The next meeting of the Dutch Petrophysical Society will be held online on Thursday 4th March, The meeting will be conducted using GoToMeeting facilitated by the SPWLA.

The theme of the meeting is:

Bayesian Methods in Petrophysics

With two talks:

Talk 1 – Petrophysical Evaluation of Thinly Laminated Depositional Sequences using Statistical Matching Procedures
Joachin Ambia, 3D UTAPWeLS Development Team, University of Texas at Austin.

Talk 2 Application of Probability Pruning by Filtering in Petrophysics. Is It Artificially Intelligent?
Mirano Spalburg, Consultant.

Abstracts for the talks are below.

A link to register via SPWLA will be posted shortly. Alternatively register your interest below by entering your name and email.

We will also be announcing the candidates for the DPS board elections. All positions are up for election, and interested members of the DPS are invited to express their interest by email to info@dps-nl.org.

Positions up for election and current incumbents are listed below:

• President – Iulian Hulea (Shell)
• Secretary – Danijela Krizanic (Independent)
• VP Technology – Chris Harris (Independent)
• VP Communications and Publications – Tom Bradley (Baker Hughes)
• Treasurer – Paul Mast (SODM)
• YP Representatives – Abdul Hamid (EBN), Jan-Bart Brinks (Schlumberger)
• Board members X3 – Marisa Sitta (Wintershall), Simon Smith (Halliburton), Igor Kim (Shell), Tracey Flynn, Tarek El-Taraboulsi (Schlumberger), Morgane Bizeray (Baker Hughes)

Applications close on 3rd March 2021 so that candidates can be announced in the Marh meeting. The election will be conducted online in May with the new board being announced at the AGM in June. Note that to be a DPS board member, you have to be a full member of the SPWLA.

Talk 1 – Petrophysical Evaluation of Thinly Laminated Depositional Sequences using Statistical Matching Procedures
Joachin Ambia, 3D UTAPWeLS Development Team, University of Texas at Austin.

Abstract

Conventional petrophysical evaluation techniques are unreliable to assess individual bed properties in laminated depositional sequences with beds thinner than the vertical resolution of standard logging tools. The main cause of this limitation is that well logs average formation properties across multiple interbedded intervals. Commonly used solutions are limited to shaly sandstone models to account for either presence of grain-coating clay in sandstones or laminated shale-sandstone systems. Other solutions rely on volumetric techniques which require subjective interpretation of shale and total porosity concentrations. Likewise, it is usually assumed that both sandstone and shale properties remain constant within siliciclastic reservoirs, which is not always the case in heterolithic bedding or in laminated sequences with strong diagenetic alterations.

To address this technical challenge, we introduce an inversion workflow that reproduces measurements via analogues of thinly laminated reservoirs. We use a Markov-Chain Monte Carlo inversion algorithm to generate independent realizations of each petrophysical property. All petrophysical properties are combined to estimate probability histograms rather than reducing them to a single value for each petrophysical property.

The method is applied to a deep water heterolithic clastic sequence of grain-coating clay sandstones where bed thickness varies from 3 to 4 inches. In addition to conventional well logs, high-resolution borehole images are used to detect bed boundaries. The statistical method is used to estimate total porosity, water saturation, and permeability based on the earth-model-derived properties. Finally, net-to-gross and hydrocarbon pore volume are estimated using the calculated statistical properties.

Compared to conventional interpretation procedures, the formation evaluation method described here enables the incorporation of non-constant matrix and shale properties in the sandstone-shale laminated sequence, and estimates individual layer properties and their uncertainties, thereby reducing subjectivity in the interpretation of static and dynamic petrophysical properties of heterolithic clastic sedimentary sequences.

Biographical Details of the Speaker

Joaquin Ambia is currently the software development lead for the Formation Evaluation Joint Industry Research Consortium at the University of Texas at Austin. He holds a BS degree from the Instituto Tecnológico de Estudios Superiores de Monterrey in Mexico and a PhD in physics from the University of Houston.

Talk 2 Application of Probability Pruning by Filtering in Petrophysics. Is It Artificially Intelligent?
Mirano Spalburg, Consultant.

Abstract

Log evaluations to estimate formation properties such as porosity and saturation are often deterministic, the uncertainty being estimated with a partial derivatives scheme or a Monte Carlo scheme around the calculated property values. Almost inevitably the results exhibit high sensitivity to evaluation parameters such as grain and fluid density and failure to acquire data for one of the logging tools used in the evaluation scheme can cause considerable concern. Uncertainty is usually only estimated in the neighbourhood of the calculated formation properties. In Bayesian uncertainty reduction, by contrast, a range of possible formation properties is defined by a (team of) specialists and subsequently reduced by one or more logging measurements. Procedures used to implement this, such as inversion through Markov Chain Monte Carlo sampling, are typically computationally intensive, but the full uncertainty in the results is captured.

The subject of this presentation is an alternative Bayesian method in which all relevant uncertainties in formation properties, and their importance for the result of the evaluation, are defined at the outset and used to generate a large set of synthetic formations together with the responses of various logging tools to these synthetic formations. This synthetic data can be used to generate histograms and cross-plots to validate that the uncertainties defined at the outset have been correctly incorporated. The data set is then reduced, or “pruned” by using the results of available logging measurements, together with prescribed uncertainties in these measurements. The Bayesian inference is thus reduced to a filtering operation using measured tool data to select elements from a large data set with all relevant and possible combinations of formation properties and associated logging tool responses.

The method has been extensively tested and applied to logging data from UK and NL clastic environments and some of the results, and lessons learned during the evaluations, are shared in this presentation. It is less computationally intensive than the MCMC method referred to above and has been cast in the form of a Javascript module, which can be run on a year 2020 mid-range home/game PC, or even on a high-end smartphone. The method appears to require only moderate expert guidance and may indicate it is a step towards petrophysical artificial intelligence.

Biographical Details of the Speaker

Mirano Spalburg is retired. He enjoyed working for Shell as a petrophysics assurer, a studies and operational petrophysicist and as a researcher over a 30+ years period. He has a PhD in Mathematics and Physics and an almost lifelong affection for Bayesian methods.