Comparison of data fusion methods for satellite-assisted determination of PM10 ambient concentration

Our recent work has demonstrated the feasibility of using satellite-derived data to draw quantitative maps of particulate loading within the planetary boundary layer. Our method, when used in conjunction with atmospheric dispersion models and ground data, can provide a comprehensive estimate of tropospheric pollution from particulate matter. Information filtering techniques are used to reduce the error of the information fusion algorithm and, consequently, produce the best possible estimate of tropospheric aerosol. Two data filtering methods have been used and their effectiveness with regard to overall error reduction is determined in this work. The first one is based on a weight scheme to take into account an empirical estimate of local error and/or uncertainty in input data. The second uses a modified Kalman filter for error reduction. The effectiveness of each of the filtering techniques depends on factors such as relative error variance across the computational domain, and precision of model input, i.e. on the accuracy of the ground emissions inventory and the reliability of measured ambient aerosol concentrations. The ICAROS NET fusion method was applied in the greater area of Athens, Greece over several days of observation in order to assess conclusively the adequacy of the information fusion filters employed.