Alexis Praga
PhD Thesis : a large-scale chemistry-transport model for massively parallel architectures.
Pangolin resources
What is my PhD about, in layman's terms
Modeling the atmospheric chemical composition is critical for weather
forecasting and climate simulations. It can provide useful information for
crisis such the Fukushima Daiichi nuclear disaster. In this case,
radioactive particles were released in the atmosphere and transported
outside Japan. Simulations of radionucleide activity can be extremely
valuable for government officials in their management of the situation.
As a stunning illustration, you can see the animation resulting of an
atmospheric dispersion modeling from the CEREA laboratory .
Of
course, it can also be used to predict the evolution of other atmosphere
particles, such as volcanic plumes during the Eyjafjallajökull eruption.
How are we going to simulate the transport ? It can be represented mathematically as two equations. Unfortunately, there are no exact solutions for them so we have to do some approximations. First, we apply a grid over all the Earth. Then, for each cell of this grid, the equation will be transformed to a simpler, numerical formulation.
Yet there are several ways of choosing the grid and the scheme (how the equation is approximated in the cells). And each possibility has its own pros and cons so there is no perfect choice. Now a common grid is this one :
However, at the North pole, you can see the meridian are converging, leading to very small cells. For some schemes, it can hurt their performances, especially for parallel architectures. That is why we have developped a grid for which all cells have the same area.
Another important aspect for simulations is performance. Major
models now run on massively parallel machines, which are equivalent to
several thousands of desktop computers stacked together. It allows for
better performances, for example finer resolutions leading to more
accurate results. But it also requires a special approach, which may not
be compatible with all models.
So my PhD is about developping an model which run efficiently on parallel
architectures. Using a custom grid, it will provide accurate 3D simulations on very
fine resolutions, up to 80 chemical species.