ESA is developing a digital Earth observation assistant based on deep learning

by alex

A new generation of digital assistants will help deal more effectively with natural disasters

Satellites produce enormous amounts of valuable data every day, but their use can be complex and does not always provide maximum benefit to users. That's why the European Space Agency (ESA) is working with technology partners to develop artificial intelligence-based applications that are betting on making it easier to work with remote Earth observation data.

One of the initiatives was the I*STAR platform. It uses AI to monitor current events such as earthquakes or volcanic eruptions so satellite operators can automatically schedule the next data collection for clients.

Another tool developed with the support of InCubed is SaferPlaces AI. It creates flood maps for disaster response teams by combining local measurements with satellite data. SaferPlaces played an important role in assessing the damage from last year's floods in Emilia-Romagna (Italy).

The next step will be an application that «understands» ChatGPT-style queries to work with Earth observation data. This task requires the collection of various components, starting with the basic model underlying the assistant.

ESA's F-lab is working to develop baseline models for Earth observation applications, using data to provide information on important environmental topics such as methane leaks and extreme weather mitigation.

READ
2000 km on one tank with a consumption of 2.9 liters per 100 km and modern filling - from $16,600. The BYD Qin L DM-i super hybrid has already appeared in Chinese showrooms

One of the foundation's projects — the PhilEO model, which began operations in early 2023 and is currently in testing. It uses global Copernicus Sentinel-2 data and will be offered to the community for joint validation of the underlying model.

ESA projects also aim to create a digital assistant that will accept queries in natural language, process the data using Earth Observation Foundation models, and provide answers in text and/or graphical format. Past digital prototype assistant showed that such a system could perform multimodal tasks, searching for information among multiple archived data, such as Sentinel-1 and Sentinel-2, to compare information.

F-lab plans to expand natural language processing capabilities to analyze information from Earth observation text sources and interpret queries from both experts and ordinary users. This is the basis for the creation of a fully functional digital assistant.

The head of F-lab, Giuseppe Borghi, expressed his opinion on the significance of the project: «The concept of a digital assistant for Earth observation, capable of providing different types of information from different sources, sounds very promising. We there are the necessary fundamental components for its implementation. Thanks to the progress with PhilEO and the previous digital assistant, I am fully confident that new projects will bring revolutionary results in the near future».

You may also like

Leave a Comment