Climate change's connection to the frequency, severity, and longevity of future storms and natural disasters is a significant concern for governments and businesses worldwide. The latest models show that hurricane rainfall rates will increase by 2100. Determining how far climate has shifted can aid researchers in their search for solutions and remedies. However, Raghu Ganti, principal researcher at IBM Research, says, "Currently, half of all scientific findings come from archived sources."
Imagine a world in which researchers have more up-to-date data. The partnership between IBM and the National Aeronautics and Space Administration (NASA) aims to provide that data by leveraging artificial intelligence (AI) in research. Ganti shares how this partnership can advance research and inform government policy and business strategy going forward.
Supercharge Research With Up-to-Date Data
When relying on archived data for research, researchers face significant challenges in studying climate change. "The goal of this IBM-NASA partnership is to advance the scientific understanding of and response to Earth and climate-related issues, such as natural disasters and warming temperatures," explains Ganti. "IBM’s AI technology quickly speeds up the categorization and analysis of the incoming data by learning data representations in a self-supervised manner."
IBM will use foundation models to analyze the raw data and find their underlying structure. "Businesses around the world are exploring what's possible with foundation models, such as ChatGPT, a foundation model that applies generative AI to language tasks," says Ganti.
The partnership will organize and categorize text from more than 300,000 journal articles from the American Geophysical Union, American Meteorological Society, and other organizations to discover new insights about earth science. It also will use satellite data, particularly the Harmonized Landsat Sentinel-2 (HLS), a record of land-use changes captured by Earth-orbiting satellites, to analyze geospatial fields to build AI applications tailored to specific questions and tasks. These first two projects of the partnership mark the first time that foundation model technology will be applied to NASA's earth science data.
The melting of polar ice caps and deforestation are just two variables that could be explored using NASA's satellite data, according to Ganti. Once the foundation model applies the information it learns to one situation, it can then apply that knowledge to other concerns with minimal modification.
According to IBM, it is equivalent to when you learn how to drive a car and can then apply that knowledge to driving other vehicles. Through the NASA-IBM partnership, the model could eventually be used to track changes to habitats, inform natural resource management, or identify natural hazards.
Flexible and AI-Powered Data Processing on Cloud
IBM's foundation models will run on its cloud-based AI supercomputer, which has been online since May 2022, and is housed in IBM Cloud's Virtual Private Cloud (VPC) service. "While it’s currently just for use by the IBM Research community, the choices we’ve made with this design give us the flexibility to scale up and readily deploy similar infrastructure in the future," explains Ganti. "This cloud-native AI supercomputer is now our go-to environment for IBM Researchers creating our most advanced AI capabilities and is where we collaborate with partners to train models of many kinds."
Researchers are adapting the language model with the help of NASA to develop earth-science specific benchmarks that can be used to measure the model's performance, according to IBM. Once ready, IBM will employ its open-source multilingual question-and-answering system, known as PrimeQA. The end goal is to equip the model so it can answer questions about climate change by citing papers, linking to relevant context, and using the latest data. Additionally, a second model will streamline the development of AI applications that can be used to analyze satellite data at scale.
IBM Research's focus on open innovation aligns with NASA's Open-Source Science Initiative, in which software, data, knowledge (eg. algorithms, papers, documents) are shared earlier in the scientific process. The rapid advancement of AI can improve data analysis, and public-private partnerships like the one between NASA and IBM can democratize access to the technology, according to Ganti.