Local Porosity-Permeability Calibration Calculator

Calculation sequence

1. Enter four to one hundred paired core measurements from a locally consistent rock population. Porosity is percent, so enter 20 for 20 percent, not 0.20. Permeability is positive mD on a consistent directional, confining-stress, fluid and gas-slip-correction basis. Use facies or rock-type separation and lab metadata before fitting; the calculator cannot detect whether the rows were taken from comparable measurements.
2. Fit log10 of permeability against porosity in percentage points using unweighted ordinary least squares. A centered, scaled design matrix and ML.js QR decomposition avoid fitting normal equations or mixing raw-mD and logarithmic residual objectives. Repeated porosity rows remain repeated observations. At least three distinct porosities and a 0.01-percentage-point span are required, with resolvable variation in log permeability.
3. Inspect the slope, intercept, log-space residual table, RMSE and R-squared. RMSE divides squared log residuals by observation count, not by N minus two. Ten raised to that RMSE is a multiplicative scatter diagnostic, not a confidence interval. The plot uses log10 permeability on its vertical coordinate so the displayed line matches the actual fitting objective rather than masking low-permeability deviations.
4. Predict at the selected porosity and review the measured interval before reuse. Predictions outside it are explicitly flagged; a nonpositive slope, a span below two percentage points or log-space R-squared below 0.5 prompts review. These warnings are application review thresholds rather than geological validity rules. Save, export and share the actual measured pairs along with the fitted case so a colleague can reopen and recalculate it.

Worked check

The five default synthetic core pairs are (10%, 1 mD), (15%, 10 mD), (20%, 100 mD), (25%, 1,000 mD) and (30%, 10,000 mD). They exactly follow log10(k in mD) = -2 + 0.2 porosity_percent. At 20% the fitted prediction is 100 mD, log RMSE is essentially zero and log R-squared is one. This exact arithmetic fixture is not an actual formation data set or an accuracy claim.

Synthetic examples for reproducibility, not measured field results.

Assumptions and limitations

• This is a locally calibrated empirical log-linear transform, not a universal porosity-to-permeability equation, Kozeny-Carman pore-geometry model, NMR transform or effective-permeability upscaling method. Porosity alone does not identify pore connectivity, grain size, diagenesis, fracture contribution or directional permeability.
• Only measured positive permeabilities from 1e-9 to 1e6 mD and porosities from 0.01 to 60 percent are accepted. Zero values, below-detection-limit/censored data and missing fields are rejected, not replaced with tiny invented numbers. No outlier removal, weighting, uncertainty interval or facies classification is performed.
• Exponentiating the fitted log trend does not estimate arithmetic mean permeability without an additional statistical correction and assumptions. No such correction is silently applied. High training R-squared is not out-of-sample validation; reserve independent core data before applying the relationship to logs or reservoir simulation.

Sources and maintenance

Published by PetroCalcHub Team. Implementation and source checks: 2026-10-14. This does not claim independent expert review.

• USGS, Nelson (1994): formation-specific log-permeability/porosity trends and geological controls (abstract inspected)
• USGS, Nelson and Kibler (2003): 70 core-data sets with formation and measurement-condition metadata
• ML.js matrix v6.14.0: QR least-squares solve and rank checks

Related tools and references