Bilan d'erreur pour la correction atmosphérique d'images hyperspectrales dans le visible et le proche infrarouge
Improvements to the predictions of climatic and climatological models depend, in part, on the quality of the modelization of the atmosphere. Its precision, in turn, depends on our theoretical understanding and on our capacity to evaluate the relevant atmospheric parameters. In the past, numerous efforts have been made but very few have dealt with an exhaustive study of the error factors, hence the lack of information on their accuracy. The current trend favouring quantitative over qualitative remote sensing necessitates improvements of our knowledge of the impact of atmospheric effects on image data. This thesis contributes to this goal. To this end, we present a simple yet efficient approach to the estimation of the error budget on the prediction of the apparent at-ground bidirectional reflectance factor (BRF) from the apparent at-sensor BRF. This method is essentially a sensibility analysis. Contributions from the different parameters are decomposed according to their relative importance to the total error. Results show that, in the case of the CASI sensor for the selected sites, the relative error (percentage error on the apparent at ground BRF) is around five percent, with a significant increase to about twenty percent in both the blue and the near-infrared. Sensor calibration appears as the largest source of error, aerosol optical depth being a distant second. The method is then validated according to its accuracy (absolute validation) through the extrapolation to the ground of the apparent at-sensor BRF acquired from multi-altitude imagery. The apparent at-ground BRF obtained is then considered representative of the ground truth and thus constitutes an absolute validation of the method. Results demonstrate the validity of the method to estimate the magnitude of the error on the atmospheric correction.