Speaker: Howard Barker, Environment and Climate Change Canada
High-resolution cloud models, such as LESs, CSRMs, GCRMs, and some NWP
models, apply 1D radiative transfer (RT) models to all columns in a
domain (the Independent Column Approximation, or ICA). When the number
of columns is large (e.g., in the millions), this can be expensive; even
when the RT models are called only once every ~30 dynamics time steps
they still account for ~30% of the cloud model's entire CPU time. This
seminar reports on an alternate to the ICA method that reduces the
number of ICA RT model executions by factors of well over 100. Like a
"lossy" file compression algorithm, however, this technique loses much
radiative information, but, as will be shown, when a well-known cloud
system-resolving model is run in radiative convective equilibrium-mode,
only minor differences, few being statistically significant at the 95%
confidence level, exist between the control simulation, using the ICA
method, and the experiments using the new method with 100, 500, and
1,000 times fewer calls to the RT models. This suggests that RT models
could be called 'every' dynamical time step and yet still have CPU time
devoted to RT be less than having the ICA call the RT models after
numerous dynamical time steps.
