Evaluating the influence of land use and land cover change on fine particulate matter

Fine particulate matter (i.e. particles with diameters smaller than 2.5 microns, PM2.5) has become a critical environmental issue in China. Land use and land cover (LULC) is recognized as one of the most important influence factors, however very fewer investigations have focused on the impact of LULC on PM2.5. The influences of different LULC types and different land use and land cover change (LULCC) types on PM2.5 are discussed. A geographically weighted regression model is used for the general analysis, and a spatial analysis method based on the geographic information system is used for a detailed analysis. The results show that LULCC has a stable influence on PM2.5 concentration. For different LULC types, construction lands have the highest PM2.5 concentration and woodlands have the lowest. The order of PM2.5 concentration for the different LULC types is: construction lands > unused lands > water > farmlands >grasslands > woodlands. For different LULCC types, when high-grade land types are converted to low-grade types, the PM2.5 concentration decreases; otherwise, the PM2.5 concentration increases. The result of this study can provide a decision basis for regional environmental protection and regional ecological security agencies.

Evaluating the influence of land use and land cover change on fine particulate matter Wei Yang 1* & Xiaoli Jiang 2 Fine particulate matter (i.e. particles with diameters smaller than 2.5 microns, PM 2.5 ) has become a critical environmental issue in China. Land use and land cover (LULC) is recognized as one of the most important influence factors, however very fewer investigations have focused on the impact of LULC on PM 2.5 . The influences of different LULC types and different land use and land cover change (LULCC) types on PM 2.5 are discussed. A geographically weighted regression model is used for the general analysis, and a spatial analysis method based on the geographic information system is used for a detailed analysis. The results show that LULCC has a stable influence on PM 2.5 concentration. For different LULC types, construction lands have the highest PM 2.5 concentration and woodlands have the lowest. The order of PM 2.5 concentration for the different LULC types is: construction lands > unused lands > water > farmlands >grasslands > woodlands. For different LULCC types, when high-grade land types are converted to low-grade types, the PM 2.5 concentration decreases; otherwise, the PM 2.5 concentration increases. The result of this study can provide a decision basis for regional environmental protection and regional ecological security agencies.
With the rapid development of China's economy and society, its rate of urbanization is accelerating. China's industrial scale is also expanding rapidly, and the problem of air pollution is becoming increasingly serious, which has a tremendous impact on the environment, economic development, and even people's health 1 . Fine particulate matter (i.e. particles with diameters smaller than 2.5 microns, PM 2.5 ) is considered a crucial protagonist among the various air pollution factors 2 . As a significant health hazard, PM 2.5 is highly associated with an increased probability of respiratory diseases 3,4 , cardiorespiratory problems 5,6 , mutagenic diseases 7 and increased mortality. Therefore, it is of vital significance to understand PM 2.5 pollution clearly, especially its distribution characteristics and influence factors, which are helpful for reducing pollution and protecting human health.
As a severe air pollutant, the concentration of PM 2.5 is influenced by meteorological factors 8-10 , human activities 11 , and the surrounding environment 12 . PM 2.5 is emitted mainly from anthropogenic sources, such as from traffic 13 and industrial production 14 . The spatial and temporal distributions of PM 2.5 are impacted by meteorological and environment factors [15][16][17] . Previous research has revealed that PM 2.5 is severely affected by meteorological factors at the macro-scale 18 in terms of temperature 19 , precipitation 20 , wind conditions 21,22 , etc., while at the micro-scale, PM 2.5 is strongly associated with land use and land cover (LULC) type 23 . Optimizing LULC type may reduce PM 2.5 pollution at the community or city level 24,25 . Land use and land cover change (LULCC) is the embodiment of human activities, which also has an obvious effect on PM 2.5 distribution 26 . To mitigate pollution, it is significant to explore the effects of LULC and LULCC on PM 2.5 pollution.
To conduct research on the relationship between LULCC and PM 2.5, relevant data are required. Remote sensing based LULCC research has a long history and is relatively mature 27,28 , which has become an effective method to obtain LULCC data. Conventional methods of obtaining PM 2.5 data employ monitoring stations at fixed sites, whose effective monitoring distances range from 0.5 to 4 km 29 , and which can provide accurate point-source data. The area among the monitoring sites can not been represented by this data. Due to the discontinuous spatial distribution of sites monitoring PM 2.5 data, several methods have been employed to solve this problem, including spatial interpolation 30 , chemical transport models 31 , land-use regression models 32 and aerosol optical depth (AOD) based statistical models 33 . However, as the use of a single approach leads to large uncertainties, some researchers have sought to integrate different methods to improve the PM 2.5 estimation accuracy, such as a combination of chemical transport models and satellite-derived AOD 34 36,37 . Due to atmospheric transport, PM 2.5 distribution is not only affected by local emissions, but also regional transport 38 . Regional land use changes can directly or indirectly affect PM 2.5 distribution. There is an insufficient amount on research at regional scale. Moreover, most of the existed researches focus on the influence of landscape patterns on PM 2.5 pollution but not LULCC types 39,40 . And the PM 2.5 data used in these studies was station monitoring data which is spatially discontinuous and cannot reveal the spatial relationship between PM 2.5 and LULCC types. Therefore, in this paper we analyze the relationship between dynamic PM 2.5 and LULCC type. To avoid the spatial discontinuity of station monitoring PM 2.5 data, the spatially continuous PM 2.5 data from the Atmospheric Composition Analysis Group (ACAG) are used. A geographical weighted regression model and a spatial analysis method are employed to identify the response mechanism between dynamic PM 2.5 and LULCC type. The results of this study can provide a decision basis for regional environmental protection and regional ecological security agencies.

Methodology
Study area. Shanxi Province is located in the middle of China (Fig. 1), which is the most important energy bases in the country and whose coal output was ranked first before 2016, and second thereafter. Due to the abundance of coal resources in Shanxi Province, its energy structure is focused on coal, which accounts for 72% of its total energy consumption. Shanxi Province is not only an important coal exporter, but also an important power exporter. The power plants in Shanxi Province are mainly coal-fired, which produce considerable amounts of emissions. Additionally, coking and steel industries are pillar industries in Shanxi Province, which also produce vast amounts of emissions. This economic structure based on energy consumption causes serious air pollution. Several cities in Shanxi Province, such as Taiyuan, Linfen, Jincheng, etc., contain the worst air pollution of all cities in China. Meanwhile, Shanxi Province had experienced obvious LULCCs, such as urban expansion caused by fast urbanization and an increase of green land owing to the growth of the 'Grain for Green' project. Therefore, Shanxi Province was selected as the study area to analyze the relationship between LULCC and PM 2.5 .
Data acquisition and preparation. PM 2.5 data. The PM 2.5 data provided by ACAG were generated based on a combination of a chemical transport model, satellite observations and ground-based observations 41 . The data have been validated in North America, which have been shown to have higher accuracy than purely geoscience-based estimates 35 . However, the accuracy of the ACAG data of China has not been validated; therefore, in this work we estimated its accuracy (see Sect. 3.1).
Pearson's correlation coefficient and the root mean square error (RMSE) were calculated in the validation as: where X i represents a PM 2.5 value from the ACAG, and Y i represents a PM 2.5 value from monitoring stations.
Land use and land cover data. The China multi-period land use land cover data set (CNLUCC) was used. The CNLUCC data were generated with a visual interpretation method based on Landsat remote-sensing data. The data set was provided by the Data Center for Resources and Environmental Sciences, Chinese Academy of Science (http:// www. resdc. cn). Data in 2000 and 2018 were used (Fig. 3). The data consist of six classes: farmlands, forests, grasslands, water, construction lands, and unused lands. The data were shown to have an accuracy of 88.95%, which meet the needs of this study. Geographical weighted regression model. Geographical weighted regression (GWR) models are a powerful tool to explore the heterogeneity of spatial relations 42 . As a local spatial regression model, GWR can effectively solve the nonstationarity of variable space, which has been widely used in the spatial analyses of different geographic elements 43 . The essence of GWR is locally weighted least squares, where the 'weight' is a distance function of spatial position between the point to be estimated and other observation points. The expression of GWR is as follows: where y is the dependent variable, x is the explanatory variable, (u i , v i ) is the coordinates of the ith point in space, a k (u i , v i ) is a realization of the continuous function a k (u, v) at point i, and ε i is the error term.
To identify the spatial relationship between LULCC and PM 2.5 , a 3 × 3 km grid map (Fig. 2) was generated of the study area. The variations of PM 2.5 between 2000 and 2018 were calculated in each grid, where the results were considered as the dependent variable in Eq. (3). The changing area of each different land type in each grid was also calculated and considered as the explanatory variable. Four main land types, farmlands, woodlands, grasslands www.nature.com/scientificreports/ and construction lands, were selected as the explanatory variables because their combined area accounted for nearly 99% of the total area.

Analysis framework.
A GWR analysis was used to determine the overall characteristic between the LULCC and PM 2.5 dynamics. After that, based on the spatial analysis tools in ArcGIS 10.2, a detail analysis was conducted from two aspects: (1) PM 2.5 distributions for the different LULC types, and (2) PM 2.5 dynamics for the different LULCC types. The analysis process is shown in Fig. 4.

Results
Validation of the PM 2.5 data. As mentioned above, station monitoring data can represent a scope from 0.5 to 4 km. Thus, a 4-km buffer from each monitoring station was generated. In the buffer, the mean values of the PM 2.5 data from ACAG were calculated and validated according to the station monitoring data. The results (Table 1) show an RMSE of 7.05 and a Pearson's correlation coefficient of 0.82, which show that the ACAG PM 2.5 data have high consistency with the ground-based observational data.

GWR analysis.
The GWR analysis showed that R 2 reached 0.94 which implies a good fitting effect. 93.56% of the standardized residuals were between − 2 and 2, which demonstrates that the model fitting was stable 44 .
The results show that there was a stable relationship between PM 2.5 and LULCC. As shown in Fig. 5   Effect of the LULCC type on PM 2.5 . LULCC matrix. As showed in Table 3, in 2000, the main land type in Shanxi Province was farmland, whose area was 6.12 × 10 4 km 2 , accounting for 39.09% of the total area. Next in total area were grasslands and woodlands, accounting for 29.16% and 28.01% respectively. Construction lands covered 0.42 × 10 4 km 2 , accounting for 2.67% of the total area. The areas of water and unused lands were very little, accounting for just 0.97% and 0.10% respectively. In 2018, although farmlands still covered the largest area, its area reduced to 5.78 × 10 4 km 2 , accounting for 36.91% of the total area. The area of woodlands increased, accounting for 28.36% of the total area, and became the second largest land type. The area of grasslands decreased by 0.16 × 10 4 km 2 and became the third largest land type. The area of construction lands increased dramatically, and its proportion increased to 5.56%, which was two times greater than in 2000. Water and unused lands still covered very little area, accounting for 0.94% and 0.08% of the total, respectively. From the perspective of land being converting from one type to another, there was a large conversion of farmlands to other land types, roughly 2.29 × 10 4 km 2 . Grasslands, woodlands, and construction lands underwent the largest amounts of change, accounting for 53.47%, 23.07%, and 20.91% of the total converted area, respectively. On the other hand, the other land types that were converted to farmlands accounted for 1.95 × 10 4 km 2 , which significant decreased the total area of farmlands. The main conversion types of woodlands to other land types were grasslands, farmlands and construction lands. The sources of woodlands were mainly grasslands and farmlands. It was seen that the amount of woodlands converted to other types, and those converted to woodlands, were nearly equivalent in total area. Grasslands showed a similar trend as seen for the woodlands, which also showed a relatively stable state. Construction lands were mainly converted to farmlands, which accounted for 82.51% of the total converted area. The total area of construction lands that were converted to other types was 0.22 × 10 4 km 2 . The main conversion sources of construction lands were farmlands, grasslands, and woodlands, and the total conversion area was 0.67 × 10 4 km 2 , which was caused by fast urbanization.   www.nature.com/scientificreports/   www.nature.com/scientificreports/ PM 2.5 dynamics. As water and unused lands covered only 1% of the total area, we only considered the other four land types (farmlands, woodlands, grasslands, and construction lands) to ascertain the influence of the LULCC types on the PM 2.5 dynamics. As discussed above, the PM 2.5 concentrations considerably increased from 2000 to 2018 for all land types, which indicated that there would be a PM 2.5 concentration increase for non-LULCC areas. The increase was mainly caused by increased pollution levels, not by LULCC. This would bring disturbance to our analysis. To avoid this disturbance, the range of PM 2.5 concentration variations in non-LULCC areas was calculated first (Table 4) and set as the reference variation range when analyzing the PM 2.5 concentration variations in the LULCC areas.
As showed in Table 4, the PM 2.5 dynamics in the different LULCC types showed two opposing trends, increasing and decreasing. The largest increase was for woodlands converted to construction lands, while the largest decline was for farmlands converted to woodlands. When farmlands were converted to woodlands and grasslands, the PM 2.5 concentrations declined, but when they were converted to construction lands, the PM 2.5 concentration increased. Increasing trends were seen when woodlands were converted to the other three land types. Conversely, declining trends were found when construction lands were converted to the other land types. When grasslands were converted to woodlands, a declining trend was witnessed, but when they were converted to the other two land types, increasing trends were seen.
As discussed above, the PM 2.5 concentrations for the four land types showed similar trends in both years: construction lands > farmlands > grasslands > woodlands. Therefore, according to the PM 2.5 concentrations, the four land types were divide into four grades: highest (construction lands), high (farmlands), medium (grasslands) and low (woodlands). As showed in Table 5, when high-grade land types are converted to low-grade types, the PM 2.5 concentrations decrease, and when low-grade land types are converted to high-grade types, the PM 2.5 concentrations increase.

Discussion
As an important energy base, the economic development of Shanxi Province has been mainly based on energy consumption, which continues to generate large quantities of harmful emissions 45 . Therefore, human activities were considered as the most important influence factor of PM 2.5 pollution 11 . However, LULC types were also representative of different human activities 46 . Different to previous studies, which mainly focused on discussing the relationship between land use type and PM 2.5 concentrations at urban scales 37,47 , in this study we discussed the impact of land use on PM 2.5 concentrations from two aspects: different LULC and LULCC types at regional scales.
The different LULC types indicated the different intensities of human activities. Construction lands represented the highest intensity because of the high population density, traffic flow, industrial and commercial activities, etc., therein. All of these generate large quantities of air pollutants and caused the highest PM 2.5 www.nature.com/scientificreports/ concentrations 48 . Farmlands were also intensively affected by human activities, which caused relatively high PM 2.5 concentrations. Firstly, straw burning in farmlands can result in a sharp increase of PM 2.5 concentration within a short time 49 . Secondly, as a great agricultural country, the use of fertilizer in China is pervasive, and emissions arising from the manufacturing and use of fertilizer have a strong relationship with PM 2.5 50 . For example, fertilizer liberated from the soil can be converted into a precursor of PM 2.5 51 . Thirdly, heating activities in rural areas in winter mainly consist of burning coal, which generates large quantities of air pollutants and has an important impact on the PM 2.5 concentration in farmlands 52,53 . Vegetation covered area, including woodlands and grasslands, had relatively low PM 2.5 concentrations. These areas were less influenced by human activities, as indicated by the lower pollutant emissions therein. Meanwhile, it has been suggested that vegetation coverage has a negative regulating effect on PM 2.5 concentration 54,55 . Thus, woodlands have the lowest PM 2.5 concentrations because of their highest vegetation coverage.
The different LULCC types represented transitions among the different intensities of human activities, which caused dynamic changes of the PM 2.5 concentrations. When other land types were converted to construction lands, the intensity of human activities increased, which caused an increase of PM 2.5 concentration. A similar conclusion was found in another study, which showed that when natural land cover is replaced by manmade areas PM 2.5 concentrations increase 56 . Furthermore, other LULCC types were also discussed in our study. Farmlands may also contain intense human activities that can increase the PM 2.5 concentration, such as agricultural activities 57,58 . This was demonstrated by the increasing trend of PM 2.5 concentration when woodlands and grasslands were converted to farmlands. As vegetation coverage had a negative effect on PM 2.5 concentration 55 , the PM 2.5 concentration also changed when the vegetation type changed; i.e. an increase trend was seen when woodlands were converted to grasslands.
Due to the limited LULC data, this study illustrated the influence of LULC and LULCC on PM 2.5 at the regional scale where human activities were considered as the most important influence factor. However, PM 2.5 pollution is both affected by human and natural factors 59 . In desert areas, natural factors including dust and wind could be the most important factors 60 , while in coastal areas, climatic elements had the most important influence on PM 2.5 pollution 37 . These situations were not discussed in the present study. Future studies at larger scales are required to demonstrate the influence of LULC on PM 2.5 more comprehensively. The relationship between PM 2.5 pollution and its influence factors is complex and non-linear 61 . Traditional linear analysis methods have certain limitations and new non-linear analysis methods should be employed. Moreover, higher spatial and temporal resolution PM 2.5 data and LULC data are also required to better understand the response mechanism of PM 2.5 pollution to LULCC.

Conclusions
In this study, high-accuracy PM 2.5 data from ACAG and LULC data were employed to explore the relationship between PM 2.5 and LULCC. A GWR method was used for the general analysis, and a spatial analysis method based on the geographic information system was used for the detailed analysis. The main conclusions can be drawn as follows: (1) The GWR analysis showed that R 2 reached 0.92, which represented a stable relationship between PM 2.5 and LULCC. High local R 2 values located in highly dynamic LULCC areas indicated that the dynamic PM 2.5 had a significant response to LULCC. (2) In both considered years, 2000 and 2018, the order of PM 2.5 concentration in the different LULC types was the same: construction lands > unused lands > water > farmlands >grasslands > woodlands, meaning that the LULC type had an important influence on the PM 2.5 concentration. (3) LULCC can also impact the dynamics of PM 2.5 concentration. When low-grade land types are converted to high-grade types, the PM 2.5 concentration increases; otherwise, it decreases. From another angle, when natural lands are converted to human-related lands, the PM 2.5 concentration increase; otherwise, the PM 2.5 concentrations decrease.