Satellite-Observed Variations and Trends in Carbon Monoxide over Asia and Their Sensitivities to Biomass Burning

As the carbon monoxide (CO) total column over Asia is among the highest in the world, it is important to characterize its variations in space and time. Using Measurements of Pollution in the Troposphere (MOPITT) and Atmospheric InfraRed Sounder (AIRS) satellite data, the variations and trends in CO total column over Asia and its seven subregions during 2003–2017 are investigated in this study. The CO total column in Asia is higher in spring and winter than in summer and autumn. The seasonal maximum and minimum are in spring and summer respectively in the regional mean over Asia, varying between land and oceans, as well as among the subregions. The CO total column in Asia shows strong interannual variation, with a regional mean coefficient of variation of 5.8% in MOPITT data. From 2003 to 2017, the annual mean of CO total column over Asia decreased significantly at a rate of (0.58 ± 0.15)% per year (or −(0.11 ± 0.03) × 1017 molecules cm−2 per year) in MOPITT data, resulting from significant CO decreases in winter, summer, and spring. In most of the subregions, significant decreasing trends in CO total column are also observed, more obviously over areas with high CO total column, including eastern regions of China and the Sichuan Basin. The regional decreasing trends in these areas are over 1% per year. Over the entire Asia, and in fire-prone subregions including South Siberia, Indo-China Peninsula, and Indonesia, we found significant correlations between the MOPITT CO total column and the fire counts from the Moderate Resolution Imaging Spectroradiometer (MODIS). The variations in MODIS fire counts may explain 58%, 60%, 36%, and 71% of the interannual variation in CO total column in Asia and these three subregions, respectively. Over different land cover types, the variations in biomass burning may explain 62%, 52%, and 31% of the interannual variation in CO total column, respectively, over the forest, grassland, and shrubland in Asia. Extremes in CO total column in Asia can be largely explained by the extreme fire events, such as the fires over Siberia in 2003 and 2012 and over Indonesia in 2006 and 2015. The significant decreasing trends in MODIS fire counts inside and outside Asia suggest that global biomass burning may be a driver for the decreasing trend in CO total column in Asia, especially in spring. In general, the variations and trends in CO total column over Asia detected by AIRS are similar to but smaller than those by MOPITT. The two datasets show similar spatial and temporal variations in CO total column over Asia, with correlation coefficients of 0.86–0.98 in the annual means. This study shows that the interannual variation in atmospheric CO in Asia is sensitive to biomass burning, while the decreasing trend in atmospheric CO over Asia coincides with the decreasing trend in MODIS fire counts from 2003 to 2017.


Study Area, Study Period, and Statistics Analysis
Asia (10 • S-60 • N, 60 • E-140 • E) is the domains of interest in this study. The study period is from 2003 to 2017. Since Asia is large and nonhomogeneous, we assessed variations and trends in CO total column over Asia and its subregions, including South Siberia ( (Figure 1a). The seasons are defined as: spring (March-May), summer (June-August), autumn (September-November), and winter (December-February). The trends are quantified using the simple linear regression. The slope of the linear fit represents the increasing or decreasing rate of CO. In this study, the statistics are reported as significant when p < 0.05.

Satellite Carbon Monoxide Data and Analysis
MOPITT (https://terra.nasa.gov/about/terra-instruments/mopitt) is an instrument on board the National Aeronautics and Space Administration (NASA) Earth Observing System Terra spacecraft launched in 1999, using near-infrared at 2.3 µm and thermal-infrared radiation at 4.7 µm. MOPITT detects CO at a spatial resolution of 22 km × 22 km at nadir. MOPITT monthly products (MOP03JM, level 3, version 7) with a horizontal resolution of 1 • × 1 • (latitude × longitude) were used [42]. To enhance data quality, the daytime data with the degree of freedom for signal higher than 0.75 were selected.
CO data from AIRS instrument (https://airs.jpl.nasa.gov/mission_and_instrument/overview) were analyzed as a supplement. AIRS is on board Aqua, launched in 2002 by NASA. AIRS resolving power is λ/∆λ = 1200, resulting in a spectral resolution~1.8 cm −1 for the 4.6 µm CO absorption. AIRS provides twice daily and near-global coverage of tropospheric CO [43]. AIRS version 6 products (AIRS3STM and AIRS3SPM) of monthly CO data at 1 • × 1 • were used. Similar to the data filtering for MOPITT measurements, AIRS daytime CO data with the degree of freedom for signal higher than 0.5 were selected.
Both MOPITT [6,7] and AIRS [8] are most sensitive to CO in the middle troposphere. Deeter et al. [7] suggested that compared with the in situ observations from National Oceanic and Atmospheric Administration (NOAA) validation sites, CO total column from MOPITT version 7 product has a correlation coefficient of 0.93 and a bias of 0.3 × 10 17 molecules cm −2 . McMillan et al. [44] suggested that compared with in the situ CO profiles, AIRS version 5 retrievals at 300-900 hPa are biased by 6-10% in the northern mid-latitudes. MOPITT and AIRS CO data have been widely used in investigating the distribution, transport, sources, and sinks of global tropospheric CO [20,45].

Fire Count and Emission Data and Land Cover Data
To fit the horizontal resolution of satellite CO data, we gridded the Moderate Resolution Imaging Spectroradiometer (MODIS) monthly active fire count data (MCD14ML) [46] into 1 • × 1 • (latitude × longitude) globally from 2003 to 2017. The data with the confidence level higher than 0.75 were selected. The Global Fire Emissions Database (GFED) is a bottom-up inventory for fire emissions based on satellite burned areas [47]. The version GFED 4.1s data [27] were used, including small fires with the spatial resolution of 0.  [46] and (c) the annual total carbon monoxide (CO) emissions from biomass burning (BB) from the Global Fire Emissions Database (GFED) data [27]. The   The annual total fire counts from the Moderate Resolution Imaging Spectroradiometer (MODIS) data [46] and (c) the annual total carbon monoxide (CO) emissions from biomass burning (BB) from the Global Fire Emissions Database (GFED) data [27]. The grid size is 1 The land cover data from the European Space Agency Climate Change Initiative (ESA CCI) with a resolution of 300 m were used. The land cover data are updated every year and available from 2003 to 2015. As the land cover data in 2016 and 2017 were not available, the land cover in 2015 was used as a substitute. We regridded the original ESA CCI land cover data to grids of 1 • × 1 • (Figure 1a). The most dominant land cover type in each of the grids of 1 • × 1 • is assigned to that grid. The original 37 classes of land cover types were reclassified to the seven land cover types: forest, cropland, shrubland, grassland, urban areas, water bodies and others as shown in Figure 1a. Reclassified land cover types is listed in detail in Supplement as Table S1. MODIS fire count data and ESA CCI land cover data were widely used and have been well validated in previous studies [46]. Figure 1b shows the spatial distribution of the annual total fire counts over Asia, averaged over 2003-2017. Fires occur more in the Indo-China Peninsula, Indonesia, and South Siberia than in other subregions in Asia. The three fire-prone subregions contribute a majority of CO fire emissions in Asia (Figure 1c) [48][49][50]. The three regions are covered mostly by forest and cropland (Figure 1a). Among seasons, fires occur over the Indo-China Peninsula mostly in spring and winter, over Indonesia mostly in summer and autumn, and over South Siberia mostly in spring and summer ( Figure 2). Dry and hot climate in spring, summer and autumn is the main reason for strong fires in Indo-China Peninsula, Indonesia, and South Siberia [51,52]. In winter, since the cold Siberian high and the East Asian winter monsoon can bring frequent and severe cold surges and/or snowstorms, leading to the surface air temperature below the point of fire ignition [53,54]. Therefore, fires are inactive in winter over South Siberia and latitudes above 23 • N. In the annual mean, CO emissions from biomass burning in Asia are~76.5 Tg during 2003-2017 (Figure 3b), which account for 23% of the global fire emissions (332.5 Tg). Forest contributes approximately 59% of the CO emissions from biomass burning (Figure 3d) annually. Seasonally, the estimated fire CO emissions in Asia rank as 25.6 Tg in spring, >23.7 Tg in autumn, >17.9 Tg in summer, >8.6 Tg in winter.     shows spatial variations in CO total column in Asia in the annual mean and by season from MOPITT and ARIS data averaged over 2003-2017. In the annual mean, CO total columns from MOPITT and AIRS show similar spatial variations over Asia. CO total column is high in North China, South China, the Sichuan Basin, and Indo-China Peninsula, and low in the Tibetan Plateau. CO total column from MOPITT is larger than that from AIRS over the land of Asia (Figures 4b and 5b), but lower than that from AIRS over the ocean (Figures 4c and 5c). The two datasets have a significant correlation over Asia, with the correlation coefficient ranging from 0.76 to 0.98 ( Figure 4). The correlation coefficient in the mean over Asia between the two datasets reaches as high as 0.86-0.98 in all seasons ( Figure 5). In the annual mean, the regional mean CO total columns over Asia retrieved from MOPITT and AIRS are (19.3    Atmospheric InfraRed Sounder (AIRS) (the 2nd column). Correlation coefficients (r) between MOPITT and AIRS CO total column (the 3rd column). The shadow areas indicate that the r is statistically significant (p < 0.05). The corresponding annual mean is shown in (a-c) (the 1st row). The corresponding seasonal mean is shown in (d-f) for spring (the 2nd row), in (g-i) for summer (the 3rd row), in (j-l) for autumn (the 4th row) and in (m-o) for winter (the 5th row). Seasonally ( Figure 6), the regional mean CO total column in Asia is higher in spring and winter than in summer and autumn. The seasonal maximum and minimum are in spring and summer over the entire Asia, respectively, varying between land and oceans, as well as among subregions. Over land and in most subregions, the maximum and minimum are in spring and autumn, respectively. However, CO total column over Indonesia peaks in autumn. The seasonality in CO total column is larger over the land than over the oceans in both MOPITT and AIRS data.

Interannual Variation in Atmospheric Carbon Monoxide over Asia
The interannual variation in CO total column from MOPITT and AIRS over Asia during 2003-2017 are shown in Figure 7. The maximum and minimum in CO total column appeared in 2003 and 2017, respectively. We use the coefficient of variation (CV) as a measure of the strength of the interannual variation in CO total column. CV is the ratio of the standard deviation to the long-term mean. CO total column in Asia shows strong interannual variation, with a regional mean CV of 5.8% in MOPITT data. The mean and CV of CO total column over Asia are both higher in MOPITT than in AIRS data. Spatially, the MOPITT CV in Asia ranges between 1.3% and 37.8% (Figure 8a). CV is generally high in the regions with strong fire activities (Figures 1b, 1c, and 8a) and is highest in Indonesia with the value of 12.7% (Figures 8a and 8b). CV is also higher over land than over oceans. The Asian mean CV is lower than the global mean CV.

Trends in Atmospheric Carbon Monoxide over Asia from 2003-2017
The trends in the annual mean CO total column derived from MOPITT and AIRS over Asia during 2003-2017 are examined spatially ( Figure 9) and on regional mean ( Figure 10, Table 1, and  Table S2). Both satellite data show that the annual mean CO total column has decreased significantly over large areas of East Asia, including South Siberia, North China, South China, the Sichuan Basin (Figures 9a,b). The decreasing trend in the annual mean CO total column is stronger in Asia than in the world ( Table 1). The decreasing trend in the annual mean CO total column is strongest in North China, South China, and the Sichuan Basin where the CO trends range from -(0.30 ± 0.04) × 10 17 molecules cm −2 yr −2 to -(0.36 ± 0.04) × 10 17 molecules cm −2 yr −1 . However, no significant trend in the annual mean CO total column is observed over the ocean of Asia, India, Indo-China Peninsula, and Indonesia in MOPITT and AIRS data. The annual mean CO total column in Asia declined at an average rate of 0.58% and 0.30% per year from 2003 to 2017, respectively, in MOPITT and AIRS data ( Figure 10a, Table 1, Table S2). The decreasing trend in the annual mean CO total column over Asia Seasonally ( Figure 6), the regional mean CO total column in Asia is higher in spring and winter than in summer and autumn. The seasonal maximum and minimum are in spring and summer over the entire Asia, respectively, varying between land and oceans, as well as among subregions. Over land and in most subregions, the maximum and minimum are in spring and autumn, respectively. However, CO total column over Indonesia peaks in autumn. The seasonality in CO total column is larger over the land than over the oceans in both MOPITT and AIRS data.

Interannual Variation in Atmospheric Carbon Monoxide over Asia
The interannual variation in CO total column from MOPITT and AIRS over Asia during 2003-2017 are shown in Figure 7. The maximum and minimum in CO total column appeared in 2003 and 2017, respectively. We use the coefficient of variation (CV) as a measure of the strength of the interannual variation in CO total column. CV is the ratio of the standard deviation to the long-term mean. CO total column in Asia shows strong interannual variation, with a regional mean CV of 5.8% in MOPITT data. The mean and CV of CO total column over Asia are both higher in MOPITT than in AIRS data. Spatially, the MOPITT CV in Asia ranges between 1.3% and 37.8% ( Figure 8a). CV is generally high in the regions with strong fire activities (Figure 1b,c and Figure 8a) and is highest in Indonesia with the value of 12.7% (Figure 8a,b). CV is also higher over land than over oceans. The Asian mean CV is lower than the global mean CV.      The means CV of CO total column in Asia, its subregions, and the world. The bar indicates the standard deviation of the mean CV. CV is the ratio of the standard deviation to the long-term mean.

Trends in Atmospheric Carbon Monoxide over Asia from 2003-2017
The trends in the annual mean CO total column derived from MOPITT and AIRS over Asia during 2003-2017 are examined spatially ( Figure 9) and on regional mean ( Figure 10, Table 1, and Table S2). Both satellite data show that the annual mean CO total column has decreased significantly over large areas of East Asia, including South Siberia, North China, South China, the Sichuan Basin (Figure 9a,b). The decreasing trend in the annual mean CO total column is stronger in Asia than in the world ( Table 1). The decreasing trend in the annual mean CO total column is strongest in North China, South China, and the Sichuan Basin where the CO trends range from -(0.30 ± 0.04) × 10 17 molecules cm −2 yr −1 to -(0.36 ± 0.04) × 10 17 molecules cm −2 yr −1 . However, no significant trend in the annual mean CO total column is observed over the ocean of Asia, India, Indo-China Peninsula, and Indonesia in MOPITT and AIRS data. The annual mean CO total column in Asia declined at an average rate of 0.58% and 0.30% per year from 2003 to 2017, respectively, in MOPITT and AIRS data ( Figure 10a, Table 1, Table S2). The decreasing trend in the annual mean CO total column over Asia is stronger over land in MOPITT data than in AIRS data, but weaker over oceans in MOPITT data than in AIRS data (Figure 10a-c).   Indonesia. The red star indicates that the trend is statistically significant at 95% level (p < 0.05). The bar indicates the 95% confident interval.
14 ± 0.08 1 Numbers in bold indicate that the trends are statistically significant at the 95% level (p < 0.05). Figure 9 shows that the area with significant decreasing trend in CO total column from MOPITT is larger in spring and winter than in summer and autumn, and the decreasing trend in AIRS data is larger in spring and summer than in autumn and winter. The relative trends in MOPITT data averaged over Asia are found to be significant in all seasons except autumn ( Figure 10a and Table 1). In North China, South China, and the Sichuan Basin, CO total column from both datasets decreases significantly in all seasons. In these subregions, CO total column observed by MOPITT declined at a rate of over 1% per year, while AIRS data declined at a rate of about 0.5% per year in all seasons (Figure 10f-h). Note that in India, CO total column from MOPITT decreases significantly only in winter (Figure 10e). In South Siberia, the decreasing trend is strongest in spring, as high as 1.16% and 0.69% per year, respectively, from MOPITT and AIRS observations (Figure 10d). In the Indo-China Peninsula, the decreasing trend is significant in winter (Figure 10i). In Indonesia, no significant trends are observed in all seasons (Figure 10j). Compared with the trends in CO total column over North China, South China, and the Sichuan Basin where anthropogenic emissions are high, the trends in CO total column over the fire-prone regions, i.e., South Siberia, the Indo-China Peninsula and Indonesia, appear weak or insignificant during the fire seasons (Figures 2, 9 and 10).

Correlations between Biomass Burning and the Interannual Variations and Trends in Atmospheric Carbon
Monoxide over Asia

Sensitivity of the Interannual Variation in Carbon Monoxide over Asia to Biomass Burning
Biomass burning in Asia has strong interannual variation (Figure 11), which can significantly influence the year-to-year variation in CO total column in Asia. Tables 2 and 3 and Figure 12 show the correlation coefficients between CO total column and MODIS fire counts and between CO total column and GFED4 fire CO emissions over Asia. The annual mean of CO total column correlates significantly with the annual total fire counts, with r being 0.74 for MOPITT and 0.75 for AIRS, respectively (Figure 12a). Seasonally, the correlations averaged over Asia are significant in spring, summer, and autumn, and insignificant in winter (Figure 12c,e,g,i). The correlations between CO total column and fire counts over Asia appear stronger than those with fire CO emission data in all seasons (Figure 12c-j).
column and GFED4 fire CO emissions over Asia. The annual mean of CO total column correlates significantly with the annual total fire counts, with r being 0.74 for MOPITT and 0.75 for AIRS, respectively (Figure 12a). Seasonally, the correlations averaged over Asia are significant in spring, summer, and autumn, and insignificant in winter (Figure 12c,e,g,i). The correlations between CO total column and fire counts over Asia appear stronger than those with fire CO emission data in all seasons (Figure 12c-j). The correlation between the CO total column and biomass burning in different subregions are variant ( Table 2). According to MOPITT data, the correlation is significant (p < 0.05) in three subregions: Indo-China (r = 0.60), South Siberia (r = 0.77), and Indonesia (r = 0.84). Seasonally, the correlation over Indonesia is significant in all seasons (r ranging between 0.78 and 0.95) and most significant in autumn, while over South Siberia, it is insignificant in autumn and winter (Table 2) when fires rarely occur (Figure 2). The correlations over the land of Indo-China Peninsula appear to be significant only in autumn and winter. Interestingly, only in autumn, the correlation is significant over the grassland (r = 0.70), shrubland (r = 0.82), and cropland (r = 0.74). Over forests, the correlation is significant in all seasons except in winter. Generally, the results from AIRS are similar to these from MOPITT (Table S3).
According to the CO fire emission data from GFED4, the correlation between CO total column and CO fire emissions is generally similar to but weaker than that between CO and MODIS fire counts in different subregions, land cover types, and seasons (Tables 2-3, Tables S3-S4).
We explore the relationship between the extreme events of biomass burning and CO total column over Asia during 2003-2017 on regional mean. The annual total fire counts in Asia averaged  Figure 13, which shows the monthly anomalies of CO total column, fire counts, and fire CO emissions by latitude during 2003-2017. The extreme CO anomalies correspond to the strong fire events over some areas in Asia. The strong fire events occurs when CO anomalies are above 6 × 10 17 molecules cm −2 in the satellite CO data in Figure 13, in which four extreme fire events are marked. The signature of the extreme fire events is well captured by MOPITT CO, MODIS fire counts, and GFED4 CO emissions from fires (see AIRS CO in Figure S1).     The correlation between the CO total column and biomass burning in different subregions are variant (Table 2). According to MOPITT data, the correlation is significant (p < 0.05) in three subregions: Indo-China (r = 0.60), South Siberia (r = 0.77), and Indonesia (r = 0.84). Seasonally, the correlation over Indonesia is significant in all seasons (r ranging between 0.78 and 0.95) and most significant in autumn, while over South Siberia, it is insignificant in autumn and winter ( Table 2) when fires rarely occur (Figure 2). The correlations over the land of Indo-China Peninsula appear to be significant only in autumn and winter. Interestingly, only in autumn, the correlation is significant over the grassland (r = 0.70), shrubland (r = 0.82), and cropland (r = 0.74). Over forests, the correlation is significant in all seasons except in winter. Generally, the results from AIRS are similar to these from MOPITT (Table S3).
According to the CO fire emission data from GFED4, the correlation between CO total column and CO fire emissions is generally similar to but weaker than that between CO and MODIS fire counts in different subregions, land cover types, and seasons (Tables 2 and 3, Tables S3 and S4).
We explore the relationship between the extreme events of biomass burning and CO total column over Asia during 2003-2017 on regional mean. The annual total fire counts in Asia averaged over 2003-2017 is approximately 0.32 million (Figure 11 Figure 13, which shows the monthly anomalies of CO total column, fire counts, and fire CO emissions by latitude during 2003-2017. The extreme CO anomalies correspond to the strong fire events over some areas in Asia. The strong fire events occurs when CO anomalies are above 6 × 10 17 molecules cm −2 in the satellite CO data in Figure 13, in which four extreme fire events are marked. The signature of the extreme fire events is well captured by MOPITT CO, MODIS fire counts, and GFED4 CO emissions from fires (see AIRS CO in Figure S1). When extreme fire events occur, positive anomalies of CO total column are observed over wide areas in Asia. Forest fires in eastern Russia in 2003, forest fires in Indonesia in 2006, wildfires in Siberia in 2012, and forest fires in Indonesia in 2015 can enhance the CO total column by 3 × 10 17 , 3.4 × 10 17 , 1.7 × 10 17 , and 3.5 × 10 17 molecules cm −2 over the fire-prone areas according to MOPITT observations (Figure 13a). The four extreme fire events largely contribute to the peaks of regional mean CO total column over Asia in 2003, 2006, 2012, and 2015 (Figure 7). Overall, biomass burning can largely explain the interannual variation in CO total column over Asia on regional mean.

Comparison of the Trend in Atmospheric Carbon Monoxide with the Trend in Biomass Burning in Asia
The decreasing trend in CO over Asia is probably associated with multiple factors, including anthropogenic and biomass emissions, atmospheric chemistry and dynamics. Few studies have focused on the impact of biomass burning on the CO trend over Asia [24]. Uncertainties in emission inventories also enhance the challenges of quantifying this impact [20,55]. In this section, we compare the trend in CO total column over Asia and the trend in biomass burning, to explore a linkage between the two. Since CO emitted from biomass burning can be transported across continents [56][57][58], the trends in biomass burning both inside and outside Asia are examined.
As shown in Section 3.1.3, CO total column over Asia from MOPITT decreases significantly in all seasons except autumn ( Figure 10). Figure 14 compares the trend in CO total column in Asia with the trends in four datasets, i.e., MODIS fire counts inside and outside Asia, and GFED4 CO fire emissions inside and outside Asia. In spring, the fire counts inside and outside Asia decrease significantly; CO fire emissions inside and outside Asia decrease at a significant level of 86% and 89%, respectively. The trends in the four datasets indicate that global biomass burning may be a potential driver of the decreasing trend in CO total column over Asia in spring. However, in other seasons, among the four datasets, only fire counts outside Asia decrease significantly in autumn and winter. In summer, global biomass burning may not be a driver for the decreasing trend in CO total column over Asia. In winter, the biomass burning outside Asia has a low probability to be a cause of the decreasing trend in CO total column over Asia.

Comparison of the Trend in Atmospheric Carbon Monoxide with the Trend in Biomass Burning in Asia
The decreasing trend in CO over Asia is probably associated with multiple factors, including anthropogenic and biomass emissions, atmospheric chemistry and dynamics. Few studies have focused on the impact of biomass burning on the CO trend over Asia [24]. Uncertainties in emission inventories also enhance the challenges of quantifying this impact [20,55]. In this section, we compare the trend in CO total column over Asia and the trend in biomass burning, to explore a linkage between the two. Since CO emitted from biomass burning can be transported across continents [56][57][58], the trends in biomass burning both inside and outside Asia are examined.
As shown in Section 3.1.3, CO total column over Asia from MOPITT decreases significantly in all seasons except autumn ( Figure 10). Figure 14 compares the trend in CO total column in Asia with the trends in four datasets, i.e., MODIS fire counts inside and outside Asia, and GFED4 CO fire emissions inside and outside Asia. In spring, the fire counts inside and outside Asia decrease significantly; CO fire emissions inside and outside Asia decrease at a significant level of 86% and 89%, respectively. The trends in the four datasets indicate that global biomass burning may be a potential driver of the decreasing trend in CO total column over Asia in spring. However, in other seasons, among the four datasets, only fire counts outside Asia decrease significantly in autumn and winter. In summer, global biomass burning may not be a driver for the decreasing trend in CO total column over Asia. In winter, the biomass burning outside Asia has a low probability to be a cause of the decreasing trend in CO total column over Asia.

Temporal-Spatial Variations and Trends in Atmospheric Carbon Monoxide over Asia
Previous studies showed that Asia is one of the most polluted regions, and CO hotspots were observed over East Asia and East India [59,60]. Over 2003-2017, the values of CO total column from MOPITT and AIRS are high in most areas of East Asia and Indo-China Peninsula (Figure 4). Intensive traffic and industrial activities, urbanization, and large population density in these areas lead to high levels of anthropogenic CO emissions [17,24,29]. On regional average, atmospheric CO column in Asia observed by MOPITT in spring and winter is higher than in summer and autumn, in agreement with the seasonality of CO columns from AIRS ( Figure 6).
The values of CO total column from MOPITT and AIRS show strong interannual variation in Asia (Figures 7 and 8), and CO peaked in the years of 2003, 2006, 2009, 2012 and 2015. Some of the peaks are probably due to large fire events under dry conditions (i.e., the years of EI Niño) [39,61], especially in spring and summer over South Siberia [3,37], and in autumn and winter in Indonesia [32,40]. According to MOPITT data, the CV of the annual mean CO total column over Asia is 5.8%, which is weaker than that over the world (8.3%). The CV is higher in the regions with more biomass

Temporal-Spatial Variations and Trends in Atmospheric Carbon Monoxide over Asia
Previous studies showed that Asia is one of the most polluted regions, and CO hotspots were observed over East Asia and East India [59,60]. Over 2003-2017, the values of CO total column from MOPITT and AIRS are high in most areas of East Asia and Indo-China Peninsula (Figure 4). Intensive traffic and industrial activities, urbanization, and large population density in these areas lead to high levels of anthropogenic CO emissions [17,24,29]. On regional average, atmospheric CO column in Asia observed by MOPITT in spring and winter is higher than in summer and autumn, in agreement with the seasonality of CO columns from AIRS ( Figure 6).
The values of CO total column from MOPITT and AIRS show strong interannual variation in Asia (Figures 7 and 8), and CO peaked in the years of 2003, 2006, 2009, 2012 and 2015. Some of the peaks are probably due to large fire events under dry conditions (i.e., the years of EI Niño) [39,61], especially in spring and summer over South Siberia [3,37], and in autumn and winter in Indonesia [32,40]. According to MOPITT data, the CV of the annual mean CO total column over Asia is 5.8%, which is weaker than that over the world (8.3%). The CV is higher in the regions with more biomass burning [27] (Figure 2a and 8a). The CV is lowest in India, likely due to small interannual variations in emissions of CO from anthropogenic activities and biomass burning.
Worden et al. [19] found that all nadir-viewing thermal infrared (TIR) satellites measurements of CO total column are with a significantly decreasing trend~1% yr −1 at the 1σ level over the Northern Hemisphere from 2000 to 2011. Our results show that the annual mean CO total column in Asia declined at an average rate of 0.58% yr −1 from 2003 to 2017 in MOPITT data (p < 0.05). CO total column in Asia has decreased continuously since 2011 after Worden's analysis. Figure 9 shows that the decreasing trends in CO total column from MOPITT is strongest in East China, with a rate of −0.2~−0.4 × 10 17 molecules cm −2 per year. Previous studies showed that the decreasing trends are mainly caused by rapid technological changes with improved combustion efficiency and emission control measures in China since 2010 [32,55,62]. Note that in India, the annual and seasonal trends in MOPITT CO columns are smallest among all the subregions and all the decreasing trends are insignificant except in winter; there even is an insignificant increasing trend in spring. Contrary to China, the anthropogenic emissions in India have increased during the study period [24,55]. The increase of anthropogenic emissions over India may partially offset the overall decreasing trend in Asia and lead to smaller decreasing trends in India than in the rest of Asia. In addition, Yuan et al. [63] reported that CO has increased in the Asian Tropopause Aerosol Layer, a planetary-scale aerosol layer situated 13-18 km above sea level partially covering the northern part of India. This amount of CO and its variation are detected by satellite instruments. This may also be a reason why the trend in CO total column over India is among the smallest in Asia.
In general, the trends in CO total column over the fire-prone regions are found to be weak during their corresponding fire seasons. This leads to the next section where we discuss the impacts of biomass burning on the interannual variation and trend in atmospheric CO over Asia in further detail.

Impacts of Biomass Burning on the Interannual Variation and Trend in Atmospheric Carbon Monoxide over Asia
The interannual variation in CO total column appears to be sensitive to biomass burning in Asia and its subregions. In Asia, MODIS fire counts and GFED4 CO emissions show strong interannual variations, and high values often coincide with EI Niño years [64]. As seen in Figures 7 and 11, the years with peak CO total column often coincide with the years of peak fire counts or CO emissions from fires or both. On the annual average, the correlation coefficients between CO total column and fire counts is 0.74 from MOPITT data and 0.75 from AIRS data (p < 0.05). However, although the GFED4 data also suggest a positive correlation between CO total column and CO emissions from biomass burning in Asia, the correlation is not significant (r = 0.31-0.36, p > 0.05) (Figure 12a,b). Seasonally (Figure 12c-j), the fire count data suggest strong correlations between fires and CO total column in all seasons except in winter using both MOPITT and AIRS CO data, while the CO emission data only show strong correlations in autumn in both CO datasets and in summer in MOPITT CO data. The fire counts in spring comprise 52.5% of the total annual fire counts and springtime CO emissions from fires are the highest among all seasons (34.7%). In autumn, the fire counts comprise 15% of the total annual fire counts and the CO emissions form fires are the second highest (30.9%).
Overall, biomass burning occurs mostly in spring and autumn over Asia in 2003-2017, CO total column is sensitive to variation in biomass burning. High CO levels were observed during the intensive biomass burning events ( Figure 13) [37,40,41]. For example, wildfires in Siberia in 2003 and forest fires in Indonesia in 2015 can enhance the abundances of CO by 3 × 10 17 and 3.5 × 10 17 molecules cm −2 over the fire-prone areas according to MOPITT measurements. Tables 2 and 3 show how the interannual variation in CO total column coincide with that in fire occurrences in different land cover types and over different subregions of Asia. Because forest fires account for~50% of the total fire counts or CO emissions from fires (Figure 3c-d), plus the fact that the emission factors of forest fires are one of the highest among all land cover types, forest fires have a significant impact on the interannual variations in CO columns from both MOPITT and AIRS data over the land of Asia (r = 0.54-0.79), in all seasons except in winter. Interestingly, over cropland, grassland, and shrubland, fire counts only significantly correlate with CO columns in autumn over the land of Asia (r = 0.70-0.82). By subregion, there is a positive and significant correlation between CO total column and biomass burning in the fire-prone seasons and areas, for example, in spring, summer over South Siberia (r = 0. 65-0.73), and in all seasons over Indonesia (r = 0.73-0.95). Positive correlations are usually found in fire-prone regions [3,24,47], e.g., in winter over the land of Indo-China Peninsula. Yin et al. [25] also reported a significant correlation (r = 0.82) between monthly MOPITT CO total column and fire counts in Southeast Asia during the fire season (December-May). However, in the regions with strong anthropogenic emissions and few fire activities [15,18,32,65], negative correlations between CO columns and fire counts can appear, e.g., India, North China, and the Sichuan Basin in the annual mean, North China in spring, and the Sichuan Basin in summer.
Interpreting the decreasing trends in atmospheric CO in Asia requires accurate evolution of CO concentrations between multiple CO emissions sources (i.e., anthropogenic emissions, biomass burning emissions, biogenic and oceanic emissions, and chemical production) and the CO sinks (i.e., the CO chemical sink and dry deposition) [2,24]. Zheng et al. [24] stated that CO from chemical production, as well as from oceanic and biogenic sources changes a little on a global scale. In contrast, atmospheric transport play a confounding role in modulating the CO trends in a receptor region [5,34,56]. Zheng et al. [24] identified a declining trend in the global CO budget in 2000-2017, driven by reduced anthropogenic emissions in the US, Europe, and China, as well as by reduced biomass burning emissions globally. In this study, the declining trend in the annual total fire counts inside Asia is revealed (Figure 14a), which is mainly attributable to declining fire counts in spring. However, the GFED4 fire emission data show no significant trend inside Asia in all seasons (Figure 14b). Outside Asia, the downward trend in fire counts is statistically significant in all seasons except in summer (Figure 14c), while no significant trends in the GFED4 CO emissions are shown ( Figure 14d). As the GFED4 data show a different sensitivity of CO trend to biomass burning, future work using numerical models and observation evidence is needed to further address the issue. The deceasing trends in fire accounts inside and outside Asia are probably attributable to the following reasons. First, land use change (i.e., the expansion of crop and pasture lands leads to more burned areas but fewer emissions globally. Second, fires are suppressed more due to increased human efforts [52]. Third, the globe becomes wetter in the second half of the twentieth century in most regions of Asia [66] (i.e., western China, Central Asia, India subcontinent, and Indonesia).
There are multiple factors affecting the spatial variations and trends in CO total column among different subregions in Asia. For example, in China, these factors include uneven distributions of economic development, population, meteorology [56], farming techniques (i.e., the control of chemical fertilizers, the strict straw open burning ban policy) [67,68], and the region-specific forest management strategies (i.e., forest fire prevention) [69], which would result in a CO trend downward or upward. In South Asia and Indonesia, land cover use change (i.e., urbanization), and farming techniques (i.e., the expansion of crop and pasture lands) [66] likely lead to strong CO emissions. The length of the fire seasons in the southern Siberia (virgin boreal forest) has been projected to increase by at least one month due to the lack of precipitation [51,70], drier climate; higher fire danger would likely lead to huge CO emissions from fires. Pan et al. [71] found that fires are always more intensive in southern Kalimantan than in southern Sumatra in all EI Niño events in 1979-2016. More intense and prolonged Indonesian drought and fires occur in the Eastern Pacific type, during which the emitted carbon amounts almost double those in the Central Pacific type.

Comparison of Atmospheric Carbon Monoxide over Asia Observed by MOPITT and AIRS
In this study, we used two sets of satellite CO data to demonstrate the spatial-temporal variations in atmospheric CO over Asia to enhance confidence for our analysis. Although MOPITT and AIRS CO columns have a significant correlation over Asia up to 0.98 in all seasons ( Figure 5), some discrepancies between the MOPITT and the AIRS CO data are observed. First, both long-term mean and standard deviation of CO total column from MOPITT are higher than those from AIRS. Second, the mean CV from MOPITT (5.8%) is higher than that from AIRS (4.1%) over Asia, suggesting a larger interannual variation. Third, the decreasing trend in the annual mean of CO total column from MOPITT (−0.58% per year) is stronger than that from AIRS (−0.30% per year) over Asia. Fourthly, discrepancies between MOPITT and AIRS CO data are apparent in different seasons, among different subregions, and over different land covers, in terms of CO abundances, temporal-spatial variations, and sensitivity of CO to biomass burning (Tables S2-S4 and Figure S1).
These differences are likely due to multiple factors. First, MOPITT and AIRS uses different instruments; one uses a gas correlation radiometer and the other uses grating spectrometer. Second, at the nadir, the ground footprint of MOPITT measurement is 22 km × 22 km, and the scan angle of 26.1 • across the satellite flight track (640 km) allows a global coverage in 2.5 days [19,59,72], while AIRS has 13.5 km × 13.5 km footprint, and its swath (1650 km) provides near global coverage twice daily [19,73]. Therefore, MOPITT can capture high CO hotspots and result in high spatial variation [8], while AIRS data cover significantly large areas daily.
The MOPITT CO retrieval algorithm is a maximum a posteriori method that incorporates a priori information of the physical and statistical variability of the trace gas distribution in the atmosphere to choose the best solution [72,74]. Compared to CO retrievals from MOPITT, the current AIRS physical retrieval algorithm seeks to minimize the weighted difference between the clear column radiance observations [75] and the radiances computed using a forward model [8,76] by varying the geophysical state.
The AIRS CO retrievals use reconstructed cloudy grids while MOPITT removes cloudy grids [72,73,77]. Due to the lack of sensitivity in the lower troposphere for down-looking spectrometers such as AIRS [8], when the total column CO amount is high over the land in the Northern Hemisphere, AIRS CO columns are lower than MOPITT CO (Figure 5b). Over the oceans, AIRS CO columns are slightly larger than MOPITT CO (Figure 5c).

Conclusions
Using MOPITT and AIRS satellite data, which have high accuracy and long-term coverage, we investigated the seasonal and interannual variations, and the long-term trends in atmospheric CO in Asia over 2003-2017. Combing with the datasets of MODIS fire counts and GFED4 fire emissions, we explored the influences of biomass burning on the long-term variation and trends in atmospheric CO in Asia over different regions and different land covers.
CO total column over Asia from MOPITT is slightly higher than that from AIRS, and the two datasets show similar spatial and seasonal patterns. On annual average, CO columns over Asia from MOPITT and AIRS are (19.3 ± 4.3) × 10 17 molecules cm −2 and (18.5 ± 2.8) × 10 17 molecules cm −2 , respectively. The correlation coefficient between the two datasets ranges from 0.86 to 0.98 over Asia, being lowest over the Tibetan Plateau and India. CO total column over most areas in Asia exhibits a strong seasonality, being higher in spring and winter than in summer and autumn. The interannual variation in CO total column is large over Asia, with a regional mean CV of 5.8% in MOPITT data, although the CV is lower than the global mean.
The seasonal and interannual variations in CO total column over Asia is greatly impacted by biomass burning, especially over South Siberia, Indo-China Peninsula, and Indonesia. CO total column in Asia correlates more closely with MODIS fire counts than with GFED4 fire emissions. On annual mean, the correlation coefficient between MODIS fire counts and MOPITT CO total column over Asia reaches 0.76. MODIS fire counts may explain 60%, 36%, and 71% of the interannual variation in the annual mean CO total column over South Siberia, Indo-China Peninsula, and Indonesia, respectively. Meanwhile, MODIS fire counts may explain 62%, 52%, and 31% of the interannual variation in the annual mean CO total column, respectively, over forest, grassland, and shrubland in Asia. During 2003-2017, the peaks of CO total column over Asia are closely correlate to the extreme fire events, for instance, the severe fires over Siberia in 2003 and 2012 and over Indonesia in 2006 and 2015. The extreme fire events can remarkably increase the CO total column over Asia, which are observed by the MOPITT and AIRS.
From 2003 to 2017, according to MOPITT data, CO total column in Asia decreased significantly at a rate of −(0.58 ± 0.15)% (or −(0.11 ± 0.03) × 10 17 molecules cm −2 ) per year. The decreasing trend is significant over land but insignificant over oceans. Over land, the decreasing trend is most obvious over North China, South China, and the Sichuan Basin, with a regional mean over 1% per year. However, over India and Indonesia, no significant trends in the annual mean CO total column are observed. Seasonally, the decreasing trend over Asia is most significant in winter, following by summer and spring, while the decreasing trend in autumn is not significant. The declines of MODIS fire counts inside and outside Asia over 2003-2017 suggest that biomass burning may be one of the reasons for the decreasing trend in CO total column in Asia, especially in spring, although the decreasing trends in the GFED4 fire emission data are insignificant.
Based on satellite observations and statistical analysis, this study demonstrates the spatial variations and long-term trends in atmospheric CO over Asia and their sensitivities to biomass burning. The results help further understand the role of CO in atmospheric chemistry, air pollution, and carbon cycle. In the future, numerical simulations are needed to further quantify the contributions of various factors to the trends in atmospheric CO over Asia, including the influences of emissions from both anthropogenic activities and biomass burning.

Supplementary Materials:
The following are available online at http://www.mdpi.com/2072-4292/12/5/830/s1, Table S1: Land cover classification on the basis of the ESA CCI Land Cover, Table S1: The trends of AIRS CO total column in Asia, its sub-regions, and the world, Table S3: Correlation coefficients (r) between AIRS CO total column and the number of MODIS fire counts over different land cover and sub-regions in Asia during 2009-2017, Table S4: Correlation coefficients (r) between AIRS CO total column and the GFED4 CO emissions from biomass burning over different land covers and sub-regions in Asia during 2003-2017, Figure S1: Monthly variations in the anomalies of AIRS CO total column.