Marysa Laguë is an assistant professor and climate scientist in UBC’s Department of Geography where she leads the Computational Climate research group. To study how changes in the land surface impact both local and global climate, she creates and employs sophisticated climate models using Sockeye, UBC’s central high-performance computing (HPC) platform.
Dr. Laguë's research program simulates real and imagined scenarios, such as reconfigurations of Earth’s ocean and continental arrangements (e.g. creating a world where the entire northern hemisphere is land and the southern is ocean), or massive changes in vegetation (e.g. removing an entire forest region to see how cloud cover is affected) in order to improve our understanding of both past and future climates, help scientists better predict the impacts of environmental change, and inform strategies for climate adaptation.
What inspired your research focus?
I did undergraduate and master's degrees in mathematics — mostly in theoretical math and number theory — and just thought math was cool. I took a lot of earth science classes for fun and when I realized you can use math to describe what's happening in the atmosphere, I got sucked in!
How does your research have impact?
The math we use in our climate modelling is mostly differential equations — fluid dynamics, motions in the atmosphere, energy conservation — and we stuff them all into a model, and basically generate a toy earth in a computer that we can manipulate. We can spin the earth backwards, move mountains and even continents around to see what happens.
These things obviously aren’t possible in real life, but exploring those really idealized, hypothetical questions helps us improve our understanding in a way that is really useful in applied contexts.
Think about making a big perturbation to the vegetation on the land surface: planting a large number of trees. It will cause increased flux of water, energy and carbon between the surface and the atmosphere, which will have a direct impact on surface climate. More evaporation makes it cooler locally, but likely also results in a change in cloud formation, which in turn lead to a change in precipitation — whether localized, or father away. We try to understand the immediate direct impact of the change, and how it impacts long-term and indirect changes.
Through understanding the role of changing land evaporation and the effects of controlling the water vapour greenhouse effect, we can help to predict future climate states, and advise on measures that may help to mitigate the negative impacts of changes to environments and climate systems.
How do you know that the models are accurate?
There's definitely some sanity checks that go on when we're building the models! Some of it is basic non-contentious physics and thermodynamics, but there is also a lot of observational data that is used to validate the models, and even inform how we represent physical processes. There are really complex processes that we can’t necessarily represent using fundamental physics equations – either they are too computationally expensive, or we don’t fully understand the physics yet. Instead, we represent these processes using parameterizations — equations trained on and able to replicate real-world observational data. Together, these give us confidence in our ability to put these models into truly wacky configurations and trust that the resulting data will be accurate to how our modern Earth might respond.
Sometimes I get a really unexpected result and think, “Oh, I must have set this up wrong,” but actually I just found a new physical pathway that I hadn't been thinking about before.
How do you use ARC in your research?
We do most of our data analysis on Sockeye and if ARC didn't exist, I don't know what I would be doing. These earth system models are very complicated computer programs, comprised of millions of lines of code and requiring specialized software packages, processors and compilers in order to function. Moving models from one supercomputer to another one is a massive pain, but when I came to UBC in 2024, I needed my climate models to migrate with me. I tried a bit on my own, but ran into a lot of challenges. The people at ARC put in quite a lot of work to sort it out and I'm so grateful because otherwise I literally wouldn't be able to do any of the climate modeling I’ve been talking about!
We also use ARC to store and backup our data — supporting the storage of 100s of terabytes of data. This is incredibly valuable because it is computationally and time-intensive to produce, so we don't want to have to recreate any of it.
How much time does it take to complete a simulation?
This is a fascinating question.
For an average model it might take a week to determine the parameters I want to test across, then I could set it up in a couple of hours and have it running in Sockeye and off it would go. It would run for a week or two before it produced enough data to look at meaningfully.
But time has a weird meaning on these high-performance computing systems. We measure the amount of time a simulation costs (or takes) in terms of what we call core hours or core years. For example, if I did the simulation on just one computer, to run a 50-year long simulation might take 20,000 core years, which is about 50 years. But using Sockeye I can get the answer in a week!
What’s next for your research?
One of the modelling tools we’re beginning to explore is called a large eddy simulator, which is basically a really high-resolution model that can capture the swirly whirly processes that go into forming a cloud.
In our current modelling the clouds are what you might call low, or coarse-resolution and actual clouds are way smaller than the resolution they present at in the models. One of the things I want to investigate is how changes in vegetation are actually modifying cloud cover, by explicitly resolving those processes in a way that our current global models can't.
I also want to dig into what I call the orphaned airspace, that neither the current land or atmosphere models are simulating. Historically these models were developed by different teams — land modellers and atmospheric modellers — but, plants and trees are tall and can intersect both those regions and the science that defines those relationships is not yet being explicitly represented.
Global warming and carbon dioxide-driven climate change have a big warming signal. I want to develop simulations of the airspace within forest canopies to explore how that signal is experienced by a complex forest canopy, and to explore the impacts of different levels of heat stress. For example, if the temperature varies by even five degrees between the leaves at the very top of the canopy and the foliage below, is that going to be equally bad for all critters and all plants trying to take up carbon at different heights and levels within the canopy? Is the degree of climate change going to vary, depending on where they are in the canopy?
ABOUT UBC ADVANCED RESEARCH COMPUTING (ARC) SOCKEYE
The ARC Sockeye platform is available free of charge to the UBC research community. It provides nearly 16,000 CPU cores and 200 GPUs to support researchers across all disciplines. Projects that require advanced research computing often involve large datasets, high-performance computing, modelling, or data visualization that exceed the capabilities of standard computing infrastructure. For more information or to apply for an allocation, visit the UBC ARC website.