30.09.2020

We have published a summary of the first season-long simulation of the global atmosphere with 1.4 km grid-spacing. The simulation was based on the Integrated Forecast System (IFS) of ECMWF and performed on the Summit supercomputer at Oak Ridge National Laboratory.
N. P. Wedi, I. Polichtchouk, P. Dueben, V. G. Anantharaj, P. Bauer, S. Boussetta, P. Browne, W. Deconinck, W. Gaudin, I. Hadade, S. Hatfield, O. Iffrig, P. Lopez, P. Maciel, A. Mueller, S. Saarinen, I. Sandu, T. Quintino, F. Vitart: A baseline for global weather and climate simulations at 1.4 km resolution, accepted in JAMES, https://doi.org/10.1029/2020MS002192, 2020.

In a collaboration between the University of Oxford and ECMWF, we have investigated how different formats of 16-bit arithmetic could be used for weather and climate models. 16-bit arithmetic is getting increasingly important in modern High-Performance Computing as deep learning applications require high flop rates at low precision.  
M. Kloewer, P. D. Dueben, T. N. Palmer: Number formats, error mitigation and scope for 16-bit arithmetics in weather and climate modelling analysed with a shallow water model, accepted in JAMES, https://doi.org/10.1029/2020MS002246, 2020.

In a collaborative effort of ETH and ECMWF, we have explored deep learning applications to improve global ensemble predictions via post-processing. Results indicate that forecast scores such as CRPS can indeed be improved.
P. Groenquist, C. Yao, T. Ben-Nun, N. Dryden, P. Dueben, S. Li, T. Hoefler: Deep Learning for Post-Processing Ensemble Weather Forecasts, https://arxiv.org/abs/2005.08748, 2020.

In a collaboration between several groups, we have published the first benchmark dataset for machine learning applications on the learning of the dynamics of the atmosphere from historical data. The benchmark dataset "WeatherBench" aims to provide Machine Learning specialists with a scientific problem that is of interest to the weather and climate community and allows for a quantitative comparison of results for different machine learning approaches.
S
. Rasp, P. D. Dueben, S. Scher, J.A. Weyn, S. Mouatadid, N. Thuerey: WeatherBench: A benchmark dataset for data-driven weather forecasting, JAMES, https://doi.org/10.1029/2020MS002203, 2020.

We present the collaborative model of the ESiWACE2 services, where research software engineers (RSEs) from the Netherlands eScience Center (NLeSC) and Atos offer their expertise to climate and earth system modeling groups across Europe. Within 6-month collaborative projects, the RSEs provide guidance and advice regarding the performance, portability to new architectures, and scalability of selected applications. We present the four projects running in 2020 as examples of this funding structure.
Gijs van den Oord, Victor Azizi, Alessio Sclocco, Georges-Emmanuel Moulard, David Guibert, Jisk Attema, Erwan Raffin, and Ben van Werkhoven: ESiWACE2 Services: RSE collaborations in Weather and Climate, accepted for publication in Research Software Engineers in HPC (RSE-HPC-2020) Workshop at Supercomputing 2020 (SC20), 2020.