| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
a Idaho National Engineering and Environmental Laboratory, P.O. Box 1625, Idaho Falls, ID 83415-2107
b Sandia National Laboratories, P.O. Box 5800, Albuquerque, NM 87185
* Corresponding author (tqw{at}inel.gov).
Received 14 May 2003.
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONEXPERIMENTAL DESIGNRESULTSSUMMARY AND CONCLUSIONSREFERENCES |
|---|
Abbreviations: INEEL, Idaho National Engineering and Environmental Laboratory
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONEXPERIMENTAL DESIGNRESULTSSUMMARY AND CONCLUSIONSREFERENCES |
|---|
Common to this conceptualization, and indeed to other studies conducted on a wide variety of fractured rock types, is the suggestion that at large spatial scales and relatively short temporal scales, fracture networks can facilitate the deep penetration of dissolved contaminants, even when the rock matrix is far from saturation (e.g., Fabryka-Martin et al., 1996; Davidson et al., 1998; Dahan et al., 1999; Faybishenko et al., 2000; Wood and Norrell, 1996; Podgorney et al., 2000; Wood et al., 1997, 2000a, 2002b, 2002c; Glass et al., 2002a, 2002b; Nicholl and Glass, 2002). Mechanistically, the behavior of individual fractures must organize to yield this large-scale behavior. To consider this issue, Glass et al. (1995)(1996) assembled understanding at the single fracture–scale and then proposed via a thought experiment that fracture intersections behaving as capillary barriers might cause both the confluence and sequestration of flow within a fracture network to create large-scale gravity-driven vertical pathways even in the absence of significant vertical heterogeneities (e.g., vertical faults).
As a next step toward understanding how existing models should evolve to make better predictions of fluid flow in fracture networks at the field scale, a comparison is needed between the behavior of flow in a fully characterized fracture network to modeled behavior using standard approaches. The similarities and differences between the observed and the predicted behaviors will serve to show what works and what does not when conventional modeling approaches are applied to unsaturated flow in a fractured rock network. We focus on this comparison in two companion papers. This paper, Part 1, describes the experimental data set collected for comparison to equivalent continuum modeling simulations described in Part 2 (Fairley et al., 2004). Our objectives are (i) to better understand phenomena that influence dynamical behavior of unsaturated flow in fractured rock networks and (ii) to collect a data set with sufficient control and resolution for comparison to numerical simulations.
We constructed the simplest fracture network that would satisfy our objectives. The experimental work by Glass et al. (2002a), where stacked bricks formed a small network, led us to conduct a similar experiment using natural rock and distributed input. Our initial results, summarized by LaViolette et al. (2003), showed that flow tended to converge in this simple analog fractured rock network even under conditions of distributed uniform flux, supporting the conjecture of Glass et al. (1995)(1996) that fracture intersections exert a first-order control on unsaturated flow at the network scale by acting as local capillary barriers that focus flow. Here, we provide the complete experimental design and give full details of experimental results used for modeling described in Part 2 (Fairley et al., 2004). Because of the similarity in our approach, we will necessarily echo some of the findings of Glass et al. (2002a), such as intersections acting as capillary barriers, flow switching, and pulsation along pathways, but our work has the following differences: (i) distributed flow to multiple fractures, (ii) natural rock, (iii) different boundary condition, and (iv) multiple realizations of reasonable duration. Taken as a whole, these studies identify several critical issues that require further study regarding the application of popular modeling approaches to unsaturated flow in fractured rock.
|
EXPERIMENTAL DESIGN |
|---|
|
TOPABSTRACTINTRODUCTIONEXPERIMENTAL DESIGNRESULTSSUMMARY AND CONCLUSIONSREFERENCES |
|---|
Experimental System
The experimental apparatus consisted of a frame to hold the blocks in place, a water inflow and outflow system, laboratory scales or weighing load cells for tracking mass in and out and time-lapsed photography. Several materials were considered for use during these tests. Our selection criteria included: (i) a relatively homogeneous permeability and porosity field, (ii) a rock easy to shape and durable during handling, and (iii) a rock that visually showed wetting for photographic imaging. We selected a building stone, which is a fine-grained dolomitic limestone from the Ordovician Oneota formation, quarried in Le Sueur County, Minnesota (Vetter Stone Company). Hydraulic tests on representative samples (Table 1) yielded average values for porosity of 15% (±1.5%) and permeability of 1.3 x 10–15 m2 (saturated hydraulic conductivity, Ks, of 7 x 10–5 cm min–1 at 21°C) (Table 1). Twelve blocks were cut to size with the nominal dimensions of 5 by 7 by 30 cm for testing. Asperities were minimized with sandpaper. The block surfaces were washed thoroughly and then assembled into the model without contaminating the fracture surfaces.
|
|
Experiments Conducted
A total of eight experiments were conducted. We conducted three different types of tests: (i) random stacking of blocks (Tests 1–3), (ii) a planned stacking of blocks to focus discharge to a particular outlet (Test 4), and (iii) repeated tests using the same stack of blocks (Tests 5–8). For all tests, mass flux and points of entry to the model were kept the same (i.e., constant, uniform flux, identical source locations, and a nominal 72-h test duration). In Tests 5 through 8 we tilted the experiment slightly forward to ensure that water ran within the fractures toward the side where we recorded the wetting history via time lapsed photography. Table 2 summarizes the tests. In the following, we describe each test in detail.
|
In Test 6 we wanted to evaluate the repeatability of the advance of the wetting front and determine if dynamical behavior occurred at the same elapsed time and same location in the same aperture field starting from similar initial moisture conditions. The blocks were initially oven dried and then stacked into place in Test 5. The bricks remained in place in the frame for Tests 6, 7, and 8 because it was impossible to remove the bricks, dry them, and restack them so that the aperture field was identical to a previous test. We air dried the blocks in place using a fan to force air through the model until reaching a target electrical resistance measured across electrodes at the side bricks. We employed a stainless-steel bolt, drilled and then threaded 1 cm into the blocks on the outside edge of the model, to increase the contact area and reduce the possibility of surface drying of the blocks dominating the resistivity reading. In cases where the electrical resistance became too low, we used a humidifier to raise the moisture content of the bricks, thus lowering the resistance to the target level.
Near the end of Test 1, the automated system for emptying the discharge bottles failed at an elapsed time about 48 h. For Test 1, all of the discharge data are presented and the unusable data noted. During Test 8, a malfunction of the water collection system caused the program collecting the discharge data to crash, preventing the use of any discharge data for Test 8. However, the visual image sequence is usable and these images are analyzed and presented in a later section.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONEXPERIMENTAL DESIGNRESULTSSUMMARY AND CONCLUSIONSREFERENCES |
|---|
2000 min) are caused by the routine calculating the flow rate from the mass of the discharge bottle and are not real events. When the bottles were emptied by the automated system it created a spike in calculated flow rate. Load cells were used for Tests 5 through 8, which are inherently noisier than scale data. We processed the data with the same smoothing routine used in Tests 1 through 4. The smoothing routine was a hamming window (i.e., weighted average) of 31 data points centered on the mass measurement being smoothed. The differences of consecutive smoothed mass data points were divided by the time interval between measurements to compute a flow rate in grams per minute. Additionally, the data were smoothed by omitting spikes and using larger symbols to visually reduce the appearance of noise caused by the weighing load cells.
|
In Test 4, we constructed the model using shims placed at the edge of the bricks so that wider apertures were in the lower left-hand corner as mentioned above. The intent was to see if we could force water to the lower right fracture. We thought that capillary forces probably controlled flow; thus fracture flow would preferentially occur in narrower fractures. As can be seen in Fig. 2, flow for about the first 120 min of the experiment did go almost exclusively to the right-hand fracture. An abrupt switch in flow occurred at approximately 120 min, and then most of the flow went to the left-hand fracture for the remainder of the test. There are at least two possible explanations for this abrupt switch. The first is that imbibition by the matrix during the first 120 min reduced flow through the fractures and thus capillary forces dominated within the fractures at early time. However, with increasing time, imbibition was reduced, flow within the fractures increased, and gravitational forces became more important. This change in relative influence of capillary and gravitational forces caused a switch in the flow path to the left-hand fracture by allowing a critical capillary barrier to be breached. A second explanation could be that with time, the critical capillary barrier simply wetted and was breached due to a reduction in the local contact angle. In this case, the expectation is that the wetting of the surrounding matrix reduces the effective contact angle from the unwetted rough surface.
We observed many instances of pathway switching where discharge from one fracture would increase at the expense of another. Examples of flow path switching can be seen to varying degrees in all tests. The switching of flow from one pathway to another occurred either gradually in intervals of hundreds of minutes or abruptly in a period of a few minutes. An example of gradual pathway switching can be seen in Test 3; beginning at an elapsed time of approximately 3800 min flow gradually shifts from V1 to V2 during the next 2000 min. Discharge to V3 remains constant. An abrupt sudden shift can be seen in Test 4 at an elapsed time of approximately 2800 min when discharge to V2 suddenly ceases and flow switches to V1. During the next 1400 min the discharge gradually switches back to V2 at the expense of V1. Examination of the outflow plots in Fig. 2 indicates that regimes of pathway switching (either gradual or abrupt) are more common than periods of steady, constant flow to the discharge points. Only about 10000 min of the approximately 25000 min of testing data presented in Fig. 2 represent periods of relatively steady, constant flow to the discharge points. The other 15000 min of data are periods with either abrupt or gradual pathway switching.
We cannot explain the causes of the pathway switching with the data set we collected. The boundary conditions of temperature, air pressure, and initial moisture content were not controlled or monitored sufficiently to correlate to pathway switching. We speculate that subtle changes in temperature, air pressure, or vibrations may trigger some of the switching events. Others may be the result of the complicated interaction of gravity, capillary, and inertial forces that develop in the fracture network. Our data corroborate observations made by Glass et al. (2002a) in an experiment with better control and monitoring of air pressure, humidity, and temperature. In their experiment, some pathway switching was explained by a simple evaporation–condensation mechanism, while other pathway switching could not be explained. They concluded that fluctuations in the outflow occur spontaneously within the fracture network and speculated that pulsation at fracture intersections along flow pathways might be the cause of flow path switching. Our experiments confirm that flow path switching appears to be a common characteristic of unsaturated flow in fracture networks.
Time Lapse Imaging
The time-lapse images provide a qualitative record of the advance of the wetting front through the fracture network. A shortcoming of this experimental setup is that the model is not truly two-dimensional—some flow also occurred back of the front face of the block wall and was not recorded with time-lapsed photography because we could not photograph the back of the block wall. However, to alleviate this problem somewhat, we tilted the frame slightly forward in Test 5 through 8 to enhance viewing of the wetting front. Figure 3
shows a side-by-side comparison for Tests 5 through 8 at times of 7 min 30 s, 22 min 49 s, 50 min 17 s, 1 h 47 min, 3 h 39 min, and 10 h 9 min. The wetted areas are dark, and the horizontal dark bands are shadows from the apparatus.
|
At the second time step the pathway dynamics became more interesting as the water began to spread laterally in the first horizontal fracture (H2). In Test 7, at the second time step, water had traversed the entire fracture network to the terminus of V3. For the other tests, the wetting front is not visible at the lower boundary, although flow may have been arriving at the lower boundary behind the plane of the front surface.
At the third time step, imbibition into the limestone matrix is apparent in the upper row of blocks (Row C) and flow appears to be equal in V1, V2, and V3 in Row C. However, in the second row of blocks (Row B), flow in the fractures tended to converge to one or two of the vertical fractures. The convergence of flow in Row B is more apparent in the fourth time step. Here it can be seen that for Test 5 flow was focused in V2 in Row B, for Test 6 flow as focused in V2 and V3 in Row B, in Test 7 the flow was focused to V1 and V2, and in Test 8 the flow was focused in V2 and V3. It is likely that small differences in initial moisture content, hysteresis, and the fact that the flow is inherently unstable may account for the variation in wetting patterns between tests.
During the last two time steps it can be seen that for all four tests, flow tended to converge into the middle fracture, V2, through the lowest Row A (this is confirmed by outflow data in Fig. 2). Matrix imbibition begins to color the entire block wall and the time-lapsed photographic record is of little use in tracking the flow of water after an elapsed time of about 12 h.
Figure 4 shows the development of the wetted structure for each test, with each color representing wetted structure at each time step from Fig. 3. As described above, water initially left the fractures in Tests 6, 7, and 8 and traveled along the face of the model at an off vertical angle, probably because the initial moisture contents of the bricks were higher in the later tests as a result of their being dried in place. Focusing of flow into the fractures of Row B and Row A and convergence to V2 can be seen in all tests. We note that flow apparently zig-zagged through Row B either going down V2, V2 and V3, or V2 and V1, but it tended to converge in all cases to V2 through Row A. Variation in the pathways through Row B is an example of nonrepeatable, dynamical behavior. Figure 4 also shows the existence of several stalled flowpaths that initially exhibited rapid movement of water, but failed to remain active and dried out as the test progressed. Stalled flow paths can be seen in several tests. In Test 5 along fracture V1 just below H1, the green color stops with only a small advance of yellow and orange. In Test 6 along the lower reach of V3 at the intersection with H2 the green color stops with only a slight advance of yellow and orange. In Test 8 along the lower reach of V1 at the intersection of H2 and in V3 at the bottom of the model the green color stops with no yellow or orange advancing forward. The stalled pathways did not always occur in the same locations during repeated tests.
|
|
Fluid cascades appear to be caused by the pooling above a capillary barrier created at the fracture intersection and the subsequent release of water as gravitational forces breech the barrier. This process of flow integration and subsequent intermittent release at a single fracture intersection has been studied recently by Wood et al. (2002). They noted that water filled the connecting horizontal fracture to create a much larger integration volume than could form only in the pool immediately above the intersection within the fracture. Water entering the pool from above causes pressure to build in the combined pool volume until a critical point is exceeded, after which a cascade ensues, creating a gravity driven finger (e.g., Nicholl et al., 1993; Su et al., 1999). Water then advances rapidly to the next barrier, thus creating a stop and start advance of water through the network. We note that the fluid cascade illustrated in Fig. 5 was not initially captured by the next lower intersection. We attribute this to two factors. First, the capillary strength of the intersection was not sufficient to stop the mass of water plus the inertial forces of the fluid cascade. Second, fluid storage above and in the horizontal fractures to either side was insufficient to attenuate the inertial forces and reduce gravity forces to the point where the capillary barrier could hold the advancing cascade. However, at later times, fluid below the intersection becomes disconnected from the fluid finger, and flow at the intersection is diverted to the left to V2 (see Fig. 3, Test 6, 1 h 47 min). This observation underscores the importance of inertial forces generated during the release of a fluid cascade.
While the formation of fluid cascades was most obvious during the initial advance of the wetting front through the dry network, it was also observed at later times when the blocks had imbibed significant amounts of moisture. We must note, however, that fluid cascades did not occur at every intersection, they did not always persist with time as noted above, and we did not always observe the reoccurrence of a fluid cascade at the same intersection under repeated tests.
In addition to the fact that individual flow paths were observed to vary from one test to the next for the constant stack tests, we noted that the percentage of wetted area (wetted area divided by the total area) varied from one test to the next. The percentage of wetted area also changed during a single test when compared with other tests. For instance, if one test began with a higher percentage of wetted area during early time steps it was not unusual for this same test to show less wetted area at later time steps when compared with other tests, or vice versa. Figure 6 is a plot that shows the percentage of wetted area vs. time for the four infiltration tests under constant aperture conditions (Tests 5–8). The switching of rank for wetted area between tests can be clearly seen in this figure.
|
|
SUMMARY AND CONCLUSIONS |
|---|
|
TOPABSTRACTINTRODUCTIONEXPERIMENTAL DESIGNRESULTSSUMMARY AND CONCLUSIONSREFERENCES |
|---|
Convergence of flow with depth.
With uniform distribution of flow across the top of this fracture network, the occurrence of converging or focused flow is at least as statistically significant as the occurrence of uniform flow.
Flow pathway switching is a common, if not characteristic, behavior of unsaturated fractured rock networks.
We observed many instances of pathway switching where discharge from one fracture would increase at the expense of another. The switching of flow from one pathway to another occurred either gradually or abruptly. Periods of pathway switching (either gradual or abrupt) were more common than periods of steady, constant flow.
Fracture intersections act to accumulate and integrate the steady flux of vertically flowing water, generating less frequent, larger pulses of water.
We observed water accumulating or pooling behind capillary barriers formed by intersections. The discharge of pooled water occurred in a fluid cascade as a critical point was exceeded. The capillary barriers formed by fracture intersections can divert vertically flowing water from one fracture into horizontal fractures, thus, increasing the pool volume relative to storage in a single fracture.
Wetting front pattern, including fluid cascades and stalled flow paths, and discharge flux to exit fractures with time were not repeatable for our experimental fracture network.
We performed four tests to examine the repeatability of system behavior in a fractured rock network under similar wetting and initial moisture conditions. The results suggest that fractured rock systems do not exactly repeat wetting advance from one test to the next even under controlled laboratory conditions. The large variations in the observed flow patterns and discharge flux among these four tests seem to exceed what might be anticipated based on the small differences in test conditions.
Our tests yielded a number of significant observations and generate more unanswered questions. Of particular interest is the possibility that fluid cascades may span multiple intersections and grow to be quite large in natural fractured rock networks. With this possibility in mind, vadose zones consisting of fracture networks might harbor large multi-intersection "pools" that gather slow downward flow for long periods of time and then discharge abruptly in a fluid cascade quickly traversing large vertical distances.
The results of this experiment suggest that common modeling approaches cannot reproduce the behavior of the experimental results at this experimental scale. The results presented here add to the growing body of evidence calling for a paradigm shift in conceptualizing unsaturated flow in fractured rock vadose zones. Popular modeling approaches commonly applied to simulation of flow in fracture networks cannot duplicate the nonideal, dynamical behavior observed in our experiments (Fairley et al., 2001, 2004). Conceptualizing flow in fractured rock networks in a volume averaged manner, as commonly assumed, may be inappropriate in many situations. Significant variation in the location and flux of network discharge may occur without apparent external causes. Changes of internal intermediate flow pathways (switching) are at least as common as continuous flow along stable flow paths. The interaction of capillary and gravitational forces can trigger events that can rapidly transmit water significant distances. The convergence of flow with depth is probably a common feature in fracture networks.
|
ACKNOWLEDGMENTS |
|---|
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONEXPERIMENTAL DESIGNRESULTSSUMMARY AND CONCLUSIONSREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  R. J. Lenhard and P. Meakin Water Behavior in Layered Porous Media with Discrete Flow Channels: Results of a Large-Scale Experiment Vadose Zone J., August 1, 2007; 6(3): 458 - 470. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| The SCI Journals | Agronomy Journal | Crop Science | |||
| Journal of Natural Resources and Life Sciences Education |
Soil Science Society of America Journal | ||||
| Journal of Plant Registrations | Journal of Environmental Quality |
The Plant Genome | |||
