Segmentation of satellite images using machine learning algorithms for cloud classification

Abstract

Clouds play a significant role in determining the state of a changing weather. Clouds offer useful information for forecasting precipitation and provide measurement for showcasing solar irradiance variability. The influence of specific types of clouds on rainfall prediction and solar radiance has been discussed in this paper. Various segmentation algorithms, clustering algorithms and supervised machine learning algorithms such as K Nearest Neighbors and Random forest have been used to segment/classify the clouds using the dataset obtained from INSAT-3DR satellite. Clouds have been classified into high level clouds (Cirrus clouds), medium level clouds (Alto clouds) and low level clouds (Stratus clouds) in accordance with the altitude and cloud densities. The performance metrics has been found for the segmented images. Parameters that provide optimum results for supervised machine learning algorithms have been explored. On the images, different machine learning algorithms have been compared.