Subsurface formations often contain multiple minerals with different wettability characteristics upon contact with nonaqueous-phase liquids (NAPLs). Constitutive relationships between microstructure heterogeneity and NAPL fate and transport in these formations are difficult to predict. Several studies have used pore-scale network models with faithful representations of rock pore space topology to predict macroscopic descriptors of two-phase flow, however wettability is usually considered as a spatially random variable. This study attempts to overcome this limitation by considering more realistic representations of rock mineralogy and wettability in these models. This is especially important for heterogeneous rocks where properties vary at the pore-scale. The work was carried out in two phases. First, pore-fluid occupancy maps during waterflooding were obtained by X-ray microtomography to elucidate the impact of pore wall mineralogy and wettability on water preferential flow paths and NAPL trapping within a heterogeneous aquifer sandstone (Arkose). Then, microtomography images of the rock were used to generate a hybrid pore network model (PNM) that incorporated both pore space topology and pore wall mineralogy. In-situ contact angles (CA) measured on the surface of different minerals were assigned to the network on a pore-by-pore basis to describe the exact wettability distribution of the rock (Pore-by-pore model). The equivalent network was used as input in a quasi-static flow model to simulate waterflooding, and the predictions of residual NAPL saturation and relative permeabilities were compared against their experimental counterparts. To examine the sensitivity of the model to the underlying fluid-solid interactions, we also used traditional methods of wettability characterization in the input data and assigned them randomly to the PNM. Wettability in this case was assessed from macroscale CA distribution of oil droplets on the surface of unpolished Arkose substrates released by spontaneous imbibition of water (Arkose model) and from pendant drop measurements on polished quartz (Quartz model). Our results revealed that the Pore-by-pore model predicted waterflooding with the highest accuracy among all three cases. The Arkose model slightly overestimated NAPL removal whereas the Quartz model failed to predict the experiments. More in-depth analysis of the Pore-by-pore and Arkose models showed that macroscopic transport quantities are less dependent to microstructure heterogeneity if minerals are distributed uniformly across the rock. The predictions herein indicate the importance of incorporating mineralogy and wettability maps to improve the prediction capabilities of PNMs especially in systems with high mineral heterogeneity, where minerals are nonuniformly distributed, or selective fluid-mineral interactions are targeted.

与非水相液体（NAPL）接触后，地下地层通常含有多种具有不同润湿性的矿物。这些结构中微观结构异质性与NAPL命运和运输之间的本构关系很难预测。几项研究使用了具有岩石孔隙空间拓扑结构忠实表示的孔隙尺度网络模型来预测两相流的宏观描述，但是润湿性通常被认为是空间随机变量。这项研究试图通过考虑这些模型中岩石矿物学和润湿性的更真实表示来克服这一限制。这对于性质在孔隙尺度上变化的非均质岩石尤其重要。这项工作分两个阶段进行。第一，通过X射线断层照相术获得了注水过程中的孔隙流体占有率图，以阐明孔隙壁矿物学和润湿性对非均质含水层砂岩（Arkose）中水优先流路径和NAPL截留的影响。然后，使用岩石的显微照片图像来生成混合孔隙网络模型（PNM），该模型结合了孔隙空间拓扑结构和孔隙壁矿物学。将不同矿物表面上测量的原位接触角（CA）逐孔分配给网络，以描述岩石的精确润湿性分布（逐孔模型）。等效网络被用作准静态流模型中的输入以模拟注水，并且将残余NAPL饱和度和相对渗透率的预测与实验对应物进行了比较。为了检查模型对基础流体-固体相互作用的敏感性，我们还在输入数据中使用了传统的润湿性表征方法，并将其随机分配给PNM。在这种情况下，可润湿性是通过自吸水释放的未抛光的Arkose基质表面上油滴的宏观CA分布（Arkose模型）和抛光石英的悬垂测量（Quartz模型）来评估的。我们的结果表明，在三种情况下，孔隙孔隙模型预测的注水精度最高。Arkose模型略微高估了NAPL去除率，而Quartz模型未能预测实验。对孔隙孔隙和阿科塞模型的更深入分析表明，如果矿物在岩石上均匀分布，则宏观运输量对微观结构异质性的依赖性较小。本文的预测表明了结合矿物学和润湿性图来改善PNM的预测能力的重要性，尤其是在矿物异质性高，矿物分布不均匀或目标流体-矿物相互作用为目标的系统中。