JUWELS Cluster and Booster: Exascale Pathfinder with Modular Supercomputing Architecture at Juelich Supercomputing Centre

JUWELS is a multi-petaflop modular supercomputer operated by Juelich Supercomputing Centre at Forschungszentrum Juelich as a European and national supercomputing resource for the Gauss Centre for Supercomputing. In addition, JUWELS serves the Earth system modeling community and the AI community within the Helmholtz Association as well. JUWELS currently consists of two modules. The first module deployed in 2018 is the so-called Cluster module. The Cluster is a BullSequana X1000 system with Intel Xeon Skylake-SP processors and Mellanox EDR InfiniBand. The second module deployed in 2020 is the so-called Booster module. The Booster is a BullSequana XH2000 system with 2nd generation AMD EPYC processors, NVIDIA Ampere GPUs and NVIDIA/Mellanox HDR Infiniband. This paper describes in detail the architecture of the system from a users perspective, and additionally provides further insights into the administrative infrastructure used to operate the supercomputer.

Modeling ocean-induced rapid Earth rotation variations: an update

We revisit the problem of modeling the ocean’s contribution to rapid, non-tidal Earth rotation variations at periods of 2–120 days. Estimates of oceanic angular momentum (OAM, 2007–2011) are drawn from a suite of established circulation models and new numerical simulations, whose finest configuration is on a "Image missing"$$^\circ $$ ∘ grid. We show that the OAM product by the Earth System Modeling Group at GeoForschungsZentrum Potsdam has spurious short period variance in its equatorial motion terms, rendering the series a poor choice for describing oceanic signals in polar motion on time scales of less than $$\sim $$ ∼ 2 weeks. Accounting for OAM in rotation budgets from other models typically reduces the variance of atmosphere-corrected geodetic excitation by $$\sim $$ ∼ 54% for deconvolved polar motion and by $$\sim $$ ∼ 60% for length-of-day. Use of OAM from the "Image missing"$$^\circ $$ ∘ model does provide for an additional reduction in residual variance such that the combined oceanic–atmospheric effect explains as much as 84% of the polar motion excitation at periods < 120 days. Employing statistical analysis and bottom pressure changes from daily Gravity Recovery and Climate Experiment solutions, we highlight the tendency of ocean models run at a 1$$^\circ $$ ∘ grid spacing to misrepresent topographically constrained dynamics in some deep basins of the Southern Ocean, which has adverse effects on OAM estimates taken along the 90$$^\circ $$ ∘ meridian. Higher model resolution thus emerges as a sensible target for improving the oceanic component in broader efforts of Earth system modeling for geodetic purposes.

Call for Papers on Machine Learning and Earth System Modeling

Contributions are invited to a new journal special collection on the use of new machine learning methodologies and applications of machine learning to Earth system modeling.

The diurnal Energy Balance Model (dEBM): a convenient surface mass balance solution for ice sheets in Earth system modeling

The surface mass balance scheme dEBM (diurnal Energy Balance Model) provides a novel interface between the atmosphere and land ice for Earth system modeling, which is based on the energy balance of glaciated surfaces. In contrast to empirical schemes, dEBM accounts for changes in the Earth’s orbit and atmospheric composition. The scheme only requires monthly atmospheric forcing (precipitation, temperature, shortwave and longwave radiation, and cloud cover). It is also computationally inexpensive, which makes it particularly suitable to investigate the ice sheets' response to long-term climate change. After calibration and validation, we analyze the surface mass balance of the Greenland Ice Sheet (GrIS) based on climate simulations representing two warm climate states: a simulation of the mid-Holocene (approximately 6000 years before present) and a climate projection based on an extreme emission scenario which extends to the year 2100. The former period features an intensified summer insolation while the 21st century is characterized by reduced outgoing longwave radiation. Specifically, we investigate whether the temperature–melt relationship, as used in empirical temperature-index methods, remains stable under changing insolation and atmospheric composition. Our results indicate that the temperature–melt relation is sensitive to changes in insolation on orbital timescales but remains mostly invariant under the projected warming climate of the 21st century.

Identifying the sensitivity of GPS to non-tidal loadings at various time resolutions: examining vertical displacements from continental Eurasia

We examine the sensitivity of the Global Positioning System (GPS) to non-tidal loading for a set of continental Eurasia permanent stations. We utilized daily vertical displacements available from the Nevada Geodetic Laboratory (NGL) at stations located at least 100 km away from the coast. Loading-induced predictions of displacements of earth’s crust are provided by the Earth-System-Modeling Group of the GFZ (ESMGFZ). We demonstrate that the hydrological loading, supported by barystatic sea-level changes to close the global mass budget (HYDL + SLEL), contributes to GPS displacements only in the seasonal band. Non-tidal atmospheric loading, supported by non-tidal oceanic loading (NTAL + NTOL), correlates positively with GPS displacements for almost all time resolutions, including non-seasonal changes from 2 days to 5 months, which are often considered as noise, intra-seasonal and seasonal changes with periods between 4 months and 1.4 years, and, also, inter-annual signals between 1.1 and 3.0 years. Correcting the GPS vertical displacements by NTAL leads to a reduction in the time series variances, evoking a whitening of the GPS stochastic character and a decrease in the standard deviation of noise. Both lead, on average, to an improvement in the uncertainty of the GPS vertical velocity by a factor of 2. To reduce its impact on the GPS displacement time series, we recommend that NTAL is applied at the observation level during the processing of GPS observations. HYDL might be corrected at the observation level or remain in the data and be applied at the stage of time series analysis.

Seasonal to multi-year soil moisture drought forecasting

Soil moisture predictability on seasonal to decadal (S2D) continuum timescales over North America is examined from the Community Earth System Modeling (CESM) experiments. The effects of ocean and land initializations are disentangled using two large ensemble datasets—initialized and uninitialized experiments from the CESM. We find that soil moisture has significant predictability on S2D timescales despite limited predictability in precipitation. On sub-seasonal to seasonal timescales, precipitation variability is an order of magnitude greater than soil moisture, suggesting land surface processes, including soil moisture memory, reemergence, land–atmosphere interactions, transform a less predictable precipitation signal into a more predictable soil moisture signal.

Easing and promoting the application of ML and AI in earth system sciences - introducing the KI:STE platform

&lt;p&gt;Earth system modeling is virtually impossible without dedicated data analysis. Typically, data are big and due to the complexity of the system, adequate tools for the analysis lie in the domain of machine learning or artificial intelligence. However, earth system specialists have other expertise than developing and deploying state-of-the art programming code which is needed to efficiently use modern software frameworks and computing resources. In addition, Cloud and HPC infrastructure are frequently needed to run analyses with data beyond Tera- or even Petascale volume, and corresponding requirements on available RAM, GPU and CPU sizes.&amp;#160;&lt;/p&gt;&lt;p&gt;Inside the KI:STE project (www.kiste-project.de), we extend the concepts of an existing project, the Mantik-platform (www.mantik.ai), such that handling of data and algorithms is facilitated for earth system analyses while abstracting technical challenges such as scheduling and monitoring of training jobs and platform specific configurations away from the user.&lt;/p&gt;&lt;p&gt;The principles for design are collaboration and reproducibility of algorithms from the first data load to the deployment of a model to a cluster infrastructure. In addition to the executive part where code is developed and deployed, the KI:STE project develops a learning platform where dedicated topics in relation to earth system science are systematically and pedagogically presented.&lt;/p&gt;&lt;p&gt;In this presentation, we show the architecture and interfaces of the KI:STE platform together with a simple example.&lt;/p&gt;