Start:

Friday, February 2, 2024 12:01 pm

End:

Friday, February 2, 2024 12:50 pm

Location:

STAG 113

Sean Santos

PNNL Earth System Modeling Group

ABSTRACT: Earth system models solve exceedingly complicated multiphysics problems by breaking down the Earth system hierarchically into smaller sub-models (e.g. atmosphere, ocean, land, and sea ice), which are composed of smaller components themselves. This decomposition of an Earth system model (which may require millions of lines of code in its software implementation) into many small modules is a vital part of model development. However, naïve coupling of modular physics packages using first-order methods can significantly reduce model accuracy, or even produce numerical instability. This talk covers two examples from the Energy Exascale Earth System Model (E3SM). First, we will see that “sequential” (Lie-Trotter) splitting is a major source of error for E3SM’s cloud and precipitation physics. We will discuss our evaluation of several proposed alternatives, including Strang splitting and multirate methods. Second, we will see that E3SM is prone to spurious “oscillations” in winds blowing over rough surfaces (e.g. forests). Using a simplified model of the atmospheric boundary layer, we show that these oscillations can be attributed to numerical instability arising from infrequent atmosphere-surface coupling. When using an “explicit” flux calculation method to supply the atmosphere with a lower boundary condition, we show that there is a maximum (linearly) stable atmosphere-surface coupling time step size, and we derive approximate formulas for this step size.
BIO: Sean Patrick Santos is an atmospheric scientist and applied mathematician at Pacific Northwest National Laboratory (PNNL). He received a Bachelor’s in Engineering Physics from the Colorado School of Mines, after which he spent 5 years as a scientific software engineer, and went on to obtain a PhD in Applied Mathematics from the University of Washington. As a postdoctoral research scientist at Columbia University/NASA GISS, he worked on developing data-driven schemes for cloud microphysics using Bayesian methods. At PNNL, he is a part of the PAESCAL team, working with scientists and mathematicians across the country to improve the accuracy of numerical methods in Earth system models.