Abstract: Land use and land cover (LULC) monitoring is a vital tool in mapping urbanization, deforestation, and climate change. In LULC analysis, each patch of a satellite image is classified as a pasture, forest, highway, industrial area, residential area, or sea/lake. EuroSAT (Helber et. al., 2019) is a labelled dataset of ~27,000 satellite images taken over Europe with the Sentinel-2A satellite. In this tutorial, we use Pytorch and EuroSAT to train a convolutional neural network for LULC classification. We explore important problems in earth observation: (1) domain shift in satellite data, (2) temperature trends of different LULC categories, and (3) LULC change detection. Using Google Earth Engine, GeoJSON, Rasterio, and Folium, we create high-quality, interactive visualizations of the satellite images and the corresponding LULC classifications.
Bio: Bio Coming Soon!
Ph.D. Student | University of California, Berkeley