New forecast evaluation tool for OpenIFS

Gabriella Szépszó, Glenn Carver


The OpenIFS team at ECMWF has released a new tool for users of OpenIFS to test the performance of this easy-to-use version of the Centre’s Integrated Forecasting System (IFS). The tool is based on two case studies. Input data, the Metview visualisation programs with detailed guidance and output figures are available for download.

Until now, OpenIFS users could only check their model installation with a short acceptance test at the lowest model resolution (T21). Our aim with the new tool is to provide benchmark forecasts for two weather events, which users can repeat with their OpenIFS version on their computing facility. They can then compare their results to the ones produced using OpenIFS at ECMWF. We also plan to use the package internally.

The meteorological performance of the IFS is extensively assessed before each model upgrade. However, the IFS evaluation system is too complex to be made available with OpenIFS. This is why a similar suite was designed to explore new OpenIFS cycles before their release. The evaluation programs have for the first time been made available together with OpenIFS 40r1v2. The input data and the figures can be retrieved from the ECMWF download server and a comprehensive description with a user guide can be found on the OpenIFS home page (see the links at the end of this article).

Twenty-four hour precipitation experiments. The plots show 48-hour forecasts of 24-hour precipitation on 6 December 2015 in T255 resolution experiments initialised from ERA-Interim (top left) and ERA5 (top right) and in a T639 resolution experiment initialised from ERA-Interim (bottom left), and 24-hour precipitation in ERA5 (bottom right). More realistic precipitation patterns are produced at the higher resolution, while the difference between initialising from ERA-Interim and ERA5 at lower resolution is less pronounced.

We selected storm events with severe impacts over Europe and governed by large-scale dynamics: Xaver and Desmond. Storm Xaver hit the North Sea region and several adjacent countries on 5 December 2013. The cyclone developed on 4 December northeast of Newfoundland and was situated between converging northerly and southerly airstreams. The operational IFS forecast predicted the cyclone eight to nine days ahead and the very strong wind gusts three to four days in advance. The orographically driven precipitation of storm Desmond caused severe flooding across northern England, parts of Scotland and Ireland on 5 December 2015 and broke the United Kingdom's 24-hour rainfall record. Although the IFS forecast the high rainfall amounts well, it underestimated the peak values and overestimated the precipitation in the lee of hills.

OpenIFS was run several times for both cases to test the effect of forecast length, initial conditions and spatial resolution on forecast quality. We initialised the experiments from the ERA-Interim and ERA5 reanalyses one to five days before the high-impact events using three different resolutions: T255L91, T639L137 and T1279L137 (approximately 80, 32 and 16 km grid spacing with 91 and 137 vertical levels, respectively). The evaluation pages on the OpenIFS website guide the users step by step through post-processing the model outputs and plotting them with Metview.

We prepared Metview macros to analyse the initial conditions, the forecast results and the reference data. Our focus was on the evolution of mean sea level pressure, 2-metre temperature, 24-hour precipitation, 3-hourly maximum 10-metre wind gusts, 850 hPa temperature, 700 hPa relative humidity, 500 hPa geopotential, and 250 hPa and 100 hPa wind fields on 5 December. As a reference, we used ERA-Interim and ERA5, which are the fourth and fifth generations of ECMWF atmospheric reanalyses. ERA5 uses a newer model version with higher resolution than ERA-Interim and includes newly reprocessed observational data.

For an overview of the large number of figures produced by the Metview macros, an album can be prepared containing all plots for each selected variable and a given investigation aspect (e.g. to study the impact of resolution changes in the forecast) in a concise format. We provide two methods for users, Microsoft Word macros and html templates, to quickly generate such an album.

The package has already been tested by students at the Eötvös Loránd University in Budapest. We encourage other OpenIFS users to implement the evaluation package and to provide feedback on their experiences.

ECMWF download server:

OpenIFS Meteorological Evaluation: