To account for errors, meteorologists will run an ensemble of weather models, inputting slightly different initial conditions into each model and then watching where the models agree and disagree. As they look deeper into the future, the models will branch further and further apart, until they bear as much resemblance to each other as to a model based on completely different inputs. This is the point of “error saturation,” when the models “lose memory” of the initial conditions, as Sheshadri wrote.

Source: Washington Post January 26, 2022 00:17 UTC