Yes, good to see someone here that essentially produces half the referenced and cited (and high quality) graphs concerning oil depletion.

The one pressing question I always have is how the interpolated and extrapolated smooth lines get drawn on these figures. We all know that the oil production curves tend to use the Logistic as a fitting function, but we don't have a good handle on what most analysts use for discovery curves and creaming curves. In particular I have seen several references to creaming curves being modeled as "hyperbolic" curves yet find little in fundamental analysis to make any kind of connection.

Based on statistical considerations I am convinced that the discovery and creaming curves result from a relatively simple model that I have outlined on TOD. I have a recent post where I make the connection from dispersive discovery to creaming curves here:
http://mobjectivist.blogspot.com/2008/03/creaming-curves-and-dispersive....

In the following figure I apply the Dispersive Discovery function to one of the data sets on your graph. This function is simple to formulate and it produces a finite asymptote which you can use to estimate the "ultimate discoverable" (150 GBoe for NG in the following).

If in fact you use the same formulation for your extrapolation, I would be interested to know.

But I have a feeling you share the same frustration as I based on:

"But the USGS past approach since 2000 has been too optimistic, where only one geologist is simply guessing solely the number of discoveries and the size to be forecast (seventh approximation sheet) and then a Monte Carlo run (50 000 transforms) These wild guesses result into a beautiful distribution that looks real. A better approach should be hoped for, based on the complete past data reviewed by several geologists. "

This sounds like the best model out there is some horribly complex Monte Carlo run that ends up being a SWAG. I do indeed agree that a better approach is needed and think we need a good model to start with.