Data Jam, EPFL, November 2017
In November 2017, two ACE projects took part in a Data Jam organised by the Swiss Data Science Center at EPFL. Scientists with datasets and problems to solve came together with data scientists, visualisation experts and programmers, to combine their expertise and promote collaboration across disciplines.
Project: Uncovering the mystery of the ocean’s false bottom
Camille Le Guen, a PhD student at the University of St Andrews took part in the Data Jam with the aim of creating a script to detect aggregations of krill and try to extract some key characteristics of krill swarms (length, height, depth, biomass).
Following presentations by each of the six teams providing data and a problem, groups formed to tackle the problems in question. An introduction to Jupyter notebooks which the teams could use to create and test code, was very useful for those who wanted to use this handy tool.
A collaboration between this research group and Benjamin Ricaud is now ongoing to extend the analysis over the rest of the ACE dataset.
Project: Observing interactions between winds waves currents and ice
The aim of Alberto Alberello, a Post-Doc at the University of Melbourne was to combine WaMoS II marine radar data and ship motion data to derive a transfer function that allows the estimation of wave conditions from the ship motion only.
A team composed by Alberto Alberello, Antoine Ratouis and Charles Antoine Kuszli was quickly set up and started working on the data. The team members had a background in ocean engineering, shipping industry and meteorology. The team worked together to write numerical code based on a simplified version of the floating body equations. Due to the complexity of the equations involved in the problem only one degree of freedom (heave) of the ship was considered. This simplification limited the accuracy of the method.
A brief report of the challenge and the codes developed during the Data Jam Days were made available on a public repository on GitHub (https://github.com/charlesantoine1/ACE-WAVE). Note that the code developed during the Data Jam Days are not final, they are first attempt to tackle the problem of reconstructing the wave spectral properties from the ship motion.
Polar Data Architecture workshop, Geneva, November 2018
In November 2018, around forty representatives from the polar community came together for technical discussions on how to advance Polar Data Management at the workshop co-convened by the Arctic Data Committee, the Southern Ocean Observing System and the Standing Committee on Antarctic Data Management. Jen Thomas, the ACE Data Manager, participated in discussions that were focussed on addressing common data infrastructure and some more domain-specific systems.
Following a day of presentations on current work, the workshop was split into two sections, one of which was “centred on achieving federated search through the exchange of standardised, well formatted discovery metadata”, which would essentially allow all polar datasets to be searched for through one central location. The focussed on architecture design, particularly on data interoperability.
The metadata workshop continued working towards recommendations for best practise for polar data managers regarding the use of metadata, and continues to map between existing metadata schemas.
For more information about this workshop, including the list of organising and involved parties, see here. A draft report of the meeting is also available.
Further work will continue along similar lines at the third Polar Data Forum to be held in late 2019.
Machine Learning for the Environmental Sciences, Zurich, January 2019
In January, Jen Thomas, the ACE Data Manager and Sebastian Landwehr, a Post-Doc working at the Paul Scherrer Institute on the ACE-DATA project attended the Machine Learning for Environmental and Geosciences Conference, held in Zurich, Switzerland. Being quite new to the topic, the day-long tutorial was a very helpful introduction which can be followed here: https://github.com/langnico/MLEG_tutorial
The second day had talks about applications of machine learning in the environmental sciences. Examples ranged hugely from robotics, to image recognition. Below are a few examples of topics related to ACE projects.
— Working with machine learning on satellite data and imagery to look at landscapes —
Many groups demonstrated examples of using freely available satellite data to look at properties of vegetation and land-use. The work done by the Ecovision group, led by Dr Jan Wegener at ETHZ, is one such example, where they are measuring vegetation height from Sentinel-2 data. Further work in this lab is helping to understand how to count things on a satellite image that are much smaller than a pixel! Bettina Weibel also uses machine learning techniques to describe the landscape, and Lauren Zweifel from the University of Basel is using machine learning to identify small landslides.
— Making predictions —
Miriam Guillevic from EMPA is looking at creating an early warning system of pollution, whilst Jonas Bhend from Meteo Swiss is using machine learning techniques to improve the accuracy of weather forecasts. Other work being undertaken by Michael Sprenger from ETHZ is looking to forecast specific weather phenomena such as the föhn winds.
— Oceanography —
One speaker used neural networks and observations of CO2 concentration across the global oceans to produce an estimate of the seasonal and inter-annual variation in the Atlantic Ocean carbon sink: https://doi.org/10.5194/bg-10-7793-2013.