Hyperspectral sensors provide researchers and governmental authorities with a wealth of information due to their fine spectral resolution, numerous bands, and wide spectral range. These sensors are used in various fields, including agriculture, environmental and forestry monitoring, geology, biology, medicine, and food quality assessment, among others. Generally, they measure across the visible and infrared parts of the electromagnetic spectrum, but they cannot penetrate thick cloud layers, which makes observations unusable under cloudy conditions. Also, the presence of thin and very thin clouds is a problem for the accurate retrieval of surface and atmospheric parameters. The PRecursore IperSpettrale della Missione Applicativa (PRISMA) is a medium-resolution hyperspectral imaging satellite, developed, owned, and operated by Agenzia Spaziale Italiana, launched in orbit on the 22 March 2019. PRISMA carries two sensor instruments, the HYC Hyperspectral Camera module and the panchromatic camera module. In this article, we present the results we obtained by testing some machine learning techniques for cloud detection on Top of Atmosphere (TOA) reflectance data. In particular, we focused on k-nearest neighbors, random forest, and extreme gradient boosting trained on a dataset of manually annotated images by the authors, after transforming the L1 TOA radiance in reflectance data. We also provide numerical comparison with the Cloud detection in hyperspectral images with atmospheric column water vapor method.
