Geospatial technology is having a ChatGPT moment

For many years, geospatial technology has been operating quietly in the background of daily life. Satellites public and private generate terabytes of imagery each day, tracking – when processed and made intelligible by state-of-the-art algorithms– emissions, deforestation, wildfires, floods, biodiversity loss. We would be much worse off without them. 

I am able to speak from experience: my company was the first in the world capable of tracking emissions of methane – an invisible, odourless gas that is 84 times more potent than carbon dioxide – around the globe. Methane was known to have a devastating effect on the planet, but until then couldn’t be tracked at scale – which meant it couldn’t be attributed, which meant it couldn’t be reduced. We managed to crack this, and our work informed the Global Methane Pledge and helped shape the European Union’s 2024 methane regulation, the first law of its kind based on real-time satellite data. We were also among the pioneers in many other use cases for traders and insurers. 

Geospatial technology has been used mainly by powerful organisations like the United Nations and the major banks, or by traders looking for an edge. The reason is simple: developing reliable machine-learning models was both time-consuming and capital-intensive. Unlike large language models (LLMs) trained on text, which is easy to get hold of and understand, geospatial machine-learning models are trained on physical ground truth. To properly calibrate a model for French firefighters, we had to visit more than 2,000 houses. This kind of drudgery, in addition to the cost, has held back geospatial technology, which in theory lets us know about anything happening anywhere in near real-time.  

And this is why the latest developments in AI,  the emergence of remote-sensing foundation models (RSFMs), are so exciting. RSFMs can interpret many different types of input data, are highly adaptable, and can spot hidden patterns in satellite imagery without much human help. In layman’s terms, these models, enabled by AI, can deliver insights at a level of precision and at a speed that quite frankly are extraordinary. They can enable the infinite number of use cases that have been too time-consuming to pursue. What would once take months now takes hours. 

These foundation models go well beyond describing the world, which until now has been the focus of Earth observation companies like mine. They look forward and can make sound predictions as to what might happen. Imagine, for example, asking an AI not ‘How best to water my garden?’ but ‘When should I water my garden?’ And then imagine getting real-time updates, based on real-time, highly accurate satellite observation, on how much moisture the plants in your garden are holding and when you should water them.  

From the home to the corner office: imagine a bank or insurer asking, ‘How at risk is this building here?’ and then getting an instant, location-specific risk profile that includes – for example – climate exposure, supply chain disruption, political instability, all of it mapped, ranked and summarised. In the near future, we will see companies and indeed national governments develop the ability to create custom AI-powered geospatial tools – one that uses satellite data to monitor whatever it is that matters to them, be it forest degradation or mining activity or the spread of wildfires.  

There are general-use foundation models and there are specialised foundation models. The foundation models I’m discussing here are in the latter category: fast, with the capacity for precise fine-tuning, done via API, and scalable. Just as language models have made searching, writing and coding quicker, geospatial foundation models will change how we see the Earth, and increase our capacity to act wisely on what we see. Moreover, it will be far more easy, cheap and quick to develop specialised foundation models than it will to develop large ones. 

It goes without saying that efforts to curb climate change have gone nowhere and that our carbon footprint is indeed getting worse. Temperatures are climbing, crops are failing, forests are burning, and COP30 has roundly – and rightly – been viewed as a failure. But change is hard to deal with if you can’t see it. It’s a cliché, but it’s true: we can’t manage what we cannot measure. And what this means is that this ‘ChatGPT moment’ for geospatial is of profound global importance. Greenwashing could only thrive because emissions were estimated, not measured. Policies have been ineffective in part because the data that guided them was poor or non-existent, or because the policies served partisan ends. Satellite data plus AI gives us the power to hold polluters accountable, to verify claims, to track commitments with a degree of certainty that was impossible before. 

Of course, technology can be used for good or ill, so these powerful developments in AI come with responsibilities. We’ll need to make sure they’re used for the public good, that they’re not hoarded or used to deepen inequality. We’ll need to be transparent, and to work closely with others – partners in industry, in government, in science – to make sure that these tools have the right guardrails and are used in the right way and for the right reasons. 

We’re entering the golden age of geospatial intelligence. And it will shape everything: not just the climate, but energy markets, financial risk, insurance models, infrastructure, even geopolitics, while making the lives of ordinary people easier. The task for governments, in Europe as elsewhere, is to support the work of homegrown companies to bring these foundation models to their fullest potential. They may involve harmonising standards and resources across the continent and integrating these models into government processes. But something new and exciting has been set in motion. The job now is to drive it forward.