Build the Reservoir Model

A reservoir model is a chain of decisions from data to forecast; seeing the whole chain first keeps you from optimizing one link and breaking the rest.

Cores, logs, seismic, and production each see a different Earth; the conceptual model is where you decide what story they are telling together.

The model is only as honest as the rock properties feeding it; poro-perm, Archie, and capillarity are the same physics the log analyst hands you.

Rock types are the vocabulary the whole model speaks; cutoffs and flow units decide which rock even enters the grid.

Horizons, faults, and zones are the skeleton every property will hang on; a bad framework cannot be rescued downstream.

Faults either compartmentalize or leak, and the layering scheme quietly sets the vertical resolution of everything above it.

The grid is where geology meets arithmetic; cell geometry and the resolution tradeoff decide both accuracy and runtime.

Zero-thickness cells and grid-orientation effects are the quiet killers of a simulation; cell-quality QC is where you catch them.

Every geostatistical map is downstream of one function; range, sill, and nugget are the dials, and this is the same spatial contract the geostatistics path teaches from the reservoir side.

Wells are drilled where rock is good, so raw statistics flatter the field; declustering and trend removal are how the model stops believing its own advertising.

Properties live inside facies; blocking wells and choosing object, indicator, or Gaussian simulation is the decision that shapes flow more than any single number.

When channels and lobes matter, two-point statistics run out of words; training images carry the shapes, and conditioning keeps them honest to the wells.

Kriging gives the best map and simulation gives the honest spread; porosity and permeability go in per facies or the model averages away the heterogeneity that controls flow.

One realization is a guess; a stack of them is an uncertainty model, and property QC is what stops a pretty map from hiding a broken one.

STOIIP with Monte Carlo turns a number into a range, and flow-based upscaling is how the fine geological grid survives the trip to the simulator without lying about connectivity.

The tornado chart names the inputs your volume actually depends on, which is the same as naming where the next dollar of appraisal should go.

The static model holds rock; the dynamic model needs fluids, relative permeability, and wettability, or the flow simulation is fiction with good graphics.

Darcy's law and mass conservation are the physics; discretizing them into transmissibilities is how the reservoir becomes a system of equations the simulator solves.

Buckley-Leverett and the mobility ratio decide how a waterflood sweeps, and the Peaceman model is how a real well enters a grid block.

The productivity index is where reservoir pressure becomes a rate on a report; it is the model's handshake with the engineer's economics.

IMPES or fully implicit, the CFL limit, and Newton convergence are the difference between a run that finishes and a run that lies; reading the output is a skill of its own.

History matching is inversion with all of inversion's non-uniqueness; a match that ignores that is a forecast built on sand.

The whole chain, seen once more from the top, ending where it should: a development decision the model was built to inform.

Framework, facies, properties, volumes: the static half of the workflow built with your own hands, proven by doing it, not by answering about it.

Upscale, initialize, drill, and run the waterflood: the dynamic half, where every static choice you made finally shows its consequence in a production profile.