COMPARISON OF THREE TYPES OF AEROSOL PRODUCTS DURING 2015-2017 IN CHINA

In this study, we use sunphotometer (AERONET) data as reference to evaluate MODIS, Himawari-08, and MERRA-2 aerosol products over China. AERONET data from Beijing, Xianghe, Xuzhou, Taihu and Hong Kong are used. The results indicate that aerosol optical depths (AODs) obtained from MODIS, Himawari-08, and MERRA-2 from 2015-2017 over China are consistent with the AERONET observations. The accuracy of these AOD products shows a spatial and temporal variation. MODIS aerosol products tend to overestimate the AODs at Beijing, Xianghe and Xuzhou, with stronger overestimation in winter. For Taihu and Hong Kong, MODIS only overestimates the AODs in winter. Himawari-08 and MERRA-2 underestimate the AODs in autumn over almost all stations except Taihu.


INTRODUCTION
Atmospheric aerosols participate directly or indirectly in radiative forcing processes and cloud-forming processes, and have an extremely important impact on global climate change.By directly scattering or absorbing sunlight, aerosols can alter surface albedo and influence radiative budget.Therefore, increasing attentions have been paid to accurately determine the aerosol radiative effects in recent years [1][2][3][4][5] .In the study of atmospheric aerosols, people often have to obtain aerosol data all over the world.But, due to the limitation of the number of ground-based observation sites, it is not conducive to derive global aerosol properties for researches worldwide.Satellite remote sensing measurements can fill in these huge data voids and achieve global coverage easily.However, influenced by surface characteristics and weather conditions easily, the accuracy of satellite inversion data is lower than that of groundbased observations.So it is necessary that the accuracy of the aerosol products should be verified before they can be put into use [6] .With the development of measurement techniques and radiative transfer calculations, varies of satellite aerosol data products are available.Nowadays MODIS, Himawari-08 and MERRA-2 aerosol products are widely used as they provide aerosol optical properties with high spatial and temporal resolution on a global scale.Since the 1990s, the inversion algorithm of MODIS aerosol products has been developed for more than 20 years and the accuracy of data is constantly improving [7][8] .Munchak et al. developed a new algorithm (MxD04 3K) for the inversion of aerosol optical properties from MODIS observations [9] .Wang et al. conducted an accuracy test of MODIS data over China [10] .Zhang et al. validated the MODIS C6 AOD product based on a station in the Beijing-Tianjin-Hebei areas [11] .Different from MODIS aerosol products, Himawari-08 aerosol products have the advantages of higher spatial and temporal resolutions and more detection channels including the band 3 (0.64 µm), band 4 (0.86 µm), and band 6 (2.3 µm) [12][13] .As the Japanese geostationary meteorological satellite Himawari-08 is located at 140.7°E, which was launched in October 2014, its aerosol products provide important information of aerosol distribution over East Asia.Lots of efforts have been done to improve the algorithm of Himawari-08 aerosol products.Fukuda et al. proposed an improved Kaufman algorithm for the inversion of aerosol optical properties from Himawari-08 observations [14] .Zhang et al. proposed an improved algorithm for monitoring and forecasting air quality from Himawari-8/AHI over East Asia [15] .However, there is less work on the accuracy of Himawari-08 aerosol products.The Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2) from the NASA Global Modeling and Assimilation Office is the new modern satellite era [16] .To our knowledge, MERRA-2 assimilates numerous data sets, like the assimilation of bias-corrected AOD from the Advanced Very High Resolution Radiometer (AVHRR) instruments and analyzed aerosol fields that are radiatively coupled to the atmosphere [17] .Because of using a larger number of new observations than MERRA, MERRA-2 has made a huge improvement in many aspects.Like MODIS and Himawari-08 aerosol products, the accuracy of its distribution characteristics also has to be evaluated.In addition, the validation of MERRA-2 aerosol product is also necessary.In previous studies, validation is mainly focused on MODIS aerosol products [7][8][9][10][11] , and the result shows that the accuracy of MODIS aerosol products is constantly improving.However, there are not much work about the evaluation of Himawari-08 and MERRA-2 aerosol products.Based on this starting point, this paper evaluates the accuracy and applicability of three satellite aerosol products, MODIS, Himawari-08 and MERRA-2, choosing Beijing, Xianghe, Xuzhou, Taihu and Hong Kong as representative stations, and analyzes regional and seasonal differences.

MODIS, Himawari-08, MERRA-2 and AERONET AODs
The MODIS data set used in our analysis is Terra and Aqua MODIS AOD products (MOD04_L2) at 550nm, covering the time period from January 2015 to December 2017 with the spatial resolution of 10km×10km [1] .The Himawari-08 data set used in our analysis covers the time period from July 2015 to December 2017 with the spatial resolution of 5km×5km [2] .The MERRA-2 data set used in our analysis covers the time period from January 2015 to December 2017 with the spatial resolution of 0.5° latitude×0.625°longitude [3] .
The sunphotometer (AERONET) data is used as the benchmark result.In our study, the AERONET stations in Beijing, Xianghe, Xuzhou, Taihu and Hong Kong are chosen as representative stations (see Table 1).The effects of the complicated underlying surface condition are considered in the analysis of the aerosol characteristics [10] .

Methodology
Due to the lack of AERONET AOD data at 550nm, we use an interpolation method to interpolate AERONET AOD at 440nm and 675nm to obtain the AOD data at 550nm [18] .The interpolation method is For the purposes of comparison, AODs products of MODIS, Himawari-08, MERRA-2 and AERONET AODs in these five stations must be corresponded in space and time.The spacetime matching scheme is that MODIS, Himawari-08 and MERRA-2 AODs are averaged over a circular field with a radius of 50 km, and AERONET AOD is averaged within half an hour before and after the satellite overpass.The black and red lines represent the slopes of linear regression and the 1:1 lines of scatter plots.The black and red lines represent the slopes of linear regression and the 1:1 lines of scatter plots.

RESULTS AND DISCUSSIONS
Figure 3 shows the scatter plots of MERRA-

Figure 1
Figure 1 shows the scatter plots of MODIS AODs versus AERONET AODs in Beijing during different seasons.From Fig.1, it is obvious that MODIS AODs show good correlation with AERONET AODs in Beijing.The squared correlation coefficients for the linear regression fits R² are 0.261, 0.849, 0.291 and 0.61 in the four seasons, respectively.The root mean square errors between MODIS and AERONET AODs are very close to zero (0.153 for spring, 0.038 for summer, 0.093 for

Figure 1 .
Figure 1.Scatter plots of MODIS AODs versus AERONET AODs in Beijing (a) spring (b) summer (c) autumn (d) winter.The black and red lines represent the slopes of linear regression and the 1:1 lines of scatter plots.

Figure 2 2 ,
Figure2shows the scatter plots of Himawari-08 versus AERONET AODs in Beijing during the four seasons.From Fig.2, It is obvious that Himawari-08 AODs show good correlation with AERONET AODs in Beijing.The R² are 0.463, 0.766, 0.625 and 0.398 in the four seasons, respectively.The root mean square errors between Himawari-08 and AERONET AODs are very close to zero (0.077 for spring, 0.069 for summer, 0.082 for autumn and 0.075 for winter).Besides, the slopes of regression line are lower than unity (0.565 for spring, 0.497 for summer, 0.344 for autumn and 0.388 for winter).To put it differently, Himawari-08 AODs in Beijing tend to underestimate the AODs in every season.

Table 1 .
Observation time and level of AERONET AODs in the representative stations

Table 2 .
The squared correlation coefficients (R 2 ), the root mean square errors (RMSE) and the slopes (K) of regression lines of MODIS, Himawari-08, MERRA-2 comparing with AERONET AODs AODs have the best applicability for the summer in North China.In addition, Himawari-08 and MERRA-2 AODs have the best applicability for the autumn in East China.The R² of Himawari-08 comparing with AERONET AODs in Xuzhou are 0.413, 0.553, 0.571, 0.566 and in Taihu, are 0.408, 0.412, 0.658 and 0.154 in the four seasons.The R² of Himawari-08 comparing with AERONET AODs for the autumn in East China are higher than other three seasons.According to the regressed slopes, we can easily know situations when the AODs are overestimated by the three types of aerosol products.When the regressed slope is higher than unity, the aerosol products tend to overestimate the AODs.It can be clearly seen from Table2that AODs are overvalued by MODIS at Beijing, Xianghe and Xuzhou during summer and winter, with the strongest overestimation occurring in winter.As for Taihu and Hong Kong, overestimation only appears in winter for MODIS AOD products.Meanwhile, except for Taihu station in autumn, the Himawari-08 and MERRA-2 aerosol products always underestimate the AODs.This paper compares the AODs of the three types of aerosol products with AERONET AODs at Beijing, Xianghe, Xuzhou, Taihu and Hong Kong respectively.The results indicate that the aerosol optical depths (AODs) obtained by MODIS, Himawari-08, and MERRA-2 in China are basically consistent with the results from AERONET.In general, the accuracy of theses AOD products varies with regions and seasons.MODIS aerosol products tend to overestimate the AODs in summer and winter at Beijing, Xianghe and Xuzhou, and the strongest overestimation are found in winter.As for Taihu and Hong Kong, overestimation only appears in winter for MODIS AOD products.Himawari-08 and MERRA-2 underestimate the AODs in autumn over almost all stations except Taihu.