Environ. Sci. Technol. 2007, 41, 4428-4434
Kinetic Gas-Water Transfer and Gas Accumulation in Porous Media during Pulsed Oxygen Sparging GERD U. BALCKE, STEFAN MEENKEN, CARSTEN HOEFER, AND SASCHA E. OSWALD* UFZ, Helmholtz Centre for Environmental Research, Department of Hydrogeology, Permoserstr. 15, 04318 Leipzig, Germany
Gas-water mass transfer and the transport of dissolved gases in variably saturated porous media are key processes for in-situ remediation by pulsed gas sparging. In this context, gas dissolution tests were conducted during pulsed oxygen gas injection into sand columns. The columns were recharged with anoxic water, effluents were analyzed for dissolved O2, and tracer tests were performed to detect accumulation of trapped gas. In a second series oxygen gas was blended with sulfur hexafluoride (SF6), and O2 and SF6 breakthrough curves were recorded. To interpret experimental results, a numerical model was applied that simulates multi-species kinetic mass transfer during gas dissolution. The model predicted breakthrough curves of dissolved gas species and delivered spatially resolved values for gas phase accumulation and composition, which are not directly accessible experimentally. It was shown how dissolved nitrogen accumulates increasingly in trapped gas phase and inhibits its complete dissolution, in case the pulsed gas injections were operated based on O2 breakthrough only. Accumulation of nitrogen also retarded dissolved oxygen transport and thus oxygen breakthrough. Experiments plus modeling demonstrated that SF6 measurements are highly sensitive to the gas dissolution processes, and provide a more sensitive criterion for determining gas injection frequencies during pulsed biosparging.
1. Introduction Due to oxygen demanding processes, the relatively low solubility of oxygen in water, and restricted direct exchange with the atmosphere, contaminated subsurface environments may become rapidly depleted in dissolved oxygen (DO). In this situation, entrapped gas phases introduced into an aquifer could offer a longer lasting source of oxygen resupply. In the subsurface gas can be trapped by a number of processes, e.g., at the capillary fringe due to rising water tables (1), inside an aquifer due to biogenic bubble formation (2), or as a result of in-situ gas sparging operations (3). If gas sparging is designed mainly to enhance intrinsic microbial degradation of groundwater pollutants, it is called biosparging. Conventional biosparging aims at maintaining aerobic conditions in groundwater via injection of air or pure oxygen gas (4). However, our knowledge of the gas-water interphase mass transfer in the subsurface largely lags behind * Corresponding author phone: +49 341 2353985; fax: +49 341 2352126; e-mail:
[email protected]. 4428
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 41, NO. 12, 2007
technical applications. For instance, undue gas injection may cause undesired effects including clogging and uncontrolled hydrocarbon vaporization, if gas breaks through to the water table. Furthermore, research indicates that it is not necessary to maintain high levels of DO in aquifers to foster hydrocarbon degradation (5, 6). Data from pulsed sparging operations showed that hydrocarbon concentrations continued to decline for several days after the gas injection stopped, although dissolved oxygen could not be detected (3, 4). This ongoing disappearance was attributed to aerobic biodegradation near zones of trapped gas that continued to supply low amounts of oxygen. For these reasons, the development of better operation criteria for pulsed gas injections could improve biosparging efficiency in the field. Differing interphase mass transfer kinetics for relevant gases are controlled by different aqueous diffusion and partitioning behavior of each individual gas component and affect composition and volume of trapped gas, as does the presence of gases already dissolved in the inflowing groundwater (7, 8). Hence, the accumulation of another gas compound, e.g., nitrogen, in a trapped gas phase will have an impact on the net mass transfer of a given gas, e.g., oxygen, mainly due to changes in the interfacial geometry and the alteration of local concentration gradients between gas and aqueous phase. Overall, at each location there is a unique gas composition, gas volume, and dissolved gas concentration, which, in order to reach a local equilibrium, undergo a temporal evolution driven by groundwater flow and changes in hydrostatic and capillary pressure. Additionally, in pulsed biosparging, the mass transfer of a gas compound can be affected by a residual, immobile gas phase remaining undissolved from previous gas injections. The combined effects of these processes cannot be resolved separately and have not been sufficiently considered. Hence, this work strives to provide a more comprehensive understanding of gas dissolution processes in relation to pulsed gas injections. Based on implications of work by Fry et al. on gas-water interactions in porous systems (9-11), early models still assumed a constant gas volume with equilibrium between gas and aqueous phases. However, when interphase mass transfer simulations failed to reflect experimental observations at high gas saturation, attempts were made to improve the model by adding a kinetic term (12, 13). Cirpka and Kitanidis recognized that a volatile partitioning tracer might change the gaseous phase saturation (14), but their model did not consider the kinetics of interphase mass transfer. Recently, a kinetic approach has been introduced that considers a changing gas-phase composition, compoundspecific transfer kinetics, and a changing gas saturation during the transport of dissolved volatile compounds (7), which was later extended for permeability changes with gas saturation (8). In this study, we extended the Kinetic Bubble Dissolution model (KBD) by Holocher (7) and applied it to quantify multispecies gas-water interphase mass transfer during pulsed gas injection events for homogeneous conditions. Closer investigation of heterogeneity effects or gas-phase flow was beyond the scope of this work.
2. Materials and Methods 2.1. Experimental Setup. The experimental setup is based on 0.70 m long, transparent glass columns (0.046 m inside diameter) packed with washed glacial quartz sand. The sand comprised the grain fractions 200-630 µm (29%), 630-1000 µm (15%), and 1000-2000 µm (56%) reflecting the grain size distribution of a polluted aquifer in Bitterfeld, Germany (15). 10.1021/es062890+ CCC: $37.00
2007 American Chemical Society Published on Web 05/12/2007
FIGURE 1. Development of trapped gas phases emerging upon pulsed gas injections (conceptual model). As each gas injection produces an entrapped, immobile gas phase inside the porous medium without gas breakthrough occurring at the column outlet, the column then has a lower section containing an immobile gas phase and an upper section still water-saturated. The gas injection height (zinj) specifies how far an approximated, homogeneous distribution of oxygen gas extends above the inlet. Subsequent injections result in a farther total gas penetration, zgas, since the injected gas stream displaces residual gas from incomplete dissolution of the previous injection-flushing cycle, which again is trapped as soon as the injection is stopped. This continues for subsequent gas pulses in similar manner. The porous medium (total pore volume 405 mL) was watersaturated at the start of each experiment as described in ref 10. Gas sparging was simulated by injecting 11.5 mL of gas at the bottom before slowly flushing the column with several column pore volumes of oxygen-free water over several days. During this time resident, trapped gas phase (in the lower part of the column) exchanged gas components with the aqueous phase. Tracer tests were performed during the experiments, breakthrough curves (BTC) of tracer and DO were recorded at the column outlet, and the columns were weighed upon full water saturation and after dissolution of an injected gas phase. Gas injection (and subsequent flushing) was repeated several times. The periodicity was set by the effluent BTC concentrations, analogously to using an observation well located downstream of a biosparging fence. The experiments were conducted at 15 °C using a temperature-controlled system with up to four columns operating in parallel (for details see Supporting Information (SI)). In our experiments, flushing a 5 mM sodium hydrogen carbonate solution with N2/CO2 95%/5% (v/v) (at 1 atm) created an anoxic influent, with maximum DO levels of 0.05 mg L-1, and balanced pH of 6.7. A constant head was maintained at the inlet by setting the inflow reservoir (in these experiments: 0.5-0.6 m) above the column base. A peristaltic pump created a constant upward water flow through the columns. The effluent passed through a gas trap enabling volumetric measurement of nondissolved gas phase. An oxygen-sensitive optode flow-through cell connected to a fiber optic device was installed to detect oxygen at each column outlet. All tubing materials were made using low gas permeability materials (glass, high-density PVC, steel). An event-controlled pump, triggered by a counter, automatically primed effluent samples through a bypass multichannel magnetic valve and an array of sensors for detecting properties such as UV-absorbance. Solute and Gas Tracers. Potassium iodide (KI) solution (40 g L-1, p.a. grade) was injected (0.8 mL) as a dissolved, non-partitioning tracer. KI breakthrough was registered by UV-absorbance at 235 nm using a HPLC detector with a preparative cell (0.5 mm light path). When sulfur hexafluoride (SF6) was applied as a gaseous partitioning tracer, a mixture of 98.2% O2 and 1.8% SF6 was produced in glass bottles containing 10 mL of water and sealed by a Mininert valve (Supelco). Following intensive flushing of the glass bottle with oxygen, the cap was loosened for a very short time to dissipate excess pressure and return the bottle to atmospheric pressure. Then pure SF6 was spiked using a gastight syringe with stopcock and samples were shaken until injection. Effluent samples (0.5 mL) were collected via a 3-way bypass valve using a gastight syringe with stopcock and injected into GC vials containing 9.5 mL deionized water, previously capped with butyl rubber septa. The samples were stored upside down in a water tank until analysis. SF6 was
analyzed from the headspace of the shaken GC vials using a GC-ECD system (Shimadzu, 60 °C isotherm, packed column Haye Q 80/100 mesh, Vici International) with valve injection using a 0.1 mL injection coil. Dissolved SF6 was calibrated against authentic standards (Linde, Germany). The detection limit for dissolved SF6 was about 0.004 mg L-1. Calibration of oxygen was done using sodium dithionite and air-equilibrated medium (see below) corrected for atmospheric pressure and temperature. The detection limit under experimental conditions was about 0.05 mg L-1 dissolved oxygen. Gas Injection Procedure. A syringe with stopcock (total volume 13.0 mL) was first filled with 1.5 mL of deionized water and then with either oxygen gas or the O2/SF6 blend at a temperature of 23 ( 2 °C. The stopcock was briefly opened to release gas and achieve atmospheric pressure within the syringe. Subsequently, the 11.5 mL of gas was slowly injected followed by the 1.5 mL of water to ensure complete injection of the tracer and seal the void volume of the injection port. 2.2. Numerical Simulations. To simulate the effect of pulsed gas injection events in porous media, a conceptual model and a MATLAB based code (KBD) of Holocher were used (7). This kinetic model includes mass-transfer kinetics at gas-water interfaces of spherical gas bubbles, changes of gas-phase volume and composition, dissolved gas components, and vertical 1-D solute transport in the aqueous phase. Gas bubbles can completely dissolve, but cannot move through the porous medium during the dissolution process. Hydrostatic and capillary pressures are considered. However, additions to the code were required in respect to the experimental procedure (Figure 1). In the extended model, a gas injection creates a section at the very bottom of the column containing the oxygen gas (or the O2/SF6 blend). If there is already a gas phase present before an injection, e.g., residual gas due to incomplete gas dissolution, another section on top of the bottom one is created that represents this residual gas-phase displaced by zinj due to the next gas injection. In the experiments, pulsed gas injections had always the same gas volume, and it can be assumed that also the resulting gas-water ratio is always the same for the injections into a particular column. Therefore, zinj was taken to be a constant for pulsed injections. The computed composition and size of gas bubbles remaining after a flushing period were used to calculate a total residual gas volume and total masses of components present in the gas phase. For simplicity, the gas volume displaced by a following injection is taken to be unchanged, and its gas composition is assumed to be identical to the average composition of the former gas phase. However, the initial bubble size resulting from the short-term, upward displacement was adjusted to the same uniform initial radius as the one appearing in the lower section. The new zgas is calculated via conservation of the gas volume and the fixed gas-water ratio of the column. During VOL. 41, NO. 12, 2007 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
4429
the injection, gas components mix at the front of the invading gas to some unknown extent. Such mixing of parts of the gas bubbles is not reflected in the modeling procedure. However, additional simulations with complete mixing were performed, but the experimental results were not better reproduced with this approach (SI). Longitudinal dispersivity values were obtained from several non-partitioning tracer experiments monitored under water-saturated conditions. Since the injected gas-phase did not change the observed dispersivity substantially, a mean longitudinal dispersivity value of 2.0 mm was used in the model. The columns were operated in two flow modes, i.e., with and without tracer sampling. Each mode had a constant discharge rate, thus the model was amended to capture this change from the otherwise constant Darcy flux. In the simulations, the carbonate buffer and possible transfer of CO2 into gas phase were neglected to limit the number of gas components. Screening simulations had shown that CO2 had no major impact on oxygen gas dissolution, as discussed in the Supporting Information (cf. Figure SI-3). Notably, the listed parameter values and settings were fixed for all simulations of the column experiments, without any parameter fitting.
3. Results and Discussion Results from single series of pulsed oxygen and oxygen/SF6 injections are presented here. In general, experiments using different columns filled with the same material in a very similar manner showed similar trends. Differences are caused by the sensitivity of the gas-phase distribution due to pore space variability. The assumption of a homogeneous distribution of gas bubbles is inherent in the conceptual and numerical models. Therefore, the column experiment with the most homogeneous conditions, indicated by the least spreading of the BTCs, was selected for evaluation of both types of gas injections. For brevity, the discussion will not explicitly cover differences between these column experiments. 3.1. Pulsed Injection of Pure Oxygen Gas. First, the reinjection of oxygen was triggered by the nearly complete breakthrough of dissolved oxygen from the previous gas injection pulse (effluent criterion 0.4 mg L-1 DO). In total, five consecutive oxygen pulses of 11.5 mL each were injected. A non-partitioning inert tracer pulse was released before the first gas injection into the entirely saturated column, with each gas injection, and at the end of each gas dissolution period. Since the inert tracer had to be sampled with higher frequency to sufficiently map those BTCs, an increased Darcy flux resulted when the inert tracer was measured (0.53 m d-1) as compared to periods when only oxygen was detected (0.40 m d-1). With each repeated injection, the O2-BTCs displayed slowing overall oxygen mass transfer reflected by an increasing retardation of DO breakthrough, lower maximum DO concentrations, and more peak tailing (Figure 2A). We operationally define an apparent retardation factor Ri50 of gas-phase compound i (Table 1 and eq 1) as indicator of mass-transfer behavior, including effects of transfer from gas into water phase, aqueous phase transport, and temporal storage in gas phase
R50 i )
Vi50 50 VKI
(1)
where V50 i is the volume of fluid passed through the column to reach a 50% breakthrough of the recovered mass of gas i (originating from the gas phase), with i ) O2, SF6. V50 KI is the fluid volume passed through the column to reach a 50% 4430
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 41, NO. 12, 2007
FIGURE 2. Five consecutive oxygen sparging pulses, when nearly complete oxygen breakthrough was used to trigger subsequent injections: measured DO breakthrough curves (A) versus simulated ones (B). breakthrough of the non-partitioning inert tracer (KI) at full water saturation. Analysis of the BTCs of non-partitioning inert tracers injected at the end of each oxygen breakthrough (at 0.4 mg L-1 DO) reflected an increase in the gas saturation Sg with each subsequent injection (Table 1). Incomplete gas dissolution was also observed through the transparent column walls while gas breakthrough at the column outlet was not detected for any of the five gas injections. Oxygen recoveries (obtained by integrating BTCs) were significantly less than 100% during subsequent oxygen pulse injections (Table 1). Errors associated with the oxygen detection are too small to fully explain the amount of oxygen not recovered. Additional tests verified close to 100% oxygen recovery with no gas phase present (see SI). Apparently, even small portions of trapped gases have the potential to significantly decrease the recovery for oxygen pulses. A minor possibility also remains that a biological oxygen demand contributed to a decreased mass recovery for oxygen, although a washed sand with low organic carbon content was used (15). Simulation Fine-Tuning of Pulsed Oxygen Injections. Since the experimental observations could neither detect the spatial distribution of trapped gas in the porous medium nor the interfacial area, both had to be fitted in the simulations, albeit within a limited range of possible values.
TABLE 1. Experimentally and Numerically Derived Parameters Sensitive to Pulsed Oxygen Gas Injection injection number (consecutive) recovery O2 [%] 50 apparent retardation for O2, RO 2 [-] recovery non-partitioning tracer [%] gas saturation Sg c,d [%],experimental gas saturation Sg c [%],simulated number of pore volumes needed to reach 0.4 mg L-1 O2 effluent concn.
saturated
100.6 0 0
1
2
3
4
5
75.6 1.23 102.6a/104.7b 4.6a/1.1b 2.8/1.0 3.2
78.9 1.29 101.0a/103.7b 6.6a/4.0b 3.9/2.0 3.4
83.3 1.57 101.4a/112.75b 7.2a/5.5b 4.9/3.1 3.7
83.7 1.81 98.1a/100.0b 5.7a/6.6b 5.9/4.2 4.4
69.3 1.94 96.0a/99.1b 8.3a/8.6b 7.0/5.4 4.4
a From tracer test concomitant to gas injection. b From tracer test at 0.4 mg L-1 effluent DO concentration, before next gas injection. c Mean value over whole column. dAn average of three columns operated separately.
The gas saturation Sg arising in the bottom part of the column during gas injection directly determines the gas injection height zinj, given the injected gas volume is known. The oxygen bubbles at this position determine the time needed for the first DO to arrive at the outlet. This time is significantly shorter than the one to discharge a full pore volume. As a model parameter, zinj is well constrained since farther gas penetration heights directly cause earlier oxygen breakthrough. Therefore, zinj, and thus Sg, could be estimated within a narrow range by comparing the rising portions of simulated and observed oxygen breakthrough curves of merely the first oxygen injection. A zinj value of 0.275 m, according to a local Sg of 7.0%, gave the best fit (data not shown). It is in good agreement with the maximum gas penetration observed through the transparent glass column walls immediately after the first gas injection. Furthermore, Sg falls in the range of values reported for medium- to coarse-grained sandy materials for situations in which no coherent flow occurs (3, 11, 16). This parameter combination was then fixed for simulations of all five consecutive injection events. The second parameter to be fitted was the initial bubble diameter. Although the model assumes the trapped gas phase to exist as spherical bubbles of uniform size, the observed trapped gas phase appeared mainly as rather compact gas clusters after the gas strings had pinched off following coherent gas flow during the injection. Simulations of the bubble development started with an initial diameter of 2 mm for the oxygen bubbles (as resulting from each gas injection). This value was then modified to achieve a good overall match of all five breakthrough curves, and a value of 4 mm gave the best fit for the initial bubble diameter. Recently, Selker et al. (17) theoretically derived double that value for gas strand diameter close to the injection point, and reported similar experimental values for sand sizes comparable to the ones used here. Thus, the fitted value lies between the pore size and the value reported for gas strand size, and should primarily be interpreted as an effective parameter. Simulation Results for Pulsed Oxygen Injections. Simulations on gas bubble evolution for pulsed gas injections provided detail on the distribution and composition of the gas phase, which were not accessible experimentally. Modeled effluent oxygen concentrations of the five consecutive oxygen pulses reflected the same tendencies as the observed results, i.e., (i) increased retardation, (ii) decreasing peak maximum, and (iii) increasing tailing, respectively (Figure 2B). The curves are fitted reasonably well, matching the gas arrival data while using a unique set of parameters for all injections. Residual gas volume and composition were taken from the previous run and thus were not fitted for subsequent injections. However, a detailed comparison of single breakthrough curves reveals differences between simulated and observed data, e.g., in the peak maximum or the tailing behavior. The clearest discrepancy is a too low observed peak maximum for the first injection, which is probably related to the incomplete experimental oxygen recovery, and which does not fit into the trend of peak maxima observed for the
other injections. Both simulations and experiments showed substantial accumulation of trapped gas in the column pore space when the injection of a subsequent pulse was triggered by the oxygen breakthrough (see gas saturations, Table 1). Discussion of Mechanisms. Net transfer of a volatile compound between aqueous and gaseous phases is directed from the phase with higher to that of lower partial pressure. The sum of partial pressures in the recharging solution does not exceed the sum of hydrostatic and capillary pressures in the gas phase except at the column top so that the entire gas-phase would be dissolved upon sufficiently long flushing. Near the inlet, the entrapped gas initially consists of pure oxygen which is more soluble and has a higher molecular diffusion coefficient than nitrogen (7), which dominates the composition of volatile compounds in the solution. Under these conditions, a monotonic decrease of gas saturation is expected (14). And in fact, the simulated radius of gas bubbles monotonically decreases near the inlet (Figure 3A, solid lines). In this process, the aqueous solution picks up oxygen but simultaneously loses nitrogen by gas-water interface mass transfer (Figure 3B, solid lines), resulting in a nitrogen dominated gas-phase composition. Farther in the column, this solution passes the bubbles remaining from the previous injection, which are mainly composed of nitrogen. Here, oxygen is transferred into the gas phase again while less nitrogen is dissolved, resulting in an intermediate increase of gas saturation (Figure 3A, dashed lines) and an accumulation of oxygen in the old gas bubbles (Figure 3B, dashed lines). Only after the gas saturation near the inlet has decreased and less oxygen has started being released into the aqueous phase, the gas saturation in the zone of old gas bubbles begins decreasing (Figure 3A, dashed lines). Concurrently, the mole fraction of oxygen in the gaseous phase decreases (Figure 3B, dashed lines). Overall, the old gas bubbles act as temporary storage for oxygen and retard the transport of DO (9-11). These old gas bubbles are slightly larger at the end of the flushing period than at the beginning (Figure 3A, dotted lines). Therefore, applying the oxygenbased criterion of initiating the next injection cycle leads to the accumulation of an immobile nitrogen gas phase. With increasing remaining gas saturation after each sparging-andflushing cycle, the retarding effect of the old gas on DO transport increases from one cycle to the next. In conclusion, use of an effluent criterion based on oxygen levels does not support operating a pulsed sparging system that provides similar oxygen levels during pulses, instead in our case this led to accumulation of an immobile nitrogen gas phase that retarded the spreading of DO. 3.2. Pulsed Gas Injections of O2/SF6 Blends. The enrichment of trapped gas motivated the introduction of a tracer that behaves conservatively within the water phase but partitions between gas and water (Figure 4A). Due to a higher volatility and a lower aqueous diffusion coefficient than oxygen or nitrogen (13), sulfur hexafluoride (SF6) features the strongest tendency to remain or accumulate in trapped gas bubbles among the gases studied. When combined with VOL. 41, NO. 12, 2007 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
4431
FIGURE 3. Simulated gas bubble evolution for the third gas injection pulse, when nearly complete oxygen breakthrough was used to trigger subsequent injections. Bubble radius development (A) and molar fractions of oxygen and nitrogen (B) of freshly injected gas bubbles and of those remaining from incomplete dissolution of the previous injection pulse. sensitive detection methods, it constitutes a tracer well suited to assess the complete dissolution of trapped gas phases. In a second experiment, three consecutive pulses of oxygen gas blended with SF6 (11.5 mL; 98.2%O2, 1.8%SF6; 23 °C) were injected into a similar sand column under equal conditions. In this experiment, the mean Darcy flux of water was increased for experimental reasons, and was 1.44 m d-1 during tracer sampling, and 1.05 m d-1 after breakthrough of the non-partitioning tracer. In contrast to the experiments above, the next pulse was triggered by “complete” SF6 breakthrough (4-6 µg L-1 dissolved effluent SF6). The nonpartitioning inert tracer KI was applied with the gas injection and also when dissolved SF6 fell below the effluent concentration mentioned. Additionally, each column was balanced gravimetrically prior to the first and after the last O2/SF6 breakthrough. It was assumed that a full SF6-breakthrough should reflect complete gas dissolution. Consequently, three consecutive injections should give congruent BTCs for O2 and SF6, respectively. Experimental results largely confirmed this assumption. In contrast to the experiment with an oxygen-based reinjection criterion, the oxygen breakthrough curves were almost not retarded for increasing number of injections (Figure 4A; cf. Figure 2). Thus, the three oxygen BTCs display similar maximum breakthrough concentrations and similar tailing. Compared to oxygen, BTCs reflected slower dissolution for SF6, expressed by more extended tailing and higher retardation for SF6 compared to O2 (Table 2). In contrast to oxygen, with recoveries significantly less than 100%, the SF6 mass recovery was complete within experimental error. In more detail, repeated gas injections were observed 50 experimentally to give slight increases of RSF6 , and increas4432
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 41, NO. 12, 2007
FIGURE 4. Three consecutive gas sparging pulses of O2/SF6 blends, when nearly complete SF6 breakthrough was used to trigger subsequent injections: measured dissolved O2 and SF6 breakthrough curves (A) versus simulated ones (B). Simulated amounts of gases and bubble radius in the lowermost gas bubbles (slice of 0.007 m thickness) for the third O2/SF6 injection pulse, shown until complete dissolution of these bubbles at about 22 pore volumes (120 h) (C). ing pore volumes of anoxic medium were needed to reach the re-injection criterion (Table 2). Gravimetric measurements (
