Group 4 Summary (Monday)

(1) What can/should climate models do?

Some evidence that climate models are fairly good at reproducing
statistical properties of extremes indices aggregated involving
multiple sites (e.g., such as those used by IPCC).

Should not directly compare station data with model output because of
scale issue, etc. Research problem to determine how to perform such
comparisons.

Climate models are fairly good at reproducing large-scale features.
Although still need for improvement in modeling El Nino phenomenon.

Only make direct use of output from climate models about such
large-scale features, not any information about local or regional
extremes. Indirectly derive other information through
statistical downscaling. Such downscaling should be performed in terms
of
distributions. But not clear yet how to obtain such distributions.
Because this downscaling exercise is uni-directional, may need to
consider
"upscaling" as well.

(2) Is extreme value theory useful in this context?

Could live without extreme value theory, but we are considerably
better off with it.

It remains unclear under what conditions extreme value theory is
applicable to climate
output?

It does not appear straightforward to apply multivariate extreme value
theory in realistic climate applications. But this theory would be
potentially of use to model complex climate extremes such as heat
waves that depend on a number of variables (temperature, humidity,
etc.)