Forecasting Numerical weather prediction (NWP) requires input of meteorological data, collected by satellites and
earth observation systems such as
automatic and crewed
weather stations, aircraft (including commercial flights), ships and
weather balloons.
Assimilation of this data is used to produce an initial state of a computer model of the atmosphere, from which an
atmospheric model is used to forecast the weather. These forecasts are typically: • medium-range forecasts, predicting the weather up to 15 days ahead • monthly forecasts, predicting the weather on a weekly basis 30 days ahead • seasonal forecasts up to 12 months ahead. Over the past three decades ECMWF's wide-ranging programme of research has played a major role in developing such assimilation and modelling systems. This improves the accuracy and reliability of
weather forecasting by about a day per decade, so that a seven-day forecast now (2015) is as accurate as a three-day forecast was four decades ago (1975).
Monthly and seasonal forecasts ECMWF's monthly and seasonal forecasts provide early predictions of events such as
heat waves, cold spells and droughts, as well as their impacts on sectors such as agriculture, energy and health. Since ECMWF runs a wave model, there are also predictions of coastal waves and storm surges in European waters which can be used to provide warnings.
Early warning of severe weather events Forecasts of severe weather events allow appropriate mitigating action to be taken and contingency plans to be put into place by the authorities and the public. The increased time gained by issuing accurate warnings can save lives, for instance by evacuating people from a
storm surge area. Authorities and businesses can plan to maintain services around threats such as high winds, floods or snow. In October 2012 the ECMWF model suggested seven days in advance that
Hurricane Sandy was likely to make landfall on the
East Coast of the United States. It also predicted the intensity and track of the
November 2012 nor'easter, which impacted the east coast a week after Sandy. ECMWF's Extreme Forecast Index (EFI) was developed as a tool to identify where the EPS (Ensemble Prediction System) forecast distribution differs substantially from that of the model climate. It contains information regarding variability of weather parameters, in location and time and can highlight an abnormality of a weather situation without having to define specific space- and time-dependent thresholds.
Satellite data ECMWF, through its partnerships with EUMETSAT, ESA, the EU and others, exploits satellite data for operational numerical weather prediction and operational seasonal forecasting with coupled atmosphere–ocean–land models. The increasing amount of satellite data and the development of more sophisticated ways of extracting information from that data have made a major contribution to improving the accuracy and utility of NWP forecasts.
Reanalysis ECMWF supports research on climate variability using an approach known as
reanalysis. This involves feeding weather observations collected over decades into a NWP system to recreate past atmospheric, sea- and land-surface conditions over specific time periods to obtain a clearer picture of how the climate has changed. Reanalysis provides a four-dimensional picture of the atmosphere and effectively allows monitoring of the variability and change of global climate, thereby contributing also to the understanding and attribution of climate change. To date, and with support from Europe's National Meteorological Services and the European Commission, ECMWF has conducted several major reanalyses of the global atmosphere: the first
ECMWF re-analysis (ERA-15) project generated reanalyses from December 1978 to February 1994; the
ERA-40 project generated reanalyses from September 1957 to August 2002. The ERA-Interim reanalysis covered the period from 1979 onwards. A reanalysis product (ERA5) with higher spatial resolution (31 km) was released by ECMWF in 2019 as part of the
Copernicus Climate Change Service.
Operational forecast model ECMWF's operational forecasts are produced from its "
Integrated Forecast System" (sometimes informally known in the United States as the "European model") which is run every twelve hours and forecasts out to ten days. It includes both a "deterministic forecast" mode and an
ensemble. The deterministic forecast is a single model run that is relatively high in resolution as well as in computational expense. The ensemble is relatively low (about half that of the deterministic) in resolution (and in computational expense), so less accurate. But it is run 51 times in parallel, from slightly different initial conditions to give a spread of likelihood over the range of the forecast. As of 2021, the ECMWF's weather model is generally considered to be the most accurate weather forecasting model. In 2025, the AIFS became ECMWF's first purely data-driven operational model. In comparison to the
Integrated Forecast System, the AIFS does not attempt to explicitly resolve the underlying physics. The AIFS uses a coarser 28 km grid compared to the IFS's 9 km grid. == Member and co-operating states ==