Robert, a good article that points up some of my own misgivings about the reliance on HL. The wise words "don't use models the way a drunk uses a lamppost - use them for illumination, not support" comes back to me. The moment you rely without thinking and questioning is the moment you make that BIG mistake.

To me the whole thing is very much a case of 'the central limit theorum' - if you know nothing more then you can assume that the combination of independent distributions tends to a normal distribution (eg 'Peak Oil'). However we DO know something more - production is NOT independent and production levels/demand are significantly influenced by politics, prices, technology, social events, etc. Although to a first order it will be OK, we know the model cannot be taken beyond that because of all the influences upon it.

Does that mean peak oil is a joke (hello CERA)? No, the basics are still there. Oil is a finite resource and production will increase with increased usage, and decline with increase problems of production (smaller, harder wells). However to get to a useful prediction of reserves, dates, and most importantly, decline rates, we need a much better model - particularly since we have to make up for hidden data.

Some try to consider fields individually, summing the resultant predictions to arrive at a global production forecast, bottom up. Some also try to capture planned developments (megaprojects) to predict future production. Neither really deal with the key issues the confuse our view of production/reserves and thus forecast - the human dimension.

It seems to me the only way to try and apply a model to the data to make predictions is to 'reverse out' the effect of the human dimension on production figures - constructing an idealised production history from the raw data by removing the human perturbations (from history, which we DO know accurately). From there we can use three more pieces of information; which fields the oil is coming from, the fact that any country/company will produce its 'easiest' oil first, and the fact that companies will produce oil so as to maximise their discounted return on investment.

Putting that together, it should be possible to arrive at an alternative history of production which would then be amenable to improved mathematical modeling to determine what was under the ground. We could then take real production to yield real world predictions of future production capability.

I'd suggest that a fuzzy systems dynamics model is the first approach to performing this 'reversing out' - constructing a model that predicts out history supply record with the known history, and then reversing the effects of the elements to arrive at the inputs. It sounds as if this should work, being based on the data we do know accurately (production totals, history, driving behaviours), rather than the data we don't (URR, etc.)

Your thoughts?