RECONSTRUCTION OF WATER MASK FROM OPTICAL CLOUDY LANDSAT SCENES USING NAIVE BAYES ALGORITHM

Floods are natural events that can generate major disasters. Remote sensing data can be used for mapping flood events. However, atmospheric conditions in images captured by optical sensors, such as the presence of clouds, affect the recorded data and impair the use of these images in hydrological studies. In this context, a machine learning algorithm – Naive Bayes – was used to determine the extent of water masks under two hypothetical cloud cover. The results, compared to the observed data, indicated that the proposed methodology could be used for the determination of water masks, since the performance metrics reached values for accuracy above 0.98 and for the critical success index above 0.70. Further tests and validations should be performed to prove the potential of reconstruction of water masks by the proposed methodology.