betaFeedback

WorksPhysics and Astronomy

Estimating Permeability of Carbonate Rocks from NMR T2 Distribution and Formation Factor Using a Physics-Based Model

Abstract

Nuclear magnetic resonance (NMR) logging provides a nondestructive method for characterizing pore structure and fluid distribution in reservoir rocks. Conventional permeability models, such as the Schlumberger-Doll Research (SDR) and Timur-Coates (TC) equations, have been widely used to estimate permeability from NMR T2 relaxation time distributions. However, these empirical models require calibration against core-measured permeability, and their fitted coefficients often vary substantially across lithologies and pore systems. In this study, we formulate and evaluate an NMR-informed implementation of a physics-based permeability relationship derived from critical path analysis (CPA) by linking the characteristic pore length scale to the dominant NMR relaxation time. We estimate permeability (k) using NMR T2 distributions and electrical formation factor (F) for 14 carbonate rock samples collected from the Morro do Chaves Formation in the Sergipe-Alagoas Basin, Brazil. We compare the predictive performance of this formulation with that of calibrated SDR and TC models. The proposed CPA-NMR formulation achieves a root mean squared logarithmic error (RMSLE) of 0.56 without calibration against measured permeability from the investigated formation. These findings indicate the potential of combining NMR relaxation measurements with electrical formation factor for physics-based permeability estimation. Further evaluation using independent datasets is needed to assess applicability across formations and lithologies.

Introduction

Permeability (kk) estimation is fundamental to aquifer and reservoir characterization because it directly governs fluid flow and transport in geologic formations. In carbonate reservoirs, permeability prediction is particularly challenging due to complex and frequently vuggy pore structures, multiscale porosity systems, and diverse diagenetic alterations [41, 52]. Although conventional core-based measurements provide reliable permeability values, they are expensive, time-consuming, and spatially limited. Consequently, numerous predictive relationships have been developed to estimate carbonate permeability from more accessible rock properties, including porosity and specific surface area [8], porosity and lithology [34], and mercury intrusion capillary pressure (MICP) measurements [25, 42]. However, although porosity and lithology information may often be obtained from well logs, in situ determination of specific surface area and capillary pressure characteristics remains difficult.

Nuclear magnetic resonance (NMR) measurements provide a non-destructive alternative for pore-scale characterization that can be performed both in laboratory experiments and through downhole logging tools. NMR-derived transverse relaxation-time (T2T_2) distributions are commonly interpreted as indicators of pore-size distribution and fluid distribution within the pore network, thereby providing a basis for permeability estimation [29, 11]. For a fully saturated porous medium, the measured NMR signal arises primarily from hydrogen-bearing fluids occupying the pore space, such that the signal amplitude associated with a given relaxation component is related to the volume of fluid contributing to that component.

Under fast-diffusion conditions and when surface relaxation dominates over bulk and diffusion-induced relaxation, the transverse relaxation rate can be approximated as 1/T2≈ρ2S/V1/T_2 \approx\rho_2 S/V, where ρ2\rho_2 is the transverse surface relaxivity and (S/V) is the pore surface-to-volume ratio [11]. Consequently, larger pores generally exhibit longer T2T_2 values, whereas smaller pores relax more rapidly. The measured T2T_2 distribution can therefore provide a proxy for the distribution of pore-scale surface-to-volume ratios, provided that surface relaxivity does not vary strongly within the sample and that the assumptions underlying the fast-diffusion approximation are reasonably satisfied.

Importantly, however, an NMR-derived pore-size distribution does not necessarily represent the pore-throat size distribution controlling permeability, particularly in heterogeneous carbonate rocks.

One of the most widely used NMR-based permeability models is the Timur–Coates (TC) equation, which relates permeability to porosity (ϕ\phi), free fluid index (FFI), and bulk volume irreducible (BVI) through the empirical power-law relationship [11]

k=CTCϕ4(FFIBVI)2(1)k = C_{TC}\phi^{4}\left(\frac{FFI}{BVI}\right)^{2} \tag*{(1)}

where CTCC_{TC} is an empirical proportionality coefficient. The free fluid index (FFI) represents the fraction of NMR-detectable pore fluid associated with relaxation times above a prescribed T2T_2 cutoff and is commonly interpreted as relatively mobile fluid, whereas bulk volume irreducible (BVI) represents the fluid fraction below the cutoff and is associated with more strongly confined or capillary-bound fluid. The TC model attempts to distinguish between mobile and immobile fluids within the pore system, but its formulation remains fundamentally empirical.

Another widely applied NMR-based permeability relationship is the Schlumberger-Doll Research (SDR) model, which relates kk to ϕ\phi and the logarithmic mean relaxation time (T2lmT_{2\mathrm{lm}}) as follows [29]:

k=CSDRϕ4T2lm2(2)k = C_{SDR}\phi^4 T_{2\mathrm{lm}}^2 \tag*{(2)}

where CSDRC_{SDR} is an empirical constant whose value depends strongly on lithology, pore geometry, and measurement conditions.

Behroozmand et al. [6] reviewed the application of NMR methods to near-surface characterization and highlighted that the predictive reliability of SDR-based permeability estimation depends on lithology and calibration strategy. Their analysis showed that the SDR model can provide reliable predictions for inorganic unconsolidated sediments when properly calibrated at site- or geology-specific scales, whereas it fails in organic-rich materials such as peat, where relaxation time no longer correlates with pore-size-controlled flow. Their study further demonstrated that generalized calibration across mixed lithologies is ineffective and that simplified formulations excluding porosity may reduce uncertainty under conditions where porosity estimation is biased by magnetic susceptibility effects.

Both the TC and SDR models, Eqs. (1) and (2), rely heavily on calibration against core-measured permeability data. Although these models have been successfully applied to sandstone reservoirs [11, 9], their applicability becomes considerably more challenging in formations characterized by heterogeneous pore structures and complex mineralogy, particularly in carbonate reservoirs. This limitation arises because the coefficients CCTC_{CT} and CSDRC_{SDR} depend strongly on pore geometry, mineralogical composition, and surface relaxivity [38], thereby restricting their transferability across different formations.

To improve predictive flexibility, both models have been generalized by replacing the fixed exponents in Eqs. (1) and (2) with adjustable fitting parameters calibrated separately for each formation or dataset [43, 15, 53]. Table 1 summarizes several generalized TC and SDR formulations reported for carbonate reservoirs worldwide. As shown, the calibrated exponents vary substantially among formations, further highlighting the strong lithology dependence of empirical NMR-based permeability relationships.

Referencekk unitϕ\phi unitT2lmT_{2lm} unitModelReservoir
Amabeoku et al. [2]mDcm3^{3}cm−3^{-3}msk=0.0574ϕ3.79T2lm1.89k = 0.0574\phi^{3.79}T_{2lm}^{1.89}Arab carbonate reservoir
Amabeoku et al. [2]mDcm3^{3}cm−3^{-3}msk=511.4ϕ2.84(FFIBVI)2k = 511.4\phi^{2.84}\left(\frac{FFI}{BVI}\right)^{2}Arab carbonate reservoir
Hassall et al. [24]mDcm3^{3}cm−3^{-3}msk=0.5ϕ2(ρ2T2lm)2k = 0.5\phi^{2}(\rho_{2}T_{2lm})^{2}Abu Dhabi, UAE
Machado et al. [33]mD%msk=7.3×10−11ϕ2.68T2lm4.56k = 7.3 \times10^{-11}\phi^{2.68}T_{2lm}^{4.56}Santos Basin, Brazil
Machado et al. [33]mD%msk=(ϕ109.64)1.74(FFIBVI)3.85k = \left(\frac{\phi}{109.64}\right)^{1.74}\left(\frac{FFI}{BVI}\right)^{3.85}Santos Basin, Brazil
Agha et al. (2018)mDcm3^{3}cm−3^{-3}msk=0.0216ϕ1.64T2lm1.68k = 0.0216\phi^{1.64}T_{2lm}^{1.68}Asmari, Iran
Agha et al. [1]mDcm3^{3}cm−3^{-3}msk=(ϕ0.222)3.9(FFIBVI)0.51k = \left(\frac{\phi}{0.222}\right)^{3.9}\left(\frac{FFI}{BVI}\right)^{0.51}Asmari, Iran
Martins [35]mD%msk=0.12×104ϕ4.48T2lm1.0k = 0.12 \times10^{4}\phi^{4.48}T_{2lm}^{1.0}Morro do Chaves, Brazil
Martins [35]mD%msk=3.45×104ϕ5.05(FFIBVI)0.78k = 3.45 \times10^{4}\phi^{5.05}\left(\frac{FFI}{BVI}\right)^{0.78}Morro do Chaves, Brazil
Shabani et al. [48]mD%msk=0.0004(ϕ10)3T2lm1.6k = 0.0004\left(\frac{\phi}{10}\right)^{3}T_{2lm}^{1.6}Asmari, Iran
Shabani et al. [48]mD%msk=0.022(ϕ10)4(FFIBVI)2k = 0.022\left(\frac{\phi}{10}\right)^{4}\left(\frac{FFI}{BVI}\right)^{2}Asmari, Iran

Table 1. The generalized TC and SDR models and their parameters calibrated for carbonate rocks from various reservoirs. kk is the permeability, ϕ\phi is the porosity, FFIFFI represents the free fluid index, BVIBVI denotes the bulk volume irreducible, and ρ2\rho_{2} is the surface relaxivity. The unit of ρ2\rho_{2} in the Hassall et al. [24] model is μm/s\mu\mathrm{m}/\mathrm{s}.

More recently, Mashhadi et al. [36] evaluated SDR-based permeability estimation for unconsolidated geological materials using 92 laboratory samples and 198 borehole screens collected from sites in Denmark and Germany. Their dataset included inorganic sediments (sand, clay, and till) as well as organic materials such as peat and gyttja. They found that the SDR model performed reasonably well for inorganic materials when geology-specific calibration was applied, whereas it failed for peat soils, where no meaningful correlation existed between relaxation time and permeability. Although gyttja followed SDR-type scaling behavior, it required calibration coefficients substantially different from those of inorganic sediments. Their results further demonstrated that generalized calibration across mixed lithologies was ineffective, whereas site- or geology-specific calibration significantly improved prediction accuracy.

Currently, the exponents in generalized TC and SDR formulations are commonly optimized to maintain an approximately constant prefactor at the formation scale.

However, Ghanbarian et al. [21] recently argued that such treatment may lead to physically inconsistent interpretations because variations in pore geometry, surface relaxivity, and electrical connectivity are implicitly absorbed into empirical exponents.

Previous studies have established theoretical connections between NMR relaxation measurements and permeability through characteristic pore dimensions and pore-scale transport relationships. In particular, Sigal [50] developed an NMR-informed permeability expression motivated by percolation theory (see his Eq. (8)), relating permeability to a characteristic NMR relaxation time and porosity through Archie's relationship (Archie, [3]). Although this formulation provided a theoretical interpretation of SDR-type permeability models, its predictive performance was not directly evaluated using measured permeability data. Moreover, replacing formation factor with Archie's relationship introduces an additional assumption that may not adequately describe electrically complex carbonate rocks. Building on this theoretical foundation, the present study integrates critical path analysis with the dominant NMR transverse relaxation time, effective surface relaxivity, and independently measured formation factor. The resulting formulation retains an explicit electrical transport constraint without requiring Archie's relationship and is quantitatively evaluated against measured permeability for 14 heterogeneous carbonate samples. The contribution of this study is therefore the implementation and evaluation of a physically motivated NMR–CPA framework, rather than the derivation of an entirely new permeability law.

The main objective of this study is to develop and evaluate a physics-based CPA–NMR formulation for permeability estimation that does not require calibration against measured permeability from the target formation. Specifically, we investigate whether the dominant NMR transverse relaxation time, combined with effective surface relaxivity and independently measured electrical formation factor, can provide a practical approximation of the characteristic length scale required by CPA. The predictive performance of the proposed formulation is evaluated using 14 heterogeneous carbonate samples from the Morro do Chaves Formation and compared with existing SDR and Timur–Coates permeability models.

NMR-Informed Permeability Framework

Unlike empirical permeability formulations, physics-based models attempt to explicitly relate permeability to pore geometry and transport properties of the porous medium [12, 7, 16]. Within the framework of critical path analysis (CPA), Katz and Thompson [27] developed a percolation-based permeability relationship involving a critical pore diameter (dcd_c) and formation factor (FF) as follows

k=dc2CCPAF(3)k = \frac{d_c^2}{C_{CPA}F} \tag*{(3)}

where CCPAC_{CPA} is a proportionality constant. In Eq. (3), dcd_c represents the characteristic constriction size associated with the onset of a sample-spanning connected pathway, rather than an arithmetic or volumetric mean pore diameter. Katz and Thompson [27] originally proposed a value of 226 for CCPAC_{CPA}; however, subsequent studies reported different values depending on pore structure and porous medium type. A review of reported CCPAC_{CPA} values can be found in Ghanbarian et al. [20]. Eq. (3) has been successfully applied to estimate the kk of sandstone and carbonate rocks from mercury intrusion porosimetry data [27, 20, 40, 37].

Katz and Thompson [27] further suggested that the critical pore diameter may be estimated from the inflection point of the capillary pressure curve, corresponding to the mode of the pore-throat size distribution [18]. In the present study, we set

dc=cSPρ2eT2peak(4)d_c = c_{SP}\rho_{2e}T_{2peak} \tag*{(4)}

in which csc_s is a pore-geometry factor equal to 4 for cylindrical pores and 6 for spherical pores, ρ2e\rho_{2e} is the effective surface relaxivity [11], and T2peakT_{2peak} represents the relaxation time corresponding to the maximum incremental porosity of the NMR T2T_2 distribution.

The approximation dc=csρ2eT2peakd_c = c_s\rho_{2e}T_{2peak} requires careful interpretation because NMR transverse relaxation is primarily sensitive to pore surface-to-volume characteristics, whereas the critical diameter, dcd_c, in CPA represents a flow-controlling pore-throat scale along the sample-spanning flow network. As discussed by Coates et al. [11], NMR-derived characteristic pore dimensions do not necessarily correspond to critical pore-throat dimensions, particularly in heterogeneous carbonate rocks with complex pore-body-to-throat relationships. To account phenomenologically for this distinction, an effective surface relaxivity, ρ2e\rho_{2e}, is introduced to relate the NMR-derived characteristic length scale to the hydraulically relevant pore-throat dimension. Unlike the intrinsic transverse surface relaxivity, ρ2\rho_2, which characterizes relaxation at the pore-fluid–solid interface, ρ2e\rho_{2e} represents an effective conversion parameter that incorporates the relationship between NMR-sensitive pore dimensions and flow-controlling constrictions. Accordingly, Eq. (4) constitutes a modeling approximation rather than a direct measurement of the critical pore-throat diameter, and its validity depends on the correspondence between NMR relaxation characteristics and the pore-network geometry governing fluid flow.

Substituting Eq. (4) into Eq. (3) yields

k=cs2ρ2e2T2peak2cCPAF(5)k = \frac{c_s^2 \rho_{2e}^2 T_{2peak}^2}{c_{\mathrm{CPA}} F} \tag*{(5)}

Eq. (5) provides a physically motivated relationship linking permeability to pore geometry, electrical connectivity, and NMR relaxation behavior. The use of T2peakT_{2peak} as a characteristic relaxation parameter has been previously explored in several studies. For example, [10], [30], and [54] successfully employed T2peakT_{2peak} in permeability-related analyses. Rezaee et al. [45] further demonstrated that permeability in tight gas sandstones was strongly correlated with porosity and T2peakT_{2peak} using multiple linear regression analysis.

Although CPA-based permeability relationships have been extensively evaluated using critical pore-throat sizes determined from MICP measurements, such measurements are not routinely available in many subsurface applications, particularly well logging.

The objective of the proposed formulation is therefore to approximate the critical pore-throat length scale required by CPA using the dominant NMR relaxation time and effective transverse surface relaxivity, thereby enabling permeability estimation without direct MICP or micro-CT characterization. The effective surface relaxivity provides a phenomenological means of accounting for differences between NMR-sensitive pore dimensions and permeability-controlling pore throats. Although this approximation does not establish a unique correspondence between T2peakT_{2peak} and the critical pore-throat diameter, its practical utility can be assessed through comparison of predicted and independently measured permeability. Accordingly, the present study evaluates the predictive performance of this NMR–CPA formulation rather than independently validating the underlying pore-throat length-scale correspondence.

Experiments

A total of 14 carbonate rock samples collected from the Morro do Chaves Formation in the Sergipe-Alagoas Basin, northeastern Brazil, were analyzed in this study. The Lower Cretaceous Morro do Chaves Formation consists predominantly of limestones characterized by variable porosity, heterogeneous pore systems, and locally developed vuggy structures.

Prior to petrophysical characterization, all samples were cleaned to remove residual hydrocarbons, salts, and other pore-filling contaminants. Cleaning was performed using Soxhlet extraction, with toluene to dissolve hydrocarbons and methanol to remove salts and residual water from the pore system. This procedure enabled continuous extraction while minimizing alteration of the rock framework. Following extraction, the samples were dried in an oven at 60°C for 12 h to ensure complete removal of residual fluids and solvents before measurements.

Porosity was determined using helium gas expansion based on Boyle’s law. Absolute permeability was measured using steady-state gas permeametry during routine core analysis. In this method, an inert gas was injected through the core plug at a controlled pressure gradient until steady-state flow was achieved. Permeability was subsequently calculated using Darcy’s law, and Klinkenberg corrections were applied to estimate equivalent liquid permeability.

The samples were then prepared for NMR measurements by full brine saturation with a 30,000 ppm KCl solution at a density of 1.067 g cm−3^{-3}. The KCl brine was selected to minimize clay swelling and fluid–rock interactions. Because the samples covered a broad permeability range, different saturation procedures were applied for high- and lowpermeability samples. For samples with permeability values greater than 100 mD, saturation was achieved by applying vacuum while the samples were immersed in brine. For the lower-permeability samples ((k<100k<100) mD), an initial vacuum treatment was applied for 8 h, after which the plugs were transferred to a pressure cell completely filled with the saturating brine. The samples were maintained under an applied fluid pressure of up to 2,000 psi for approximately 24 h to promote complete saturation. This pressure was applied during the saturation procedure and should not be interpreted as an independently imposed external confining pressure. Electrical measurements were subsequently conducted under ambient laboratory conditions. After saturation, the samples were reweighed, and the difference between dry and saturated masses was used to estimate the volume of brine occupying the pore system. Prior to NMR measurements, the samples were wrapped with Teflon tape to preserve saturation conditions and minimize evaporative losses during acquisition.

The electrical formation factor, FF, was determined from resistance measurements performed on fully brine-saturated core plugs using a VINCI Technologies EPS-A (Electrical Properties System at Ambient Conditions) equipped with a two-electrode configuration. The electrodes were positioned at opposite end faces of each cylindrical plug, with pneumatic loading applied to maintain reproducible electrode-to-core contact under ambient laboratory conditions without imposing external confining pressure. Electrical resistance was measured at 10 kHz using an external LCR meter, with three measurements obtained for each sample. This frequency is consistent with previous investigations of rock electrical properties. For example, Zisser and Nover [57] reported negligible electrode polarization effects at 10 kHz in their experimental configuration, while Revil et al. [44] obtained comparable formation factors over frequencies ranging from 10 Hz to 10 kHz (see their Figs. 9 and 10). The arithmetic mean of the three recorded resistance values, Rˉ\bar{R}, was used to calculate the resistivity of the fully saturated rock as Ro=Rˉ(A/L)R_o = \bar{R}(A/L), where AA and LL represent the cross-sectional area and length of the core plug, respectively. The formation factor was subsequently calculated as F=Ro/RwF = R_o/R_w, using the reported resistivity of the 30,000-ppm KCl saturating brine, Rw=0.147 Ω⋅mR_w = 0.147\ \Omega\cdot\mathrm{m}, at 23°C. Although the measurements were conducted under ambient laboratory conditions, the exact sample temperature and the equivalent-circuit configuration of the LCR meter were not documented. Consequently, it could not be established whether the recorded resistance represented the real component of complex impedance or another instrument-defined resistance quantity.

The NMR measurements were performed using a low-field GeoSpec2 spectrometer (Oxford Instruments, UK) operating at a magnetic field strength of 0.047 T and a resonance frequency of 2 MHz. All measurements were conducted in the NMR and Petrophysics Applications Laboratory (UFFLar), and acquisition control was implemented using GIT software (Green Imaging Technologies, Fredericton, Canada).

The transverse relaxation-time (T2\mathrm{T}_2) measurements were acquired using the Carr–Purcell–Meiboom–Gill (CPMG) pulse sequence, which is widely used in petrophysical applications because of its sensitivity to pore-scale relaxation processes. The sequence consisted of an initial 90∘90^\circ radio-frequency pulse followed by a train of 180∘180^\circ refocusing pulses. Echo signals were recorded with signal-to-noise ratios greater than 100 to ensure reliable reconstruction of the T2\mathrm{T}_2 relaxation-time distributions. Additional acquisition parameters comprised 16,250 echoes, an echo time of 200 μ\mus, 64 averages, a waiting time of 10 s between averages, and a measurement temperature of 25 °C.

The measured CPMG echo-decay signals were represented as multiexponential relaxation functions, and the corresponding T2T_2 distributions were obtained through mathematical inversion using an inverse Laplace transform. The resulting distributions were represented on a common logarithmically spaced grid comprising 121 relaxation-time bins ranging from 0.01 to 10,000 ms, with a spacing of Δlog⁡10(T2)=0.05\Delta\log_{10}(T_2) = 0.05. The specific regularization method and associated parameter settings employed during the original NMR inversion could not be recovered from the available experimental records.

The dominant relaxation time, T2peakT_{2\mathrm{peak}}, was defined as the relaxation time corresponding to the maximum incremental-porosity amplitude in the T2T_2 distribution. For multimodal distributions, the global maximum was selected, irrespective of the presence of secondary peaks. This criterion was applied consistently to all 14 samples. Figure 1 (PDF p. 17) presents the resulting NMR T2T_2 distributions for the investigated carbonate samples.

Image not extracted

Figure 1 (PDF p. 17). The NMR T2T_2 distributions for 14 samples analyzed in this study.

The bulk volume irreducible (BVI) and free fluid index (FFI) were determined from the T2T_2 distributions using a fixed T2T_2 cutoff of 100 ms for all samples, consistent with the original experimental procedure reported by Martins (2019) and Lima (2020). The integrated signal amplitude below the cutoff was assigned to BVI, whereas the integrated amplitude above the cutoff was assigned to FFI. T2T_2 cutoff values are ideally established from laboratory calibration and depend on lithology. Whereas a value near 33 ms is commonly adopted for sandstones, substantially higher and more variable cutoffs, typically within approximately 90–200 ms, have been reported for carbonate rocks [47]. Table 2 summarizes the experimentally measured petrophysical properties of the analyzed samples, including gas porosity, NMR porosity, gas permeability, Klinkenberg-corrected permeability, and formation factor.

Sampleϕgas\phi_{\mathrm{gas}} (%)ϕNMR\phi_{\mathrm{NMR}} (%)kgask_{\mathrm{gas}} (mD)kgas−Klinkenbergk_{\mathrm{gas-Klinkenberg}} (mD)FFT2lmT_{2lm} (ms)T2peakT_{2peak} (ms)FFIBVI
118.217.770.868.339.3262.5562.314.23.5
216.315.424.122.958.864.8141.37.08.5
319.617.5504.4482.337.8529.01122.015.52.0
410.29.45.34.757.251.0112.23.85.7
58.795.24.8115.7133.089.15.04.0
619.817.9232.9227.035.1437.61122.015.22.7
718.517.9103.599.027.956.1177.88.39.6
818.418649.9636.648.1398.11000.015.03.0
924.924.684.081.218.9105.7281.814.310.3
108.78.90.40.4130.223.339.81.97.0
1112.212.15.74.846.175.4501.26.45.7
1212.113.622.118.962.3145.1631.08.74.9
1316.816.3245.1173.168.0749.31412.514.51.8
1412.112.4152.4107.650.2461.2891.310.91.5

Table 2. Salient properties of the carbonate samples analyzed in this study.

Fourteen plots of incremental porosity versus T₂ (ms), labelled Sample 1 through Sample 14, with axes ranging from 0.01 to 10000 ms and incremental porosity from 0 to 1

Figure 4.

Estimating Permeability from NMR T2T_2 Distribution

The pore-geometry factor csc_s accounts for the relationship between pore surface area and volume in the NMR relaxation model. For idealized cylindrical and spherical pores, cs=4c_s = 4 and cs=6c_s = 6, respectively. However, natural carbonate rocks exhibit complex and irregular pore geometries that cannot be adequately described by either idealization. In the absence of independent pore-scale geometric measurements, we adopted an intermediate effective value of cs=5c_s = 5 to estimate kk using the physics-based model, Eq. (5). This assumption does not imply that carbonate pores possess a uniform or intermediate geometry, but provides a practical approximation for converting NMR relaxation times into characteristic pore sizes. Although permeability predictions depend quadratically on csc_s in the adopted CPA formulation, varying this factor between 4 and 6 changes predicted permeability by factors of 0.64–1.44 relative to cs=5c_s = 5, with all other parameters held constant. Therefore, uncertainty associated with this approximation is expected to be relatively modest compared with the potentially larger uncertainties associated with surface relaxivity and the representation of pore-network connectivity. Nevertheless, csc_s should be regarded as an effective parameter whose value may vary among carbonate pore systems.

Following Skaggs [51], the CPA proportionality constant was assumed to be CCPA=72.2C_{CPA} = 72.2. In the CPA permeability model, Eq. (3), the coefficient CCPAC_{CPA} depends on the underlying theoretical assumptions, the definition of the critical pore-throat length scale, and pore-network characteristics. Katz and Thompson [27] originally proposed CCPA=226C_{CPA} = 226, whereas subsequent studies have considered alternative values. Ghanbarian et al. [20] systematically evaluated several previously proposed coefficients for tight-gas sandstones and found that CCPA=72.2C_{CPA} = 72.2 was among those yielding accurate permeability estimates, although CCPA=53.5C_{CPA} = 53.5 provided the best agreement with their measurements. Subsequently, Ghanbarian (2020) applied CPA with CCPA=72.2C_{CPA} = 72.2 to 105 uniform sand and glass-bead packs and obtained an RMSLE of 0.16. Based on these previous evaluations, we adopted CCPA=72.2C_{CPA} = 72.2 as the baseline coefficient in the present study and further assessed its influence on permeability predictions through sensitivity analysis.

Because effective transverse surface relaxivity ρ2e\rho_{2e} was not independently determined for the Morro do Chaves samples, a literature-based estimate was adopted. Jácomo et al. [26] reported effective surface relaxivity values ranging from 0.22 to 18.65 μm/s\mu\mathrm{m}/\mathrm{s} for nine carbonate samples from the Barra Velha Formation, based on the integration of NMR relaxation measurements and mercury intrusion capillary pressure (MICP) data. The arithmetic mean of these values, ρ2e=7.2\rho_{2e} = 7.2 μm/s, was adopted in the present study. This estimate is also reasonably close to the mean value of 8.05 μm/s reported by August et al. [5] for four lithofacies from the same formation. However, the Morro do Chaves Formation is dominated by reworked bioclastic coquinas and is more closely comparable, in terms of lithofacies, to the bioclastic coquina intervals of the Itapema Formation than to the carbonate facies represented by the Barra Velha samples. To the best of our knowledge, no published, methodologically comparable surface relaxivity estimates are available for the Itapema Formation. Furthermore, because the literature-derived values were inferred from relationships between NMR relaxation characteristics and MICP-derived pore-throat dimensions, ρ2e\rho_{2e} should be interpreted as an effective conversion parameter rather than a direct measurement of intrinsic surface relaxivity. Such relationships do not necessarily establish that the inferred characteristic length scale corresponds to the critical pore-throat diameter governing permeability. Accordingly, the adopted value of ρ2e=7.2\rho_{2e} = 7.2 μm/s represents an externally constrained approximation whose applicability to the Morro do Chaves carbonates remains uncertain because of potential differences in mineralogy, pore structure, and pore-body-to-throat relationships. Independent measurements of NMR relaxation characteristics and critical pore-throat dimensions on the same samples would be necessary to evaluate this approximation directly.

To compare the predictive performance of the proposed physics-based model with empirical NMR permeability relationships, we evaluated the generalized TC and SDR formulations calibrated by Machado et al. [33] for carbonate samples from the Santos Basin, Brazil, as well as other models reported in Table 1. Martins [35] previously calibrated generalized TC and SDR models using 45 carbonate samples from the Morro do Chaves Formation. However, those models were not directly evaluated in the present study because the current dataset partially overlaps with the database used by Martins [35]. More specifically, the 14 samples analyzed here represent a subset of the larger dataset, and formation factor measurements were available only for these 14 samples.

To evaluate the predictive performance of the permeability models, two complementary statistical metrics were employed: the root mean squared logarithmic error (RMSLE) and the mean logarithmic prediction error (MLPE) defined as

RMSLE=∑i=1N[log⁡10(kesti)−log⁡10(kmeasi)]2N(6)RMSLE = \sqrt{\frac{\sum_{i=1}^{N}\left[\log_{10}\left(k_{\mathrm{est}}^{i}\right)-\log_{10}\left(k_{\mathrm{meas}}^{i}\right)\right]^2}{N}} \tag*{(6)}
MLPE=1N∑i=1Nlog⁡10(kestikmeasi)(7)MLPE = \frac{1}{N}\sum_{i=1}^{N}\log_{10}\left(\frac{k_{\mathrm{est}}^{i}}{k_{\mathrm{meas}}^{i}}\right) \tag*{(7)}

where N=14N=14 is the number of samples, and kestk_{\mathrm{est}} and kmeask_{\mathrm{meas}} are the estimated and measured permeability values, respectively.

RMSLE quantifies the overall magnitude of prediction errors, whereas MLPE characterizes systematic prediction bias. Both metrics were calculated using base-10 logarithms. An MLPE value of zero indicates no systematic multiplicative bias, whereas positive and negative values indicate tendencies toward overestimation and underestimation of permeability, respectively. The combined use of RMSLE and MLPE enables assessment of both predictive accuracy and systematic bias without requiring calibration of model parameters against the investigated permeability dataset.

Results and Discussion

Figure 2 shows the relationships between Klinkenberg-corrected permeability and T2lmT_{2lm}, NMR porosity, T2peakT_{2peak}, and the reciprocal of formation factor for the 14 carbonate samples from the Morro do Chaves Formation. As shown in Figure 2a, permeability generally increases with increasing T2lmT_{2lm} following a power-law trend with an exponent of approximately 1.61. However, the relatively low coefficient of determination (R2=0.28R^2 = 0.28) indicates substantial scatter in the data. We also observed a weak relationship between permeability and NMR porosity (Figure 2b). Although stronger porosity–permeability correlations have been reported for some carbonate datasets (e.g., Shabani et al. [49]), such behavior is often associated with smaller datasets, more homogeneous lithologies, or carbonate systems with less complex pore architecture (Lucia [32]). For example, Ng and Santamarina (2023) analyzed several hundred carbonate samples compiled from the literature and showed that permeability spanning six orders of magnitude was only weakly correlated with porosity because of the highly variable and heterogeneous pore connectivity characteristic of carbonate rocks.

Four scatter plots of Klinkenberg-corrected permeability against T2lm, NMR porosity, T2peak, and reciprocal of formation factor

Figure 2. Klinkenberg-corrected permeability against (a) T2lmT_{2lm}, (b) NMR porosity, (c) T2peakT_{2peak}, and (d) reciprocal of formation factor for 14 carbonate samples from the Morro do Chaves formation.

Figure 2c presents Klinkenberg-corrected permeability as a function of T2peakT_{2peak} on a logarithmic scale. Compared with T2lmT_{2lm}, T2peakT_{2peak} exhibited a stronger association with permeability (R2=0.41R^2 = 0.41 vs. 0.28). This result supports its use as the characteristic relaxation parameter evaluated here, but does not establish that uniquely represents the critical pore-throat diameter. Rather, the proposed correspondence between T2peakT_{2peak} and the flow-controlling length scale should be regarded as a testable approximation.

Figure 2d shows the Klinkenberg-corrected permeability versus the reciprocal of the formation factor (1/F1/F). Following the theoretical expectation of CPA-based permeability models [27, 51], a linear relationship passing through the origin was fitted to the data. Although permeability generally increased with increasing 1/F1/F, the data remained scattered, further reflecting the heterogeneous and complex pore structure of the Morro do Chaves carbonate samples.

Figure 3 compares permeability estimated using the proposed physics-based model, Eq. (5), with predictions from generalized TC and SDR models. The physics-based model produced a root mean square logarithmic error (RMSLE) of 0.56 (Figure 3a), substantially lower than the errors associated with the generalized TC and SDR formulations (Figure 3b and Figure 3c). To assess the sensitivity of the CPA–NMR permeability predictions to the coefficient CCPAC_{CPA}, we evaluated three previously proposed values, CCPA=53.5C_{CPA} = 53.5, 72.2, and 226, while maintaining the effective surface relaxivity at ρ2e=7.2 μm/s\rho_{2e} = 7.2\ \mathrm{\mu m/s} and keeping all other model inputs unchanged. The corresponding RMSLE values were 0.59, 0.56, and 0.67, respectively. Thus, the baseline value of CCPA=72.2C_{CPA} = 72.2 yielded the lowest prediction error among the three coefficients considered, although the differences were relatively modest. These results indicate that the overall prediction error is less sensitive to the choice of CCPAC_{CPA} within the investigated range than to variations in effective surface relaxivity. Nevertheless, the appropriate coefficient may depend on the underlying CPA formulation, pore-network geometry, and definition of the characteristic critical pore-throat dimension [51]. Accordingly, the present sensitivity analysis supports the use of CCPA=72.2C_{CPA} = 72.2 for the investigated dataset but does not establish its universal applicability to heterogeneous carbonate rocks.

Estimated versus measured permeability for 14 carbonate samples

Figure 3. Estimated versus measured permeability for 14 carbonate samples analyzed in this study. Estimated permeabilities were obtained using the physics-based model (Eq. (5)) as well as the generalized TC and SDR models calibrated by Machado et al. [33].

To further investigate the influence of the characteristic NMR relaxation time on permeability prediction, we evaluated Eq. (5) by replacing the dominant relaxation time, T2peakT_{2peak}, with the geometric mean relaxation time, T2lmT_{2lm}, while keeping all other model parameters unchanged. The resulting RMSLE increased from 0.56 to 0.77, corresponding to an approximately 38% increase in prediction error. Furthermore, the MLPE was -0.6 (compared to 0.1 found for T2peakT_{2peak}), indicating systematic underestimation of permeability when T2lmT_{2lm} was used. These results suggest that T2peakT_{2peak} is a more suitable characteristic relaxation time than T2lmT_{2lm} for permeability estimation using the proposed CPA–NMR formulation in the investigated carbonate samples. Nevertheless, the improved predictive performance does not necessarily establish that T2peakT_{2peak} directly represents the critical pore-throat diameter controlling fluid flow.

The proposed CPA–NMR formulation, Eq. (5), achieved an RMSLE of 0.56, with 13 of the 14 estimated permeability values falling within a factor of 10 of their corresponding measured values. The only exception was Sample 11, for which the estimated permeability was 97.7 mD compared with a measured value of 4.79 mD, representing an overprediction by a factor of approximately 20.6. This discrepancy highlights that, despite the relatively low overall prediction error, substantial deviations may occur for individual samples, potentially reflecting limitations in representing complex pore-throat characteristics using a single NMR-derived characteristic relaxation time.

The generalized TC model, however, systematically underestimated permeability and the generalized SDR model considerably overestimated permeability. Similar overestimation and underestimation behavior of TC- and SDR-type models has been widely reported in the literature [13, 56, 23]. For instance, Kendrick et al. [28] reported that an SDR model calibrated for unconsolidated alluvial aquifers substantially overestimated hydraulic conductivity when applied to glacial aquifers, despite comparable hydraulic conductivity ranges among the studied formations.

The differences in predictive performance among the evaluated models were further examined using MLPE to distinguish systematic overestimation from underestimation. The proposed CPA–NMR formulation, E. (5), exhibited a relatively small positive bias (MLPE = 0.1), whereas the generalized TC and SDR formulations showed varying degrees of systematic underestimation or overestimation, depending on their calibration datasets and parameterizations. These results indicate that the proposed formulation achieved comparatively low prediction error and systematic bias for the investigated carbonate samples without local permeability calibration. Nevertheless, the limited sample size and formation-specific nature of the evaluation preclude conclusions regarding universal predictive superiority.

Recently, Chen et al. [9] developed a data-driven approach for estimating matrix permeability in sandstone aquifers using NMR T2T_{2} relaxation data and machine learning algorithms. Using a dataset of 132 sandstone core samples spanning seven orders of magnitude in permeability, they demonstrated that classical NMR permeability models such as SDR and TC produced relatively large prediction errors, whereas machine-learning approaches substantially improved predictive accuracy. In particular, the gradient boosting decision tree model achieved the best performance with a correlation coefficient of approximately 0.92. Their SHAP analysis further showed that different portions of the T2T_{2} spectrum contribute differently to permeability prediction, highlighting the important role of intermediate pore sizes in controlling flow.

Bias in generalized TC and SDR permeability models has been attributed to several factors, including uncertainty in selecting T2cutoffT_{2cutoff} [55], the presence of large but poorly connected vugs [52], organic matter [36], and strong formation heterogeneity [14]. In carbonate rocks, pore connectivity may be as important as pore size itself. Wang et al. (2018), for example, analyzed Madison limestone samples and showed that strong long-T2T_2 responses may originate from large but poorly connected vugs, causing SDR-type interpretations to overestimate permeability relative to the actual flow capacity of the rock.

An important distinction between the proposed CPA–NMR formulation, Eq. (5), and conventional SDR and TC models concerns their calibration requirements. The proposed model estimates permeability without fitting its parameters to measured permeability from the investigated carbonate samples, whereas SDR and TC formulations generally require empirical calibration to establish formation-specific coefficients. Accordingly, the present comparison evaluates the ability of a locally uncalibrated, physics-based formulation to predict permeability relative to empirical relationships calibrated using data from other formations. Although locally calibrated SDR and TC models might achieve improved predictive accuracy, such calibration requires measured permeability from the target formation and therefore represents a different application scenario. Consequently, the results should be interpreted as demonstrating the potential of the proposed approach for permeability prediction without local core-permeability calibration, rather than establishing the universal superiority of its functional form over empirical SDR and TC relationships.

The improved predictive performance of Eq. (5) relative to the generalized TC and SDR formulations may be partly attributed to the explicit incorporation of formation factor and surface relaxivity. The formation factor provides information about electrical transport through the connected pore network and is commonly related to effective electrical tortuosity through τe=ϕF\tau_e = \phi F [19], assuming negligible surface conduction. Here, τe\tau_e represents electrical rather than geometrical or hydraulic tortuosity. Although electrical and hydraulic transport are both influenced by pore-network connectivity, tortuosity, and constrictions, their respective pathways and sensitivities to pore-throat geometry may differ, particularly in heterogeneous carbonate rocks. Thus, formation factor should be interpreted as an indirect indicator of pore-network transport characteristics rather than a direct measure of hydraulic tortuosity or connectivity. Arns et al. [4] used three-dimensional tomographic images of Fontainebleau sandstone to evaluate permeability predictors based on pore volume-to-surface-area ratio, critical channel diameter, NMR-derived length scales, and formation factor. Their results demonstrated the importance of critical pore dimensions and formation factor in permeability estimation, highlighting the complementary roles of pore-throat geometry and electrical transport properties. Accordingly, incorporating formation factor alongside NMR-derived characteristic pore dimensions provides a physically motivated means of accounting for aspects of pore-network structure not explicitly represented in conventional TC and SDR formulations, although it does not fully resolve the differences between electrical and hydraulic transport pathways.

Table 3 summarizes the RMSLE and MLPE calculated for the permeability models listed in Table 1, excluding the formulations of Martins [35], whose calibration dataset partially overlaps with the samples investigated in this study. Among the evaluated models, the proposed physics-based CPA–NMR formulation, Eq. (5), achieved the lowest RMSLE (0.56) and the smallest MLPE (0.1), indicating the highest predictive accuracy and lowest systematic logarithmic bias for the investigated samples. The positive MLPE indicates a slight overall tendency toward permeability overestimation.

Table 3. Root mean squared logarithmic error (RMSLE) and mean logarithmic prediction error (MLPE) for the proposed CPA–NMR formulation, Eq. (5), and previously published SDR and Timur-Coates permeability models evaluated using 14 carbonate samples from the Morro do Chaves Formation. Both metrics were calculated using base-10 logarithms. The models of Martins [35] were excluded from the quantitative comparison because their calibration dataset overlaps with the samples investigated in this study.

ReferenceModelRMSLEMLPE
Amabeoku et al. [2]k=0.0574ϕ3.79T2lm1.89k = 0.0574\phi^{3.79}T_{2lm}^{1.89}1.91-1.9
Amabeoku et al. [2]k=511.4ϕ2.84(FFIBVI)2k = 511.4\phi^{2.84}\left(\frac{FFI}{BVI}\right)^20.77-0.6
Hassall et al. [24]k=0.5ϕ2(ρ2T2lm)2k = 0.5\phi^2(\rho_2T_{2lm})^22.592.5
Machado et al. [33]k=7.3×10−11ϕ2.68T2lm4.56k = 7.3 \times10^{-11}\phi^{2.68}T_{2lm}^{4.56}2.061.4
Machado et al. [33]k=(ϕ109.64)1.74(FFIBVI)3.85k = \left(\frac{\phi}{109.64}\right)^{1.74}\left(\frac{FFI}{BVI}\right)^{3.85}2.23-1.9
Aghda et al. [1]k=0.0216ϕ1.64T2lm1.68k = 0.0216\phi^{1.64}T_{2lm}^{1.68}1.01-0.9
Aghda et al. [1]k=(ϕ0.222)3.9(FFIBVI)0.51k = \left(\frac{\phi}{0.222}\right)^{3.9}\left(\frac{FFI}{BVI}\right)^{0.51}2.18-2.1
Martins [35]k=0.12×104ϕ4.48T2lm1.0k = 0.12 \times10^{4}\phi^{4.48}T_{2lm}^{1.0}--
Martins [35]k=3.45×104ϕ5.05(FFIBVI)0.78k = 3.45 \times10^{4}\phi^{5.05}\left(\frac{FFI}{BVI}\right)^{0.78}--
Shabani et al. [48]k=0.0004(ϕ10)3T2lm1.6k = 0.0004\left(\frac{\phi}{10}\right)^3T_{2lm}^{1.6}1.04-1.0
Shabani et al. [48]k=0.022(ϕ10)4(FFIBVI)2k = 0.022\left(\frac{\phi}{10}\right)^4\left(\frac{FFI}{BVI}\right)^22.04-2.0
This studyEq. (5)0.560.1

Table 3.

The second-lowest RMSLE (0.78) was obtained using the generalized Timur–Coates (TC) formulation proposed by Amabeoku et al. [2], which was calibrated using seven carbonate samples from an Arab reservoir. However, its MLPE of −0.6 indicates systematic underestimation of permeability. The generalized SDR formulations developed by Aghda et al. [1] and Shabani et al. [48] yielded comparable RMSLE values of 1.01 and 1.04, respectively, with corresponding MLPE values of −0.9 and −1.0, indicating systematic underestimation. Both formulations were developed using carbonate samples from the Asmari Formation in Iran. Notably, Shabani et al. [48] reportedly calibrated their generalized SDR model using a single carbonate sample. Estimating three independent fitting coefficients from one sample would constitute an underdetermined calibration problem unless additional constraints were imposed, raising concerns regarding parameter identifiability and generalizability. Overall, these results indicate that the proposed CPA–NMR formulation, Eq. (5), provides more accurate permeability predictions than the evaluated externally calibrated empirical relationships for the present dataset, without requiring calibration against measured permeability from the investigated formation.

The results presented in Table 3 demonstrate that some empirically calibrated TC and SDR formulations can provide reasonable permeability estimates for carbonate rocks. However, empirical data-driven models are inherently database dependent, meaning that their predictive capability is often restricted to formations similar to those used during calibration [46, 17]. In contrast, physics-based formulations, provided that their input parameters are reasonably estimated, offer the potential to predict permeability from physically meaningful transport relationships rather than purely empirical fitting.

An important limitation of the present study is the relatively small number of investigated samples and their origin from a single geological formation. The objective was not to establish the statistical representativeness of these samples with respect to the Morro do Chaves Formation, but rather to assess whether the CPA–NMR formulation, Eq. (5), could estimate permeability with reasonable accuracy without calibration against locally measured permeability, given the availability of NMR T2T_{2} relaxation distributions and independently measured electrical formation factors. Although the results support the feasibility of this approach for the investigated samples, the limited sample size introduces uncertainty into the estimated prediction errors, and the observed performance does not establish applicability to other carbonate formations or lithologies. Broader validation is constrained by the limited availability of datasets containing these complementary measurements for the same samples. Furthermore, the effective surface relaxivity adopted from a different carbonate formation may not adequately represent rocks with substantially different mineralogical compositions or pore structures. Consequently, the present findings should be interpreted as an initial evaluation of the proposed formulation, and its predictive reliability and transferability should be investigated using larger, independent datasets encompassing a broader range of geological formations and pore-system characteristics.

Sensitivity to the surface relaxivity

To evaluate the sensitivity of the proposed permeability model to the assumed effective transverse surface relaxivity, ρ2e\rho_{2e}, we performed a scenario-based sensitivity analysis using values of 0.22, 2.0, 7.2, 12.0, and 18.65 μm/s. The lower and upper bounds correspond to the range reported by Jácomo et al. [26] for carbonate samples from the Barra Velha Formation, whereas 7.2 μ\mum/s represents the baseline value adopted in this study. For each scenario, permeability was estimated for all 14 carbonate samples using Eq. (5), while maintaining the remaining model parameters unchanged. The corresponding RMSLE was then calculated to quantify the influence of surface relaxivity on prediction accuracy. Because permeability scales quadratically with ρ2e\rho_{2e}, this analysis provides a direct assessment of the sensitivity of the model predictions to uncertainties associated with transferring surface relaxivity estimates between carbonate formations.

Results presented in Table 4 demonstrate that permeability predictions are highly sensitive to the assumed surface relaxivity. At the lower bound of ρ2e=0.22\rho_{2e} = 0.22 μ\mum/s, the model systematically underestimated permeability, with an MLPE of −2.9 and an RMSLE of 2.98. Increasing ρ2e\rho_{2e} to 2.0 μ\mum/s substantially reduced the RMSLE to 1.15, although permeability remained predominantly underestimated. The lowest RMSLE among the five scenarios (0.56) was obtained using the baseline value of 7.2 μ\mum/s, for which the MLPE was approximately 0.1, indicating relatively small overall systematic bias. Further increases in ρ2e\rho_{2e} to 12.0 and 18.65 μ\mum/s increased the RMSLE to 0.77 and 1.08, respectively, accompanied by increasingly positive MLPEs of 0.5 and 0.9. These results indicate a transition from systematic underestimation at low surface relaxivity to systematic overestimation at higher values.

Sampleρ2e=0.22\rho_{2e}=0.22 μm/s kestk_{est} (mD)ρ2e=0.22\rho_{2e}=0.22 μm/s log⁡(kest/kmeas)\log(k_{est}/k_{meas})ρ2e=2\rho_{2e}=2 μm/s kestk_{est} (mD)ρ2e=2\rho_{2e}=2 μm/s log⁡(kest/kmeas)\log(k_{est}/k_{meas})ρ2e=7.2\rho_{2e}=7.2 μm/s kestk_{est} (mD)ρ2e=7.2\rho_{2e}=7.2 μm/s log⁡(kest/kmeas)\log(k_{est}/k_{meas})ρ2e=12\rho_{2e}=12 μm/s kestk_{est} (mD)ρ2e=12\rho_{2e}=12 μm/s log⁡(kest/kmeas)\log(k_{est}/k_{meas})ρ2e=18.65\rho_{2e}=18.65 μm/s kestk_{est} (mD)ρ2e=18.65\rho_{2e}=18.65 μm/s log⁡(kest/kmeas)\log(k_{est}/k_{meas})
10.1347-2.711.13-0.8144.30.3400.80.8968.11.2
20.0057-3.60.47-1.76.1-0.616.9-0.140.90.3
30.5580-2.946.12-1.0597.70.11660.30.54010.40.9
40.0037-3.10.30-1.24.0-0.111.00.426.50.8
50.0012-3.60.10-1.71.2-0.63.4-0.18.30.2
60.6012-2.649.69-0.7643.90.51788.70.94320.61.3
70.0190-3.71.57-1.820.3-0.756.5-0.2136.50.1
80.3482-3.328.77-1.3372.9-0.21035.80.22502.00.6
90.0705-3.15.83-1.175.50.0209.80.4506.60.8
100.0002-3.20.02-1.30.2-0.20.60.21.50.6
110.0913-1.77.540.297.81.3271.61.8656.02.1
120.1071-2.28.85-0.3114.80.8318.81.2770.01.6
130.4917-2.540.63-0.6526.60.51462.80.93533.31.3
140.2652-2.621.91-0.7284.00.4788.90.91905.51.2
MLPE-2.9-1.00.10.50.9
RMSLE2.981.150.560.771.08

Table 4. Sensitivity of permeability estimates from Eq. (5) to effective transverse surface relaxivity ρ2e\rho_{2e} for 14 carbonate samples from the Morro do Chaves Formation. Estimated permeability kestk_{est}, 10-based logarithmic residuals log⁡(kest/kmeas)\log(k_{est}/k_{meas}), mean logarithmic residuals, and RMSLE are reported for five assumed ρ2e\rho_{2e} values. The baseline value of 7.2 μm/s and the lower and upper literature bounds of 0.22 and 18.65 μm/s are based on Jácomo et al. [26].

The pronounced sensitivity of permeability predictions to ρ2e\rho_{2e} is consistent with the quadratic dependence in Eq. (5). Relative to the baseline value of 7.2 μ\mum/s, adopting the lower and upper bounds changes the predicted permeability by multiplicative factors of approximately 0.00093 and 6.71, respectively. Nevertheless, the baseline value produced the lowest RMSLE among the five scenarios, suggesting that the literature-based estimate of 7.2 μm/s provides a reasonable effective surface relaxivity for the investigated samples. This agreement should not, however, be interpreted as independent validation of the assumed value or evidence that surface relaxivity is uniform across the samples. The sensitivity analysis assumes a common ρ2e\rho_{2e} for all 14 samples, whereas variations in mineralogy, pore geometry, and surface chemistry may produce sample-specific differences in effective surface relaxivity. Furthermore, the literature-derived range represents variability observed in the Barra Velha Formation rather than a statistically defined uncertainty interval for the Morro do Chaves samples. Consequently, although the proposed model exhibits its lowest prediction error near the adopted baseline value among the scenarios evaluated, its predictive accuracy and transferability remain dependent on reliable characterization of surface relaxivity. Independent measurements of ρ2e\rho_{2e} on the investigated samples and additional carbonate formations are therefore needed to establish the broader applicability of the proposed framework.

Conclusions

In this study, we developed and evaluated a physics-based permeability estimation approach that integrates NMR transverse relaxation characteristics, electrical formation factor, and effective surface relaxivity within a critical path analysis (CPA) framework.

The proposed CPA–NMR formulation, Eq. (5), was evaluated using 14 heterogeneous carbonate samples from the Morro do Chaves Formation without calibrating model parameters against their measured permeability. The model achieved a root mean squared logarithmic error (RMSLE) of 0.56, with 13 of the 14 permeability predictions falling within one order of magnitude of the measured values. Its predictive accuracy exceeded that of the externally calibrated generalized Timur–Coates (TC) and Schlumberger–Doll Research (SDR) formulations considered in this study. However, this comparison reflects the performance of the selected models under conditions without local permeability calibration and does not establish the inherent superiority of the CPA–NMR formulation over locally calibrated empirical alternatives.

The results suggest that combining NMR-derived characteristic length scales with independently measured electrical formation factor provides a physically motivated approach to permeability estimation in heterogeneous carbonate rocks. Within the proposed formulation, the dominant relaxation time, T2peakT_{2\mathrm{peak}}, yielded more accurate predictions than the geometric mean relaxation time, T2lmT_{2\mathrm{lm}}. Nevertheless, the improved predictive performance does not independently establish that T2peakT_{2\mathrm{peak}} represents the critical pore-throat dimension controlling fluid flow. Similarly, although formation factor provides complementary information about electrical transport through the connected pore network, the relative contributions of pore geometry, electrical connectivity, and effective surface relaxivity to the observed predictive improvement remain unresolved.

Sensitivity analysis further demonstrated a strong dependence of predicted permeability on effective surface relaxivity, with RMSLE increasing from 0.56 at the baseline value of 7.2 μm/s7.2\ \mu\mathrm{m}/\mathrm{s} to 2.98 and 1.08 at the lower and upper literature bounds, respectively.

The principal limitations of this study are the relatively small dataset, the absence of independently measured effective surface relaxivity and critical pore-throat dimensions, and the uncertainty associated with applying externally determined model parameters to the investigated carbonate samples. One sample exhibited a permeability overprediction of approximately 20-fold, illustrating the potential limitations of representing complex pore systems using a single characteristic NMR relaxation time. Consequently, the present findings constitute an initial evaluation of the CPA–NMR formulation rather than evidence of its general transferability across geological formations or lithologies. Future studies should evaluate the approach using larger, independent datasets containing NMR relaxation distributions, electrical formation factor, measured permeability, and independently constrained effective surface relaxivity for the same samples. Complementary pore-network characterization, including mercury intrusion capillary pressure measurements or three-dimensional imaging, would also help establish the relationship between NMR-derived characteristic dimensions and the critical pore-throat constrictions governing permeability.

Acknowledgement

BG is grateful to the University of Texas at Arlington for financial support through faculty startup fund and the STARs award. The authors used AI-assisted language editing tools to improve grammar and readability. All scientific interpretations and conclusions remain the responsibility of the authors.

Conflict of Interest

The authors declare no conflicts of interest relevant to this study.

Data Availability Statement

Data are available and are given by Ghanbarian et al. [22].

References

  1. [1]Aghda, S. F., Taslimi, M., & Fahimifar, A. (2018). Adjusting porosity and permeability estimation by nuclear magnetic resonance: a case study from a carbonate reservoir of south of Iran. Journal of Petroleum Exploration and Production Technology, 8(4), 1113-1127.DOI
  2. [2]Amabeoku, M. O., Funk, J. J., Al-Dossary, S. M., & Al-Ali, H. A. (2001). Calibration of permeability derived from NMR logs in carbonate reservoirs. In SPE Middle East Oil and Gas Show and Conference. pp. SPE-68085.
  3. [3]Archie, G. E. (1942). The electrical resistivity log as an aid in determining some reservoir characteristics. Trans. AIME, 146, 54-62.DOI
  4. [4]Arns, C. H., Knackstedt, M. A., & Martys, N. S. (2005). Cross-property correlations and permeability estimation in sandstone. Physical Review E, 72(4), 046304.pubmed.ncbi.nlm.nih.gov/16383530
  5. [5]August, H., Azizoglu, Z., Heidari, Z., Goncalves, L., de Oliveira, L. A. B., do Nascimento Neto, M. S., & Victor, R. A. (2025). Novel estimation of pore-throat size and permeability in carbonates from the integration of NMR and resistivity data in the Santos Basin of Brazil. Petroleum Geoscience, 31(2), petgeo2024-093.DOI
  6. [6]Behroozmand, A. A., Keating, K., & Auken, E. (2015). A review of the principles and applications of the NMR technique for near-surface characterization. Surveys in Geophysics, 36(1), 27-85.DOI
  7. [7]Bernabé, Y., Zamora, M., Li, M., Maineult, A., & Tang, Y. B. (2011). Pore connectivity, permeability, and electrical formation factor: A new model and comparison to experimental data. Journal of Geophysical Research: Solid Earth, 116, B11204.DOI
  8. [8]Cardona, A., & Santamarina, J. C. (2020). Carbonate rocks: Matrix permeability estimation. AAPG Bulletin, 104(1), 131-144.DOI
  9. [9]Chen, X., Zhao, X., Tahmasebi, P., Luo, C., & Cai, J. (2023). NMR-data-driven prediction of matrix permeability in sandstone aquifers. Journal of Hydrology, 618, 129147.DOI
  10. [10]Chi, L., & Roth, S. (2018). A new directional permeability model combining NMR and directional resistivity measurements for complex formations. In SPWLA Annual Logging Symposium. p. D033S003R011.
  11. [11]Coates, G.R., Xiao, L., and Prammer, M.G. (1999). NMR Logging: Principles and Applications. Halliburton Energy Services.
  12. [12]David, C., Gueguen, Y., & Pampoukis, G. (1990). Effective medium theory and network theory applied to the transport properties of rock. Journal of Geophysical Research: Solid Earth, 95(B5), 6993-7005.DOI
  13. [13]Dlubac, K., Knight, R., Song, Y. Q., Bachman, N., Grau, B., Cannia, J., & Williams, J. (2013). Use of NMR logging to obtain estimates of hydraulic conductivity in the High Plains aquifer, Nebraska, USA. Water Resources Research, 49(4), 1871-1886.DOI
  14. [14]Eeva, S., Karjalainen, A., Koivisto, E., Korkka-Niemi, K., Rautio, A., Räisänen, O., ... & Birt, B. (2023). Hydrogeological characterization of crystalline bedrock using borehole magnetic resonance. Groundwater, 61(6), 793-815.DOI
  15. [15]Gao, Z., Niu, F., Wang, Y., Lin, Z., & Wang, W. (2021). Superpermafrost groundwater flow and exchange around a thermokarst lake on the Qinghai–Tibet Plateau, China. Journal of Hydrology, 593, 125882.DOI
  16. [16]Ghanbarian, B., & Javadpour, F. (2017). Upscaling pore pressure-dependent gas permeability in shales. Journal of Geophysical Research: Solid Earth, 122(4), 2541-2552.DOI
  17. [17]Ghanbarian, B., & Pachepsky, Y. (2022). Machine learning in vadose zone hydrology: A flashback. Vadose Zone Journal, 21(4), e20212.DOI
  18. [18]Ghanbarian, B., & Skaggs, T. H. (2022). Soil water retention curve inflection point: Insight into soil structure from percolation theory. Soil Science Society of America Journal, 86(2), 338-344.DOI
  19. [19]Ghanbarian, B., Hunt, A. G., Ewing, R. P., & Sahimi, M. (2013). Tortuosity in porous media: a critical review. Soil Science Society of America Journal, 77(5), 1461-1477.DOI
  20. [20]Ghanbarian, B., Torres-Verdín, C., & Skaggs, T. H. (2016). Quantifying tight-gas sandstone permeability via critical path analysis. Advances in Water Resources, 92, 316-322.DOI
  21. [21]Ghanbarian, B., Bonnie, R.J.M., & Chen, J. (2027). The Generalized SDR permeability model revisited: NMR-based interpretation, physical constraints, and sources of uncertainty. Petrophysics (Accepted).
  22. [22]Ghanbarian, B., Bermudez, S. L. B., Souza, A., Ribeiro, G. S. S., Lima, M. C. O., Couto, P. (2026). Morro do Chaves Formation carbonate rock samples, HydroShare, http://www.hydroshare.org/resource/422500b9fac44ba2863a180739610d90
  23. [23]Gonçalves, L., Victor, R. A., & Barroso, E. V. (2024). Enhancing permeability estimation methods in Brazilian pre-salt complex carbonate reservoirs. Geoenergy Science and Engineering, 243, 213345.DOI
  24. [24]Hassall, J. K., Ferraris, P., Al-Raisi, M., Hurley, N. F., Boyd, A., & Allen, D. F. (2004). Comparison of permeability predictors from NMR, formation image and other logs in a carbonate reservoir. In Abu Dhabi International Petroleum Exhibition and Conference. pp. SPE-88683.
  25. [25]Hulea, I. N., & Nicholls, C. A. (2012). Carbonate rock characterization and modeling: Capillary pressure and permeability in multimodal rocks—A look beyond sample specific heterogeneity. AAPG Bulletin, 96(9), 1627-1642.DOI
  26. [26]Jácomo, M. H., Hartmann, G. A., Lucas-Oliveira, E., Rojas, D., & Leite, E. P. (2025). Estimating nuclear magnetic resonance transverse surface relaxivity in pre-salt carbonates of the Santos Basin, Brazil. Journal of Applied Geophysics, 105913.DOI
  27. [27]Katz, A.J., and Thompson, A.H. (1986). Quantitative prediction of permeability in porous rock. Physical Review B, 34(11), 8179–8181.pubmed.ncbi.nlm.nih.gov/9939522
  28. [28]Kendrick, A. K., Knight, R., Johnson, C. D., Liu, G., Knobbe, S., Hunt, R. J., & Butler Jr, J. J. (2021). Assessment of NMR logging for estimating hydraulic conductivity in glacial aquifers. Groundwater, 59(1), 31-48.
  29. [29]Kenyon, W.E., Day, P.I., Straley, C., and Willemsen, J.F. (1988). A three-part study of NMR well logging of sedimentary rocks: Theory and laboratory results. SPE Formation Evaluation, 3(3), 622–636.
  30. [30]Lai, J., Wang, G., Fan, Z., Zhou, Z., Chen, J., & Wang, S. (2018). Fractal analysis of tight shaly sandstones using nuclear magnetic resonance measurements. AAPG Bulletin, 102(2), 175-193.DOI
  31. [32]Lucia, F. J. (1983). Petrophysical parameters estimated from visual descriptions of carbonate rocks: a field classification of carbonate pore space. Journal of Petroleum Technology, 35(03), 629-637.DOI
  32. [33]Machado, V. F., Azeredo, R. B. V., Ramos, P. F., Coutinho, M. R., & Carneiro, S. R. (2008). A NMR permeability model based on partial least square (PLS) regression analysis for a low permeability gas sand in Santos Basin. In International Symposium of the Society of Core Analysts (pp. 29-10).
  33. [34]Male, F., & Duncan, I. J. (2020). Lessons for machine learning from the analysis of porosity-permeability transforms for carbonate reservoirs. Journal of Petroleum Science and Engineering, 187, 106825.DOI
  34. [35]Martins, L.P. (2019). Analysis and estimation of permeability by nuclear magnetic resonance technique in coquinas from Morro Do Chaves formation, Sergipe-Alagoas Basin. Master’s thesis. Federal University of Rio de Janeiro.
  35. [36]Mashhadi, S. R., Keating, K., Petersen, R. J., Parra, A. O., Costabel, S., Beisembina, G., ... & Grombacher, D. (2025). Hydraulic conductivity estimation by NMR data in unconsolidated geological materials: insights from SDR model calibrations. Journal of Hydrology, 134060.DOI
  36. [37]Meng, F., Chen, Y., Reuschle, T., Baud, P., & Wong, T. F. (2026). Transport Properties of Dual-Porosity Limestones: Laboratory Measurement, Critical Path Analysis Prediction, and Image-Based Simulation. Journal of Geophysical Research: Solid Earth, 131(4), e2025JB033032.DOI
  37. [38]Mohnke, O., & Yaramanci, U. (2008). Pore size distributions and hydraulic conductivities of rocks derived from magnetic resonance sounding relaxation data using multi-exponential decay time inversion. Journal of Applied Geophysics, 66(3-4), 73-81.DOI
  38. [40]Nishiyama, N., & Yokoyama, T. (2017). Permeability of porous media: Role of the critical pore size. Journal of Geophysical Research: Solid Earth, 122(9), 6955-6971.DOI
  39. [41]Pirrone, M., & Galli, G. (2018). The dynamic character of porosity by nuclear magnetic resonance and production logging for a comprehensive permeability prediction in carbonate reservoirs. In Abu Dhabi International Petroleum Exhibition and Conference. Society of Petroleum Engineers.DOI
  40. [42]Rashid, F., Glover, P. W. J., Lorinczi, P., Hussein, D., Collier, R., & Lawrence, J. (2015). Permeability prediction in tight carbonate rocks using capillary pressure measurements. Marine and Petroleum Geology, 68, 536-550.DOI
  41. [43]Ren, S., Parsekian, A. D., Zhang, Y., & Carr, B. J. (2019). Hydraulic conductivity calibration of logging NMR in a granite aquifer, Laramie Range, Wyoming. Groundwater, 57(2), 303-319.DOI
  42. [44]Revil, A., Le Breton, M., Niu, Q., Wallin, E., Haskins, E., & Thomas, D. M. (2017). Induced polarization of volcanic rocks–1. Surface versus quadrature conductivity. Geophysical Journal International, 208(2), 826-844.DOI
  43. [45]Rezaee, R., Saeedi, A., & Clennell, B. (2012). Tight gas sands permeability estimation from mercury injection capillary pressure and nuclear magnetic resonance data. Journal of Petroleum Science and Engineering, 88, 92-99.DOI
  44. [46]Schaap, M. G., & Leij, F. J. (1998). Database-related accuracy and uncertainty of pedotransfer functions. Soil Science, 163(10), 765-779.DOI
  45. [47]Schön, J. H. (2015). Physical Properties of Rocks: Fundamentals and Principles of Petrophysics (2nd ed.). Elsevier.DOI
  46. [48]Shabani, M., Ghaffary, S., & Yarmohammadi, S. (2021). Evaluation of rock properties determined from core and NMR data: A case study on Asmari carbonate reservoir. Journal of Petroleum Science and Technology, 11(3), 2-10.DOI
  47. [49]Shabani, M., Yarmohammadi, S., & Ghaffary, S. (2023). Reservoir quality investigation by combination of core measured data and NMR technique analysis: a case study of Asmari carbonate reservoir in Gachsaran field. Carbonates and Evaporites, 38(1), 1.DOI
  48. [50]Sigal, R. (2002). Coates and SDR permeability: two variations on the same theme. Petrophysics- The SPWLA Journal, 43(01), 38-47.
  49. [51]Skaggs, T.H. (2011). Predicting permeability from electrical conductivity in saturated soils. Advances in Water Resources, 34, 1088–1096.
  50. [52]Wang, H., Alvarado, V., McLaughlin, J. F., Bagdonas, D. A., Kaszuba, J. P., Campbell, E., & Grana, D. (2018). Low-field nuclear magnetic resonance characterization of carbonate and sandstone reservoirs from rock spring uplift of Wyoming. Journal of Geophysical Research: Solid Earth, 123(9), 7444-7460.DOI
  51. [53]Weller, A., & Zhang, Z. (2022). A comparative study of permeability prediction for Eocene sandstones—Part 2: Application and modification of the Schlumberger-Doll Research equation. Geophysics, 87(5), M179-M188.DOI
  52. [54]Weller, A., & Zhang, Z. (2023). Modification of the SDR equation for permeability prediction. In E3S Web of Conferences. Vol. 367, p. 01014. EDP Sciences.DOI
  53. [55]Xiao, L., Liu, X. P., Zou, C. C., Hu, X. X., Mao, Z. Q., Shi, Y. J., ... & Li, G. R. (2014). Comparative study of models for predicting permeability from nuclear magnetic resonance (NMR) logs in two Chinese tight sandstone reservoirs. Acta Geophysica, 62(1), 116-141.DOI
  54. [56]Yoneda, J., Jin, Y., Muraoka, M., Oshima, M., Suzuki, K., Waite, W. F., & Flemings, P. B. (2022). Comprehensive pressure core analysis for hydrate-bearing sediments from Gulf of Mexico Green Canyon Block 955, including assessments of geomechanical viscous behavior and nuclear magnetic resonance permeability. AAPG Bulletin, 106(5), 1143-1177.DOI
  55. [57]Zisser, N., & Nover, G. (2009). Anisotropy of permeability and complex resistivity of tight sandstones subjected to hydrostatic pressure. Journal of Applied Geophysics, 68(3), 356-370.DOI

Paper details

vlnc:W_zxrwrsng
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

AI-assisted language editing tools

View statement
Looking for related papers…
Contents