Generalized Extreme Value Statistics, Physical Scaling and Forecasts of Gas Production in the Barnett Shale

We develop a method of predicting field-wide gas (or oil) production from unconventional reservoirs, using the Barnett shale as an illustration. Our method has six steps. First, divide a field of interest (here Barnett) into geographic/depositional regions, where -- upon statistical testing -- gas and/or oil production are statistically uniform. Second, in each region i, fit a generalized extreme value distribution to every cohort of gas/oil wells with 1,2,…,n_i years on production. Third, obtain accurate estimates of uncertainties in the distribution parameters for each regional well cohort. As a result, obtain n_i points for the stable mean (P₅₀) well prototypes for each region i, and the corresponding high/low (P₁₀/P₉₀) bounds on well production. Fourth, by adjusting the producible gas/oil in place and pressure interference times between the adjacent hydrofractures, fit each statistical P₅₀ well prototype with a physics-based scaling curve that also accounts for late-time external gas inflow. The physics-scaled well prototypes now extend 10-20 years into the future. Fifth, for each region, time-shift the dimensional, scaled well prototype and multiply it by the number of well completions during each year of field production. Add the production from all regions to match the past field production and predict decline of all wells up to current time. These well productivity estimates are more accurate and better quantified than anything a production decline curve analysis might yield. Sixth, by assuming different future drilling programs in each region, predict field production futures. We hope that the US Securities and Exchange Commission will adopt our robust, transparent approach as a new standard for booking gas (and oil) reserves in shale wells.