Biological and Environmental Research - Earth and Environmental System Sciences
Earth and Environmental System Modeling

Implementation, Optimization, and Science Tests of Photolysis in CAM5

Monday, May 12, 2014 - 07:00
Add to Calendar

Implementation of the new Fast-J photolysis code into CAM5 at LLNL enables full interaction between chemistry, aerosols, and clouds in climate modeling. Fast-J is now being put onto the CESM trunk in a form that will allow WACCM to employ a mixed wavelength J-code adding Fast-J rates (for wavelengths >200 nm) to the original WACCM lookup tables truncated to <200 nm. These longer wavelengths reach the lower-middle stratosphere and troposphere where clouds, aerosols, and geo-engineered particles alter the photochemistry, and thus Fast-J capability is necessary. This new hybrid approach allows for more accurate SRM geo-engineering studies. Update of Fast-J to include sub-grid cloud fields in a full flexible format (Cloud-J) is complete as a standalone code and within the UCI chemistry-transport model. The Cloud-J code will replace various hard-wired, simplistic approximations for clouds with a range of options, including approximations more consistent with CAM5 heating codes. In tests at UCI, Fast-J operates at a fraction the cost of the chemistry module; nevertheless, the chemical-solver codes may find new approaches to optimization, and approaches for more realism in the cloud fields will lead to several times more radiative transfer calculations within in a single column grid-cell. Thus UCI has profiled the choke-points in the code (primarily the block tridiagonal solver) and implemented Fast-J on a CPU/GPU system (Nvidia Tesla C2070). Fast-J solves the radiative transfer problem 18 times (wavelengths) in each air column, but this is inefficient on GPUs. For Cloud-J, the number of air columns per cell will rise to about 50, still not enough. If 80 grid-cell columns are grouped with one CPU/GPU, then the GPU could take on 4,000 air columns and the speedup over the CPU (Core i7) is 10x, with possible 18x from further study of load balancing and more direct use of CUDA. Taking the chemical model capability we have developed and LLNL and UCI for CAM5, we examine an important photochemical uncertainty -- that of the molecular oxygen cross sections in the Hertzberg continuum (200-240 nm) -- and its impact on climate. Photolysis of oxygen in this wavelength region is the source of ozone in most of the lower stratosphere and thus determines the heating rates across the tropopause as well as from tropics to extra-tropics. We varied the cross sections by +-30% (the laboratory 90%-confidence range) and found differences in stability and wave propagation, but little impact on the residual circulation.

Presentation file(s):