Predicting drainage capillary pressure curves in natural porous media by NMR-T

Mercury injection capillary pressure (MICP) tests are used to characterize capillary profiles of geological formations. For CO2 geo-storage, sealing and trapping capacity is often assessed by converting air–mercury curves into brine–CO2 curves. These tests, however, are typically performed only at sparse intervals along the well. To extend coverage, many studies have attempted to derive continuous pseudo-capillary pressure curves from NMR T2​ measurements. Conventional approaches match T2​ and MICP curves to estimate surface relaxivity as a conversion factor, but this assumes uniform relaxivity and simple pore geometry, conditions rarely valid for complex lithologies and requiring calibration by rock type. Because conversion of cumulative T2​ response into capillary curve depends on lithology, we introduce J parameter based on sample gas porosity and permeability (J​), bimodality index (BI) for quantifying the degree of peak separation in T2 spectrum, and the bin-weighted T2​ logarithmic mean (T2lmA​) to allow autonomous rock typing by the model. Using these features, a CatBoost model was trained on 30 core samples (carbonates, sandstones, tight mudstones) spanning permeabilities from 0.001 to 11,050 mD. For the training set, the MICP and NMR T2​ distributions of 30 rock cores were subsampled resulting in 3112 ​​ datapoints. Blind test subset comprised 6 additional cores kept outside the training loop. In the blind test phase, the model reproduced MICP curves across 10− 1–105 psi (7 × 10− 4–7 × 102 MPa) pressure range with mean R2 = 0.94, and SHg saturation mean absolute error of 3.6% indicating reliable estimation of air–mercury and by conversion to brine–CO2 curves.