In collaboration with the Institute of Electrical and Electronics Engineers Geoscience and Remote Sensing Society (IEEE GRSS) Earth Science Informatics Technical Committee (ESI TC), the Interagency Implementation and Advanced Concepts Team (IMPACT) organized a successful summer school session titled "Geospatial Data Science: An Experiential Journey," for thirty diverse students from around the world.
Held from 5/29-6/2/23, at the University of Iceland and hosted by the High-performance and Disruptive Computing in Remote Sensing (HDCRS) working group, the session aimed to provide technical guidance and training on end-to-end data science for Earth science. With a focus on open science, collaboration, and skill development, the summer school program included in-depth discussions and practical sessions on the topic of data harmonization, focusing specifically on the production, utilization, interactive exploration of Harmonized Landsat Sentinel-2 (HLS) datasets. The program further covered the development of the foundation model using HLS data with pre-training and a hands-on exercise of fine-tuning the foundation model. Additionally, participants used the model to inference with new HLS data. Participants had the opportunity to actively engage and exchange ideas and conduct hands-on exercise in an open environment leveraging NASA's Visualization, Exploration, and Data Analysis (VEDA) platform. The session was jointly developed by IMPACT and IBM Research, with contributions from experts in the field, including Manil Maskey (IMPACT/ESI TC Chair), Brian Freitag (IMPACT), Muthukumaran Ramasubramanian (UAH/IMPACT), Iksha Gurung (UAH/IMPACT), Linsong Chu (IBM Research), Paolo Fraccaro (IBM Research), Blair Edwards (IBM Research), Johannes Jakubik (IBM Research), and Sean Harkins (Development Seed/IMPACT). The students successfully fine-tuned the foundation model for flood water detection and burn scar identification. Resources used in the summer school are available at https://github.com/NASA-IMPACT/summer-school-2023.
