Study On Remote Sensing Retrieval Method For Atmospheric Inhalable Particulate Matter

These years, as the hazes frequently appearing in Beijing, airpollution is turning into our country’s hot theme, especially PM2.5. It’sclear that two ways can be used to obtain particulates’ spatialdistribution, including field monitoring and remote sensing. Remotesensing is best characterized by wide cover, real-time operation, whichis focus on building tie-in equation between retrieval parameters (AOT)and particulate matters. Further research on the field measurement findssome drawback. For example, many researchers adopt random sampling andstatistical interpolation to analyze particulate pollution in a largescale, but ignore the importance of operational synchronism, the essenceof enough samples, and the time-consuming and labor-intensive feature.Thus, the PM study centre has undoubtedly transferred into remote sensingimages.This paper takes Beijing as the main study field, use Landsat and MODISimages to inverse AOT. In addition, the introduction of meteorologicalfactors can improve the accuracy of AOT retrieval. The results are takenas following:(1) Based on different data sources, namely Landsat7、Landsat8、MODIS, this study adopts5kinds of AOT retrieval algorithms(SARAAlgorithm, Improved SARA Algorithm, DDV Algorithm, IARA-LandsatAlgorithm, Landsat8Algorithm). And correlation analysis betweenretrieved AOT and MODIS MOD04AOT products is used to evaluate theperformance of these models. It’s not hard to find that SARA algorithmhas a higher accuracy and AOT retrieved in this method has a0.95coefficient with48samples of AERONET AOT; although the R value of AOTgained in Improved SARA retrieval is only0.83, this model solves the AOTretrieval problem that some remote places cannot reach the coverage of the AERONET stations; The DDV model cannot be compared with others dueto less verification points.(2) Its originality is the application of IARA-Landsat Algorithm,solving the surface reflectance calculation of "dark pixel" and "brightpixel". Based on the satellite images of "clean days" and "dirty days"over the same period, the IARA-Landsat model calculates dark objections’reflectance by normal DDV and assigns reflectance values in clean daysto bright pixels. Then taking ground-measurement AERONET AOT as a standard,the goodness of fittest differences between retrieved and MODIS MOD04AOTcan highlight the IARA-Landsat algorithm.(3) AOT retrieval based on Landsat8images introduces the NASA V5.2algorithm that can contributes to calculating pixels’ surface reflectance.Finally, using traditional multivariate linear and BP neural network andcombining meteorological factors can improve the accuracy of AOTretrieval, the latter of which has a better fitting.