IBM Japan IBM and NASA release open source AI model suitable for weather and climate applications on Hugging Face

[IBM Japan] IBM and NASA release open source AI model suitable for weather and climate applications on Hugging Face ​
IBM Japan Press release: September 27, 2024 IBM and NASA release open source AI model suitable for weather and climate applications on Hugging Face New AI-based models provide insights beyond prediction to help scientists, developers and businesses better understand and analyze weather and climate data [Yorktown Heights, New York, USA – September 23, 2024 (local time)] IBM today announced new AI foundation models for a variety of weather and climate use cases that are available as open source for the scientific, developer, and business communities. Developed jointly by IBM and NASA in collaboration with Oak Ridge National Laboratory, this foundational model is a flexible, scalable model to address a variety of challenges associated with short-term and long-term climate prediction. provide a method. The Foundation Model for Weather and Climate, as outlined in the recently published paper Prithvi WxC: Foundation Model for Weather and Climate, has a unique design and learning regime that allows it to build upon existing weather AI model Many more applications can be considered. Potential applications include creating targeted forecasts based on local observations, detecting and predicting extreme weather patterns, increasing the spatial resolution of global climate simulations, and improving the spatial resolution of numerical weather and climate models. This includes improvements to the way physical processes are represented. In one of the experiments in the above paper, the underlying model accurately reconstructs the Earth’s surface temperature from just a 5% sample of data randomly drawn from the original data, which is an important contribution to the data assimilation problem. It suggests broader applications. The basic model announced this time was pretrained using 40 years of Earth observation data from NASA’s Modern-Era Retrospective analysis for Research and Applications, Version 2 (MERRA-2). Featuring a unique architecture that can be fine-tuned at global, regional, and local scales, this flexibility as a foundational model makes it suitable for a variety of climate studies. In addition to this basic model available for download on Hugging Face, two fine-tuned models are also available for specific scientific and industrial applications: Downscaling of climate and weather data: Downscaling is a common implementation in meteorology that infers high-resolution outputs from low-resolution variables. is. Typical input data includes temperature, precipitation, and surface winds, which can vary in resolution. This foundational model can display both weather and climate data at up to 12x higher resolution and generate localized forecasts and climate projections. The fine-tuned downscaling model is published on Hugging Face’s IBM Granite page. Gravitational wave parameterization: Gravitational waves are ubiquitous in the atmosphere and can influence many atmospheric processes related to climate and weather, such as cloud formation and aircraft turbulence. Traditionally, existing numerical climate models have not captured gravity waves well, leading to uncertainty in how precisely they influence climate processes. This weather and climate infrastructure model will help scientists more accurately estimate the occurrence of gravitational waves, improve the accuracy of numerical weather and climate models, and reduce uncertainty when simulating future weather and climate phenomena. Helpful. This gravitational wave parameterization model is published on Hugging Face as part of the NASA-IBM Prithvi model family. “We are thrilled to be working with NASA on this mission,” said Karen St. Germain, director of NASA’s Earth Science Division, Science Mission Directorate. “Advancing NASA Earth science for the benefit of humanity means delivering actionable science in a way that helps people, organizations, and communities. This strategy must be implemented with urgency to respond to the rapid changes that are occurring. NASA’s foundational models provide weather, seasonal, and climate predictions that help inform decisions about how to prepare, respond, and mitigate. , to help create tools that people can use.” Juan Bernabe-Moreno, director of IBM Research Europe and IBM’s Climate & Sustainability lead for Accelerated Discovery, said: “This space is seeing the emergence of large-scale AI models that focus on fixed datasets and single use cases (primarily prediction). The weather and climate infrastructure models developed by IBM and NASA are designed to go beyond these constraints and be tuned for a variety of inputs and applications. For example, the model can be run on a global or local scale. This technical flexibility allows the model to understand weather phenomena such as hurricanes and atmospheric rivers, increase the resolution of climate models to infer potential future climate risks, and Ultimately, it is suitable to help understand impending extreme weather phenomena.” “We are thrilled to be partnering with the National Center for Computational Science at Oak Ridge National Laboratory,” said Arjun Shankar, director of the National Center for Computational Sciences at Oak Ridge National Laboratory. “As a premier research institution and computing facility, we are committed to supporting teams achieving research breakthroughs in various fields of science. In collaboration with IBM and NASA, we are working with the Prithvi Weather and Climate Foundation Model has supported the development of advanced computing and It was an important part of our goal to provide data.” IBM is already working with Environment and Climate Change Canada (ECCC) to test the model’s flexibility in new and different weather forecasting use cases. ECCC is using this model to explore very short-term precipitation forecasting using a technique called precipitation nowcasting that takes real-time radar data as input. The team is also testing a downscaling approach from global model forecasts at 15 km resolution to the km scale. This weather and climate model is part of a larger collaboration between IBM Research and NASA to explore Earth using AI techniques, and joins the Prithvi family of AI-based models. Last year, IBM and NASA made the Prithvi geospatial AI foundation model the largest open source geospatial AI model available on Hugging Face. This geospatial-based model is used by governments, businesses, and public agencies to investigate changes in disaster patterns, biodiversity, land use, and other geophysical processes. The foundation model and gravitational wave parameterization model are NASA-IBM Hugging Face From the page, the downscaling model is IBM Granite Hugging Face It can be accessed from the page. This press release is based on an excerpt of a press release issued by IBM Corporation on September 23, 2024 (local time). Original text: Here Please refer to IBM and ibm.com are trademarks or registered trademarks of International Business Machines Corporation in the United States and other countries. Other product and service names may be trademarks of IBM or other companies. For a current list of IBM trademarks, visit .