Latest updates
  • 21/09/2022 - Added hackathons signup in programme (under Friday) and added restaurant recommendations to logistics
  • 20/09/2022 - Punting location + map added. Information about Thursday wine tasting social tomorrow at 5:30pm.
  • 19/09/2022 - Updated time of departure for walking group from Jesus College to Maths, 08:15am. Added link for signups for 1-1 chats with RSEs.
  • 18/09/2022 - Code-of-conduct added to logistics. Resources tab added.

During the week, you can book in a session with one of the RSE team for advice, or to discuss ongoing projects.

Monday 19th September, Jesus College

Due to the state funeral for the Queen which is happening on this day, we have had to re-organise and merge Monday’s scientific programme into the rest of the week. Instead, we will have a low-key, informal start with informal networking during the day at Jesus College.

Start End Event
07:30 09:00 Breakfast in Main Hall, Forum building for those staying in College
12:00 13:00 Informal buffet lunch at Jesus College Dining Room, West Court
14:00 19:00 Informal networking in the Webb Library
15:00 16:00
19:00 -

Tuesday 20th September, Centre for Mathematical Sciences (MR2)

The walk from Jesus College to the CMS takes about 25 minutes. There will be a mini-bus taxi available leaving from Jesus College at 08:30. Those wishing to walk can join Dominic at 08:15 at the Porter’s lodge to walk over, going a very scenic route through the old colleges and past the library.

Location of MR2 is in the ‘core’ part of the Centre for Mathematical Sciences, downstairs

Start End Event
07:30 09:00 Breakfast in Main Hall, Forum building for those staying in College
09:00 09:30 Welcome and Introduction ICCS directors and RSEs
09:30 10:30 DanWorkshop: Using Git and GitHub effectively Dominic Orchard, ICCS
10:30 11:00 Tea
11:00 11:30 ChrisWorkshop: Follow-up to using Git and GitHub effectively Dominic Orchard, ICCS
11:30 12:30 Opening keynote - Bringing Scale and Trust to Carbon Credits Through Computer Science S. Keshav, University of Cambridge Carbon credits–especially those derived from nature-based solutions such as reforestation or averted deforestation–are deservedly viewed as being untrustworthy and their use by airlines and oil companies a barely-concealed form of greenwashing. In this talk, I will present a solution to these issues that leverages advances in earth observation, AI, cloud storage, and blockchain. This solution is being prototyped by the Cambridge Center for Carbon Credits (https://4c.cst.cam.ac.uk ) and I will discuss the current status and our vision for the future.
12:30 13:30 Lunch
13:30 14:20 DominicScience talk (SASIP) Multi-scale sea ice and ocean modeling. Why and why now? Chris Horvat, Brown University Significant changes to Earth’s climate are most prominent in the polar regions — especially in the Arctic, where surface temperatures have risen by up to 3x the global mean. In turn, the decline of Arctic sea ice, land ice, and permafrost has ushered in a new status quo for local food webs, peoples, and climate. New under-ice ecosystems and chemical exchange, alterations to transportation and ways of living, and disrupted atmospheric and oceanic variability are all features of the emergent “New Arctic.” In spite (perhaps because) of this, climate models have repeatedly failed to capture these changes, so doubts loom over whether forecasts of Arctic and Antarctic change can be relied upon in the future. Here I’ll discuss several newly observed features of the Arctic coupled system and how many are driven by a similar quality: the fragmented and fractal nature of the Arctic sea ice cover, which contemporary modeling frameworks largely ignore. I’ll discuss efforts through the Scale Aware Sea Ice Project to observe, diagnose and rectify problems associated with the incorrect treatment of Arctic sea ice. These are led by (1) improved resolved-scale modeling of fragmented sea ice through the neXtSIM sea ice model, (2) new scale-aware parameterizations of ocean turbulence, waves, and air-sea exchange that drive polar change in climate models, and (3) new observations of sea ice and ocean variability for assimilating into cutting-edge forecast models.
14:20 15:05 Science talk (DataWave) Towards an improved understanding and representation of atmospheric gravity waves Aditi Sheshadri, Stanford University Atmospheric gravity waves (GWs) are ubiquitously excited on the Earth and are critical drivers of the atmospheric circulation, however, they present a challenge to climate prediction: waves on scales of 102-105m can neither be systematically measured with conventional observational systems, nor properly resolved in atmospheric models. I will describe recent work in my group aimed at understanding the effects of gravity waves on climate variability and improving their representation in GCMs. a) We have leveraged high-resolution data from balloon flights launched by Loon LLC, originally deployed for internet access. The opportunistic Loon dataset, though not from a scientific campaign, gives us access to thousands of balloon flights with measurements of position, pressure, and temperature from which we have inferred statistics of gravity wave motions in the lower stratosphere. b) We have developed a machine learning GW parameterization, coupled it to a global climate model, showed that it is stable and accurate when run online, and that it reproduces features of the climate that depend critically on GWs. c) I will describe recent results with regard to calibration and uncertainty quantification of a popular gravity wave parameterization.
15:05 15:30 Tea break
15:30 17:00 JacobWorkshop: Testing Chris Edsall, ICCS
17:00 17:40 Short talks 1
      • ML-based emulators of sea-ice models - Charlotte Durand, École des Ponts

  •Sensitivity Analysis and Machine Learning of a Sea Ice Melt Pond Parametrisation, Simon Driscoll, University of Reading In this study we seek to understand and characterise the sensitivity of the state-of-the-art sea ice column physics model, Icepack, to its level-ice melt pond parametrisation and see if machine learning can learn and replace this parametrisation. A global sensitivity analysis of all its melt pond parameters indicate that parameters controlling the amount of melt water allowed to run off to the ocean (and in particular that meltwater added to the melt ponds in early melting season) plays a substantial effect on the sea ice properties.
We perform simulations of the Icepack model forced by hourly data from the Climate Forecasting System Version 2 dataset, over a range of Arctic locations, and show that neural networks demonstrate the ability to learn and predict output given by the level-ice melt pond parametrisation, and furthermore do not suffer from drift or instability when used in the Icepack model replacing the melt pond parametrisation itself. With uncertainty around the precise values of many sea ice parameters, our work opens the possibility of, for example, applying hybrid data assimilation and machine learning techniques that have been used to incorporate direct (often spare and noisy) data to infer unresolved scale parametrisations, in sea ice models.
17:45 18:30 Taxis to dinner
18:30 -

Wednesday 21st September, Centre for Mathematical Sciences (MR2)

Start End Event
07:30 09:00 Breakfast in Main Hall, Forum building for those staying in College
09:00 10:30 DominicWorkshop: CI and GitHub actions Ben Orchard, ICCS
10:30 11:00 Tea
11:00 11:30 JacobScience talk (LEMONTREE) Relationships between resprouting and fire regimes Yicheng Shen, University of Reading Resprouting is a resilience trait that allows individuals to regenerate rapidly following fire. It has profound effects on the speed of post-fire ecosystem recovery and therefore on water- and energy-exchanges with the atmosphere and the carbon cycle. However, the ability to resprout requires investing in carbon storage. Balancing the benefits of rapid recovery of photosynthesis against the costs of carbon storage implies that resprouting is an optimal behaviour in environments where fire is neither too frequent nor too infrequent. Although there is anecdotal support for this assertion, there has been little quantitative investigation of the types of fire regime where resprouting is an optimal strategy. In this study, we use data on the abundance of woody species in Europe and Australia derived from the sPlotOpen database combined with information on whether the species present can resprout or not, derived from regional and global plant trait databases and field information, to examine how changes in the abundance of resprouting species varies with fire return interval and with fire intensity. We show that the proportion of resprouting species decreases as fire return intervals increase, while the abundance of resprouters is maximal at intermediate levels of fire intensity. This work suggests that it should be possible to model the occurrence and abundance of resprouting using an eco-evolutionary optimality approach based on balancing the costs and benefits of resprouting under different fire regimes.
11:30 12:30 Workshop: Bridging Fortran and Python for ML Athena Elafrou and Simon Clifford, ICCS
12:30 13:30 Lunch and poster presentations (posters will be displayed in the lunch area).
13:30 15:00 DanAutomating forward and inverse geoscientific simulation in Firedrake. David Ham, Imperial College London Creating simulations of continuous systems, such as the ocean, atmosphere, or cryosphere, usually involves numerically solving partial differential equations. Creating these numerical solvers is a complex task that requires the composition of the right differential equation with suitable discretisations, parametrisations, solvers and preconditioners. In addition, many geoscientific simulation challenges are inverse problems which require the numerical solution of the adjoint PDE. Combine all of this with increasingly sophisticated parallel computing systems and creating geoscientific solutions becomes complex, labour-intensive and error-prone. Here I will present a radically different alternative. The Firedrake system allows model developers to express the simualation they wish to conduct in a high-level mathematical language embedded in Python. High performance parallel implementations are automatically generated and the solution returned. Parallelisation and the evaluation of adjoint simulations are fully automated. Here I will present a short overview of the Firedrake system before we move on to a practical hands-on demonstration using Jupyter notebooks in the cloud. Participants should bring their laptops and be ready to participate. No software installation will be required.
15:00 15:30 Tea
16:30 17:30
19:00 22:00

Thursday 22nd September, Centre for Mathematical Sciences (MR2)

Start End Event
07:30 09:00 Breakfast in Main Hall, Forum building for those staying in College
07:00 - Optional 5k guided run along the river in Cambridge, starting from Jesus College. 25-30 minute pace with Dominic, and a leisurely 45+ minute pace with Marla for non-runners.
09:00 10:30 DanWorkshop: Training ML models Will Handley, University of Cambridge
10:30 11:00 Tea
11:00 11:30 JacobWorkshop: Questions and followup to Training ML models Will Handley
11:30 12:30 Science talk (CliMA) Exploring parallel programming in Julia Valentin Churavy, MIT Parallel programming is required to solve large scale computational models in climate science. In this session we will explore the fundamentals of parallel programming with MPI and GPU, as well as performance engineering in Julia. The goal is to provide an intuition of what approaches for parallelism are out there and how one could apply them in their own work. We will use the Julia programming language to explore these concepts, but no prior knowledge of it is required.
12:30 13:30 Lunch
13:30 14:10 DominicShort talks 2
      • Chaotic and forced variability in the North Atlantic Eighteen Degree Water - Olivier Narinc Recommendations laid out for CMIP6 allow climate models to use a resolution of 1/4° in the ocean. At these scales, chaotic variability is introduced in the ocean, as evidenced by the presence of large oceanic eddies. Observational studies of the North Atlantic Subtropical Mode Water (STMW) have found that not all of its variability can be explained by atmospheric variability. The STMW is a water mass formed in the winter mixed layer south of the Gulf Stream, and is the most abundant T,S class of water in the surface North Atlantic. Consequently it plays a key role in air-sea exchanges over the basin. These elements have motivated the present model investigation of the STMW’s ocean-driven (intrinsic) chaotic variability using a NEMO-based, 1/4°, 50-member ensemble simulation of the Northern Atlantic ocean. The ensemble is generated using a realistic, stochastic parameterisation of subgrid noise. The model is assessed against the ARMOR3D ocean reanalysis, based on in situ data collected from ARGO floats and satellite observations. Using this dataset, six STMW-wide integrated variables are defined and analysed: total volume, and averaged potential vorticity, depth, temperature, salinity and density. The water mass’ chaotic variability is estimated from the time-averaged ensemble standard deviation, and is compared to the total variability estimated from the ensemble mean of the temporal standard deviations of all members. Initial results show that chaotic variability is significant for STMW properties at interannual timescales, representing almost half of the total variability of its average temperature. This suggests that as climate models move towards finer spatial resolution in the ocean, oceanic chaos can be expected to introduce more variability at interannual and longer timescales. This study also highlights the existence of an unknown uncertainty caused by chaotic variability in the ocean.
  • Tailoring data assimilation for discontinuous Galerkin models - Ivo Pasmans Satellites provide observational coverage of sea ice motion, coverage and thickness, often at resolutions finer than the grid resolution of the current generation of sea ice models. Data assimilation (DA) aims to combine these observations with model output to arrive at a better estimate of the true sea-ice state. As part of the Scale Aware Sea-Ice Project (SASIP), a new sea ice model is being developed using a discontinuous Galerkin (DG) approach to solve the model equations. Contrary to the current generation of finite-difference and finite-volume models, DG models are able to resolve the model solution on a subgrid scale. In this work, we explore the possibility to exploit this ability of DG models to improve the DA using an idealized 1D model and Ensemble Transform Kalman Filter (ETKF). In particular, we look at 1) the benefits of using higher-order polynomial interpolation over linear interpolation in the observation operator, 2) the possibility to assimilate multiple observations per grid cell instead of averaging all observations in a grid cell into one superobservation as is the current practice, 3) the idea to remove spurious long-distance correlations stemming from the use of a limited-size ensemble by constructing a scale-sensitive DG localisation operator. A more detailed description of these ideas and some preliminary results will be presented.
14:10 15:00 Science talk (M2LInES) Software and Infrastructure for Data-Intensive Climate Science Ryan Abernathey, Columbia University Physics-informed machine learning for climate modeling is data-intensive; before any ML can begin, many terabytes of observational or model data often must be processed in order to prepare appropriate training data. Performing this work and effectively sharing data and code in a large, collaborative project is challenging. In this talk, I will give an overview of different open-source software and data infrastructure components used in the M2LInES project and the broader Pangeo community, including xarray, zarr, xgcm, xbatcher, Pangeo Forge, commercial cloud computing, and Open Storage Network. I will conclude with a future vision for a decentralized data platform for collaborative data-intensive science.
15:00 15:30 Tea
15:30 17:00 ChrisWorkshop: Pairing and code review Ben Orchard, and Dominic Orchard, ICCS
17:00 17:30 Hackathon pitches and introduction
17:30 18:30 (in MR4) with additional time for extra poster presentations

Friday 23rd September - William Gates Building, Intel Lab

Make a Hackathon pitch

Join a Hackathon team please add your name and e-mail (and mark if you are a virtual participant by putting a capital V after your name).

Start End Event
07:30 09:00 Breakfast in Main Hall, Forum building for those staying in College
09:00 10:30 Hackathon group work
10:30 11:00 Tea
11:00 12:30 Hackathon group work
12:30 13:30 Lunch
13:30 15:00 Hackathon group work
15:00 15:30 Tea
15:30 16:00 Finishing up and preparing short presentation
16:00 17:00 Group presentations
17:00 17:15 Short break
17:15 17:30 Prize giving and closing remarks
17:30 19:00

Show session chairs