NASA and IBM Research have collaborated to develop a new AI model for weather and climate called Prithvi-weather-climate. This model is pre-trained using 40 years of data from NASA’s MERRA-2 and aims to incorporate AI and machine learning methods into weather and climate applications like storm tracking and forecasting. According to NASA’s open science policies, Prithvi-weather-climate will be openly available to the public through hugging face, an open-source machine learning model repository.
Foundation models (FMs) are essential for AI systems to process large amounts of data and generate results based on data associations. The development of FMs pre-trained on vast amounts of unlabeled data can serve as a basis for various tasks with minimal tuning. The Prithvi Geospatial FM was created in 2023 using NASA’s HLS data, leading to the development of Prithvi-weather-climate.
The team behind Prithvi-weather-climate includes experts from NASA, IBM Research, Oak Ridge National Laboratory, NVIDIA Corporation, and several universities dedicated to AI and large-scale computing. The model development was initiated at a workshop in 2023, where the team planned the next steps in developing and pre-training the model. The focus was on creating a model that could support a wide range of scientific applications rather than just forecasting.
MERRA-2 data forms the basis of Prithvi-weather-climate, providing insights from 40 years of global reanalysis data assimilating space-based observations of aerosols. The collaboration between NASA and IBM Research highlights the importance of open-source technologies for advancing research in areas like climate change.
The applications of Prithvi-weather-climate range from enhancing resolution in long-term climate models to improving public safety through more accurate forecasts of severe weather events. The model can also generate targeted forecasts using localized observations and improve predictions of various weather patterns.
The development team for Prithvi-weather-climate includes experts from NASA, IBM, universities, and other research institutions. Their work aims to leverage AI and machine learning to advance climate science and improve resilience to weather-related hazards. Through open collaboration and cutting-edge technology, Prithvi-weather-climate represents a significant step forward in the field of weather and climate forecasting.
Article Source
https://www.earthdata.nasa.gov/learn/blog/nasa-ibm-weather-climate-foundation-model