Presented two papers and co-chaired my first session at the 45th EARSeL Symposium in Athens, Greece

From 29 September to 2 October 2026 I was in Athens for the 45th EARSeL Symposium, hosted by Harokopio University. It was a full week: I presented two papers at the 9th Workshop on Global South, co-authored two more, and co-chaired a conference session for the first time.

My first session as co-chair

I co-chaired a session of the inaugural workshop of the new EARSeL Special Interest Group on Biodiversity and Ecosystem Services (EO-BES), chaired by my NEOGEO colleagues Felicia O. Akinyemi and Jan Haas. More than 80 people attended the oral sessions, and the day ended with an interactive Mentimeter session on the SIG’s future direction, possible collaborations and priorities for the next workshops. It was my first time chairing, and a great way to start.

EO-BES workshop session, 45th EARSeL Symposium

What do satellite embeddings bring to mapping deprived urban areas?

At the 9th Workshop on Global South, our DynEO4SLUMS team presented three studies on this question from different angles. I presented two of them and co-authored the third.

Evaluating satellite embeddings for neighborhood-scale mapping of deprived urban areas (presented)

We compared AlphaEarth and TESSERA embeddings with hand-crafted Sentinel-2 features and a CNN across 10 Global South cities. The embeddings matched or outperformed the spectral features and showed useful signal from as few as 500 labels, while the CNN needed far more labelled data to catch up. Zero-shot transfer to unseen cities works, but only to a point: city characteristics matter more than the choice of representation.

Presenting at the 9th Workshop on Global South

Informal-settlement mapping with foundation-model embeddings: a comparison of TESSERA and AlphaEarth (presented)

In a controlled comparison in Windhoek, both embedding products reached high discrimination with only 10% of the positive training samples. Their performance was nearly identical, and combining them raised the F1 score to 0.95, which makes TESSERA a competitive open alternative. Settlement boundaries and semi-formal areas remain the main challenge.

Do satellite embeddings improve population estimation in deprived urban areas? A survey-informed evaluation across three cities (co-author, presented by Stefanos Georganos)

Using survey-informed, bottom-up population estimates for Nairobi, Lagos and Buenos Aires, we found that global gridded products differ substantially, with WorldPop 51–72% lower than the survey-informed estimates. 2.5D building structure (height and volume) was the strongest predictor, cutting error by 27–52% compared with basic footprints. Embeddings added value mainly when combined with building features.

Stefanos Georganos presenting our population estimation study

The bigger picture

Embeddings are a strong, low-effort option for mapping deprived areas, especially when labels are scarce.
For population, explicit building structure matters most, and local survey data remain essential to get absolute numbers right.
Transferability across cities is the key frontier. That is where we go next: multi-city, multi-year analyses of how deprived urban areas change over space and time.

I was also a co-author on Event-based windstorm exposure in Sweden using hazard and building footprints, presented by my colleague Eleni Georgali.

Thanks

Thank you to my co-authors Stefanos Georganos, Jan Haas, Konstantinos Karagiorgos, Sabine Vanhuysse, Eleonore Wolff and Monika Kuffer; to our partner projects DEPRIMAP, ONEKANA and IDEAtlas for data and collaboration; and to BELSPO for supporting DynEO4SLUMS. Thanks also to the EARSeL organisers, the workshop co-hosts and everyone who joined the discussions.

My participation was partly funded by an Erasmus+ mobility scholarship from Karlstad University, which I’m grateful for.

The conference proceedings and full papers will follow once they are online.