IBM and NASA have jointly released Prithvi, an advanced AI foundational model focused on climate and weather research. The model was launched on the Hugging Face platform during Climate Week 2024. Developed collaboratively by IBM, NASA, and Oak Ridge National Laboratory, Prithvi extends beyond traditional forecasting tasks, providing valuable insights for scientists, developers, and enterprises.
The name "Prithvi" is derived from the Sanskrit word for "Earth." It was first introduced in May this year and is scheduled for open-source release later in the year. Unlike existing models that primarily focus on short-term forecasts, Prithvi is versatile, suitable for applications ranging from real-time weather predictions to long-term climate change studies. Its unique architecture and training methodologies enhance accuracy, processing speed, and reduce costs.
This release of Prithvi includes two specialized versions:
- A refined model that enhances meteorological and climate data resolution by 12 times, aiding in generating more detailed local-level forecasts and predictions.
- Another is a gravity wave parameterization model, which assists researchers in better understanding these atmospheric phenomena and their impact on processes such as cloud formation.
Prithvi can be fine-tuned with minimal additional data, efficiently accomplishing a variety of tasks. For example, in an experiment, the model accurately reconstructed global surface temperature distribution using only 5% of the original data.
The model was pre-trained on forty years of Earth observation data from NASA's MERRA-2 dataset. It is designed for applications on global, regional, and even local scales, making it suitable for a wide range of weather and climate research areas.
The launch of Prithvi is part of a collaboration between IBM's research division and NASA, aiming to leverage artificial intelligence to explore geophysical processes. Previously, the two organizations released large-scale open-source geospatial AI models on Hugging Face, intended for government agencies to study disaster patterns, biodiversity, and land-use changes.
Additionally, IBM collaborated with Canada's Environment and Climate Change Department to test Prithvi's capabilities in real-time precipitation forecasting and high-resolution climate modeling. These partnerships highlight Prithvi's versatility and its potential to advance climate science and support public safety.
As the global challenges posed by climate change intensify, open-source models like Prithvi aim to accelerate related research and deepen human understanding of complex weather systems and climate change. Now, researchers worldwide can access Prithvi and its specialized models through Hugging Face, enhancing their capability to address climate change and improve the accuracy of weather forecasts.
