ECMWF scientists constantly work to refine and develop all aspects the Integrated Forecasting System (IFS) to push the quality of our weather predictions to the limit. They do so in collaboration with experts across our Member and Co-operating States and beyond. A good example is the close collaboration of ECMWF scientists with the European Space Agency and other organisations on new wind observations from the ground-breaking Aeolus satellite. This work and its impact on our forecasts are described separately in this Report.

In 2019, ECMWF also developed quintic vertical interpolation to address temperature biases in forecasts for the stratosphere; a new version of weak-constraint 4D-Var to improve the handling of model bias in data assimilation; and continuous long-window data assimilation to make the best possible use of observations, all in readiness for the next IFS upgrade. ECMWF scientists prepared the IFS for the operational use of cloud radar and lidar satellite data, and they made progress in coupled ocean–atmosphere data assimilation.

The new ERA5 reanalysis began to be used to initialise re-forecasts, and ECMWF scientists picked up on work carried out in our Member States on a new vertical numerical scheme, which was adapted for use in the IFS. ECMWF also introduced modifications to its wave physics package to improve the prediction of ocean waves. The wave physics changes had previously been implemented by Météo-France, and ECMWF scientists adapted and optimised them for use in the IFS.

Quintic vertical interpolation

ECMWF has over the years repeatedly increased the horizontal resolution of its forecasts to today’s grid spacing of 9 km in high-resolution forecasts (HRES) and 18 km in ensemble forecasts (ENS). The resolution increases greatly improved forecast quality in most parts of the atmosphere but led to unphysical cooling in the lower to mid-stratosphere. Investigations showed that this was a result of numerical errors accumulating due to insufficient vertical resolution in the stratosphere.

Accurately representing the stratosphere is important because variability in the winter- and spring-time stratosphere can influence tropospheric weather patterns, and because accurate model information in the stratosphere aids the use of satellite data to obtain the best possible estimate of the state of the Earth system at the start of forecasts.

In 2019, ECMWF scientists worked on a method to address stratospheric temperature biases without introducing a computationally costly increase in vertical resolution. They showed that, in ECMWF’s Integrated Forecasting System (IFS), fifth-order (quintic) vertical interpolation leads to more physical model behaviour, reduced sensitivity to horizontal resolution, and better forecast skill in the lower to mid-stratosphere. Quintic vertical interpolation was therefore scheduled to be implemented in the IFS upgrade planned for 2020.

A warmer stratosphere in forecasts
A warmer stratosphere in forecastsDifferences in zonally (latitudinally) averaged temperature between quintic and cubic vertical interpolation forecasts for (a) a grid spacing of about 79 km and (b) a grid spacing of about 9 km. Mean values over 31 forecasts starting in July 2017 and valid at day 10 are shown. The plots show that quintic vertical interpolation warms the stratosphere more at high horizontal resolution.

Handling model bias in data assimilation

Data assimilation combines a short-range forecast (the first guess) with the most recent observations to estimate the state of the Earth system at the start of forecasts (the analysis). A high-quality analysis is hugely important for successful forecasts. For the atmosphere, ECMWF uses the 4D-Var data assimilation method. In the standard formulation, known as strong-constraint 4D-Var, the model is assumed to be perfect and any systematic model errors (biases) which gradually accumulate in the short-range forecasts are not taken into account.

When it turned out that there are significant model errors that develop during the data assimilation cycle for temperature in the stratosphere, ECMWF scientists worked on relaxing the assumption of a perfect model. They did that by introducing a forcing term η into the model to correct for the model bias which builds up in the model trajectory.

The resulting version of 4D-Var, known as weak-constraint 4D-Var, was found to reduce temperature biases in the analysis of the stratosphere by up to 50%. The improvement was achieved by correctly handling large-scale systematic errors as model deficiencies. In view of these results, the revised version of 4D-Var was due to be implemented in the next upgrade of the Integrated Forecasting System planned for 2020.

Weak-constraint 4D-Var
Weak-constraint 4D-Var In the case illustrated here, for a single parameter x (e.g. temperature), the forcing term η cools the trajectory at every time step to correct for the temperature warm bias in the model. In this formulation of 4D-Var, the error statistics of the model bias need to be calculated offline. They then enter into the calculations through which the data assimilation system determines the optimal combination of initial state and forcing term adjustments.

Assimilating cloud radar and lidar data

Successful weather forecasts start from accurate estimates of the current state of the Earth system. Such estimates are obtained by combining short-range forecasts with the latest Earth system observations in a process called data assimilation. Work carried out at ECMWF in 2019 demonstrated for the first time that assimilating cloud observations from satellite radar and lidar instruments into a global, operational forecasting system using a 4D-Var data assimilation system is feasible and improves weather forecasts.

The assimilation experiments used the full range of regularly assimilated observations at ECMWF to test the impact of adding the radar and lidar data on ten-day forecasts. They showed improvements in forecast quality across a range of key variables and altitudes. For example, as shown in the figure, between forecast days 4 to 8 the error in predictions of temperature at 1,000 hPa was reduced by about 1%. There are improvements for other variables too, but not yet at the same level of confidence.

Historical CloudSat radar reflectivity and CALIPSO lidar backscatter data were used to carry out the assimilation experiments. In the next few years, new satellite missions with cloud radar and lidar are planned, such as EarthCARE from the European Space Agency (ESA) and the Japan Aerospace Exploration Agency (JAXA).

Reduced errors
Reduced errors The charts show the impact from assimilating space-borne cloud radar and lidar observations on forecast errors (root-mean-square error) for a range of variables computed against ECMWF’s own analysis, up to 10 days ahead. The zero line represents errors without the assimilation of cloud radar and lidar observations, so that negative values indicate reduced errors. Bars indicate 95% confidence intervals. All scores are for the whole globe over the period of August to October 2007.

Coupled data assimilation

ECMWF’s Integrated Forecasting System (IFS) uses separate data assimilation systems for the atmosphere, the ocean, ocean waves, the land surface and sea ice. This may produce an internally inconsistent analysis if the data assimilation systems are independent from each other. Coupled data assimilation aims to ensure that the analysis of different Earth system components is consistent.

In 2019, research resulted in the implementation of ‘weakly coupled’ data assimilation for the atmosphere and sea-surface temperature in the IFS. In weakly coupled data assimilation, the observations of one Earth system component influence the analysis in other components with a certain delay. Previously, weakly coupled data assimilation had been implemented for the atmosphere and sea ice.

Weakly coupled data assimilation of the atmosphere with the sea-surface temperature of the ocean was implemented in the tropics and not the extratropics. This is because the ocean model used has greater effective resolution in the tropics than in the extratropics, where it is unable to resolve eddies. Experiments confirmed that the adoption of coupled atmosphere/sea-surface temperature/sea-ice data assimilation significantly improves the analysis of atmospheric variables such as temperature and humidity in the tropics and the polar regions.

Better analyses
Better analyses Normalised difference in root-mean-square deviation (RMSD) of forecasts from the experiment’s own analysis with and without weakly coupled data assimilation (EXP minus CTR) for forecasts of temperature at 1,000 hPa 24 hours ahead, for the period 9 June 2017 to 21 May 2018. Blue shades mean that the differences between forecasts and the analysis are smaller when weakly coupled data assimilation is used.

Use of ERA5 to initialise re-forecasts

Reanalysis, in other words the combination of observations with model information to reconstruct past weather and climate, plays an important role in numerical weather prediction. An example of this is the use of reanalysis to initialise re-forecasts. Re-forecasts are forecasts produced at the current time but starting from some point in the past. They are used to estimate a forecast model climate, which is needed to calibrate forecast products. Re-forecasts also serve to assess extended-range forecast skill and the evolution of forecast skill from year to year.

Like all forecasts, re-forecasts require a set of initial conditions, which reanalysis can readily supply. In 2019, ECMWF’s new ERA5 reanalysis replaced the older ERA-Interim to initialise re-forecasts. Tests showed that this resulted in better re-forecasts, better Extreme Forecast Index (EFI) skill scores, and improvements in the prediction of extended-range anomalies.

Better EFI skill scores
Better EFI skill scores The chart shows how using re-forecasts initialised using ERA5 instead of ERA-Interim improves EFI skill for global 2-metre temperature during the summer of 2018. Skill is here measured by a ROC area score (2 x ROC area – 1) so that ‘1’ corresponds to a perfect forecast and ‘0’ to ‘no skill’. The vertical bars show 95% confidence intervals.

New numerical scheme

ECMWF has worked with experts in its Member and Co-operating States to test a new numerical scheme for calculations over the vertical grid used in its Integrated Forecasting System (IFS). The results are very encouraging. The IFS employs a spectral method to solve the equations describing atmospheric dynamics in the horizontal and a finite element method to solve them in the vertical. A team led by Jozef Vivoda from the Slovak Hydrometeorological Institute (SHMI) and Petra Smolíková from the Czech Hydrometeorological Institute (CHMI) has developed a new vertical finite element (VFE) scheme.

In 2019, the new scheme was adapted to the IFS and was shown to meet three key requirements: the need to enhance flexibility in the chosen accuracy in the vertical; robustness with reduced precision, which is computationally more efficient; and the need to prepare the IFS for higher resolutions, including the ability to run a nonhydrostatic version of the model that is compatible with the current hydrostatic one.

As the figure illustrates, a 10-day forecast at a grid spacing of 5 km with the new nonhydrostatic scheme for Hurricane Dorian is remarkably similar to the equivalent hydrostatic forecast using the new scheme. The forecasts also agree very well with the observed hurricane track.

Forecasts using the new vertical finite element scheme
Forecasts using the new vertical finite element scheme The charts show maximum 10 m wind speed for a 10-day forecast of Hurricane Dorian, starting from 12 UTC on 31 August 2019, at a grid spacing of 5 km, for every 12 hours starting 12 hours into the forecast. The left-hand panel shows the forecast using the hydrostatic IFS with the new VFE and the right-hand panel the forecast using the NH-IFS with the new VFE. Both forecasts were run in single precision, using the same time step of 240 s and, where applicable, identical settings for model dynamics.

Ocean wave upgrade

Ocean waves are an important part of the Earth system: they depend on conditions in the atmosphere and the ocean, and in turn they influence those conditions, for example by slowing down winds. As part of ECMWF’s Earth system approach, the wave model component of the Integrated Forecasting System (IFS) is coupled to both the atmosphere and the ocean modelling subsystems.

In 2019, ECMWF introduced modifications to its wave physics package to improve the prediction of ocean waves. The wave physics package models how the wind generates waves, how different waves interact with each other, and how wave energy gradually dissipates.

The changes in the physics package included new parametrizations for wind input and deep-water dissipation of waves as previously implemented by Météo-France, based on work by Fabrice Ardhuin (Ifremer, France) and collaborators. ECMWF scientists adapted and optimised the changes so that they would run efficiently in the IFS. They ensured that the new parametrizations return a similar level of feedback from the modelled sea state to the atmosphere model, and they assessed the impact of ocean waves as modelled by the new package on the modelled ocean circulation.

The main impact of the changes was increased accuracy and realism of ocean wave parameters, such as significant wave height (roughly the average height of the highest one third of waves). The new formulation reduced the overprediction of long period swell energy and the small wave height underestimation in the storm tracks. Forecasts were generally improved up to 10 days ahead.

Differences between the new and the old wave physics
Differences between the new and the old wave physics The chart shows the annual mean difference in significant wave height between the new and old wave physics (new minus old). The data are from standalone wave model runs with atmospheric and ocean variables provided by the ERA5 reanalysis.