About the challenge

The ThinkingEarth Hackathon at Big Data from Space (BiDS) 2025, invites AI and Earth observation enthusiasts to explore the power of Copernicus-scale foundation models. Organised by the Horizon Europe project ThinkingEarth, this challenge gives participants hands-on access to state-of-the-art resources including Earth Observation (EO) foundation models, graph-based weather forecasting tools, and vision-language models for satellite data.

Participants will work on one of three open-ended tracks:

  • EO Foundation Models: fine-tune and adapt large models for geospatial tasks

  • Weather Forecasting: evaluate and extend AI-based weather models

  • Vision-Language Models for EO: connect imagery and text to enhance EO data interpretation

Over a three-week remote development phase and a final onsite sprint at BiDS25, teams will prototype solutions, receive expert mentoring, and compete for prizes. The challenge is open to individuals and small teams with strong skills in deep learning and data science. No prior EO experience is required—just curiosity, creativity, and a drive to innovate.

A webinar was organised on September 3rd at 10:00 AM CEST to walk participants through the tracks, tools, and participation process. You can watch the recording of the webinar here: https://youtu.be/Bmtw8U-acfQ 

Get started

To help you get started we provide a GitHub page with the following material:

  • Access to Earth Observation foundation models (e.g., CopernicusFM, DOFA)

  • Pre-trained weather forecasting and vision-language models

  • Example EO datasets and downstream task descriptions (e.g., land cover classification, cyclone tracking, image captioning)

  • Starter notebooks with example code for each track

Access all the necessary resources here: https://thinkingearth-hackathon.devpost.com/resources

To participate, you’ll need to have coding experience in Python and deep learning libraries such as PyTorch and NumPy. You can run your solutions locally or use cloud-based environments like Colab or AWS.

Submission steps

  1. Create your project on Devpost using the platform's template
  2. Develop your solution remotely until September 28th
  3. Bring your laptop and finalise your project during the final coding sprint at BiDS'25 in Riga
  4. Submit your project on September 29th by 17:00 local time

Requirements

What to Build

Participants will develop technical solutions within one of the three open-ended tracks:

Track 1: EO Foundation Models
Build applications or tools using EO foundation models (e.g., CopernicusFM, DOFA). You may explore fine-tuning strategies, adapt models across geographies or modalities, identify limitations, or develop novel use cases related to tasks such as land cover classification or biomass estimation. A demo notebook will be provided to get started.

Track 2: Weather Forecasting Models
Experiment with state-of-the-art weather forecasting models. You may assess model capabilities, design evaluation frameworks, or apply models to downstream tasks such as cyclone tracking or extreme weather event prediction. Demo notebooks will introduce selected models, datasets, and practical considerations.

Track 3: EO Vision-Language Models (VLMs)
Use pre-trained VLMs to interact with EO data through natural language. Possible directions include image captioning, visual question answering, EO data retrieval, or text-based image search. Starter notebooks, datasets, and model access will be provided to support experimentation.

Projects will be evaluated on effectiveness, efficiency, novelty, clarity, and impact potential. Participants may work individually or in teams of up to three people.

What to Submit

Participants must submit the following material through the Devpost platform ("My projects" --> "Create project") by the end of the development phase:

  • Project description outlining the problem addressed, approach taken, and relevance to the selected track (follow the Devpost project submission template)

  • GitHub repository with source code used to develop the solution and documentation explaining how to run the code, key methods used, and any dependencies or requirements

  • A ppt presentation and a short video showcasing the solution, its functionality, innovation, and impact potential

Submissions will be evaluated by a panel of expert judges based on:

  1. Effectiveness

  2. Efficiency

  3. Novelty

  4. Clarity

  5. Impact potential

Only projects submitted by participants physically attending the Hackathon at the BiDS’25 Satellite Events in Riga will be evaluated and considered for prizes. Remote submissions are not eligible.

All submissions must respect the intellectual property and data protection guidelines outlined in the Hackathon terms. Participants will retain IP rights over their demonstrators; background IP remains with the original license holders.

Hackathon Sponsors

Prizes

€5,000 in prizes
First Place
€2,500 in cash
1 winner

Second Place
€1,500 in cash
1 winner

Third place
€1,000 in cash
1 winner

Devpost Achievements

Submitting to this hackathon could earn you:

Judges

Stefanie Lumnitz

Stefanie Lumnitz
ESA

Pieter Kempeneers

Pieter Kempeneers
JRC, European Commission

Dainis Jakovels

Dainis Jakovels
Institute for Environmental Solutions

Michael Schick

Michael Schick
EUMETSAT

Despina-Athanasia Pantazi

Despina-Athanasia Pantazi
National and Kapodistrian University of Athens

Judging Criteria

  • Effectiveness
  • Efficiency
  • Clarity
  • Impact potential
  • Novelty

Questions? Email the hackathon manager

Tell your friends

Hackathon sponsors

This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.