date: Mon, 07 Jun 2004 10:15:34 -0600 from: Tom Wigley subject: tuning to: Sarah Raper Sarah, Very interesting. So the issue is whether RLO(t) *should* be constant in time for step forcing (noting that, if RLO(t) is constant then sensitivity is constant). [I use RLO(t) here to distinguish the RLO that we input from the backed-out transient land/ocean temperature ratio.] It seems from the Gregory plot that you think RLO(t) *is* constant in time in at least one AOGCM -- although what you say is 'pattern of warming is constant' which is more general. Am I interpreting you right here? But some GCMs have non-constant sensitivity. So, presumably, these would also have non-constant RLO(t) ???? My physical intuition says that, in general, RLO(t) should *not* be constant in this type of experiment. The processes that affect RLO(t) are the input RLO, the input XKLO (and to a lesser extent XKNS). Greater RLO-in can be offset by greater XKLO. One can easily quantify these effects in a 2-box model. However, we do not know these things a priori so we cannot a priori say whether RLO-in wins or XKLO wins or they balance. What we can say is that there is no a priori reason why they should balance since they involve separate and largely independent physics. For the PCM case that you show the XKLO that gives RLO(t) constant is 16, which seems to be a very high number. This is for RLO-in = 1.4. If you used a smaller RLO you would get a smaller (and more realistic?) XKLO to give constant RLO(t). So perhaps the PCM RLO we use is wrong? I am not sure how we can 'fix' this -- i.e., how we can back out XKLO and RLO independently. I presume that if we are just concerned with global-mean temperature it doesn't matter -- there must be a whole range of XKLO/RLO-in combinations that will work. One can't use ocean heat flux (or expansion) because this seems to be pretty much independent of the XKLO/RLO(t) choice. My guess is that fitting both land and ocean temperatures won't solve this problem, but I could be wrong. Perhaps SO4 forcing cases may help, given the big spatial differences in forcing. Have you looked at all the CMIP 1% CO2 model runs to see how RLO(t) varies? I think you only had those 20-year average results when you did this before, and it would be better to use full time series. I can easily do this for the 50+ PCM experiments Very interesting. Tom.