Arps' deterministic decline curve analysis has been the industry standard for forecasting production and estimating reserves despite its subjectivity and potentially large uncertainty, particularly early in well life. Probabilistic decline curve analysis (PDCA) methods have been proposed to quantify the significant uncertainty in reserves estimation in shale gas plays. However, all the published PDCA methodologies have been tested using Arps' decline curve model. New decline curve analysis (DCA) models have been developed for hydraulically fractured horizontal wells. However, these models have been primarily applied deterministically, without quantification of uncertainty.
We performed hindcast tests of six DCA models presented in the literature (Arps, Arps with a minimum decline, modified Arps, Power Law, Stretched Exponential, and Duong) with the Markov Chain Monte Carlo (MCMC) probabilistic methodology with different amounts of production data. Our results show that the MCMC probabilistic method reliably quantifies the uncertainty in production hindcasts and cumulative production estimates for decline curve models developed for shale gas wells. Even with DCA models based on Arps' equations, the probabilistic methodology is reasonably well calibrated. Hindcast uncertainty decreases as the amount of matched production data increases, but the MCMC probabilistic method is reasonably well calibrated regardless of the amount of production data matched. Furthermore, the MCMC probabilistic method yields P50 estimates that are more accurate than deterministic estimates at early times.
Uncertainty will always be present in production forecasts and reserves estimates and uncertainty can be quite large early in the producing lives of shale gas reservoirs. Reliable quantification of uncertainty can improve decision making in early stages of production, which can lead to more efficient exploitation of these reservoirs.