Each of the three cities of Munich, Hannover and Berlin is now home to an AI Future Lab – an International Future Lab for Artificial intelligence. These bring together both early career researchers and experienced scientists from all over the world to explore scientific questions surrounding the future use of artificial intelligence (AI). The labs are the outcome of a competition launched by the Federal Ministry of Education and Research (BMBF). Funded by the BMBF, the international teams started work in May 2020.

Research at the AI4EO lab in Munich is focused on AI in Earth observation. AI4EO stands for Artificial Intelligence for Earth Observation: Reasoning, Uncertainties, Ethics and Beyond. The nucleus of the AI4EO team comprises 13 ‘core scientists’ who are experts from various disciplines. They are now able to carry out research in Munich for three years, supported by 70 early career scientists who each come to Germany for six months under a fellowship programme. There are still fellowship places available – if you would like to find out more, further information is available on the AI4EO website.

How artificial intelligence can improve Earth observation

Some 800 satellites circle the globe day and night, ‘observing’ Earth (Source: UCS Satellite Database, December 2020). Recording data and images over long periods of time, they document changes in land surfaces, the oceans and the atmosphere.

Using satellite imagery, it is possible to create digital, three-dimensional maps of the Earth’s surface. This is where AI4EO comes in with artificial intelligence, because such maps involve very large quantities of data. Or to be more accurate, measurements – as Professor Xiaoxiang Zhu, Director of AI4EO, points out. “For us, every pixel of an image is a measurement,” she explains. “The volume of such measurements in Earth observation is growing incredibly fast and is now so large that we can only analyse them using AI techniques.”

But what can analysing satellite imagery with artificial intelligence achieve? “We want to use AI techniques to measure the global growth of urban and rural regions,” Zhu says. Images can be analysed using AI techniques such as deep learning to recognise things such as different types of buildings. Is a specific building in an image residential or commercial? To tell them apart, the researchers need to train the algorithms to detect and correctly classify buildings. They have to teach the machine to recognise an apartment building and label it as such. The resulting information can be used as the basis for creating digital 3D models of built-up areas. Professor Zhu provides an example of how such models could be used: “They can provide policymakers and urban planners with useful pointers for making cities sustainable for the future.” AI-based analysis of satellite data also has many other uses. Examples include earlier detection and containment of forest fires and identifying where nature conservation measures should best be targeted. It may also be possible to use the outcomes from AI analysis to better manage global nutrition.

For the months ahead, however, AI4EO still has a number of more fundamental questions to explore with regard to AI in connection with Earth observation. How reliable and how accurate is the information provided by AI models? What about data access and data protection?

The lab: high-power servers and open workspace

So what does the AI4EO lab look like inside – huge screens with people viewing massive images of the Earth’s surface? “Our screens are not actually that big,” says Dr. Daniela Espinoza-Molina, Science Manager at AI4EO. What does stand out is the computing power of the servers. These high-performance machines are located on a floor which is set out as an open office, with large spaces conducive to discussion and collaboration. The Science Manager had her first visit to the lab in mid-September and is now in charge of the equipment. “We already have the latest IT. A high-power server arrived in the middle of December.”

So what’s next? Espinoza-Molina now has to prepare for the two-day kick-off meeting, which is planned for May 2021. “It would be nice if we could do at least one day of that in-person,” she hopes. The agenda includes presentations on AI4EO’s research priorities, a panel discussion and a poster session for early career researchers.

“We hope that all the team members who were unable to travel this year because of the pandemic will be able to come along in 2021,” Espinoza-Molina says. She nevertheless sees 2020 as a successful year. “I’m able to say that we now have our team together for the lab,” she says with some pride.