You say "less risky" but does either method produce a probability with its forecast? Or does probability remain a secondary, fuzzy, human, meta-judgement?

This is a really good point.

The simplest model which HAS worked for all commodities in the past in an exponential model.

If Q = total fraction of resource recovered then exponentional growth is given by

dQ/dt = AQ and hence Q = exp(At) where A is some constant, a single parameter fit as khebab says.

Well the interesting thing is if you form the ratio dQ/dt/Q and plot it as function of time you should get
dQ/dt / Q = A - a constant as a function of time.

This is the well-known, (to TOD readers), Hubbert Linearization plot and rather than being a simple constant, shows a decrease as a function of time for all mature reqions, including the whole world.

A better approximation is to use

dQ/dt = AQ*(1-Q)

Which gives a Bell-shaped curve for dQ/dT and is the basis of many predictions for near term peaks in world Oil production, including my own.

However this is also an approximation and the real-world will almost certainly show some deviations from this. As I said before, if I had to bet on future world Oil production I'd go with Khebab's loglet analysis, although even that is not currently sensitive to new technologies that would come into play with Oil prices sustained above $60 Barrel.

In any case, Oil consumption accounts for approximately 40% of world CO2 emmissions, gas accounts for 20% and coal accounts for aboout the remaining 40%. Even if we have ways to increase Oil production in the future we should not it.

We should use our Fossil fuel resources to transition to new ways of providing the wealth and energy we want and need.