An automatic method for open water detection using MUX/CBERS-4 images

Most of the rivers in the planet have been dammed for providing water storage for human needs such as energy production, irrigation and for domestic and industrial use. At present, there are thousands of reservoirs in the world in need of frequent monitoring to support proper water resources management. Remote sensing provides a large amount of data for monitoring the Earth surface but faces time consuming processes for extracting the information needed hampers its use for real time management. This paper describes an automatized method for mapping open water bodies using MUX/CBERS-4 images in order to speed the process. The method consists in applying colour transformation in all RGB combinations of MUX bands transformed to HSV (Hue Saturation Value) images and empirically defining the optimum Hue interval for splitting image pixels between two classes: water and non-water. In order to do that, all RGB compositions of MUX bands were transformed to HSV images and tested to select the set providing the best separation between water and non-water. The Hue interval was used as input for the LEGAL (Linguagem Espacial para Geoprocessamento Algébrico) available at Spring 5.2.7 to split the Hue image into water and non-water pixels. A statistical analysis was applied to aid the choice of the best composition. The RGB 587 was chosen as the best composition to identify water bodies in MUX images. Future work recommendations include applying distinct confidence intervals and performing a pre-processing of the images, including image calibration and atmospheric correction.