Dealing with Cloud Shadows in Surface Reflectance Retrieval from Drone Hyperspectral Imagery

Drone-based hyperspectral imaging is a powerful tool for environmental monitoring, yet its reliability in mid-latitude regions is often compromised by non-ideal weather conditions. Specifically, cloud shadows and light scattering by Cirrus clouds are influencing the irradiance and are difficult to parameterize. Standard atmospheric correction tools typically assume clear skies, leading to significant inconsistencies in retrieved surface reflectance values when clouds are present. This paper presents a new approach on the basis of the DROACOR atmospheric correction method which couples an empirical cloud detection routine with physical cloud transmittance calculations. For the detection of the spatially variable cloud shadows, an algorithm based on statistical analysis of spatially smoothed images is used. This is coupled with a physical model for cloud transmittance and spherical albedo retrieval. This combination is tested for the adjustment of reflectance in shaded areas to match illuminated conditions. Applied to data from HySpex VS620 and AVIRIS-4 systems, the method successfully retrieves surface reflectance values in the VNIR spectrum, comparable to fully illuminated areas. However, challenges remain in detecting faint shadows, correcting diffuse shadow borders, and retrieval of accurate surface reflectance in the SWIR spectral range.