A cloud detection algorithm-generating method for remote sensing data at visible to short-wave infrared wavelengths

To realize highly precise and automatic cloud detection from multi-sensors, this paper proposes a cloud detection algorithm-generating (CDAG) method for remote sensing data from visible to short-wave infrared (SWIR) bands. Hyperspectral remote sensing data with high spatial resolution were collected and used as a pixel dataset of cloudy and clear skies. In this paper, multi-temporal AVIRIS (Airborne Visible/Infrared Imaging Spectrometer) data with 224 bands at visible to SWIR wavelengths and a 20 m spatial resolution were used for the dataset. Based on the pixel dataset, pixels of different types of clouds and land cover were distinguished artificially and used for the simulation of multispectral sensors. Cloud detection algorithms for the multispectral remote sensing sensors were then generated based on the spectral differences between the cloudy and clear-sky pixels distinguished previously. The possibility of assigning a pixel as cloudy was calculated based on the reliability of each method. Landsat 8 OLI (Operational Land Imager), MODIS (Moderate Resolution Imaging Spectroradiometer) Terra and Suomi NPP VIIRS (Visible/Infrared Imaging Radiometer) were used for the cloud detection test with the CDAG method, and the results from each sensor were compared with the corresponding artificial results, demonstrating an accurate detection rate of more than 85%.