Multi-time small-area estimation of oil and gas production capacities by Bayesian multilevel modeling

TL;DR

This paper introduces a Bayesian multilevel modeling framework for estimating oil and gas production capacities at small geographic areas over multiple time periods, incorporating well-specific and temporal factors with uncertainty quantification.

Contribution

It develops a novel Bayesian multilevel model that captures spatial and temporal variations in production capacities, including well-specific features and resource usage, for small-area estimation.

Findings

The model effectively estimates production capacities with quantified uncertainties.

Incorporating resource usage improves the accuracy of predictions.

The approach is flexible and robust for small-area, multi-time period analysis.

Abstract

This paper presents a Bayesian multilevel modeling approach for estimating well-level oil and gas production capacities across small geographic areas over multiple time periods. Focusing on a basin, which is a geologically and economically distinguishable drilling region, we model the production capacities of its wells grouped by area and time. Regularizing our inferences with priors, we model area-level and time-level variations as well as well-level variations, incorporating lateral length, water usage, and sand usage at each well. The Maidenhead Coordinate System is used to define uniform geographic areas, many of which contain only a small number of wells in a given time period. First, a Bayesian small-area model is built, using data from the Bakken region from February 2012 to June 2024. Then, the model is expanded to contain temporal dynamics in the production capacities. In addition to general time components, water and sand usage intensities are modeled in estimating production capabilities over time. We find the Bayesian multilevel modeling approach provides a flexible and robust framework for modeling and estimating oil and gas production capacities at area and time levels and for informing area-time predictions with uncertainties.