Future forest, crop, and climate change monitoring AI will be built on the open-source approach.
A quarter million gigabytes of data are expected to be produced by NASA's Earth scientific missions in 2024 alone, according to their predictions. IBM, HuggingFace, and NASA worked together to create an open-source geospatial foundation model that will serve as the basis for a new class of climate and Earth science AIs that can track deforestation, forecast crop yields, and tally greenhouse gas emissions. This model will help climate scientists and the research community efficiently sift through these mountains of raw satellite data.
IBM used a year's worth of NASA's Harmonized Landsat Sentinel-2 satellite data (HLS) as the basis for this research, with its recently launched Watsonx.ai as the foundational model. Sentinel-2 satellites, which were developed by the European Space Agency (ESA) to collect data across land and coastal regions in 13 spectral bands, are responsible for gathering that information.
The model is being hosted by HuggingFace on its open-source AI platform. The team was able to boost the model's performance by 15% compared to the current state of the art by using half as much data by fine-tuning it using "labeled data for flood and burn scar mapping," according to IBM.
Sriram Raghavan, vice president of IBM Research AI, stated in a press release that "the essential role of open-source technologies to accelerate critical areas of discovery such as climate change has never been clearer." "By combining NASA's Earth satellite data archive with IBM's foundation model efforts to build flexible, reusable AI systems, and making it available on the top open-source AI platform, Hugging Face, we can leverage the power of collaboration to implement faster and more significant solutions that will improve our planet."