Artificial Intelligence wizards at ESA Φ-lab exploit Sentinel data to the max

02 November 2023

As part of its joint initiative with ESA Φ-lab, UNICEF appointed two researchers to work on a project that forms part of the UN Secretary-General’s Digital Cooperation Roadmap.

The project maps current access to electricity and the Internet for schools around the world, in support of an ambitious UNICEF target to provide connectivity for every child by 2030.

In an effort to close the digital gap of countries with no Internet access and children without connectivity at home, UNICEF and the International Telecommunication Union created Giga, a global initiative whose goal is to understand the current connectivity status.

This is where remote sensing and ESA Φ-lab come in—with its expertise in enhancing the value of satellite-derived data through artificial intelligence.

The team at Φ-lab is employing Artificial Intelligence (AI) to analyse multi-modal Earth Observation and terrestrial datasets.

As part of its joint initiative with ESA Φ-lab, UNICEF appointed two researchers to work on a project that forms part of the UN Secretary-General’s Digital Cooperation Roadmap. The project maps current access to electricity and the Internet for schools around the world, in support of an ambitious UNICEF target to provide connectivity for every child by 2030.

This is where remote sensing and ESA Φ-lab come in—with its expertise in enhancing the value of satellite-derived data through artificial intelligence.

The team at Φ-lab is employing Artificial Intelligence (AI) to analyse multi-modal Earth Observation and terrestrial datasets.

Rochelle Schneider is passionate about Earth observation missions and an advocate of building opportunities to introduce the benefits of satellite technologies into public health. She holds a PhD degree in Geospatial Analytics, MSc in GIS, and MRes in Remote Sensing.

Before arriving at ESA Φ-lab as the AI Applications Lead, Rochelle was Senior Research Fellow in Geospatial Data Science at the London School of Hygiene and Tropical Medicine (LSHTM).

ESA: Why Giga and what is this collaboration all about?

Rochelle: I previously led a UNICEF and Φ-lab project using satellite-based products, epidemiological and socio-economic data as inputs in an AI model, to forecast dengue outbreaks across Brazil. This turned into a multi award-winning project, receiving recognition from UNESCO and UNICEF Innocenti. After we concluded this project, the Giga team approached us to continue this successful collaboration, joining forces in a new bold and transformative idea: using AI models and satellite products to ensure no child is left in the digital dark.

ESA: How did you identify the best ways to exploit satellite data to detect schools without internet connection?

Rochelle: Giga aims to connect every school in the world to the Internet, but first we need to identify where the schools are and if they have access to electricity. Therefore, I looked for a satellite product that could give us an initial status for the known school locations provided by Giga. Exploring nightlight satellite images enabled us to detect which schools were in the dark and determine how far they were from the first spot of light. Then we could support Giga in determining what would be the best investment for a power source (for example installing solar panels, buying a power generator, or extending the electricity network, etc).

ESA: What do you enjoy most about your role within this project?

Rochelle: I feel most proud of the possibility of supporting children from Space. But I enjoy many elements of this project: firstly, the opportunity to demonstrate the value of satellites in contributing to very basic educational needs (like access to electricity and the internet). Secondly, giving more visibility to which satellites could help, reducing the information barrier in some communities, who believe that satellites focus only on weather, GPS signals or missions to the Moon or Mars. Finally, the challenge of thinking outside the box to solve societal problems using AI and satellite products.

Born and bred in Denmark, Casper Fibaek leads a passionate team of researchers as Φ-lab’s technical head on the project. With an industrial PhD in Earth Observation and Artificial Intelligence, he gained valuable experience working on international development projects with the engineering consultancy NIRAS.

Before joining Φ-lab, Casper collaborated with Africa's financial sector. He developed spatial decision-support systems tailored for fin-techs and major financial players. His endeavours ranged from drafting population maps, to detailing transportation infrastructure and deriving socio-economic insights. Notably, he integrated these EO-derived datasets and analysis tools directly into Excel, making it more straightforward for clients to incorporate them into their decision-making processes.

ESA: How did your experience/studies help you to shape this project?

Casper: Over the years, I've had the privilege of leading projects that focus on transforming Earth Observations into actionable data, thanks to artificial intelligence. When working in regions where authoritative data are scarce, merging AI and EO becomes a game-changer, making data more democratic and accessible. Collaborating with UNICEF/Giga, organisations with a profound mission, is a natural extension of my expertise and area of interest.

ESA: Just how do Sentinel data play a part in this collaboration project?

Casper: Sentinel data and the insights derived from them form the backbone of our analysis. To give you an idea, we employ models that predict building footprints using Copernicus Sentinel-1 and -2 data. This allows us to significantly narrow the search space needed to find missing schools, streamlining our approach before we turn to higher-resolution data. Moreover, the Global Human Settlement Layer data, which are based on Sentinel data, are instrumental in our multi-criteria analysis, aiding us in gauging the completeness of existing school datasets.

ESA: What are the main challenges and rewards?

Casper: The sheer scale of Earth Observation data can sometimes be a double-edged sword. On one hand, we have a global dataset at our fingertips; on the other, we have very few quality labels available, and they are spatially clustered. Cleaning, organising, and designing methods and models to meaningfully conduct the multi-layered analysis is challenging. The reward is witnessing the impact of our work with UNICEF and Giga, and the joy of utilising the potential of these vast datasets for tangible challenges and seeing the transformative effects they bring.

A British national, Abi Riley is currently a third year PhD student at Imperial College London, joining the project for a six-month research internship, jointly hosted by UNICEF and ESA. Supported by the Medical Research Council Centre for Environment and Health, her doctoral research uses Bayesian spatiotemporal modelling applied to investigating the long-term mental health effects of air pollution exposure, in a cohort of London-based children from 37 schools. Previously, she completed a Master’s in Mathematics and aspires to use large spatial data and novel statistical methods in pressing environmental, health and humanitarian areas.

ESA: How did your specific studies bring you here?

Abi: In 2022 I met Rochelle at a conference, where I presented my PhD work on air pollution modelling, using both ground data and satellite-derived products. We got talking, and I was really interested in the work ESA is doing, especially as an opportunity to bridge the gap between Bayesian statistics and machine learning—and this internship was the perfect route into this! I am also really enjoying working with UNICEF and Giga, to see the interplay of the technical work we are doing and the subsequent real-world actions.

ESA: How do Statistics come into the picture?

Abi: My current part of the project focuses on the validation of present known school locations and the prediction of new school locations, more loosely, the probability of a school location actually being a school. This approach has meant that I can pull in methods of statistical probability modelling and a way to express the confidence in our results. We are also particularly interested in identifying areas where we expected to find a school, but didn’t, which could indicate areas with non-typical school buildings, areas with limited access to education, or places with inaccurate reporting.

I have also been able to use many spatial data approaches from my PhD work, bringing in and processing large geospatial datasets, including data tables of current known schools; attributes from OpenStreetMap; and EO datasets, such as population and built environment products, nightlights imagery, and Sentinel-2 high-resolution visible and IR images.

ESA: Where do you see yourself in the future?

Abi: I have always wanted to work on real-world research projects such as this one, and I especially want to take methods and ideas from this experience to work more on EO data and using interdisciplinary methods. I would love to either join ESA for a post-doctoral position or continue in academia, but staying connected to the type of work Φ-lab does and projects linked to organisations such as UNICEF and WHO. But with almost 2 years left on my PhD, I mainly hope to use what I’ve learnt here to improve and compare my statistical methods with novel approaches in geospatial ML and AI tools.