Name: EDUARDO SOARES BORGES DOS SANTOS
Publication date: 13/03/2026
Examining board:
Name |
Role |
|---|---|
| ELIAS SILVA DE OLIVEIRA | Examinador Interno |
| GUILHERME DANIEL AVANSI | Examinador Externo |
| MATHIAS JOSÉ KREUTZ ERDTMANN | Coorientador |
| THOMAS WALTER RAUBER | Presidente |
Summary: The energy transition has intensified the search for solutions to mitigate CO2 emissions,
and among the relevant fronts, geological storage stands out. A fundamental step in this
type of project is the preliminary selection of injection locations, which requires evaluating
large volumes of petrophysical information. Full-flow simulations tend to have a high
computational cost, motivating the development of lighter descriptors that preserve the
mechanisms of CO2 plume migration.
Carbon dioxide migration in heterogeneous porous media is strongly controlled by hydraulic
connectivity and the internal architecture of the reservoir, including vertical anisotropy and
barriers that induce lateral deviations and bottlenecks. Although classical heterogeneity
metrics synthesize petrophysical variability in a predominantly statistical manner, they
do not explicitly represent the spatial organization of connectivity, which limits their
applicability. In this context, directed tortuosity is adopted as a structural descriptor, as
it interpretably condenses the combined effects of connectivity, anisotropy, and barrierinduced flow restriction.
The proposed methodology represents the discretized reservoir as a graph, in which cells
are nodes and orthogonal face connections form the edges. Connection weights translate
flow resistance, defined from the inverse of permeability. With this construction, the
minimum-cost path is computed between candidate injection points and the upper region
of the reservoir, producing a spatial field that highlights regions of higher and lower flow
resistance. The adherence of the structural proxy is evaluated by comparison with full
multiphase flow simulations, adopting as the observable the maximum elevation reached
by the plume within the analyzed time window.
At this point, the proxy results organize the domain into structural regimes, preserving
relevant information even in the absence of explicit multiphase dynamics. A supervised
predictive model is then trained exclusively on the structural attributes extracted from
the trajectories and demonstrates the ability to estimate the vertical reach with notable
performance, indicating that structural connectivity contains sufficient informative signal
to support screening tasks. As an implication, a workflow is established in which the
predictive model reduce the search space and prioritize scenarios, preserving multiphase
simulation as a subsequent stage for validation and detailed assessment
