Earth science researchers routinely hit a basic but stubborn bottleneck: figuring out what datasets exist, what they actually contain, and whether they can be used together. According to HPCwire, this discovery and assessment problem is a major time sink for scientists working with geospatial data.
To address it, the San Diego Supercomputer Center is helping develop GeoCroissant, an effort aimed at improving AI-ready geospatial data. The work involves Doug Fils of UC San Diego's Halıcıoğlu School of Data, who is part of the push to make datasets more findable and usable for machine learning workflows.
The underlying goal is to reduce the manual effort required to evaluate whether disparate geospatial datasets are interoperable, making it easier for researchers to build AI models on top of them.