Interview with Arben Shtuka

Today, we meet Arben Shtuka , our Head of Research at Seisquare , who is behind our innovative and advanced algorithms.

?

Arben, can you tell us about your background?

After my training as a geophysical engineer, I completed a PhD in image processing and applied geostatistics. After my thesis, I led the GoCAD? research team, working on geostatistical and stochastic reservoir characterization tools. This implied many collaborations with Stanford University, Elf and others. After a time as a lecturer at the Nancy School of Geology, I worked as an independent consultant for several years. I contributed to Elf’s Jacta project, which aimed to take all uncertainties into account, including in faulted fields. In 2008, I joined Seisquare as Head of Research and set out to use all my experience and training to build the most scientifically accurate and advanced algorithms.

?

What are you currently working on?

I’m writing Udomore? Suite’s core algorithms, that bring cutting-edge geostatistics, subsurface modelling, data filtering and data analysis to the users’ fingertips.

?

Can you explain, in simple terms, the challenges you’re facing?

If you try to build a model from raw exploration data analysis, you will have to use stationary, linear models, because other models are too complex to derive from data only.

But real-world models usually present non-linear trends and non-stationary covariances. So simple linear, stationary models are flawed and show discrepancies that increase as you get farther from the wells.

The best way to build accurate models is to include a priori geophysical models as well as all the available data and their associated uncertainties. This method is called Bayesian kriging developed by excellent Norwegian geostatistical school in O&G applications.

My humble contribution in the implementation of the Bayesian kriging (and cokriging) was the extension of this method that allows nonlinear trends models.

?

What are the advantages of this method?

Geophysical data (velocity data, seismic, data, well data, etc.) present many sources of uncertainties because of the way they are obtained. To compensate this, you need to use all the available data sources to cross check them and minimize the uncertainty of the model.

Using our stochastic methods, we use two extra sources compared to deterministic methods: the a priori geophysical model and all the uncertainties related to each data source.

Also, data quality control is intrinsic to the process, so flawed data is detected and can be treated, filtered, etc.

This results in a more accurate model and ultimately more successful drilling.

?

Can you give me an example of the advantages?

Take the layer cake approach, for instance. To simplify things, classic software compute layer models individually, completely ignoring correlations between layers, which can give aberrant results (zero thickness layers, crisscrossing layers, etc.).

Our method uses a multilayer model which computes all layers at once and takes horizons’ correlations into account, avoiding those anomalies and increasing the reliability of the model.

?

But with so much data, isn’t the software complicated to use?

There is complex parametrization to do, but the software is built to guide the user through it. For instance, the a priori model can be verified before the main computation is launched.

This allows the user to find the best possible fit between the a priori model and the data, which optimizes the estimation variance and lowers the uncertainties. Our software helps the user optimize the parametrization to reach the best possible estimate, for optimized exploration and production.

?

If you would like to have more technical and detailed answers, or want to see the theory in action, please contact us at [email protected] .