A translator who hears only the meaning, not the noise in the hall, can convey a conversation accurately. Climatologists have long sought such a translator for the ocean–atmosphere pair: their interaction resembles a dialogue where voices drown in the roar of extraneous sounds. In a new study, tools honed on brain signals have been applied to climate data for the first time to understand who controls whom.

The methods named rPDC and DPC originated in neuroscience. They act like an attentive listener: they do not merely note that two signals change together, but compute which one sounds first and how confidently. To do this, a model is built where the current value of a series depends on its own past and the past of the other series, while observation noise is accounted for separately. The classical Granger test, which ignores noise, was also included for comparison.

First, the methods were tested on synthetic data with a known direction of influence. Where the classical test gave a false signal in the opposite direction, the new tools clearly indicated the correct direction. Then came real climate indices: sea surface temperature in the tropical Pacific and rainfall over India. All three methods agreed: El Niño influences the monsoon significantly more strongly than the monsoon influences El Niño, and the connection lags by one to two years.

For the more complex South Atlantic Convergence Zone, where summer rains occur over Brazil and the ocean, the picture proved subtler. Year-by-year comparison of methods revealed a high correlation (0.85) for the direction from ocean to atmosphere, confirming that in some years the ocean indeed forces the atmosphere. And in 1983, the new method detected a negative influence of the atmosphere on the ocean, something the classical test cannot see in principle.

A translator who hears only the meaning, not the noise in the hall, has finally got down to work. Climatologists now have a tool to test hypotheses about the sign and strength of interactions in other regions, for example, the influence of the ocean on droughts or monsoons. What remains is to teach it to automatically select the correct model order and to check the discovered connections against surrogate data, to be sure that this is not an artifact but a real conversation.