Using Satellite Remote Sensing Data For Monitoring Respirable Particulate Matter Concentration

concentration is one of the most important quantitative indicators to air quality. Efficient collection of the concentration and distribution information of PM10 at a large scale benefits to control and improve air quality. PM10 can be measured using ground instruments that provide accurate information reflecting the spatial and temporal change of pollutants in a single location. However, ground observation is impractical if measurements are to be made over relatively large areas or for continuous monitoring. Therefore, satelite remote sensing has been administered to compensate for the limitations of ground instruments. Currently, the satellite data has been successfully applied atmosphere research, e.g. inversion of aerosol products, pollutant monitoring, study of sudden atmospheric pollution events, study of pollutant regional transportation etc.Conventional PM10 inversion algorithms are executed in two steps. Firstly, aerosol optical depth (AOD) data is extracted from satellite images; then, PM10 is estimated by modeling between AOD and ground measured data. However,due to nasa’s operatinal aod product has low spatial resolution,the PM10 inversion result based on this can not reflect the details of the distribution of ground’s PM10. Therefore, it is difficult to guarantee a high correlation between them while using the average daily as input for the model.Aim to provide new solutions to the above issue, a new approach for monitoring PM10 concentration using satellite data (e.g. MODIS) is proposed. In this approach, two analysis methods i.e. time-frequency analysis, statistical learning and an atmospheric correction algorithm based on radiative transfer model are evaluated. Major contributions of this thesis are as follows:(1) image (including images after wavelet analysis) characteristics, e.g. grayscale, edge energy etc. together are quantitatively measured; and the correlations between these measurements and PM10 concentration are studied; (2) generating the regression relationship between AOD and filed measured PM10 concentration via using filed-based aerosol remote sensing data to modify MODIS AOD product. An AOD inversion procedure is then followed. (3) considering the atmospheric conditions of Zhejiang Province, NASA’s operational AO-D inversion algorithm is modified to a better output with higher spatial resolution. Base-d on this modified algorithm, field PM10 concentration is then inversed. (4) modeling the difference map (difference between before and after atmospheric correction) from MODIS imagery with PM10 concentration using support vector machine regression (SVR).