A new model-driven correction factor for BRDF effects in hyperspectral remote sensing data

Abstract

Interpretation of hyperspectral remotely sensed (HRS) imagery can be degraded by bi-directional reflectance distribution function (BRDF) effects that contribute an unknown amount of error to reflectance values. In this work we test a new empirical approach, using in-flight records concerning sensor and sun geometry at the time of acquisition, and laboratory BRDF measurements of selected land-cover classes from the new Israeli Goniometric Facility (IGF) at the remote sensing laboratory, Tel-Aviv University. BRDF datasets are then nadir-normalized (i.e. transformed to anisotropy), spectrally resampled to the required sensor and inverted to form correction vectors for the real imagery. These correction vectors are finally applied only to class-specific pixels of interest. We demonstrate this application for a non-georeferenced CASI image and four georeferenced HyMAP images and discuss results. The CASI data preliminary average anisotropy reaches 20% and corrects down to 5%. Its RMSE values are reduced by about 40-70%. HyMAP data Uses the same model and reduces average pre-correction ANIF by 25-50%, depending on wavelength. Since angular information about the sensor is the base for this correction, natural variability of the corrected land-cover classes is maintained. Therefore this method allows a class-specific BRDF correction that improves interpretation capability and quantitative analysis.