Use of soil moisture dynamics and patterns at different spatio-temporal scales for the investigation of subsurface flow processes

Spatial patterns as well as temporal dynamics of soil moisture have a major influence on runoff generation. The investigation of these dynamics and patterns can thus yield valuable information on hydrological processes, especially in data scarce or previously ungauged catchments. The combination of spatially scarce but temporally high resolution soil moisture profiles with episodic and thus temporally scarce moisture profiles at additional locations provides information on spatial as well as temporal patterns of soil moisture at the hillslope transect scale. This approach is better suited to difficult terrain (dense forest, steep slopes) than geophysical techniques and at the same time less cost-intensive than a high resolution grid of continuously measuring sensors. Rainfall simulation experiments with dye tracers while continuously monitoring soil moisture response allows for visualization of flow processes in the unsaturated zone at these locations. Data was analyzed at different spacio-temporal scales using various graphical methods, such as space-time colour maps (for the event and plot scale) and binary indicator maps (for the long-term and hillslope scale). Annual dynamics of soil moisture and decimeterscale variability were also investigated. The proposed approach proved to be successful in the investigation of flow processes in the unsaturated zone and showed the importance of preferential flow in the Malalcahuello Catchment, a datascarce catchment in the Andes of Southern Chile. Fast response times of stream flow indicate that preferential flow Correspondence to: T. Blume (blume@gfz-potsdam.de) observed at the plot scale might also be of importance at the hillslope or catchment scale. Flow patterns were highly variable in space but persistent in time. The most likely explanation for preferential flow in this catchment is a combination of hydrophobicity, small scale heterogeneity in rainfall due to redistribution in the canopy and strong gradients in unsaturated conductivities leading to self-reinforcing flow paths.

Abstract. Spatial patterns as well as temporal dynamics of soil moisture have a major influence on runoff generation. The investigation of these dynamics and patterns can thus yield valuable information on hydrological processes, especially in data scarce or previously ungauged catchments. The combination of spatially scarce but temporally high resolution soil moisture profiles with episodic and thus temporally scarce moisture profiles at additional locations provides information on spatial as well as temporal patterns of soil moisture at the hills lope transect scale. This approach is better suited to difficult terrain (dense forest, steep slopes) than geophysical techniques and at the same time less cost-intensive than a high resolution grid of continuously measuring sensors. Rainfall simulation experiments with dye tracers while continuously monitoring soil moisture response allows for visualization of flow processes in the unsaturated zone at these locations. Data was analyzed at different spacio-temporal scales using various graphical methods, such as space-time colour maps (for the event and plot scale) and binary indicator maps (for the long-term and hillslope scale). Annual dynamics of soil moisture and decimeterscale variability were also investigated. The proposed approach proved to be successful in the investigation of flow processes in the unsaturated zone and showed the importance of preferential flow in the Malalcahuello Catchment, a datascarce catchment in the Andes of Southern Chile. Fast response times of stream flow indicate that preferential flow Correspondence to: T. Blume (blume@gfz-potsdam.de) observed at the plot scale might also be of importance at the hillslope or catchment scale. Flow patterns were highly variable in space but persistent in time. The most likely explanation for preferential flow in this catchment is a combination of hydrophobicity, small scale heterogeneity in rainfall due to redistribution in the canopy and strong gradients in unsaturated conductivities leading to self-reinforcing flow paths.

Introduction
Identification of patterns of soil moisture response to rainfall and especially the vertical dynamics of soil moisture at the hillslope or plot scale can be useful for the investigation of runoff generation processes in previously ungauged or data scarce catchments (runoff generation is here referring to all components of streamflow: groundwater, subsurface stormflow and surface runoff). When investigating runoff generation processes in a previously ungauged catchment it becomes obvious from the start that the equipment we are about to install is insufficient. There will be neither enough data points in time nor in space to characterize these processes in their temporal and spatial variability. A possible way to overcome this problem is the approach where a multitude of experimental methods is applied within a relatively short time frame, producing a data set that highlights a multitude of angles and aspects of catchment functioning. This type of study was carried out in the Malalcahuello Catchment in the Chilean Andes and is described in Blume et a1. (2008a).
The hydrologic behaviour of young volcanic ash soils in Chile is little understood and no studies of high temporal Published by Copernicus Publications on behalf of the European Geosciences Union. resolution soil moisture dynamics were found in our literature search. However, in other parts of the world such as New Zealand or Japan the soil moisture dynamics of volcanic ash soils has been investigated to some extent: Hasegawa (1997) used hourly TDR data to investigate soil water conditions and movement, Eguchi and Hasegawa (2008) used TDR and tensiometer data to estimate preferential flow, Musiake et aI. (1988) used tensiometric observations and a numerical model to study infiltration and drying behaviour of these soils and Van' t Woudt (1954) used 19 smalllysimeters to investigate subsurface stormflow.
Soil moisture data has been used as a means to understand runoff generation in other parts of the world (e.g. Kienzler and Naef, 2007;Meyles et aI. , 2003;McNamara et aI. , 2005;Frisbee et aI. , 2007;Germann and Zimmermann, 2005;Zhou et aI., 2002;Hino et aI., 1988) or for the investigation of the effects of changes in land use or management on hydrological processes (Williams et aI. , 2003;Starr and Timlin, 2004). See also Robinson et aI. (2008) and Vereecken et aI. (2008) for comprehensive reviews on measurement techniques and the value of soil moisture data, respectively. In most studies soil moisture was measured either with high spatial or with high temporal resolution, thus producing either spatial soil moisture patterns (Bardossy and Lehmann, 1998;Brocca et aI. , 2007;Meyles et aI. , 2003;Williams et aI., 2003;Western et aI. , 2004;Rezzoug et aI. , 2005;Nyberg, 1996) or information on the dynamics (e.g. McNamara et aI., 2005;Starr and Timlin, 2004;Frisbee et aI. , 2007). A combination of both can only be achieved with either a large number of probes measuring continuously such as in Starr and Timlin (2004) and Taumer et aI. (2006) or with geophysical methods such as described for example in Zhou et aI. (2001), were electric resistivity tomography was used to investigate soil moisture dynamics on a 3.5 x 3.5 m plot at hourly resolution. However, the first of these two options is cost-intensive while the second is predominantly carried out on grassland, fields or bare soils with little topography and is not feasible in complex terrain. Furthermore, as no general petrophysical relation to derive soil moisture from specific resistivity values is available, careful inversion and site specific calibration is needed. Combining data sets with different spatio-temporal resolution thus might be a viable cost-efficient alternative for difficult terrain. A recent study using a similar approach has been carried out in a small catchment in Australia, where 500 datapoints sampled on a weekly basis for 4 weeks were combined with 7 continuously measuring stations (Martinez et aI. , 2008).
At the Malalcahuello Catchment soil moisture was measured on two steep hills lope transects. Data was collected with a data logger at high temporal resolution at three points and manually at irregular intervals at 11 additional points. Each measurement produces soil moisture data for 6 different depths along a vertical profile. While this is still a very small number of data points it is nevertheless possible to get a general understanding of the major processes occur-HydroI. Earth Syst. Sci., 13, 1215-1234, 2009 ring within the unsaturated zone of this catchment. Data was analyzed using various graphical methods allowing for data exploration at different spatio-temporal scales. By carrying out rainfall simulation experiments using a dye tracer over each of the continuously measuring probes it was possible to corroborate our perception of flow in the unsaturated zone at these locations. This combination of high temporal resolution soil moisture measurements, rainfall simulation experiments and the use of dye tracers to reassess the conclusions gained from the soil moisture time series is noteworthy and has a high potential for synergetic effects. However, only one other study (Weiler and Naef, 2003) using a similar approach but a slightly different layout was found in our literature search. The study at the Malalcahuello catchment furthermore included the analysis of response times at the event scale, yearly soil moisture dynamics, spatial patterns and their long-term dynamics for 14 locations and 6 depths and the investigation of small scale soil moisture variability at the decimeter scale.
The four main questions of the study in the Malalcahuello Catchment were: become significant: on average only 80% of total precipitation reaches the forest floor as throughfall (measured with a raster of throughfall collectors with a diameter of 10.5 cm). However, throughfall amounts are highly variable and can in places also exceed total precipitation (measured outside the forest) by a factor of 2 or even 3 (Blume et aI. , 2008a). Above the tree line (20% of the catchment area) there is no significant vegetation cover.
The soils are young, little developed and strongly layered volcanic ash soils (Andosols, in Chile known as Trumaos) (Iroume, 2003;Blume et aI. , 2008a). High permeabilities (saturated and unsaturated), high porosities (60-80%) and low bulk densities (0.4-0.8 g/cm3) are typical for volcanic ash soils. They also usually show a strong hysteresis and irreversible changes (e.g. in water retention) with air-drying (Shoji et aI., 1993). Soil hydraulic conductivities for the soils in the Malalcahuello catchment were determined for soil cores (8 cm diameter) in the lab with the constant head method and range from 1.22 x 10-5 to 5.53 x 10-3 mls for the top 45 cm (independent of soil horizon), with an average of 5.63 x 10-4 mls (42 samples). The mean conductivity for the fine gravel and pumice layers is 1.88 x 10-3 mls (9 samples with a standard deviation of 6.6 % (16 samples). Grain size distributions for the upper horizons resulted in an average of 66.5 % sand, 30.4% silt and 3% clay. In the coarse layers the grain size fraction 2:2 mm ranges from 38-86% (Blume et aI. , 2008a). Layer thickness is also highly heterogeneous, and can range from 2-4 cm to several meters. It was not possible to establish a soil catena along the hillslope, probably due to the young age of the very little developed ash soils. For details on the soil physical characteristics of the major soil horizons in this catchment see Table 1. Depth to bedrock is unknown, however, manual augering to depths of 2-3 m, at one occasion even 7 m was possible (Blume et aI. , 2008a). At the locations of the 4 wells at the lower end of this slope ( Fig. 1) groundwater was found in depths of 1.8-3.2 m below the surface. However, at many other locations on this slope no ground water was found in auger holes of similar depths . For a more detailed description of the Malalcahuello Catchment see Blume et al. (2008a). The climate of this area can be described as temperate/humid with altitudinal effects. There is snow at higher elevations during winter and little precipitation during the summer months January and February. Annual rainfall amounts range from 2000 to over 3000 mm, depending on elevation.
Hydrol. Earth Syst. Sci., 13, 1215-1234,2009  An overview of catchment layout and topography as well as instrumentation is given in Fig. 1.
The current perception of the hydrological processes in this catchment can be summarized as follows: Rainfallrunoff response is generally fast, in part due to high hydraulic conductivities and thus fast matrix flow. Lateral subsurface flow is assumed to be important as strongly differing soil layers offer interfaces either for flow along capillary barriers or impeding layers. Lateral flow has furthermore been observed in the dufflayer during a high intensity rainfall simulation experiment (Blume et aI., 2008b). A large subsurface storage is indicated by the deep unsaturated zone, the high porosities and the fact that event runoff coefficients are low, with 1-10% for 17 events analyzed in 2004/2005, of which a third are smaller than 2% (Blume et aI. , 2007), while yearly runoff coefficients (> 60%) as well as the baseflow index (> 7 5 %) calculated for the years 2004 and 2005 are high (Blume et aI. , 2008a). Furthermore a shift in processes from dry to wet season (summer to winter) is indicated by a change in flow patterns observed through dye tracer experiments and a change in ground water surface water interaction observed close to the catchment outlet (Blume et aI. , 2008b).

Approach
The approach of this study is based on the measurement of spatially scarce but high temporal resolution soil moisture profiles on the one hand and episodic and therefore temporally scarce soil moisture profiles on the other hand. These two datasets combined with additional experiments Hydrol. Earth Syst. Sci., 13, 1215-1234,2009 were used to investigate soil moisture response patterns and thus flow in the unsaturated zone. This included the analysis of yearly soil moisture dynamics as well as of event response patterns resulting in the deduction of flow processes, the use of rainfall simulation experiments with dye tracers to corroborate these deductions, but also the analysis of response times at the event scale. The episodic measurements along the hillslope transects allow for the analysis of spatial patterns and their long-term dynamics for 14 locations and 6 depths and for the investigation of small scale soil moisture variability at the decimeter scale. Figure 2 gives an overview of the data sets used in this study. The synergies arising from their combination are described in Sects. 4 and 5.

Streamftow, groundwater levels and rainfall
Water levels in stream and groundwater were measured with capacitive water level sensors (WT-HR Trutrack) at 5-10 min time intervals. Stream water levels were converted to discharge with the help of a rating curve and groundwater levels are reported in reference to the well datum or the datum of the stream gauge. Rainfall was measured with a tipping bucket rain gauge with a resolution of 0.27 mm. A climate station maintained by the Universidad Austral de Chile is located in a nearby forest plantation at 1270 m elevation. This climate station has been logging the parameters rainfall, temperature, relative humidity, wind direction and velocity as well as global radiation at hourly intervals since 1999. During the winter of 2005 an ultra-sonic snow height sensor was also installed at this climate station. For more details on the experimental methods applied in the Malalcahuello Catchment see Blume et al. (2008a

Soil moisture profiles
To investigate the dynamics as well as the trends in soil moisture profiles along the hillslope, measurements were carried out at two transects with FDR (frequency domain reflectometry) profile probes (Delta-T) in 10,20, 30, 40, 60 and 100 cm depth. Both transects are located on the eastern slope close to the main stream gauging station SI (Fig. 1). This hillslope was selected for being quite typical for the catchment in slope and vegetation as well as for its accessibility. At each depth soil moisture is measured in a soil volume of about 2500 ccm, a cylinder with a radius of 10 cm. The absolute measurement error of about 3% (as given by the manufacturer) is for the measurement of the dynamics of soil moisture of less importance. The error of the measured dynamics, i.e. the error of the values relative to each other is likely to be smaller than the absolute error. As a result of the special characteristics of the volcanic ash soil, such as the extremely high porosities and the fact that volcanic glass is a primary constituent, the built-in standard calibrations were not applicable. It was thus necessary to calibrate the probe specifically for this type of soil with gravimetrically determined water contents of 19 soil samples of the upper horizons (horizons 1 and 2 in Table 1), which generally cover depths from 5 to 130 cm. Water contents for the calibration curve ranged from 16 to 51 Vol%. The calibration resulted in a correction of the supplied generalised soil calibration given by the manwww.hydrol-earth-syst-sci.net/13/1215/2009/ ufacturer: e corr = 0.8126 x e + 0.1145. With this calibration curve it was possible to reproduce the gravimetrically determined values with an R2 of 0.94 and a median absolute error of only 1 Vol%. These profile probes do not measure within a purely circular field as the sensor only extends about two thirds around the probe. For manual measurements it is possible to cover the full circle by taking three measurements, turning the probe by 120 0 each time. While many of the soil cores used in the soil physical characterisation (Table 1) were taken at or close to this hillslope, there is no soil physical information available for the exact location of each of the sensors. Due to the large variability of layer thickness, it is also difficult to attribute the sensors to a specific soil horizon.
Three profile probes were connected to a data-logger and were measuring continuously with a temporal resolution of 10 min. The data set extends from March 2003 to May 2006 for the lowest probe and from October 2004 to May 2006 for the two upper probes. For easier reference the three probes are numbered: probe 1 is located at the lower end and probe 3 at the upper end of the hillslope transect. A fourth probe was used for manual measurements at 11 points along the transects . 5 of these measurement locations supplement the transect of the continuously measuring probes, while the remaining 6 form a second transect located to the north of the first (Fig. 1). The points on the transects were roughly evenly spaced. These manual measurements were carried out at 41 occasions at irregular time intervals during field campaigns (December 2003-February 2004, October 2004-December 2004, November 2005-December 2005 and April-May 2006).

Rainfall simulation experiments
To investigate subsurface flow patterns 10 dye tracer experiments with brilliant blue were carried out in the Malalcahuello catchment in 2004 and 2005 (for details see Blume et al. (200Sb)).
Furthermore, similar rainfall simulation experiments were conducted in May 2006 at the locations of the continuously measuring soil moisture probes. It was thus possible to test our perception of flow in the unsaturated zone at these profiles. The plot size was 1.2 mx 1.2 m with the probe situated in the center. For all experiments the dye tracer Brilliant Blue with a concentration of 4 gl l was used. The dye was applied with a hand pressurized pesticide spray er in order to simulate rainfall. 30 I of the dye were sprayed over a period of 3 h. This corresponds to a total of 25 mm at application rates of S.3 mml h on average. Profiles of the plots were excavated the following day and photos of the dye patterns were taken with a digital camera.

Response times
Response times were calculated for both, dry and wet season from the time series of rainfall, soil moisture, groundwater levels at well Wl ( Fig. 1) and streamflow (all with lOmin resolution). Response time in this case was defined as the time period between begin of precipitation and first response of soil, ground-and stream water. The following threshold values were chosen to identify the point of first response in the time series: an increase of 0.2 Vol % in soil moisture, an increase of 0.005 m in ground water level and an increase of 0.01 m 3 Is in stream flow. As the main interest are the relative changes in response times from winter to summer, the absolute values of these thresholds are of less importance as long as consistent thresholds (> noise of the time series) are used for the entire analysis.

Data analysis
Data was analyzed at different space-time scales using various graphical methods. The space-time scales analyzed included event and longterm scale, point and hillslope scale.
Simple line plots were used to analyze annual dynamics as well as small scale variability of soil moisture.
Soil moisture patterns at the hillslope scale are investigated with the help of binary indicator maps for each depth. These maps show locations where soil moisture is above/below a certain threshold. Here we selected the median value to indicate wetter/drier than average locations as well as the 75 % quantile to indicat wet spots. These thresholds are calculated Hydrol. Earth Syst. Sci., 13, 1215Sci., 13, -1234Sci., 13, , 2009 for each depth and sampling occasion at all sampling locations on both transects and depend thus on time and depth. The temporal aspect of these binary patterns is included in the indicator map by plotting location on the slope on the yaxis and time on the x-axis, thus giving an idea of pattern persistency.
Event scale datasets with 10 min temporal resolution were analyzed with the help of colour maps which included temporal dynamics similar to those used by Weiler and Naef (2003). Here, time is plotted on the x-axis while depth is plotted on the y-axis. Soil water content at each depth and point in time is visualized by color, changing from blue to red with increasing wetness. For additional information the response of streamflow and groundwater, as well as the rainfall intensity at each point in time were also included. Color scales were adapted from one event to the next in order to get the best "color resolution" possible, producing clearer patterns of response. It thus becomes possible to explore and identify patterns in moisture response, patterns in space and time that are much more difficult to identify in the classical line plots of soil moisture dynamics. In a next step flow processes were deduced from these patterns.

Estimation of potential for self-reinforcing flow paths
A rough estimation was used to investigate the potential of flow paths being self-reinforcing due to a strong gradient in moisture content: potential "flow within the flow path" was compared with "flow perpendicular to the flow path interface" . The Van Genuchten parameters were obtained through fitting the Van Genuchten equation to the soil moisture characteristic curves. The soil moisture characteristic curves were determined with a pressure chamber for the first two horizons below the humus layer (3 samples each). These horizons are the two most commonly found soil layers which can extend to a depth of 130 cm. Then the gradient in matric potential was determined from the soil moisture characteristic curves for different gradients in soil moisture. The Van Genuchten equation can then be used to determine the unsaturated conductivities for the chosen matric potential. A longitudinal distance of 10 cm was chosen, as this is the range of the soil moisture probes. The ratio of the gradients in potential (across interface/within flow path) was then compared with the ratio of the unsaturated hydraulic conductivities (within the flow path/across the interface). The effective unsaturated hydraulic conductivity across the flow path interface was calculated by treating the interface as two layers of differing conductivity (due to the differences in water content) and therefore using the harmonic mean for its calculation. The gradient in potential within the flow path is assumed to be equal to 1 [cm H20!cm]. In case the ratio of the unsaturated hydraulic conductivities (Ke) is much larger than the gradient in matric potential (/11jJ) across the  Dekker and Ritsema (1994) for 12 air dried soil samples from 4 different locations and depths from 5 to 80 cm. The WDPT test is a simple test for the persistency of water repellency where a water drop is applied to a soil sample and the time between application of the water drop and its penetration into the soil is measured. Water drop penetration times for air dried soil have been classified by Dekker and Ritsema (1994) into 5 classes: wettable «5 s), slightly water repellent (5-60 s), strongly water repellent (60-600 s), severely water repellent (600-3600 s) and extremely water repellent (> 3600 s  (Fig. 3). However, as soon as the first rainfall starts in autumn, soil moisture values rebound to their previous level. Overall, the temporal variability of soil moisture at each of the sensors is much lower than the spatial variability along the depth profile (see also Blume et al. (2008a), for more detailed statistics on this topic). The profile at probe 1 (lowest on the slope) had generally higher moisture contents, but otherwise no trends along the transect could be established. Interestingly, events that lead to strong responses in discharge and ground water level (e.g. in May and June 2005), do not show a comparably strong response in soil moisture (Fig. 3). As even the slope ground water (well 5 -for location see Fig. 1) responds strongly, this difference in response can not be attributed to snow melt events in the upper part of the catchment. A more likely explanation is that the unsaturated subsurface in this catchment is close to equilibrium or steady state (see also the explanations to event 3, in Sect. 4.4). This means that soil moisture change is close to zero, but waterflow itself is non-zero. Perturbations of rainfall and drought only lead to small and short responses in soil moisture as water is efficiently being transported to greater depths. While this efficient transport of water in the www.hydrol-earth-syst-sci.net/13/1215/2009/ unsaturated zone only results in small changes in soil moisture in the upper meter, it also leads to a fast and pronounced response in ground and surface water. Subsurface flow for the three events indicated with arrows in Fig. 3 is analyzed in more detail in Sect. 4.4. 4.2 Soil moisture spatial patterns at the hillslope scale Using the manually measured soil moisture profiles as well as the corresponding logged data sets it was possible to establish a picture of spatial as well as coarse resolution temporal soil moisture characteristics. Soil moisture patterns at the two transects are depicted in Fig. 4 using the binary indicator maps.
Surprisingly, no general trends along the transects could be identified, apart from for the 10 cm sensors: There seems to be a correlation with position on the slope for the 10 cm depth, but not for the deeper sensors. At 10 cm depth the lower half of the slope is generally wetter than the upper part of the slope. This is probably due to shading effects: the deeper in the steep valley the fewer hours of direct sunshine. Another possible explanation is downslope flow accumulation in the o-horizon. The northern transect is wetter than the southern transect, which is probably due to denser vegetation. It was found that spatial patterns are generally persistent over time.
4.3 Variability of soil moisture at the decimeter scale and preferential flow patterns Figure 5 shows the small scale variability in soil moisture measured by twisting the probes at the manual measurement points H4 and H5 (for location of these points see Fig. 1). Differences in soil moisture around the probe can be very pronounced, e.g. it is wetter/drier in one direction than in the others. These patterns of small scale variability are generally persistent over time while the temporal variability of soil moisture at this time resolution (irregular intervals during field campaigns) is generally low as the dynamics of response act on much shorter time scales (compare Fig. 2  . Directional or small scale variability at manual measurement points H4 and H5. Measurements 1, 2, 3 at each date are repetitions within the same access tube, after rotating the probe by 120°. Measurement 1 is carried out with the sensor directed ups lope, while for measurements 2 and 3 the sensor is angled downslope to the right and left, respectively. Note that the measurements are taken at irregular time intervals . Samples "forest 1-3" were taken at the slope of the soil moisture transect, while samples named "pine" were taken in a pine plantation downstream of the catchment outlet. The sampling sites of "forest 1-3" do not correspond to the locations of the soil moisture probes 1-3. location depth wettable slightly strongly severely extremely (cm) water repellent water repellent water repellent water repellent study in this catchment (Blume et aI. , 2008b), where preferential flow proved to be the rule for all forested sites (9 experiments), with slightly varying flow patterns during dry and wet season (Fig. 6). If a soil moisture sensor was located near the interface of such a preferential flow path, soil moisture would differ considerably depending on the direction of the measurement. The scale or width of the flow paths identified with the dye tracer is in the order of < 3 decimeters, thus matching the scale of the measurement. A sensor showing no directional variation must therefore be located either in the center of a flow path or in the center of the matrix with no flow path within reach of the measurement. Ritsema and Dekker (1996) also used small scale (5-lOcm) variability of soil moisture as a measure for preferential or finger flow. In their study moisture gradients between flow paths and non-flow areas ranged between 3 and 6 Vol%. Assuming small scale soil moisture variability does indeed indicate the presence of a preferential flow path, the fact that in Malalc-Hydrol. Earth Syst. Sci., 13, 1215Sci., 13, -1234Sci., 13, , 2009 ahuello 68 % of all sensors show this type of variability also gives us a measure of the importance of preferential flow in this catchment. There are five possible explanations for the surprising persistency of the small scale soil moisture patterns (or preferential flow patterns) over the course of more than one and a half years (Fig. 5): 1. These patterns might be caused by air gaps between access tube and the surrounding soil due to faulty installation. However, special care was taken to avoid this problem, by using the auger supplied by the manufacturer of the probes. Furthermore no noticeable air gaps were found during excavation of the probes at the end of the field study, on the contrary, probes were sitting tightly in the soil. Air gaps are also likely to cause low soil moisture readings, but small scale variability was also found at higher moisture contents (Fig. 5) www.hydrol-earth-syst-sci.net/13/1215/2009/ Table 3. Response characteristics and antecedent conditions for the three events shown in Fig. 7 (P tot = rainfall amount, PInt = maximum rainfall intensity, antec.l1 = antecedent mean soil moisture content for the top 30 cm, max . /";.() = max . increase in soil moisture of all sensors, antec.Q = antecedent streamflow, max . /";.Q = max. increase in streamfl ow, antec.GW = antecedent ground water level at well Wl relative to the well datum at 2.55 m below the soil surface, max . /";.GW = max. rise in groundwater levels 2. They might also be due to textural differences. However, as the sensors have only a range of 10 cm the measured volume is likely to be located within a single layer.
3. These patterns might also be induced by roots, which are not likely to change position on this time scale. However, roots were only found in some of the instances where these preferential flow patterns were observed during dye tracer experiments.
4. They might be due to hydrophobicity in some parts of the soil, which would produce self reinforcing patterns likely to persist if not subjected to long periods of saturation. The change in flow patterns from dry to wet season, which was found in the dye tracer experiments (Fig. 6, for details see Blume et aI. , 2008b), supports this theory (Fig. 2). Potential water repellency of soil samples from 5 to 80 cm depth was tested with the Water Drop Penetration Time test. It was found that while the top horizons show strong to extreme potential water repellency, samples from greater depths are wettable (Table 2). However, this test determines only potential water repellency, measured in air dried soil. Hydrophobicity under field conditions is likely to be less pronounced and spatially heterogeneous as a result of water redistribution by canopy, litter, microtopography or by variability in soil organic matter Ritsema and Dekker, 1995). The O-horizon in this catchment only has a thickness of about 5 cm, but due to its higher organic matter content, it is likely to enforce redistribution processes due to spatially heterogeneous water repellency.
5. These patterns could also be self reinforcing due to the strong gradient in soil moisture itself, leading to faster vertical transport within the wetter area (the flow path) than lateral flow into the drier area as a result of the gradient in matric potential.
Possibility no.5 was investigated with a simple estimation: If the ratio of the of the unsaturated hydraulic conductivities is much larger than the gradient in matric potential across the interface (Eq. 1), flow paths are likely to be self-reinforcing where most of the variability was observed. The gradient of 5 Vol% chosen to investigate this phenomenon was in the upper range of gradients observed in the field. The gradient in potential within the flow path is assumed to be equal to 1 [cm H20/cm]. It was found that flow paths would indeed be self-reinforcing for a pure sand (with a ratio of Ke up to 11 times larger than the ratio of ,6.1jJ), however, in these ash soils, which have a fraction of at least 20% silt, it is very difficult to achieve these conditions (the ratio of Ke is less than half that of the ratio of ,6.1jJ). It is thus unlikely that solely the gradient in soil moisture causes the flow paths to persist in time. Nevertheless, if the unsaturated conductivity across the interface is further diminished by the effects of hydrophobicity a persistant pattern becomes more probable. Furthermore this type of soil is known to be hysteretic (Shoji et aI. , 1993;Musiake et aI., 1988) thus causing a shift in the wetting curve compared to the here used draining curve, which could also change the outcome of this rough estimation. Persistent fingers as a result of hysteresis of the soil moisture characteristic curves were described by Selker et al. (1996) and Nieber (1996). Nieber (1996) explains that fingers will persist if the water entry pressure on the main wetting curve is smaller then the air entry pressure on the main drainage curve. However, due to lack of information on the main wetting curve, this effect cannot be assessed for the soils in the Malalcahuello Catchment.

Soil moisture dynamics on event basis
The soil moisture response was analyzed with the help of space-time colour maps for 34 rainfall events during the period from December 2004 to December 2005. The temporal resolution of these plots is 10 min. Rainfall amounts for the events analyzed ranged from 16-150mm (2 day periods) with a median value of 48 mm. Maximum rainfall intensities ranged from 0.5-8.6 mm/lO min with a median value of 2 mm/lO min.  Explanation of color bars from top to bottom: The uppermost bar shows lO-min rainfall intensity: dark blue is equivalent to ° mm/lO min, dark red is equivalent to ::::4 mm/lO min. The two following bars show the increase of discharge and rise of groundwater level (at well WI), respectively. The color scale is stretched from minimum to maximum values. Down below follow the three wide bars representing the soil moisture response at the hillslope transect. The upper bar corresponds to the profile probe at the upper end of the slope (P3), the middle bar to the mid-slope probe (P2) and the lowest bar to the profile probe at the lower end of the slope (PI) . Within these three wide colour bars, each stripe corresponds to a certain depth: 10, 20, 30, 40, 60 and 100 cm. ° on the soil water color scale corresponds to antecedent moisture content. The arrows indicate the most prominent features and are numbered for easier reference. The small plots to the left of each soil moisture bar show antecedent moisture conditions for each depth as well as the median values of soil moisture content as reference.
Three typical events are shown in Fig. 7. The timing of the events is indicated by arrows in Fig. 3. Probe 1 is located at the lower end and probe 3 at the upper end of the hillslope transect. Details of event response and antecedent conditions are listed in Table 3.
For the first event, the event on 3 March 2005 (Fig. 7a), total precipitation amounted to 52 mm with a highest inten-Hydrol. Earth Syst. Sci., 13, 1215Sci., 13, -1234Sci., 13, , 2009 sity of 8.6 mm/IO min. The maximum change in soil moisture was high with 8.6 Vol%, which is due to the fact that this event was the rainfall event with the lowest antecedent moisture content of all events studied (Moisture contents for  1 in Fig. 7a), due to high rainfall intensities and high hydraulic conductivities, and b) very little reaction at the 10 cm depth for probes 1 and 3 (arrow 2 in Fig. 7a). This is probably due to hydrophobicity resulting from the dry antecedent moisture conditions. This pattern was observed only for the 3 driest occasions. The dry antecedent conditions also make steady state conditions unlikely, where flow without change in moisture content is possible. Soil moisture increase below the hydrophobic layer thus must be due to lateral inputs, either at the the decimeter scale or at the hillslope scale.
The rainfall event on April 6th 2005 (Fig. 7b) has a total precipitation of 28 mm and only low rainfall intensities. The maximum increase in soil moisture, as well as streamflow and ground water levels are low with 3.8 Vol %, 0.06 m 3 / s and 3 cm, respectively (Table 3). Antecedent moisture contents correspond to the median values for most depths, apart from the 10 and 20 cm depths at probe 1. The major patterns identified for this event are: a) fast vertical water transport, due to high hydraulic conductivities (arrow 3 in Fig. 7b), and b) late but persistent response at 100 cm depth for probes 2 and 3, while no such reaction can be seen at the 60 cm sensor (arrow 4 in Fig. 7b). As water is apparently not transported to this point vertically, this seems to be the result of lateral water input, causing a slow trailing "wave" at this depth.
The event on 27 May 2005 (Fig. 7c) has a very high total precipitation of 124 mm with a highest intensity of 3.2 mm/lO min. However, as this event is probably a rain on snow event, it is difficult to estimate the actual amount of water entering the soil. Antecedent moisture contents are high (at or above median values). While the response of discharge (3.22 m 3 / s increase), and ground water levels (120cm increase) is extremely strong, soil moisture shows a much less pronounced reaction. One possible explanation is that this is not only an event with high rainfall amounts, but that snow was also present in the catchment at this time (30 cm of www.hydrol-earth-syst-sci.net/13/1215/2009/ snow were measured at the climate station just outside of the research catchment at 1270 m elevation, while the soil moisture transect is located at about 1140 m elevation.). Therefore some of the runoff might be generated at the snow surface or within the snow layer. However, the extreme response at well W5 (see also Fig. 3) proves that large amounts of water did indeed enter the subsurface. A more likely explanation is therefore that as all water in excess of field capacity is being transported quickly to greater depths, soil moisture increases most for dry antecedent conditions and less in conditions of high antecedent wetness despite the fact that large amounts of water are being transported during nearly steady state flow conditions. This efficient vertical water transport will result in a pronounced response in the saturated zone without causing similarly pronounced peaks in moisture content. The most prominent patterns for this event are: a) slow vertical water transport, probably due to lower rainfall intensities (arrow 5 in Fig. 7 c and b) strong response at 40 cm depth for probe 1, very local and short-term (arrow 6 in Fig. 7c). This reaction might be due to an underlying capillary barrier, causing the water to pond above it until breakthrough. This pattern was observed at this location quite frequently (for 15 events out of 34).

Dye tracer rainfall simulation
In May 2006 rainfall simulation experiments with blue dye were carried out at the locations of the three continuously measuring probes. The soil moisture dynamics of these three experiments are shown in Fig. 8, with space-time colour maps equivalent to Fig. 7. As the same amount of dye was applied over the same amount of time during each of these experiments, the three experiments are plotted in one single figure as if corresponding to a single rainfall event. The time period and intensity of dye application is plotted in the top bar. The same colour scale was applied for the intensity of application as for the rainfall intensities in Fig. 7. Neither streamflow nor groundwater level dynamics are plotted as there was no reaction to these small scale experiments (small in comparison to the size of the hillslope). Antecedent moisture contents are quite high (at median values for all but the 10 cm sensors at probes 1 and 2). The dynamic moisture response patterns show fast/preferential vertical flow for probes 1 and 2 and slow vertical water transfer for probe 3 (Fig. 8). One day after the sprinkling experiment, cross sections of the infiltration plots were excavated and the dye stain patterns marking the flow paths of the dye in the unsaturated zone were photographed. The three photos of the cross sections at the locations of the soil moisture probes are shown in Fig. 9. Preferential flow is found at all three plots (compare Fig. 2). Flow occurred in plumes, which are separated by distinct areas of little or no flow unmarked by blue dye. Figure 9a shows the flow paths of probe 1 (located at the bottom of the slope). While blue dye can be seen in the top 5 cm, hardly any dye stains could be found in depths of 5-Hydrol. Earth Syst. Sci., 13, 1215-1234, 2009 ca. 30 cm (arrow 1 in Fig. 9a). This is most likely the suspected zone of hydrophobicity which was also found in the analysis of the time-space maps of soil moisture response to rainfall events. This zone of hydrophobicity or water repellency is most pronounced after summer dry periods but is still visible at the time of the sprinkling experiment where only little reaction was seen at the 10 cm sensor of probe 1 (Fig. 8). Distinct plumes of dye can be found at depths of ca. 30-60 cm (arrow 2 in Fig. 9a) (also at the location of the soil moisture probe), just above a very pronounced layer interface between the silty sand layer above and the gravelly layer below (arrow 3 in Fig. 9a). This confirms the hypothesis that a capillary barrier could be the cause of the ponding at the 40 cm sensor which was seen in the event response analysis (Fig. 8, arrow A). The dye stains also indicated the locations were water leached into the capillary barrier (arrow  flow patterns caused by water repellency of the soil. In late summer the 10 cm and sometimes also the 20 cm sensor are surrounded by hydrophobic soil (Fig. 7a and b).
The cross section at probe 2 ( Fig. 9b) shows as most distinct feature the saprolite layer (weathered bedrock) starting at the location of the 60cm sensor (arrow 1 in Fig. 9b). The 100 cm sensor is thus located within the saprolite. It was also found that the probe is located within a preferential flow path coinciding with a concentration of fine roots (arrow 2 in Fig. 9b). Maximum infiltration depth is about 80 cm in the three major plumes. A hydrophobic layer with very little staining can be seen in the cross section (arrow 3 in Fig. 9b). However, this layer was not identified in the soil moisture www.hydrol-earth-syst-sci.net/13/1215/2009/ data, as the probe is located within the preferential flow path and not in a hydrophobic patch. The high velocity of flow and the strong response in this preferential flow path is also visible in Fig. 8 (arrow B) and was also a feature of the soil moisture response space-time maps at this location.
The soil at probe 3 ( Fig. 9c) differed compared to the two others as the vegetation at this plot included a thicket of low shrubs, causing a higher density of roots in the top 20 cm (arrow 1 in Fig. 9c). Here, the probe was located in between dye stained preferential flow paths. While blue dye is found in the vicinity of the probe at depths 10-20 cm, very little of it is found close to the probe at greater depths. The 60 cm sensor is located just at the interface between the silty sand Hydrol. Earth Syst. Sci., 13, 1215-1234,2009 and a layer of fine gravel (arrow 2 in Fig. 9c), thus probably measuring in both layers, while the 100 cm sensor is situated in a layer of more compacted silty sand starting at a depth of approx. 75 cm. Maximum depth of infiltration is 80 cm. The fact that the layer at the 100 cm sensor is more compacted might explain why reaction at this sensor occurs delayed and prolonged. This would correspond to lower hydraulic conductivities in the compacted layer causing a delay in response and a prolonged peak. However, as the response at the 60 cm sensor is often very weak, the water causing the peak at 100 cm depths is most likely transported to this point not vertically but laterally. The dense root zone in the top 20 cm explains the strong reaction at the 20 cm sensor (Fig. 8, arrow C). Probe 3 shows a slower reaction to rainfall compared to the other two probes ( Fig. 8 and also Fig. 7), which is explained by the fact that this probe is not situated within a preferential flow path.

Response times
To test if the seasonal change in subsurface flow has an effect on overall catchment response, response times for soil moisture, groundwater and stream flow were calculated for the wet and dry season separately (Fig. 10). This analysis is based on 27 rainfall events between December 2004 and April 2006. S20, S30, S40 are the response lags of the soil moisture sensors at 20, 30 and 40 cm depth, GW is the response lag of ground water level at well Wl (Fig. 1) and Q is the response lag of stream flow. Groundwater response is generally slower than stream flow response. (At this hillslope the ground water surface at well Wl in the vicinity of the stream is generally about 60 cm below the level of the stream bed.) Surprisingly, the soil moisture sensors often react slower than stream flow. This could either mean that rainfall is not uniformly distributed over the catchment or that these sensors are bypassed by preferential flow paths. Surface runoff is unlikely, due to high infiltration rates and porosities and has not been observed during field campaigns. Another possible explanation would be lateral flow in the duff layer/organic horizon. Response lags of all parameters show similar behavior over time: response times are short from January to April (summer and early fall) when compared to the winter months. This is probably the result of higher rainfall intensities on the one hand (median maximum intensities are 2.1 mm/1 0 min in winter and 3.0 mm/lO min in summer) and enhanced preferential flow due to hydrophobicity on the other hand. Snow was present/melting during 4 of the winter events, however, this does not lead to a consistent change in response time: both, faster and slower than median responses have been observed for these events. Streamflow response times for the events shown in the space-time colour maps (Fig. 7) range from 30 minutes for the first event (driest antecedent moisture conditions), 1:50 h for the second event up to 6:40 h for the event in May (wettest antecedent conditions).

Summary and conclusions
The soil moisture data obtained in this study provided diverse insights in subsurface flow and runoff generation processes in this catchment. It was shown that high resolution time series of soil moisture in combination with manual measurements at irregular time intervals can be a valuable addition to time series of precipitation and discharge when investigating runoff generation processes. This is especially true for catchments where only short time series of data are available, as in the Malalcahuello Catchment. The approach of combining datasets with different spatio-temporal resolution allowed for the investigation of soil moisture dynamics as well as patterns and proved to be less expensive than high density installation of continuously logging sensors while also being applicable to difficult terrain, i.e. densely forested and steep hillslopes. The synergetic effects achieved with this combination of datasets is visualized in Fig. 2.
The time series of soil moisture for the 19 month period in Fig. 3 show that spatial variability of soil moisture is much higher than its temporal variability. Both, rainfall events and droughts only cause small, fast and short perturbations and the moisture content quickly rebounds to previous levels. This behaviour corresponds to a system at or close to a dynamic equilibrium, i.e. close to steady state flow conditions in the subsurface. This seems to prevail most time of the year. This is corroborated by the high annual precipitation, high baseflow index and annual runoff and the very low event runoff coefficients.
The use of space-time colour maps facilitated the analysis of soil moisture response dynamics, especially concerning the timing and extent of response along the vertical profile. It was thus possible to identify a number of patterns which can be attributed to different phenomena of flow in the unsaturated zone. The very subdued response of soil moisture in the upper soil horizon at two locations during the driest period (late summer) was attributed to the formation or reinforcement of hydrophobicity in this layer. The accumulationiponding of water at certain depths was assumed to be due to the effect of capillary barriers. This was confirmed by the dye tracer experiment carried out at this location.
Strong response at certain depths while the layers just above show little reaction indicate the importance of lateral flow processes. However, we do not know on what scale these thus "observed" lateral flow paths are active (several decimeters , meters or hills lope scale). Lateral flow has also been directly observed in a dye tracer experimenet with an application intensity of 20 mmlh. Here lateral flow occurred in the duff layer (Blume et aI., 2008b). The short response times of streamflow also indicate that lateral subsurface storm flow is likely to be important.
It was furthermore found that infiltration dynamics differed from summer to winter, which could be due to differences in rainfall intensities as well as the amplification of preferential flow due to hydrophobicity in the top layer.
www.hydrol-earth-syst-sci.net/13/1215/2009/ Persistency of potential water repellency was found to be strong to extreme for the upper horizon. Hydrophobicity has also been observed in Chilean young volcanic ash soils by other researchers (Bachmann et aI., 2000;Ellies, 1975) and is also of importance in volcanic ash soils of Ecuador (Poulenard et aI. , 2004). In addition, differences in flow patterns from dry to wet period were found in the Malalcahuello Catchment during a more extensive study involving a total of 10 dye tracer experiments (Blume et aI., 2008b). The change in flow pattern observed in this study further supports the theory that preferential flow in this catchment is reinforced by hydrophobicity (application intensities were the same for dry and wet season experiments). Similar flow patterns also attributed to hydrophobicity were observed in other studies Ritsema et aI. , 1998;Ritsema and Dekker, 1994;Dekker and Ritsema, 2000;de Rooij , 2000). The fact that throughfall amounts are highly heterogenous in this catchment (Blume et aI. , 2008a) is likely to be the reason why some locations (probably on the decimeter scale) are drier than others and thus more likely to develop water repellency. Spots of high water input are therefore likely to become preferential flow paths. These observed patterns in dynamics were found to be spatially and temporally persistent insofar as the event pattern dynamics of soil moisture observed in Fall 2005 (Fig. 7) matched well with the flow patterns found during the dye tracer experiments one year later. The persistency of the spatial patterns of soil moisture for 14 locations and 6 depths (Fig. 4) shows that spatial variability is much higher than temporal variability and that wetter locations are likely to remain wet. Furthermore the patterns of soil moisture variability at the decimeter scale, which were attributed to the presence/absence of preferential flow paths, were also found to be persistent over a period of more than one and a half years. The stationarity of these flow patterns is another indicator that steady state conditions prevail in this catchment.
Hydrophobicity is the most likely explanation for the flow patterns found here. However, the effects of hydrophobicity are probably aggravated by root channels, strong gradients in matric potential and the hysteresis of the soil moisture characteristic curves of volcanic ash soils as described by Shoji et aI. (1993).
The last and maybe most important question is the question of how important this locally observed preferential flow is for the system as a whole, i.e. runoff response/runoff generation at the catchment scale.
The small response of soil moisture dynamics to perturbations as well as persistency/stationarity of flow paths corroborate our perception that this undisturbed, forested catchment is at or close to steady state, i.e. a dynamic equilibrium. This perception was originally based on integral observations such as the fact that we observe high annual runoff and a high baseflow index, while event runoff coefficients are low, and has now been corroborated by internal observations, e.g. soil moisture response and flow patterns. Efficient water transwww.hydrol-earth-syst-sci.net/13/1215/2009/ port, as observed locally, therefore would indeed have an effect at the catchment scale.
Several additional findings indicate that while preferential flow was only observed at the plot scale it might indeed be an important factor of runoff generation at the catchment scale. That preferential flow occurs throughout the catchment is indicated by the fact that additionally to the three tracer experiments shown in this study all 9 dye tracer experiments carried out under forest at various locations in the catchment showed preferential flow patterns (Blume et aI. , 2008b). The fact that 68 % of the sensors at the 11 manual measurement points showed small scale soil moisture variability is another indicator for the importance of these preferential flow paths. Last but not least the analysis of response times for soil moisture, groundwater and streamflow reveiled that response lags are generally much shorter during the summer months were preferential flow is also likely to be further enforced by stronger hydrophobicity. Interestingly streamflow often shows faster response than both ground water and soil water. This might be due to non-uniform rainfall distribution (i.e. earlier onset of rainfall further up in the catchment causing stream levels to respond while soil moisture at the slope close to the catchment outlet remained unchanged). However, as soil moisture response measurements are restricted to only three locations it is also likely that there are other preferential flow paths with even faster response than the ones measured by our instruments. In this case preferential flow in the vertical and then a fast reaction along a horizontal layer interface might be the reason for the short response lags of streamflow found in this catchment. (Finger flow is known to cause faster breakthrough as investigated by de Rooij and deVries (1996) in a modelling study.) The question whether or not these preferential flow processes are important for catchment response could be investigated further by application of a physically based hydrological model either on the hillslope or on the catchment scale.
To summarize the main conclusions in short: 1. the synergy of soil moisture datasets with different spatio-temporal resolution proved to be useful for the investigation of subsurface flow processes. Continuously monitored rainfall experiments at the location of the moisture probes with subsequent excavation of dye stained soil profiles facilitated testing/corroboration of the perception of subsurface flow gained from the moisture patterns.
2. data-visualization with space-time colour maps permits a much more detailed analysis of soil moisture response than simple line plots alone 3. soil moisturelftow patterns in the here investigated young volcanic ash soils were shown to be persistent in time and highly variable in space 4. the most likely explanation for the observed flow patterns is a combination of hydrophobicity with strong HydroI. Earth Syst. Sci., 13, 1215Sci., 13, -1234Sci., 13, ,2009 gradients in unsaturated conductivities, where flow paths are initiated either by the presence of roots or the highly heterogeneous distribution of throughfall and thus water input 5. this soil moisture data set has provided us with internal observations corroborating our perception that the catchment is at or close to steady state/dynamic equilibrium, which was originally based on integral data, mainly rainfall and runoff time series.
6. the flow patterns observed at the local scale are likely to be important for runoff response at the catchment scale.