An Investigation into Performance of NMR Fluid Substitution Methods on Log Data Sets from Contrasting Reservoirs

Nuclear magnetic resonance (NMR) logs provide in‐situ estimates of pore‐size distribution and permeability,properties that are difficult to infer from conventional measurements. However, most models linking these rock properties to the NMR relaxation‐time spectrum are calibrated on fully water‐saturated rocks. When multiphase fluids occupy the pore space, these models can misrepresent pore sizes and permeability. In downhole logging operation in the field, the presence of hydrocarbons or invasion of oil‐based‐mud filtrate shifts the T1 or T2 relaxation‐time spectrum, and geological complexities such as laminated shale can further overlap with the water response. These effects impair pore‐size interpretation, permeability estimation, and rock‐typing results. To address the complications introduced by immiscible fluids, several NMR fluid‐substitution (FS) workflows have emerged over the past two decades. This paper evaluates three published FS approaches using logging‐while‐drilling (LWD) and wireline NMR data from contrasting reservoirs, including low‐porosity chalk, clastics (sandstones) with unimodal and multimodal pore-size distribution, and an anisotropic thinly laminated shale-sand system. The methods assessed are: (A) a chalk‐optimized FS using a T2-saturation‐ratio approach (Thern/Christensen); (B) an irreducible/movable‐water reconstruction based on weighting fractions and primary drainage/imbibition behavior (Li et al.); and (C) a pore‐size‐dependent FS (PSDFS) incorporating step‐function (Medellín et al.). For this paper, PSDFS method is slightly modified without fundamentally changing its principles. Each workflow is implemented with lithology‐specific preprocessing and quality control (QC), followed by a comparison of substituted T2 spectra, their geometric mean, and permeability trends across reservoir types and acquisition modes. In chalk reservoirs, all three fluid‐substitution (FS)methods reliably reconstruct the water‐wet T2 spectrum and support consistent flow‐unit discrimination. In sandstones with a unimodal pore‐size distribution,Methods B and C produce comparable fluid-substituted spectra. In more complex clastic systems with multimodal pore‐size distributions, particularly when wells are drilled with water‐based mud (WBM) and movable water may coexist with residual hydrocarbons,Method B tends to better preserve the separation of porosity and saturation components when the individual fluid peaks remain distinguishable. In oil‐based‐mud (OBM) environments where only irreducible water remains, both Methods B and C perform reliably, with Method C yielding an averaged pore‐size response. For thinly laminated shale-sand sequences, Methods B and C remain effective only after explicit removal of the laminated‐shale NMR contribution, after which the workflows recover sand‐lamina pore‐size distributions and permeability trends. None of the methods succeed when water and hydrocarbon signals substantially overlap (e.g., residual heavy oil in the pore space),making multidimensional NMR fluid‐typing essential in such intervals. Overall, no single workflow performs best across all reservoir conditions; outcomes depend on lithology, pore‐size modality, mud‐system effects,spectral separability, shale content, and data quality.