FEASIBILITY OFAEROSOL RETRIEVALS FROM HYPERSPECTRAL DATA Fabio Del Frate 1 , Antonio Di Noia 1 , Pasquale Sellitto 1 , and Gabriele Curci 2 1 Tor Vergata University, via del Politecnico 1, 00133 Rome, Italy 2 CETEMPS, via Vetoio 1, 67010 Coppito, L’Aquila, Italy ABSTRACT In this paper, the results of an experiment, aimed at dis- criminating between different aerosol types from hyper- spectral data, will be shown. Several simulations of re- flectances at high spectral resolution, similar to those ac- quired by an hyperspectral sensor, were performed under different aerosol typologies and loads, and different at- mospheric conditions. A neural network approach was used in order to classify the aerosol type. Although pre- liminary, the obtained results seem to indicate that hyper- spectral measurements, thanks to the contiguity of their spectral channels, can be effective for aerosol classifica- tion. Key words: Aerosols, Radiative Transfer, Neural Net- works. 1. INTRODUCTION Satellite remote sensing of atmospheric aerosols is a very important task for both climate research and air quality monitoring. Aerosols absorb and scatter electromagnetic radiation mostly in a spectral range which extends from ultraviolet to near infrared. Over the last two decades, satellite measurements made in this spectral interval have been widely exploited with the aim to infer chemical, physical and radiative properties of atmospheric aerosols on a global scale [1]. For the aerosol physical properties to be determined, its size distribution and its behaviour in terms of several pa- rameters, such as optical depth, single scattering albedo and complex refractive index, as a function of wave- length, must be known. Retrieving such quantities from satellite measurements is very hard for two reasons. The first reason is that the retrieval problem is ill-posed, be- cause different combinations of optical depth, single scat- tering albedo and size distribution can lead to the same measured signal [2]. Figure 1 gives an example of this fact: aerosols of different typologies, characterized by different optical depths, can lead to very similar measured reflectances. The second reason is represented by the in- herent difficulties encountered in isolating the effect of Figure 1. Similarity between simulated reflectance spectra generated assuming different aerosol typologies, characterized by different optical depths. aerosols from those of atmospheric gases and Earth’s sur- face on the radiance measured by a satellite sensor. For these reasons, any operational algorithm for aerosol re- trievals must rely on a priori assumptions about Earth’s surface albedo, atmospheric state and aerosol type or op- tical depth [2]. While, in some circumstances, the sur- face albedo can be directly estimated from satellite mea- surements performed by the same sensor, the a priori in- formations about the other parameters are often obtained by simply using climatological data and complementary considerations. On the other hand, directly estimating the predominant type of aerosols within a given scene from satellite measurements could lead to more accurate re- trievals of aerosol optical depth. The aerosol typology affects the reflectance at the Top Of Atmosphere (TOA). An example of the influence of the aerosol type on the TOA reflectance is shown in Figure 2. In this figure, simulated reflectance spectra correspond- ing to four different aerosol types, are drawn. The figure was obtained by performing radiative transfer simulations under the same atmospheric conditions, i.e. the same ver- tical profiles of temperature, pressure, and concentration of the most relevant trace gases, and assuming the same surface albedo. Furthermore, the same Aerosol Optical Depth (AOD) and asymmetry parameter were assumed for the four simulations. _____________________________________________________ Proc. ‘Hyperspectral 2010 Workshop’, Frascati, Italy, 17–19 March 2010 (ESA SP-683, May 2010)