Data Science
Sciences and Exploration Directorate - NASA's Goddard Space Flight Center

Innovation Lab Use Case

CNN outperforms Random Forest for cloud detection

True color (red, green, blue) WorldView imagery overlain with Random Forest (RF) and Convolutional Neural Network (CNN) results (left to right). RF incorrectly classifies building structures and water sources as cloud pixels (top, delineated with red circles and ovals). CNN outperforms the RF in the correct identification of cloud edges and thin clouds (middle). RF incorrectly classifies bright building structures and canals as clouds (bottom). © 2013, 2016, 2019 DigitalGlobe, Inc., a Maxar company, NextView License.

False color, RF and CNN plots