LTEOI-2025

Satellite Event Overview

Recent advances in satellite technology have enabled frequent, high-resolution, and global-scale monitoring of the Earth. This explosion in the availability of Earth Observation (EO) data presents unprecedented opportunities to support critical societal challenges. Artificial Intelligence, knowledge graphs, and semantic technologies are emerging as powerful enablers for bridging the gap between raw EO data and actionable insights.

LTEOI-2025 will walk participants through the full pipeline of building intelligent EO services—from ingesting raw satellite data to creating responsive, natural language-powered applications, using technologies and tools that are being applied to research projects like FAIR2ADAPT. The event is hands-on, enabling participants to learn by doing, while also offering a blueprint that can be replicated across other regions and domains.

Schedule

LTEOI-2025 is a half-day tutorial (3 hours), as seen in following schedule:

Time	Session
09:00 - 09:30	Introduction Welcome and introduction by the presenters. Presentation of tools and challenges to be demonstrated.
09:30 - 10:30	From Earth Observation Data to the Triple Store Part 1: From Unstructured EO Data to RDF with Python Part 2: Storing and Querying RDF Data with GraphDB
10:30 - 11:00	Coffee Break
11:00 - 11:40	Pulling down the SPARQL barrier for RDF access Part 3: Developing a Natural Language Interface by employing TerraQ-as-a-Service
11:40 - 12:00	Visualization of results Part 4: Visualizing Geospatial Results with Leaflet.js
12:00 - 12:20	Optimizing for realtime performance Part 5: Query execution optimization with JedAI-Spatial and GoST
12:20 - 12:30	Summary and discussion

Detailed Description of the Event

The tutorial will cover tools and methodologies to transform raw EO data into semi-structured RDF data which can be easily and efficiently queried by both novice and expert users. Although the tutorial will be grounded in a use case focused on France, the methodologies and tools presented are generalizable to EO scenarios across domains and geographies.

Part 1: From Unstructured EO Data to RDF with Python

Participants will begin by taking raw Earth Observation datasets (satellite images with accompanying metadata) and converting them into RDF format using a simple ontology. This transformation is performed using a custom semantic parsing library developed by our group, capable of extracting structured knowledge from unstructured EO data.
Tools: ToposKG.

Part 2: Storing and Querying RDF Data with GraphDB

The newly-formed RDF data, along with geospatial data concerning administrative divisions and natural features, is loaded into GraphDB. In addition to the built-in GeoSPARQL support, we present a plug-in for supporting stSPARQL. These extensions allow the representation and querying of geometries (points, polygons, etc.) using geospatial functions, making GraphDB ideal for storing Earth observation data with spatial dimensions.
Tools: GraphDB, stSPARQL extension.

Part 3: Natural Language Interface by employing TerraQ-as-a-Service

Now that the RDF knowledge graph is operational, participants will integrate TerraQ, a Text-to-SPARQL system which provides a natural language interface for RDF stores. It does so by translating natural language questions into semantically equivalent SPARQL queries. TerraQ is a Question-Answering engine that simplifies interaction with EO data, by removing the need for users to understand SPARQL. As a result, EO data is made more easily accessible to users, while maintaining accuracy and computational efficiency.
Tools: TerraQ

Part 4: Visualizing Geospatial Results with Leaflet.js

Until now, results are presented in a raw, CSV-like format. When dealing with satellite image data, it is of paramount importance to be able to see the images, and their visualizations on a map. We show how such functionality can be implemented using Leaflet.js, a lightweight and interactive JavaScript library for mapping.
Tools: Leaflet.js

Part 5: Query execution optimization with JedAI-Spatial and GoST

For the last part of our tutorial we showcase the complexity of geospatial calculations when using large polygons. This results in long response times, making our system frustrating to use. To deal with this we employ geospatial interlinking and query rewriting to dramatically speed-up query execution.
Tools: JedAI-Spatial, GoST

Note: Participants are encouraged to bring their own laptops to the event. This will enable them to follow along hands-on during the tutorial sessions and execute the code in real-time. Runnable code checkpoints will be provided a few days before the event to ensure that participants can easily set up their environments and follow the tutorial without any issues.

Organizers

Manolis Koubarakis

Manolis Koubarakis is a Professor and Director of Graduate Studies in the Dept. of Informatics and Telecommunications of the National and Kapodistrian University of Athens, where he directs the AI Team. He is also an affiliated researcher of the Archimedes Unit, a Fellow of EurAI since 2015 and Member of the EurAI Board since 2022. He has published more than 250 papers that have been widely cited (8973 citations and h-index 52 in Google Scholar) in the areas of Artificial Intelligence (especially Knowledge Representation and Constraint Programming), Databases (especially temporal and spatial databases), Semantic Web, Knowledge Graphs and Linked Data (especially geospatial and Earth observation data), and Deep Learning for Natural Language Processing. Manolis has taught at university level since 1995 and is a recognized AI researcher in spatial reasoning, geospatial databases, and knowledge graphs. He has given many tutorials and has led many laboratories for undegraduate and graduate courses.

Website Google Scholar

Sergios-Anestis Kefalidis

Sergios-Anestis Kefalidis is a Ph.D. student and Research Assistant in the AI Team of the National and Kapodistrian University of Athens. He holds a B.Sc. and a MSc. in Computer Science from the National and Kapodistrian University of Athens. He has actively participated in multiple European research projects (DA4DTE, GeoQA, STELAR). His research focuses on universal question-answering over knowledge graphs and applications for geospatial question-answering systems. His Ph.D. research is supported by the “NIK. D. XRYSOVERGI” scholarship of the Greek State Scholarship Foundation. Before his involvement with the AI Team, during his undergraduate studies, he participated in Google Summer of Code 2020, 2021 and 2022. During this time he was an active contributor of open-source software and has even served as a core developer and maintainer for the widely used Xfce desktop-environment for UNIX-like systems.

Website Google Scholar GitHub

Kostas Plas

Kostas Plas is a Ph.D. student and Research Assistant in the AI Team of the National and Kapodistrian University of Athens. He holds a BSc. and a MSc. in Computer Science from the National and Kapodistrian University of Athens. His research focuses on optimizing graph databases and geospatial triple stores as well as spatial reasoning in LLMs. His Ph.D. is supported by the scholarship with code “GD.402. ARCHI-YPPhD-0824” from the Archimedes Research Unit. He has gathered experience in geospatial knowledge graph and knowledge graph augmentation and manipulation through the usage of LLMs in his participation on the European projects DA4DTE and STELAR.

Website Google Scholar GitHub

Despina-Athanasia Pantazi

Despina-Athanasia Pantazi is a Research Associate in the Dept. of Informatics and Telecommunications at the National and Kapodistrian University of Athens, and a PhD candidate under the supervision of Professor Manolis Koubarakis. She holds a B.Sc. and M.Sc. from the Dept. of Informatics and Telecommunications of the National and Kapodistrian University of Athens. Her research interests focus on the area of Artificial Intelligence, especially in the fields of Natural Language Processing, Machine Learning, and Semantic Web. She contributed to the development of numerous Large Language Models (LLMs) such as the GreekLegalBERT model, and her PhD research focuses on the development of new improved methods for the efficient finetuning of state of the art LLMs. She also gained research experience relevant to the Earth Observation (EO) field as she participated in several European projects (Copernicus App Lab, ExtremeEarth, AI4Copernicus) over the last eight years.

Website Google Scholar GitHub

George Stamoulis

George Stamoulis is a research associate in the Dept. of Informatics and Telecommunications at the National and Kapodistrian University of Athens and a PhD candidate under the supervision of Prof. Koubarakis. He holds a B.Sc. and M.Sc. from the Department of Informatics and Telecommunications of the National and Kapodistrian University of Athens. He has worked in several European projects, building systems and applications over Earth Observation data, using Semantic Web technologies. His research interests focus in the areas of Semantic Web, Data Visualization and Integration and User Interfaces. He has contributed to various tutorials presenting Semantic Web technologies for Earth Observation data.

Website Google Scholar GitHub