Satellite Image Future Landscape Prediction using Conditional Adversarial Networks

Satellite images are important to get information related to the Earth's resources and environment. Landsat, a series of Earth-observing satellite missions, tracks use of land and records changes in it due to natural and human activities. Automated temporal analysis and prediction of landscape has various benefits in its planning and development. In this work, we use Conditional Adversarial Networks for the Landsat satellite image prediction. We not only learned a mapping from input images to output images, i.e., mapping from satellite images from 1999–2000 to satellite images from 2009–2010, but we also learned a loss function to train this mapping. The results are satisfactory in the sense that they capture the general trends over time.