Our Fellowship programme fosters links between ECMWF and individual scientists carrying out pioneering work in areas relevant to our strategic goals. In the ten years since its launch, collaborations with current and former Fellows have contributed significantly to progress in global Earth system modelling. Areas of joint work have included data assimilation, novel numerical methods, the atmospheric water cycle, predictability, and the calibration and post-processing of forecasts. The outcomes of this work have been shared through joint publications as well as at ECMWF seminars and workshops.

Several ECMWF Fellows have recently begun new terms, including Dr Sebastian Lerch from the Karlsruhe Institute of Technology in Germany, who was appointed for a first term in December 2023. His work will focus on advancing the use of methods from artificial intelligence and machine learning to improve the accuracy and reliability of probabilistic weather forecasts on subseasonal-to-seasonal timescales.

Here, Sebastian and other Fellows outline the research they are undertaking with colleagues at ECMWF and their plans for future work.

• Sebastian Lerch, Karlsruhe Institute of Technology (KIT)
• Sándor Baran, University of Debrecen
• Marc Bocquet, CEREA, École des Ponts and EDF R&D
• Hannah Cloke, University of Reading
• Daniela Domeisen, University of Lausanne and ETH Zurich
• Patrick Eriksson, Chalmers University of Technology
• Louise Nuijens, Delft University of Technology (TU Delft)
• Maria-Helena Ramos, INRAE (French National Research Institute for Agriculture, Food and the Environment)

Dr Sebastian Lerch, Karlsruhe Institute of Technology, Germany

The overarching aim of the collaboration within the fellowship is to advance the use of methods from artificial intelligence and machine learning for improving the accuracy and reliability of probabilistic weather forecasts on subseasonal-to-seasonal timescales.

Specifically, this will include the curation and development of joint datasets and testbeds to systematically evaluate and compare existing post-processing models from the groups at both ECMWF and KIT, for example those developed within the context of the “S2S AI Challenge” organised by the WMO in 2021.

Further, the availability of large archives of forecast and observation data will enable the development of novel modelling approaches regarding, for example, the use of additional predictors to better incorporate spatial and temporal information, as well as knowledge about physical processes, into building and estimating machine learning models for post-processing.

Prof. Dr. Sándor Baran, Full Professor, Department of Applied Mathematics and Probability Theory, Faculty of Informatics, University of Debrecen, Hungary

My ECMWF-related research has concentrated on statistical post-processing of ensemble predictions. During the first period of the Fellowship (2021–2023), we proposed a distributional regression network model and a truncated version of the generalized extreme value ensemble model output statistics (EMOS) method for calibrating wind speed ensemble predictions.

We also developed a general two-step machine learning-based approach for parametric post-processing of ensemble weather forecasts. In collaboration with Estibalíz Gascón, we investigated parametric post-processing of dual-resolution precipitation ensemble forecasts. We also compared the forecast skill of various two-step multivariate post-processing approaches using global ECMWF temperature, wind speed and precipitation ensemble predictions. More recently, we investigated both discrete and continuous post-processing of ECMWF visibility ensemble forecasts.

I have also helped Ivana Aleksovska in implementing EMOS models for various weather quantities and hosted her in Debrecen. In a most recent project with Martin Leutbecher, we derived the multivariate version of the fair logarithmic score.

My next plan is to study the effect of post-processing on the operational dual-resolution forecasts provided by Cycle 48r1 of the ECMWF Integrated Forecasting System (IFS).

Prof. Marc Bocquet, CEREA, École des Ponts and EDF R&D, Île-de-France, France

Alban Farchi (also from École des Ponts) and I have collaborated with Massimo Bonavita, Marcin Chrust, and Patrick Laloyaux at ECMWF to leverage a methodological approach that we developed several years ago. We proved that extracting information from analysis increments is not only a mathematically valid technique for correcting model trajectories, akin to classical data assimilation, but also for refining the model itself.

We enforce model correction by augmenting the physical model with a neural network, resulting in a hybrid model. Our combined data assimilation and machine learning formalism parallels the implementation of the weak-constraint 4D-Var for correcting the IFS in the stratosphere, and it is tailored to meet the specifications of OOPS-IFS.

The algorithm was tested on increasingly complex models. Following the publication of three papers detailing the method and its implementation, in March this year we submitted our initial findings on the IFS, showcasing the forecast performance enhancements achieved through our approach.

Prof. Hannah L Cloke OBE, Professor of Hydrology, University of Reading, UK

I work daily with ECMWF scientists on global Earth system modelling and applications of numerical weather prediction, and on river flow forecasting, flood prediction, land surface hydrology with ERA5-Land and the Global Flood Awareness System (GloFAS). I also undertake research on tropical cyclones, uncertainty and predictability, and forecasting with machine learning, and supervise a number of joint PhD researchers working together with ECMWF.

As part of the University of Reading–ECMWF collaboration “Advancing the Frontiers of Earth System Prediction”, from 2024 I will lead a five-year project investigating the use of machine learning for improving hydrological representation and predictability in Earth system models.

In the last three years, I have published 20 joint research publications with ECMWF, including work on ensembles, tropical cyclones, the multi-layer snow scheme, ERA5-HEAT and flood forecasting.

I also continue to work with ECMWF on a number of research projects, including the CLimate Adaptation and REsilience programme’s REPRESA project – Resilience and preparedness to tropical cyclones across southern Africa.

Prof. Dr. Daniela Domeisen, University of Lausanne and ETH Zurich, Switzerland

The first part of my fellowship was dedicated to stratosphere–troposphere coupling, to understand how the stratosphere can influence tropospheric weather on subseasonal timescales of weeks to months. In our collaboration between ECMWF (Inna Polichtchouk) and the University of Lausanne (Wolfgang Wicker), we found that increasing vertical model resolution allows for a better representation of gravity waves, which strongly reduces the cold bias in the lower stratosphere, a crucial region for determining the downward impact onto the troposphere.

We are further in the process of analysing the uncertainty of stratospheric predictions and of tracing the origins of this uncertainty back to the troposphere, in a collaboration between ETH Zurich (Rachel Wu) and ECMWF (Inna Polichtchouk).

The second part of my fellowship concerns the subseasonal prediction of extreme and compound weather events such as cold extremes, heatwaves, and precipitation extremes arising from a range of predictors, including the stratosphere, tropical phenomena, as well as extratropical wave dynamics, in collaboration with Frédéric Vitart and his research group at ECMWF.

(Photo: Selina Betschart, ETHZ)

Prof. Patrick Eriksson, Professor in global environmental measurements, Department of Space, Earth and Environment, Chalmers University of Technology, Sweden

My present research focuses on providing better data on humidity, clouds and precipitation to deepen our understanding of the atmospheric water cycle. I combine advanced radiative transfer simulations with machine learning for data extraction to reach this goal. Special attention is given to making better use of existing passive microwave measurements and preparing for the Arctic Weather Satellite (AWS) and Ice Cloud Imager (ICI) missions. These two missions will extend the coverage of microwave data into the sub-mm domain. For the ECMWF fellowship, my role in the Atmospheric Radiative Transfer Simulator (ARTS) is of particular relevance.

The work during my first fellowship period focused on applying the database of single scattering data that my research group has developed. Data from this source are now integrated into RTTOV-SCATT, the tool applied by ECMWF and others for simulating microwave “all-sky” radiances. RTTOV-SCATT was also compared to ARTS to ensure that it can simulate future high-frequency channels of the AWS and ICI. These efforts are described in several joint publications. The current fellowship period will focus on using the novel data from the AWS and ICI to assess the particle models assumed in RTTOV and, in this way, increase the general impact of satellite observations affected by clouds.

At ECMWF’s Annual Seminar in 2021, I gave a presentation on Considerations for radiative transfer observation operators in data assimilation.

Dr Louise Nuijens, Department of Geosciences and Remote Sensing, Delft University of Technology (TU Delft), the Netherlands

As an ECMWF Fellow, I take a closer look at parametrized and resolved convective momentum transport (CMT) in the operational IFS. CMT is the process whereby clouds modify winds through up-and downdrafts and mesoscale circulations.

Through systematic wind measurements by scanning wind lidars and cloud radars at the Ruisdael Observatory (CMTRACE campaigns), my team has derived high-resolution (in time and height) profiles of horizontal and vertical wind and their (co-)variances through the boundary layer, and, for the first time, visualised the 1–10 km circulations accompanying convection.

This month, PhD student Alessandro Savazzi started a three-month collaborative visit to the physics team of ECMWF in Bonn. The goal is to run the IFS at 9 km and 4 km resolution at the time of the campaigns and study the detailed wind evolution by convection in light of the observations, in an effort to unravel systematic biases in winds over land, following an earlier collaboration that focused on wind biases over the ocean.

Dr Maria-Helena Ramos, INRAE (French National Research Institute for Agriculture, Food and the Environment), France

The fellowship started in 2019, in time for a first visit to ECMWF in Reading before the COVID-related lockdowns. This challenging situation prompted other types of interactions and, fortunately, we could keep working on common research interests. A main achievement was the Joint Virtual Workshop on Connecting global to local hydrological modelling and forecasting: scientific advances and challenges in 2021. It was a collaborative effort by ECMWF, the Copernicus Emergency Management Service (CEMS), the Copernicus Climate Change Service (C3S), the Hydrological Ensemble Prediction EXperiment (HEPEX), and the Global Flood Partnership (GFP). More than 600 participants attended, and we published a paper summarising the achievements and opportunities for forecasting and decision-making in hydrological forecasting and warning. It complements other discussions we have raised in the hydrological forecasting community on forecast communication and forecast usefulness.

Both extremes in hydrology (flood and drought) have been explored in our work, and a vision for how to improve hydrometeorological forecasting systems and products, including advancing numerical weather and hydrological models, was presented in a publication summarising the main lessons learnt from the EU Horizon 2020 IMPREX project.

Now, excitingly, we are involved in another EU project, which started in November 2023, the MedEWSa project. It is coordinated by the WMO, and we will have the opportunity to continue working together on innovative tools and training material for interoperable multi-hazard, impact-based early warning systems.

(Photo: Irstea)

Information about all current and former ECMWF Fellows is available on the Serving meteorology page. This way of working with top scientists around the world is part of a bigger practice at ECMWF of forging strong partnerships. This is, for example, also pursued through hosting WMO Fellows and DWD Early-career Fellows at ECMWF.