Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain
the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in
Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles
and JavaScript.
The technique that can find a system’s state through data alone
Pinpointing the state of a complex system is tricky, especially when the underlying mathematical equations aren’t known. But a data-driven technique makes light work of it — and could even change the way that models are formulated.
Imagine you’re on a hill, flying a kite, and someone asks you to describe exactly what the kite is doing. “It’s over to the right, just above the treeline,” you might say, “and the wind is making it spin anticlockwise and move to the left.” This is an example of state estimation: the process of determining the true state of a complex system, typically from noisy, incomplete and often indirect observations. In a world replete with data, progress in data-driven modelling is rapid, but compatible state-estimation techniques are lagging behind. Writing in Nature, Course and Nair1 propose a state-estimation technique that enables efficient forecasts using data, without the details of any underlying model. In doing so, they offer a fresh take on the general approach to model discovery.