To make a forecast we need to know the current state of the atmosphere and the Earth's surface (land and oceans). The weather forecasts produced at ECMWF use data assimilation to estimate initial conditions for the forecast model from meteorological observations.
ECMWF is a world leader in data assimilation research and development. The quality of our forecasts depends on how well we use information received in real-time from the global observing system, which consists of numerous satellite instruments, weather stations, ships, buoys, and other components.
The purpose of data assimilation is to determine a best possible atmospheric state and its uncertainties using observations and short range forecasts. Data assimilation is typically a sequential time-stepping procedure, in which a previous model forecast is compared with newly received observations, the model state is then updated to reflect the observations, a new forecast is initiated, and so on. The update step in this process is usually referred to as the analysis; the short model forecast used to produce the analysis is called the background.
A gentle introduction to data assimilation principles and more specific information about the operational ECMWF data assimilation system is provided in the dedicated 'An Introduction to Data Assimilation' e-learning module on our Learning Platform (login required).
We also use data assimilation to monitor climate change based on past observations – this is called reanalysis.
While the ECMWF analyses are among the most accurate in the world, they are affected by uncertainties arising from the inevitable uncertainties of the observations and the model. To quantify these uncertainties ECMWF runs an ensemble of 51 lower resolution 4D-Var assimilation systems (Ensemble of Data Assimilations, EDA) where observations and model are perturbed. The EDA provides a sample of the analysis uncertainties and a starting point from which an ensemble of forecasts can be initialised.
The atmosphere is chaotic, meaning that even small differences in its initial state can lead to very different weather patterns occurring several days later – this is sometimes referred to as the butterfly effect. To account for the chaotic nature of the atmosphere and the associated uncertainty in prediction, we run an ensemble of 51 forecasts simultaneously; the forecast using the best possible initial state plus 50 other forecasts where the initial conditions and the model are perturbed. Our ensembles provide a probabilistic forecast which is an estimate of how predictable a particular weather situation is.
Land data assimilation
The ECMWF data assimilation system also includes a specific Land Data Assimilation System to initialise the land surface model variables.
Ocean data assimilation
The ocean and sea ice are now two important components in the ECMWF's Earth System model. To determine the best available ocean and sea ice states, observations from both satellite instruments and in-situ measurements (Argo floats, moorings, CTDs ...) are assimilated into a coupled ocean and sea ice model via a specific Ocean Data Assimilation System.