An unmixing-based BRDF correction in spectral remote sensing data

While extensive research addressed the Bidirectional Reflectance Distribution Function (BRDF) effect and approaches to correct it, too few works have considered the influence of spectral mixture on the correction results. This work studies the BRDF effect in spectral data and presents an approach to correct its undesired impact, considering the likely presence of mixed pixels. We propose an unmixing-based semiempirical model, incorporating endmembers’ (EMs) fractions within the data correction. To evaluate the performance of the proposed methodology, we conducted experiments with laboratory and aerial hyperspectral image data. The outcomes from all experiments reveal vital insights into the influence of the spectral mixture on the BRDF correction accuracy. Nevertheless, the results clearly show that the accuracy of the corrected reflectance significantly improved using the proposed model for reducing the BRDF effect, regardless of the pixel's microtopography arrangement. Most importantly, a quantitative assessment of the difference in reflectance values within the overlapping region between two aerial images shows that the unmixing-based model outperforms the commonly used one. While the traditional method reduces the influence of the BRDF in the corrected data by a factor of two, with an average Mean Absolute Error (MAE) of≈2.5%, the proposed approach reduces it by a factor of four with an average MAE≈1.2%.