Location
Virtual
This workshop explores the evolving landscape of Geospatial Artificial Intelligence (GeoAI) in High-Performance Computing (HPC) environments. We'll start with the foundational concepts of GeoAI, explaining common tasks and how AI is applied to geospatial data like satellite imagery.
Next, we'll introduce the practical tools that facilitate this work and set up a working environment on HPC Clusters. We'll examine the TorchGeo library, which offers a comprehensive toolkit for training models on geospatial datasets within PyTorch. We'll also cover TerraTorch, a fine-tuning and benchmarking toolkit that extends capabilities for Geospatial Foundation Models.
Finally, we'll present several case studies to demonstrate the transition from traditional, single-task GeoAI models—such as a model trained to identify specific land cover types—to the latest advancements in Geoscience Foundation Models. These larger, more flexible models can perform specific geoscience tasks with significantly less new data than required for traditional models. We will also introduce three methods for performing these tasks with HPC, with a main focus on two of them through online experiments: interactive jobs via OoD and batch jobs.