Understanding the Influence of Crop Residue Burning on PM2.5 and PM10 Concentrations in China from 2013 to 2017 Using MODIS Data

In recent years, particulate matter (PM) pollution has increasingly affected public life and health. Therefore, crop residue burning, as a significant source of PM pollution in China, should be effectively controlled. This study attempts to understand variations and characteristics of PM10 and PM2.5 concentrations and discuss correlations between the variation of PM concentrations and crop residue burning using ground observation and Moderate Resolution Imaging Spectroradiometer (MODIS) data. The results revealed that the overall PM concentration in China from 2013 to 2017 was in a downward tendency with regional variations. Correlation analysis demonstrated that the PM10 concentration was more closely related to crop residue burning than the PM2.5 concentration. From a spatial perspective, the strongest correlation between PM concentration and crop residue burning existed in Northeast China (NEC). From a temporal perspective, the strongest correlation usually appeared in autumn for most regions. The total amount of crop residue burning spots in autumn was relatively large, and NEC was the region with the most intense crop residue burning in China. We compared the correlation between PM concentrations and crop residue burning at inter-annual and seasonal scales, and during burning-concentrated periods. We found that correlations between PM concentrations and crop residue burning increased significantly with the narrowing temporal scales and was the strongest during burning-concentrated periods, indicating that intense crop residue burning leads to instant deterioration of PM concentrations. The methodology and findings from this study provide meaningful reference for better understanding the influence of crop residue burning on PM pollution across China.


Ground-Observed PM2.5 and PM10 Concentrations Data
The PM2.5 and PM10 concentrations data used for this study were obtained from website PM25.in (http://pm25.in/about), which collects official real-time air quality data provided by China National Environmental Monitoring Center (CNEMC). The real-time air quality data include hourly PM2.5 concentration data (μg/m 3 ), hourly PM10 concentration data (μg/m 3 ), Air Quality Index (AQI), and other airborne pollutants concentration data. Before 1 January 2015, the published PM data supplied

Ground-Observed PM 2.5 and PM 10 Concentrations Data
The PM 2.5 and PM 10 concentrations data used for this study were obtained from website PM25.in (http://pm25.in/about), which collects official real-time air quality data provided by China National Environmental Monitoring Center (CNEMC). The real-time air quality data include hourly PM 2.5 concentration data (µg/m 3 ), hourly PM 10 concentration data (µg/m 3 ), Air Quality Index (AQI), and other airborne pollutants concentration data. Before 1 January 2015, the published PM data supplied by PM25.in (http://pm25.in/about), covered 190 monitoring cities in China, and this number has increased to 367 since 1 January 2015 [30]. The location of ground-monitoring air quality stations can be seen in Figure 2. by PM25.in (http://pm25.in/about), covered 190 monitoring cities in China, and this number has increased to 367 since 1 January 2015 [30]. The location of ground-monitoring air quality stations can be seen in Figure 2. By calling the specific API document on website PM25.in (http://pm25.in/about), we collected hourly PM2.5 and PM10 concentrations data for all monitoring cities in China from 18 January 2013 to 31 December 2017. The daily PM concentration data for each region were calculated by averaging all available hourly PM data from all monitoring cities.

MODIS Active Fire Data
The Moderate Resolution Imaging Spectroradiometer (MODIS) is an optical remote sensing instrument widely used in the fields of Geoscience, Environmental Science, and so on. Owing to its multi-spectral bands (36) and broad spectrum, ranging from 0.4 μm (visible band) to 14.4 μm (thermal infrared band), MODIS can provide a great deal of geographic and atmospheric information. Meanwhile, terra (AM) and aqua (PM) with MODIS transits China four times per day on 10:30, 22:30, 01:30, and 13:30, respectively [31]. Concerning the capability of fire detection, MODIS can monitor conflagration areas over 1000 m 2 . If the weather is suitable (e.g., little/no smoke and relative homogeneous land surface) for observing, one tenth of burning fire spots would be detected. Light fires covering around 50 m 2 can be detected under the most favorable weather conditions [32].
We utilized MOD14A1/MYD14A1 daily Level 3 fire products (MODIS Thermal Anomalies/Fire products) with a spatial resolution of 1 km, which are available at NASA's LAADS DACC ftp server [33], to extract crop residue burning spots in China. In addition, a contextual algorithm was applied to detect fire spots according to the strong radiation from mid-infrared bands [34]. The products also classified the reliability of fire detection into three levels, including low-confidence fires, nominalconfidence fires, and high-confidence fires. MOD14A1/MYD14A1 were stored as a single file that consisted of eight days' data for convenience, representing eight-day continuous collection of fire data. To get daily fire spots map (Figure 3a), a maximum value composite method was employed for processing the data integration of MOD14A1/MYD14A1 products.

Land-Use and Land-Cover Data
Although fire spots could be extracted from MODIS fire products, it cannot be directly defined as the crop residue burning spots. Owing to the existence of such burning types as forest fire and urban solid waste incineration, the extraction of crop residue burning spots was further processed with a dataset of Land-Use and Land-Cover Change (LUCC) provided by Resources and Environmental Sciences Data Center, Chinese Academy of Sciences (RESDC) [35]. The dataset reflects changes of land-use and land-cover in China every five years with a high spatio-resolution of 1 km, which is similar to that of MODIS fire products'. This data set has six classes, including cropland, forest, grassland, waters, urban and rural & industrial and residential areas, and unused land. The By calling the specific API document on website PM25.in (http://pm25.in/about), we collected hourly PM 2.5 and PM 10 concentrations data for all monitoring cities in China from 18 January 2013 to 31 December 2017. The daily PM concentration data for each region were calculated by averaging all available hourly PM data from all monitoring cities.

MODIS Active Fire Data
The Moderate Resolution Imaging Spectroradiometer (MODIS) is an optical remote sensing instrument widely used in the fields of Geoscience, Environmental Science, and so on. Owing to its multi-spectral bands (36) and broad spectrum, ranging from 0.4 µm (visible band) to 14.4 µm (thermal infrared band), MODIS can provide a great deal of geographic and atmospheric information. Meanwhile, terra (AM) and aqua (PM) with MODIS transits China four times per day on 10:30, 22:30, 01:30, and 13:30, respectively [31]. Concerning the capability of fire detection, MODIS can monitor conflagration areas over 1000 m 2 . If the weather is suitable (e.g., little/no smoke and relative homogeneous land surface) for observing, one tenth of burning fire spots would be detected. Light fires covering around 50 m 2 can be detected under the most favorable weather conditions [32].
We utilized MOD14A1/MYD14A1 daily Level 3 fire products (MODIS Thermal Anomalies/Fire products) with a spatial resolution of 1 km, which are available at NASA's LAADS DACC ftp server [33], to extract crop residue burning spots in China. In addition, a contextual algorithm was applied to detect fire spots according to the strong radiation from mid-infrared bands [34]. The products also classified the reliability of fire detection into three levels, including low-confidence fires, nominal-confidence fires, and high-confidence fires. MOD14A1/MYD14A1 were stored as a single file that consisted of eight days' data for convenience, representing eight-day continuous collection of fire data. To get daily fire spots map (Figure 3a), a maximum value composite method was employed for processing the data integration of MOD14A1/MYD14A1 products. classification precision of this dataset for each region varies from 73% to 89%, and the overall accuracy of whole nation is up to 81% [36]. In this study, for more reliable extraction of crop residue burning spots, we used the LUCC data in year 2010 and year 2015 (Figure 3b) to generate cropland-masks on study area. Here, the extracted fire spots in year 2013 and 2014 corresponded to cropland-mask in 2010, and fire spots in other years corresponded to cropland-mask in 2015 (Figure 3c).

Methods
Firstly, due to a tremendous amount of pixels comprised, we conducted mosaic processes to compose complete remote sensing images of China. Meanwhile, we extracted "fire-mask" from Science Dataset for obtaining fire spots maps of the study area. Given the long research period and the large quantity of data, we employed batch processing using a specific tool named MODIS Reprojection Tool (MRT) provided by the Land Processes Distributed Active Archive Center. Secondly, in order to summarize overall fire spots in one day, a maximum value composite strategy was proposed and developed to count the number of daily fire spots [18]. The principle of this strategy is to set corresponding attribute values (7 means low-confidence fire spots, 8 means nominalconfidence fire spots, and 9 means high-confidence fire spots) to each pixel based on the maximum value in the daily four observations. In the process of composite, if fire spots detected in the same pixel were recorded several times for a day, we only counted them as one spot to avoid repeat counting. Clouds and haze had significant influences on the detection of fire spots. Since the same area was rarely covered by clouds in the four observations per day, this strategy reduced the occlusion effects and guaranteed the accuracy of fire spots detection. Thirdly, we employed LUCC dataset for extracting crop residue burning spots from the preprocessed data. Cropland-masks were selected from the dataset and combined with corresponding fire spots maps, then daily fire pixels located in croplands (daily crop residue burning spots) were extracted. On the other hand, hourly PM2.5 and PM10 concentration data were collated into a daily format and the city-level observation

Land-Use and Land-Cover Data
Although fire spots could be extracted from MODIS fire products, it cannot be directly defined as the crop residue burning spots. Owing to the existence of such burning types as forest fire and urban solid waste incineration, the extraction of crop residue burning spots was further processed with a dataset of Land-Use and Land-Cover Change (LUCC) provided by Resources and Environmental Sciences Data Center, Chinese Academy of Sciences (RESDC) [35]. The dataset reflects changes of land-use and land-cover in China every five years with a high spatio-resolution of 1 km, which is similar to that of MODIS fire products'. This data set has six classes, including cropland, forest, grassland, waters, urban and rural & industrial and residential areas, and unused land. The classification precision of this dataset for each region varies from 73% to 89%, and the overall accuracy of whole nation is up to 81% [36]. In this study, for more reliable extraction of crop residue burning spots, we used the LUCC data in year 2010 and year 2015 ( Figure 3b) to generate cropland-masks on study area. Here, the extracted fire spots in year 2013 and 2014 corresponded to cropland-mask in 2010, and fire spots in other years corresponded to cropland-mask in 2015 (Figure 3c).

Methods
Firstly, due to a tremendous amount of pixels comprised, we conducted mosaic processes to compose complete remote sensing images of China. Meanwhile, we extracted "fire-mask" from Science Dataset for obtaining fire spots maps of the study area. Given the long research period and the large quantity of data, we employed batch processing using a specific tool named MODIS Reprojection Tool (MRT) provided by the Land Processes Distributed Active Archive Center. Secondly, in order to summarize overall fire spots in one day, a maximum value composite strategy was proposed and developed to count the number of daily fire spots [18]. The principle of this strategy is to set corresponding attribute values (7 means low-confidence fire spots, 8 means nominal-confidence fire spots, and 9 means high-confidence fire spots) to each pixel based on the maximum value in the daily four observations. In the process of composite, if fire spots detected in the same pixel were recorded several times for a day, we only counted them as one spot to avoid repeat counting. Clouds and haze had significant influences on the detection of fire spots. Since the same area was rarely covered by clouds in the four observations per day, this strategy reduced the occlusion effects and guaranteed the accuracy of fire spots detection. Thirdly, we employed LUCC dataset for extracting crop residue burning spots from the preprocessed data. Cropland-masks were selected from the dataset and combined with corresponding fire spots maps, then daily fire pixels located in croplands (daily crop residue burning spots) were extracted. On the other hand, hourly PM 2.5 and PM 10 concentration data were collated into a daily format and the city-level observation data were also recalculated into a regional scale. Finally, we employed statistical and Spearman's rank correlation analysis to examine the correlation between crop residue burning and PM pollution for each region at different temporal scales.

Results
To better understand the following study, the spatial distribution of crop residue burning and PM concentrations in the different regions of China was shown in Figure 4. data were also recalculated into a regional scale. Finally, we employed statistical and Spearman's rank correlation analysis to examine the correlation between crop residue burning and PM pollution for each region at different temporal scales.

Results
To better understand the following study, the spatial distribution of crop residue burning and PM concentrations in the different regions of China was shown in Figure 4.     10 concentrations was witnessed in CC and NWC, respectively. The decrease of PM concentrations in NEC was relatively higher than that of other regions. Furthermore, we analyzed the PM 2.5 /PM 10 ratio, which could reveal different characteristics and origins of particle pollution [36]. A higher ratio usually indicated that PM pollution was caused by anthropogenic activities, while a lower ratio demonstrated that natural factors were the main contribution source of PM pollution [37]. According to Figure 6, the PM 2.5 /PM 10 ratio in each region all dropped to a much lower level with small fluctuations that occasionally arose during 5-year period. Meanwhile, the most obvious decline of PM 2.5 /PM 10  concentrations was witnessed in CC and NWC, respectively. The decrease of PM concentrations in NEC was relatively higher than that of other regions. Furthermore, we analyzed the PM2.5/PM10 ratio, which could reveal different characteristics and origins of particle pollution [36]. A higher ratio usually indicated that PM pollution was caused by anthropogenic activities, while a lower ratio demonstrated that natural factors were the main contribution source of PM pollution [37]. According to Figure 6, the PM2.5/PM10 ratio in each region all dropped to a much lower level with small fluctuations that occasionally arose during 5-year period. Meanwhile, the most obvious decline of PM2.5/PM10 ratio was shown in CC (from 0.85 in 2013 to 0.63 in 2017) and the lowest ratio appeared in NWC (average value is about 0.47) for each year.

Seasonal Variations and Characteristics
For better understanding seasonal variations and characteristics of PM2.5 and PM10 concentrations, we divided twelve months into four seasons as follows: Spring (March, April, May), summer (June, July, August), autumn (September, October, November), and winter (December, January, February). As can be seen from Figure 7, the seasonal variation of PM10 concentrations in the same region is similar to that of PM2.5 concentrations, whereas seasonal characteristics and variations of these two PM concentrations vary significantly across regions. Besides, concentrations of PM10 and  Regarding characteristics of PM10 concentrations in different regions, the highest value always appeared in NWC and the lowest concentration of PM10 was usually observed in SC. In addition, throughout a whole year, the average PM10 concentration of NC always maintained a much higher level than that of other regions'. For CC and NEC, the PM10 pollution usually deteriorated in autumn and winter. Moreover, from a temporal perspective, the maxima of PM10 concentrations in each region appeared in winter, and the minima appeared in summer. In spring, PM10 concentrations evidently decreased in NWC and slightly decreased in other regions. For the decline of PM10 concentration in summer, the maximum change appeared in NWC, with NEC in the second place. In autumn, the declines from 2013 to 2015 were evident in all regions and increases appeared in northern and western China in 2016, when PM10 concentrations in CC, NEC, and NWC greatly reduced (40 μg/m 3 approximately) compared to the previous high concentration. For winter, the major decrease of PM10 concentrations was witnessed in NEC, NC, and CC.

Seasonal Variations and Characteristics
For better understanding seasonal variations and characteristics of PM 2.5 and PM 10 concentrations, we divided twelve months into four seasons as follows: Spring (March, April, May), summer (June, July, August), autumn (September, October, November), and winter (December, January, February). As can be seen from Figure 7, the seasonal variation of PM 10 concentrations in the same region is similar to that of PM 2.5 concentrations, whereas seasonal characteristics and variations of these two PM concentrations vary significantly across regions. Besides, concentrations of PM 10  very small in each region and the largest decrease of 16 μg/m 3 appeared in NC. Different from slight variations in spring and summer, PM2.5 concentrations in autumn and winter decreased significantly in each region. Particularly, maximum changes were observed in CC (reduced about 50 μg/m 3 ) and NEC (reduced about 35 μg/m 3 ). Besides, for NC, the decreased-concentration in winter was much higher than that in autumn. Other seasonal-interannual variations of PM concentrations could be found in Figure 7.  Regarding characteristics of PM 10 concentrations in different regions, the highest value always appeared in NWC and the lowest concentration of PM 10 was usually observed in SC. In addition, throughout a whole year, the average PM 10 concentration of NC always maintained a much higher level than that of other regions'. For CC and NEC, the PM 10 pollution usually deteriorated in autumn and winter. Moreover, from a temporal perspective, the maxima of PM 10 concentrations in each region appeared in winter, and the minima appeared in summer. In spring, PM 10 concentrations evidently decreased in NWC and slightly decreased in other regions. For the decline of PM 10 concentration in summer, the maximum change appeared in NWC, with NEC in the second place. In autumn, the declines from 2013 to 2015 were evident in all regions and increases appeared in northern and western China in 2016, when PM 10 concentrations in CC, NEC, and NWC greatly reduced (40 µg/m 3 approximately) compared to the previous high concentration. For winter, the major decrease of PM 10 concentrations was witnessed in NEC, NC, and CC.
Similar to PM 10 concentrations, PM 2.5 concentrations in different regions were the lowest in summer and highest in winter. Spatially, the peak of PM 2.5 concentrations usually appeared in CC and NC, which was different from that of PM 10 concentrations. Meanwhile, the lowest PM 2.5 concentration showed in SC, which was similar to that of PM 10 concentrations. For other regions, the PM 2.5 concentration of NEC always kept at a much higher level in spring, autumn, and winter. Although the PM 2.5 concentration of NWC was not the highest in these seven regions, it remained at a relatively high level throughout the year. The higher PM 2.5 concentration was also observed in EC in spring, summer, and winter. PM 2.5 concentration in SWC was lower than other regions except for SC. For spring, the notable decline of PM 2.5 concentrations was witnessed in NWC and CC, whilst the decrease in other regions was much smaller. For summer, the decline of PM 2.5 concentrations was very small in each region and the largest decrease of 16 µg/m 3 appeared in NC. Different from slight variations in spring and summer, PM 2.5 concentrations in autumn and winter decreased significantly in each region. Particularly, maximum changes were observed in CC (reduced about 50 µg/m 3 ) and NEC (reduced about 35 µg/m 3 ). Besides, for NC, the decreased-concentration in winter was much higher than that in autumn. Other seasonal-interannual variations of PM concentrations could be found in Figure 7.

Seasonal Variations
According to Figure 9, we can see clear seasonal variations of crop residue burning spots for each region. Crop residue burning in CC usually took place in summer and autumn. During 2013 to 2017, the proportion of crop residue burning in spring increased gradually, and decreased notably in summer and autumn, whilst it demonstrated slight variations in winter. The variation of crop residue burning in EC were generally consistent with that in CC. For NC, crop residues were often burnt in summer and autumn. However, the proportion of crop residue burning spots in these two seasons decreased year by year, while the ratio in spring gradually increased to one third of the total amount. The number of crop residue burning spots were limitedly distributed in winter. As an agriculturally developed region, NEC experienced very intense crop residue burning, which mainly concentrated in spring and autumn. Meanwhile, the proportion of crop residue burning in autumn decreased from 67% in 2013 to 34% in 2017, and the proportion in spring increased from 27% in 2013 to 64% in 2017. For NWC, crop residue burning mainly took place in spring and autumn. A sudden increase appeared in the spring of 2014, whilst the proportion in autumn plummeted to 20%. Following this, crop residue burning in spring and autumn decreased dramatically, and gradually concentrated in summer. During this period, the proportion of crop residue burning in autumn decreased whilst the proportion in spring stabilized between 30% and 40%. Finally, crop residue burning spots in NWC presented similar proportion in spring, summer and winter in 2017. Unlike the northern part of China, crop residue burning in SC was usually observed in winter. Whereas, in recent years, proportions of crop residue burning in other seasons increased without clear pattern. Furthermore, crop residue burning of SWC usually concentrated in spring and summer. During this period, the proportion of crop residue burning increased in summer and decreased in spring.

The Correlation between PM Concentrations and Crop Residue Burning at an Annual Scale
We employed Spearman's rank correlation for establishing the correlation between daily PM data and daily crop residue burning spots data. The result (Table 1) showed that the correlation between PM concentration and crop residue burning in NEC and SC were much stronger than that in other regions. According to Figure 10, variations were different in these two regions. In NEC, correlations between PM10 concentration and crop residue burning were generally upward with fluctuations, except for a notable decrease in 2015. The overall trend of the correlation between crop residue burning and PM2.5 concentrations was similar, yet the significance of this correlation was much weaker. In SC, correlation coefficients between PM concentrations and crop residue burning generally decreased, except for a slight increase in 2015. In addition, a significant phenomenon was

The Correlation between PM Concentrations and Crop Residue Burning at an Annual Scale
We employed Spearman's rank correlation for establishing the correlation between daily PM data and daily crop residue burning spots data. The result (Table 1) showed that the correlation between PM concentration and crop residue burning in NEC and SC were much stronger than that in other regions. According to Figure 10, variations were different in these two regions. In NEC, correlations between PM 10 concentration and crop residue burning were generally upward with fluctuations, except for a notable decrease in 2015. The overall trend of the correlation between crop residue burning and PM 2.5 concentrations was similar, yet the significance of this correlation was much weaker. In SC, correlation coefficients between PM concentrations and crop residue burning generally decreased, except for a slight increase in 2015. In addition, a significant phenomenon was that the correlation between PM 10 concentrations and crop residue burning was stronger than that between PM 2.5 concentrations and crop residue burning.  that the correlation between PM10 concentrations and crop residue burning was stronger than that between PM2.5 concentrations and crop residue burning.

The Correlation between PM Concentrations and Crop Residue Burning at a Seasonal Scale
We analyzed correlations between PM concentrations and crop residue burning for each region from a seasonal perspective. The results (in Figure 11 and Table 2) showed that correlations in autumn were significantly stronger for the north part of China, including CC, EC, and NEC. For SC, correlations were stronger throughout four seasons and the largest correlation coefficient appeared in winter. Correlations in SWC were relatively poor and only significant in spring and summer. The correlation coefficient in NEC was the strongest among seven regions and the strongest correlation usually appeared in spring and autumn, when crop residues were intensely burnt in NEC. For EC, the correlation between PM concentrations and crop residue burning was significant in four seasons and were much stronger in autumn and winter. Similar to annual analysis, PM10 concentrations were more strongly correlated with crop residue burning than PM2.5 concentrations.

The Correlation between PM Concentrations and Crop Residue Burning at a Seasonal Scale
We analyzed correlations between PM concentrations and crop residue burning for each region from a seasonal perspective. The results (in Figure 11 and Table 2) showed that correlations in autumn were significantly stronger for the north part of China, including CC, EC, and NEC. For SC, correlations were stronger throughout four seasons and the largest correlation coefficient appeared in winter. Correlations in SWC were relatively poor and only significant in spring and summer. The correlation coefficient in NEC was the strongest among seven regions and the strongest correlation usually appeared in spring and autumn, when crop residues were intensely burnt in NEC. For EC, the correlation between PM concentrations and crop residue burning was significant in four seasons and were much stronger in autumn and winter. Similar to annual analysis, PM 10 concentrations were more strongly correlated with crop residue burning than PM 2.5 concentrations.  With different time of crop ripening in each region, periods of crop residue burning are different accordingly. Therefore, in order to better analyze the change of PM concentrations when crop residues were intensely combusted, for each year, we selected a burning-concentrated period for each

The Correlation between PM Concentrations and Crop Residue Burning in Burning-Concentrated Periods
With different time of crop ripening in each region, periods of crop residue burning are different accordingly. Therefore, in order to better analyze the change of PM concentrations when crop residues were intensely combusted, for each year, we selected a burning-concentrated period for each region during 2013-2017. The principle of selection was based on the appearance of peak months of crop residue burning spots and prior knowledge of agricultural production. In total, we acquired five periods for each region and analyzed the correlation between the number of crop residue burning spots during the burning-concentrated period and corresponding PM 2.5 concentrations. The results are shown in Table 3. Except for NC, correlations between PM concentrations and crop residue burning were significant in all regions. Generally, correlations in NC and SWC were the weakest, and correlations in NEC were the strongest. Meanwhile, the correlation between PM 10 concentrations and crop residue burning was significantly stronger than that of PM 2.5 concentrations. This result indicated that the variation of PM 10 concentrations was more sensitive to crop residue burning than that of PM 2.5 concentrations during the process of crop residue burning. Correlation between PM concentrations and crop residue burning increased significantly with the narrowing temporal scales and was the strongest during burning-concentrated periods, indicating that intense crop residue burning exerts a much stronger influence on the short-term than long-term variation of PM concentrations.

The Attribution of Variations of PM 10 and PM 2.5 Concentrations during 5-Year Period
In this study, we analyzed variations and characteristics of PM concentrations from interannual and seasonal perspectives. Meanwhile, we selected some crop residue burning-concentrated periods to explore variations of PM concentrations during the burning processes. Generally, concentrations of PM 10 and PM 2.5 have decreased notably since 2013. Besides, PM 2.5 /PM 10 ratios also declined during the 5-year period which indicates that the composition of PM 10 occupied by PM 2.5 is decreasing. Meanwhile, some studies have shown that the high PM 2.5 /PM 10 ratio can be attributed to human activities, while the lower ratio is related to natural factors [37,38]. In other words, PM 2.5 pollution has been mitigated significantly, due to a series of emission-reduction measures. Firstly, in autumn and winter, the variation of PM concentrations in northern China can be attributed to the control of crop residue burning, traffic exhaust, and coal combustion for large-scale central heating [39]. Secondly, with the implementation of Red and Orange alert measures for reducing PM pollution, PM 2.5 concentrations have decreased remarkably [40]. Thirdly, as a result of traffic control, the exhaust-emission of vehicles has been cut down dramatically and leads to the reduction of PM concentrations [41]. Fourthly, some environmental-meteorological projects have been implemented to address PM pollution issues [42]. In burning-concentrated periods, the variation trend of PM concentrations is consistent with that of crop residue burning in all regions, indicating intensive crop residue burning leads to instant deterioration of PM concentrations. Hence, more strict and effective policies should be proposed and implemented to encourage more efficient utility of crop residues and reduce large scale and intensive crop residue burning.

The Attribution of Correlations between PM Concentration and Crop Residue Burning
The correlation between PM concentrations and crop residue burning was discussed in this paper. Firstly, it is found that the correlation between PM concentrations and crop residue burning is significant and strong, especially in burning-concentrated periods, which is consistent with findings from previous studies [43]. Awasthi [23] mainly introduced the temporal variation of both crop residue burning and PM 2.5 concentrations in China and did not discuss the correlation from different temporal scales. From a spatial perspective, the correlation in NEC is the strongest among the seven regions, especially in spring and autumn, suggesting that the PM concentration is closely related to crop residue burning in the burning-concentrated periods. This phenomenon was consistent with findings from previous studies suggesting that crop residue burning is related to PM 2.5 concentration [23,24]. The main reason for the poor correlation in NC is that the source of PM is high exhaust-emission of vehicles and industrial production, instead of crop residue burning [41]. For NWC, petroleum exploitation is also an important contributor to PM pollution [44], which may be the reason why PM 10 demonstrates a weaker correlation with crop residue burning than PM 2.5 . To sum up, the burning of crop residues has a great contribution to PM pollution, though the relative contribution of crop residue burning to PM concentrations, compared with other emission sources, including industry and traffic exhaust, should be further investigated.

Limitations and Prospect
Although the paper comprehensively examined correlations between PM concentration and crop residue burning, some limitations remain. Firstly, due to the fact that crop residue burning usually lasts for a short period, the correlation analysis should be more reliable if it is conducted based on a finer temporal resolution, such as hourly. Thus, considering the finer temporal resolution of Himawari-8, it is a better choice to extract fire spots on the hourly scale. Secondly, due to the limited spatial resolution of MODIS data, some actual burning spots may be lost in the process of fire spots extraction and statistics. That means remote sensing data with higher temporal resolution are required for extracting fine-scale crop reside burning spots. Furthermore, due to complicated interactions between PM and meteorological factors, commonly used correlation analysis may be biased significantly. To reduce the influence from other factors and better investigate the influence of crop residue burning on PM concentrations, advanced causality methods, such as cross convergent mapping (CCM) [45] and chemical transport models (CTM), such as WRF-CAMx [46], should be employed in future studies. Whereas, the difficulty for examining the causality of crop residue burning on PM concentration without other influencing factors, using above models lies in the short time series of the concentrated crop residue burning periods. Meanwhile, the MODIS data extracted crop residue burning spots are mainly based on a daily scale and thus the time series of intensive crop residue burning is limited to less than 30 numbers, not sufficient for a robust CCM or CTM analysis. Therefore, to implement CCM or CTM analysis, fire spots should be extracted using remote sensing data with a much higher temporal resolution, such as Himawari 8 with 10-min temporal resolution. In the future, with growing availability and accuracy of Himawari data sources, it is possible to conduct robust causality analysis based on CCM or CTM using long time series data of crop residue burning and PM pollution. In this case, the influence of crop residue burning on PM concentrations can be better extracted by filtering the biases of other influencing factors.

Conclusions
This paper analyzed interannual and seasonal variations of PM 10 and PM 2.5 concentrations and simultaneous variations of crop residue burning in several regions across China. The results showed that the PM concentration was in a downward trend from interannual and seasonal perspectives and PM 2.5 /PM 10 ratios in different regions decreased gradually. The peak value of PM 10 concentrations usually appeared in NWC and winter whilst the peak value of PM 2.5 concentrations appeared in NC and CC. Temporal variations of PM 2.5 are similar to that of PM 10 concentrations. For the number of crop residue burning spots in China, it remained a downward tendency during the 5-year period in most regions, except for an evident increase in NEC in 2017. Furthermore, we analyzed correlations between PM concentration and crop residue burning and explored at different temporal scales. The variation of PM 10 concentration was more sensitive to crop residue burning than that of PM 2.5 concentrations and the strongest correlation between PM concentrations and crop residue burning appears in NEC. Correlation between PM concentrations and crop residue burning increased significantly with the narrowing temporal scales and was the strongest during burning-concentrated periods, indicating that intense crop residue burning exert a much stronger influence on the short-term than long-term variation of PM concentrations. The methodology and conclusions from this study provide useful reference for better understanding the influence of crop residue burning on PM concentrations at different scales and suggest that intensive crop residue burning leads to instant increases of PM concentrations. Given the major contribution of crop residue burning to PM pollution, more strict and effective policies should be proposed and implemented to encourage more efficient utility of crop residues and reduce large scale and intensive crop residue burning.