Earth Systems Intelligence: sensing the pulse of our planet.

Artificial Intelligence has quietly begun to shape how we perceive and protect our planet. And that’s a big deal. 

The Frontier Development Lab has spent over a decade at the intersection of AI, space science, and planetary health. Partnering with ESA’s Phi-Lab, our work has achieved many applied AI firsts - the kind of pragmatic achievements that move the window of possibility forward one discovery at a time. Among them: the first ever flood segmentation from space, and the first detection of methane plumes from space using ML (work that has directly informed the United Nations’ Methane Observatory.) This year we’re excited to extend this research with a project called StarCop 2.0: Atmospheric Anomaly detection from onboard - detailed in the 2025 results document for the community.

At the heart of this new capability is a new toolbox supercharging how data from multiple instruments and vantage points in space can be woven together. Instrument-to-Instrument translation (ITI), for example, enables different spacecraft to act in concert, forming a single hybrid observation network. This orchestration produces dynamic, three-dimensional views of phenomena like storm clouds in real time - work that has been extended in this year’s research, 3D Clouds for Climate Extremes.

Meanwhile, the transformer architectures that allow language systems such as Google Gemini, Claude and ChatGPT to converse and reason are now reshaping geospatial intelligence. These foundation models - vast neural networks trained on oceans of data - are learning to combine and interpret satellite imagery, radar, and climate signals and allowing us to converse to the data in natural language. Models like Google DeepMind’s AlphaEarth hint at a coming generation of systems capable of continuous planetary understanding.

But how can we trust such models when even language AIs hallucinate? The answer lies in systems like Shrug FM, the output of this year’s, Foundation Models in Extreme Environments challenge, which estimates its own confidence, mapping the regions where it believes its predictions are strong or weak. For the first time, our algorithms are learning to doubt themselves. This self-awareness of limits, nascent but profound, is a step toward the safe AI that planetary stewardship demands.

Taken together, hybrid observation, orbital machine learning, and self-aware geospatial models, these advances suggest a new paradigm for planetary intelligence. No single organisation has yet unified them; they remain distributed across research silos. But the frontier has shifted: it is now possible to integrate them into a cohesive framework. Imagine a decision intelligence system that links satellites, sensors, and simulation engines into one planetary operating system: a system that can predict floods, monitor carbon flux, or guide reforestation with scientific precision and temporal foresight.

In this emerging world, models can describe predictions in natural language. A scientist or a policymaker can ask a model, “Where are regions most vulnerable to flooding?” and receive not an inert map, but an intelligent explanation grounded in physics, hydrology, and probability. The model can construct a digital twin of the Earth’s near-future conditions, allowing humanity to explore “what-if” scenarios before catastrophe strikes. What once required weeks of expert computation can now be done in seconds.

However, prediction alone is not enough. The ultimate task is to connect predictability with purpose. AI is more than a tool to model the world; it must align with the levers that shape it - finance, policy, and collective action. The same intelligence that predicts a flood could inform an insurance bond, guide disaster relief funding, or redirect investment toward climate resilience. When knowledge flows into the architecture of capital, the invisible hand of the market can begin to heal rather than harm.

It is easy, in the noise of our age, to see AI through a pessimistic lens. We see a future where AI acts as an instrument of stewardship, a critical collaborator in the long project of sustaining life on a finite world.