Seismic Stratigraphy and Facies
The interpreter reads geometries; the modeler fills a grid. This path walks the whole distance between them: terminations and sequences on the section, channels and fans in map view, then variograms, facies methods, and indicator geostatistics until the depositional system you saw on seismic lives inside the reservoir model.
You can name the four reflection terminations and the surfaces they diagnose, classify a channel system from a horizon slice and predict its net-to-gross, place your prospect on a turbidite fan, fit a variogram you can defend, choose between object-based and pixel-based facies methods for a given geology, and condition the resulting model to wells, trends, and seismic probability.
Read the geometries
Onlap, downlap, toplap, truncation: four ways a reflector can end, and the entire alphabet of seismic stratigraphy. Every sequence boundary you will ever draw rests on this evidence.
The terminations combine into sequence boundaries, flooding surfaces, and systems tracts driven by relative sea level; this is how you correlate across a basin with no well control.
Fluvial, deltaic, shoreface, carbonate, deep-water: each system has an architectural signature and a reservoir prediction attached. Seismic geomorphology assembles them into one readable scene on a horizon slice.
Channels are the one depositional body that announces itself in plan view, and channel type sets net-to-gross before the first well is drilled. Classifying them fast is a core workstation skill.
The fan is not a uniform sand body: proximal, mid-fan, and distal positions carry different architectures, net-to-gross, and drilling targets, and modern deep-water exploration is built on reading that gradient.
Spatial continuity groundwork
The variogram is how a model remembers geology: average squared difference versus lag, with nugget, sill, and range setting the texture of every realization built on it.
Continuity in a reservoir is directional, with the azimuth taken from the depositional system you mapped in stage one, and a trend breaks the stationarity the variogram assumes until you remove it and model the residual.
The experimental variogram is a handful of noisy points; kriging and simulation need a valid smooth model, pinned to the pair-rich short lags and checked against what you know of the rock.
Depositional systems into the model
Facies are the first-order control on porosity and permeability, which is why the workflow models facies before properties; blocking the fine well logs to grid-cell scale is the unglamorous step that decides what the model can honor.
Object-based modeling drops whole channel bodies into the grid with the width, sinuosity, and orientation you read from seismic in stage one; the price of that realism is that net-to-gross becomes an outcome, not a setting.
Truncated Gaussian cuts one continuous field into ordered facies bands; sequential indicator simulation handles unordered facies cell by cell. Between them they trade the crisp shapes of objects for exact proportions and easy conditioning.
Facies by geostatistics
The indicator transform turns a category or a cutoff into a 0/1 field whose variogram measures the connectivity of the zones you care about, a different question than the continuous variogram answers.
Krige the indicator field and the estimate reads directly as a facies probability, no Gaussian assumption required; order-relation violations are the housekeeping the method demands in return.
Indicator methods are distribution-free and categorical-native, but they carry five canonical failure modes; knowing when SISIM is inadequate and an object-based or multipoint approach is warranted is the judgment this path exists to build.
The finished facies model must reproduce the well facies exactly while trends, proportion curves, and seismic-derived probability steer it between the wells; this is the seam where the interpretation from stage one finally constrains the grid.