A decade of Firsts

FDL.ai and FDL Europe have been at the forefront of applied AI for science for a decade. 

Dear colleagues, this list is based on literature searches conducted at the time of publication. It is not our intention to over-claim or garner credit at the expense of other research groups. If there are any “firsts” here that have are highlighted erroneously, please do let us know and we will be all to happy to adjust accordingly. We know how hard it is to get original research into the world and are grateful to all of the brilliant researchers around the world of whom we have built on. We are standing on your shoulders.

Ad astra, Trillium

“FDL showed Google Cloud the value of AI for science.”
— Massy Mascaro, Distinguished Technical Director of Applied AI

Outstanding Achievements & Research Contributions -
 AI Accelerator/Incubator

Best Innovation - Space 

Planetary Defense 

  • Demonstrated for the first time meteorite hunting using an ML enabled drone 

  • First use of ML for automated asteroid shape modeling 

  • First application of Decision Intelligence for Planetary Defense.

First deployed system for fireball monitoring and long-comet detection (including the discovery of new meteor showers.)

Heliophysics and space weather

  • The first virtual instrument recreating a broken instrument (SDOML) 

  • the first live inference of synthetic EUV (EUV ARD)

  • the first ML prediction of thermospheric density (KARMAN)

  • The first demonstration of ML instrument autocorrection 

  • The first (known) demonstration of ML as a tool for discovery (geomagnetic ring currents) while predicting the Kp index. 

  • The first investigation of substructures within the solar wind using NLP. (Deep-WIND)

  • The first ML-enabled upscaling of historical data to modern instruments (HMI) 

  • The first prediction of GNSS scintiallion based on auroral structures 

The first AI twin of the Sun in 4D showing a fly-by of the solar poles. (SPi3ES)

Exoplanets, habitability and Astrobiology

  • The first method for extracting exoplanets from TESS data using ML. 

  • The largest synthetic dataset of exoplanet spectra, (INARA)

  • The first demonstration of ML retrieval of exoplanet atmospheres from spectral data 

  • Methods for determining ‘living’ (Gaian) planets using ML. 

  • The first AI that for robotic exploration that can detect complex biological signatures.

  • The first ML methods for determining ‘star spots’ on alien suns. (STARY-SPOTTER)

Advanced warning of disasters

  • The first demonstration of AI detection of the lighting ‘jump signature’ prior to tornadoes or heavy hail storms. 

  • The first statistically relevant detection of earthquake and tsunami precursors in ionospheric data.

The use of physics informed methods to detect storm-surge prior to a inundation event. (COASTAL-TWIN) 

The first ML able to detect an incoming geomagnetic storm 30 minutes in advance. (DAGGER-SHEATH)

Wildfires

  • The first demonstration of wildfire ignition detection from orbit using ML image stacking

  • The first AI to provide unsupervised post-fire burn analysis

The first use of AI to understand the causal drivers of pyro-CB fires and detection from space. (PYROCAST)

The first unsupervised ML burn-scar detection from orbit. (RAVAEN)

Flooding, landslides and hydrology 

  • The first surrogate model of ground contamination transport linked to climate projections. 

  • The first LLM knowledge aggregator for disaster scenarios (FLOODBRAIN)

The first automated flood mapping from Orbit (WORLDFLOODS)

The first landslide classifier in SAR data.

he use of AI to monitor low-flow streams and monitor hydrological systems. (OTTERWORKS).

The first virtual stream gauge and AI flood level projection (FLOODCAST)

Climate change and weather 

  • The first use of ML to determine the persistence of clouds based on measured aerosols (albedo). 

  • The first surrogate model of Earth systems (Digital Twin) to achieve parity with ECMWF’s ERA 5 using AI to predict global rainfall 72 hours in advance. 

  • The first ‘satellite image from the future’ using physics informed generative AI (Earth Intelligence Engine.)

The first AI cloud classifier and training data. (CUMULO)

Sustainability and the SDGs 

  • The first use of multi-model AI to demonstrate the potential post-flood and earthquake damage from orbit. (MULTINET)

  • The first AI to map informal settlements from satellite data. 

  • The first mult-image super-resolution (MISR) to upscale ESA’s Sentinel 2 to Planetscope data while retaining spectral integrity. 

  • The first SAR foundation model precursor 

  • The first use of ‘ADI’ (Artificial Design Intelligence) to CSP optimize power systems.

The first self-supervised image search tool for NASA’s Earth Science archive. (GIBS WORLDVIEW)

Space Medicine and astronaut health

  • The first use of transformers to generate medical data. 

  • The first tool for causal inference using mammalian and astronaut data (CRISP)

  • The first federated causal inference tool for protecting patient data (CRISP-FL)

Mission operations and space traffic management 

  • The first codification of human operator skill for ground station pass management

  • The first ML tool for space traffic collision avoidance and assessment, collision data messages (CDMs) and synthetic training-set (KESSLER). 

Lunar exploration 

  • The first images from the Moon’s permanently shadowed craters (HORUS) 

  • The first super-resolution of the lunar surface with uncertainty maps (ANUBIS)

  • The first ML change detection pipeline (NEPHTHYS)

The first metallic anomaly map of the Moon. (NEBU)