5, PM₁₀, CO, NO₂, O₃ and SO₂, with winter maxima and summer minima across the three urban agglomerations. For example, wintertime PM_2.5 in 2015 averaged 101.76 µg/m³, substantially higher than the corresponding summer level. This pattern is consistent with enhanced emissions during the heating season and less favorable dispersion conditions in winter. In the summer of 2015, O₃ concentration was 80.92 µg/m³, peaked at 111.91 µg/m³ in the summer of 2022, and then showed a declining trend, reaching 104.58 µg/m³ in the summer of 2024. Notably, while PM₁₀ reached a decade low in 2022 (96.26 µg/m³), a slight rebound was observed thereafter. This likely reflects the rapid recovery of regional infrastructure investment and industrial production activities in the post-pandemic era.

3.2. Spatial Characterization and Seasonal Variations in Multi-Pollutant Concentrations

The six criterion pollutants exhibit pronounced spatial heterogeneity and distinct seasonal disparities across the three major urban agglomerations of the YRB. Overall, high-value clusters of PM_2.5, PM₁₀, CO, NO₂, O₃ and SO₂ are primarily concentrated in the CPUA and the western portion of the SPUA. In contrast, relatively lower concentrations are observed in the GZPUA and the coastal areas of the SPUA. This spatial gradient is likely linked to the high density of industrial emissions and population in the central plains, coupled with the “cleansing effect” of maritime air masses in coastal regions [43,44,45]. Furthermore, SO₂ hotspots are localized in a few energy-intensive industrial cities, while O₃ displays a spatial pattern nearly inverse to that of particulate matter (PM_2.5 and PM₁₀), with higher values emerging in parts of the eastern SPUA and central-eastern regions.

3.2.1. Seasonal Dynamics and Regional Gradients of Key Air Pollutants

At the seasonal scale, as shown in Figure 3 and Figure 4, PM_2.5 and PM₁₀ distributions typically follow an increasing southwest-to-northeast gradient during the annual, spring, autumn, and winter periods. During autumn and winter, high-concentration zones form continuous, large-scale clusters. This spatial continuity suggests significant regional transport and the “trapping effect” of the basin’s complex topography during periods of atmospheric stability [46,47]. In summer, overall concentrations decrease substantially, and the spatial extent of high-value zones contracts significantly, persisting only at moderate levels in specific central cities. Conversely, the spatial distribution of O₃ is characterized by a “high in summer/east and low in winter/west” pattern. During summer, a continuous high-value belt forms across the SPUA and central cities. This eastward-increasing gradient reflects the synergy between higher precursor (NOx and VOCs) concentrations and stronger solar radiation in the eastern plains compared to the western highlands. This range contracts slightly during spring and autumn and remains at a low baseline level across the entire region in winter.

The spatial distribution of NO₂ annually and in winter closely mirrors that of PM_2.5, with the core of the CPUA and cities with dense transportation networks serving as prominent hotspots. The spatial overlap between NO₂ peaks and major traffic corridors highlights the dominant contribution of vehicular emissions to urban air pollution [48,49,50]. In summer, NO₂ levels weaken overall, leaving only linear high-value ribbons along primary transportation axes. SO₂ distribution is dominated by localized hotspots across all seasons. Annual and winter high-value centers are concentrated in specific cities characterized by coal-fired power plants or heavy chemical industries. During summer and autumn, the overall concentration decreases, and spatial variance diminishes.

3.2.2. Spatial and Seasonal Patterns of Temporal Trend Slopes of Air Pollutants

On an annual scale, as shown in Figure 5 and Figure 6, the most significant negative slopes (indicating the fastest improvement) for PM_2.5 and PM₁₀ were concentrated in the eastern portion of the CPUA and western Shandong Province, where localized stations exhibited substantial reduction rates. In contrast, the negative slopes in the GZPUA and coastal areas of Shandong were relatively smaller, forming a spatial gradient described as “gradually weakening from east to west and from core cities to the periphery.” The more pronounced decreases in NO₂ and CO were concentrated within urban areas. For PM_2.5, PM₁₀, NO₂, and CO, negative slopes were more evident in spring, autumn, and winter, whereas the magnitude of decrease tended to be smaller in summer. In winter, several core urban areas exhibited the most significant declines, aligning with strengthened emission control measures during the heating season, though interannual meteorological variations may also be a contributing factor. The positive values in portions of the central-eastern and coastal regions were slightly higher in summer than in other seasons. Consequently, these areas represent critical focus zones for future synergistic control strategies targeting both particles and ozone.

3.2.3. Results Comparison with Other Regions

Compared with studies from other regions, the spatiotemporal patterns identified in the three YRB urban agglomerations are broadly consistent with the global picture of decreasing particulate pollution and persistent or increasing ozone. In Europe, long-term analyses in Germany and Italy have reported significant declines in PM_2.5 and PM₁₀ following stringent emission controls, while surface O₃ trends are mixed, with many urban sites showing either weak increases or only limited decreases, particularly in warm seasons [51,52]. Similar behavior has been documented in the United States, where NO_x and primary PM emissions have fallen under the Clean Air Act, yet summertime O₃ formation in regions such as California remains problematic because of non-linear NO_x–VOC chemistry and meteorological sensitivity [53,54]. In rapidly urbanizing India, recent work has shown strong reductions or stabilization of PM_2.5 in some cities but continued high levels and frequent exceedances of WHO guidelines, accompanied by rising O₃ exposure risks linked to increasing anthropogenic VOCs and intense photochemistry [55,56]. By contrast, Japan and South Korea—both of which have implemented long-term national and regional air-pollution control strategies—generally exhibit declining PM_2.5 concentrations and more moderate O₃ trends, although positive O₃ tendencies are still observed in several urban areas [57]. Overall, the YRB pattern of marked wintertime reductions in primary pollutants but weakly increasing, meteorology-sensitive O₃, especially in coastal and downwind regions, aligns more closely with the experience of East Asian and some European city clusters than with that of North America, underscoring the need for synergistic PM_2.5–O₃ control that explicitly accounts for regional transport and season-specific photochemical regimes.

3.3. Partial Correlation Analysis Among the Six Criterion Air Pollutants

The spatial distribution and inter-pollutant correlations vary across the three major urban agglomerations due to differences in geographical location (land–sea distribution) and socio-economic conditions. By analyzing the partial correlation coefficients among the six criterion pollutants, strong associations were identified within particulate matter categories and between particulates and combustion-related gases [58,59].

The partial correlation coefficients for PM_2.5–PM₁₀ in the GZPUA, CPUA, and SPUA were 0.67, 0.78, and 0.73, respectively, all showing significant positive correlations, as shown in Figure 7. This indicates that the two types of particulate matter are highly coupled in terms of sources and transport processes, co-regulated by primary emissions and secondary particle formation. The PM_2.5–CO correlation was prominent in the CPUA (0.65) and SPUA (0.69). Combined with the moderate negative correlations observed for PM₁₀–CO (−0.19 to −0.47), these results suggest that fine particles and carbon monoxide are markedly influenced by combustion sources. The inverse relationship between coarse particles and CO in certain regions may be linked to dust/fugitive dust processes and varying meteorological conditions.

O₃ exhibited inverse correlations with its precursor gases. The O₃–NO₂ partial correlation coefficients across the three regions were −0.60, −0.54, and −0.65, representing strong negative associations. This reflects the dual role of NO₂ as a vital O₃ precursor and a participant in the photochemical equilibrium that titrates O₃ [60,61]. O₃–PM_2.5 showed moderate negative correlations (−0.25 to −0.49), consistent with the previously discussed trend of “declining PM_2.5 and slowly rising O₃”. Regional disparities in O₃–CO correlations (−0.28 to 0.19) suggest that CO may act as a weak precursor in some regions while being modulated by boundary layer structures and transport conditions in others.

The PM_2.5–NO₂ relationship was weak overall and varied from −0.20 to 0.06, suggesting that the association between NO₂ and PM_2.5 differs with secondary formation conditions and local emission characteristics. The PM_2.5–SO₂ correlation was close to zero in the GZPUA (−0.01) but became clearly negative in the CPUA (−0.46) and SPUA (−0.35). One possible explanation is that coal-reduction efforts in the Central Plains and the Shandong Peninsula have reduced the co-variation between SO₂ and secondary PM. By comparison, PM₁₀–SO₂ remained weak to moderately positive (0.14–0.41), which may be related to a mix of coal combustion emissions and industrial dust contributions.

3.4. Analysis of Influencing Drivers Based on GeoDetector

In this study, nine influencing factors were selected to quantify the contribution of each factor to the pollutants. The data format is described in Section 2.2. Given the inherent correlation between Digital Elevation Model (DEM) and Slope Aspect, the spatial distribution map for Aspect is not redundantly displayed in the text. In the data preprocessing phase, a long-term averaging treatment was applied to the nine categories of data from 2015 to 2024. This approach was adopted to eliminate the influence of inter-annual fluctuations on the established spatial patterns. The spatial distributions of the eight representative driving factors within the study area exhibit significant spatial stratified heterogeneity (Figure 8). For the purpose of simplifying subsequent analysis, the factors are defined as follows: Wind Speed (Wind, X1), Temperature (TMP, X2), Solar Radiation (SRAD, X3), Relative Humidity (RHU, X4), Atmospheric Pressure (PRES, X5), Precipitation (PREC, X6), Aspect (X7), Population Density (POP, X8), and Elevation (DEM, X9).

The q-statistic of the factor detector indicates that for particulate matter (PM_2.5 and PM₁₀), the top four driving factors, in order, are wind speed (X1, q = 0.30), temperature (X2, q = 0.27), elevation (X9, q = 0.23), and atmospheric pressure/precipitation (X5/X6, q = 0.20), suggesting that the spatial variation in particulate matter is closely related to large-scale circulation conditions and underlying surface topographic features. For ozone (O₃), which is most sensitive to photochemical reaction conditions and boundary layer structure, the dominant factors are wind speed (X1, q = 0.44), elevation (X9, q = 0.37), atmospheric pressure (X5, q = 0.37), and solar radiation (X3, q = 0.37).

Seasonally, in summer, O₃ shows the strongest response to radiation (X3) and temperature (X2) (q = 0.47), consistent with its dependence on high temperatures and strong radiation conditions. In autumn and winter, PM_2.5 and PM₁₀ revert to a pattern dominated by wind speed and temperature (X1, X2; q = 0.32), highlighting the influence of atmospheric stability and heating-season emissions. Conversely, in winter, SO₂ and CO are primarily driven by relative humidity and precipitation (X4, X6; q = 0.54), emphasizing the role of humidity-related scavenging processes and regional emission structures.

The interaction detector shows that, for paired factors, the interaction q value is higher than the q value of either factor alone, corresponding to a bi-factor enhancement effect. This suggests that pollutant patterns across the three urban agglomerations are associated with multiple drivers acting together rather than one dominant factor. On an annual scale, the highest explanatory power for particulate matter (PM_2.5 and PM₁₀) was observed in the combinations of Precipitation ∩ Temperature (X6 ∩ X2, q = 0.64) and Elevation ∩ Temperature (X9 ∩ X2, q = 0.61), followed by Wind Speed ∩ Temperature (X1 ∩ X2, q = 0.60). These results indicate that temperature, atmospheric humidity, and precipitation have a joint influence on the spatial distribution of particulate matter. For ozone (O₃), the strongest interactions are radiation ∩ temperature (X3 ∩ X2, q = 0.68) and precipitation ∩ radiation (X6 ∩ X3, q = 0.66), indicating that radiative and thermal conditions, together with humidity-related processes, are all related to the spatial pattern of ozone. For sulfur dioxide (SO₂) and carbon monoxide (CO), the highest interaction q values occur in humidity combinations related to wind speed or temperature (X4 ∩ X1/X4 ∩ X2, q = 0.71), indicating that these combustion-related pollutants often have higher concentrations under humid conditions, especially when dispersion is relatively weak.

In spring, O₃ showed the strongest interaction effects for Radiation ∩ Temperature and Elevation ∩ Radiation (X3 ∩ X2 and X6 ∩ X3; q = 0.73). This pattern is in line with springtime conditions when radiation increases and atmospheric mixing becomes more active. In summer, O₃ was still mainly associated with Radiation ∩ Temperature and Elevation ∩ Radiation (X3 ∩ X2 and X9 ∩ X3; q = 0.71), indicating that radiative and thermal conditions remain important for ozone-related interactions. In autumn, the leading interactions for particulate matter and NO₂ shifted to Pressure ∩ Temperature and Wind Speed ∩ Temperature (X5 ∩ X2 and X1 ∩ X2; q = 0.63), suggesting a larger role of stable meteorological conditions during the seasonal transition. In winter, when pollution episodes are more frequent during the heating period, interactions involving stability-related variables (e.g., low wind and higher pressure) together with humidity and precipitation showed relatively high explanatory power. For PM_2.5 and PM₁₀, q reached 0.67 under Pressure ∩ Temperature and Elevation ∩ Temperature. The detailed data descriptions are provided in Appendix A.

4. Discussion

This study analyzed the spatiotemporal distribution patterns of air pollutants across three urban agglomerations in the YRB from 2015 to 2024, utilizing partial correlation analysis and the GeoDetector model. The results indicate that particulate matter (PM_2.5 and PM₁₀) and gaseous pollutants (NO₂, SO₂, CO) exhibited downward trends; O₃ concentrations followed an upward trend, peaking annually during the summer.

The observed decline in most pollutants aligns with previous studies focusing on the YRB, the Central Plains, the Guanzhong Plain, and key cities such as Xi’an and Zhengzhou [62,63,64]. These improvements are attributable to the stringent air pollution prevention and control policies implemented since 2013, alongside advancements in industrial governance and traffic management [65]. The rising trend of O₃ as particulate levels are brought under control is consistent with recent nationwide observations. Wang and Liu et al. have identified O₃ as a primary target for future atmospheric governance in China [66,67]; the choice of discretization method and the number of classes is a key step affecting the robustness of GeoDetector results. Existing research has provided methodological references, generally suggesting 5–8 classes and commonly adopting equal-interval or natural-break schemes for factor discretization in GeoDetector applications [68,69,70,71,72]. We conducted a sensitivity analysis of different classification methods and class numbers (see Supplementary Materials for details). We found that the natural-break classification with nine classes produced more stable and higher explanatory performance across pollutants and seasons.

The stronger interaction between elevation and radiation further suggests that topography modifies both the radiative environment and vertical exchange, amplifying O₃ formation and accumulation at higher altitudes [73,74]. Observations along elevation gradients and at high-altitude or mountainous sites have documented enhanced spring O₃ linked with stronger background radiation, more frequent coupling with the free troposphere, and more efficient vertical mixing, again pointing to a non-linear combined effect of elevation and radiative forcing on surface O₃ [75,76].

Partial correlation analysis revealed a significant positive correlation between PM_2.5 and PM₁₀. A strong negative correlation exists between O₃ and NO₂, reflecting the role of NO₂ as a critical precursor and the titration effect in the photochemical cycle [77]. While PM_2.5 and O₃ are negatively correlated across the three regions, PM₁₀ and O₃ exhibit a positive correlation, particularly in the coastal Shandong Peninsula. The “declining PM/rising O₃” phenomenon is a pivotal shift in China’s pollution landscape, necessitating research into co-existence and synergistic governance to support high-quality regional development [78]. NOx requires close attention not only as a primary pollutant but as a driver of secondary inorganic aerosol formation through interactions with volatile organic compounds (VOCs). Future strategies must prioritize the synergistic control of multiple pollutants [79]. By targeting emissions co-emitted with CO₂, we can simultaneously improve air quality and reduce carbon emissions, thereby serving the national “Dual Carbon” strategic goals.

5. Conclusions

Using station-based daily observations from 2015 to 2024, this study quantified trends, seasonality, spatial heterogeneity, and key drivers of six criterion pollutants across the Guanzhong Plain, Central Plains, and Shandong Peninsula urban agglomerations in the middle–lower YRB. Based our findings, we observed that PM_2.5, PM₁₀, NO₂, SO₂, and CO generally decreased over the study period, whereas O₃ showed a weak increasing tendency with pronounced summer maxima. Spatially, higher particulate matter and combustion-related pollutants were concentrated in the core areas of the Central Plains urban agglomeration (e.g., Zhengzhou and Xinxiang), while elevated O₃ occurred more frequently in coastal areas and parts of the Shandong Peninsula region. GeoDetector results indicate that meteorological conditions—especially wind speed and temperature and their interactions with precipitation/topography—provide the strongest explanatory power for pollutant heterogeneity, highlighting the importance of season-specific mechanisms. These results suggest that future control policies should jointly address wintertime PM reduction and summertime ozone formation, with coordinated strategies tailored to the distinct regional contexts of the three urban agglomerations. In daily practice, this implies region-oriented, multi-pollutant management that simultaneously considers local emissions, regional transport and meteorology. Evidence from other major Chinese urban agglomerations shows that joint PM_2.5–O₃ control can yield substantial additional health and economic benefits compared with single-pollutant strategies, but requires differentiated reduction pathways for NOx, VOCs, SO₂ and primary PM across cities. For the middle–lower YRB, the findings here suggest several actionable directions. First, in the Central Plains UA, where particulate pollution remains highest, continued deep cuts in primary PM_2.5, SO₂ and NOx from heavy industry and wintertime residential combustion should be prioritized, while preventing NOx-only control from exacerbating summertime O₃ under VOC-limited regimes. Second, in the coastal and Shandong Peninsula UAs, where O₃ burdens and meteorological sensitivity are stronger, emission reduction should focus more on VOCs from industry and solvents, coordinated with region-wide NOx control and seasonal O₃-targeted measures in spring–summer. Third, across all three urban agglomerations, cross-boundary joint-prevention mechanisms are needed to address strong spillover and transport identified in other clusters such as the Beijing–Tianjin–Hebei Urban Agglomeration and YRD regions, including shared early warning, unified emission-reduction schedules during regional episodes, and city-cluster-level health-risk assessments to guide period- and pollutant-specific interventions.