Study On Atmospheric Correction Of Hyperspectral Imagery Over Taihu Lake

Atmospheric correction methods of hyperspectral imageries (i.e. HSI and MERIS) for inland case II water are researched,based on the existing atmospheric-radiative-transfer theory and atmospheric correction methods.Neural network model is built and Gordon model for atmospheric correction is improved based on water-reflectance relationships. This research can provide technical support for quantitative retrieval applications of MERIS and HSI data.The major achievements are as follows:1. Through radiative transfer simulation, the impact on water image in Taihu Lake is analyzed, which is made by aerosol type, optical thickness change,besides, the error is computed caused by using default aerosol situation settings. Through the research, some conclusion can be drawn that large errors (40%) will be caused by wrong assumptions about aerosol type and optical thickness, and the optical thickness can make bigger impact and larger errors than aerosol type on water image.2. Based on the neural network, a model can be built that extracts water reflectance with satellite radiance, which doesn't need the synchronous aerosol information. Through the radiative transfer model, a data set that is simulated accurately can be built which contains different spectrum, aerosol types, and optical thicknesses. According to the optimization research, a BP artificial neural network (ANN) with a single hidden layer,10 hidden nodes, LM train function works efficiently and accurately in atmospheric correction of MERIS data; moreover, it has good generalization ability.3. After researching about spectrum of Taihu Lake, a linear relation is found between near-infrared spectrums, which is stable and isn't influenced by the change of the groups of water. In this paper the linear relationship is introduced to the Gordon model for atmospheric correction, then the aerosol indexεcan be achieved by optimization with the relationship between water scattering and aerosol scattering in the model, futher, the atmospheric correction method of HSI data for caseⅡwater can be realized. Using improved atmospheric correction, there are no negative values, that situation which Gordon model is applied to caseⅡwater directly; the corrected reflect spectrum accord with caseⅡwater characteristics; compared with the ANN results, they are similar both in form and value, and the two results confirm each other.