ECMWF’s seasonal forecasts are based on our medium-range forecasts but extend to seven to thirteen months. Stephanie Johnson has a key role in developing and optimising the seasonal forecasting system.
Inspired by her fascination with astronomy as a child, Stephanie first studied physics and subsequently did a PhD in astrophysics at Harvard University in the US.
Stephanie’s PhD was about star formation in galaxies. This image from a 2010 article by Stephanie et al. in The Astrophysical Journal shows the evolution of stellar surface density and gas surface density of spiral-induced star formation in the outer disks of galaxies. ‘Gyr’ stands for ‘gigayear’, which is one billion years. © AAS. Reproduced with permission.
In 2010, she moved into meteorology for her post-doc work at the University of Reading, UK.
“My work in astronomy was similar to weather prediction in that we represent dynamical equations in a numerical way to create a simulation, although the methodology is quite different,” she explains.
“I changed fields because I wanted to do something that would help people, such as predicting an El Niño or La Niña event in the tropical Pacific Ocean, which has lots of repercussions around the world.”
The work during her post-doc concerned improving rainfall predictions in the south Asian monsoon. Stephanie considered how changing different convection parameters affected rainfall, and she looked into the impact of the model’s resolution.
This image is from an article Stephanie et al. published in Climate Dynamics in 2015. It presents the climatological 850 hPa circulation in June–July–August–September in ECMWF’s ERA-Interim reanalysis and different configurations of the UK Met Office Unified Model (ensemble averaged). Grey lines indicate the ERA-Interim and Unified Model orography on their native grid at approximately 200, 60 and 25 km resolution. (Creative Commons Attribution 4.0 International Licence)
Towards the end of the project, she was working more on seasonal forecasting, which is what she has been concerned with at ECMWF since January 2016.
Seasonal forecasts
Stephanie’s work at ECMWF has revolved around improving the Centre’s seasonal forecasting system. Her position is funded by the Copernicus Climate Change Service (C3S), which ECMWF implements for the EU and which publishes seasonal forecasts from a range of forecasting centres.
ECMWF’s seasonal forecasts extend to between 7 and 13 months ahead. “The goal of a seasonal forecast is to say whether a particular season is going to be warmer or colder than average, or wetter or drier than average,” Stephanie says.
Such forecasts are thus not concerned with day-to-day variations in the weather, which can only be forecast with different degrees of confidence up to about 10 days in advance. They are, however, useful for several applications. “For example, a government may decide to buy more grit if there is a prospect of a particularly cold winter,” Stephanie explains.
Commensurate with their longer timescale, seasonal forecasts are often presented as three-month averages. “We are trying to predict what those three months will be like compared to what we’re used to.”
This chart illustrates the skill of seasonal 2 m temperature predictions for ECMWF’s SEAS5 system, which is currently in use. It shows the anomaly correlation for the mean of the forecast ensemble for December–January–February from 1 November. An anomaly correlation of 1 corresponds to a perfect deterministic forecast, while 0 means no skill.
Predictability
Seasonal forecasts can be provided because to some extent the weather depends on processes that are predictable in the long term. These include, for example, processes in the ocean.
“Physical processes in the atmosphere are quite fast, whereas physical processes in the ocean are much slower,” Stephanie says. “The development of an El Niño or La Niña event has an impact on the atmosphere.”
The development of weak La Niña conditions was predicted by ECMWF as early as from August 2024 (left), and it was confirmed in the forecast from November 2024 (right). The charts show ensemble forecasts of expected anomalies in the NINO3.4 region in the Pacific Ocean. The red lines are the ensemble members and the dotted blue line corresponds to the verifying analysis.
Likewise, land surface properties, such as soil moisture, and processes in the stratosphere impact the troposphere and change relatively slowly.
“My work consists in taking all the progress achieved at the Centre to improve these various processes and using it to make seasonal forecasts better,” Stephanie says. “It’s about testing changes to, for example, the cloud scheme or the ocean to see what kind of performance they give us in the seasonal forecast.”
This is an example of an ECMWF seasonal mean precipitation anomaly forecast, issued on 1 August 2024 for October–November–December 2024. Shaded areas show where the distribution of possible rainfall amounts for this period differs at a 10% significance level from the usual distribution of possible rainfall amounts, while solid contours indicate significance at 1%.
The chart below shows that the forecast successfully predicted some developments, particularly around the equator.
This chart indicates the precipitation anomaly for October–November–December 2024 in ECMWF’s ERA5 reanalysis.
Technical developments
Stephanie also carries out some technical work. For example, ECMWF is currently moving to a new file format in its data output. This requires work across the different forecasts we produce, including seasonal forecasts.
Machine learning methods that have initially been developed for the medium range will be tested for longer ranges, too, including the seasonal range.
“I have to make sure such developments are correctly integrated into the seasonal forecasting system,” Stephanie says. “My job is to take the work done at the Centre and apply it to the seasonal forecasting system.”
Further reading
More information on our sub-seasonal (extended-range) and seasonal (long-range) forecasts can be found in an ECMWF fact sheet.