In this work, five representative sediment samples (i.e., W04-1 to W04-5) were acquired from the gas hydrate-bearing sediments in the Shenhu area (Figure 1). These samples were depressurized and stored after physical property measurements. For experiments on hydrate formation and dissociation, samples were initially evacuated and then saturated with distilled water at 12 MPa for 48 h.

The characterization of mineral composition, particle size, and pore size was performed in this work. Mineralogical data were analyzed using the X-ray diffraction (XRD) method. The grain size analysis of sediment samples for the nano- or micrograins was performed using the Zetasizer Nano ZS90 particle size analyzer and HELOS-OASIS particle size analyzer, respectively. The ASAP 2460 instrument was used for N₂ adsorption measurements to determine pore size distributions of sediments.

Figure 2 shows a schematic of the LF-NMR monitoring apparatus, which was used to perform the in situ hydrate formation and dissociation experiments. The experimental apparatus system mainly comprises an LF-NMR analysis system, a core holder, a gas booster, a confining pressure tracking pump, a refrigeration cycle system, and the gas-liquid separator. The LF-NMR system is a Mini MR60 spectrometer with a magnetic field strength of 0.5 T and resonance frequency of 23 MHz. The confining pressure tracking pump is used to maintain certain confining pressure for CO₂ injection. The refrigeration cycle system contains a circulation pump for cooling the fluorocarbon oil to control the experimental temperature.

In experiments, all the sediment samples were enclosed with heat-shrink tubing as a core with a length of 3.9 cm and a diameter of 2.5 cm. They were saturated with distilled water in the core holder, and then the CO₂ was injected into the core holder with a constant flux until the water was displaced to specific content. The outlet valve of the core holder was then closed, and the CO₂ supplement was enabled by a high-pressure pump to maintain the pore pressure at 4 MPa. During the process, some gases dissolved in water across gas–water interfaces. In situ NMR measurements started with decreasing temperature. Temperature controlled by the refrigeration cycle system decreased at a speed of 0.01°C/min from room temperature to 4°C and then kept stable until the end of hydrate formation. Then, the temperature increased at a speed of 0.02°C/min to 15°C for the hydrate dissociation. Gas permeability of the core was measured before and after the experiment. NMR measurements were performed using the Carr–Purcell–Meiboom–Gill (CPMG) pulse sequence during the experiment. Parameters of the CPMG pulse sequence included the echo number of 18,000, echo spacing of 0.15 ms, and waiting time of 1,000 ms.

The cumulative value of NMR signal intensity can quantify the water content and volume fraction in porous media, and the transverse relaxation time (T₂) of water, proportional to the surface-to-volume ratio (S/V), can be used to evaluate the pore structure (Ji et al., 2019). For CO₂ hydrate (CO₂·5.75 H₂O) formation and dissociation, the volume of CO₂ hydrate (V _h ) can be calculated by water reduction as follows: V h = m h ρ h = m w d 18 × 5.75 ( 44 + 18 × 5.75 ) ρ h , (1) where m h is the mass of CO₂ hydrate, m w d is the mass reduction of water, and ρ h is the density of CO₂ hydrate. CO₂ hydrate saturation is the ratio of hydrate volume and pore volume.

The evolution of permeability subject to hydrate saturation is estimated by experimental data and theoretical models. The SDR method is usually used to calculate permeability from NMR data. It is based on the measured T ₂ distribution being a pseudo-pore size distribution during hydrate formation or dissociation. Eqn. 2 is described as follows: K S D R = C ∅ 4 T 2 g m 2 , (2) where K is the permeability, Φ is the porosity (%), and T _2gm is the logarithmic mean of T ₂ distribution (mD). The parameter C in the model is estimated by NMR measurements with known porosity and permeability of samples.

Theoretical permeability models differ from each other by using various assumptions of hydrate pore habits. The parallel capillary model, Kozeny grain model, and Masuda model are used in this paper.

1) Parallel capillary model (Kleinberg, 2003).

This model describing porous media as a bundle of straight, parallel cylindrical capillaries was developed by Kleinberg (2003). In this model, hydrate forms in the center of cylindrical pores, uniformly coating the walls of capillary. With the increase in hydrate saturation, an annular flow path is left for fluid. For the hydrate coating capillary walls model, the permeability can be calculated using Eqn. 3: K ( S h ) = K 0 ( 1 − S h ) 2 , (3) where K ₀ is the permeability before hydrate formation and S _h is the hydrate saturation. If hydrate occurs in the center of the capillary model, Eqn. 4 is used: K ( S h ) = K 0 ( 1 − S h 2 + 2 ( 1 − S h ) 2 ln ( S h ) ) . (4)

2) Kozeny grain model (Spangenberg, 2001).

Spangberg (2001) assumed porous media as Kozeny grain models and hydrate distribution patterns of both coating the grains and occupying pore centers can be calculated. For the former pattern, the permeability is evaluated as shown in Eqn. 5: K ( S h ) = K 0 ( 1 − S h ) n + 1 , (5) where the saturation exponent n is 1.5 for 0 < S h < 0.8. When S h > 0.8, the saturation exponent diverges. Eqn. 6 is used to calculate the permeability when hydrate occupies the center of the pore, n = 0.7 S h + 0.3 : K ( S h ) = K 0 ( 1 − S h ) n + 2 ( 1 + S h 0.5 ) 2 . (6)

3) Masuda model.

Masuda model (Masuda et al., 1997) is commonly applied to interpret the relationship between permeability and hydrate saturation as Eqn. 7: K ( S h ) = K 0 ( 1 − S h ) N , (7) where N with regard to pore structure is the permeability decreasing index and changes from 2 to 15 (Kumar et al., 2010).

The XRD analysis of sediment samples shows that the predominant minerals are quartz (30.9%∼37.0 wt. %), carbonate minerals (22.8%∼30.8 wt. %), and mica (13.6∼31.6 wt. %) and are followed by clay minerals. The lower contents of clay minerals were determined in this work compared to previous studies (Li et al., 2019; Wang et al., 2021), which reported the weight percent of clay minerals high to 40% for samples from the SCS.

The results of the grain size analysis show similar properties of particle size distributions for five sediment samples. The frequency curves of all the samples are unimodal (Figure 3) and negatively skewed, with peak values varying over a broad range of 20∼40 um (Figure 3B). In particular, the median grain size of W04-1, W04-2, W04-3, W04-4, and W04-5 is 5.47, 18.57, 17.09, 11.84, and 17.48 um, respectively. It indicates that the sediments from the SCS mainly comprise clay (< 4 μm) and fine sand grains (4∼64 μm). Particle size distributions of nanosized grains also highlight the grains of samples are very fine. The frequency curves are also unimodal with a range of 396∼1,106 nm, and peak values of curves vary from 600 to 800 nm (Figure 3A).

The N₂ adsorption–desorption isotherms of five samples display similar patterns (Figure 4A). The adsorption isotherms increase slightly at a relative pressure (p/p₀) <0.45, followed by a sharp increase when p/p₀ is close to 1.0. The hysteresis loops of adsorption–desorption isotherms are indistinctive, which indicates the pore system of sediment samples is composed of cylindrical, wedge-shaped, or slit pores.

Figure 4B shows pore size distributions of W04-1 to W04-5. For W04-1, W04-3, and W04-4, results of pore size with the bimodal distribution mainly range from 1.3 to 80 nm. In particular, the first peak value is 1.27, 1.59, and 1.48 nm, and the second peak value is 54.42, 68.50, and 79.87 nm for W04-1, W04-3, and W04-4, respectively. The pore size curves are unimodal with a peak value of 93.13 nm for W04-2 and W04-5. Based on pore size analysis, a large number of nanopores in these sediments can provide considerable space for hydrate accumulation.

Figure 5 shows the procedure of the experiments with an increase and then a decrease in the temperature. The whole process can be divided into several periods in terms of the variation of hydrate saturation. For the period A-B, an induction time for hydrate formation is observed with the limited variation of hydrate. During period B-C, the pressure and temperature conditions are favorable for hydrate formation, and hydrate saturation increased rapidly up to 50%. The rate of hydrate formation is around 12%/min, estimated by variations of hydrate saturation (%) over time. The hydrate saturation is almost consistent within the period C-D resulting from the consumption of water and the limited contact between water and gas. After the D point, the temperature increases gradually and results in the dissociation of hydrates. During period D-E, the rate of dissociation gradually increases to 3%/min with the higher temperature. It is because the experimental condition is far away from the phase equilibrium condition, leading to a larger driving force for hydrate dissociation. At the end of the experiments, all the CO₂ hydrates dissociate, and the hydrate saturation decreases to 0.

T₂ distribution spectra of NMR provide insights into the pore size distribution of sediments and the change of pore space subjected to the hydrate formation or dissociation. As shown in Figure 6A, there are two distinct parts for the initial signal of sediment samples, i.e., Part 1 (0.2∼2 ms) and Part 2 (10∼300 ms), representing the complex pore structure with mixed grains. At the early stage of hydrate formation, there is an obvious decrease in Part 2, whereas there is a slight increase in the peak of Part 1. It indicates that the pore space of Part 2 is occupied by the formation of hydrate and leads to a faster relaxation time of the NMR signal. The peak area of Part 2 decreases 40% from 275 to 350 min with a preferential formation of hydrate in larger pores. Some scholars believed that the enhanced performance of heat and mass transfer makes it easier for large pore spaces to achieve nucleation (Qin et al., 2021). With the continuous increase of hydrate saturation, Part 2 became vanished, while Part 1 was still with a low peak at the end of hydrate formation. The phenomenon shows the movement of fluid is quite limited with the presence of hydrates, and thus the local hydrate formation is restrained due to the lack of water or gas.

An opposite tendency of NMR signal variation can be observed in the process of hydrate dissociation (Figure 7). The value and width of peaks increase with the hydrate dissociation. Compared to the process of formation, the increase of Part 2 is slightly later than that of Part 1. The rate of hydrate dissociation gradually increases in both Part 1 and Part 2 over time confirmed by the cumulative intensity of the NMR signal. A noticeable difference is shown between the T₂ distribution spectra of the initial situation and the end of dissociation. Similar behaviors have been observed by Lei et al. (2022b). They analyzed microscopic pore characteristics and flowability of five samples from the Shenhu area, showing better connectivity in hydrate-bearing sediment samples after hydrate dissociation compared to samples from the underlying layer. It demonstrates that pore structure may be altered after the hydrate dynamics in both pore size and pore quantity.

Hydrate saturation and permeability in the porous medium were determined by NMR measurements, and the tendency of permeability evolution subjected to the hydrate formation or dissociation is shown in Figure 8. With the increase of hydrate saturation, a remarkable decrease in permeability can be observed. It indicates that the permeability of porous media obviously decreases with occupying gas hydrates in pore spaces, and the evolution of permeability differs in terms of the hydrate quantity and distribution. In particular, an order of magnitude lower is in permeability with the hydrate saturation of 32%, while three orders of magnitude are lower with the hydrate saturation of 50% compared to the initial permeability.

By comparing permeability results obtained from experimental data and theory models, different tendencies of permeability evolution are shown in Figure 8. Generally, a sharper change of permeability is determined by experimental data at the early stage, while a gentler change at the middle stage is compared to other models. In particular, NMR results show a good agreement with the parallel capillary model (pore-filling) when the hydrate saturation is smaller than 0.1, indicating hydrates prefer to form in the pore central space. Compared to other patterns of hydrate distribution, e.g., a thin film of hydrate coating grain walls, the changes in the permeability are expected to be larger if hydrates occupy flow paths. When the hydrate saturation is larger than 0.3, permeability evolution can be especially captured by the Masuda model (n = 12). Some studies demonstrated that grain-coating hydrates appear to be more common than pore-filling ones with a low porosity and permeability (Lei et al., 2022a). Moreover, a drastic reduction in permeability may occur if formed hydrates plug the pore throats in addition to pore bodies, especially when a high hydrate saturation is in porous media. It is clear that the tendency of permeability evolution is differentiated induced by various patterns of hydrate growth in pores and is more complicated when hybrid patterns exist. In this work, most experimental permeability data points fall between fitted lines from different theoretical permeability models, i.e., the pore wall coating and pore center occupying models. It highlights that hybrid pore habits of hydrates affect the evolution of permeability in hydrate-bearing sediments of the Shenhu area, and a preponderant pattern of hydrate growth changes spatially and temporally.

During the process of hydrate formation and dissociation, pore structure characteristics of sediments influence the hydrate saturation and distribution, as well as gas–water conditions. Previous studies pointed out that various gas–water flow patterns result in excess-gas, excess-water, or dissolved-gas conditions for hydrate formation in sediments and have a profound impact on hydrate growth (Wang et al., 2021). In the excess-gas condition, a limited amount of water mainly cements hydrophilic grain surfaces or contacts, which results in grain-coating hydrates with gas diffusing into water layers. But when the ratio of gas to water is low, the formation of hydrate starts close to trapped gas bubbles or dissolved-gas regions, and gradually exhibits a pore-filling morphology. In this experiment, excess-gas, excess-water, and dissolved-gas conditions coexist in sediments, and gas–water ratios and distributions vary with hydrate dynamics (illustrated in Figure 8). Hydrates formed at gas–water interfaces can continuously grow to coat grains or occupy pore central spaces subjected to the specific gas–water condition. In this regard, a theoretical model assuming a single pattern of hydrate growth is expected to fail to predict the hydrate–saturation-dependent permeability in a geosystem in terms of heterogeneous pore structures and gas–water conditions.