Wednesday, January 02, 2019

Predictions

Tim Harford says that the act of making predictions makes for better people (based on this paper). I've always enjoyed making predictions so I suppose I should be pretty wonderful by now. Hmmm...well in fairness he was only suggesting an association not a guarantee. In the hope of improving myself a little further, I offer the following:

  1. Brexit won't happen (p=0.95).
  2. I will run a time (just!) under 2:45 at Manchester marathon (p=0.6).
  3. Jules and I will finish off the rather delayed work with Thorsten and Bjorn (p=0.95).
  4. We will also submit a highly impactful paper in collaboration with many others (p=0.85).
  5. 2019 will be warmer than most years this century so far (p=0.75 - not the result of any real analysis).
  6. The level of CO2 in the atmosphere will increase (p=0.999).


16 comments:

jules said...

Not (2) surely? I guess (sorry - I mean, "predict") 2:46 to 2:47. :? :o Although, if he goes off at 2:45 pace and then fails, it could be sooo much worse... ;-)

James Annan said...

Magic shoes :-)

Though it's far from certain. Perhaps I should have attached probabilities.

James Annan said...

ok probabilities added

Pieter Zijlstra said...

Nr 6
Why not p = 1.0000

William M. Connolley said...

And the spiral stairs?

jules said...

Oh, I'm optimistic about the spiral stairs! P(0.8) But if he doesn't reciprocate by making the workshop neat (I've only been asking for 4 years, but he's finally agreed, hurrah!) I might deliberately load the spiral stairs up with special obstacles designed to trip up excessively tall men! :-)

James Annan said...

I usually think of logarithmic scoring rules, where there is no real advantage in a scoring system going from p=0.999 to 1 for an event that happens, but a huge (infinite) problem going from 0.001 to 0 for an event that happens! What about a big global war that wipes out fossil fuel production (to the point at which ocean uptake exceeds production)?

https://en.wikipedia.org/wiki/Scoring_rule

David B. Benson said...

Remain!

crandles said...

A big global war would use lots of fossil fuels or it would take lots of fuel bunkers and production facilities being blown up to prevent this. So perhaps a more likely scenario would be terrorist action blowing up almost all refining facilities worldwide?

Whatever the scenario causing huge production plunge, would ocean uptake also be reduced quite considerably?

Finally, if we are entering weak El Nino episode that normally results in larger than normal CO2 rise.

Think I am more inclined to go for p=0.99999 but then I have advantage of 12 days production, not forgetting all worldwide stocks ...

crandles said...

>"2019 will be warmer than most years this century so far (p=0.75 - not the result of any real analysis)."

A glance would seem to suggest temp per gisstemp loti would have fall by about 0.17 to 0.18C. We have had bigger drops like 1981 to 1982, 1991 to 1992 and 1998 to 1999. Obviously 1999 is coming off super El Nino year which we didn't have in 2018. Pinatubo June 1991 and El Chicheon late March 1982 volcanoes seem likely cause of the other two. Seems like a major cause is needed but there are other years with slightly smaller falls that with a bit of extra noise/luck/... could reach the threshold. If anything, we are getting a weak El Nino developing which should make it less likely.

Other temperature indexes may be a little different but I don't see why the probability isn't well over 90% rather than ~75%. Again that is just a glance rather than a real analysis.

James Annan said...

Yes you're probably right about the temps. I didn't want to suggest it was a sure thing but at a glance it certainly seemed likely. As for CO2, what's a few more 9s between friends? Any disruption would only have to (very roughly) halve anthropogenic production of CO2, as the amount going into the ocean wouldn't change (that just depends on the large imbalance between atmos and surface ocean concs, not the emissions on the annual scale)

crandles said...

Perhaps I have a model in my head which is just wrong: Upwelling water, whether 20 years or 1000 years since last on surface would continue to absorb over 95% of what would be absorbed had emissions continued as normal. However most absorption is by surface waters staying on the surface and with this, if the typical time to equilibrium timescale is say a month or less then there isn't very much catching up to do once emissions cease.

I could easily have got this hopelessly wrong, so feel free to enlighten me.

James Annan said...

I haven't modelled this myself but I do think you are wrong. The upwelling continues indefinitely, the skin that rapidly reaches equilibrium is mixed down and replenished by unsaturated water. Just as for warming, really. If we stopped all emissions then the atmos conc would start to drop, quite rapidly at first (2ppm/y) and then gradually slowing as most of the emitted CO2 ends up in the ocean at equilibrium (but this will take O(1000y) or even longer). See Ken Caldeira's work.

crandles said...

Trying to plug a few numbers in:
Assume 50m of mixed surface water depth, 1000 year time frame for overturning circulation, and average ocean depth of 3688m per wikipedia.

Then upwelling water in one year seems to represents about 7% of the 50m mixed surface water. (7% is more than I expected.)

Ignoring biology effects which might not be a sensible assumption,

7% of surface waters absorbing at a CO2 concentration differential of 410-280 ppm
versus 93% of surface waters absorbing at a CO2 concentration differential of 2.3ppm

then it looks like the upwelling water absorbs around 4.5 times the amount by surface waters staying on the surface.

So my error could well be the "most absorption is by surface waters staying on the surface" part of my answer.

Re timeframe for surface waters remaining on the surface, I am thinking a couple of minutes to equilibriate the top 0.01m (partial pressures are fast acting) then that gets mixed with 5000 times the quantity so the timeframe will be more than 2 min * 5000 so a week doesn't seem enough but my suggested month sounds to be of the right sort of order so I am inclined to stick with that.

This could still be hopelessly inaccurate but suggests that only around 20% of the ocean sink varies/disappears with the emissions. I guess land sink would slowly disappear over a couple of decades. But the big one, being 80% of the ocean sink only disappears very slowly over about 1000 years.

Yes, you were always more likely to be right than me ;)

jules said...

Any updates on 1 and 2? :-)

James Annan said...

Someone's been checking my training...despite today's poor effort, 2 is still somewhere in the highly uncertain range I'd say. I remain cautiously (perhaps stupidly) optimistic :-)

As for 1, nothing has changed.....yet. If brexit wasn't going to happen, this is one of the plausible pathways. Ie the likelihood P(obs|no brexit) seems quite high to me.

If crandles has any specific scenarios to test as regards emissions, that would be easy now I've found (and coded up) the Myhrvold and Caldeira equation. Though it's only a simple equation that might not work that well for extreme cases.