The four-dimensional variational analysis (4DVAR) system, introduced in November 1997, uses an optimisation procedure whereby the initial condition is adjusted to obtain an optimal fit through all the observations in the assimilation interval and at the same time tries to stay as close as possible to the first guess.
[Image Source: https://www.rmets.org/sites/default/files/event_documents/3/Ftom%20Variational%20Data%20Assimilation%20to%20leading%20the%20European%20Centre%20for%20Medium%20range%20Weather%20Forecast%20-%20Florence%20Rabier.pdf]
The "four-dimensional" nature of 4DVAR reflects the fact that the observation set spans not only three-dimensional space but also a time domain.