Next Article in Journal
Parametric Multi-Objective Optimization of Urban Block Morphology Using NSGA-II: A Case Study in Wuhan, China
Previous Article in Journal
Spatial Optimization Strategies for Rural Tourism Villages: A Behavioral Network Perspective—A Case Study of Wulin Village
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Sustainability
Volume 17
Issue 21
10.3390/su17219721
sustainability-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Multi-Scale Analysis of Influencing Factors for Temporal and Spatial Variations in PM2.5 in the Yangtze River Economic Belt

by
Yufei Zhang
1,2,
Yu Chen
1,* and
Yongming Wei
3
1
International Research Center of Big Data for Sustainable Development Goals, Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100094, China
2
University of Chinese Academy of Sciences, Beijing 100049, China
3
Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100094, China
*
Author to whom correspondence should be addressed.
Sustainability 2025, 17(21), 9721; https://doi.org/10.3390/su17219721
Submission received: 2 October 2025 / Revised: 27 October 2025 / Accepted: 29 October 2025 / Published: 31 October 2025
(This article belongs to the Section Air, Climate Change and Sustainability)
Browse Figures
Review Reports Versions Notes

Abstract

PM2.5 is the primary source of urban atmospheric pollution, as it not only damages the ecological environment but also poses a threat to human health. Taking the Yangtze River Economic Belt as the research object, this study analyzes the spatiotemporal variation characteristics of PM2.5 concentrations in the region from 2005 to 2020. Furthermore, by combining the Geodetector model with Geographically and Temporally Weighted Regression (GTWR) model, the spatiotemporal heterogeneity of its influencing factors is revealed at three scales: municipal, watershed, and grid. The results show that, from 2005 to 2020, the annual average PM2.5 concentration in the Yangtze River Economic Belt exhibited an inverted U-shaped trend with 2013 as the inflection point, showing distinct spatial clustering characteristics. Overall, the spatiotemporal variation in annual average PM2.5 concentration demonstrated a significant downward trend during this period, with slower decline rates in the western region and faster rates in the central and eastern regions. Spatial differentiation of annual average PM2.5 concentrations within the region was primarily influenced by three factors: PFA, PISA, and PD. NDVI and PWA exerted their effects mainly at large scales, while MAT and SDE primarily acted at small scales. Within the region, NDVI and CVO predominantly suppressed PM2.5 concentrations, whereas MAT, PFA, PD, and SDE primarily promoted PM2.5 pollution. The spatial distribution of effects for factors within the same category is broadly consistent across the three scales, though details vary. This study overcomes previous limitations of administrative-scale research, yielding more refined results. It provides new methodologies and insights for future research while offering more precise scientific support for regional PM2.5 governance.

1. Introduction

PM2.5 refers to fine particulate matter with a diameter of 2.5 μm or smaller. It is a primary source of urban air pollution [1] and poses a serious threat to human health. According to a report by the World Health Organization, approximately 4.2 million people worldwide die each year as a result of environmental air pollution [2]. Related studies have confirmed that prolonged exposure to high concentrations of PM2.5 may trigger at least 64 severe diseases [3,4,5], and there is a significant positive correlation between PM2.5 concentration levels and mortality rates for certain diseases [6,7]. The Yangtze River Economic Belt spans China’s eastern, central, and western regions and is one of the three major strategies prioritized by the central government. Since the reform and opening-up, the Yangtze River Economic Belt has developed into one of China’s most comprehensive and strategically significant regions, playing a crucial role in urban development and economic construction. In September 2014, the State Council issued the “Guiding Opinions on Promoting the Development of the Yangtze River Economic Belt by Leveraging the Golden Waterway,” outlining plans to develop the Yangtze River Economic Belt into a globally influential inland waterway economic belt, a coordinated development belt for east–west–central cooperation, a comprehensive opening-up belt for domestic and international cooperation along the coast, rivers, and borders, and a pioneer demonstration belt for ecological civilization construction. Therefore, studying the spatiotemporal distribution characteristics and influencing factors of PM2.5 concentrations in the Yangtze River Economic Belt is of great significance for reducing health risks to residents and promoting urban ecological civilization construction.
Currently, domestic scholars have conducted extensive research on the influencing factors of PM2.5. Multiple studies have shown that changes in PM2.5 concentrations are related to both natural environmental factors and human activities [8,9]. In terms of natural environmental factors, precipitation, temperature, humidity, wind speed, and other factors influence PM2.5 concentrations [10]; human activity-related factors such as population density, Gross Domestic Product (GDP), nighttime lighting, and energy consumption [11] also influence PM2.5 pollution. Huang [12] used the Logarithmic Mean Divisia Index method to identify key factors affecting PM2.5 concentrations in 236 cities in China from 2011 to 2020, and found that emission intensity inhibited PM2.5 concentrations in all cities, energy intensity inhibited it in 157 cities; economic output stimulated it in some less economically developed regions, and population stimulated it in 135 cities, which were mainly concentrated in economically developed eastern regions. Xu [13] used a geographic detector to investigate the influencing factors of PM2.5 in the Beijing-Tianjin-Hebei Urban Agglomeration from 2000 to 2021, finding that temperature, elevation, and road network density are the primary factors influencing the spatial variation in PM2.5 concentrations, the interactive effects of temperature with precipitation, elevation, and road network density constitute the main factor combinations influencing the spatial differentiation of PM2.5 in the Beijing-Tianjin-Hebei Urban Agglomeration. Li [14] utilized the Spatial Durbin Model (SDE) to analyze the influencing factors of PM2.5 concentrations in the three major urban agglomerations within the Yangtze River Economic Belt from 2000 to 2020. They found that changes in PM2.5 concentrations are jointly influenced by factors such as the secondary industry, urban built-up area, population density, annual precipitation, and Normalized Difference Vegetation Index (NDVI). The dominant factors influencing PM2.5 concentrations in the three major urban agglomerations share similarities but also exhibit differences. Fang [15] studied 74 countries involved in the Silk Road Economic Belt and the 21st-Century Maritime Silk Road Initiative (BRI), using a combination of spatial autocorrelation and regression analysis to identify the factors influencing PM2.5 concentrations. They found that energy intensity and per capita electricity consumption are the primary drivers of PM2.5 concentrations, while the expansion of forest area significantly contributes to the reduction of PM2.5 concentrations.
In terms of research methods, to conduct a more comprehensive analysis of the mechanisms underlying changes in PM2.5 concentrations and their influencing factors, scholars both domestically and internationally have employed various methods, including correlation analysis [16], Spatial Econometric models [17], Geodetector [18], Geographically Weighted Regression models [18], and some machine learning methods [19,20,21,22,23,24] such as Random Forest (RF), Generalized Additive Models, eXtreme Gradient Boosting (XGB), and Artificial Intelligence (XAI), etc. However, each individual method has its limitations in research. Therefore, this study adopts a combined approach of Geodetector and GTWR models. In terms of research scale, most existing studies use provincial, municipal, or county administrative regions as research units and primarily focus on a single scale, neglecting the differences in spatiotemporal variations and influencing factors of PM2.5 concentrations across different regions and scales. Therefore, this study aims to break through the limitations of previous research confined to administrative boundaries, analyzing the influencing factors of PM2.5 at three scales: municipal, watershed, and grid, to explore the commonalities and differences in influencing factors across different scales.
In summary, aiming to analyze the spatiotemporal variation patterns of PM2.5 concentrations in the Yangtze River Economic Belt, explore the influencing factors of these spatiotemporal changes, provide scientific support for air pollution prevention and control measures, assist policymakers in effectively addressing regional air pollution, and promote the continuous improvement of air quality. This study takes the Yangtze River Economic Belt as the research object. Based on remotely sensed PM2.5 data from 2005 to 2020, combined with relevant data on the natural environment and human activities during the same period, an integrated global-to-local analytical approach is established by coupling the Geodetector and GTWR models. This study reveals the spatiotemporal variation characteristics of PM2.5 concentrations, explores the key influencing factors, and analyzes the spatiotemporal heterogeneity of these key factors across three scales: municipal, watershed, and grid.

2. Materials and Methods

2.1. Study Area

The Yangtze River Economic Belt is centered around the Yangtze River’s golden waterway, covering 11 provinces and municipalities including Shanghai, Jiangsu, Zhejiang, Anhui, Jiangxi, Hubei, Hunan, Chongqing, Sichuan, Guizhou, and Yunnan, with a total area of approximately 2.0523 million square kilometers, accounting for 21.4% of the national total [25]. It spans the eastern, central, and western regions of China’s mainland and is one of the three major strategies prioritized by the central government. It serves as a globally influential inland economic belt, a coordinated development belt for east–west and central–west cooperation, a comprehensive opening-up belt along the coast, rivers, and borders, and a pioneer demonstration belt for ecological civilization construction. The spatial scope is between 97°20′ E and 123°30′ E, and 21°30′ N and 35°20′ N. The overall terrain is lower in the east and higher in the west, with diverse topography within the region, including the Qinghai–Tibet Plateau, the Yungui Plateau, the Sichuan Basin, and the Yangtze River Middle and Lower Reaches Plain. Except for the Qinghai–Tibet Plateau region, which has a highland mountain climate, most of the region falls under the subtropical monsoon climate zone, characterized by a warm climate, abundant rainfall, and numerous tributary lakes. The region boasts convenient transportation, access to rivers and seas, and a prominent geographical advantage; it also has a distinct industrial advantage, dense urban areas, and high economic density, with both population and GDP totaling over 40% of China’s national figures (Figure 1). From west to east, the level of economic development gradually increases [26]. Historically, it has been one of China’s most important industrial corridors, home to a large number of high-energy-consuming, high-value-added enterprises.

2.2. Research Scale

This study examines three spatial scales: municipal, watershed, and grid (10 km). At the municipal scale, 130 units were identified (Figure 2a), with China’s administrative boundary data sourced from the Tianditu National Geographic Information Public Service Platform. The watershed scale comprises 1041 units (Figure 2b), utilizing the HydroBASINS [27] watershed dataset. Specifically, the Asia Level 07 dataset was employed and cropped according to the vector boundaries of the Yangtze River Economic Belt. The grid scale consists of 20,477 units (Figure 2c), generated using the ArcGIS 10.8 Grid tool to create a 10 KM grid.

2.3. Data Sources and Preprocessing

The data used in this paper include(as shown in Table 1): (1) PM2.5 data; (2) natural factor data (Normalized Difference Vegetation Index (NDVI), Annual Rainfall (AR), Mean Annual Temperature (MAT), Proportion of Water Area (PWA)); (3) Human factor data (Proportion of Farmland Area (PFA), Proportion of Impervious Surface Area (PISA), Population Density (PD), Nighttime Lights (NTL), Gross Domestic Product (GDP), Sulfur Dioxide Emissions (SDE), Civil Vehicle Ownership (CVO)). The PM2.5 data is sourced from the ChinaHighPM2.5 dataset Version 4 [28,29], a yearly 1 km product covering the period from 2000 to present. This study selecting annual average data from 2005 to 2020. NDVI data is sourced from the NASA MOD13A3 dataset (MODIS/Terra Vegetation Indices Monthly L3 Global 1 km SIN Grid V061|Earthdata Search (nasa.gov)), from which annual average NDVI values for 2005–2020 were calculated. AR [30,31,32,33,34] and MAT [30,31,32,33,35] data were obtained from the National Tibetan Plateau Data Center, selecting annual data from 2005 to 2020; GDP [36,37,38] data sourced from Global 1 km × 1 km gridded revised real gross domestic product, selecting data from 2005 to 2020. PD data sourced from the WorldPop dataset Unconstrained individual countries 2000–2020 (1 Km resolution) version [39]. NTL data sourced from the global 500-m resolution “NPP-VIIRS-like” nighttime light dataset [40], selecting data from 2005 to 2020. PWA, PFA and PISA were calculated using land cover data. The aforementioned land cover data were sourced from the China Land Cover Dataset (CLCD) [41], with data selected from 2005 to 2020. SDE and CVO data were obtained from the National Bureau of Statistics’ annual provincial-level data, with statistics collected from 2005 to 2020.SDE and CVO data are based on provincial statistics, necessitating data disaggregation. This paper primarily employs a simple area-weighted method using auxiliary data [42,43] to disaggregate both datasets to a 10 KM grid scale. For SDE data, provincial-level data is decomposed to the 10 KM grid scale using the kernel density of factory POI points as weights. For CVO data, provincial-level data is decomposed to the 10 KM grid scale using road network density as weights. POI data and road network data were sourced from OpenStreetMap (OSM).
For PM2.5 data and these nine categories—NDVI, AR, MAT, GDP, PD, NTL, SDE, CVO—zonal statistics were conducted. The mean values were calculated at three scales: municipal, watershed, and grid. The three data categories—PFA, PWA, PISA—represent the area proportions of each land type within units at the three scales, calculated using CLCD [41]. This process was implemented using Arcpy (ArcGIS Pro-py3) and Python 3.9, generating datasets for all three scales covering 16 years from 2005 to 2020.

2.4. Method and Technique Process

This study’s experimental process primarily consists of three steps. First, data collection and preprocessing: PM2.5 data, influencing factor data, and auxiliary data were gathered to create datasets for subsequent analysis; Second, spatiotemporal analysis of annual average PM2.5 concentrations in the Yangtze River Economic Belt: Box plots and visualization techniques were employed to examine spatiotemporal variations, while the MK trend analysis method assessed the rates and significance of these changes; Third, analysis of factors influencing annual average PM2.5 concentrations in the Yangtze River Economic Belt. Geodetector were first employed at three scales—municipal, watershed, and grid (10 KM)—to identify key factors driving spatial variation in PM2.5 concentrations. Subsequently, GTWR analysis was conducted on several key factors to further examine the spatiotemporal heterogeneity of factors affecting annual average PM2.5 concentrations within the region. The technical workflow is illustrated in Figure 3 below.
The technical methods involved in this paper mainly include Mann–Kendall Trend Analysis, Geodetector, and GTWR.
The Mann–Kendall significance test is a trend analysis method independent of data distribution, characterized by high robustness and broad applicability. It can be used to validate the reliability of Theil–Sen Median slope estimates. This study performs pixel-by-pixel calculations on PM2.5 data from 2005 to 2020 to analyze its significance. Details regarding the methodological descriptions and formulas can be found in Appendix B.1.
The Geodetector, proposed by Wang Jinsong’s research group [44,45], is a statistical tool for measuring spatial hierarchical heterogeneity and exploring its driving forces. It primarily comprises four detectors: Factor Detector, Interaction Detector, Risk Detector, and Ecological Detector. This paper employs the Factor Detection, and the spatial distribution relationship between variables is measured by the statistic q-value (see Appendix B.2 for its formula). By calculating the q-values between PM2.5 concentrations and various influencing factors from 2005 to 2020 across three scales, the key influencing factors of PM2.5 concentrations were identified, laying a foundation for subsequent analysis. Details regarding the methodological descriptions and formulas can be found in Appendix B.2.
The GTWR is a regression model developed by Huang Bo [46] by incorporating temporal parameters into the Geographically Weighted Regression (GWR) [47] framework. In this study, GTWR was applied to the key factors derived from Geodetector and PM2.5 concentration data across three scales. This approach yielded the regression coefficients of each factor at different spatiotemporal locations. Subsequently, these regression coefficients were visualized to analyze their variation trends, and a comprehensive comparison was conducted to explore the similarities and differences in the variation trends across the three scales. Details regarding the methodological descriptions and formulas can be found in Appendix B.3.

3. Results

3.1. Spatial–Temporal Variability of PM2.5

The temporal variation in annual average PM2.5 concentrations along the Yangtze River Economic Belt is shown in Figure 4. Between 2005 and 2020, the temporal changes in PM2.5 concentrations within the region can be divided into two distinct phases. During the first phase (2005–2013), the annual average PM2.5 concentration remained at a relatively high level overall, exhibiting a fluctuating upward trend with the regional mean and median fluctuating between 45–49.5 μg/m3. In 2013, the average PM2.5 concentration reached its peak value of 49.5 μg/m3 within the 2005–2020 period. During the second phase, from 2014 to 2020, it exhibited a pronounced downward trend. The mean value gradually decreased, with annual averages predominantly concentrated in the lower range. By 2020, the regional average reached its lowest value within the 2005–2020 period at 25.0 μg/m3, indicating substantial improvement in PM2.5 pollution levels since 2013. This indicates that the Air Pollution Prevention and Control Action Plan, approved and implemented by the State Council in 2013, has achieved positive effects.
The spatial variation in annual average PM2.5 concentrations in the Yangtze River Economic Belt is shown in Figure 5. Between 2005 and 2020, PM2.5 pollution exhibited distinct clustering patterns within the region. Overall, PM2.5 pollution showed a distribution pattern characterized by lower levels in the west and higher levels in the central and eastern parts. High-pollution zones were primarily concentrated in the three major urban agglomerations: the Chengdu-Chongqing Urban Agglomeration, the Middle Yangtze River Urban Agglomeration, and the Yangtze River Delta Urban Agglomeration. These areas feature well-developed urban infrastructure, dense populations, and advanced industrialization. Low-pollution zones, primarily located in western Sichuan and Yunnan, consistently maintained PM2.5 concentrations below 35 μg/m3 annually. These regions predominantly comprise ethnic minority settlements with relatively underdeveloped urbanization, sparse populations, and low industrialization levels. From 2005 to 2013, the annual average PM2.5 concentration in the central and eastern parts of the Yangtze River Economic Belt consistently exceeded 35 μg/m3, with some areas reaching over 70 μg/m3. PM2.5 pollution peaked in 2013. In contrast, most western regions maintained annual averages below 35 μg/m3. From 2014 to 2020, PM2.5 pollution within the Yangtze River Economic Belt saw significant improvement. Annual average PM2.5 concentrations in the central and eastern regions showed a marked decline, with most areas falling below 50 μg/m3 by 2016. Western regions saw further declines in annual PM2.5 concentrations, with western Sichuan reaching below 10 μg/m3 by 2019. By 2020, most areas within the Yangtze River Economic Belt maintained annual PM2.5 concentrations below 35 μg/m3.
To further investigate the spatio-temporal change rates of annual average PM2.5 concentrations in the Yangtze River Economic Belt from 2005 to 2020 and determine whether these changes were significant, slope analysis and MK trend analysis were conducted on the annual average PM2.5 concentrations within the region during this period. The slope analysis results, as shown in Figure 6a, indicate that the annual average PM2.5 concentration change slope within the region fluctuated between −3.422 and 0.034 μg/(m3·a) from 2005 to 2020. Among these, Yunnan Province and western Sichuan exhibited a slow decline in PM2.5 concentrations, with slopes ranging from −1 to 0 μg/(m3·a). while the Sichuan Basin, parts of the middle Yangtze River region, and areas near Lake Tai saw faster declines with slopes below −2 μg/(m3·a). The remaining areas exhibited slopes between −2 and −1 μg/(m3·a). Only a few isolated points within the region exhibited PM2.5 concentration change rates > 0. Overall, the annual average PM2.5 concentration change rates in the Yangtze River Economic Belt from 2005 to 2020 were predominantly negative, showing a spatial trend of slower decline in the west and faster decline in the central and eastern regions. The MK trend analysis results are shown in Figure 6b. No significant increase in PM2.5 pollution was observed in the Yangtze River Economic Belt between 2005 and 2020. Most regions (67.45%) exhibited a significant decline in PM2.5 pollution exceeding 99%, while a very small portion (0.01%) showed no significant change. This latter category was primarily concentrated in western Sichuan, where PM2.5 pollution levels have consistently remained low, resulting in an insignificant decline trend.

3.2. Analysis of Factors Influencing PM2.5

3.2.1. Geodetector Results

The Geodetector in this paper was implemented using the R (4.4.2) package gdversep (1.3.3) [48]. Data discretization was performed automatically during factor detector by the opgd function, which selected an appropriate discretization method and number of categories. Figure 7 presents the geodetector factor detector results across three scales.
As shown in Figure 7a, at the municipal scale, the mean q-statistic values for factor detector results of these 11 explanatory factors over the 16-year period from 2005 to 2020, ranked from highest to lowest, were: PFA > PD > PISA > NDVI > PWA > SDE > CVO > GDP > MAT > NTL > AR. The NTL factor detection results showed p-values greater than 0.01 from 2005 to 2011, indicating no significant factor detector. For AR, p-values exceeded 0.01 in 2005, 2009, 2012 to 2017, 2019, and 2020, indicating no significant factor detector results. For all other factors, p-values were consistently below 0.01 across all years, demonstrating significant results. As shown in Figure 7c, at the watershed scale, the mean q-statistic values for factor detector results of these 11 explanatory factors over the 16-year period from 2005 to 2020, ranked from highest to lowest, were: PFA > PISA > PD > SDE > PWA > MAT > CVO > GDP > NTL > AR > NDVI. Moreover, the p-values for all 11 factor detector results from 2005 to 2020 were less than 0.01, indicating significant factor detector results. As shown in Figure 7e, at the grid scale, the mean q-statistic values for factor detector results of these 11 explanatory factors over the 16-year period from 2005 to 2020 were, in descending order: PFA > PISA > PD > SDE > MAT > CVO > PWA > GDP > AR > NDVI > NTL. Furthermore, the p-values for factor detector results of these 11 factors were all less than 0.01 from 2005 to 2020, indicating significant factor detector results.
Among the four natural environmental factors—NDVI, PWA, MAT, and AR. NDVI consistently ranked among the top two in q-value magnitude among the four natural environmental factors at the municipal scale (Figure 7b) during 2005–2020. However, at the watershed scale (Figure 7d) and grid scale (Figure 7f), its q-values consistently ranked among the bottom two. At the grid scale, it occupied the last position in most years, indicating that as the study scale becomes finer, NDVI’s explanatory power for spatial variation in regional annual average PM2.5 concentrations diminishes. The q-values of PWA consistently ranked among the top two across the three spatial scales from 2005 to 2020, but decreased from the municipal to the grid scale. These results indicate that PWA exhibits strong explanatory power for the spatial heterogeneity of the annual mean PM2.5 concentration within the region across all three scales, although its explanatory capacity diminishes as the research scale becomes more refined. AR exhibited significant fluctuations in its q-values across all three scales during 2005–2020, alternating between high and low values annually while showing an overall downward trend. For MAT, the ranking of its q-values among the four natural environmental factors progressively increased from the municipal scale to the grid scale—from the bottom two positions at the municipal scale to the top position at the grid scale. This indicates that the explanatory power of MAT for spatial variation in regional annual average PM2.5 concentrations increases as the study scale becomes finer.
Among the seven human activity-related factors—PFA, PD, PISA, NTL, CVO, SDE, and GDP. PFA, PISA, and PD consistently ranked among the top three in terms of q-value magnitude across all three spatial scales from 2005 to 2020. Specifically, PFA consistently ranked first in q-value magnitude at the municipal scale, while at the watershed and grid scales, its q-value magnitude was comparable to that of PISA, each taking the top position at their respective scales. Although both factors exhibited fluctuations in q-values magnitude at the municipal scale, their values increased only marginally from 2005 to 2020. At the watershed and grid scales, however, both factors showed a marked increase in q-values from 2005 to 2020. This indicates that at these two scales, the explanatory power of PFA and PISA for the spatial variation in annual average PM2.5 concentrations within the region increased over time. At the municipal scale, PD exhibited a declining trend in q-values from 2005 to 2020. However, at the watershed and grid scales, its q-values fluctuated without a significant decrease. This indicates that PD explanatory power for spatial variation in regional annual average PM2.5 concentrations declined over time at the municipal scale, while maintaining a consistently high level at the watershed and grid scales. The q-values of SDE exhibited a declining trend across all three spatial scales from 2005 to 2020, while the average q-values increased progressively from the municipal scale to the grid scale. This indicates that the explanatory power of SDE for the spatial heterogeneity of regional annual mean PM2.5 concentrations weakened over time but strengthened as the research scale became more refined. CVO exhibited a increasing trend in q-values across all three scales from 2005 to 2020, this indicates that CVO’s explanatory power for spatial variation in regional annual average PM2.5 concentrations is increasing. For GDP and NTL, their q-values consistently ranked among the bottom two across all three scales from 2005 to 2020, though both showed an upward trend. NTL exhibited a more pronounced increase, indicating that these factors maintain relatively low explanatory power for spatial variation in regional annual average PM2.5 concentrations, albeit with an upward trajectory in explanatory magnitude.
Overall, the results from the geodetector factor detector indicate that among the 11 factors studied in this paper, the spatial variation in the annual average PM2.5 concentration within the region is primarily influenced by three types of factors: PFA, PISA, and PD. AR and NTL exert relatively minor effects. Furthermore, the explanatory power of different factor types for the spatial variation in annual average PM2.5 within the region varies depending on the scale of the study, NDVI and PWA exerts a stronger influence at larger scale, while MAT and SDE exert greater influence at smaller scale.

3.2.2. GTWR Results

Geodetector can only determine the overall magnitude of influence factors on PM2.5 concentrations within a region, but cannot detect the direction of influence or the varying magnitudes of influence factors across different areas within the region. Therefore, this study employs GTWR to detect changes in the magnitude of influence across different geographic locations and time points for various influence factors within the region. The top six factors identified by the geodetector are selected as explanatory variables for GTWR: at the municipal scale, these are PFA, PD, PISA, NDVI, PWA, and SDE; at the watershed scale, they are PFA, PISA, PD, SDE, PWA, and MAT; and at the grid scale: PFA, PISA, PD, SDE, MAT, and CVO. Ultimately, eight explanatory variables were selected: PFA, PISA, PD, SDE, MAT, CVO, NDVI, and PWA.
This paper primarily implements GTWR using Python mgtwr package. Since different types of data possess varying unit dimensions, the dataset undergoes normalization prior to regression to unify dimensions. Before conducting GTWR, multicollinearity detection must be performed on each factor. The Variance Inflation Factor (VIF) (formula is provided in Appendix B.4) can characterize the degree of collinearity among independent variables, and its value reflects whether there is multicollinearity among the observed values of independent variables as well as the extent of such multicollinearity. The VIF are shown in Table 2, with all factor VIF values below 7. This indicates no multicollinearity exists among the variables across the three scales.
GTWR was conducted at three scales. After multiple experiments to select an appropriate bandwidth, the final bandwidth was set to 100,000 and the temporal bandwidth to 2. The adjusted R-squared (Adj-R2) at the municipal scale was 0.960, at the watershed scale 0.961, and at the grid scale 0.949 (as shown in Table 3).
Visualize the regression coefficients for each factor to intuitively display their spatiotemporal variations (see Appendix A for details).
The NDVI GTWR coefficients are shown in Appendix A.1. Temporally, they exhibit a decreasing trend across all three scales. From 2005, when nearly half the regions showed positive regression coefficients, to 2020, the vast majority of regions now exhibit negative coefficients. This indicates that over time, the NDVI’s suppression effect on PM2.5 within the region has progressively strengthened. Spatially, the distribution of regression coefficients exhibits consistency across all three scales. Regions with positive coefficients are primarily concentrated in the northeast (near Hefei, Nanjing, Shanghai, and Hangzhou), central (near Wuhan, Changsha, and Nanchang), and the southwest (near Guizhou and Kunming). However, differences exist between scales: at the municipal scale, Tibetan Autonomous Prefecture of Garzê and Aba Tibetan and Qiang Autonomous Prefecture exhibited positive regression coefficients from 2005 to 2011, whereas at the watershed and grid scales, this region consistently showed negative coefficients. Similarly, Shanghai demonstrated negative coefficients at the municipal scale but predominantly positive values at the grid scale. Overall, NDVI exerts a suppressing effect on PM2.5 pollution across most areas within the Yangtze River Economic Belt, indicating that vegetation planting can effectively mitigate PM2.5 pollution.
The MAT GTWR coefficients are shown in Appendix A.2. Overall, most regions exhibited positive values across all three scales from 2005 to 2011. Starting in 2012, negative values began appearing near Wuhan, Hefei, Changsha, and Nanchang. By 2020, a spatial distribution pattern emerged: positive values were concentrated in western areas near Chongqing, Chengdu, Kunming, and Guizhou, as well as eastern regions near Shanghai and Hangzhou; while negative values dominated central areas near Wuhan, Hefei, Changsha, and Nanchang. However, spatial distribution varies across scales. For instance, at the municipal and watershed scales, the northwest region exhibited positive regression coefficients in 2020, whereas some areas showed negative values at the grid scale. Overall, by 2020, MAT will promote PM2.5 pollution in the western and eastern coastal areas of the region while suppressing PM2.5 pollution in the central and eastern parts.
The PWA GTWR coefficients are shown in Appendix A.3. Overall, a decrease is observed across all three scales. In 2005, most areas within the region exhibited positive regression coefficients across all three scales. By 2020, the distribution pattern at the Municipal scale showed positive coefficients in the western and northeastern regions, with negative coefficients elsewhere. At the watershed and grid scales, positive coefficients were observed in the northwestern, central-southern, and northeastern regions, while negative coefficients prevailed in the remaining areas.
The PFA GTWR coefficients are shown in Appendix A.4. At the municipal scale, these coefficients exhibit a decreasing trend, shifting from predominantly positive values across most regions in 2005 to negative values appearing in the northwest, near Chengdu, and near Chongqing by 2020. At both the watershed and grid scales, regression coefficients also show a decreasing trend, though they remain predominantly positive from 2005 to 2020. Negative values were observed only in a few areas, primarily concentrated near Chengdu and Chongqing, as well as in western Changsha. The results indicate that at the municipal scale, the PFA exerts a more pronounced inhibitory effect on PM2.5 pollution. However, at the watershed and grid scales, PFA continues to primarily promote PM2.5 pollution. Yet, since crops are also considered vegetation, and an increase in farmland proportion should lead to increased vegetation, thereby increasing NDVI values and suppressing PM2.5 pollution. The opposite effect observed may be related to the increase in polluting agricultural activities associated with increased agricultural land use, such as straw burning and the use of agricultural machinery. This ultimately results in PFA predominantly promoting PM2.5 pollution at the watershed and grid scales.
The PISA GTWR coefficients are shown in Appendix A.5. Overall, a decreasing trend is observed across all three scales. While over half of the regions exhibited positive regression coefficients in 2005, only a small portion maintained positive coefficients by 2020, primarily concentrated in the northeastern region and a limited area in the west. This indicates that PISA contribution to PM2.5 pollution has diminished over time. The proportion of PISA reflects a municipal scale of urban development. Results indicate that during the early stages of development, an increase in PISA does indeed exacerbate PM2.5 pollution. However, over time, the contribution of PISA to PM2.5 pollution has gradually diminished, and in some cases, even reduced PM2.5 pollution. This may be related to increased attention to environmental issues and growing emphasis on urban greening during the urban development process.
The PD GTWR coefficients are shown in Appendix A.6. Temporally, a slight increasing trend is observed at the municipal scale, while no significant changes are evident at the watershed and grid scales. Spatially, the regression coefficients predominantly exhibit positive values across the region, primarily distributed near Shanghai, Hangzhou, Wuhan, Nanchang, Changsha, Chengdu, Chongqing, Guizhou, and Kunming, while negative values are concentrated in the northwest, central-north, and northeast regions near Hefei and Nanjing. The results indicate that PD primarily exerts a promoting effect on regional PM2.5 pollution.
The SDE GTWR coefficients, as shown in Appendix A.7, exhibit a pronounced increasing trend across all three scales. From predominantly negative values across most regions in 2005, the majority of regions turned positive by 2020. The results indicate that over time, the enhancing effect of SDE on PM2.5 pollution has gradually increased, becoming predominant by 2020.
The CVO GTWR coefficients, as shown in Appendix A.8, exhibit a trend of decreasing followed by increasing across all three scales. From predominantly positive values in 2005, they shifted to predominantly negative values by 2020. This indicates that over time, the contribution of CVO to PM2.5 pollution has been diminishing. This trend may be related to the vigorous promotion of new energy vehicles and the widespread adoption of public transportation.
In summary, among the three natural environmental factors—NDVI, MAT, and PWA—NDVI primarily exerts an inhibitory effect on PM2.5 pollution within the region. An increase in NDVI helps mitigate PM2.5 pollution. Both MAT and PWA exhibit significant spatial variations in their impact on PM2.5 pollution, with inhibitory and promoting regions each accounting for approximately half of the area. Among the five human activity-related factors—PFA, PISA, PD, SDE, and CVO—PFA, PD, and SDE primarily exert a promoting effect on PM2.5 pollution within the region. PISA and CVO predominantly promoted PM2.5 pollution during the early period from 2005 to 2020, but their promoting effect diminished over time. By 2020, their distribution range had narrowed, with most areas showing a predominantly inhibitory effect. The spatial distribution of influence for the same factor was broadly consistent across the three scales, though certain differences existed in the details.

4. Discussion

4.1. Multi-Scale Factor Analysis

Analysis of influencing factors at the municipal, watershed, and grid scales reveals that across all three scales, the key factors driving spatial variation in PM2.5 concentrations are PFA, PISA, and PD. The impacts of NTL and AR factors are relatively minor. Compared to natural environmental factors, human activity-related factors exert a greater influence, though differences exist across scales. NDVI exhibits greater explanatory power at the city level but less so at the watershed and grid scales. MAT and SDE show lower explanatory power at the city level but higher at the watershed and grid scales. PWA demonstrates greater explanatory power at the city and watershed scales but less at the grid scale. This indicates that NDVI and PWA primarily influence PM2.5 concentrations at large scales, while MAT and SDE exert greater effects at small scales. The spatiotemporal distribution of these factors’ impacts on PM2.5 concentrations across the three scales is broadly consistent. NDVI and CVO predominantly exhibit negative correlations with PM2.5 concentrations; MAT, PFA, PD, and SDE primarily show positive correlations. PWA and PISA exert dual effects on PM2.5 concentrations within the region, each covering approximately half the area. However, the influence of these factors varies across different scales. At the city scale, PFA exhibits a negative correlation with PM2.5 concentrations in more areas by 2020. This may be attributed to NDVI’s stronger explanatory power at larger scales, as increased PFA may promote crop cultivation—which is also considered vegetation—and vegetation growth can suppress PM2.5 pollution. Thus, PFA demonstrates a more pronounced inhibitory effect at the city scale. At the watershed scale, PWA shows a negative correlation with PM2.5 concentration in more areas, while at the city level, it exhibits a positive correlation in more areas. Other factors reveal minor variations in detail. Smaller scales yield more refined results and provide richer insights. For instance, the city level only reveals the overall impact across the entire city, whereas the grid scale allows for distinguishing differences between urban and suburban areas within the city.

4.2. Policy Recommendations

Based on the research findings, this paper proposes the following recommendations for addressing PM2.5 pollution in the Yangtze River Economic Belt: (1) Given the pronounced clustering characteristics of PM2.5 pollution within the region, provinces, cities, and districts must collaborate closely during remediation efforts. They should actively explore coordinated mechanisms for regional joint prevention and control to achieve synergistic spatial effects. (2) NDVI can effectively mitigate PM2.5 pollution. Therefore, rational urban greening, enhanced protection of forest and grassland resources, and strict prohibition of indiscriminate logging can alleviate PM2.5 pollution. (3) PFA primarily promotes PM2.5 pollution at watershed and grid scales. Thus, agricultural activities like crop planting and harvesting must strictly control polluting practices such as straw burning. During mechanized operations, exhaust emissions should be reasonably managed by selecting clean-type machinery. (4) SED significantly promotes PM2.5 pollution. Strictly control SO2 emissions; treat polluting gases before discharge. Enterprises and factories should improve pollutant treatment systems and discharge pollutants legally and reasonably. (5) For regions with severe PM2.5 pollution, such as Chengdu, Zigong, Wuhan, and Xuzhou, it is advisable to strengthen regulatory oversight, strictly control the emission of polluting gases, develop public transportation, and encourage residents to utilize public transport or adopt green travel modes.

4.3. Shortcomings and Outlook

In this paper, an analytical approach encompassing global-to-local analysis was established by integrating the Geodetector and GTWR methods. Meanwhile, research at both the watershed and grid scale was introduced, which breaks through the limitation of previous studies that primarily adopted administrative divisions as the basic research unit. The inclusion of the watershed and grid scales is not only intended to obtain more refined results but also driven by the fact that irregular research units may exert certain impacts on the study. In contrast, the grid scale features research units that are regular and uniform in size, thereby effectively eliminating the influences caused by variations in the shape and size of research units. It is hoped that this approach may provide a new research methodology and perspective for future related studies.
However, this study has certain limitations. In terms of data selection, all remote sensing data utilized in this study are derived from publicly available datasets on the internet (see Table 1 for details). Specifically, the PM2.5 [28,29] data yield a cross-validation coefficient of determination (CV-R2) ranging from 0.86 to 0.90; the overall accuracy of the Chinese land cover data [41] reaches 79.31%; the AR [30,31,32,33,34] data exhibit an observational bias of approximately −0.5 mm, while the MAT [30,31,32,33,35] data show a bias of around 0.01 °C relative to observations; the NTL [40] data achieve a coefficient of determination (R2) of about 0.95; for the GDP [36,37,38] data, the normalized validation R2 is 0.998 and the test R2 is 0.996. The two types of statistical data, namely SDE and CVO, employed in this research are at the provincial scale. To meet the requirements of the grid scale study, this paper approximates their spatial distribution through simple spatial disaggregation based on auxiliary data. Although the selected data demonstrate high accuracy, they still contain certain inherent errors that may exert a certain impact on the research outcomes. Regarding the selection of influencing factors, to meet the requirements of the 10 km grid scale, the currently selected factors cannot cover all aspects, resulting in an incomplete analysis of the factors affecting PM2.5 concentrations. Regarding the study period, due to the inherent lag in remote sensing data and statistical data, the temporal coverage could not extend to recent years, limiting the analysis to data from 2005 to 2020. In terms of research methods, the GTWR employed in this study is a linear model. Linear models have inherent limitations in fully elucidating the complex relationships between PM2.5 concentrations and various influencing factors. Furthermore, PM2.5 pollution is the result of the combined effects of multiple factors, while this study mainly focuses on the role of single factors and lacks an analysis of the interactive effects among different factors. This study primarily focuses on the effects of individual factors and lacks an analysis of the interactive effects among various factors. Meanwhile, the research methods employed in this paper can only quantify the magnitude of each factor’s impact on PM2.5, without conducting an in-depth exploration of the underlying mechanisms of action.
Accordingly, for future research, the following improvements should be considered: in terms of data selection, efforts should be made to prioritize high-quality datasets; in factor selection, a more comprehensive range of influencing factors ought to be included; and the temporal scope of the analysis should be extended as much as possible. Regarding analytical methodologies, more sophisticated algorithms such as machine learning or deep learning could be employed to fully account for the mutual coupling effects among various influencing factors, while conducting a more in-depth exploration of the underlying mechanisms through which each factor operates.

5. Conclusions

This study analyzes the annual average PM2.5 concentrations in the Yangtze River Economic Belt from 2005 to 2020 through time-series analysis, spatial distribution visualization, and MK trend analysis, revealing its spatiotemporal evolution patterns. Using geodetector, PM2.5 influencing factors were analyzed at three scales—municipal, watershed, and grid—to identify key factors. The top-ranked factors were selected and analyzed using the GTWR model to further reveal the spatiotemporal heterogeneity of each influencing factor across the three scales. The following conclusions were drawn:
  • From 2005 to 2020, the annual average PM2.5 concentration in the Yangtze River Economic Belt exhibited an inverted U-shaped trend over time, with 2013 serving as the inflection point. Spatially, the annual average PM2.5 concentration in the Yangtze River Economic Belt showed a distribution pattern of lower levels in the west and higher levels in the central and eastern regions, revealing distinct agglomeration characteristics.
  • From 2005 to 2020, the annual average PM2.5 concentration in the Yangtze River Economic Belt showed a significant overall downward trend in both spatiotemporal variations, with no regions exhibiting a notable increase. PM2.5 concentrations decreased slowly in the western region, while they declined more rapidly in the central and eastern regions.
  • At the municipal, watershed, and grid scales, the spatial variation in annual average PM2.5 concentrations along the Yangtze River Economic Belt is primarily influenced by three factors: PFA, PISA, and PD. The impacts of NTL and AR are relatively minor. NDVI and PWA exert a stronger influence at larger scale, while MAT and SDE exert greater influence at smaller scale. Human activity-related factors exert a greater influence on the spatial variation in PM2.5 concentrations within the region.
  • NDVI, CVO, and PM2.5 concentration primarily exhibit a negative correlation; MAT, PFA, PD, and SDE primarily exhibit a positive correlation with PM2.5 concentration; and PWA and PISA exert dual effects on PM2.5 concentrations within the region, each covering approximately half the area. At different scales, the spatial distribution of the same factor’s impact on PM2.5 concentrations is largely consistent, though some variations exist. A smaller scale would yield more refined results.

Author Contributions

Conceptualization, Y.Z. and Y.C.; Methodology, Y.Z. and Y.C.; Software, Y.Z. and Y.C.; Validation, Y.Z. and Y.C.; Formal analysis, Y.Z. and Y.C.; Investigation, Y.Z. and Y.C.; Data curation, Y.Z. and Y.C.; Writing—original draft, Y.Z.; Writing—review & editing, Y.C. and Y.W.; Visualization, Y.Z.; Supervision, Y.C. and Y.W.; Funding acquisition, Y.C. All authors have read and agreed to the published version of the manuscript.

Funding

The research was supported by the International Research Center of Big Data for Sustainable Development Goals (CBAS) [No. CBASYX0906], the key project of sustainable development international cooperation program by NSFC [No.42361144883] and the National Natural Science Foundation of China [No.42271422].

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

The original contributions presented in this study are included in the article. Further inquiries can be directed to the corresponding author(s).

Conflicts of Interest

The authors declare no conflicts of interest.

Appendix A

Appendix A.1

Sustainability 17 09721 g0a1
Figure A1. Spatio-temporal Distribution Map of NDVI Regression Coefficients at the Municipal scale.
Figure A1. Spatio-temporal Distribution Map of NDVI Regression Coefficients at the Municipal scale.
Sustainability 17 09721 g0a1
Sustainability 17 09721 g0a2
Figure A2. Spatio-temporal Distribution Map of NDVI Regression Coefficients at the Watershed scale.
Figure A2. Spatio-temporal Distribution Map of NDVI Regression Coefficients at the Watershed scale.
Sustainability 17 09721 g0a2
Sustainability 17 09721 g0a3
Figure A3. Spatio-temporal Distribution Map of NDVI Regression Coefficients at the Grid scale.
Figure A3. Spatio-temporal Distribution Map of NDVI Regression Coefficients at the Grid scale.
Sustainability 17 09721 g0a3

Appendix A.2

Sustainability 17 09721 g0a4
Figure A4. Spatio-temporal Distribution Map of MAT Regression Coefficients at the Municipal scale.
Figure A4. Spatio-temporal Distribution Map of MAT Regression Coefficients at the Municipal scale.
Sustainability 17 09721 g0a4
Sustainability 17 09721 g0a5
Figure A5. Spatio-temporal Distribution Map of MAT Regression Coefficients at the Watershed scale.
Figure A5. Spatio-temporal Distribution Map of MAT Regression Coefficients at the Watershed scale.
Sustainability 17 09721 g0a5
Sustainability 17 09721 g0a6
Figure A6. Spatio-temporal Distribution Map of MAT Regression Coefficients at the Grid scale.
Figure A6. Spatio-temporal Distribution Map of MAT Regression Coefficients at the Grid scale.
Sustainability 17 09721 g0a6

Appendix A.3

Sustainability 17 09721 g0a7
Figure A7. Spatio-temporal Distribution Map of PWA Regression Coefficients at the Municipal scale.
Figure A7. Spatio-temporal Distribution Map of PWA Regression Coefficients at the Municipal scale.
Sustainability 17 09721 g0a7
Sustainability 17 09721 g0a8
Figure A8. Spatio-temporal Distribution Map of PWA Regression Coefficients at the Watershed scale.
Figure A8. Spatio-temporal Distribution Map of PWA Regression Coefficients at the Watershed scale.
Sustainability 17 09721 g0a8
Sustainability 17 09721 g0a9
Figure A9. Spatio-temporal Distribution Map of PWA Regression Coefficients at the Grid scale.
Figure A9. Spatio-temporal Distribution Map of PWA Regression Coefficients at the Grid scale.
Sustainability 17 09721 g0a9

Appendix A.4

Sustainability 17 09721 g0a10
Figure A10. Spatio-temporal Distribution Map of PFA Regression Coefficients at the Municipal scale.
Figure A10. Spatio-temporal Distribution Map of PFA Regression Coefficients at the Municipal scale.
Sustainability 17 09721 g0a10
Sustainability 17 09721 g0a11
Figure A11. Spatio-temporal Distribution Map of PFA Regression Coefficients at the Watershed scale.
Figure A11. Spatio-temporal Distribution Map of PFA Regression Coefficients at the Watershed scale.
Sustainability 17 09721 g0a11
Sustainability 17 09721 g0a12
Figure A12. Spatio-temporal Distribution Map of PFA Regression Coefficients at the Grid scale.
Figure A12. Spatio-temporal Distribution Map of PFA Regression Coefficients at the Grid scale.
Sustainability 17 09721 g0a12

Appendix A.5

Sustainability 17 09721 g0a13
Figure A13. Spatio-temporal Distribution Map of PISA Regression Coefficients at the Municipal scale.
Figure A13. Spatio-temporal Distribution Map of PISA Regression Coefficients at the Municipal scale.
Sustainability 17 09721 g0a13
Sustainability 17 09721 g0a14
Figure A14. Spatio-temporal Distribution Map of PISA Regression Coefficients at the Watershed scale.
Figure A14. Spatio-temporal Distribution Map of PISA Regression Coefficients at the Watershed scale.
Sustainability 17 09721 g0a14
Sustainability 17 09721 g0a15
Figure A15. Spatio-temporal Distribution Map of PISA Regression Coefficients at the Grid scale.
Figure A15. Spatio-temporal Distribution Map of PISA Regression Coefficients at the Grid scale.
Sustainability 17 09721 g0a15

Appendix A.6

Sustainability 17 09721 g0a16
Figure A16. Spatio-temporal Distribution Map of PD Regression Coefficients at the Municipal scale.
Figure A16. Spatio-temporal Distribution Map of PD Regression Coefficients at the Municipal scale.
Sustainability 17 09721 g0a16
Sustainability 17 09721 g0a17
Figure A17. Spatio-temporal Distribution Map of PD Regression Coefficients at the Watershed scale.
Figure A17. Spatio-temporal Distribution Map of PD Regression Coefficients at the Watershed scale.
Sustainability 17 09721 g0a17
Sustainability 17 09721 g0a18
Figure A18. Spatio-temporal Distribution Map of PD Regression Coefficients at the Grid scale.
Figure A18. Spatio-temporal Distribution Map of PD Regression Coefficients at the Grid scale.
Sustainability 17 09721 g0a18

Appendix A.7

Sustainability 17 09721 g0a19
Figure A19. Spatio-temporal Distribution Map of SDE Regression Coefficients at the Municipal scale.
Figure A19. Spatio-temporal Distribution Map of SDE Regression Coefficients at the Municipal scale.
Sustainability 17 09721 g0a19
Sustainability 17 09721 g0a20
Figure A20. Spatio-temporal Distribution Map of SDE Regression Coefficients at the Watershed scale.
Figure A20. Spatio-temporal Distribution Map of SDE Regression Coefficients at the Watershed scale.
Sustainability 17 09721 g0a20
Sustainability 17 09721 g0a21
Figure A21. Spatio-temporal Distribution Map of SDE Regression Coefficients at the Grid scale.
Figure A21. Spatio-temporal Distribution Map of SDE Regression Coefficients at the Grid scale.
Sustainability 17 09721 g0a21

Appendix A.8

Sustainability 17 09721 g0a22
Figure A22. Spatio-temporal Distribution Map of CVO Regression Coefficients at the Municipal scale.
Figure A22. Spatio-temporal Distribution Map of CVO Regression Coefficients at the Municipal scale.
Sustainability 17 09721 g0a22
Sustainability 17 09721 g0a23
Figure A23. Spatio-temporal Distribution Map of CVO Regression Coefficients at the Watershed scale.
Figure A23. Spatio-temporal Distribution Map of CVO Regression Coefficients at the Watershed scale.
Sustainability 17 09721 g0a23
Sustainability 17 09721 g0a24
Figure A24. Spatio-temporal Distribution Map of CVO Regression Coefficients at the Grid scale.
Figure A24. Spatio-temporal Distribution Map of CVO Regression Coefficients at the Grid scale.
Sustainability 17 09721 g0a24

Appendix B

Appendix B.1

The Mann–Kendall significance test is a trend analysis method independent of data distribution, characterized by high robustness and broad applicability. It can be used to validate the reliability of Theil–Sen Median slope estimates. The calculation formula is as follows:
Z M K = S − 1 n n − 1 2 n + 5 18 , S > 0 S + 1 n n − 1 2 n + 5 18 , S < 0
s = ∑ k = 1 n − 1 ∑ j = k + 1 n s g n x j − x k
s g n x j − x k = 1 , x j − x k > 0 0 , x j − x k = 0 − 1 , x j − x k < 0
In the formula, S is the Mann–Kendall test statistic, x j and x k represent the inverted results of PM2.5 concentrations for years j and k respectively, n denotes the length of the data series, and sgn is the sign function. Z M K is the standardized test statistic, where the positive or negative sign indicates the direction of change in PM2.5 concentrations during the study period. A larger absolute value indicates a more significant trend. When the absolute value of Z M K exceeds 1.96, it indicates that the trend in PM2.5 concentration during the study period is statistically significant at the 95% confidence level. When the absolute value of Z M K exceeds 2.58, it indicates that the trend in PM2.5 concentration during the study period is statistically significant at the 99% confidence level.

Appendix B.2

The Geodetector [44,45], proposed by Wang Jinsong’s research group, is a statistical tool for measuring spatial hierarchical heterogeneity and exploring its driving forces. It primarily comprises four detectors: Factor Detector, Interaction Detector, Risk Detector, and Ecological Detector. The advantages of the Geodetector are (1) It can examine the influence of individual variables on the dependent variable, offering greater reliability than correlation coefficients; (2) It can investigate the strength of mutual influence among multiple variables. (3) Simultaneously examining multiple factors’ effects on the dependent variable. Originally applied to spatial health risk assessment, geodetector are now widely used across ecological environments, meteorology and hydrology, and socioeconomic fields. They are extensively employed in air quality assessment, particularly regarding PM2.5 influencing factors.
q = 1 − ∑ h = 1 L N h σ h 2 N σ 2 = 1 − S S W S S T
S S W = ∑ h = 1 L N h σ h 2
S S T = N σ 2
Among these, h denotes the stratification or partitioning of the dependent variable, q represents the explanatory power of the PM2.5 influencing factor, N h and N denote the influencing factor h and the total number of influencing factors, respectively; σ 2 and σ h 2 denote the overall sample size of the study area and the variance of stratification h.
When considering a single factor, the magnitude of the q-value indicates the strength of that factor’s influence on the dependent variable Y. When multiple factors are considered, the q-value represents the relative strength of each factor’s influence on Y, with its numerical value being proportional to the influence. When q = 1 indicates that the spatial distribution of Y is completely determined by the spatial distribution of X, while q = 0 indicates no spatial differentiation effect between X and Y. The distribution of q-values follows a skewed F-distribution.
F = N − 1 L − 1 q 1 − q ~ F L − 1 , N − 1 ; λ
λ = 1 σ 2 ∑ h = 1 L y ¯ h 2 − 1 h ∑ h = 1 L N h y ¯ h 2
where λ is the asymmetry parameter and y ¯ h 2 is the stratified mean.

Appendix B.3

GWR [47] is a modeling approach tailored for local characteristics. Its core principle lies in the fact that regression coefficients are not fixed but vary with geographical location. This model represents a deepening and extension of traditional linear regression by introducing a spatial dimension, assigning a distinct set of local variable regression coefficients to each sample point. Specifically, the closer a sample point is geographically to the fitted points, the greater its influence (i.e., weight) on the regression coefficients. This design enables GWR to capture non-uniform spatial variations in parameters, thereby demonstrating superior performance compared to traditional linear regression models when handling spatially heterogeneous data.
The fundamental formula for the Geographically Weighted Regression model is
y i = β 0 u i , v i + ∑ k = 1 p β k u i , v i x i k + ε i                 i = 1,2 , … , n
In the equation, u i , v i denotes the coordinates of the i-th sample point; y i represents the dependent variable of the i-th sample point; x i k indicates the k-th explanatory variable of the i-th sample point; β k u i , v i denotes the regression coefficient for the k-th explanatory variable of the i-th sample point; β 0 u i , v i denotes the regression constant for the i-th sample point; ε i ~ N ( 0 , σ 2 ) denotes the model error.
The GTWR model [46] is a regression model developed by Huang Bo by incorporating temporal parameters into the GWR framework. It employs spatial data, time-series data, and a weight matrix to perform local regression, thereby exploring the heterogeneous effects of explanatory variables on the response variable across different regions and time periods. It can accurately quantify the influence of spatial locations or temporal points and predict future trends by incorporating weights. The GTWR model primarily estimates parameters through iterative weighted least squares. The specific principle formula is as follows:
y i = β 0 u i , v i , t i + ∑ k = 1 p β k u i , v i , t i x i k + ε i                 i = 1,2 , … , n
In the equation, u i , v i , t i denotes the spatio-temporal coordinates of the i-th sample point, where t i is the observation time; y i denotes the dependent variable for the i-th sample point; x i k denotes the k-th explanatory variable for the i-th sample point; β k u i , v i , t i x i k denotes the regression coefficient for the k-th explanatory variable at the i-th sample point; β 0 ( u i , v i , t i ) denotes the regression constant for the i-th sample point; ε i ~ N ( 0 , σ 2 ) represents the model error.

Appendix B.4

Multicollinearity refers to the existence of a certain linear relationship among explanatory variables in a multiple regression model. The VIF can characterize the degree of collinearity among independent variables, and its value reflects whether there is multicollinearity among the observed values of independent variables as well as the extent of such multicollinearity. The formula is presented as follows:
V I F i = 1 1 − R i
where R i denotes the goodness of fit obtained by regressing the i-th explanatory variable (treated as the dependent variable) against the other explanatory variables. The higher the degree of collinearity between x i and other explanatory variables, the larger the value of R i , and consequently the larger the value of VIF. In empirical analysis, a common criterion is that if VIF > 10, the i-th explanatory variable is considered to potentially have a severe collinearity issue with other explanatory variables.

References

  1. Lin, G.; Fu, J.; Jiang, D.; Hu, W.; Dong, D.; Huang, Y.; Zhao, M. Spatio-Temporal Variation of PM2.5 Concentrations and Their Relationship with Geographic and Socioeconomic Factors in China. Int. J. Environ. Res. Public. Health 2014, 11, 173–186. [Google Scholar] [CrossRef] [PubMed]
  2. Yousefi, S.; Shahsavani, A.; Hadei, M. Applying EPA’s instruction to calculate air quality index (AQI) in Tehran. J. Air Pollut. Health 2019, 4, 81–96. [Google Scholar] [CrossRef]
  3. Zhu, S.; Lian, X.; Wei, L.; Che, J.; Shen, X.; Yang, L.; Qiu, X.; Liu, X.; Gao, W.; Ren, X.; et al. PM2.5 forecasting using SVR with PSOGSA algorithm based on CEEMD, GRNN and GCA considering meteorological factors. Atmos. Environ. 2018, 183, 20–32. [Google Scholar] [CrossRef]
  4. Zhou, C.; Chen, J.; Wang, S. Examining the effects of socioeconomic development on fine particulate matter (PM2.5) in China’s cities using spatial regression and the geographical detector technique. Sci. Total Environ. 2018, 619–620, 436–445. [Google Scholar] [CrossRef]
  5. Juarez, P.D.; Tabatabai, M.; Burciaga Valdez, R.; Hood, D.B.; Im, W.; Mouton, C.; Colen, C.; Al-Hamdan, M.Z.; Matthews-Juarez, P.; Lichtveld, M.Y.; et al. The Effects of Social, Personal, and Behavioral Risk Factors and PM2.5 on Cardio-Metabolic Disparities in a Cohort of Community Health Center Patients. Int. J. Environ. Res. Public Health 2020, 17, 3561. [Google Scholar] [CrossRef] [PubMed]
  6. Cao, Q.; Liang, Y.; Niu, X. China’s Air Quality and Respiratory Disease Mortality Based on the Spatial Panel Model. Int. J. Environ. Res. Public Health 2017, 14, 1081. [Google Scholar] [CrossRef] [PubMed]
  7. Han, X.; Liu, Y.; Gao, H.; Ma, J.; Mao, X.; Wang, Y.; Ma, X. Forecasting PM2.5 induced male lung cancer morbidity in China using satellite retrieved PM2.5 and spatial analysis. Sci. Total Environ. 2017, 607–608, 1009–1017. [Google Scholar] [CrossRef]
  8. Yun, G.; He, Y.; Jiang, Y.; Dou, P.; Dai, S. PM2.5 Spatiotemporal Evolution and Drivers in the Yangtze River Delta between 2005 and 2015. Atmosphere 2019, 10, 55. [Google Scholar] [CrossRef]
  9. Yan, J.-W.; Tao, F.; Zhang, S.-Q.; Lin, S.; Zhou, T. Spatiotemporal Distribution Characteristics and Driving Forces of PM2.5 in Three Urban Agglomerations of the Yangtze River Economic Belt. Int. J. Environ. Res. Public Health 2021, 18, 2222. [Google Scholar] [CrossRef]
  10. He, J.; Gong, S.; Yu, Y.; Yu, L.; Wu, L.; Mao, H.; Song, C.; Zhao, S.; Liu, H.; Li, X.; et al. Air pollution characteristics and their relation to meteorological conditions during 2014–2015 in major Chinese cities. Environ. Pollut. 2017, 223, 484–496. [Google Scholar] [CrossRef]
  11. Guo, X.; Mu, X.; Ding, Z.; Qin, D. Nonlinear effects and driving mechanism of multidimensional urbanization on PM2.5 concentrations in the Yangtze River Delta. Acta Geogr. Sin. 2021, 76, 1274–1293. [Google Scholar] [CrossRef]
  12. Huang, H.; Jiang, P.; Chen, Y. Analysis of the Social and Economic Factors Influencing PM2.5 Emissions at the City Level in China. Sustainability 2023, 15, 16335. [Google Scholar] [CrossRef]
  13. Xu, Y.; Guo, Z.; Zheng, Z.; Dai, Q.; Zhao, C.; Huang, W. Study of the PM2.5 Concentration Variation and its Influencing Factors in the Beijing-Tianjin-Hebei Urban Agglomeration Using Geo-Detector. Res. Environ. Sci. 2023, 36, 649–659. [Google Scholar] [CrossRef]
  14. Li, S.; Wei, G.; Liu, Y.; Bai, L. Multi-Scale Analysis of PM2.5 Concentrations in the Yangtze River Economic Belt: Investigating the Combined Impact of Natural and Human Factors. Remote Sens. 2023, 15, 3356. [Google Scholar] [CrossRef]
  15. Fang, K.; Wang, T.; He, J.; Wang, T.; Xie, X.; Tang, Y.; Shen, Y.; Xu, A. The distribution and drivers of PM2.5 in a rapidly urbanizing region: The Belt and Road Initiative in focus. Sci. Total Environ. 2020, 716, 137010. [Google Scholar] [CrossRef]
  16. Liang, M.; Li, H.; Li, W.; Jia, X.; Wang, M.; Feng, C. Spatial-Temporal Distribution Characteristics of PM2.5 and PM10 in the Northeast Tibetan Plateau and the Influence of Meteorological Factors. Northwest. Geol. 2024, 57, 285–296. [Google Scholar] [CrossRef]
  17. Fu, Z.; Li, R. The contributions of socioeconomic indicators to global PM2.5 based on the hybrid method of spatial econometric model and geographical and temporal weighted regression. Sci. Total Environ. 2020, 703, 135481. [Google Scholar] [CrossRef] [PubMed]
  18. Zhao, R.; Zhan, L.; Yao, M.; Yang, L. A geographically weighted regression model augmented by Geodetector analysis and principal component analysis for the spatial distribution of PM2.5. Sustain. Cities Soc. 2020, 56, 102106. [Google Scholar] [CrossRef]
  19. Lan, T.; Han, L.; Tian, J.; Qi, C.; Xiao, Q. Important factors affecting the multi-scale variation of PM2.5 and O3 in Xingtai City and their contributions. China Environ. Sci. 2024, 44, 4786–4795. [Google Scholar] [CrossRef]
  20. Huang, C.; Lo, K.; Wu, Y.; Wang, F.; Shiu, Y.; Chen, C.; Lin, Y.; Kuo, C.; Liao, H.; Lin, T.; et al. Estimating and characterizing spatiotemporal distributions of elemental PM2.5 using an ensemble machine learning approach in Taiwan. Atmos. Pollut. Res. 2025, 16, 102463. [Google Scholar] [CrossRef]
  21. Wang, S.; Ren, Y.; Xia, B. PM2.5 and O3 concentration estimation based on interpretable machine learning. Atmos. Pollut. Res. 2023, 14, 101866. [Google Scholar] [CrossRef]
  22. Liao, D.; Hong, Y.; Huang, H.; Choi, S.; Zhuang, Z. Machine learning exploring the chemical compositions characteristics and sources of PM2.5 from reduced on-road activity. Atmos. Pollut. Res. 2024, 15, 102265. [Google Scholar] [CrossRef]
  23. Lundberg, S.M.; Erion, G.; Chen, H.; DeGrave, A.; Prutkin, J.M.; Nair, B.; Katz, R.; Himmelfarb, J.; Bansal, N.; Lee, S.I. From local explanations to global understanding with explainable AI for trees. Nat. Mach. Intell. 2020, 2, 56–67. [Google Scholar] [CrossRef] [PubMed]
  24. Kim, M.; Kim, S.; Lee, K.; Kim, P.; Kim, H.; Song, H. Development and interpretation of PM2.5 estimation model for the Seoul Metropolitan Area using machine learning and explainable AI. Atmos. Pollut. Res. 2025, 16, 102672. [Google Scholar] [CrossRef]
  25. He, J.; Jing, Y.; Ran, D. Is There a Relationship between Increased Land-Use Intensity and the Rise in PM2.5 Pollution Levels in the Yangtze River Economic Belt, China (2000–2021)? Atmosphere 2023, 14, 1097. [Google Scholar] [CrossRef]
  26. He, Q.; Kong, F.; Wen, C. Evolution of Harmonious Coexistence Between Humans and Nature in the Yangtze River Economic Belt: A Production-Living-Ecological Balance Perspective. J. Hubei Univ. Econ. 2020, 18, 109–119. [Google Scholar] [CrossRef]
  27. Lehner, B.; Grill, G. Global river hydrography and network routing: Baseline data and new approaches to study the world’s large river systems. Hydrol. Process. 2013, 27, 2171–2186. [Google Scholar] [CrossRef]
  28. Wei, J.; Li, Z.; Lyapustin, A.; Sun, L.; Peng, Y.; Xue, W.; Su, T.; Cribb, M. Reconstructing 1-km-resolution high-quality PM2.5 data records from 2000 to 2018 in China: Spatiotemporal variations and policy implications. Remote Sens. Environ. 2021, 252, 112136. [Google Scholar] [CrossRef]
  29. Wei, J.; Li, Z.; Cribb, M.; Huang, W.; Xue, W.; Sun, L.; Guo, J.; Peng, Y.; Li, J.; Lyapustin, A.; et al. Improved 1 km resolution PM2.5 estimates across China using enhanced space-time extremely randomized trees. Atmos. Chem. Phys. 2020, 20, 3273–3289. [Google Scholar] [CrossRef]
  30. Peng, S.; Ding, Y.; Wen, Z.; Chen, Y.; Cao, Y.; Ren, J. Spatiotemporal change and trend analysis of potential evapotranspiration over the Loess Plateau of China during 2011–2100. Agric. For. Meteorol. 2017, 233, 183–194. [Google Scholar] [CrossRef]
  31. Ding, Y.; Peng, S. Spatiotemporal Trends and Attribution of Drought across China from 1901–2100. Sustainability 2020, 12, 477. [Google Scholar] [CrossRef]
  32. Peng, S.; Ding, Y.; Liu, W.; Li, Z. 1 km monthly temperature and precipitation dataset for China from 1901 to 2017. Earth Syst. Sci. Data 2019, 11, 1931–1946. [Google Scholar] [CrossRef]
  33. Peng, S.; Gang, C.; Cao, Y.; Chen, Y. Assessment of climate change trends over the loess plateau in China from 1901 to 2100. Int. J. Climatol. 2017, 38, 2250–2264. [Google Scholar] [CrossRef]
  34. Peng, S. 1-km Monthly Precipitation Dataset for China (1901–2024). National Tibetan Plateau/Third Pole Environment Data Center. 2020. Available online: https://data.tpdc.ac.cn/en/data/faae7605-a0f2-4d18-b28f-5cee413766a2 (accessed on 28 October 2025).
  35. Peng, S. 1-km Monthly Mean Temperature Dataset for China (1901–2024). National Tibetan Plateau/Third Pole Environment Data Center. 2019. Available online: https://cstr.cn/18406.11.Meteoro.tpdc.270961 (accessed on 28 October 2025).
  36. Chen, J.; Gao, M.; Cheng, S.; Hou, W.; Song, M.; Liu, X.; Liu, Y. Global 1 km × 1 km gridded revised real gross domestic product and electricity consumption during 1992–2019 based on calibrated nighttime light data. Sci. Data 2022, 9, 202. [Google Scholar] [CrossRef] [PubMed]
  37. Chen, J.; Gao, M. Global 1 km × 1 km Gridded Revised Real Gross Domestic Product and Electricity Consumption During 1992–2019 Based on Calibrated Nighttime Light Data. 2021. Available online: https://figshare.com/articles/dataset/Global_1_km_1_km_gridded_revised_real_gross_domestic_product_and_electricity_consumption_during_1992-2019_based_on_calibrated_nighttime_light_data/17004523/1?file=31456837 (accessed on 28 October 2025).
  38. Gao, M.; Chen, J. PSO-BP. Codes for Modelling the Relationships Between the National GDP and Nighttime Light Data. 2022. Available online: https://figshare.com/articles/dataset/PSO-BP_codes_for_modelling_the_relationships_between_the_national_GDP_and_nighttime_light_data/19517272/1?file=34691536 (accessed on 28 October 2025).
  39. WorldPop. Global High Resolution Population Denominators Project. Available online: https://hub.worldpop.org/doi/10.5258/SOTON/WP00674 (accessed on 28 October 2025).
  40. Chen, Z.; Yu, B.; Yang, C.; Zhou, Y.; Yao, S.; Qian, X.; Wang, C.; Wu, B.; Wu, J. An Extended Time-Series (2000–2018) of Global NPP-VIIRS-like Nighttime Light Data. Earth Syst. Sci. Data 2021, 13, 889–906. [Google Scholar] [CrossRef]
  41. Yang, J.; Huang, X. The 30 m annual land cover dataset and its dynamics in China from 1990 to 2019. Earth Syst. Sci. Data 2021, 13, 3907–3925. [Google Scholar] [CrossRef]
  42. Wu, J.; Wang, X.; Wang, C.; He, X.; Ye, M. The status and development trend of disaggregation of socio-economic data. J. Geo-Inf. Sci. 2018, 20, 1252–1262. [Google Scholar] [CrossRef]
  43. Guo, H.; Zhu, W. A review on the spatial disaggregation of socioeconomic statistical data. Acta Geogr. Sin. 2022, 77, 2650–2667. [Google Scholar] [CrossRef]
  44. Wang, J.; Li, X.; Christakos, G.; Liao, Y.; Zhang, T.; Gu, X.; Zheng, X. Geographical detectors-based health risk assessment and its application in the neural tube defects study of the Heshun region, China. Int. J. Geogr. Inf. Syst. 2010, 24, 107–127. [Google Scholar] [CrossRef]
  45. Wang, J.; Xu, C. Geodetector: Principle and prospective. Acta Geogr. Sin. 2017, 72, 116–134. [Google Scholar] [CrossRef]
  46. Huang, B.; Wu, B.; Barry, M. Geographically and temporally weighted regression for modeling spatio-temporal variation in house prices. Int. J. Geog. Inf. Sci. 2010, 24, 383–401. [Google Scholar] [CrossRef]
  47. Fotheringham, A.; Charlton, M.; Brunsdon, C. The geography of parameter space: An investigation of spatial non-stationarity. Int. J. Geogr. Inf. Syst. 1996, 10, 605–627. [Google Scholar] [CrossRef]
  48. Lv, W.; Lei, Y.; Liu, F.; Yan, J.; Song, Y.; Zhao, W. gdverse: An R Package for Spatial Stratified Heterogeneity Family. Trans. GIS. 2025, 29, e70032. [Google Scholar] [CrossRef]
Sustainability 17 09721 g001
Figure 1. Study area.
Figure 1. Study area.
Sustainability 17 09721 g001
Sustainability 17 09721 g002
Figure 2. Study scale: (a) Municipal scale; (b) watershed scale; (c) grid scale.
Figure 2. Study scale: (a) Municipal scale; (b) watershed scale; (c) grid scale.
Sustainability 17 09721 g002
Sustainability 17 09721 g003
Figure 3. Technology Roadmap.
Figure 3. Technology Roadmap.
Sustainability 17 09721 g003
Sustainability 17 09721 g004
Figure 4. Box-and-Whisker Plot of Annual Average PM2.5 Concentration in the Yangtze River Economic Belt, 2005–2020.
Figure 4. Box-and-Whisker Plot of Annual Average PM2.5 Concentration in the Yangtze River Economic Belt, 2005–2020.
Sustainability 17 09721 g004
Sustainability 17 09721 g005
Figure 5. Spatial Distribution Map of Annual Average PM2.5 Concentrations in the Yangtze River Economic Belt, 2005–2020.
Figure 5. Spatial Distribution Map of Annual Average PM2.5 Concentrations in the Yangtze River Economic Belt, 2005–2020.
Sustainability 17 09721 g005
Sustainability 17 09721 g006
Figure 6. Distribution Map of Annual Average PM2.5 Concentration Change Slope in the Yangtze River Economic Belt from 2005 to 2020 (a) and Distribution Map of Change Significance (b).
Figure 6. Distribution Map of Annual Average PM2.5 Concentration Change Slope in the Yangtze River Economic Belt from 2005 to 2020 (a) and Distribution Map of Change Significance (b).
Sustainability 17 09721 g006
Sustainability 17 09721 g007
Figure 7. GeoDetector Factor Detector Results: (a) Municipal-scale 2005–2020 q mean histogram; (b) shows the line chart of q values at the Municipal-scale from 2005 to 2020; (c) shows the histogram of q mean values at the watershed-scale from 2005 to 2020; (d) shows the line chart of q values at the watershed-scale from 2005 to 2020; (e) shows the histogram of q mean values at the grid-scale from 2005 to 2020; (f) shows the line chart of q values at the grid-scale from 2005 to 2020.
Figure 7. GeoDetector Factor Detector Results: (a) Municipal-scale 2005–2020 q mean histogram; (b) shows the line chart of q values at the Municipal-scale from 2005 to 2020; (c) shows the histogram of q mean values at the watershed-scale from 2005 to 2020; (d) shows the line chart of q values at the watershed-scale from 2005 to 2020; (e) shows the histogram of q mean values at the grid-scale from 2005 to 2020; (f) shows the line chart of q values at the grid-scale from 2005 to 2020.
Sustainability 17 09721 g007
Table 1. Data Introduction.
Table 1. Data Introduction.
CategoryAbbreviationDescriptionUnitSource
PM2.5PM2.52.5-micrometer Particulate Matterμg/m3https://zenodo.org/records/6398971 (accessed on 28 October 2025)
natural factorNDVINormalized Difference Vegetation Index-https://search.earthdata.nasa.gov/search/granules?p=C2327962326-LPCLOUD&pg[0][v]=f&pg[0][gsk]=-start_date&q=MOD13A3 (accessed on 28 October 2025)
ARAnnual Rainfallmmhttps://data.tpdc.ac.cn/en/data/faae7605-a0f2-4d18-b28f-5cee413766a2 (accessed on 28 October 2025)
MATMean Annual Temperature°Chttps://cstr.cn/18406.11.Meteoro.tpdc.270961 (accessed on 28 October 2025)
PWAProportion of Water Area-https://zenodo.org/records/15853565 (accessed on 28 October 2025)
Human factorPFAProportion of Farmland Area-https://zenodo.org/records/15853565 (accessed on 28 October 2025)
PISAProportion of Impervious Surface Area-https://zenodo.org/records/15853565 (accessed on 28 October 2025)
PDPopulation Densitypeople/km2https://hub.worldpop.org/geodata/listing?id=76 (accessed on 28 October 2025)
NTLNighttime Lights-https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/YGIVCD (accessed on 28 October 2025)
GDPGross Domestic Productdollarshttps://figshare.com/articles/dataset/Global_1_km_1_km_gridded_revised_real_gross_domestic_product_and_electricity_consumption_during_1992-2019_based_on_calibrated_nighttime_light_data/17004523/1?file=31456837 (accessed on 28 October 2025)
SDESulfur Dioxide Emissionstonhttps://www.stats.gov.cn/ (accessed on 28 October 2025)
CVOCivil Vehicle Ownershiptonhttps://www.stats.gov.cn/ (accessed on 28 October 2025)
Table 2. VIF for each factor across three scales.
Table 2. VIF for each factor across three scales.
FactorVIF
MunicipalWatershedGrid
NDVI6.784.903.88
MAT2.093.583.30
PWA3.542.351.74
PD5.793.122.50
SDE1.921.381.24
CVO3.141.611.48
PFA2.832.682.43
PISA3.743.563.20
Table 3. GTWR Model Parameters and Results.
Table 3. GTWR Model Parameters and Results.
Spatial BandwidthTemporal
Bandwidth		R-SquaredAdjusted
R-Squared		AICc
Municipal100,00020.9680.960−201.894
Watershed0.9630.961−5434.839
Grid0.9490.949−42,395.559
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Zhang, Y.; Chen, Y.; Wei, Y. Multi-Scale Analysis of Influencing Factors for Temporal and Spatial Variations in PM2.5 in the Yangtze River Economic Belt. Sustainability 2025, 17, 9721. https://doi.org/10.3390/su17219721

AMA Style

Zhang Y, Chen Y, Wei Y. Multi-Scale Analysis of Influencing Factors for Temporal and Spatial Variations in PM2.5 in the Yangtze River Economic Belt. Sustainability. 2025; 17(21):9721. https://doi.org/10.3390/su17219721

Chicago/Turabian Style

Zhang, Yufei, Yu Chen, and Yongming Wei. 2025. "Multi-Scale Analysis of Influencing Factors for Temporal and Spatial Variations in PM2.5 in the Yangtze River Economic Belt" Sustainability 17, no. 21: 9721. https://doi.org/10.3390/su17219721

APA Style

Zhang, Y., Chen, Y., & Wei, Y. (2025). Multi-Scale Analysis of Influencing Factors for Temporal and Spatial Variations in PM2.5 in the Yangtze River Economic Belt. Sustainability, 17(21), 9721. https://doi.org/10.3390/su17219721

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop