The contemporary world of climate science is characterized by swift changes in environmental conditions. Although there's a surge in data availability - NASA's projection shows a whopping 250,000 terabytes of data from upcoming missions by 2024 - data analysis remains a complex challenge for researchers and scientists. IBM's initiative earlier in the year, in association with NASA via a Space Act Agreement, was geared towards addressing this issue by creating an AI foundation model tailored for geospatial data.

This step, in conjunction with Hugging Face, an acclaimed leader in open-source AI and a popular hub for transformer models, aims to foster democratized access to AI. The broader goal is to spur innovations in both Earth and climate science realms.

Sriram Raghavan, Vice President at IBM Research AI, emphasized the significance of open-source tech in crucial research areas, particularly climate change. "Our collaborative endeavor with NASA's Earth-satellite data, channelized through Hugging Face's open-source AI platform, has the potential to bring forth transformative solutions for our planet," he noted.

Hugging Face's Jeff Boudier shared similar sentiments. He highlighted that the consistent progression of AI hinges on collaboration and information dissemination. "Open-source AI, along with the release of datasets and models, is indispensable to ensure the benefits of AI permeate a wider audience."

Kevin Murphy, NASA's Chief Science Data Officer, articulated the potential of foundation models to revolutionize the analysis of observational data, offering a deeper comprehension of our Earth. "Open sourcing these models can enhance their influence manifold," he remarked.

The combined efforts of IBM and NASA have borne fruit. The model, honed over a year using Harmonized Landsat Sentinel-2 satellite data from the US, showcases a 15 percent efficiency enhancement over traditional techniques, and this is achieved using merely half the labeled data. This base model, subject to further fine-tuning, holds promise in diverse areas, be it tracking deforestation, predicting agricultural yields, or greenhouse gas monitoring. In fact, a collaborative endeavor with Clark University is underway to adapt this model for niche applications like time-series segmentation and research on similarities.

This development builds on IBM's prior announcement of their collaboration with NASA, aiming to expedite satellite image analysis and foster scientific revelations. Additionally, it dovetails with NASA's decade-long Open-Source Science Initiative, which envisages a scientific community that's more accessible and encourages collaboration. In line with this spirit, 2023 has been heralded as the Year of Open Science by NASA, the White House, and other federal institutions, underscoring the myriad advantages of open data and knowledge sharing.

IBM's current model benefits from the foundational technology of the company and is aligned with their broader vision of devising AI models that cater to varied tasks, transferring information seamlessly across different contexts. Earlier in June, IBM rolled out watsonx, an AI-data platform designed to amplify the AI impact on enterprises. A commercial variant of this geospatial model, integrated with IBM watsonx, is set to launch through the IBM Environmental Intelligence Suite (EIS) in the coming months.

Related Links
Hugging Face
Earth Observation News - Suppiliers, Technology and Application