Arithmetic Mean
Arithmetic mean is the sum of a set of values divided by the count of values, expressed as k_A = (k_1 + k_2 + ... + k_n) / n for n measurements or, in the thickness-weighted form used in reservoir engineering, as k_A = (Σ k_i × h_i) / Σ h_i where k_i is the permeability and h_i is the thickness of each layer. In petroleum engineering and petrophysics, the arithmetic mean is specifically the appropriate average for calculating the effective permeability of layers arranged in parallel flow (that is, layers stacked in the same direction as flow, each contributing independently to the total throughput). This is in contrast to the harmonic mean (appropriate for layers in series, perpendicular to flow) and the geometric mean (appropriate for randomly arranged heterogeneous media). The distinction among the three averaging methods has direct economic consequences: the arithmetic mean gives the highest of the three averages for any set of positive values with variance greater than zero, the harmonic mean gives the lowest, and the geometric mean lies between them. For a formation with a few high-permeability streaks and a majority of low-permeability matrix (common in argillaceous sandstones and carbonates), the arithmetic mean permeability can be 5 to 20 times higher than the harmonic mean, meaning a formation modelled with the arithmetic mean will be predicted to produce far more quickly than one modelled with the harmonic mean. Selecting the wrong averaging method for the reservoir geometry can produce waterflood forecast errors of 50 to 200 percent in cumulative production at a given time step, making average permeability calculation one of the most consequential technical decisions in reservoir simulation model construction.
Key Takeaways
- The arithmetic mean gives the effective permeability of parallel-flow layers, which is always equal to or greater than the geometric and harmonic means for any distribution with variance greater than zero: For a formation consisting of n layers each with individual permeability k_i and thickness h_i, flowing in parallel (so that the pressure drop across each layer is the same and the flow rates add), the total flow rate is the sum of individual layer flow rates by Darcy's law: Q = Σ (k_i × h_i × A × delta_P) / (mu × L). Dividing by the total thickness Σh_i and comparing to the Darcy equation for a single equivalent layer gives k_eff = Σ (k_i × h_i) / Σ h_i, the thickness-weighted arithmetic mean. A three-layer system with permeabilities 1,000, 50, and 2 mD and equal thicknesses has arithmetic mean (1000+50+2)/3 = 350.7 mD — far above the geometric mean (1000×50×2)^(1/3) = 21.5 mD or the harmonic mean 3/(1/1000+1/50+1/2) = 5.8 mD. This 350.7/5.8 = 60-fold range in effective permeability depending on flow direction is the fundamental reason why horizontal-to-vertical permeability ratio (k_H / k_V) in layered reservoirs is often 10 to 100, with k_H reflecting the arithmetic average of parallel layers and k_V reflecting the harmonic average of layers in series perpendicular to flow.
- The Dykstra-Parsons permeability variation coefficient and the Lorenz coefficient both use the arithmetic-mean-weighted permeability distribution to characterise heterogeneity in waterfloods: The Dykstra-Parsons coefficient V_DP = (k_50 - k_84) / k_50 is defined from the permeability distribution on a log-probability plot (where k_50 is the median and k_84 is the 84th percentile of the cumulative log-normal distribution). The Lorenz coefficient is the area between the permeability-thickness capacity curve (cumulative k_i × h_i / total k × h, plotted against cumulative h_i / total h) and the line of perfect homogeneity. Both coefficients are constructed from the set of layer-by-layer k_i × h_i products (flow capacity), which is the numerator of the arithmetic mean weighting formula. In the Pembina Cardium waterflood, V_DP of 0.45 to 0.60 and Lorenz coefficient of 0.25 to 0.40 indicate moderate heterogeneity, consistent with the dominance of a few higher-permeability conglomerate and coarse sandstone intervals (k = 50 to 300 mD) in the total flow capacity (arithmetic-mean contribution) despite the majority of net pay being in moderate-permeability (k = 2 to 15 mD) fine-to-medium sandstone. The waterflood breakthrough time is primarily controlled by the high-k layers (arithmetic-mean-weighted), while the total oil recovery is controlled by the distribution across all layers (integrated Lorenz coefficient).
- In reservoir simulation, the arithmetic mean is used to upscale fine-scale permeability grids to coarser simulation cells for parallel-flow grid blocks, while the harmonic mean is used for series-flow directions: Modern reservoir simulation models upscale fine-scale geological model permeabilities (calculated from core measurements at plug scale, 2.5 to 5 cm) to coarser simulation cells (typically 50 to 200 metres in the areal directions and 0.5 to 5 metres vertically). The upscaling procedure uses arithmetic mean for permeability within a simulation layer (horizontal direction, parallel flow across the cell face) and harmonic mean for permeability across layer boundaries (vertical direction, series flow between upper and lower cells). For a simulation layer containing 20 fine-scale grid cells each with different permeability, the arithmetic mean of those 20 values gives the horizontal effective permeability for the upscaled cell; dividing the layer into 2 sub-layers and computing the harmonic mean gives the vertical transmissibility between them. If the fine-scale permeability distribution has a geometric mean of 10 mD but includes a 2-cm plug measuring 2,000 mD (representing a natural fracture or microfault), the arithmetic mean for a 50-metre horizontal simulation cell including that plug is approximately 10.8 mD (the 2,000 mD plug at 2 cm barely affects the 50-metre arithmetic average), while the harmonic mean for a vertical 5-cm column including that plug is 10,000 / (10,000/1.5 + 10,000/2,000 + 10,000/1.5) × n^-1 = substantially lower, illustrating how the averaging method changes dramatically depending on whether the high-permeability plug is in the flow-parallel or flow-perpendicular direction.
- The arithmetic mean of formation water resistivity (Rw) across a vertically inhomogeneous water column provides the input to regional water saturation calculations in formation evaluation studies: In a regional formation evaluation study covering hundreds of wells, the Rw value used in the Archie Equation must represent the formation water at each well location. Where the formation water composition varies with depth due to salinity stratification (common in Devonian carbonates and some Cretaceous sandstones that were influenced by freshwater recharge), the Rw at each depth point is estimated from the SP log or from produced water chemistry. For a well evaluation that averages over a 10-metre gross pay interval with varying salinity (Rw ranging from 0.025 to 0.045 ohm-m over depth), the arithmetic mean Rw is the correct input for the Archie Equation applied at the mean porosity of the interval, because the Archie Equation's linear relationship between conductance (1/Rw) and the water-phase conductance in parallel pores makes the arithmetic mean of Rw the appropriate pore-volume-weighted equivalent for a vertically heterogeneous water column. Using a single minimum or maximum Rw rather than the arithmetic mean across the interval introduces a systematic error in Sw of 5 to 15 percent in formations with moderate Rw gradients, which can be material in thin-pay borderline economic wells.
- Statistical arithmetic means are used in drilling engineering to characterise average drilling performance metrics including rate of penetration, bit wear, and non-productive time distribution: The arithmetic mean of rate of penetration (ROP) over a well section provides the basis for estimating rig time and cost for subsequent wells in the same formation or area. For a 1,200-metre Cardium horizontal well lateral drilled with a PDC bit in 14 days, the arithmetic mean ROP is 1,200 / (14 × 24) = 3.57 m/h; if the same section is drilled in 4 runs with individual average ROPs of 2.5, 4.0, 3.8, and 3.9 m/h, the overall section arithmetic mean is (2.5 + 4.0 + 3.8 + 3.9) / 4 = 3.55 m/h, very close to the overall average, because the run lengths are approximately equal. If run lengths differ substantially (e.g., 100 m at 2.5 m/h and 1,100 m at 3.9 m/h), the length-weighted arithmetic mean gives (100×2.5 + 1100×3.9) / 1200 = 3.68 m/h, which is the correct overall average ROP for cost forecasting. The distinction between simple (unweighted) and length-weighted arithmetic mean is exactly analogous to the distinction between simple and thickness-weighted permeability averaging, and failure to use the appropriate weighting can bias the drilling performance benchmark and lead to over- or under-estimation of the budget for subsequent wells in the same play.
Arithmetic Mean in Reservoir Engineering, Petrophysics, and Drilling Operations
The classical Dykstra-Parsons waterflood performance prediction method, which remains widely used in WCSB waterflood management, relies entirely on the arithmetic-mean-weighted flow capacity (k_i × h_i) distribution. The method ranks individual flow layers from highest to lowest k_i × h_i, computes the cumulative flow capacity at each percentile of cumulative layer thickness, and uses the resulting Lorenz plot to predict breakthrough time and post-breakthrough water-oil ratio as a function of cumulative pore volumes injected. The arithmetic mean permeability (Σ k_i × h_i / Σ h_i) appears as the normalising factor in this analysis, setting the time scale for the waterflood advance. An accurate arithmetic mean permeability is therefore the fundamental input to the Dykstra-Parsons waterflood forecast: errors in the layer-by-layer k_i values propagate into the arithmetic mean and then into the breakthrough time prediction and the ultimate recovery forecast. In the Pembina Cardium Unit, where individual well arithmetic mean permeabilities range from 2 to 45 mD across the pool area (reflecting local variations in conglomerate and sandstone facies proportions), the spatial distribution of the arithmetic mean permeability is the primary control on the spatial distribution of waterflood recovery efficiency and is used to guide the priority ranking for infill drilling locations and pattern rebalancing decisions.
The relationship between arithmetic mean and geometric mean permeability in a log-normally distributed formation has a well-known closed-form expression: k_A = k_G × exp(sigma_ln^2 / 2), where k_G is the geometric mean and sigma_ln is the standard deviation of the natural log of permeability. For a moderately heterogeneous sandstone with geometric mean permeability of 10 mD and sigma_ln = 2.0 (equivalent to V_DP approximately 0.73), the arithmetic mean is 10 × exp(4/2) = 10 × 7.39 = 73.9 mD. This factor of 7.4 between arithmetic and geometric means illustrates why using the arithmetic mean for a heterogeneous reservoir without considering the flow direction will grossly overestimate horizontal permeability (appropriate for layer-parallel flow) compared to the geometric mean (appropriate for flow through a random heterogeneous medium such as a core analysis sample). In practice, plug-scale core permeabilities (measured in the axial direction through the plug, which is typically in the flow-perpendicular direction if the plug is cut vertically from a horizontal core) are closer to the harmonic mean of the formation's layered structure, while wireline log-derived permeabilities (estimated from NMR or empirical transforms applied to porosity logs) sample the full thickness of the bed and are closer to arithmetic-mean-weighted averages.
In petrophysical formation evaluation, the arithmetic mean of log-derived porosity across a net pay interval is the correct input to the volumetric OOIP calculation: OOIP = A × h × phi_mean × (1-Sw_mean) / Bo, where phi_mean and Sw_mean are both thickness-weighted arithmetic means over the net pay interval. However, for computing the arithmetic mean from discrete log readings, the appropriate weighting is the depth sampling interval (typically 0.15 metres or 6 inches for standard digital log recordings in WCSB wells), not the gross thickness of the interval. For a 6-metre Cardium pay zone with 40 log readings at 0.15-metre spacing, the thickness-weighted arithmetic mean porosity is (Σ phi_i × 0.15) / (40 × 0.15) = (Σ phi_i) / 40, the simple average of the 40 readings, provided the readings are evenly spaced. If depth-selective averaging is performed (including only the readings where GR less than 40 API and phi greater than 10 percent as net pay), the arithmetic mean porosity of the net pay is computed from the qualifying depth samples only, giving a higher mean porosity than the gross interval average because the excluded argillaceous intervals have lower porosity than the clean sand pay.