Interpolation

Key Takeaways

• Kriging, the geostatistical interpolation method developed by the South African mining engineer Danie Krige in the 1950s and formalized by Georges Matheron in the 1960s, is the most widely used spatial interpolation method in petroleum reservoir modeling because it uses the variogram (a spatial correlation function that describes how the property's variance changes with separation distance between data points) to weight the known data points optimally, minimizing the estimation error at unsampled locations; unlike simpler methods such as inverse distance weighting (which gives more influence to nearby points without regard to the actual spatial correlation structure of the data), kriging uses the statistical relationship between the data values and their spatial arrangement to produce interpolated estimates that honor the geological continuity implied by the variogram; kriging also provides an estimation variance (uncertainty measure) for each interpolated value, allowing the interpreter to identify where the interpolation is well-constrained (many nearby data points, small estimation variance) and where it is highly uncertain (sparse data coverage, large estimation variance); the limitation of kriging is that it produces smooth interpolated surfaces that reproduce the input data exactly at data points but may not represent the full variability of the property between wells, which is why multiple stochastic realizations (using sequential Gaussian simulation or similar methods) are often used alongside kriging to represent the geological uncertainty in the interpolated distribution.
• The variogram (or its equivalent, the covariance function) used in kriging interpolation is one of the most important and often most poorly constrained inputs in petroleum reservoir geostatistics: the variogram quantifies the spatial correlation structure of the property being interpolated — the correlation range (the distance beyond which data points have negligible influence on the interpolated value), the sill (the total variance of the property), and the nugget (the variability at zero separation distance, reflecting measurement error and sub-scale variability); fitting a variogram model to experimental variograms computed from the available well data is straightforward when the data density is high and the property is relatively homogeneous, but becomes highly uncertain when the well spacing is larger than the geological correlation range or when the geological variability is strongly directional (higher correlation in the depositional direction than perpendicular to it); the variogram must be estimated from well data alone if seismic-derived property estimates are not available, and with typical North Sea or Permian Basin well spacings of 500-2,000 meters, the variogram is often determined by only a handful of data pairs at short separation distances and many uncertain assumptions about the geological process that controls the spatial correlation structure.
• Seismic attribute interpolation (also called trace interpolation or data regularization) addresses the requirement that seismic migration algorithms need data at regular spatial intervals to produce artifact-free images, while real acquisition footprints have irregular geometry due to cable feathering (in marine surveys), surface obstacles (in land surveys), and source-receiver offset distribution variations: the interpolation fills in missing or sparse traces using information from neighboring traces, exploiting the fact that the seismic wavefield at a missing location can be predicted from the wavefield at nearby recorded locations using plane-wave decomposition, Fourier-domain interpolation (F-K or F-X interpolation), or prediction-error filtering; successful trace interpolation allows surveys with irregular acquisition footprints to be processed with migration algorithms that assume regular geometry, significantly improving image quality; the limitation is that interpolation cannot reliably reconstruct data at locations separated from the nearest recorded traces by more than the spatial aliasing wavelength (the distance at which the wavefield changes by more than half a wavelength between adjacent traces), and aggressive interpolation of highly aliased data can introduce artifacts that reduce rather than improve image quality.
• Temporal interpolation between well test pressure points is used in pressure transient analysis when the downhole pressure gauge recording rate is low relative to the rate of pressure change during early-time buildup or drawdown: downhole pressure gauges in modern wells typically sample at 1 second or finer intervals during transients, but legacy gauges or cost-optimized recording schemes may sample at 1-5 minute intervals, missing the early-time pressure behavior that contains information about the wellbore storage coefficient, the damaged zone permeability, and the skin factor; interpolating between sparse pressure measurements to create a denser time series for derivative calculation may produce a smooth-looking derivative that appears to confirm the expected radial flow regime but actually obscures real features in the early-time behavior; the best practice for wells where early-time data is critical is to use high-frequency recording during the period of maximum rate of pressure change (the first few minutes to hours of buildup), and to store data at lower frequency only during the late-time period when the derivative has stabilized and the rate of pressure change is slow enough that sparse sampling is adequate.
• The distinction between interpolation (estimating values within the range of the known data) and extrapolation (estimating values beyond the range of the known data) is critical in petroleum engineering because extrapolated values have much higher uncertainty than interpolated values and may be physically unrealistic: a porosity-depth relationship calibrated from wells with depths between 1,000 and 3,000 meters can be interpolated within that range with reasonable confidence, but extrapolating the same relationship to 5,000 meters depth (beyond the calibration range) may predict negative porosity or porosities inconsistent with the mechanical compaction physics of the formation; similarly, a production decline curve fit to the first two years of well production can be used to interpolate the expected rate at any time within those two years with good confidence, but extrapolating the same curve to predict the 20-year cumulative production requires assumptions about decline behavior that may not hold if the well undergoes artificial lift, recompletion, or reservoir depletion effects not represented in the early production data; distinguishing interpolation from extrapolation in both the spatial and temporal domains is a fundamental quality control step in petroleum engineering analysis.