Stochastic and deterministic low-order dynamical system modeling to analyze climate sensitivity and climate variability

Professor Sergei Soldatenko1, Dr Robert Colman1

1Bureau Of Meteorology, Docklands, Australia

 

There is emerging (and tantalizing) evidence of links between “natural” climate variability, and climate change.  Recent results suggest that climate feedbacks (including cloud feedback) are correlated between forced climate change and climate variability on interannual and decadal timescales.  There are hints, too, of relationships between the magnitude of climate variability on long (e.g. decadal) timescales in coupled climate models (CCM) and their climate sensitivity.  However, such relationships remaining poorly understood, limiting our confidence in using them as constraints for future forced climate change. Here we use low-order deterministic and stochastic dynamical systems (DS) to understand the fundamental processes driving such relationships and ask how robust they might be, how important are radiative feedbacks, and what basic physical process are driving them. To answer this, we use the DS and CMIP5 (Coupled Model Intercomparison Project Phase 5 of the World Climate Research Programme) results to address the following questions: Why is the range of natural climate variability so large in CCMs in the first place, and what is the role of radiative feedbacks in this spread? What is the tightest relationship we should “expect” to find between climate sensitivity and climate variability on annual, decadal or multi-decadal timescales? How important are climate radiative feedbacks in this relationship? Would we “expect” to find a closer relationship between variability and transient climate response than equilibrium climate sensitivity, and what do the CCMs say? It is highly important using new methods to process and analyse climate data in our study.

 


Biography:

Sergei Soldatenko was graduated from Military Aerospace University (MAU), St. Petersburg with MSc degree in Atmospheric Sciences. In 1983 he got the PhD from the same university, in 1992 the Doctor of Science degree in Math and Physics from the state university of St. Petersburg. Since 1990 to 1999 he was a Chairman of the Department of Meteorology at, Hydrology and Geophysics at MAU and then for a number of years he was a Lab Head at the Institute of Computer Science of the Russian Academy of Sciences. Since 2011 he is a senior professional officer at the Australian Bureau of Meteorology.

Data-Driven Identification of Coal Reservoir Families

Dr Tauqir Moughal1, Dr  Irina Emelyanova1, Dr Marina Pervukhina1

1CSIRO Energy, 26 Dick Perry Avenue, Kensington, Australia

 

Introduction
Identification of coal reservoir families is important for the prediction of gas in place and potential sweet spots. This task is challenging especially for ultra-deep coal seams. In this study, we propose an approach for identifying coal reservoir families which is fully data-driven and based on hierarchical clustering of target coals.

Study Area and Data
The Cooper Basin is selected as a study area which is a world-class unconventional gas resource located in the South Australia and Queensland. A large amount of information about the physical parameters of the deep coal seams from 1194 wells is available. We proceeded with ten relevant target coal characterisation parameters (DCCPs).

Methods
In order to identify the coal gas reservoir families, we develop a decision tree (DT) for partitioning the original dataset into clusters that can be interpreted as reservoir families. At each node of the DT, the quality of the cluster is evaluated and clustering is conducted.  The partitioning process stops when the cluster quality is acceptable. The cluster quality is assessed statistically through a cluster validity criterion.

Results
Ten meaningful clusters are generated to identify coal reservoir families in the Cooper Basin. Furthermore, these clusters are analysed and interpreted by a geologist.

Conclusions
A hierarchical clustering approach has been applied to identify coal reservoir families in the Cooper Basin. The method can assist geologists in detection and mapping of potential sweet spots. This application demonstrates how traditional practices of reservoir modelling can be improved using machine learning.


Biography:

Tauqir Moughal is a data scientist with significant experience in machine learning. He is a member of CSIRO Geoscience Data Analytics team, a team of computational scientists who develop and implement data-driven techniques in various geoscience applications.

The Australian Climate Science Data-enhanced Virtual Laboratory

Ms Clare Richards1, Dr Ben Evans1, Dr Kate Snow1, Mr Chris Allen1, Mr Matt Nethery1, Ms Paola Petrelli2,5, Ms Claire Trenham6, Mr Aurel Moise7, Mr Sean Pringle1, Dr Claire Carouge4,5, Mr Scott Wales3,5, Mr Tim Erwin6

1NCI Australia, Acton, Australia, 2University of Tasmania, Hobart, Australia, 3University of Melbourne, Parkville, Australia, 4University of New South Wales, Sydney, Australia, 5ARC Centre of Excellence for Climate Extremes, , Australia, 6CSIRO, Aspendale, Australia, 7Bureau of Meteorology, Docklands, Australia

 

The Australian Climate Science Data-enhanced Virtual Laboratory (DeVL) focus has been to further develop the Australian research environments and data management capabilities for Australia’s contributions for the Coupled Model Intercomparison Project Phase 6 (CMIP6). This includes a) major upgrades and improvements to data organisation of the Earth Systems Grid Federation Node at NCI b) upgraded data analysis environments at NCI c) improved data FAIRness and d) updated user support information.

The Climate DeVL is a collaborative project between NCI, CSIRO, the Bureau of Meteorology, the ARC Centre for Climate Extremes and co-funded by the Australian Research Data Commons. The Climate DeVL builds on previous Australian e-infrastructure programs to support national and international collaboration on priority climate research across these partners, and other government programs.

The project has several areas of focus:

  • Data management for improved use within a High Performance Data (HPD) environment.
  • Access to international best-practice tools and common workflows for enhanced collaboration and innovative digital research methods.
  • Easy access to new climate data for use across a number of different research areas both directly at NCI as well as remotely.
  • Data publication services that support a range of access methods for communication and translation of research outputs with other research communities, for policy development, and industry.
  • Training and user support materials for more effective use of new capabilities.

While this project addresses the Community’s immediate goals for CMIP6, the need to extend this capability to prepare for future technical challenges is already emerging.


Biography:

Clare Richards is an experienced project manager for major research infrastructure and modelling software projects including engagement with internal and external stakeholders to develop project plans for implementing new business systems, applications, services and websites. Before joining NCI she was with the Australian Bureau of Meteorology. Clare holds a B.Sc. degree in Applied Physics from the University of South Australia, a Master’s degree in Media Practice from the University of Sydney and an MBA from La Trobe University.

Feature engineering for catalysis, material in-between periodic and molecular systems

Dr Julia Melisande  Fischer1, Ms Michelle  Hunter2, Dr Marlies  Hankel2, Prof Debra  Searles2,3, Dr Amanda  Parker1, Dr Amanda  Barnard1

1Data61 CSIRO, Docklands , Australia, 2Centre for Theoretical and Computational Molecular Science, Australian Institute for Bioengineering and Nanotechnology, The University of Queensland, Brisbane, Australia, 3School of Chemistry and Molecular Biosciences, The University of Queensland, Brisbane, Australia

 

Most machine learning research on chemical systems study either materials, with periodic structures, or molecules. Often for surface catalysis, the relevant information is in-between both systems. For this study, our dataset contains molecules on different surface defects, namely catalytic active centres. These are N-doped graphene pores supporting one or two single metal atoms.

The binding energies (BE) is an indicator for the strength of the interaction of the molecule with the surface and is used as the label. The data includes around 1600 structures of molecules on various different active centres and their BE. The geometric data comprises of the 3-dimensional coordinates of all atoms in a periodic cell. The main problem is describing features which are meaningful for a wide number of structures, as different structures have different atom types, adsorbed species and number of atoms. Through different descriptions of the geometrical structure in the systems, we try to link the BE to essential geometric features.


Biography:

Meli Fischer completed her Bachelor and Master of Science at Ulm University, Germany. She specialised on physical and theoretical chemistry with an emphasis on material science. For her further studies, she received an international postgrad scholarship from the University of Queensland and graduated with a Doctor of Philosophy in chemistry in 2018. Since January 2019, she is working on material data science at Data61, Docklands.

Deep learning based yield estimation and fruit quality prediction platform

Dr Dadong Wang1, Dr Mark Thomas2, Dr Everard Edwards2, Dr Ryan Lagerstrom1, Dr Changming Sun1, Dr Stephen Gensemer2, Dr Maoying Qiao1, Dr Rizwan  Khokher1

1CSIRO, Sydney, Australia, 2CSIRO, Adelaide, Australia

 

Accurate yield estimation at early stages of growth is critical to the competitiveness of Australian agriculture industry. For example, in viticulture, yield estimation is currently undertaken by the growers in a manual, destructive, and spatially sparse sampling regime. This method is inherently limited and typically results in ~33% error rates with the following factors as contributing to the variability: bunch number (~60%), berry number (~30%), and berry weight (~10%). Being able to robustly and accurately estimate the bunch number and berry number would be of major benefit to the Australian industry.

We are developing an AI based platform for yield estimation and fruit quality prediction. We have done some research on the yield estimation by using deep learning to detect inflorescences and fruit sets from video footages acquired at vineyards in SA and the results look very promising. We are also using machine learning for analysis of hyperspectral data for fruit quality prediction and classification. We would like to further develop these technologies into a generic platform for various fruits including grapes, avocados etc.

Once such a platform is developed for the grape industry it could be applied to other horticulture fruit crops, and the application to fruit quality is not only in the field to determine when to harvest but also throughout the supply chain from farms to consumers.


Biography:

Dr Dadong Wang is a Principal Research Scientist & the research team leader of CSIRO Quantitative Imaging Research Team, part of the CSIRO Data61. The research team has extensive experience and a track record in developing intelligent end-to-end imaging and image analysis solutions and via collaboration with innovation partners bringing them to the market. One of the solutions has been licensed to a world leading cellular imaging company and about 1,500 licenses have been sold so far, giving research laboratories around the world access to the most advanced image analysis algorithms developed by the team.

Dadong is also an adjunct Professor at the University of Technology, Sydney (UTS) and a Conjoint Associate Professor at the University of New South Wales (UNSW).

 

Machine Learning for Rapid Material Characterisation

Mr Alex Pitt1, Mr Paul McPhee1, Dr Chad Hargrave1

1CSIRO, Pullenvale, Australia

 

Accurate characterisation of the microscopic structure of mined resources such as coal is fundamental to understanding its utility and environmental impact. Such information is also critical to understanding the provenance and potentially hazardous nature of dust, sediments and other environmental samples. Such microscopic analysis typically requires an expert petrographer or environmental scientist to characterise the samples manually, a time consuming and expensive process.

CSIRO has developed an automated Component Grain Analysis (CGA) system to reduce the time required to segment and characterise coal and dust samples. The system manipulates 300 GB sample images with sub-micron pixel resolution, using automated processing to resolve this data intensive task. CGA provides reliable statistics on the distribution of maceral types and impurities in coal samples, and component materials in dust samples.

A recent breakthrough development is the incorporation of machine learning (ML) algorithms to the complex task of particle segmentation. Specifically, convolutional neural networks (CNNs) have been employed due to their demonstrated successes in the semantic segmentation of natural images, their capacity for learning texture, their computational efficiency, and their suitability for distributed computation. State-of-the-art CNN models were trained on microscopy images and ground-truth labels (provided by CSIRO petrographers) and consistently converged to segmentation accuracies on validation data of more than 95% – exceeding the estimated noise in the expert labelling.

Further steps for the development of the ML system include hyper-parameter refinement, label noise reduction, and model augmentation for the automatic characterisation of component materials and intra-particle segmentation for complex particles.


Biography:

Alex Pitt is a software engineer in the Energy division at CSIRO. He graduated from the University of Queensland in 2011 with a B.E. (Software), and worked on data-engineering products for Microsoft in Redmond, Washington. After returning to Australia, he joined the CSIRO where he has worked for the last 6 years on sensor integration, signal processing and computer vision.

 

Extracting Meaningful Features from Early-Science Radio Data

Matthew Alger1,2, Prof. Naomi McClure-Griffiths1, Dr Cheng Soon Ong3

1Research School of Astronomy and Astrophysics, The Australian National University, Canberra, Australia, 2Data61, CSIRO, Acton, Australia, 3Research School of Computer Science, The Australian National University, Acton, Australia

 

Early-science data from the Australian Square Kilometre Array Pathfinder (ASKAP) are coming fast. Wide-area radio projects such as the Evolutionary Map of the Universe (EMU) and the Polarisation Sky Survey of the Universe’s Magnetism (POSSUM) already have terabytes of ASKAP observations for use in early-science and survey planning. We have applied unsupervised machine learning methods to learn a meaningful representation of these early-science observations and demonstrated that this representation generalises across different sets of observations. We use this representation to address physical problems such as polarised source characterisation and physical model-fitting. Our approach provides a way to use early-science data even without full understanding of the unique instrumentation effects brought to the table by ASKAP.


Biography:

Matthew is an astrophysics student at the Australian National University and Data61. They are interested in applications of machine learning to wide-area radio surveys, with a focus on the upcoming EMU (continuum) and POSSUM (polarisation) surveys to be conducted with the Australian Square Kilometre Array Pathfinder.

Advanced Use of Jupyter Notebooks – Working with Open Geoscience Data

Dr Carina Kemp1, Dr Frankie Stevens1, Ms Ingrid Mason1

1AARNET, Sydney, Australia

 

DESCRIPTION

This workshop is targeted at researchers and research support specialists interested in accessing and working with open geoscience data.   Workshop participants will learn about a range of data sources for open geoscience data, and want to know techniques of accessing and processing that data using a Jupyter notebook.   The workshop is targeted* towards researchers using AARNet’s CloudStor service (with Jupyter notebooks integrated)

SCHEDULE

The Advanced Use of Jupyter Notebook workshop is ~3 hours (including breaks) and will cover:

  • Where open geoscience data can be sourced. 30 minutes
  • How to access and process open geoscience data using a Jupyter notebook. 45 minutes
  • What different programming languages and techniques can be used for data processing. 30 minutes
  • Advanced programming of an open geoscience dataset using a Jupyter notebook. 30 minutes
  • Wrap up. 10 minutes

REQUIREMENTS

  • Workshop preparation: Come with a laptop.
  • Workshop breakdown: Workshop is half-day, includes a hands-on component.
  • Workshop requirements: Wifi for ~40 attendees, no special seating.

Workshop content is relevant to all means of accessing and using Jupyter notebooks.


Biography:

Dr Carina Kemp leads the AARNet eResearch team and engagement in Australian eResearch community.  She has also worked as a research scientist, a company geophysicist and a consultant to industry. Her enduring interests include Open Data, Data Analytics, Machine Learning, and High Performance Computing.

 

Introduction to Jupyter Notebooks – for Research Support Specialists

Ms Ingrid Mason1, Dr Frankie Stevens1

1AARNET, Sydney, Australia

 

DESCRIPTION

This workshop is targeted at data librarians, research support, and eResearch professionals interested in: what the Jupyter notebook does; how the notebook works; and how the notebook can be used to train researchers in basic programming skills; and some key research data management and research collaboration considerations (e.g. higher education, government and industry partnerships).  Workshop participants will want to know where the Jupyter notebook fits into different researchers’ toolkits (along with Excel, SPSS, Stata, RebExr, RStudio, or MatLab).  The workshop is targeted* towards researchers using AARNet’s CloudStor service (with Jupyter notebooks integrated).

SCHEDULE

The Introduction to Jupyter Notebook workshop is ~3 hours (including breaks) and will cover:

  • What a Jupyter notebook is and how it functions. 30 minutes
  • Where a Jupyter notebook fits in different researcher toolkits and workflows. 45 minutes
  • When/why to use a Jupyter notebook on a desktop and in the cloud. 30 minutes
  • Basic Python programming using a Jupyter notebook. 30 minutes
  • Wrap up. 10 minutes

REQUIREMENTS

  • Workshop preparation: Come with a laptop.
  • Workshop breakdown: Workshop is half-day, includes a hands-on component.
  • Workshop requirements: Wifi for ~40 attendees, no special seating.

Workshop content is relevant to all means of accessing and using Jupyter notebooks.


Biography:

Ingrid Mason (Deployment Strategist), Dr Frankie Stevens (Research Engagement Strategist) for AARNet, are eResearch specialists with extensive experience in researcher engagement, training, and have expertise in research data and technologies across HASS and STEM research areas.

3D material micro-structure characterization and properties modelling

Dr Sam Yang1, Mr Clement Chu1, Dr Tony Murphy1

1CSIRO Manufacturing, Clayton, Australia

 

This presentation is a overview of our recent development in data-constrained modelling (DCM) methodology for quantitative and sample-non-destructive (SND) characterization of 3D microscopic composition distribution in materials, and microstructure-based predictive modelling of material multi-physics properties. Potential impacts are illustrated with examples in a range of R&D disciplines. Data and computational challenges will be presented.


Biography:

Dr Sam Yang (PhD in Statistical Physics) is a Principal Research Scientist in CSIRO Manufacturing BU. He has invented data-constrained modelling (DCM) in 2007 and has been leading its development. DCM has been applied for quantitative microstructure characterization in a range of disciplines worldwide. He has also researched on fundamental and applied statistical physics, applied complex system science, bio-physics, computational and simulations sciences, and has published over 100 research papers in refereed journals, conference proceedings, book chapters and patents. Dr Yang is leading a major project in non-destructive quality evaluation of additively-manufactured (AM) metal components funded by the Australian Department of Industry, Innovation and Science (DIIS) and five global industries.

ABOUT AeRO

AeRO is the industry association focused on eResearch in Australasia. We play a critical coordination role for our members, who are actively transforming research via Information Technology. Organisations join AeRO to advance their own capabilities and services, to collaborate and to network with peers. AeRO believes researchers and the sector significantly benefit from greater communication, coordination and sharing among the increasingly different and evolving service providers.

Conference Managers

Please contact the team at Conference Design with any questions regarding the conference.
© 2017 Conference Design Pty Ltd