Search

Machine learning (ML) and in particular artificial neural networks (ANNs) push state-of-the-art solutions for many hard problems e.g., image classification, speech recognition or time series forecasting. In the domain of climate science, ANNs have good prospects to identify causally linked modes of climate variability as key to understand the climate system and to improve the predictive skills of forecast systems. To attribute climate events in a data-driven way with ANNs, we need sufficient training data, which is often limited for real world measurements. The data science community provides standard data sets for many applications. As a new data set, we introduce a collection of climate indices typically used to describe Earth System dynamics. This collection is consistent and comprehensive as we use control simulations from Earth System Models (ESMs) over 1,000 years to derive climate indices. The data set is provided as an open-source framework that can be extended and customized to individual needs. It allows to develop new ML methodologies and to compare results to existing methods and models as benchmark. Exemplary, we use the data set to predict rainfall in the African Sahel region and El Niño Southern Oscillation with various ML models. We argue that this new data set allows to thoroughly explore techniques from the domain of explainable artificial intelligence to have trustworthy models, that are accepted by domain scientists. Our aim is to build a bridge between the data science community and researchers and practitioners from the domain of climate science to jointly improve our understanding of the climate system.

Lagrangian particle dispersal simulations are widely used for studying directed connectivity between different locations in the ocean. They are used, both, for the understanding of ocean physics and for interdisciplinary questions. One biological example is the dispersal of passively drifting marine organisms.The typical modus operandi of such “bio-physical” studies is to design an underlying Lagrangian simulation in close synchronisation with a specific biological research question. This leads to a conflation of concerns between physical and biological aspects of the study. This conflation might result in repeated and slow development cycles of re-calculation for different scenarios and hence inhibit scientific progress.We aim at improving the separation of concerns between biological and physical components for bio-physical Lagrangian studies, by aggregating physical Lagrangian data into directed multigraphs encoding locations as nodes and multiple parallel pathways as directed edges. Those graphs condense the physics-based information on directed oceanic relations and thus serve as a basis for simultaneously answering various biological questions on connectivity. As the proposed aggregation retains the distinction of different pathways between locations, it can, to some extent, also provide information of underway environmental conditions. This greatly enhances the range of applications of our approach over existing aggregations of Lagrangian data as connectivity probability graphs.We present a specific set of biological case studies — the multi-year spreading of two oyster diseases in the North Sea — and develop a framework that facilitates efficiently and simultaneously testing multiple biological hypotheses for marine diseases of various species based on the same processed physical data set.

In this study, we use a joint observation–model approach to investigate the mixed-layer heat and salt annual mean as well as seasonal budgets in the eastern tropical Atlantic. The regional PREFCLIM (PREFACE Climatology) observational climatology provides the budget terms with a relatively low spatial and temporal resolution compared to the online NEMO (Nucleus for European Modeling of the Ocean; Madec, G., 2014) model, and this is later resampled as in PREFCLIM climatology. In addition, advection terms are recomputed offline from the model as PREFCLIM gridded advection computation. In the Senegal, Angola, and Benguela regions, the seasonal cycle of mixed-layer temperature is mainly governed by surface heat fluxes; however, it is essentially driven by vertical heat diffusion in the equatorial region. The seasonal cycle of mixed-layer salinity is largely controlled by freshwater flux in the Senegal and Benguela regions; however, it follows the variability of zonal and meridional salt advection in the equatorial and Angola regions, respectively. Our results show that the time-averaged spatial distribution of NEMO offline heat and salt advection terms compares much better to PREFCLIM horizontal advection terms than the online heat and salt advection terms. However, the seasonal cycle of horizontal advection in selected regions shows that NEMO offline terms do not always compare well with PREFCLIM, sometimes less than online terms. Despite this difference, these results suggest the important role of small-scale variability in mixed-layer heat and salt budgets.

Observations of the eastern subpolar North Atlantic in the 2010s show exceptional freshening and cooling of the upper ocean, peaking in 2016 with the lowest salinities recorded for 120 years. Published theories for the mechanisms driving the freshening include: reduced transport of saltier, warmer surface waters northwards from the subtropics associated with reduced meridional overturning; shifts in the pathways of fresher, cooler surface water from the Labrador Sea driven by changing patterns of wind stress; and the eastward expansion of the subpolar gyre. Using output from a high-resolution hindcast model simulation, we propose that the primary cause of the exceptional freshening and cooling is reduced surface heat loss in the Labrador Sea. Tracking virtual fluid particles in the model backwards from the eastern subpolar North Atlantic between 1990 and 2020 shows the major cause of the freshening and cooling to be an increased outflow of relatively fresh and cold surface waters from the Labrador Sea; with a minor contribution from reduced transport of warmer, saltier surface water northward from the subtropics. The cooling, but not the freshening, produced by changing proportions of source waters is mitigated by reduced along-track heat loss to the atmosphere in the North Atlantic Current. We analyse modelled boundary exchanges and water mass transformation in the Labrador Sea to show that since 2000, while inflows of lighter surface waters remain steady, the increasing output of these waters is due to reduced surface heat loss in the Labrador Sea beginning in the early 2000s. Tracking particles further upstream reveals the primary source of the increased volume of lighter water transported out of the Labrador Sea is increased recirculation of water, and therefore longer residence times, in the upper 500–1000 m of the subpolar gyre.

Lagrangian experiments for particle tracing in atmosphere or ocean models and their analysis are a cornerstone of earth-system studies. They cover diverse study objectives such as the identification of pathways or source regions. Data for Lagrangian studies are generated by releasing virtual particles in one or in multiple locations of interest and simulating their advective-diffusive behavior backwards or forwards in time. Identifying main pathways connecting two regions of interest is often done by counting the trajectories that reach both regions. Here, the exact source and target region must be defined manually by a researcher. Manually defining the importance and exact location of these regions introduces a highly subjective perspective into the analysis. Additionally, to investigate all major target regions, all of them must be defined manually and the data must be analyzed accordingly. This human element slows down and complicates large scale analyses with many different sections and possible source areas.We propose to significantly reduce the manual aspect by automatizing this process. To this end, we combine methods from different areas of machine learning and pattern mining into a sequence of steps. First, unsupervised methods, i.e., clustering, identify possible source areas on a randomized subset of the data. In a successive second step, supervised learning, i.e., classification, labels the positions along the trajectories according to their most probable source area using the previously automatically identified clusters as labels. The results of this approach can then be compared quantitatively to the results of analyses with manual definition of source areas and border-hitting-based labeling of the trajectories. Preliminary findings suggest that this approach could indeed help greatly to objectify and fasten the analysis process for Lagrangian Particle Release Experiments.

The distribution of passively drifting particles within highly turbulent flows is a classic problem in marine sciences. The use of trajectory clustering on huge amounts of simulated marine trajectory data to identify main pathways of drifting particles has not been widely investigated from a data science perspective yet. In this paper, we propose a fast and computationally light method to efficiently identify main pathways in large amounts of trajectory data. It aims at overcoming some of the issues of probabilistic maps and existing trajectory clustering approaches. Our approach is evaluated against simulated larvae dispersion data based on a real-world model that have been produced as part of work in the marine science domain.

In the earth-physics community Lagrangian trajectories are used within multiple contexts – analyzing the spreading of pollutants in the air or studying the connectivity between two ocean regions of interest. Huge amounts of data are generated reporting the geo position and other variables e.g. temperature, depth or salinity for particles spreading in the ocean. As state-of-the-art, these experiments are analyzed and visualized by binning the particle positions to pre-defined rectangular boxes. For each box a particle density is computed which then yields a probability map to visualize major pathways. Identifying the main pathways directly still remains a challenge when huge amounts of particles and variables are involved.We propose a novel method that focuses on linking the net fluctuation of particles between adaptable hexagonal grid cells. For very small areas the rectangular boxing does not imply big differences in area or shape, though when gridding larger areas it introduces rather large distortions. Using hexagons instead provides multiple advantages, such as constant distances between the centers of neighboring boxes or more possibilities of movement due to 6 edges instead of 4 with a lower number of neighbors at the same time (6 instead of 9). The net fluctuation can be viewed as transition strength between the cells.Through this network perspective, the density of the transition strength can be visualized clearly. The main pathways are the transitions with the highest net fluctuation. Thus, simple statistical filtering can be used to reveal the main pathways. The combination of network analysis and adaptable hexagonal grid cells yields a surprisingly time and resource efficient way to identify main pathways.

The direct response of the tropical mixed layer to near-inertial waves (NIWs) has only rarely been observed. Here, we present upper-ocean turbulence data that provide evidence for a strongly elevated vertical diffusive heat flux across the base of the mixed layer in the presence of a NIW, thereby cooling the mixed layer at a rate of 244 W m−2 over the 20 h of continuous measurements. We investigate the seasonal cycle of strong NIW events and find that despite their local intermittent nature, they occur preferentially during boreal summer, presumably associated with the passage of atmospheric African Easterly Waves. We illustrate the impact of these rare but intense NIW induced mixing events on the mixed layer heat balance, highlight their contribution to the seasonal evolution of sea surface temperature, and discuss their potential impact on biological productivity in the tropical North Atlantic. How the tropical mixed layer of the ocean reacts to near-inertial waves has rarely been observed directly. Here, the authors present new data that shows strongly elevated vertical diffusive heat flux in the presence of near-inertial waves, causing a cooling of the mixed layer that is particularly strong in summer.

Howfar domarine larvae disperse in the ocean? Decades of population genetic studies have revealed generally low levels of genetic structure at large spatial scales (hundreds of kilometres). Yet this result, typically based on discrete sampling designs, does not necessarily imply extensive dispersal. Here, we adopt a continuous sampling strategy along 950 km of coast in the northwestern Mediterranean Sea to address this question in four species. In line with expectations, we observe weak genetic structure at a large spatial scale. Nevertheless, our continuous sampling strategy uncovers a pattern of isolation by distance at small spatial scales (few tens of kilometres) in two species. Individual- based simulations indicate that this signal is an expected signature of restricted dispersal. At the other extreme of the connectivity spectrum, two pairs of individuals that are closely related genetically were found more than 290 km apart, indicating long-distance dispersal. Such a combination of restricted dispersal with rare long-distance dispersal events is supported by a high-resolution biophysical model of larval dispersal in the study area, and we posit that it may be common in marine species. Our results bridge population genetic studies with direct dispersal studies and have implications for the design of marine reserve networks, RESERVEBENFIT, SI

Metamorphic testing seeks to validate software in the absence of test oracles. Our application domain is ocean modeling, where test oracles often do not exist, but where symmetries of the simulated physical systems are known. In this short paper we present work in progress for automated generation of metamorphic test scenarios using machine learning. Metamorphic testing may be expressed as f(g(X))=h(f(X)) with f being the application under test, with input data X, and with the metamorphic relation (g, h). Automatically generated metamorphic relations can be used for constructing regression tests, and for comparing different versions of the same software application. Here, we restrict to h being the identity map. Then, the task of constructing tests means finding different g which we tackle using machine learning algorithms. These algorithms typically minimize a cost function. As one possible g is already known to be the identity map, for finding a second possible g, we construct the cost function to minimize for g being a metamorphic relation and to penalize for g being the identity map. After identifying the first metamorphic relation, the procedure is repeated with a cost function rewarding g that are orthogonal to previously found metamorphic relations. For experimental evaluation, two implementations of an ocean-modeling application will be subjected to the proposed method with the objective of presenting the use of metamorphic relations to test the implementations of the applications., Shot paper:2 pages, 2020 IEEE International Conference On Artificial Intelligence Testing (AITest)

This presentation will detail the design, implementation, and operation of ERDA, which is a collection of external version-controlled research datasets, of multiple synchronized deployments of the data, of a growing set of minimal examples using the datasets from various deployments, of stand-alone tools to create, maintain, and deploy new datasets, and of documentation targeting different audiences (users, maintainers, developers).ERDA was designed with the following principles in mind: Provide clear data provenance and ensure long-term availability, minimize effort for adding data and make all contents available to all users immediately, ensure unambiguous referencing and develop transparent versioning conventions, embrace mobility of scientists and target independence from the infrastructure of specific institutions.The talk will show how the data management is done with Git-LFS, demonstrate how data repositories are rendered from human-readable data, and give an overview of the versioning scheme that is applied.

Demand on high-end high performance computer (HPC) systems by the Earth system science community today encompasses not only the handling of complex simulations but also machine and deep learning as well as interactive data analysis workloads on large volumes of data. This poster addresses the infrastructure needs of large-scale interactive data analysis workloads on supercomputers. It lays out how to enable optimizations of existing infrastructure with respect to accessibility, usability and interactivity and aims at informing decision making about future systems. To enhance accessibility, options for distributed access, e.g. through JupyterHub, will be evaluated. To increase usability, the unification of working environments via the operation and the joint maintenance of containers will be explored. Containers serve as a portable base software setting for data analysis application stacks and allow for long-term usability of individual working environments and repeatability of scientific analysis. Aiming for interactive big-data analysis on HPC will also help the scientific community in utilizing increasingly heterogeneous supercomputers, since the modular data-analysis stack already contains solutions for seamless use of various architectures such as accelerators. However, to enable day-to-day interactive work on supercomputers, the inter-operation of workloads with quick turn-around times and highly variable resource demands needs to be understood and evaluated. To this end, scheduling policies on selected HPC systems are reviewed with respect to existing technical solutions such as job preemption, utilizing the resiliency features of parallel computing toolkits like Dask. Presented are preliminary results focussing on the aspects of usability and interactive use of HPC systems on the basis of typical use cases from the ocean science community.

A new Earth system model, the Flexible Ocean and Climate Infrastructure (FOCI), is introduced. A first version of FOCI consists of a global high-top atmosphere (European Centre Hamburg general circulation model; ECHAM6.3) and an ocean model (Nucleus for European Modelling of the Ocean v3.6; NEMO3.6) as well as sea-ice (Louvain-la-Neuve sea Ice Model version 2; LIM2) and land surface model components (Jena Scheme for Biosphere Atmosphere Coupling in Hamburg; JSBACH), which are coupled through the OASIS3-MCT software package. FOCI includes a number of optional modules which can be activated depending on the scientific question of interest. In the atmosphere, interactive stratospheric chemistry can be used (ECHAM6-HAMMOZ) to study, for example, the effects of the ozone hole on the climate system. In the ocean, a biogeochemistry model (Model of Oceanic Pelagic Stoichiometry; MOPS) is available to study the global carbon cycle. A unique feature of FOCI is the ability to explicitly resolve mesoscale ocean eddies in specific regions. This is realized in the ocean through nesting; first examples for the Agulhas Current and the Gulf Stream systems are described here. FOCI therefore bridges the gap between coarse-resolution climate models and global high-resolution weather prediction and ocean-only models. It allows to study the evolution of the climate system on regional and seasonal to (multi)decadal scales. The development of FOCI resulted from a combination of the long-standing expertise in ocean and climate modeling in several research units and divisions at the Helmholtz Centre for Ocean Research Kiel (GEOMAR). FOCI will thus be used to complement and interpret long-term observations in the Atlantic, enhance the process understanding of the role of mesoscale oceanic eddies for large-scale oceanic and atmospheric circulation patterns, study feedback mechanisms with stratospheric processes, estimate future ocean acidification, and improve the simulation of the Atlantic Meridional Overturning Circulation changes and their influence on climate, ocean chemistry and biology. In this paper, we present both the scientific vision for the development of FOCI as well as some technical details. This includes a first validation of the different model components using several configurations of FOCI. Results show that the model in its basic configuration runs stably under pre-industrial control as well as under historical forcing and produces a mean climate and variability which compares well with observations, reanalysis products and other climate models. The nested configurations reduce some long-standing biases in climate models and are an important step forward to include the atmospheric response in multidecadal eddy-rich configurations. © 2020 Copernicus GmbH. All rights reserved.

Antarctic Intermediate Water (AAIW) is an important conduit for nutrients to reach the nutrient‐poor low‐latitude ocean areas. In the Atlantic, it forms part of the return path of the Atlantic Meridional Overturning Circulation (AMOC). Despite the importance of AAIW, little is known about variations in its composition and signature during the prominent AMOC and climate changes of the last deglaciation. Here, we reconstruct benthic foraminiferal Mg/Ca‐based intermediate water temperatures (IWTMg/Ca) and intermediate water neodymium (Nd) isotope compositions at sub‐millennial resolution from unique sediment cores located at the northern tip of modern AAIW extent in the tropical W‐Atlantic (850 and 1018 m water depth). Our data indicate a pronounced warming of AAIW in the tropical W‐Atlantic during Heinrich Stadial 1 (HS1) and the Younger Dryas (YD). We argue that these warming events were induced by major AMOC perturbations resulting in the pronounced accumulation of heat in the surface Southern Ocean. Combined with published results, our data suggest the subsequent uptake of Southern Ocean heat by AAIW and its rapid northward transfer to the tropical W‐Atlantic. Hence, the rapid deglacial northern climate perturbations directly controlled the AAIW heat budget in the tropical W‐Atlantic after a detour via the Southern Ocean. We speculate that the ocean heat redistribution via AAIW effectively dampened Southern Hemisphere warming during the deglaciation and may therefore have been a crucial player in the climate seesaw mechanisms between the two hemispheres.

[1] We use an eddying realistic primitive-equation model of the Southern Ocean to examine the spatial and temporal distribution of near-inertial wind-power input (WPI) and near-inertial energy (NIE) in the Southern Ocean. We find that the modeled near-inertial WPI is almost proportional to inertial wind-stress variance (IWSV), while the modeled NIE is modulated by the inverse of the mixed-layer depth. We go on to assess recent decadal trends of near-inertial WPI from trends of IWSV based on reanalysis wind stress. Averaged over the Southern Ocean, annual-mean IWSV is found to have increased by 16% over the years 1979–2011. Part of the increase of IWSV is found to be related to the positive trend of the Southern-Annular Mode over the same period. Finally, we show that there are horizontal local maxima of NIE at depth that are almost exclusively associated with anticyclonic eddies.

A realistic primitive-equation model of the Southern Ocean at eddying spatial resolution is used to examine the effect of ocean-surface-velocity dependence of the wind stress on the strength of near-inertial oscillations. Accounting for the ocean-surface-velocity dependence of the wind stress leads to a large reduction of wind-induced near-inertial energy of approximately 40 percent and of wind power input into the near-inertial frequency band of approximately 20 percent. A large part of this reduction can be explained by the leading-order modification to the wind stress if the ocean-surface velocity is included. The strength of the reduction is shown to be modulated by the inverse of the ocean-surface-mixed-layer depth. We conclude that the effect of surface-velocity dependence of the wind stress should be taken into account when estimating the wind-power input into the near-inertial frequency band and when estimating near-inertial energy levels in the ocean due to wind forcing.

Near-inertial oscillations are ubiquitous in the ocean and are believed to play an important role in the global climate system. Studies on wind power input to near-inertial motions (WPI) have so far focused primarily on estimating the time-mean WPI, with little attention being paid to its temporal variability. In this study, a combination of atmospheric reanalysis products, a high-resolution ocean model and linear regression models are used to investigate for the first time the relationship between interannual variability of WPI in the North Atlantic and the North Atlantic Oscillation (NAO), motivated by the idea that the NAO serves as a good indicator for storminess over the North Atlantic and that storms account for the majority of WPI. It is found that WPI at low and high latitudes of the North Atlantic is significantly correlated to the NAO, owing to its influence on the configuration of the storm track. Positive (negative) NAO conditions are associated with increased WPI in the subpolar (subtropical) ocean. Basin-wide WPI is found to be significantly enhanced under negative NAO conditions, but is not significantly different from the climatological average under positive NAO conditions. This indicates a weak inverse relationship between basin-wide WPI and the NAO, contradicting intuitive expectations. The asymmetric impact of the NAO on basin-wide WPI results from greater sensitivity of WPI to near-inertial wind forcing at lower latitudes due to the variation of the Coriolis parameter with latitude.

We investigate the discrete chiral transformation of a Majorana fermion on a torus. Depending on the boundary conditions the integration measure can change sign. Taking this anomalous behavior into account we define a chiral order parameter as a ratio of partition functions with differing boundary conditions. Then the lattice realization of the Gross-Neveu model with Wilson fermions is simulated using the recent `worm' technique on the loop gas or all-order hopping representation of the fermions. An algorithm is formulated that includes the Gross-Neveu interaction for N fermion species. The critical line m_c(g) is constructed for a range of couplings at N = 6 and for N = 2, the Thirring model, as examples., 21 pages, 1 figure. Typos corrected and a few sentences improved. To appear in Nucl. Phys. B [FS]