In 2023, NASA partnered with IBM Research to create an AI geospatial base model, artificial intelligence. The model, formed with large amounts of harmonized Landsat and Sentinel-2 (HLS) data widely used by NASA, provides a foundation for a variety of AI-driven studies to address environmental challenges. In accordance with the principles of open sciencethe model is freely available for anyone to access.
New Integrated Forecasting/Artificial Intelligence System at the European Centre, ECMWF
NASA joins forces with IBM
The basic models serve as a foundation from which scientists can develop a diverse set of applications, enabling powerful and efficient solutions. “Basic models only know what things are represented in the data“he explained Manil Maskey, data science leader at NASA’s Office of the Chief Scientific Officer (OCSDO). “It’s like a Swiss army knife: it can be used for many different things“.
Once it is created a basic modelcan train with a small amount of data to perform a specific task. To date, the Interagency Implementation and Advanced Concept Team (IMPACT), along with its collaborators, have demonstrated the capabilities of the geospatial-based model by fine-tuning it to detect fire scars, delineate flood water, and classify crops and other use categories. from the earth.
Due to the computational resources required to create the initial basic model, a partnership was necessary to be successful. In this case, NASA provided the data and scientific knowledge while IBM provided computing power and expertise in optimizing AI algorithms. The team’s shared commitment to making their research accessible through open science principles ensures that their model can be useful to as many researchers as possible.
“To build a basic model at scale, we realized from the beginning that it is not feasible for a single institution to build it“Maskey said.”Everything we’ve done on our base models has been open to the public, from pre-training data, code, best practices, model weights, training data tuning, and publications. There is transparency, so researchers can trace why certain things were used in terms of data or model architecture.either”.
After the success of his geospatial basis modelNASA and IBM Research continue their partnership to create a new and similar model for meteorological and climate studies. They are collaborating with Oak Ridge National Laboratory (ORNL), NVIDIA, and several universities to bring this model to life.
Like the geospatial-based model, The weather and climate model is being developed with an open scientific approach and will be available to the public in the near future.
To cover all aspects of Earth sciences would require several basic models trained on different types of data sets. However, Maskey believes those future models could one day be combined into a comprehensive model, leading to a “digital twin” of Earth that would provide unparalleled analysis and predictions for all types of climate and environmental events.
This entry was published in News on Apr 29, 2024 by Francisco Martín León
