Here's why IBM and NASA have teamed up to study the effects of climate change using AI

The AI will sift through terabytes of satellite data to determine how climate change and natural disasters affect agricultural output and wildlife habitats. (Photo via Pixabay/Pexels)

NASA and IBM have teamed up to push a novel use of AI technology on Earth observation data from Nasa satellites.

The two groups have announced their intention to create many programmes to mine Earth observation data for information regarding climate change. For example, one project will employ NASA's Harmonized Landsat Sentinel-2 dataset, a record of land cover and land use changes acquired by Earth-orbiting satellites, to train an IBM geospatial intelligence foundation model. The AI will sift through terabytes of satellite data to determine how climate change and natural disasters affect agricultural output and wildlife habitats.

In the meantime, another initiative will likely involve constructing a sizeable language model from Earth science literature. IBM has built a natural language processing model based on approximately 300,000 Earth science journal articles to facilitate the organization of the literature and the discovery of new knowledge. In addition, the researchers have announced that both models would be freely available to everyone in the scientific community.

Foundation models are classes of AI models that can be used for various tasks, are trained on many unlabelled data, and can transfer knowledge from one context to another. Over the past five years, these models have quickly expanded the natural language processing (NLP) technology field. As a result, IBM is at the forefront of using foundation models for applications outside of language.

Earth observation data are being gathered at unprecedented rates and volumes, allowing scientists to study and monitor our world. However, novel and creative ways are needed to extract knowledge from these enormous data resources. This effort aims to make it simpler for academics to examine and extrapolate information from these vast datasets. In addition, the ability to identify and analyze these data more quickly because of IBM's foundation model technology could help increase science's comprehension of Earth and its reaction to climate-related problems.

Alongside, to get knowledge from Earth's observational data, IBM and NASA intend to create several innovative technologies. In one project, NASA's Harmonized Landsat Sentinel-2 (HLS) dataset, a record of land cover and land use changes acquired by Earth-orbiting satellites, will be used to train an IBM geospatial intelligence foundation model. This foundation model technology will let researchers analyze our planet's environmental processes by analyzing petabytes of satellite data.

In the past 10 years, artificial intelligence (AI) has made enormous strides, and it is hoped that its use in climate science will help increase the precision of upcoming climate estimates. Let's explore some of the existing approaches.

In 2020, a team of researchers from the United States demonstrated that the dynamics of the Earth system might be accurately modelled using time-series models and computer vision.

In 2021, experts from the Indian Institute of Tropical Meteorology in Pune, IIT Delhi, and others predicted that advancements in this area will allow artificial intelligence to model the physics of clouds and rainfall processes and lessen uncertainty in the current systems.

According to a group of researchers from Google and Microsoft, AI has also been recommended for climate change mitigation applications in addition to enhancing the representation of natural systems in climate models by leveraging the currently accessible high-quality data. The technologies of carbon capture, building information systems, better transportation systems, and effective waste management, to name a few, are other fields where AI is leading the way.

However, according to New York University research, there are limits to the current deep learning models, such as their inability to distinguish between causation and correlation. In addition, Moore's law is anticipated to end around 2025 as it runs into fundamental physical constraints like quantum tunnelling. Alternate hardware developments are becoming necessary due to the growing need for deep learning and other software paradigms.

A searchable corpus of Earth science literature is anticipated as another result of this IBM and NASA partnership. To organize the literature and make it simpler to identify new knowledge, IBM created an NLP model trained on nearly 300,000 Earth science journal articles. The fully trained model employs PrimeQA, IBM's open-source, multilingual question-answering system, and is one of the largest AI workloads introduced on Red Hat's OpenShift infrastructure to date. In addition, NASA's scientific data management and stewardship operations may use the Earth science language model.

Furthermore, using MERRA-2, IBM and NASA will also analyse and interpret Earth's atmospheric data to better understand climate change. NASA's Open-Source Science Initiative, which includes this partnership, is an ongoing effort to create a welcoming, open, and cooperative scientific community over the next decade.