On April 16, 2012, Washington experienced heat no one expected. The thermometer climbed 18 degrees above normal. Sixty days earlier, Bucharest and Kyiv had weather unusual for them. The link between these events was found by an algorithm that scans pairs of cities worldwide, looking for recurring coincidences.
The idea is simple: if in the past an anomaly at point A was followed by an extreme at point B at least four times, that rule can be used for forecasting. No complex atmospheric models. Just a history of repetitions. For Washington, three such rules were found, and they predicted nearly a fifth of all hot days over a three-year test.
Importantly, the method does not peek into the future. The forecast relies only on data available at the time of prediction. It is like reading old signs, but the signs are backed by statistics. The probability of the results being a coincidence is less than five percent.
Echoes of distant events warning of a storm: that is how one might describe this approach. It requires no supercomputers or global models. Just patiently sift through pairs of cities and wait for the echo to repeat.
The heat in Washington on April 16, 2012 was predicted from temperatures in Bucharest and Kyiv 60 days before the event, even though these cities are more than 7000 km apart.