Atmospheric Chemistry Modeling, or Earth System Modeling in general, brings up a lot of computational challenges. Recent advances in computer technology have greatly enhanced our capability in simulating our earth. I am really excited about applying cutting-edge computational and mathematical techniques in atmospheric research.
I have a broad interest in model development, numerical algorithms and software engineering:
Projects in model development
GCHP (The High-Performance Version of GEOS-Chem)
GCHP is the next generation of GEOS-Chem, with the ability to perform global simulations at arbitrary resolutions using any number of CPUs. I have been developing GCHP since 2015, and now it is relatively mature and ready for scientific analysis.
FlexGrid GC (GEOS-Chem custom nested simulation)
GEOS-Chem currently has a quite limited grid-nesting capability -- only few domains (China, North America and Europe) are available, and there are a lot of redundant codes for different domains. In fact, it is not difficult to set up a new nested run (see the steps I wrote), but getting the required meteorological data is still a trouble.
I have built a new functionality to regrid and crop metfields online through HEMCO, which allows users to set up a new nested domain without touching the source code. It is expected to be available in v11-02.
Projects in numerical algorithms
Vertical resolution and tracer transport
Thanks to the growing computing power, atmospheric models can now run at unprecedentedly high resolutions. However, while the horizontal resolution of global models has changed from ~100km to ~10km or even ~1km during the past years, the vertical resolution is seldom touched. I am testing the potential benefit of higher vertical resolution using the GFDL-FV3 model.
Accelarate atmospheric chemistry solver
Speed-up the KPP solver by various numerical techniques to make atmospheric chemistry affordable in earth system models.
Projects in software engineering
Cloud-GC (Running GEOS-Chem on cloud computing platforms)
This exiciting new technology will make the model much more accessible to everyone, with a potentially much lower price compared to local HPC clusters. See the tutorial here.
GCPy (A free, new python tool to replace IDL/gamap)
I am the original developer of the GCPy package. The initial purpose was for GCHP benchmarking, but it is soon going to be developed jointly by a larger commnunity in order to completely retire the expensive IDL language.