4th INTERNATIONAL CONFERENCE
ON THE IMPORTANCE OF BUILDING ACCURATE
SEDIMENTOLOGICAL MODELS IN THE PETROLEUM INDUSTRY
C. Zavala1, 2, M. Arcuri 1, 2, L. Blanco Valiente3, J. Iparraguirre1,4, M. Di Meglio1, 2, and A.
1GCS Argentina SRL. Molina Campos 150, 8000 Bahía Blanca, Argentina.
2Departamento de Geología, Universidad Nacional del Sur, San Juan 670, 8000 Bahía Blanca, Argentina.
3REPSOL SINOPEC, Praia de Botafogo 300, 7º andar, 22250-040 Rio de Janeiro, Brazil.
4Universidad Tecnológica Nacional, 11 de Abril 461, 8000 Bahía Blanca, Argentina
Keyword: Geological model, static model, reservoir analysis, computer vision, Machine learning.
Hydrocarbons are contained on rock pore spaces. Porous sedimentary rock bodies have shapes,
thickness, and continuity that depend mainly on active depositional processes during accumulation.
In this context, the sedimentological model (or static model) represents an approximation to the
reality of the oil field. An accurate model is necessary to reduce the uncertainty during exploration
and production phases. At present, two different methodologies are applied in the industry to build
the geological model: 1) the traditional “pigeonholing” procedure and 2) the custom made
Traditional (1) sedimentological models are commonly constructed following the main guidelines
provided by the regional stratigraphic framework. Depositional environments are mainly
determined analyzing the shape of GR logs, and are complemented with some observations
performed on cores. Models showing facies distribution and dimensions of sedimentary bodies are
general, and are taken from analog systems found in literature and outcrops (if any). The final
model is chosen from a set of possible classes using a pigeon-hole approach. The main advantage
of these models is that they can be constructed relatively fast, at a low cost, by junior geologists.
However, there is a high risk involved. Maps and scales are often interpretative and poorly precise.
As a conceptual analogy, in this approach the model tries to explain the data.
By contrast, custom-made sedimentological models (2) involve detailed core studies, a meticulous
facies characterization, and determination of sedimentary process. Core description can be
facilitated using specific software and computer vision techniques, allowing the construction of
detailed sedimentological sheets and accurate determination of core shifts. The use of specific
software also allows to upload all observations into standard oil industry tools as a set of .las files.
Facies and facies sequences recognized on cores are calibrated with well logs, conventional
analysis, as well as petrographic, diagenetic and biostratigraphic studies. Depositional elements
recognized during facies analysis are characterized by a set of lithologic logs using machine
learning, allowing the definition of electrofacies and rock types. Machine learning also allows to
propagate core-based interpretation to a large well log database in mature fields. Additionally, the
construction of detailed correlations supported by 3D seismic survey allows to precisely determine
the dimensions and distribution of clastic bodies. The geological model built from core analysis
can be refined incorporating field studies performed on outcrops (if available). Population of
electrofacies with conventional analysis permits the construction of detailed maps showing not
only the distribution of depositional elements, but also reservoir properties for the different
analyzed intervals. In this scenario, the model follows the data. As a conclusion, the custom-made
model drastically reduces the exploration risk, since data is at the center of model building process.
This last approach requires a qualified multidisciplinary team with high experience on computer
and geological sciences.