Lightning Talks Schedule

Thursday 02 October

16:00 - 17:30

CEOS Analysis Ready Data - setting sails to new shores

Peter Strobl (EC-JRC)

CEOS-ARD is a well established series of specifications allowing space agencies to benchmark their products against a set of requirements intended to enable easy uptake and best possible interoperability. In a recent effort CEOS has embarked on revamping the CEOS-ARD development by bringing it to Github, modularising it, linking it to STAC and other community efforts to maximise transparency and user engagement. The talk will highlight the new strategy and how to get involved.

Find me purple trees in Mexico city. Could geospatial (gen)AI help our generation see the world through fresh A-Eyes?

Lea Kierbel (Airbus)

Have you ever imagined how AI can help you find in a very easy way even the most challenging objects in an image?
(gen)AI can do just that and more!

Bringing science to market: a path for researchers who don’t see themselves as entrepreneurs

Nikolai Adamovitch (CEO of Commercialization Reactor)

Many researchers hesitate to step into entrepreneurship, yet their discoveries have the potential to shape industries. Commercialization Reactor has spent 15 years helping scientists turn ideas into startups without leaving the lab. This talk will share how early-stage scientific projects can be transformed into deep tech ventures, using proven methodology, international experience, and the support of ESA BIC Latvia.

A framework for global highway network change detection applied to Landsat data

Johannes Uhl (European Commission, Joint Research Centre)

Multi-temporal geospatial data measuring the evolution of transportation networks are scarce, impeding our knowledge on the dynamics of highway networks at global scale. We present a framework that integrates contemporary road network data with road presence probabilities extracted from historical Landsat data, enabling the measurement of highway network growth from 1990 onwards, providing important baseline data for multi-temporal accessibility studies.

ACCURATE MAPPING OF MEKONG DELTA’S MANGROVE DISTRIBUTIONS THROUGH LARGE-SCALE EXTENT LABEL VALIDATION

Quan Le (University College Dublin)

In this study, we present an expert-validated pipeline for labelling mangroves of Vietnam. We annotated multiple layers of satellite imageries to generate Label V1, and systematically refined it to Label V2. We trained and evaluated LightGBM models on the two versions of labels using multispectral features as input, demonstrating a +4.6 % gain in F1-accuracy. Our best map achieved 99.7 % F1-accuracy on an independent set of reference points provided by Global Mangrove Watch.

Atlas of Human Settlements

Georgios Ouzounis (Vice President of AI Research)

The Atlas of Human Settlements (AHS) is a built-up base-map of global extent, delivered at 10m spatial resolution. It is updated monthly with historical records covering the past decade. The AHS layers are instrumental for understanding human activity and detecting change in a wide-area monitoring context, free of spatial constraints and/or prior knowledge. This talk will highlight some of its key features and present indicative commercial use-cases.

Processing and Data Access of the Global Flood Monitoring Service

Tobias Raiger-Stachl (EODC Earth Observation Data Centre GmbH)

The Global Flood Monitoring (GFM) service, part of Copernicus Emergency Management Service (CEMS), uses Sentinel-1 SAR data and a three-algorithm ensemble to deliver rapid flood maps and impact indicators. Hosted on EODC cloud infrastructure, it provides results in under 5 hours - often within 90 minutes. Data is accessible via GloFAS, EFAS, APIs, a web portal, and a STAC-compliant catalog. This talk highlights GFM’s operations and future scaling plans.

Unlocking the power of Big Data in Earth Observation

veronica mercuri (EO59)

This talk presents SARPROZ as a scalable solution for Earth Observation, used in the field of big data from space and specialized in radar-based analysis. SARPROZ processes large volumes of satellite radar data to measure ground displacements with high precision, supporting both small-scale investigations and regional monitoring.

GeoAI.js: Bringing Earth Observation AI to the Browser with WebGPU

Shoaib Burq (Decision Labs)

Earth Observation is gaining traction, but tools for developers to run GeoAI remain limited. GeoAI.js is an open-source JavaScript library that enables AI tasks on satellite imagery directly in the browser, powered by WebGPU. This lightning talk introduces how developers can bring interactive geospatial AI into modern web apps.

How Agentic AI Works: A Primer for Earth Observation

Shoaib Burq (Decision Labs)

This Lightning Talk teaches the fundamentals of agentic AI and how it can transform Earth Observation workflows. We’ll break down how agents plan, chain models, and deliver interpretable insights. The aim is to give the audience a clear understanding of agentic flows, practical EO examples, and ideas to apply in their own projects.

What would you do if you could have a simple into your data? Without installing anything!

Lazaro Alonso (Max Planck Institute for Biogeochemistry)

In this talk we introduce Browzarr, an open-source framework designed to democratize multidimensional data exploration for domain experts. By combining GPU-accelerated rendering with WebGPU compute pipelines. Built on the Zarr format's chunked data model, Browzarr streams only necessary data when constructing visuals, enabling seamless exploration and spatial computing of datasets exceeding 100s of millions of data points.

High resolution mid wave thermal data - a new modality for extending foundation models

James O'Connor (Satellite Vu)

SatVu, recently joining the Copernicus CCM programme, will fly a constellation of high resolution thermal imaging satellites capturing data at up to 3.5m GSD. EO Foundation models focus more on the optical and SAR, owing to the success of the Copernicus and Landsat programmes. I will discuss what the future looks like for new data sources and the relationship with calibration and integration into big and massive data initiatives, where data volumes of the new sources are relatively smaller.
Back to home