Fine tuning of a haze detection threshold in atmospheric correction of spaceborne optical imagery

The focus of this work concentrates on the atmospheric haze phenomenon. Haze as a common image pollution is a known effect for any kind of photography. It originates in the scattering of light in the atmosphere due to water vapor or aerosol particles (Makarau et al. 2014, 5895). In remote sensed data it is essential to correct the effect of haze since the data is processed later to retrieve valuable information about the surface. Compared to thick clouds, that limit the possibility to retrieve surface data, haze is mostly characterized as transparent enough to still retrieve the information from the surface below. At first, the workflow of the module is explained. The haze module, just as many other modules from PACO are inherited from the DLR-developed ATCOR program. For ATCOR, two methods are available with the Haze Optimized Transform (HOT) always being executed but limited to visible range channels, while the calculation of a Haze Thickness Map (HTM) provides more accurate results and is currently embedded in PACO (Richter & Schläpfer 2019, 102, Makarau et al. 2014, 5895). The calculation is based in a Dark Object Subtraction Method (ref. Chapter 1.1), where dark pixels such as shadows are used to determine the thickness of both haze and aerosols. An extrapolated band (usually from two bands from the blue area of the spectrum) is used for the dark pixel search. This is done to avoid overcorrection, which can occur when using only one blue band. Later, a band specific HTM is calculated with the help of a regression coefficient. This ensures that the decreasing influence of Rayleigh-scattering for the longer wavelengths is considered while dehazing. Before the dehazing is done, a binary mask is employed to label hazy and haze-free regions. This is done by taking an additional HTM with larger window size than the band specific ones. By thresholding, the mask is created. The threshold tHTM is calculated as the following with m(HTM) being the mean of the HTM, hs being haze sigma and \sigma being the standard deviation of the HTM