Frontier Development Lab (FDL)

What do we hope this challenge will show? A good score on old data is not the same as a forecast you can rely on. Conditions in space change all the time. For three months, every Benchlab model is checked against new solar wind observations as they arrive. We are looking at more than accuracy: are the results reproducible, and does the workflow keep running reliably, day after day? The results will show which approaches hold up under real operating conditions, and how forecast skill changes from one hour ahead to three days ahead. Live evaluation starts 19 November. May the best model win. Watch this short video from Dr Enrico Camporeale below and find out more at https://fdl.ai/benchlab #Benchlab #FDLAI #SolarWind #SpaceWeather #MachineLearning

A huge thank you to Scan Computers and the wonderful team: James Gorbold, Glyn Merga, Thomas B., Harry Bloxham, and Elliot Fishwick for making the work possible. Scan Computers have shared an update from the Lunarlab Technical Showcase, to read please see here: https://lnkd.in/edVmbAXy

What makes Benchlab different? Most benchmarks test models on the past. Benchlab tests them on the future. From 19 November, submitted workflows run live, forecasting ambient solar wind speed 72 hours ahead using only the information available at the time they run. Every six hours, for three months, each workflow fetches the public data it needs and makes its forecast in a controlled environment. Models are frozen before the run starts. No retuning, no hindsight: what is tested is exactly what was submitted. May the best model win. Watch this short video from Dr Enrico Camporeale below and find out more at https://fdl.ai/benchlab

Traditional benchmarking has hit a wall. Complex models can bend reality to fit by tuning parameters until they match historical data perfectly. Benchlab Live is a new kind of AI tournament that tests models against live, unseen data. The first iteration of Benchlab Live will centre around space weather, teams will build models that are evaluated live against future solar wind observations. Having models predict the unknown (and ‘impossible to know’) will help unlock true operational capability and gives us a unique insight into how artificial intelligence can help us rethink how we forecast space weather. Watch the short video from Dr Enrico Camporeale below and find out more on https://fdl.ai/benchlab

How do we make the most of the explosion AI models and merge the best bits together? This is the exact question the Mixture of Earths (MoEarths) team is tackling as part of the FDL Earth Systems Lab 2026. Inspired by a mixture of experts, the team has been building a context-conditioned fusion approach to combine both geospatial and general foundation models. This journey has revealed many open questions, including how best to assess the coverage and complementarity of embeddings, and how to merge these to make a difference downstream. If you are interested in what the frontier of AI looks like, join us August 7 for our Frontier Development Lab (FDL) Earth Systems Lab AI SOTA (State Of The Art) Showcase. Come meet the minds behind the work and explore how AI is becoming a trusted ally for planetary-scale challenges 💫 📅 Date: August 7 ⏰ Time: 14:00 - 15:00 (CEST) 📌 Sign up here: https://lnkd.in/eP_PcjAu A huge thank you to our partners European Space Agency - ESA , Scan Computers, NVIDIA, Google, Pasteur Labs & ISI and University of Oxford for continuing to support this mission. And our team for the work Kai-Hendrik Cohrs, Shruti N., Christian Lange, Carmen Oliver Huidobro, Georgios Athanasiou, Choukouriyah Arinloye and Nikolaos Ioannis Bountos. Ad astra per algorithmos. FDL Earth Systems Lab Team 🌍 #FDL2026 #ESL2026 #EarthSystemsLab #AI4Science #EarthObservation #AI4Good #AI #ESAPhilab

Every year at FDL (Frontier Development Lab (FDL)) I wonder "can we do it again"? 🤖 🌏 This year we set some really hard problems with crucial implications for #AI4good 🌳 1. Can we understand the living dynamics of our rainforests? 🧠 2. Can we merge models to create a deeper understanding? ⛈️ 3. And can we predict if a cyclone will weaken or intensify? The good news is that, yes, we can :) Join us this Friday at 2.00 CET and 1.00pm BST for an epic showcase of SOTA in applied AI. https://eslab.ai/live Thanks to our friends at European Space Agency - ESA and Scan Computers for supporting this research.

Destructive tropical cyclones transform from mild to monsters overnight 🌀 As weather forecasting improves using both traditional and AI approaches, accurate modeling of rapid intensifications remains elusive. The Tropical Cyclone Dynamics team is right at the eye of this storm as part of FDL Earth Systems Lab 2026. Can we use aIl to produce 6-to-24-hour wind field forecasts at 3 km resolution? This work bridges upper-level cloud data with surface winds and comes with some big questions, particularly how to enforce physical constraints when combining heterogeneous atmospheric layers, or how to validate models on these rare, rapid intensification events. Come meet the minds behind the work and explore how AI is becoming a trusted ally for planetary-scale challenges. 📅 Date: August 7 ⏰ Time: 14:00 - 15:00 (CEST) ✍ Sign up here: https://lnkd.in/eP_PcjAu A huge thank you to our partners European Space Agency - ESA, Scan Computers, NVIDIA, Google, Pasteur Labs & ISI and University of Oxford for continuing to support this mission. And our awesome team for the groundbreaking work Emiliano Díaz Salas-Porras, Anna Jungbluth, Cristina Radin, Simon Donike, Pritthijit Nath, William Jones, Ruben C. and Arthur Avenas. Ad astra per algorithmos FDL Earth Systems Lab Team 🌍 #FDL2026 #ESL2026 #EarthSystemsLab #AI4Science #EarthObservation #AI4Good #AI

Can AI quantify forest degradation and help protect the lungs of our planet? This is the question that the Living Forest team has been exploring as part of the FDL Earth Systems Lab 2026. Using the ESA Biomass mission’s P-band radar to probe deep under tree foliage and novel JEPA-driven 3D tomogram reconstruction, a new generation of tools for planetary stewardship are now possible. Come meet the minds behind the work and explore how AI is becoming a trusted ally for planetary-scale challenges 💫 📅 Date: August 7 ⏰ Time: 14:00 - 15:00 (CEST) 📌 Sign up here: https://lnkd.in/eP_PcjAu A huge thank you to our partners European Space Agency - ESA, Scan Computers, NVIDIA, Google, Pasteur Labs & ISI and University of Oxford for continuing to support this mission. And our awesome team for the groundbreaking work Joppe Massant 🔸, Felipe Begliomini, Grace Colverd, Ayush Prasad, Besma Guesmi, Palok Biswas and Gabriel da Rocha Bragion Ad astra per algorithmos. Earth Systems Lab Team 🌍 #FDL2026 #ESL2026 #EarthSystemsLab #AIforScience #EarthObervation #AIforGood #AI #ESAPhilab

Can AI quantify forest degradation and help protect the lungs of our planet? This is the question that the Living Forest team has been exploring as part of the FDL Earth Systems Lab 2026. Using the ESA Biomass mission’s P-band radar to probe deep under tree foliage and novel JEPA-driven 3D tomogram reconstruction, a new generation of tools for planetary stewardship are now possible. Come meet the minds behind the work and explore how AI is becoming a trusted ally for planetary-scale challenges 💫 📅 Date: August 7 ⏰ Time: 14:00 - 15:00 (CEST) 📌 Sign up here: https://lnkd.in/eP_PcjAu A huge thank you to our partners European Space Agency - ESA, Scan Computers, NVIDIA, Google, Pasteur Labs & ISI and University of Oxford for continuing to support this mission. And our awesome team for the groundbreaking work Joppe Massant 🔸, Felipe Begliomini, Grace Colverd, Ayush Prasad, Besma Guesmi, Palok Biswas and Gabriel da Rocha Bragion Ad astra per algorithmos. Earth Systems Lab Team 🌍 #FDL2026 #ESL2026 #EarthSystemsLab #AIforScience #EarthObervation #AIforGood #AI #ESAPhilab