<The Oil Drum is Terribly Two Today The Oil Drum DrumBeat: March 23, 2007>

278 comments on Does the Hubbert Linearization Ever Work?

Comments can no longer be added to this story.

| Show without comments | PDF version

WebHubbleTelescope on March 22, 2007 - 10:22pm Permalink | Subthread | Parent | Comments top

I just used the non-dog variation on the Logistic model:
http://mobjectivist.blogspot.com/2007/03/derivation-of-logistic-growth-v...

This is the "drunk looking for his car-keys under the lamp-post" problem. Somebody asks the drunk why he chose to look under the lamp-post. "Of course, that's where the light is".

But the dog variation is actually more funny and perhaps more telling.

Comments can no longer be added to this story.
Slippery Slope on March 22, 2007 - 11:58pm Permalink | Subthread | Parent | Comments top

WHT

Perhaps you could post an article here describing your Shock Model. In particular I would appreciate your view of recent Saudi production with regard to your model, assuming that they are pumping to the capacity of refiners to accept their crude oil.

Comments can no longer be added to this story.
memmel on March 23, 2007 - 1:21am Permalink | Subthread | Parent | Comments top

The only reason I'm not convinced the shock model is right is I'm not sure we have enough data to support it. I think it is probably a better model but on the same hand given the data a simple analytic model suffices. I think you could drop even drop the
logistic curve and simply use parabolas. It would be nice if you picked a simple model of your choosing that you did like so we could see how the shock model varied against that.

Here you basically do that.

http://mobjectivist.blogspot.com/2007/02/quardratic-linearization.html

And here you compare to a Gaussian. Isn't a Gaussian good enough for our purposes ?

http://mobjectivist.blogspot.com/2007/01/missing-link.html

Next you added in discovery for the US but KSA for example has had many fields that were discovered yet not produced because of technical problems.

Again in general your right and I agree its just hard to see
if your model has to many parameters.

I actually favor a Gaussian because it fits what I know about movement of molecules through a matrix so it scales nicely a sample of the matrix all the way up to the field.
So you don't have to change equations. Because of the noise in the data I did not see that the choice makes a lot of difference.

I'd say the best approach is a Gaussian first approximation then move to something like the shock model for more complex cases. In general the only reason to model is when your within 20% of peak on the front side as say 10% on the backside. So I'm not sure we needs a super good model.

With that said I'd love to see you model Russia.
Thats one that HL basically fails on and probably any other simple analytical model. And if you do what I said and pick a simple model this would be a good way to show when it fails.

Comments can no longer be added to this story.
Robert Rapier on March 23, 2007 - 3:01am Permalink | Subthread | Parent | Comments top

The only reason I'm not convinced the shock model is right is I'm not sure we have enough data to support it.

Yet you latched on to the HL and continue to insist that it has value, even as case after case falls upon scrutiny. Now, scanning down the responses I can see that you have been reduced to aspersion casting. You still haven’t produced a case that works, but insist that I have demonstrated nothing. It is like someone who insists there are pink unicorns living in their house. I go from room to room, and I find no pink unicorn. But you insist that the burden of proof is on me to show that the unicorn is not there. Actually, the burden of proof is on you. Show me the unicorn.

Thats one that HL basically fails on…

Show me one that it didn’t basically fail on.

Comments can no longer be added to this story.
memmel on March 23, 2007 - 11:37am Permalink | Subthread | Parent | Comments top

Robert its a model. In this post you already showed it works for KSA in your examples of how it does not work.

The 2000 production data gave a good fit and when you went to 2006 the fit was worse and the URR changed but the model indicated that it had more noise. The good thing about HL is its noise or error term is realistic.

So you already posted a great example of HL working if you understand modeling.

So what is the model hinting at after 2000 ? Its says that a lot of "new" production was brought online but yet KSA did not
find some large new field during this time period.

So what happened ?

They had spare capacity in 2000 by 2006 this was gone either it was rotated out or depletion is catching up since prices are still high depletion is a good possibility.

HL cannot predict when someone changes the extraction pattern abruptly but it does show in the data and the fit as noise. Its also giving us a estimate on how much spare capacity the used to have. And we have confirming statement from KSA that say they where pumping at capacity in 2005.
If you had bothered to read on the noise distribution in HL you would have realized that a increase in the noise itself
means something disrupted the model since the noise goes to zero. And your using the URR estimate without a error term
if you showed the real error in URR you would see that most of what happened through 2006 was simply to increase the error in the estimate the error terms probably still overlap.

I'm comfortable with your KSA plots and believe since KSA was a confirmed swing producer and the model shows that they have brought all their excess capacity online they have already peaked from the HL data.

This post is junk science at its worst.

Comments can no longer be added to this story.
garyp on March 23, 2007 - 3:02am Permalink | Subthread | Parent | Comments top

Its a different metaphor. One is about using the lamp post as a tool to help, rather than trusting it to do all the work - the other is about looking to find answers only where you have the data.

Lots of lamp post metaphors.

Ahh well, looks like nobody is going to come back on the meat of the suggestion. Pity really, I think systems dynamics and working from the data we really do have is a promising approach. Maybe I'll have to do it myself one day.

Comments can no longer be added to this story.

278 comments on Does the Hubbert Linearization Ever Work?

Comments can no longer be added to this story.

| Show without comments | PDF version