model

Why we included this project

Teams working with satellite imagery and environmental data will find Clay a practical starting point for geospatial machine learning. It is a pretrained Vision Transformer that takes Sentinel-1 and Sentinel-2 imagery along with location and time information and produces semantic embeddings of any area on Earth's surface. Those embeddings can be used directly to locate features like surface mines or aquaculture, or fine-tuned for tasks such as crop type mapping, biomass estimation, or tracking deforestation and flooding. The repo ships the model code, pretrained weights, and Jupyter tutorials that walk through embedding generation, reconstruction, and fine-tuning on real datasets, so you can go from install to a working pipeline without assembling pieces from several projects. Since the weights and code are Apache-licensed, Clay also works as a backbone inside your own downstream models.