Threshold Value Determination for Cloud Masking Process using Landsat 8 Imagery

Optical remote sensing is inevitable from cloud cover problems, especially in tropical country. Cloud cover could reduce the potential usage of the imagery, for instance it would be impractical to use it for land cover classification. The existance of cloud and it’s shadow in an imagery could hinder image analysis such as image transformation and image classification. Furthermore, cloud cover could reduce the obervable area of an imagery. Automatic or semi-automatic cloud masking is considered an effective means of removing the cloud cover. Threshold value in the cloud masking process is essential to provide a clean cloud removal.Cloud masking was conducted using Landsat 8 imagery, which has been through pre processing such as Top of Atmospheric (ToA) correction, Bidirectional Reflectance Distribution Function (BDRF) correction, and radiometric terrain correction. LAPAN’s (National Institute of Aeronautics and Space) cloud removal algorithm was used since it provide a semi-automatic procedure. Four threshold value options were chosen based on pixel’s statictics in cloud, cloud shadow, water, and other objects that are potentially identify as cloud or cloud’s shadow. Cloud was defined based on albedo value using visible channels, whereas the cloud’s shadow is defined based on Near Infrared and Short Wave Infrared bands.Four threshold combinations were successfully made based on three different acquisition date of Landsat 8 imagery. The best threshold value should be able to identify cloud and it’s shadow, but shows minimal effect to the objects resemble to cloud or it’s shadow. The result shows that the most effective threshold is, 1650 and 3600 for cloud, 11000 for lower limit of shadow, 12000 for upper limit of shadow, -1350 for a set of cloud shadow, and 900 for water.