Published in Vadose Zone Journal 2:338-348 (2003)
© 2003 Soil Science Society of America
677 S. Segoe Rd., Madison, WI 53711 USA
ORIGINAL RESEARCH PAPER
Discrimination of Flow Regions on the Basis of Stained Infiltration Patterns in Soil Profiles
Beatrice Kulli*,a,
Christian Stammb,
Andreas Papritza and
Hannes Flühlera
a Institute of Terrestrial Ecology, Soil Physics Group, Swiss Federal Institute of Technology (ETH Zürich), Grabenstrasse 3, CH-8952 Schlieren, Switzerland
b Swiss Federal Institute for Environmental Sciences and Technology (EAWAG), Water and Agriculture, CH-8600 Dübendorf, Switzerland
* Corresponding author (kulli{at}ito.umnw.ethz.ch).
Received 23 September 2002.
 |
ABSTRACT
|
|---|
Dye tracer experiments have often been used to study the prevailing flow regime, such as in investigating the role and extent of preferential flow of water. Flow patterns in two-dimensional profiles provide qualitative information on the infiltration regime but are difficult to analyze and compare quantitatively. The scope of this study was to develop a quantitative method to analyze the spatial distribution of the stained areas in vertical profiles, to identify differing transport mechanisms on the basis of the pattern information, and to analyze how the discriminated patterns correspond with soil properties and structure. Dye tracer infiltration experiments were performed on 25 plots at eight sites. The spatial distribution of the stained areas in vertical profiles was analyzed and compared using digital image processing. We first split the flow patterns into similarly stained horizontal layers based on the width distribution of stained areas. All of these layers identified in the flow patterns of all 25 plots were then partitioned into groups of layers with similar patterns by hierarchical clustering. The sequence of layers found in the pattern was finally interpreted with respect to transport mechanisms and qualitatively compared with the sequence of morphological layers observed in the soil profiles. The obtained classification reliably distinguished between zones of homogeneous infiltration and zones of preferential flow, but also between zones of narrow stained structures and zones of lateral spreading (e.g., sand or gravel lenses). Dye coverage and mean width of stained structures were the most indicative factors for the different clusters. We often found an agreement between the sequences of layers found in the flow patterns and the soil horizons. However, in all profiles we observed layers in the flow patterns that did not correspond to textural and structural layers observed in the filed. It seems that knowing the pedological horizons is important but not sufficient to understand the observed flow patterns. Since the flow pattern in a given layer always depends on the overlying soil layers, similarly textured soil layers do not necessarily exhibit equal patterns. However, soil layers with a given textural sequence (e.g., fine-course-fine) are reflected by typical flow patterns.
|
INTRODUCTION
|
|---|
DYE TRACER INFILTRATION EXPERIMENTS are widely used to study water flow and transport of solutes in soils. They have often been used to examine the prevailing flow regime, as for instance, the role and extent of preferential flow of water (Ghodrati and Jury, 1990; Flury et al., 1994; Gjettermann et al., 1997; Zehe and Flühler 2001; Weiler and Naef, 2003) or to guide selective sampling of flow paths and bypassed soil matrix (Bundt et al., 2000; Sinaj et al. 2002).
To compare stained flow patterns in a quantitative and objective way or to distinguish site specific variation from that occurring at different sites or under differing boundary conditions, we need to capture the geometric and distributional features of the pattern. Forrer et al. (2000) quantified the two-dimensional concentration distributions of dye tracer plumes, which infiltrated from line sources applied at the soil surface. They calculated horizontal and vertical distributions of the tracer mass and thereby reduced the data to one-dimensional quantities. This procedure provides information about mass balance, but misses the spatial characteristics of the flow pattern. Pattern features have been characterized with different methods. Flury and Flühler (1995) modeled the patterns and their development in time using a diffusion-limited aggregation algorithm. Baveye et al. (1998) used fractal geometry to describe flow patterns quantitatively.
Flühler et al. (1996) proposed a model distinguishing conceptually between (i) an "attractor zone", where homogeneous flow is funneled into preferential flow ports; (ii) a "transmission zone", where rapid and bypassing preferential flow is the main transport mechanism, and (iii) a "dispersion zone", where the solute is spreading out laterally. These concepts imply that a soil profile is a composite of discernible flow regions. Rather than assessing the flow regime for the entire profile, we therefore split up the flow patterns into similarly stained regions. One of the important factors affecting infiltration patterns are soil texture and soil structure. We often find layers in flow patterns that correspond to differently textured soil horizons, such as the sandy layer at about 0.6 m depth in the soil profile shown in Fig. 1, which lead to lateral spreading of the tracer. Therefore, we propose as a first step in the flow pattern analysis to distinguish layers of a given infiltration behavior. The flow patterns in observed layers of a given soil (e.g., a loamy layer above a sandy layer) may differ considerably, while on the other hand, different soils may exhibit similarly stained horizons, as for instance, the plow layer.
An obvious difference for the discrimination of different regions in a vertical flow pattern is the fineness of the stained areas. Whether the tracer solution infiltrates as a homogeneous infiltration front, as a broad plume or as a network of narrow flow paths depends on soil structure and texture as well as on the experimental conditions. This experimental observation can be explained by the changing extent and rate of lateral mixing in different soil materials and under different infiltration conditions (Flühler et al., 1996). An indicator for the extent of lateral mixing is the broadness of the stained areas. It can easily be obtained from a profile image by measuring the horizontal width of the stained areas. We consider these widths of the stained areas as a suitable indicator for changes of typical transport mechanisms within a soil profile.
In this study we developed a method for the detection of boundaries between layers with different staining patterns. Using a hierarchical classification method, these layers are then grouped into clusters of layers with similar patterns and the larger sequences of layers in the pattern are qualitatively compared with soil morphological layers at the same plot.
|
MATERIAL AND METHODS
|
|---|
Figure 2 summarizes the steps of the method described here.
Field Experiments
The field experiments were conducted at eight sites where 25 plots were investigated, resulting in 171 single images of flow patterns available for further analysis. Information on soil and experimental setup at the sites is given in Table 1.
A solution of the dye Brilliant Blue FCF (C.I. 42090, Hoechst, Frankfurt, Germany; dye content 
85%) at a concentration of 4 g L-1 was used to stain the flow paths. At some of the agriculturally used sites, the effect of plowing was simulated by homogenizing the topsoil before sprinkling. A total amount of 40 L m-2 of tracer solution was applied at the sites where no plowing was simulated, and 100 L m-2 was applied at the sites where plowing was simulated. The tracer solution was applied at a constant rate of 5 mm h-1 using the sprinkling apparatus described by Flury et al. (1994).
One day after sprinkling, a soil pit was excavated and vertical profiles were prepared in the central square meter of the plot. Using the method by Forrer et al. (2000), a Kodak (Eastman Kodak Co., Rochester, NY) color scale and gray scale as well as a 1 by 1 m frame were attached to the profile and photographed with the pattern to allow corrections by digital image analysis. The stained flow patterns were photographed on 4 to 10 profiles per plot. The distance between adjacent profiles was 0.1 m.
Image Processing
The color slides were developed and digitized onto a photo CD by a commercial laboratory. Image analysis included (i) geometrical correction, (ii) correcting inhomogeneous illumination, (iii) correcting variations in color between the pictures, and (iv) classifying each pixel as stained or unstained based on its red, green, and blue value. This procedure is described in detail by Forrer et al. (2000). The resulting binary images show the flow paths of the tracer solution in the soil. Each picture consists of 1000 by 1000 pixels, each pixel representing 1 mm2 of the original soil profile. Figure 3 illustrates the sequential image processing.
Defining and Discriminating of Layers in the Flow Patterns
The purpose of this step is to objectively discriminate homogeneous layers in the flow patterns of every plot. Different layers can be identified by differences in the spatial distribution of the stained area. The discrimination of different regions in a vertical flow profile was performed based on the fineness of the stained pattern. Therefore the horizontal width of the flow paths was chosen as input variable for the analysis. We measured the widths of the stained areas in 1-mm depth increments (horizontal line of the flow pattern images) and determined their statistical distribution for each depth. For that purpose, each row of the pattern image was extracted, and the lengths of the black pixel chains, indicating that the pixel belongs to the stained area, were measured as shown in the upper part of Fig. 4. The flow patterns as well as the widths of the flow paths are considered to be horizontally translation invariant. To look for layer boundaries representative for an entire plot and not only for one single profile, we merge the width distributions for a given depth of all replicated profile images.
View larger version (21K):
[in this window]
[in a new window]
|
Fig. 4. Analysis of the width distributions of the stained areas. Top: width distributions are determined for every horizontal line (1-mm depth increment). Bottom: for each depth zL the width distributions below and above the respective depth are pooled and compared.
|
|
We then separate the profile into two layers by maximizing the variability of the width distributions of the stained areas between those layers. Therefore we divide the profile images into two layers at a certain depth zL. We denote the upper part as the X layer and the lower as the Y layer. The pooled width distributions of the stained flow paths of all rows of the X layer are then compared with the pooled width distributions of all rows of the Y layer by the Wilcoxon Rank-Sum Test (Fig. 4, lower panel).
The null hypothesis is that the width distributions of the layers X and Y have the same median. Thus it is tested against the hypothesis that they differ. The test value Z is
 | [1] |
The elements of the pooled width distributions of all depth increments of the layers X and Y are ranked according to their width and the ranks of the X layer are summed up to WX. NX and NY are the number of width measurements of the two distributions. The module RS_TEST (Research Systems, Inc., 1997, p. 786–787) was used to carry out the rank sum test. An increasing Z indicates an increase of the difference between the two distributions.
Z was calculated for each position zL (separation depth of layer X and Y) of the flow patterns, resulting in Z as a function of depth. The position zL of the largest value of Z delineated a layer boundary for the initial iteration. The two parts above and below this new boundary were then analyzed separately following the same procedure, which yields two new boundaries (Fig. 5). This procedure was iteratively repeated. The Z value of the first iteration is at the boundary that coincides with the most characteristic change of the pattern. The maximum of the Z values decreases with each iteration, resulting in many boundaries with small Z values. In a first analysis, we stopped the iterations when no more Z > 1 were found. The number of boundaries found depending on Z was then analyzed to find a suitable Z value for every plot to eliminate the least significant boundaries.
View larger version (31K):
[in this window]
[in a new window]
|
Fig. 5. Maximizing Z to detect layer boundaries. Top: iterative discrimination of dissimilar layers. Bottom: all images of a given plot lined up.
|
|
Z might be used to calculate the probability by which two distributions differ. But in our case the probabilities of the Wilkoxon Test are not a suitable criterion to stop iteration because Z is not calculated independently for each depth and cannot be used to judge the statistical significance.
To find a suitable cut-off value for Z to discriminate the least meaningful boundaries, the number of boundaries was analyzed according to their Z value. Threshold values between 1 and 50 were tested, and the boundaries with Z values larger than that particular threshold were counted. The resulting monotonously decreasing function has a highly negative slope for small thresholds followed by a very small negative slope after a certain threshold. We consider the inflection point as suitable to cut off unnecessary boundaries. The curve can be approximated by two straight line segments of different slopes (Fig. 6). The first line was fitted to the first part of the function and the second line segment to the second part. The best fit of the two lines was calculated by minimizing the chi-square error statistic depending on the threshold value, where the discontinuity between the two lines was located. Figure 6 shows the number of boundaries as a function of the threshold value as well as the best fit of the two lines. The figure also shows the chi-square error depending on the location of the discontinuity. Only the boundaries with Z values larger than the threshold value of the discontinuity were taken into account for further analysis. This evaluation was performed independently for every plot.
View larger version (14K):
[in this window]
[in a new window]
|
Fig. 6. Top: number of boundaries depending on the threshold (black) and best fit of the two straight line segments with the least chi-square error (gray). Bottom: sum of squared errors for the fit of the line segments depending on the threshold.
|
|
Cluster Analysis
A hierarchical cluster analysis was performed to group the identified layers according to pattern similarity. This analysis was performed based on the width distributions of the individual layers. The input variables for the cluster analysis were "minimum", "maximum", "mean", "standard deviation", and "median" of the width distribution of the stained structures per layer, and the "dye coverage" (fraction of stained area in a layer). The clustering of the data was performed by the Ward's minimum variance method (Ward, 1963). The routines "dist", "hclust" and "cutree" of the programming environment R were used (Ihaka and Gentleman, 1996).
The method of Xu et al. (1993) was used to find the optimal number of clusters. Based on the minimum of "between cluster distances" (MBCD) and the sum of squared errors (SSE) within the clusters, we calculated the parameter E indicating the number of clusters most suitable for grouping of the data. E is calculated as follows:
 | [2] |
 | [3] |
 | [4] |
 | [5] |
 | [6] |
where i is the total number of clusters; djk is the distance between the jth and the kth cluster (Ward, 1963); M(i) is the minimum of djk when i clusters are distinguished; J is the SSE for i clusters; nj and nk are the number of samples in cluster j and k, respectively; and mj and mk are the centroids of the clusters. High values, or a global maximum of E(i), indicate cluster numbers for suitable partitioning of the data.
|
RESULTS AND DISCUSSION
|
|---|
Figure 7 summarizes the results for six plots, representing the spectrum of flow patterns observed in our experiments. Due to the homogenized topsoil at the agriculturally used sites, many of the remaining flow patterns not shown here looked similar to those of the plots Fi_B and Ru_R. Each row shows the results of one plot. The first column shows the background images of one of the replicated profile images per plot. These background images contain only the image information of the blue channel. At the respective wavelengths, the reflection (and light absorption) by the tracer is eliminated, but structural information is retained. The black horizontal lines show the textural and structural soil layers observed in the field. The second column illustrates the pattern variability found in the experiments, by showing one example flow pattern per plot. The third column shows the superimposed 4 to 10 flow patterns of each plot. The dark pixels in the resulting overlay image indicate many profiles having a stained flow path at that location, whereas light pixels indicate no or only few profiles being stained at this point. The results of the layer boundary detection are shown as red horizontal lines in the overlay images. The fourth column shows the results of the cluster analysis for a total number of 20, 7, and 3 clusters. The cluster memberships of a layer are indicated by the color in the respective depth. The cluster numbers are identified by the large color scale on the right-hand side of Fig. 7.
View larger version (75K):
[in this window]
[in a new window]
|
Fig. 7. Overview of the results for six sites (rows). From left to right: background image with textural and structural layer boundaries observed in the field; example flow pattern of each plot; overlay image of all flow patterns of the plot with the layer boundaries detected on the patterns as red horizontal lines; results of the cluster analysis for a total of 20, 7, and 3 clusters given by color bars. The color scale on the right identifies the cluster numbers by the colors.
|
|
Layer Detection
In total 535 layers were identified in the flow patterns of the 25 plots. Examples are shown in the third row of Fig. 7. For most of the plots, many narrow layers have been detected by our method. However, many of the boundaries turn out to be irrelevant after the classification because consecutive layers may belong to the same cluster.
Cluster Analysis
Figure 8 shows the dependence of the parameter E on cluster number obtained by the method of Xu et al. (1993). The curve exhibits three peaks. The global maximum occurs at a cluster number of 20. Two local maxima can be found at cluster numbers of 3 and 7, whereas the peak at a cluster number of 7 is almost as high as the global maximum at a cluster number of 20. To investigate the effect of the cluster number on the resulting layer classification, all three numbers of clusters indicated by the maxima of the parameter E were taken into account for the further analysis.
The results of the cluster analysis are shown in the fourth row of Fig. 7. The three color scales indicate the membership of each layer for a total number of clusters of 20, 7, and 3. On the right-hand side of Fig. 7 the cluster numbers are identified by the colors. Within a certain total number of clusters, layers identified by the same color belong to the same cluster. For different total numbers of clusters the colors cannot be compared directly.
The total number of clusters affects the degree of the detail of layer discrimination. Three clusters seem to be sufficient to capture the main characteristics of the infiltration patterns at the plot Fi_B and Ru_R, where zones of homogeneous infiltration and zones of preferential flow are nicely distinguished. At the other four plots (WE_03, WE_13, FF_R, and FF_4) the classification is definitely not sensitive enough to capture the main characteristics of the pattern. A total of seven clusters results in a more detailed sequence of layers, which captures the sequence of preferential-flow layers and tracer-spreading layers appropriately. In case of Ru_R and Fi_B, additional layers are detected showing the transition of the homogeneous infiltration zone to the zone where almost no tracer was found. At Ru_R these gradual changes are much closer together than at Fi_B, but both plots show a sequence of layers with the same cluster memberships in their flow patterns. Dividing the layers into a total of 20 clusters increases the complexity of the layer sequence for all sites, revealing the existence of several very small layers of enhanced or suppressed spreading (e.g., Clusters 18 and 19 at the plots WE_13 and FF_4). Generally the complexity of the sequence of layers for a total of 20 clusters goes beyond the elaborateness of layer differentiation in the field. In our case, a total number of seven clusters provides sufficiently detailed information to characterize these flow patterns and compare them with the sequence of soil horizons found in the field.
We think that the optimal number of clusters is specific for the degree of detail which is of interest. The distribution of the data in the variable space shows clusters on different scales. There may be more than one total number of clusters, dividing the data into reasonable groups, giving information with a different degree of detail. The degree of detail suitable for the specific question has to be found independently.
Analysis of the Characteristics of the Flow Patterns
Figure 9 shows box plots of the variables used for the cluster analysis (minimum, maximum, mean, standard deviation, and median of the pooled line sequence width distribution per layer as well as the dye coverage), depending on the cluster membership, for a total number of clusters of 3 and 7. Table 2 shows the main characteristics of the clusters for the same total numbers of groups. The division of all layers into 20 clusters is not discussed in this context, because the degree of detail exceeds the information needed for this purpose.
View larger version (33K):
[in this window]
[in a new window]
|
Fig. 9. Box plots, showing the variability of the variables for every cluster for a total number of clusters of 3 (above) and 7 (below).
|
|
In case of a total number of clusters of 3, the layers belonging to the Clusters 1 and 2 have high maxima of the width distribution of the stained flow pattern per layer and large dye coverage. The layers of Cluster 2 have a higher mean and median of the width distribution than those of Cluster 1, and a higher dye coverage. The layers belonging to Cluster 3 have low maximum, mean, median, and dye coverage. Cluster 2 represents almost closed homogeneous infiltration fronts. Cluster 1 contains layers with homogeneous infiltration with some unstained areas. Cluster 3 contains layers of preferential flow, but also layers with almost no stained areas.
Since we conducted our cluster analysis by a hierarchical algorithm, an increase of the total number of clusters usually leads to a division of clusters into several new groups. Increasing the total number of clusters from 3 to 7 leads to a division of Cluster 1 and 3 into three clusters each, while Cluster 2 remains the same (Table 2). For a total number of seven clusters, the layers belonging to Cluster 1, 2, and 3 have high maxima of the width distribution of stained areas and high dye coverage. However, they differ in mean and median, whereas the layers belonging to Cluster 2 have the highest mean and median of the width distribution of the stained areas, followed by Cluster 3 and Cluster 1. All three clusters contain layers of homogeneous infiltration. Cluster 4 contains layers where the flow changes from homogeneous infiltration to preferential flow. The three remaining clusters contain layers with preferential flow, reaching from layers with dye coverage around 0.5 to layers with almost no dye coverage. Layers with many flow paths, broader structures and higher dye coverage belong to Cluster 5. The layers belonging to Cluster 6 show fewer flow paths, finer structures, and less dye coverage. Cluster 7 contains zones of very few or no flow paths and low dye coverage.
Comparison of the Layers Found in the Patterns and the Soil Horizons
The next step is a qualitative comparison between the layer sequence found in the flow patterns and the soil horizons found in the field. Table 1 shows information about the soils at the experimental sites, including those of the example plots shown in Fig. 7. The first column of Fig. 7 shows the background images of one of the replicated profile images per plot. These background images contain only the image information of the blue channel. At the respective wavelengths the reflection (and light absorption) by the tracer is eliminated, but structural information is retained. The black horizontal lines indicate soil structural or textural changes observed in the field.
Based on the flow patterns, the sites can be roughly divided into two groups. Fi_B and Ru_R exhibit a transition from intensive staining to tracer absence at a depth of 1 m. To mimic the effect of plowing, the topsoil was homogenized on both plots before sprinkling. This top layer with a low bulk density shows a homogeneous infiltration pattern and no indication of preferential flow. At both plots, it was followed by a well-structured horizon with higher density. In the case of Fi_B, this layer was a clayey loam with many earthworm (Lumbricus terrestris) burrows and cracks. At Ru_R underneath the plow layer, we found a very homogeneous silt loam with definitely fewer worm burrows than at Fi_B. In these subsoil layers, the flow became preferential. In case of Fi_B, worm burrows were the most important flow paths. Further down the profile, the soil density gradually increased, and the stained flow paths (root channels and worm burrows) gradually disappeared. In the case of Ru_R, preferential flow was less pronounced, probably due to a smaller number of worm burrows and cracks.
Although only two structural soil layers were distinguished in the profiles at Fi_B and Ru_R, more than two layers were detected in the flow pattern, showing the transition of the homogeneous infiltration zone to the zone of preferential flow. Both plots showed the same sequence of layers in the pattern. Hence, changes in a flow pattern do not always match with textural layers or soil horizons detected in the field. Soil properties and texture may gradually change, causing additional layers in the flow pattern.
The other four plots show another type of flow pattern. The dye coverage generally is decreasing with depth, but also increases in some layers, indicating enhanced lateral spreading.
The two plots WE_03 and WE_13 were located at the same site. The soil was strongly layered and very heterogeneous due to the alluvial deposition of the material. Similar to the plots Fi_B and Ru_R, a layer with a homogeneous infiltration pattern was followed by a layer where preferential flow was initiated. On both plots, the topsoil was not homogenized. Grass was growing at that site and the topsoil was densely rooted. The infiltrating water first followed the root network. As the root density decreased with depth, the flow paths became thinner and dye coverage decreased. In the case of WE_03, the loamy layer of the top soil was followed by a sandy layer, at a depth of about 20 cm. This can be seen by the texture of the background image as shown in the first column of Fig. 7. In this sandy layer, the dye tracer spread laterally, the flow tongues widened, and the dye coverage increased. The sandy layer was followed by a layer with clay loam at a depth of about 40 cm. The front of the flow pattern stops across the entire profile at about the same depth. There are two explanations for this. The change in soil texture allowed only very few flow paths to be continuous to greater depth. On the other hand, infiltration may have been stopped at the time when the front reached the respective depth. Flow patterns are not only influenced by the soil structure, but also by the experimental conditions. The rate as well as the cumulated amount of infiltration and the initial conditions, especially the initial water content, play an important role for the infiltration regime.
In case of plot WE_13, the water infiltrated to greater depth on the right-hand side of the profile. As the background image shows (Fig. 7), the sandy layer at the 15- to 40-cm depth was only present on the left-hand side of the profile. Water storage capacity and lateral flow were probably greater in the sand lens on the left-hand side than in the loamy soil on the right-hand side. The dye tracer, which reached greater depth on the right-hand side of the profile, spread laterally once again in the gravel layer at the 75- to 90-cm depth. This example shows the influence of the upper soil layers on the flow patterns in greater depth.
The two plots FF_R and FF_4 show the same tendency of having zones with a decreasing dye coverage followed by zones with increasing dye coverage. However, unlike most of the other sites, they do not show a homogeneous infiltration front near the soil surface. The topsoil was a sandy loam. Layers of coarser material were found in the subsoil. Plot FF_4 was strongly compacted by multiple passages of a sugar beet harvester, while plot FF_R plot was left as it was (Kulli et al., 2003). Soil compaction decreased the infiltration capacity of the fine pore network in the main root zone of the grass covering the site. Some vertical earthworm burrows remained open and acted as preferential flow paths. Because of this difference in the topsoil, the two flow patterns look entirely different, but sand and gravel layers affected the patterns in both cases, giving them a characteristic change between layers with higher and lower dye coverage.
At the plot FF_R the maximum infiltration depth was not big enough to cover the whole profile depth. As described for the plot WE_03, the maximum infiltration depth depends on the total amount of tracer solution applied, but also on the water storage capacity of the soil (depending on the porosity and initial water content). The effect of the zone below the maximum infiltration depth of the tracer on the flow pattern could not be studied at FF_R. If the transport through the whole profile needs to be assessed in an experiment, one should make sure that the tracer solution reaches the depth needed. On the other hand, if we are interested in the effect of a certain amount of rainfall at certain intensity, the maximal infiltration depth is one of the results of an experiment for the given experimental conditions.
Not only the total amount of tracer solution applied, but also the sprinkling rate may affect the outcome of the experiment. At FF_4, because of a compaction of the top few centimeters of the soil, the water was ponding at the soil surface and preferential flow was initiated. At FF_R, the water was infiltrating without ponding, probably under unsaturated conditions at the same sprinkling intensity, and the earthworm burrows did not act as flowpaths. At a higher irrigation rate, under saturated conditions, the transport regime might drastically change at FF_R and the worm burrows may be activated for preferential flow. Several studies showed that preferential flow is more pronounced under high irrigation rates, especially under flooding conditions than under low irrigation rates (Flury et al., 1994; Gjettermann et al., 1997; McIntosh et al., 1999).
The importance of larger-scale structures for preferential flow and the transport regime were ascertained at the plot FF_4. On the background image some small layers can be recognized at depth of 0.4 to 0.5 m, which did not lead to a classification as different flow types, although the little changes in the patterns are reflected by the layer boundaries found. These layers seem to be too small to lead to an amount of spreading relevant for the cluster analysis as compared with the effect of the worm burrow on the flow pattern, which masked the small differences caused by the layers.
|
CONCLUSIONS
|
|---|
Conceptually, flow patterns may be divided into three categories associated with different flow regimes (Flühler et al., 1996) (i) an attractor zone, where homogeneous flow is funneled into preferential flow ports; (ii) a transmission zone, where rapid and bypassing preferential flow is the main transport mechanism; and (iii) a dispersion zone, where the solute is spreading out laterally. Looking at the sequence of clusters, to which the individual layers were assigned, the above categories of flow regimes are hardly found in the sequence (i) to (iii). In reality, different combinations can be observed. Sites with strongly layered alluvial soils showed one or more dispersion zone, while soils without sharp textural layer boundaries had a typical sequence of layers of a homogeneous infiltration front followed by layers of preferential flow with gradually decreasing widths of stained areas and decreasing dye coverage. A dispersion zone is lacking in these cases.
A total number of seven clusters turned out to be suitable for the partitioning of the layers found in the flow patterns. The groups of layers found in the cluster analysis can easily be assigned to the two flow regimes "homogeneous infiltration front" and "preferential flow". However, to find out whether a layer acts as a transmission or dispersion zone, the entire sequence of layers has to be studied.
The study showed that there is indeed a strong influence of the soil horizons on the infiltration of water into the soil and that many of the textural layer boundaries in the soil were reflected in the flow patterns. However, additional boundaries in the flow pattern were also found within the textural soil layers at Fi_B and Ru_R. Possible explanations for this effect are (i) that a certain structural differentiation may occur within a macroscopically observed textural soil layer and (ii) that textural layer changes are not always sharp boundaries but may also include gradients of changing soil properties leading to variations in the flow pattern. In contrast to this effect, small textural layers may become irrelevant for the transport regime and are not detected by the flow pattern analysis in the presence of larger-scale structures like earthworm burrows, as shown at FF_4. These two observations have to be considered for any assessment of infiltration patterns and transport regimes based on soil structural and textural information.
In addition, it is important to consider the influence of the experimental setup, boundary, and initial conditions on the results of infiltration experiments. The flow patterns are influenced by the total amount of tracer solution applied, the sprinkling rate, and boundary and initial conditions. Because upper layers in the profile have an influence on the flow patterns in the layers below, the pattern in a certain depth can never be analyzed independently. Finally, the chemical properties of the dye tracer applied and differences in sorption in different soil horizons may affect the outcome of analysis, an effect that was not examined in our experiments.
|
ACKNOWLEDGMENTS
|
|---|
Financial support for this work was provided by the Research and Development Fund of the Swiss Gas Industry (FOGA). We acknowledge the excellent collaboration of Dr. Michael Gysi and the staff of the Swiss Federal Research Station for Agricultural Economics and Engineering (FAT), and the opportunity for collaboration at the experiments at the FAT site in Frauenfeld. The help of J. Leuenberger in planning and conducting the field work and the contribution of M. Jaquillard in the field and his work on the evaluation of the pictures from FF during his Masters thesis are gratefully acknowledged.
|
REFERENCES
|
|---|
- Baveye, P., C.W. Boast, S. Ogawa, J.Y. Parlange, and T. Steenhuis. 1998. Influence of image resolution and thresholding on the apparent mass fractal characteristics of preferential flow patterns in field soils. Water Resour. Res. 34:2783–2796.
- Bundt, M., A. Albrecht, P. Froidevaux, P. Blaser, and H. Fluhler. 2000. Impact of preferential flow on radionuclide distribution in soil. Environ. Sci. Technol. 34:3895–3899.
- Flühler, H., W. Durner, M. Flury. 1996. Lateral solute mixing processes—A key for understanding field-scale transport of water and solutes. Geoderma 70:165–183.[ISI]
- Flury, M., H. Flühler, W.A. Jury, and J. Leuenberger. 1994. Susceptibility of soils to preferential flow of water: A field study. Water Resour. Res. 30:1945–1954.
- Flury, M., and H. Flühler. 1995. Modeling solute leaching in soils by diffusion-limited aggregation: Basic concepts and application to conservative solutes. Water Resour. Res. 31:2443–2452.
- Forrer, I.N., A. Papritz, R. Kasteel, H. Flühler, and D. Luca. 2000. Quantifying dye tracer in soil profiles by image processing. Eur. J. Soil Sci. 51:313–322.
- Ghodrati, M., and W.A. Jury. 1990. A field study using dyes to characterize preferential flow of water. Soil Sci. Soc. Am. J. 54:1558–1563.[Abstract/Free Full Text]
- Gjettermann, B., K.L. Nielsen, C.T. Petersen, H.E. Jensen, and S. Hansen. 1997. Preferential flow in sandy loam soils as affected by irrigation intensity. Soil Technol. 11:139–152.
- Ihaka, R., and R. Gentleman. 1996. R: A language for data analysis and graphics. J. Comput. Graph. Stat. 5(3):299–314.
- Kulli, B., M. Gysi, and H. Flühler. 2003. Visualizing soil compaction based on flow pattern analysis. Soil Tillage Res. 70:29–40.
- McIntosh, J., J.J. McDonnel, and N.E. Peters. 1999. Tracer and hydrometric study of preferential flow in large undisturbed soil cores from the Georgia Piedmont, USA. Hydrol. Processes 13:139–155.
- Research Systems, Inc. 1997. IDL reference guide. Vol. 2. Research Systems, Inc., Boulder, CO.
- Sinaj, S., Ch. Stamm, G.S. Toor, L.M. Condron, T. Hendry, H.J. Di, K.C. Cameron, and E. Frossard. 2002. Phosphorus exchangeability and leaching losses from two grassland soils. J. Environ. Qual. 31:319–330.[Abstract/Free Full Text]
- Ward J.H. 1963. Hierarchical grouping to optimize an objective function. J. Am. Stat. Assoc. 58(301):236–244.
- Weiler, M., and F. Naef. 2003. An experimental tracer study of the role of macropores in infiltration in grassland soils. Hydrol. Processes 17:477–493.
- Xu, S., M.V. Kamath, and D.W. Capson. 1993. Selection of partitions from a hierarchy. Pattern Recognit. Lett. 14:7–15.
- Zehe, E., and H. Flühler, 2001. Slope scale variation of flow patterns in soil profiles. J. Hydrol. 247(1–2):116–132.
TOP
ABSTRACT
INTRODUCTION
