Develop numerical methods for modeling through collaborations

Numerical models are the basis for assessing, understanding, and forecasting the complex interactions between our environment and human activities. The increasingly ambitious goals of NCAR’s models of the Earth System, the Sun, and the Sun-Earth System motivate CISL’s scientific research on algorithms, numerical methods, and techniques for parallelizing and optimizing computations. CISL focuses on the numerical algorithms and computational science that will accelerate the simulation rate of Earth System models by crafting scalable numerical algorithms that can take advantage of large numbers of processors and coprocessors.

A priority in geophysical modeling is to increase resolution because higher resolution can resolve important processes to improve the accuracy of prediction and perhaps uncover unexpected interactions within the physical system. However, increasing resolution is very computationally expensive because it scales slightly faster than the cube of the resolution improvement.

CISL research strategy focuses on ways to numerically increase the effective model resolution without sacrificing the methods’ scalability or computational efficiency. The concept of effective resolution is distinct from a computational grid’s lattice spacing because it also considers the solution quality given by the numerical method – i.e., the fidelity of the flow features that it delivers in practice. However, there is a tension between the complexity of a particular numerical method and its efficiency on computer architectures. The strength of CISL’s research program is that it combines excellence in numerical methods development with computational optimization and parallelization expertise. For example, important improvements in simulation speed can be achieved by combining better time integration algorithms with innovative computer science optimization techniques.

Many of these new modeling strategies arise as basic scientific research that begin on idealized problems and are later transferred to more complex ones to meet the practical requirements of fully realized production community models.

This work is supported by NSF Core funding.