Why IBM and NASA are tackling the world’s problems with geospatial AI

As technology has evolved from kites to airplanes and eventually satellites, the fundamental goal has remained the same: collecting geospatial data to understand features of the Earth for disaster assessment, environmental monitoring, and more.

As the world focuses on escalating climate change and environmental challenges, the merging of technology and data insights offers ways to solve it.

An artificial intelligence (AI)-powered collaboration between two industry giants – IBM and NASA – is set to redefine our ability to understand and respond to our planet’s dynamics and could reshape disaster management, environmental monitoring and climate change adaptation.

When and where is geodata used?

Geospatial data plays a central role in disaster management, which includes the phases of preparation, response and recovery. In the event of events such as earthquakes, floods and forest fires, real-time geospatial data facilitate damage assessment, identification of affected regions and efficient planning of relief measures.

In environmental monitoring, geospatial data serves as an indicator of change, tracking deforestation, urban growth and climate-related changes.

This data enables policymakers to formulate sustainable strategies, protect delicate ecosystems, and manage resources effectively.

To combat climate change, geospatial data is used to monitor emissions, temperature variations and sea level rise. This information feeds into the development of strategies aimed at mitigating and adapting to these impacts.

In times of crisis, geospatial data plays a crucial role in humanitarian response, helping to map affected regions, assess the extent of damage and coordinate relief efforts.

Use AI to analyze geospatial data

Although geospatial data plays a central role in tasks such as disaster management, environmental monitoring, and climate monitoring, the complex nature of geospatial imagery poses significant difficulties for manual interpretation.

The proliferation of satellites and drones has led to an increase in geospatial data, making manual analysis ineffective, time-consuming and impractical in terms of scalability.

This situation is exacerbated by the lack of competent professionals to carry out these analyses, leading to delays.

In addition, human analysts may be faced with limited capacities and subjective perspectives, leading to inaccuracies and different results.

These analysts can also struggle to fully understand context, affecting the precision of their decisions.

Meanwhile, AI has gained the remarkable ability to rapidly process massive amounts of image data at scale.

Thanks to this ability, AI is able to analyze data streams in real time, which is particularly important in scenarios that require quick reactions, such as. B. in disaster management, is of crucial importance.

AI’s ability to recognize complex patterns helps mitigate the inherent subjectivity of human interpretation, which can ensure consistent and accurate results.

By understanding the complex relationships in geospatial data, the AI ​​can make better decisions.

In addition, the potential of AI to reduce the dependency on experts contributes to the democratization of geospatial analysis and enables non-experts to carry out sophisticated analyzes in this field.

The challenge of AI for geospatial analysis

Although AI holds great promise for geographic applications, its effectiveness is limited by the scarcity and high costs associated with obtaining high-quality geospatial data.

In addition, training models on large-scale, high-resolution geospatial data requires significant computational resources.

This is a particular challenge considering that the NASA 250,000 terabytes of data by 2024 from new missions for scientists and researchers.

Training AI models on such large datasets comes with high costs and environmental impacts, but the benefits may outweigh the costs.

What is a Foundation Model in AI?

A viable approach to addressing the above challenges is to create a foundation model for geospatial data.

A foundation model in AI is a pre-trained model that is trained on a large dataset using self-supervised learning to learn common patterns and characteristics from the data. This general purpose model serves as a basis for the development of more specialized and refined models.

When creating a specialized AI model for a specific task or domain, the base model is refined or fine-tuned with a smaller, task-specific dataset. This process allows the model to take the knowledge gained during pre-training and refine it for a specific task.

Using a base model speeds up the development process, minimizes the data and costs required for specialized AI training, and boosts the model’s performance through its existing knowledge.

This approach has found acceptance in various AI applications and enables the creation of powerful and effective models with less time and resources required for training.

IBM’s Geospatial Foundational Model

IBM recently partnered with NASA Foundation Model based on geospatial data developed.

The main goals are to reduce reliance on large geospatial data, reduce training costs, and reduce the environmental impact of training AI models.

The model was based on harmonized Landsat Sentinel-2 satellite data (HLS)covering the entire American continent for a year, and underwent an intensive training process and further fine-tuning with tagged data for tasks such as flood and burn scar mapping.

Through this training, the model has shown a remarkable 15% improvement over current methods, achieved with only half the amount of tagged data normally required.

With further refinement, this fundamental model can be used for various tasks such as monitoring deforestation, predicting crop yields, and detecting greenhouse gases.

To encourage wider access and application of AI, the model is over hugging face, a renowned open-source AI model library. This democratization should stimulate new innovations in climate and geosciences.

In July, IBM introduced watsonx proposes a state-of-the-art AI and data platform designed to make it easier for companies to apply advanced AI with trusted data at scale and at an accelerated pace.

As an extension of this effort, a business-oriented version of the geomodel integrated with IBM watsonx will be available in the coming months IBM Environmental Intelligence Suite (EIS) be accessible.

The conclusion

IBM’s collaboration with NASA has resulted in a foundational AI model for geospatial data that addresses challenges in disaster management, environmental monitoring and urban planning.

This AI solution offers improved accuracy and consistency, and overcomes the complexities associated with manually analyzing geospatial data.

Despite the potential of AI, obstacles such as data scarcity and high costs remain. The IBM model trained on Landsat Sentinel-2 data has shown significant improvements over existing methods with only half the data labeled.

This innovation, accessible via Hugging Face, democratizes geospatial insight and promises new advances in climate and earth science applications.