Hosted by the Institute of Computing for Climate Science
Recordings of recent meetings
The cross-VESRI monthly journal club is held on the second Tuesday of each month, hosted virtually by ICCS. Members of the VESRI teams and ICCS present research papers or concepts that have been significant to their research and are of interest to the other teams. This is an opportunity for researchers to see the cross-cutting work that is being done across the VESRIs, and draw connections with their own work.
Recordings of the journal club are taken and posted online on our Youtube page; recent videos have been embedded below, with the most recent recording first.
June 2025
Use of Large Language Models in Research and Coding
Presented by Dr. Omar Jamil, Engineering Lead, Institute of Computing for Climate Science (ICCS)
Abstract: I will provide an overview of some of the ways LLMs, and associated techniques, can be used to improve the research and coding process. I will outline some of the latest tool development in this space. Finally, I will also discuss some of the issues that can come with the use of LLMs in our work.
May 2025
How Insurance Companies Quantify Natural Disaster Risk in a Changing World
Presented by Dr. Ruth Petrie & Dr. Patrick Ball, INIGO Insurance
Abstract: In 2024, economic losses from natural disasters reached $368 billion. $145 billion of these losses, or approximately 40%, were recouped by those affected through insurance mechanisms, improving the resilience of communities and aiding recovery (Aon, 2025). The largest monetary losses in 2024 were caused by hurricanes Milton and Helene which killed 278 people and generated $110 and $37.5 billion economic and insured losses, respectively. Inigo has partnered with Institute of Computing for Climate to create InSPiRe – Inigo Storm Prediction and Impact Research. This project aims to leverage the latest developments in climate science and AI to better predict present-day and future hurricane risk. In this talk, we will describe how insurers use catastrophe models to estimate the likelihood, severity and impact of natural disasters. We will also provide examples of how the insurance industry collaborates with academics to improve the understanding and modelling of these phenomena.
April 2025
Bipolar sea-saw in modern climate
Presented by Dr Ali Mashayek, Associate Professor, Department of Earth Sciences, University of Cambridge
Abstract: An energetic area of research in the climate community concerns the potential disruptions to ocean meridional circulation (MOC). Most of the work on the topic is based on modelling. I will provide a brief overview of asymmetric polar warming and its implications for modern climate change. I will then discuss a series of analyses that explore the changes in polar oceans over the past few decades, highlighting observational evidence for synchronous polar warming and disruption to upper and lower MOC branches. I will also discuss the turbulent mixing of processes within the context of MOC variability and finish by highlighting some computational and AI tools needed to explore this topic further.
March 2025
OceanBioME: a flexible ocean biogeochemical modelling environment
Presented by Professor John Taylor , Professor in Oceanography, Department of Applied Mathematics and Theoretical Physics, University of Cambridge
Abstract: In this talk I will discuss OceanBioME, a new software package building on the Oceananigans ocean model to allow simulations of coupled ocean physics and biogeochemistry. I will discuss the motivation behind the development of OceanBioME to study marine carbon dioxide removal interventions and will show an example of OceanBioME used to study flow through a giant kelp forest.
February 2025
What is Past is Prologue-- recreating the earth’s climates of the past, present, and future
Presented by Dr Allegra LeGrande, NASA Goddard Institute for Space Studies / Center for Climate Systems Research, Columbia University
Abstract: How do you reproduce the climate present day earth with a few thousand lines of FORTRAN code? How do you know if your simulations are any good at predicting the future? Simulations of the past can help vet model response and provide an out-of-sample test for your model. But pulling them together is tricky.
January 2025
Fluid: Data-Transparent Visulaisations
Presented by Dr Roly Perera - ICCS Early Career Advanced Fellow, University of Cambridge & Research Fellow University of Bristol
Abstract: Charts and other visual summaries, curated by journalists and scientists from real-world data and simulations, are how we understand our changing world. But interpreting visual outputs is a challenge, even for experts with access to the source code and data. Fluid is a new “transparent” programming language, being developed at the Institute of Computing for Climate Science in Cambridge, that can be used to create charts and figures which are linked to data so a user can interactively discover what visual elements actually represent.
November 2024
What RSE can do for Researchers
Presented by Chris Edsall, University of Cambridge, co-director of the ICCS
Abstract: Research Software Engineering is a relatively new field in the realm of research. Given that much research is built on simulation and data processing it is critical for reproducibility that these software are written correctly. In order for researchers and research groups to have impact it is a competitive advantage if their software can be reused by others in the group, collaborators, and the community. We will look at how RSE can elevate proof of concept code in to a publishable artefact in its own right. A case will be made that for a little extra cost at development time a much larger return on that investment can be reaped.
September 2024
Pure Fortran Machine Learning: Why and How?
Presented by Milan Curcic (Univeristy of Miami)
Abstract: Fortran is the oldest high-level programming language still in use. Although no longer as widely used as some other languages like C++, Python, or JavaScript, it remains relevant in the traditional HPC environments for science and engineering fields including weather, ocean, and climate prediction, molecular dynamics, and computational fluid dynamics more generally. More recently, machine learning has exploded in use and adoption in science, both for new discoveries and to accelerate traditional physics-based numerical solvers. In this talk, I will discuss: the place and role of Fortran in the world dominated by Python-based ML frameworks and heterogeneous processor architectures; recent updates to Fortran, both as a language and its ecosystem of compilers, libraries, and tools; and, the design and implementation of neural-fortran, a pure Fortran framework for deep learning.
May 2024
Datawave: The Graft-versus-host problem
Presented by Hamid Pahlavan and Ed Gerber
Abstract: In the event we successfully develop a well performing data-driven
parameterization of un- and under-resolved gravity wave momentum
transport, we do not expect it to produce desirable results when first
`grafted’ into a `host’ atmospheric model. First, atmospheric models
are capable of producing part of the gravity wave spectrum, and worse,
likely simulating it poorly; a parameterization must be calibrated to
account for what the host model does on its own. (We can turn this
question around, asking how to properly compute the gravity wave
momentum transport from a high resolution integration. There are
several potential strategies to isolate gravity waves, all which return
slightly different answers, and it’s unclear what scales count as
unresolved with respect to a global atmospheric model.) Second, many of
the sources of gravity waves (convection, frontogenesis) are themselves
not well represented in atmospheric models -- and errors in surface
winds can bias topographically induced wave production -- requiring a
calibration of the gravity wave source spectrum. Finally, success is
measured from on macroscopic behavior, e.g., a reduction in
climatological bias, or a good representation of natural variability
such as the Quasi-Biennial Oscillation. Current parameterizations serve
a dual role of representing a missing process and correcting other
biases in the host. We present a toy 1-dimensional model of the
Quasi-Biennial Oscillation that allows us to explore calibration
strategies in an idealized context. We show that offline-online
learning may provide a route for calibration, though its computational
feasibility in a full 3D model remains an open question. Even here,
however, we worry whether too much calibration defeats the initial
purpose of using a data-driven parameterization to accurately represent
gravity waves: does a calibrated scheme get the right answer for the
right reasons?
March 2024
CESM3 - Focusing on the atmospheric component CAM7 and its vertical resolution
Presented by Isla Simpson & Peter Lauritzen (NCAR)
Abstract: Isla Simpson and Peter Lauritzen, from the NSF National Center for Atmospheric Research (NCAR) of the United States of America, presented the new version of CESM at the ICCS cross-VESRI Journal Club, which took place online on the 12th of March 2024. The primary goal of the Community Earth System Model (CESM) project is to develop a state-of-the-art model and to use it to perform the best possible science to understand Earth system variability and global change. The main focus of the presentation was the atmospheric component of CESM, although other components were addressed too. The session was chaired by Colm-cille Caulfield, Head of Department of Applied Mathematics and Theoretical Physics, University of Cambridge.
February 2024
MIT's BC3 and Climate and Weather Extremes
Presented by Paul O'Gorman & Raffaele Ferrari
Abstract: Anthropogenic climate change is already leading to substantial increases in Earth’s surface temperature, and storms, droughts, fires, and flooding are becoming more prevalent, more destructive, and stronger. It is imperative that the best strategies to mitigate avoidable changes and adapt to unavoidable ones can be identified and evaluated systematically. However, climate change projections from numerical models of the Earth system remain highly uncertain and are too complex to be readily used to assess climate resilient solutions on a large scale. The Bringing Computation to the Climate Challenge (BC3) team aims to democratize access to reliable information about climate change. This requires: (1) improving the accuracy of climate projections, and (2) developing fast emulators that allow a wide range of stakeholders to query those projections.
January 2024
Introduction to the FETCH4 Project
Presented by Vasilii Petrenko, Alex Turner and Lee Murray
Abstract: FETCH₄: Fate, Emissions, and Transport of CH₄, is an international collaboration of scientists focused on improving our understanding of the past and modern methane cycle, which is a key component of the global carbon cycle. Atmospheric methane concentrations have exhibited large variations over time, and have rapidly grown since 2020, yet the drivers of these variations remain scientifically elusive. The FETCH4 team will collect data from both Greenland ice cores and air samples from stations around the world and measure their unique chemical fingerprints. Their goal is to single out individual aspects of the methane cycle, such as those coming from fossil fuel emissions, and identify both their sources and sinks. Using what they learn, the team will develop and sharpen the capability of global climate models to account for methane. Scientists hope that by creating more efficient models, which will be accelerated by machine learning, they can better interpret these chemical fingerprints and more efficiently capture the methane feedback mechanism in global climate models.
November 2023
Introduction to the CALIPSO Project
Presented by Dr Philippe Ciais, Dr Corinne Le Quére and Dr Daniel Goll
Abstract: This is an introduction to one of the new VESRI teams called CALIPSO (Carbon Loss in Plants, Soils, and Oceans). Led by research teams at the Université de Versailles Saint-Quentin-en-Yvelines, University of East Anglia, and University of Exeter, CALIPSO quantifies vulnerabilities of terrestrial and oceanic carbon stocks under climate change. By integrating novel observations, theoretical understanding, and machine learning tools, the project assesses risks associated with carbon cycle tipping points and provides refined emissions reduction strategies.
October 2023
Software Reproducibility
Presented by Dr Marion Weinzierl, Senior Research Software Engineer Institute of Computing for Climate Science (ICCS).
Abstract: Reproducibility is a fundamental principle of research, and this does not only apply to experimental and empirical research, but also to research relying on computer programs and simulations. In the past decade or so, the so-called replication crisis has led to a number of initiatives and networks which address issues with computational reproducibility. This talk explains what is meant by software/computational reproducibility and why it is important. It showcases a number of initiatives dealing with this topic, and gives guidance on how to improve the reproducibility of your research code.
ICCS is hosting a ReproHack event in March 2024 where individuals or teams try to reproduce scientific findings using code and data that has been openly published. Marion encourages VESRI researchers to submit their papers and codes in order to receive feedback on their reproducibility, and get hands-on experience with software reproducibility.
September 2023
Energy Efficiency in High Performance Computing
Presented by Chris Edsall, Co-director Institute of Computing for Climate Science (ICCS), Head of Software Engineering at the University of Cambridge.
Abstract: At the SuperComputing 2022 conference the Test Of Time Award was presented to Chung-hsing Hsu and Wu Feng for their 2005 paper "A Power-Aware Run-Time System for High Performance Computing". In 2006 they proposed the Green500, an alternative ranking of supercomputers to the Top500 that measured not only their raw compute power but also the energy used while running the benchmark. Current HPC systems can consume significant amounts of electrical energy. For example, the current number one system, while being near the state of the art for energy efficiency, still requires 22 MegaWatts of power to provide its 1.1 ExaFLOPs. As in most cases energy generation results in emission of CO2 , HPC community should consider its carbon impact. This talk will look at some of the considerations of energy efficient supercomputing.
August 2023
How sea-ice rheology should be considered as a subgridscale parameterisation of stress relaxation with the ice
Presented by Dr Einar Ólason, Research Leader at the Nansen Environmental and Remote Sensing Center (SASIP)
Abstract: Sea ice is a crucial component of the earth system and a key component in climate and earth system models. Sea ice is a highly effective insulator, so there is very little heat exchange between the ocean and atmosphere possible through sea ice and no moisture or gas exchange where the ocean is ice covered. But the ice is neither homogenous nor stationary; it drifts under the influence of wind and ocean currents, and for it to move, it first has to break. The cracks that form can let massive amounts of heat, moisture and gasses escape the ocean into the atmosphere, impacting both it and the sea below. In this talk, I will explore some intriguing properties of ice drift and deformation and relate them to how the ice breaks and how we can model this breaking with so-called brittle rheologies. We will see how we can use the concepts of fractures propagating from small to large spatial scales to inform our choices of how to model the internal ice stress in the ice. I will argue that sea-ice rheology should be considered a sub-grid-scale parameterisation of stress relaxation to be consistent with the observed spatial scaling of sea-ice deformation. The group behind the SASIP project has been working on implementing these ideas in a large-scale sea-ice model, and I will conclude by showing some of our latest results with the neXtSIM model.
June 2023
Lightweight code verification for science
Presented by Dr Dominic Orchard, Institute of Computing for Climate Science (ICCS), University of Cambridge
Abstract: This journal club talk will give an overview of two papers "A computational science agenda for programming language research" (Orchard, Rice, 2014) and "Evolving Fortran types with inferred units-of-measure" (Orchard, Rice, Oshmyan, 2015) which represents some of my earlier work looking at how programming languages, tools, and systems could be developed to support the work of computational scientists. I'll go over some principles of how tools can be used to rule out bugs in code, in particular looking at the CamFort suite of tools which we developed for scientists working with Fortran. I'll show how you could deploy CamFort in your projects to gain greater assurance and speed up development practices, and also give an idea of other available tools that you might like to consider using in other languages.
May 2023
What is Bayesian optimization and can we use it for history-matching?
Presented by Dr Henry Moss, Institute of Computing for Climate Science (ICCS), University of Cambridge
Abstract: In this month's edition of the journal club, I will be discussing Bayesian optimization. This powerful machine learning method for calibrating expensive complicated systems is a very active area of research in the ML community and has numerous successful applications across science and industry. Recently, very similar methods have been deployed to help calibrate large climate models (e.g. history matching in DataWave). After covering some core concepts, I will stress the link between Bayesian optimization and history matching. Finally, I will provide a high-level overview of recent progress in Bayesian optimization and some initial thoughts on how we could apply these ideas to help ease climate model calibration.
March 2023
Warming of Global Abyssal and Deep Southern Ocean Waters between the 1990s and 2000s: Contributions to Global Heat and Sea Level Rise Budgets
Presented by Dr Laura Cimoli, Institute of Computing for Climate Science (ICCS), University of Cambridge
February 2023
Machine Learning to Parametrise Convection in Climate Models
Presented by Simon Driscoll, University of Reading
December 2022
The challenges (and opportunities) in using ML for climate and turbulence modelling
Presented by Pedram Hassanzadeh, Rice University
The challenges and opportunities in reconciling balloon observations with gravity wave parameterisations
Presented by Francois Lott, LMD, Paris