A comparison between six model-based methods to retrieve surface reflectance and water vapor content from hyperspectral data: A case study using synthetic AVIRIS data

Six model-based methods, ACORN, ATREM, ATCOR4, CAM5S, FLAASH and HATCH, were used to retrieve reflectance information and water vapor content from synthetic AVIRIS data. The results were compared using objective method in four spectral segments across the sun radiation light. In the synthetic data, the highest spectral difference was found to be 26% and the lowest 1.1% in reflectance units as estimated from the Average Sum of Deviations Squared (ASDS) values. The reflectanceequivalent ASDS values were lower than the traditional max and min values, as ASDS averages reflectance in the spectral segment in question. Two kinds of reflectance retrieval difference were identified: 1) albedo drift and 2) spectral artifacts. Judging from the four selected spectral segments examined (VIS, NIR, SWIR-1, SWIR-2) a sequence that allocate the best method in each segment was established. It was recommended that for optimum atmospheric correction all methods are to be run and then a method for each pixel will be selected using look-up-table based on the surface basic coverage and the atmospheric condition.. A similar comparison was done with the retrieval of water vapor content by each of the six methods used. The water vapor content was found to be overestimated by ATREM (≈60%) and underestimated by HATCH (≈ 15%), whereas ACRON and ATCOR provided the most realistic results. It was found that the water vapor retrieval difference varies within the actual water vapor content and is significantly affected by the vegetation content, mainly because of the liquid water contribution. The prediction difference for the water vapor retrieval was found to have two sources: 1) liquid water occurrence and 2) miscalculation of the method used. It was concluded that the results of a model based correction technique, may vary both in retrieving reflectance values and water vapor content quite significantly from one method to another. This is true even when optimal conditions are used (zero spatial and spectral noise, no spectra line curvature ect.) and the environment signals are controlled. When adding other obstacles to the raw data (e.g. spectral and spatial noise, optical effects and more) more study has to be applied in order to pinpoint on the best method to be used. Although this study shows a methodology to judge method's performance to remove atmospheric attenuation, a separate study has to be applied in order to generate a similar look-up-tables per sensor and per mission. Keyword list: Hyperspectral remote sensing, Atmospheric correction, AVIRIS