from: "C G Kilsby" <c.g.kilsbyatXYZxyzcastle.ac.uk>
to: "Phil Jones" <p.jonesatXYZxyz.ac.uk>
Phil - below Lenny's abstracts at AGU.
See you tomorrow.
Lenny Smith - Applications and limitations of using large state of the art climate ensembles for risk management
How do we communicate what we know to people in the energy, insurance and finance industries? Business decisions look 1, 3, 5 years into the future; energy and insurance looks further ahead.
1. Head to head probabilistic hindcast comparisons of GCM simulations and statistical extrapolation would be of great value
2. All climate is local
3. The role of initial condition uncertainty
At what space and time scales do we have decision relevant information? Is a 40 year window too long? What should we be using for some local variable of interest - 5 years?
Global mean temperature is irrelevant to industry: all climates are local.
Once we get down to the regional scale there is not a lot of difference between 3�C and 5�C climate sensitivity - i.e. pdfs overlap as there is a lot of natural variability. This is especially the case for precipitation: natural variability dominates the signal. For decisions made on the ground there is often significant overlap: global mean temperature is a poor index variable for local decisions!
When will natural variability be swamped by climate change? Used a 64 member initial condition ensemble of HadSM3. CP.net range of change is enormous depending on initial conditions.
UKCIP08 claim to be able to provide "probabilities for 2080 on hourly 5km scales" - can anyone provide 5km hourly information for the 2080s? Can we do this if there is not credible support for the science behind it? The timescale for losing credibility as a climate scientist is not 2080 but rather whenever we upgrade the probability generator!
See LA Smith 2002, 2003, 2000
Lenny Smith - Questioning the relevance of model-based probability statements on extreme weather and future climate
Climate change has stimulated much work at the science-industry interface. Therefore, there is a huge need for expectation management and sanity checking.
UKCIP08 weather generator claims to be able to provide hourly data for the 2080s. But does our current understanding support claims to deliver information like this? Numerical decision-makers already see a paradox with the variability in AR4 global models. Climate science benefits when it communicates its limitations more clearly, so users can see progress in science as a good thing.
Are pdfs decision-support relevant? Utility (probabilistic similarity) does not require a perfect model, just one that is fit for purpose. We should not sell what we have now as the basis for making decisions as they are not true probabilities.
Who poses the greatest risk to climate science credibility? The climate scientists.